fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53 #include "free-space-cache.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static const struct inode_operations btrfs_dir_inode_operations;
  61 static const struct inode_operations btrfs_symlink_inode_operations;
  62 static const struct inode_operations btrfs_dir_ro_inode_operations;
  63 static const struct inode_operations btrfs_special_inode_operations;
  64 static const struct inode_operations btrfs_file_inode_operations;
  65 static const struct address_space_operations btrfs_aops;
  66 static const struct address_space_operations btrfs_symlink_aops;
  67 static const struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  88 static int btrfs_truncate(struct inode *inode);
  89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, int *page_started,
  93                                    unsigned long *nr_written, int unlock);
  94
  95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  96                                      struct inode *inode,  struct inode *dir,
  97                                      const struct qstr *qstr)
  98 {
  99         int err;
 100
 101         err = btrfs_init_acl(trans, inode, dir);
 102         if (!err)
 103                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 104         return err;
 105 }
 106
 107 /*
 108  * this does all the hard work for inserting an inline extent into
 109  * the btree.  The caller should have done a btrfs_drop_extents so that
 110  * no overlapping inline items exist in the btree
 111  */
 112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 113                                 struct btrfs_root *root, struct inode *inode,
 114                                 u64 start, size_t size, size_t compressed_size,
 115                                 int compress_type,
 116                                 struct page **compressed_pages)
 117 {
 118         struct btrfs_key key;
 119         struct btrfs_path *path;
 120         struct extent_buffer *leaf;
 121         struct page *page = NULL;
 122         char *kaddr;
 123         unsigned long ptr;
 124         struct btrfs_file_extent_item *ei;
 125         int err = 0;
 126         int ret;
 127         size_t cur_size = size;
 128         size_t datasize;
 129         unsigned long offset;
 130
 131         if (compressed_size && compressed_pages)
 132                 cur_size = compressed_size;
 133
 134         path = btrfs_alloc_path();
 135         if (!path)
 136                 return -ENOMEM;
 137
 138         path->leave_spinning = 1;
 139         btrfs_set_trans_block_group(trans, inode);
 140
 141         key.objectid = inode->i_ino;
 142         key.offset = start;
 143         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 144         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 145
 146         inode_add_bytes(inode, size);
 147         ret = btrfs_insert_empty_item(trans, root, path, &key,
 148                                       datasize);
 149         BUG_ON(ret);
 150         if (ret) {
 151                 err = ret;
 152                 goto fail;
 153         }
 154         leaf = path->nodes[0];
 155         ei = btrfs_item_ptr(leaf, path->slots[0],
 156                             struct btrfs_file_extent_item);
 157         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 158         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 159         btrfs_set_file_extent_encryption(leaf, ei, 0);
 160         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 161         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 162         ptr = btrfs_file_extent_inline_start(ei);
 163
 164         if (compress_type != BTRFS_COMPRESS_NONE) {
 165                 struct page *cpage;
 166                 int i = 0;
 167                 while (compressed_size > 0) {
 168                         cpage = compressed_pages[i];
 169                         cur_size = min_t(unsigned long, compressed_size,
 170                                        PAGE_CACHE_SIZE);
 171
 172                         kaddr = kmap_atomic(cpage, KM_USER0);
 173                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 174                         kunmap_atomic(kaddr, KM_USER0);
 175
 176                         i++;
 177                         ptr += cur_size;
 178                         compressed_size -= cur_size;
 179                 }
 180                 btrfs_set_file_extent_compression(leaf, ei,
 181                                                   compress_type);
 182         } else {
 183                 page = find_get_page(inode->i_mapping,
 184                                      start >> PAGE_CACHE_SHIFT);
 185                 btrfs_set_file_extent_compression(leaf, ei, 0);
 186                 kaddr = kmap_atomic(page, KM_USER0);
 187                 offset = start & (PAGE_CACHE_SIZE - 1);
 188                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 189                 kunmap_atomic(kaddr, KM_USER0);
 190                 page_cache_release(page);
 191         }
 192         btrfs_mark_buffer_dirty(leaf);
 193         btrfs_free_path(path);
 194
 195         /*
 196          * we're an inline extent, so nobody can
 197          * extend the file past i_size without locking
 198          * a page we already have locked.
 199          *
 200          * We must do any isize and inode updates
 201          * before we unlock the pages.  Otherwise we
 202          * could end up racing with unlink.
 203          */
 204         BTRFS_I(inode)->disk_i_size = inode->i_size;
 205         btrfs_update_inode(trans, root, inode);
 206
 207         return 0;
 208 fail:
 209         btrfs_free_path(path);
 210         return err;
 211 }
 212
 213
 214 /*
 215  * conditionally insert an inline extent into the file.  This
 216  * does the checks required to make sure the data is small enough
 217  * to fit as an inline extent.
 218  */
 219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 220                                  struct btrfs_root *root,
 221                                  struct inode *inode, u64 start, u64 end,
 222                                  size_t compressed_size, int compress_type,
 223                                  struct page **compressed_pages)
 224 {
 225         u64 isize = i_size_read(inode);
 226         u64 actual_end = min(end + 1, isize);
 227         u64 inline_len = actual_end - start;
 228         u64 aligned_end = (end + root->sectorsize - 1) &
 229                         ~((u64)root->sectorsize - 1);
 230         u64 hint_byte;
 231         u64 data_len = inline_len;
 232         int ret;
 233
 234         if (compressed_size)
 235                 data_len = compressed_size;
 236
 237         if (start > 0 ||
 238             actual_end >= PAGE_CACHE_SIZE ||
 239             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 240             (!compressed_size &&
 241             (actual_end & (root->sectorsize - 1)) == 0) ||
 242             end + 1 < isize ||
 243             data_len > root->fs_info->max_inline) {
 244                 return 1;
 245         }
 246
 247         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 248                                  &hint_byte, 1);
 249         BUG_ON(ret);
 250
 251         if (isize > actual_end)
 252                 inline_len = min_t(u64, isize, actual_end);
 253         ret = insert_inline_extent(trans, root, inode, start,
 254                                    inline_len, compressed_size,
 255                                    compress_type, compressed_pages);
 256         BUG_ON(ret);
 257         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 258         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 259         return 0;
 260 }
 261
 262 struct async_extent {
 263         u64 start;
 264         u64 ram_size;
 265         u64 compressed_size;
 266         struct page **pages;
 267         unsigned long nr_pages;
 268         int compress_type;
 269         struct list_head list;
 270 };
 271
 272 struct async_cow {
 273         struct inode *inode;
 274         struct btrfs_root *root;
 275         struct page *locked_page;
 276         u64 start;
 277         u64 end;
 278         struct list_head extents;
 279         struct btrfs_work work;
 280 };
 281
 282 static noinline int add_async_extent(struct async_cow *cow,
 283                                      u64 start, u64 ram_size,
 284                                      u64 compressed_size,
 285                                      struct page **pages,
 286                                      unsigned long nr_pages,
 287                                      int compress_type)
 288 {
 289         struct async_extent *async_extent;
 290
 291         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 292         BUG_ON(!async_extent);
 293         async_extent->start = start;
 294         async_extent->ram_size = ram_size;
 295         async_extent->compressed_size = compressed_size;
 296         async_extent->pages = pages;
 297         async_extent->nr_pages = nr_pages;
 298         async_extent->compress_type = compress_type;
 299         list_add_tail(&async_extent->list, &cow->extents);
 300         return 0;
 301 }
 302
 303 /*
 304  * we create compressed extents in two phases.  The first
 305  * phase compresses a range of pages that have already been
 306  * locked (both pages and state bits are locked).
 307  *
 308  * This is done inside an ordered work queue, and the compression
 309  * is spread across many cpus.  The actual IO submission is step
 310  * two, and the ordered work queue takes care of making sure that
 311  * happens in the same order things were put onto the queue by
 312  * writepages and friends.
 313  *
 314  * If this code finds it can't get good compression, it puts an
 315  * entry onto the work queue to write the uncompressed bytes.  This
 316  * makes sure that both compressed inodes and uncompressed inodes
 317  * are written in the same order that pdflush sent them down.
 318  */
 319 static noinline int compress_file_range(struct inode *inode,
 320                                         struct page *locked_page,
 321                                         u64 start, u64 end,
 322                                         struct async_cow *async_cow,
 323                                         int *num_added)
 324 {
 325         struct btrfs_root *root = BTRFS_I(inode)->root;
 326         struct btrfs_trans_handle *trans;
 327         u64 num_bytes;
 328         u64 blocksize = root->sectorsize;
 329         u64 actual_end;
 330         u64 isize = i_size_read(inode);
 331         int ret = 0;
 332         struct page **pages = NULL;
 333         unsigned long nr_pages;
 334         unsigned long nr_pages_ret = 0;
 335         unsigned long total_compressed = 0;
 336         unsigned long total_in = 0;
 337         unsigned long max_compressed = 128 * 1024;
 338         unsigned long max_uncompressed = 128 * 1024;
 339         int i;
 340         int will_compress;
 341         int compress_type = root->fs_info->compress_type;
 342
 343         actual_end = min_t(u64, isize, end + 1);
 344 again:
 345         will_compress = 0;
 346         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 347         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 348
 349         /*
 350          * we don't want to send crud past the end of i_size through
 351          * compression, that's just a waste of CPU time.  So, if the
 352          * end of the file is before the start of our current
 353          * requested range of bytes, we bail out to the uncompressed
 354          * cleanup code that can deal with all of this.
 355          *
 356          * It isn't really the fastest way to fix things, but this is a
 357          * very uncommon corner.
 358          */
 359         if (actual_end <= start)
 360                 goto cleanup_and_bail_uncompressed;
 361
 362         total_compressed = actual_end - start;
 363
 364         /* we want to make sure that amount of ram required to uncompress
 365          * an extent is reasonable, so we limit the total size in ram
 366          * of a compressed extent to 128k.  This is a crucial number
 367          * because it also controls how easily we can spread reads across
 368          * cpus for decompression.
 369          *
 370          * We also want to make sure the amount of IO required to do
 371          * a random read is reasonably small, so we limit the size of
 372          * a compressed extent to 128k.
 373          */
 374         total_compressed = min(total_compressed, max_uncompressed);
 375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 376         num_bytes = max(blocksize,  num_bytes);
 377         total_in = 0;
 378         ret = 0;
 379
 380         /*
 381          * we do compression for mount -o compress and when the
 382          * inode has not been flagged as nocompress.  This flag can
 383          * change at any time if we discover bad compression ratios.
 384          */
 385         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 386             (btrfs_test_opt(root, COMPRESS) ||
 387              (BTRFS_I(inode)->force_compress) ||
 388              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 389                 WARN_ON(pages);
 390                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 391                 BUG_ON(!pages);
 392
 393                 if (BTRFS_I(inode)->force_compress)
 394                         compress_type = BTRFS_I(inode)->force_compress;
 395
 396                 ret = btrfs_compress_pages(compress_type,
 397                                            inode->i_mapping, start,
 398                                            total_compressed, pages,
 399                                            nr_pages, &nr_pages_ret,
 400                                            &total_in,
 401                                            &total_compressed,
 402                                            max_compressed);
 403
 404                 if (!ret) {
 405                         unsigned long offset = total_compressed &
 406                                 (PAGE_CACHE_SIZE - 1);
 407                         struct page *page = pages[nr_pages_ret - 1];
 408                         char *kaddr;
 409
 410                         /* zero the tail end of the last page, we might be
 411                          * sending it down to disk
 412                          */
 413                         if (offset) {
 414                                 kaddr = kmap_atomic(page, KM_USER0);
 415                                 memset(kaddr + offset, 0,
 416                                        PAGE_CACHE_SIZE - offset);
 417                                 kunmap_atomic(kaddr, KM_USER0);
 418                         }
 419                         will_compress = 1;
 420                 }
 421         }
 422         if (start == 0) {
 423                 trans = btrfs_join_transaction(root, 1);
 424                 BUG_ON(IS_ERR(trans));
 425                 btrfs_set_trans_block_group(trans, inode);
 426                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 427
 428                 /* lets try to make an inline extent */
 429                 if (ret || total_in < (actual_end - start)) {
 430                         /* we didn't compress the entire range, try
 431                          * to make an uncompressed inline extent.
 432                          */
 433                         ret = cow_file_range_inline(trans, root, inode,
 434                                                     start, end, 0, 0, NULL);
 435                 } else {
 436                         /* try making a compressed inline extent */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end,
 439                                                     total_compressed,
 440                                                     compress_type, pages);
 441                 }
 442                 if (ret == 0) {
 443                         /*
 444                          * inline extent creation worked, we don't need
 445                          * to create any more async work items.  Unlock
 446                          * and free up our temp pages.
 447                          */
 448                         extent_clear_unlock_delalloc(inode,
 449                              &BTRFS_I(inode)->io_tree,
 450                              start, end, NULL,
 451                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 452                              EXTENT_CLEAR_DELALLOC |
 453                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 454
 455                         btrfs_end_transaction(trans, root);
 456                         goto free_pages_out;
 457                 }
 458                 btrfs_end_transaction(trans, root);
 459         }
 460
 461         if (will_compress) {
 462                 /*
 463                  * we aren't doing an inline extent round the compressed size
 464                  * up to a block size boundary so the allocator does sane
 465                  * things
 466                  */
 467                 total_compressed = (total_compressed + blocksize - 1) &
 468                         ~(blocksize - 1);
 469
 470                 /*
 471                  * one last check to make sure the compression is really a
 472                  * win, compare the page count read with the blocks on disk
 473                  */
 474                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 475                         ~(PAGE_CACHE_SIZE - 1);
 476                 if (total_compressed >= total_in) {
 477                         will_compress = 0;
 478                 } else {
 479                         num_bytes = total_in;
 480                 }
 481         }
 482         if (!will_compress && pages) {
 483                 /*
 484                  * the compression code ran but failed to make things smaller,
 485                  * free any pages it allocated and our page pointer array
 486                  */
 487                 for (i = 0; i < nr_pages_ret; i++) {
 488                         WARN_ON(pages[i]->mapping);
 489                         page_cache_release(pages[i]);
 490                 }
 491                 kfree(pages);
 492                 pages = NULL;
 493                 total_compressed = 0;
 494                 nr_pages_ret = 0;
 495
 496                 /* flag the file so we don't compress in the future */
 497                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 498                     !(BTRFS_I(inode)->force_compress)) {
 499                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 500                 }
 501         }
 502         if (will_compress) {
 503                 *num_added += 1;
 504
 505                 /* the async work queues will take care of doing actual
 506                  * allocation on disk for these compressed pages,
 507                  * and will submit them to the elevator.
 508                  */
 509                 add_async_extent(async_cow, start, num_bytes,
 510                                  total_compressed, pages, nr_pages_ret,
 511                                  compress_type);
 512
 513                 if (start + num_bytes < end) {
 514                         start += num_bytes;
 515                         pages = NULL;
 516                         cond_resched();
 517                         goto again;
 518                 }
 519         } else {
 520 cleanup_and_bail_uncompressed:
 521                 /*
 522                  * No compression, but we still need to write the pages in
 523                  * the file we've been given so far.  redirty the locked
 524                  * page if it corresponds to our extent and set things up
 525                  * for the async work queue to run cow_file_range to do
 526                  * the normal delalloc dance
 527                  */
 528                 if (page_offset(locked_page) >= start &&
 529                     page_offset(locked_page) <= end) {
 530                         __set_page_dirty_nobuffers(locked_page);
 531                         /* unlocked later on in the async handlers */
 532                 }
 533                 add_async_extent(async_cow, start, end - start + 1,
 534                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 535                 *num_added += 1;
 536         }
 537
 538 out:
 539         return 0;
 540
 541 free_pages_out:
 542         for (i = 0; i < nr_pages_ret; i++) {
 543                 WARN_ON(pages[i]->mapping);
 544                 page_cache_release(pages[i]);
 545         }
 546         kfree(pages);
 547
 548         goto out;
 549 }
 550
 551 /*
 552  * phase two of compressed writeback.  This is the ordered portion
 553  * of the code, which only gets called in the order the work was
 554  * queued.  We walk all the async extents created by compress_file_range
 555  * and send them down to the disk.
 556  */
 557 static noinline int submit_compressed_extents(struct inode *inode,
 558                                               struct async_cow *async_cow)
 559 {
 560         struct async_extent *async_extent;
 561         u64 alloc_hint = 0;
 562         struct btrfs_trans_handle *trans;
 563         struct btrfs_key ins;
 564         struct extent_map *em;
 565         struct btrfs_root *root = BTRFS_I(inode)->root;
 566         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 567         struct extent_io_tree *io_tree;
 568         int ret = 0;
 569
 570         if (list_empty(&async_cow->extents))
 571                 return 0;
 572
 573
 574         while (!list_empty(&async_cow->extents)) {
 575                 async_extent = list_entry(async_cow->extents.next,
 576                                           struct async_extent, list);
 577                 list_del(&async_extent->list);
 578
 579                 io_tree = &BTRFS_I(inode)->io_tree;
 580
 581 retry:
 582                 /* did the compression code fall back to uncompressed IO? */
 583                 if (!async_extent->pages) {
 584                         int page_started = 0;
 585                         unsigned long nr_written = 0;
 586
 587                         lock_extent(io_tree, async_extent->start,
 588                                          async_extent->start +
 589                                          async_extent->ram_size - 1, GFP_NOFS);
 590
 591                         /* allocate blocks */
 592                         ret = cow_file_range(inode, async_cow->locked_page,
 593                                              async_extent->start,
 594                                              async_extent->start +
 595                                              async_extent->ram_size - 1,
 596                                              &page_started, &nr_written, 0);
 597
 598                         /*
 599                          * if page_started, cow_file_range inserted an
 600                          * inline extent and took care of all the unlocking
 601                          * and IO for us.  Otherwise, we need to submit
 602                          * all those pages down to the drive.
 603                          */
 604                         if (!page_started && !ret)
 605                                 extent_write_locked_range(io_tree,
 606                                                   inode, async_extent->start,
 607                                                   async_extent->start +
 608                                                   async_extent->ram_size - 1,
 609                                                   btrfs_get_extent,
 610                                                   WB_SYNC_ALL);
 611                         kfree(async_extent);
 612                         cond_resched();
 613                         continue;
 614                 }
 615
 616                 lock_extent(io_tree, async_extent->start,
 617                             async_extent->start + async_extent->ram_size - 1,
 618                             GFP_NOFS);
 619
 620                 trans = btrfs_join_transaction(root, 1);
 621                 BUG_ON(IS_ERR(trans));
 622                 ret = btrfs_reserve_extent(trans, root,
 623                                            async_extent->compressed_size,
 624                                            async_extent->compressed_size,
 625                                            0, alloc_hint,
 626                                            (u64)-1, &ins, 1);
 627                 btrfs_end_transaction(trans, root);
 628
 629                 if (ret) {
 630                         int i;
 631                         for (i = 0; i < async_extent->nr_pages; i++) {
 632                                 WARN_ON(async_extent->pages[i]->mapping);
 633                                 page_cache_release(async_extent->pages[i]);
 634                         }
 635                         kfree(async_extent->pages);
 636                         async_extent->nr_pages = 0;
 637                         async_extent->pages = NULL;
 638                         unlock_extent(io_tree, async_extent->start,
 639                                       async_extent->start +
 640                                       async_extent->ram_size - 1, GFP_NOFS);
 641                         goto retry;
 642                 }
 643
 644                 /*
 645                  * here we're doing allocation and writeback of the
 646                  * compressed pages
 647                  */
 648                 btrfs_drop_extent_cache(inode, async_extent->start,
 649                                         async_extent->start +
 650                                         async_extent->ram_size - 1, 0);
 651
 652                 em = alloc_extent_map();
 653                 BUG_ON(!em);
 654                 em->start = async_extent->start;
 655                 em->len = async_extent->ram_size;
 656                 em->orig_start = em->start;
 657
 658                 em->block_start = ins.objectid;
 659                 em->block_len = ins.offset;
 660                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 661                 em->compress_type = async_extent->compress_type;
 662                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 663                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 664
 665                 while (1) {
 666                         write_lock(&em_tree->lock);
 667                         ret = add_extent_mapping(em_tree, em);
 668                         write_unlock(&em_tree->lock);
 669                         if (ret != -EEXIST) {
 670                                 free_extent_map(em);
 671                                 break;
 672                         }
 673                         btrfs_drop_extent_cache(inode, async_extent->start,
 674                                                 async_extent->start +
 675                                                 async_extent->ram_size - 1, 0);
 676                 }
 677
 678                 ret = btrfs_add_ordered_extent_compress(inode,
 679                                                 async_extent->start,
 680                                                 ins.objectid,
 681                                                 async_extent->ram_size,
 682                                                 ins.offset,
 683                                                 BTRFS_ORDERED_COMPRESSED,
 684                                                 async_extent->compress_type);
 685                 BUG_ON(ret);
 686
 687                 /*
 688                  * clear dirty, set writeback and unlock the pages.
 689                  */
 690                 extent_clear_unlock_delalloc(inode,
 691                                 &BTRFS_I(inode)->io_tree,
 692                                 async_extent->start,
 693                                 async_extent->start +
 694                                 async_extent->ram_size - 1,
 695                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 696                                 EXTENT_CLEAR_UNLOCK |
 697                                 EXTENT_CLEAR_DELALLOC |
 698                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 699
 700                 ret = btrfs_submit_compressed_write(inode,
 701                                     async_extent->start,
 702                                     async_extent->ram_size,
 703                                     ins.objectid,
 704                                     ins.offset, async_extent->pages,
 705                                     async_extent->nr_pages);
 706
 707                 BUG_ON(ret);
 708                 alloc_hint = ins.objectid + ins.offset;
 709                 kfree(async_extent);
 710                 cond_resched();
 711         }
 712
 713         return 0;
 714 }
 715
 716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 717                                       u64 num_bytes)
 718 {
 719         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 720         struct extent_map *em;
 721         u64 alloc_hint = 0;
 722
 723         read_lock(&em_tree->lock);
 724         em = search_extent_mapping(em_tree, start, num_bytes);
 725         if (em) {
 726                 /*
 727                  * if block start isn't an actual block number then find the
 728                  * first block in this inode and use that as a hint.  If that
 729                  * block is also bogus then just don't worry about it.
 730                  */
 731                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 732                         free_extent_map(em);
 733                         em = search_extent_mapping(em_tree, 0, 0);
 734                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 735                                 alloc_hint = em->block_start;
 736                         if (em)
 737                                 free_extent_map(em);
 738                 } else {
 739                         alloc_hint = em->block_start;
 740                         free_extent_map(em);
 741                 }
 742         }
 743         read_unlock(&em_tree->lock);
 744
 745         return alloc_hint;
 746 }
 747
 748 /*
 749  * when extent_io.c finds a delayed allocation range in the file,
 750  * the call backs end up in this code.  The basic idea is to
 751  * allocate extents on disk for the range, and create ordered data structs
 752  * in ram to track those extents.
 753  *
 754  * locked_page is the page that writepage had locked already.  We use
 755  * it to make sure we don't do extra locks or unlocks.
 756  *
 757  * *page_started is set to one if we unlock locked_page and do everything
 758  * required to start IO on it.  It may be clean and already done with
 759  * IO when we return.
 760  */
 761 static noinline int cow_file_range(struct inode *inode,
 762                                    struct page *locked_page,
 763                                    u64 start, u64 end, int *page_started,
 764                                    unsigned long *nr_written,
 765                                    int unlock)
 766 {
 767         struct btrfs_root *root = BTRFS_I(inode)->root;
 768         struct btrfs_trans_handle *trans;
 769         u64 alloc_hint = 0;
 770         u64 num_bytes;
 771         unsigned long ram_size;
 772         u64 disk_num_bytes;
 773         u64 cur_alloc_size;
 774         u64 blocksize = root->sectorsize;
 775         struct btrfs_key ins;
 776         struct extent_map *em;
 777         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 778         int ret = 0;
 779
 780         BUG_ON(root == root->fs_info->tree_root);
 781         trans = btrfs_join_transaction(root, 1);
 782         BUG_ON(IS_ERR(trans));
 783         btrfs_set_trans_block_group(trans, inode);
 784         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 785
 786         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 787         num_bytes = max(blocksize,  num_bytes);
 788         disk_num_bytes = num_bytes;
 789         ret = 0;
 790
 791         if (start == 0) {
 792                 /* lets try to make an inline extent */
 793                 ret = cow_file_range_inline(trans, root, inode,
 794                                             start, end, 0, 0, NULL);
 795                 if (ret == 0) {
 796                         extent_clear_unlock_delalloc(inode,
 797                                      &BTRFS_I(inode)->io_tree,
 798                                      start, end, NULL,
 799                                      EXTENT_CLEAR_UNLOCK_PAGE |
 800                                      EXTENT_CLEAR_UNLOCK |
 801                                      EXTENT_CLEAR_DELALLOC |
 802                                      EXTENT_CLEAR_DIRTY |
 803                                      EXTENT_SET_WRITEBACK |
 804                                      EXTENT_END_WRITEBACK);
 805
 806                         *nr_written = *nr_written +
 807                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 808                         *page_started = 1;
 809                         ret = 0;
 810                         goto out;
 811                 }
 812         }
 813
 814         BUG_ON(disk_num_bytes >
 815                btrfs_super_total_bytes(&root->fs_info->super_copy));
 816
 817         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 818         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 819
 820         while (disk_num_bytes > 0) {
 821                 unsigned long op;
 822
 823                 cur_alloc_size = disk_num_bytes;
 824                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 825                                            root->sectorsize, 0, alloc_hint,
 826                                            (u64)-1, &ins, 1);
 827                 BUG_ON(ret);
 828
 829                 em = alloc_extent_map();
 830                 BUG_ON(!em);
 831                 em->start = start;
 832                 em->orig_start = em->start;
 833                 ram_size = ins.offset;
 834                 em->len = ins.offset;
 835
 836                 em->block_start = ins.objectid;
 837                 em->block_len = ins.offset;
 838                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 839                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 840
 841                 while (1) {
 842                         write_lock(&em_tree->lock);
 843                         ret = add_extent_mapping(em_tree, em);
 844                         write_unlock(&em_tree->lock);
 845                         if (ret != -EEXIST) {
 846                                 free_extent_map(em);
 847                                 break;
 848                         }
 849                         btrfs_drop_extent_cache(inode, start,
 850                                                 start + ram_size - 1, 0);
 851                 }
 852
 853                 cur_alloc_size = ins.offset;
 854                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 855                                                ram_size, cur_alloc_size, 0);
 856                 BUG_ON(ret);
 857
 858                 if (root->root_key.objectid ==
 859                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 860                         ret = btrfs_reloc_clone_csums(inode, start,
 861                                                       cur_alloc_size);
 862                         BUG_ON(ret);
 863                 }
 864
 865                 if (disk_num_bytes < cur_alloc_size)
 866                         break;
 867
 868                 /* we're not doing compressed IO, don't unlock the first
 869                  * page (which the caller expects to stay locked), don't
 870                  * clear any dirty bits and don't set any writeback bits
 871                  *
 872                  * Do set the Private2 bit so we know this page was properly
 873                  * setup for writepage
 874                  */
 875                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 876                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 877                         EXTENT_SET_PRIVATE2;
 878
 879                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 880                                              start, start + ram_size - 1,
 881                                              locked_page, op);
 882                 disk_num_bytes -= cur_alloc_size;
 883                 num_bytes -= cur_alloc_size;
 884                 alloc_hint = ins.objectid + ins.offset;
 885                 start += cur_alloc_size;
 886         }
 887 out:
 888         ret = 0;
 889         btrfs_end_transaction(trans, root);
 890
 891         return ret;
 892 }
 893
 894 /*
 895  * work queue call back to started compression on a file and pages
 896  */
 897 static noinline void async_cow_start(struct btrfs_work *work)
 898 {
 899         struct async_cow *async_cow;
 900         int num_added = 0;
 901         async_cow = container_of(work, struct async_cow, work);
 902
 903         compress_file_range(async_cow->inode, async_cow->locked_page,
 904                             async_cow->start, async_cow->end, async_cow,
 905                             &num_added);
 906         if (num_added == 0)
 907                 async_cow->inode = NULL;
 908 }
 909
 910 /*
 911  * work queue call back to submit previously compressed pages
 912  */
 913 static noinline void async_cow_submit(struct btrfs_work *work)
 914 {
 915         struct async_cow *async_cow;
 916         struct btrfs_root *root;
 917         unsigned long nr_pages;
 918
 919         async_cow = container_of(work, struct async_cow, work);
 920
 921         root = async_cow->root;
 922         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 923                 PAGE_CACHE_SHIFT;
 924
 925         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 926
 927         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 928             5 * 1042 * 1024 &&
 929             waitqueue_active(&root->fs_info->async_submit_wait))
 930                 wake_up(&root->fs_info->async_submit_wait);
 931
 932         if (async_cow->inode)
 933                 submit_compressed_extents(async_cow->inode, async_cow);
 934 }
 935
 936 static noinline void async_cow_free(struct btrfs_work *work)
 937 {
 938         struct async_cow *async_cow;
 939         async_cow = container_of(work, struct async_cow, work);
 940         kfree(async_cow);
 941 }
 942
 943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 944                                 u64 start, u64 end, int *page_started,
 945                                 unsigned long *nr_written)
 946 {
 947         struct async_cow *async_cow;
 948         struct btrfs_root *root = BTRFS_I(inode)->root;
 949         unsigned long nr_pages;
 950         u64 cur_end;
 951         int limit = 10 * 1024 * 1042;
 952
 953         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 954                          1, 0, NULL, GFP_NOFS);
 955         while (start < end) {
 956                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 957                 BUG_ON(!async_cow);
 958                 async_cow->inode = inode;
 959                 async_cow->root = root;
 960                 async_cow->locked_page = locked_page;
 961                 async_cow->start = start;
 962
 963                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 964                         cur_end = end;
 965                 else
 966                         cur_end = min(end, start + 512 * 1024 - 1);
 967
 968                 async_cow->end = cur_end;
 969                 INIT_LIST_HEAD(&async_cow->extents);
 970
 971                 async_cow->work.func = async_cow_start;
 972                 async_cow->work.ordered_func = async_cow_submit;
 973                 async_cow->work.ordered_free = async_cow_free;
 974                 async_cow->work.flags = 0;
 975
 976                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 977                         PAGE_CACHE_SHIFT;
 978                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 979
 980                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 981                                    &async_cow->work);
 982
 983                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 984                         wait_event(root->fs_info->async_submit_wait,
 985                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 986                             limit));
 987                 }
 988
 989                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 990                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 991                         wait_event(root->fs_info->async_submit_wait,
 992                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 993                            0));
 994                 }
 995
 996                 *nr_written += nr_pages;
 997                 start = cur_end + 1;
 998         }
 999         *page_started = 1;
1000         return 0;
1001 }
1002
1003 static noinline int csum_exist_in_range(struct btrfs_root *root,
1004                                         u64 bytenr, u64 num_bytes)
1005 {
1006         int ret;
1007         struct btrfs_ordered_sum *sums;
1008         LIST_HEAD(list);
1009
1010         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1011                                        bytenr + num_bytes - 1, &list);
1012         if (ret == 0 && list_empty(&list))
1013                 return 0;
1014
1015         while (!list_empty(&list)) {
1016                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1017                 list_del(&sums->list);
1018                 kfree(sums);
1019         }
1020         return 1;
1021 }
1022
1023 /*
1024  * when nowcow writeback call back.  This checks for snapshots or COW copies
1025  * of the extents that exist in the file, and COWs the file as required.
1026  *
1027  * If no cow copies or snapshots exist, we write directly to the existing
1028  * blocks on disk
1029  */
1030 static noinline int run_delalloc_nocow(struct inode *inode,
1031                                        struct page *locked_page,
1032                               u64 start, u64 end, int *page_started, int force,
1033                               unsigned long *nr_written)
1034 {
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         struct btrfs_trans_handle *trans;
1037         struct extent_buffer *leaf;
1038         struct btrfs_path *path;
1039         struct btrfs_file_extent_item *fi;
1040         struct btrfs_key found_key;
1041         u64 cow_start;
1042         u64 cur_offset;
1043         u64 extent_end;
1044         u64 extent_offset;
1045         u64 disk_bytenr;
1046         u64 num_bytes;
1047         int extent_type;
1048         int ret;
1049         int type;
1050         int nocow;
1051         int check_prev = 1;
1052         bool nolock = false;
1053
1054         path = btrfs_alloc_path();
1055         BUG_ON(!path);
1056         if (root == root->fs_info->tree_root) {
1057                 nolock = true;
1058                 trans = btrfs_join_transaction_nolock(root, 1);
1059         } else {
1060                 trans = btrfs_join_transaction(root, 1);
1061         }
1062         BUG_ON(IS_ERR(trans));
1063
1064         cow_start = (u64)-1;
1065         cur_offset = start;
1066         while (1) {
1067                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1068                                                cur_offset, 0);
1069                 BUG_ON(ret < 0);
1070                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1071                         leaf = path->nodes[0];
1072                         btrfs_item_key_to_cpu(leaf, &found_key,
1073                                               path->slots[0] - 1);
1074                         if (found_key.objectid == inode->i_ino &&
1075                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1076                                 path->slots[0]--;
1077                 }
1078                 check_prev = 0;
1079 next_slot:
1080                 leaf = path->nodes[0];
1081                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1082                         ret = btrfs_next_leaf(root, path);
1083                         if (ret < 0)
1084                                 BUG_ON(1);
1085                         if (ret > 0)
1086                                 break;
1087                         leaf = path->nodes[0];
1088                 }
1089
1090                 nocow = 0;
1091                 disk_bytenr = 0;
1092                 num_bytes = 0;
1093                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094
1095                 if (found_key.objectid > inode->i_ino ||
1096                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1097                     found_key.offset > end)
1098                         break;
1099
1100                 if (found_key.offset > cur_offset) {
1101                         extent_end = found_key.offset;
1102                         extent_type = 0;
1103                         goto out_check;
1104                 }
1105
1106                 fi = btrfs_item_ptr(leaf, path->slots[0],
1107                                     struct btrfs_file_extent_item);
1108                 extent_type = btrfs_file_extent_type(leaf, fi);
1109
1110                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1111                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1112                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1113                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1114                         extent_end = found_key.offset +
1115                                 btrfs_file_extent_num_bytes(leaf, fi);
1116                         if (extent_end <= start) {
1117                                 path->slots[0]++;
1118                                 goto next_slot;
1119                         }
1120                         if (disk_bytenr == 0)
1121                                 goto out_check;
1122                         if (btrfs_file_extent_compression(leaf, fi) ||
1123                             btrfs_file_extent_encryption(leaf, fi) ||
1124                             btrfs_file_extent_other_encoding(leaf, fi))
1125                                 goto out_check;
1126                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127                                 goto out_check;
1128                         if (btrfs_extent_readonly(root, disk_bytenr))
1129                                 goto out_check;
1130                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1131                                                   found_key.offset -
1132                                                   extent_offset, disk_bytenr))
1133                                 goto out_check;
1134                         disk_bytenr += extent_offset;
1135                         disk_bytenr += cur_offset - found_key.offset;
1136                         num_bytes = min(end + 1, extent_end) - cur_offset;
1137                         /*
1138                          * force cow if csum exists in the range.
1139                          * this ensure that csum for a given extent are
1140                          * either valid or do not exist.
1141                          */
1142                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143                                 goto out_check;
1144                         nocow = 1;
1145                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1146                         extent_end = found_key.offset +
1147                                 btrfs_file_extent_inline_len(leaf, fi);
1148                         extent_end = ALIGN(extent_end, root->sectorsize);
1149                 } else {
1150                         BUG_ON(1);
1151                 }
1152 out_check:
1153                 if (extent_end <= start) {
1154                         path->slots[0]++;
1155                         goto next_slot;
1156                 }
1157                 if (!nocow) {
1158                         if (cow_start == (u64)-1)
1159                                 cow_start = cur_offset;
1160                         cur_offset = extent_end;
1161                         if (cur_offset > end)
1162                                 break;
1163                         path->slots[0]++;
1164                         goto next_slot;
1165                 }
1166
1167                 btrfs_release_path(path);
1168                 if (cow_start != (u64)-1) {
1169                         ret = cow_file_range(inode, locked_page, cow_start,
1170                                         found_key.offset - 1, page_started,
1171                                         nr_written, 1);
1172                         BUG_ON(ret);
1173                         cow_start = (u64)-1;
1174                 }
1175
1176                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1177                         struct extent_map *em;
1178                         struct extent_map_tree *em_tree;
1179                         em_tree = &BTRFS_I(inode)->extent_tree;
1180                         em = alloc_extent_map();
1181                         BUG_ON(!em);
1182                         em->start = cur_offset;
1183                         em->orig_start = em->start;
1184                         em->len = num_bytes;
1185                         em->block_len = num_bytes;
1186                         em->block_start = disk_bytenr;
1187                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1188                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189                         while (1) {
1190                                 write_lock(&em_tree->lock);
1191                                 ret = add_extent_mapping(em_tree, em);
1192                                 write_unlock(&em_tree->lock);
1193                                 if (ret != -EEXIST) {
1194                                         free_extent_map(em);
1195                                         break;
1196                                 }
1197                                 btrfs_drop_extent_cache(inode, em->start,
1198                                                 em->start + em->len - 1, 0);
1199                         }
1200                         type = BTRFS_ORDERED_PREALLOC;
1201                 } else {
1202                         type = BTRFS_ORDERED_NOCOW;
1203                 }
1204
1205                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1206                                                num_bytes, num_bytes, type);
1207                 BUG_ON(ret);
1208
1209                 if (root->root_key.objectid ==
1210                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1211                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1212                                                       num_bytes);
1213                         BUG_ON(ret);
1214                 }
1215
1216                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1217                                 cur_offset, cur_offset + num_bytes - 1,
1218                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1219                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1220                                 EXTENT_SET_PRIVATE2);
1221                 cur_offset = extent_end;
1222                 if (cur_offset > end)
1223                         break;
1224         }
1225         btrfs_release_path(path);
1226
1227         if (cur_offset <= end && cow_start == (u64)-1)
1228                 cow_start = cur_offset;
1229         if (cow_start != (u64)-1) {
1230                 ret = cow_file_range(inode, locked_page, cow_start, end,
1231                                      page_started, nr_written, 1);
1232                 BUG_ON(ret);
1233         }
1234
1235         if (nolock) {
1236                 ret = btrfs_end_transaction_nolock(trans, root);
1237                 BUG_ON(ret);
1238         } else {
1239                 ret = btrfs_end_transaction(trans, root);
1240                 BUG_ON(ret);
1241         }
1242         btrfs_free_path(path);
1243         return 0;
1244 }
1245
1246 /*
1247  * extent_io.c call back to do delayed allocation processing
1248  */
1249 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1250                               u64 start, u64 end, int *page_started,
1251                               unsigned long *nr_written)
1252 {
1253         int ret;
1254         struct btrfs_root *root = BTRFS_I(inode)->root;
1255
1256         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1257                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258                                          page_started, 1, nr_written);
1259         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1260                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1261                                          page_started, 0, nr_written);
1262         else if (!btrfs_test_opt(root, COMPRESS) &&
1263                  !(BTRFS_I(inode)->force_compress) &&
1264                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1265                 ret = cow_file_range(inode, locked_page, start, end,
1266                                       page_started, nr_written, 1);
1267         else
1268                 ret = cow_file_range_async(inode, locked_page, start, end,
1269                                            page_started, nr_written);
1270         return ret;
1271 }
1272
1273 static int btrfs_split_extent_hook(struct inode *inode,
1274                                    struct extent_state *orig, u64 split)
1275 {
1276         /* not delalloc, ignore it */
1277         if (!(orig->state & EXTENT_DELALLOC))
1278                 return 0;
1279
1280         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1281         return 0;
1282 }
1283
1284 /*
1285  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1286  * extents so we can keep track of new extents that are just merged onto old
1287  * extents, such as when we are doing sequential writes, so we can properly
1288  * account for the metadata space we'll need.
1289  */
1290 static int btrfs_merge_extent_hook(struct inode *inode,
1291                                    struct extent_state *new,
1292                                    struct extent_state *other)
1293 {
1294         /* not delalloc, ignore it */
1295         if (!(other->state & EXTENT_DELALLOC))
1296                 return 0;
1297
1298         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1299         return 0;
1300 }
1301
1302 /*
1303  * extent_io.c set_bit_hook, used to track delayed allocation
1304  * bytes in this file, and to maintain the list of inodes that
1305  * have pending delalloc work to be done.
1306  */
1307 static int btrfs_set_bit_hook(struct inode *inode,
1308                               struct extent_state *state, int *bits)
1309 {
1310
1311         /*
1312          * set_bit and clear bit hooks normally require _irqsave/restore
1313          * but in this case, we are only testeing for the DELALLOC
1314          * bit, which is only set or cleared with irqs on
1315          */
1316         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1317                 struct btrfs_root *root = BTRFS_I(inode)->root;
1318                 u64 len = state->end + 1 - state->start;
1319                 int do_list = (root->root_key.objectid !=
1320                                BTRFS_ROOT_TREE_OBJECTID);
1321
1322                 if (*bits & EXTENT_FIRST_DELALLOC)
1323                         *bits &= ~EXTENT_FIRST_DELALLOC;
1324                 else
1325                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326
1327                 spin_lock(&root->fs_info->delalloc_lock);
1328                 BTRFS_I(inode)->delalloc_bytes += len;
1329                 root->fs_info->delalloc_bytes += len;
1330                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332                                       &root->fs_info->delalloc_inodes);
1333                 }
1334                 spin_unlock(&root->fs_info->delalloc_lock);
1335         }
1336         return 0;
1337 }
1338
1339 /*
1340  * extent_io.c clear_bit_hook, see set_bit_hook for why
1341  */
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343                                 struct extent_state *state, int *bits)
1344 {
1345         /*
1346          * set_bit and clear bit hooks normally require _irqsave/restore
1347          * but in this case, we are only testeing for the DELALLOC
1348          * bit, which is only set or cleared with irqs on
1349          */
1350         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1351                 struct btrfs_root *root = BTRFS_I(inode)->root;
1352                 u64 len = state->end + 1 - state->start;
1353                 int do_list = (root->root_key.objectid !=
1354                                BTRFS_ROOT_TREE_OBJECTID);
1355
1356                 if (*bits & EXTENT_FIRST_DELALLOC)
1357                         *bits &= ~EXTENT_FIRST_DELALLOC;
1358                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1359                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360
1361                 if (*bits & EXTENT_DO_ACCOUNTING)
1362                         btrfs_delalloc_release_metadata(inode, len);
1363
1364                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365                     && do_list)
1366                         btrfs_free_reserved_data_space(inode, len);
1367
1368                 spin_lock(&root->fs_info->delalloc_lock);
1369                 root->fs_info->delalloc_bytes -= len;
1370                 BTRFS_I(inode)->delalloc_bytes -= len;
1371
1372                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1373                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1374                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375                 }
1376                 spin_unlock(&root->fs_info->delalloc_lock);
1377         }
1378         return 0;
1379 }
1380
1381 /*
1382  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1383  * we don't create bios that span stripes or chunks
1384  */
1385 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1386                          size_t size, struct bio *bio,
1387                          unsigned long bio_flags)
1388 {
1389         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1390         struct btrfs_mapping_tree *map_tree;
1391         u64 logical = (u64)bio->bi_sector << 9;
1392         u64 length = 0;
1393         u64 map_length;
1394         int ret;
1395
1396         if (bio_flags & EXTENT_BIO_COMPRESSED)
1397                 return 0;
1398
1399         length = bio->bi_size;
1400         map_tree = &root->fs_info->mapping_tree;
1401         map_length = length;
1402         ret = btrfs_map_block(map_tree, READ, logical,
1403                               &map_length, NULL, 0);
1404
1405         if (map_length < length + size)
1406                 return 1;
1407         return ret;
1408 }
1409
1410 /*
1411  * in order to insert checksums into the metadata in large chunks,
1412  * we wait until bio submission time.   All the pages in the bio are
1413  * checksummed and sums are attached onto the ordered extent record.
1414  *
1415  * At IO completion time the cums attached on the ordered extent record
1416  * are inserted into the btree
1417  */
1418 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1419                                     struct bio *bio, int mirror_num,
1420                                     unsigned long bio_flags,
1421                                     u64 bio_offset)
1422 {
1423         struct btrfs_root *root = BTRFS_I(inode)->root;
1424         int ret = 0;
1425
1426         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1427         BUG_ON(ret);
1428         return 0;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1440                           int mirror_num, unsigned long bio_flags,
1441                           u64 bio_offset)
1442 {
1443         struct btrfs_root *root = BTRFS_I(inode)->root;
1444         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1445 }
1446
1447 /*
1448  * extent_io.c submission hook. This does the right thing for csum calculation
1449  * on write, or reading the csums from the tree before a read
1450  */
1451 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1452                           int mirror_num, unsigned long bio_flags,
1453                           u64 bio_offset)
1454 {
1455         struct btrfs_root *root = BTRFS_I(inode)->root;
1456         int ret = 0;
1457         int skip_sum;
1458
1459         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460
1461         if (root == root->fs_info->tree_root)
1462                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463         else
1464                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465         BUG_ON(ret);
1466
1467         if (!(rw & REQ_WRITE)) {
1468                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1469                         return btrfs_submit_compressed_read(inode, bio,
1470                                                     mirror_num, bio_flags);
1471                 } else if (!skip_sum) {
1472                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1473                         if (ret)
1474                                 return ret;
1475                 }
1476                 goto mapit;
1477         } else if (!skip_sum) {
1478                 /* csum items have already been cloned */
1479                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1480                         goto mapit;
1481                 /* we're doing a write, do the async checksumming */
1482                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483                                    inode, rw, bio, mirror_num,
1484                                    bio_flags, bio_offset,
1485                                    __btrfs_submit_bio_start,
1486                                    __btrfs_submit_bio_done);
1487         }
1488
1489 mapit:
1490         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1491 }
1492
1493 /*
1494  * given a list of ordered sums record them in the inode.  This happens
1495  * at IO completion time based on sums calculated at bio submission time.
1496  */
1497 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1498                              struct inode *inode, u64 file_offset,
1499                              struct list_head *list)
1500 {
1501         struct btrfs_ordered_sum *sum;
1502
1503         btrfs_set_trans_block_group(trans, inode);
1504
1505         list_for_each_entry(sum, list, list) {
1506                 btrfs_csum_file_blocks(trans,
1507                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1508         }
1509         return 0;
1510 }
1511
1512 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1513                               struct extent_state **cached_state)
1514 {
1515         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1516                 WARN_ON(1);
1517         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1518                                    cached_state, GFP_NOFS);
1519 }
1520
1521 /* see btrfs_writepage_start_hook for details on why this is required */
1522 struct btrfs_writepage_fixup {
1523         struct page *page;
1524         struct btrfs_work work;
1525 };
1526
1527 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1528 {
1529         struct btrfs_writepage_fixup *fixup;
1530         struct btrfs_ordered_extent *ordered;
1531         struct extent_state *cached_state = NULL;
1532         struct page *page;
1533         struct inode *inode;
1534         u64 page_start;
1535         u64 page_end;
1536
1537         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1538         page = fixup->page;
1539 again:
1540         lock_page(page);
1541         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1542                 ClearPageChecked(page);
1543                 goto out_page;
1544         }
1545
1546         inode = page->mapping->host;
1547         page_start = page_offset(page);
1548         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1549
1550         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1551                          &cached_state, GFP_NOFS);
1552
1553         /* already ordered? We're done */
1554         if (PagePrivate2(page))
1555                 goto out;
1556
1557         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1558         if (ordered) {
1559                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1560                                      page_end, &cached_state, GFP_NOFS);
1561                 unlock_page(page);
1562                 btrfs_start_ordered_extent(inode, ordered, 1);
1563                 goto again;
1564         }
1565
1566         BUG();
1567         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1568         ClearPageChecked(page);
1569 out:
1570         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1571                              &cached_state, GFP_NOFS);
1572 out_page:
1573         unlock_page(page);
1574         page_cache_release(page);
1575         kfree(fixup);
1576 }
1577
1578 /*
1579  * There are a few paths in the higher layers of the kernel that directly
1580  * set the page dirty bit without asking the filesystem if it is a
1581  * good idea.  This causes problems because we want to make sure COW
1582  * properly happens and the data=ordered rules are followed.
1583  *
1584  * In our case any range that doesn't have the ORDERED bit set
1585  * hasn't been properly setup for IO.  We kick off an async process
1586  * to fix it up.  The async helper will wait for ordered extents, set
1587  * the delalloc bit and make it safe to write the page.
1588  */
1589 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1590 {
1591         struct inode *inode = page->mapping->host;
1592         struct btrfs_writepage_fixup *fixup;
1593         struct btrfs_root *root = BTRFS_I(inode)->root;
1594
1595         /* this page is properly in the ordered list */
1596         if (TestClearPagePrivate2(page))
1597                 return 0;
1598
1599         if (PageChecked(page))
1600                 return -EAGAIN;
1601
1602         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1603         if (!fixup)
1604                 return -EAGAIN;
1605
1606         SetPageChecked(page);
1607         page_cache_get(page);
1608         fixup->work.func = btrfs_writepage_fixup_worker;
1609         fixup->page = page;
1610         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1611         return -EAGAIN;
1612 }
1613
1614 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1615                                        struct inode *inode, u64 file_pos,
1616                                        u64 disk_bytenr, u64 disk_num_bytes,
1617                                        u64 num_bytes, u64 ram_bytes,
1618                                        u8 compression, u8 encryption,
1619                                        u16 other_encoding, int extent_type)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         struct btrfs_file_extent_item *fi;
1623         struct btrfs_path *path;
1624         struct extent_buffer *leaf;
1625         struct btrfs_key ins;
1626         u64 hint;
1627         int ret;
1628
1629         path = btrfs_alloc_path();
1630         BUG_ON(!path);
1631
1632         path->leave_spinning = 1;
1633
1634         /*
1635          * we may be replacing one extent in the tree with another.
1636          * The new extent is pinned in the extent map, and we don't want
1637          * to drop it from the cache until it is completely in the btree.
1638          *
1639          * So, tell btrfs_drop_extents to leave this extent in the cache.
1640          * the caller is expected to unpin it and allow it to be merged
1641          * with the others.
1642          */
1643         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1644                                  &hint, 0);
1645         BUG_ON(ret);
1646
1647         ins.objectid = inode->i_ino;
1648         ins.offset = file_pos;
1649         ins.type = BTRFS_EXTENT_DATA_KEY;
1650         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1651         BUG_ON(ret);
1652         leaf = path->nodes[0];
1653         fi = btrfs_item_ptr(leaf, path->slots[0],
1654                             struct btrfs_file_extent_item);
1655         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1656         btrfs_set_file_extent_type(leaf, fi, extent_type);
1657         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1658         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1659         btrfs_set_file_extent_offset(leaf, fi, 0);
1660         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1661         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1662         btrfs_set_file_extent_compression(leaf, fi, compression);
1663         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1664         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1665
1666         btrfs_unlock_up_safe(path, 1);
1667         btrfs_set_lock_blocking(leaf);
1668
1669         btrfs_mark_buffer_dirty(leaf);
1670
1671         inode_add_bytes(inode, num_bytes);
1672
1673         ins.objectid = disk_bytenr;
1674         ins.offset = disk_num_bytes;
1675         ins.type = BTRFS_EXTENT_ITEM_KEY;
1676         ret = btrfs_alloc_reserved_file_extent(trans, root,
1677                                         root->root_key.objectid,
1678                                         inode->i_ino, file_pos, &ins);
1679         BUG_ON(ret);
1680         btrfs_free_path(path);
1681
1682         return 0;
1683 }
1684
1685 /*
1686  * helper function for btrfs_finish_ordered_io, this
1687  * just reads in some of the csum leaves to prime them into ram
1688  * before we start the transaction.  It limits the amount of btree
1689  * reads required while inside the transaction.
1690  */
1691 /* as ordered data IO finishes, this gets called so we can finish
1692  * an ordered extent if the range of bytes in the file it covers are
1693  * fully written.
1694  */
1695 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         struct btrfs_trans_handle *trans = NULL;
1699         struct btrfs_ordered_extent *ordered_extent = NULL;
1700         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1701         struct extent_state *cached_state = NULL;
1702         int compress_type = 0;
1703         int ret;
1704         bool nolock = false;
1705
1706         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1707                                              end - start + 1);
1708         if (!ret)
1709                 return 0;
1710         BUG_ON(!ordered_extent);
1711
1712         nolock = (root == root->fs_info->tree_root);
1713
1714         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1715                 BUG_ON(!list_empty(&ordered_extent->list));
1716                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717                 if (!ret) {
1718                         if (nolock)
1719                                 trans = btrfs_join_transaction_nolock(root, 1);
1720                         else
1721                                 trans = btrfs_join_transaction(root, 1);
1722                         BUG_ON(IS_ERR(trans));
1723                         btrfs_set_trans_block_group(trans, inode);
1724                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1725                         ret = btrfs_update_inode(trans, root, inode);
1726                         BUG_ON(ret);
1727                 }
1728                 goto out;
1729         }
1730
1731         lock_extent_bits(io_tree, ordered_extent->file_offset,
1732                          ordered_extent->file_offset + ordered_extent->len - 1,
1733                          0, &cached_state, GFP_NOFS);
1734
1735         if (nolock)
1736                 trans = btrfs_join_transaction_nolock(root, 1);
1737         else
1738                 trans = btrfs_join_transaction(root, 1);
1739         BUG_ON(IS_ERR(trans));
1740         btrfs_set_trans_block_group(trans, inode);
1741         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1742
1743         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1744                 compress_type = ordered_extent->compress_type;
1745         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1746                 BUG_ON(compress_type);
1747                 ret = btrfs_mark_extent_written(trans, inode,
1748                                                 ordered_extent->file_offset,
1749                                                 ordered_extent->file_offset +
1750                                                 ordered_extent->len);
1751                 BUG_ON(ret);
1752         } else {
1753                 BUG_ON(root == root->fs_info->tree_root);
1754                 ret = insert_reserved_file_extent(trans, inode,
1755                                                 ordered_extent->file_offset,
1756                                                 ordered_extent->start,
1757                                                 ordered_extent->disk_len,
1758                                                 ordered_extent->len,
1759                                                 ordered_extent->len,
1760                                                 compress_type, 0, 0,
1761                                                 BTRFS_FILE_EXTENT_REG);
1762                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1763                                    ordered_extent->file_offset,
1764                                    ordered_extent->len);
1765                 BUG_ON(ret);
1766         }
1767         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1768                              ordered_extent->file_offset +
1769                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1770
1771         add_pending_csums(trans, inode, ordered_extent->file_offset,
1772                           &ordered_extent->list);
1773
1774         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1775         if (!ret) {
1776                 ret = btrfs_update_inode(trans, root, inode);
1777                 BUG_ON(ret);
1778         }
1779         ret = 0;
1780 out:
1781         if (nolock) {
1782                 if (trans)
1783                         btrfs_end_transaction_nolock(trans, root);
1784         } else {
1785                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1786                 if (trans)
1787                         btrfs_end_transaction(trans, root);
1788         }
1789
1790         /* once for us */
1791         btrfs_put_ordered_extent(ordered_extent);
1792         /* once for the tree */
1793         btrfs_put_ordered_extent(ordered_extent);
1794
1795         return 0;
1796 }
1797
1798 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1799                                 struct extent_state *state, int uptodate)
1800 {
1801         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1802
1803         ClearPagePrivate2(page);
1804         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1805 }
1806
1807 /*
1808  * When IO fails, either with EIO or csum verification fails, we
1809  * try other mirrors that might have a good copy of the data.  This
1810  * io_failure_record is used to record state as we go through all the
1811  * mirrors.  If another mirror has good data, the page is set up to date
1812  * and things continue.  If a good mirror can't be found, the original
1813  * bio end_io callback is called to indicate things have failed.
1814  */
1815 struct io_failure_record {
1816         struct page *page;
1817         u64 start;
1818         u64 len;
1819         u64 logical;
1820         unsigned long bio_flags;
1821         int last_mirror;
1822 };
1823
1824 static int btrfs_io_failed_hook(struct bio *failed_bio,
1825                          struct page *page, u64 start, u64 end,
1826                          struct extent_state *state)
1827 {
1828         struct io_failure_record *failrec = NULL;
1829         u64 private;
1830         struct extent_map *em;
1831         struct inode *inode = page->mapping->host;
1832         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1833         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1834         struct bio *bio;
1835         int num_copies;
1836         int ret;
1837         int rw;
1838         u64 logical;
1839
1840         ret = get_state_private(failure_tree, start, &private);
1841         if (ret) {
1842                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1843                 if (!failrec)
1844                         return -ENOMEM;
1845                 failrec->start = start;
1846                 failrec->len = end - start + 1;
1847                 failrec->last_mirror = 0;
1848                 failrec->bio_flags = 0;
1849
1850                 read_lock(&em_tree->lock);
1851                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1852                 if (em->start > start || em->start + em->len < start) {
1853                         free_extent_map(em);
1854                         em = NULL;
1855                 }
1856                 read_unlock(&em_tree->lock);
1857
1858                 if (IS_ERR_OR_NULL(em)) {
1859                         kfree(failrec);
1860                         return -EIO;
1861                 }
1862                 logical = start - em->start;
1863                 logical = em->block_start + logical;
1864                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1865                         logical = em->block_start;
1866                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1867                         extent_set_compress_type(&failrec->bio_flags,
1868                                                  em->compress_type);
1869                 }
1870                 failrec->logical = logical;
1871                 free_extent_map(em);
1872                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1873                                 EXTENT_DIRTY, GFP_NOFS);
1874                 set_state_private(failure_tree, start,
1875                                  (u64)(unsigned long)failrec);
1876         } else {
1877                 failrec = (struct io_failure_record *)(unsigned long)private;
1878         }
1879         num_copies = btrfs_num_copies(
1880                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1881                               failrec->logical, failrec->len);
1882         failrec->last_mirror++;
1883         if (!state) {
1884                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1885                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1886                                                     failrec->start,
1887                                                     EXTENT_LOCKED);
1888                 if (state && state->start != failrec->start)
1889                         state = NULL;
1890                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1891         }
1892         if (!state || failrec->last_mirror > num_copies) {
1893                 set_state_private(failure_tree, failrec->start, 0);
1894                 clear_extent_bits(failure_tree, failrec->start,
1895                                   failrec->start + failrec->len - 1,
1896                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1897                 kfree(failrec);
1898                 return -EIO;
1899         }
1900         bio = bio_alloc(GFP_NOFS, 1);
1901         bio->bi_private = state;
1902         bio->bi_end_io = failed_bio->bi_end_io;
1903         bio->bi_sector = failrec->logical >> 9;
1904         bio->bi_bdev = failed_bio->bi_bdev;
1905         bio->bi_size = 0;
1906
1907         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1908         if (failed_bio->bi_rw & REQ_WRITE)
1909                 rw = WRITE;
1910         else
1911                 rw = READ;
1912
1913         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1914                                                       failrec->last_mirror,
1915                                                       failrec->bio_flags, 0);
1916         return ret;
1917 }
1918
1919 /*
1920  * each time an IO finishes, we do a fast check in the IO failure tree
1921  * to see if we need to process or clean up an io_failure_record
1922  */
1923 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1924 {
1925         u64 private;
1926         u64 private_failure;
1927         struct io_failure_record *failure;
1928         int ret;
1929
1930         private = 0;
1931         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1932                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1933                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1934                                         start, &private_failure);
1935                 if (ret == 0) {
1936                         failure = (struct io_failure_record *)(unsigned long)
1937                                    private_failure;
1938                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1939                                           failure->start, 0);
1940                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1941                                           failure->start,
1942                                           failure->start + failure->len - 1,
1943                                           EXTENT_DIRTY | EXTENT_LOCKED,
1944                                           GFP_NOFS);
1945                         kfree(failure);
1946                 }
1947         }
1948         return 0;
1949 }
1950
1951 /*
1952  * when reads are done, we need to check csums to verify the data is correct
1953  * if there's a match, we allow the bio to finish.  If not, we go through
1954  * the io_failure_record routines to find good copies
1955  */
1956 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1957                                struct extent_state *state)
1958 {
1959         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1960         struct inode *inode = page->mapping->host;
1961         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1962         char *kaddr;
1963         u64 private = ~(u32)0;
1964         int ret;
1965         struct btrfs_root *root = BTRFS_I(inode)->root;
1966         u32 csum = ~(u32)0;
1967
1968         if (PageChecked(page)) {
1969                 ClearPageChecked(page);
1970                 goto good;
1971         }
1972
1973         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1974                 return 0;
1975
1976         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1977             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1978                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1979                                   GFP_NOFS);
1980                 return 0;
1981         }
1982
1983         if (state && state->start == start) {
1984                 private = state->private;
1985                 ret = 0;
1986         } else {
1987                 ret = get_state_private(io_tree, start, &private);
1988         }
1989         kaddr = kmap_atomic(page, KM_USER0);
1990         if (ret)
1991                 goto zeroit;
1992
1993         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1994         btrfs_csum_final(csum, (char *)&csum);
1995         if (csum != private)
1996                 goto zeroit;
1997
1998         kunmap_atomic(kaddr, KM_USER0);
1999 good:
2000         /* if the io failure tree for this inode is non-empty,
2001          * check to see if we've recovered from a failed IO
2002          */
2003         btrfs_clean_io_failures(inode, start);
2004         return 0;
2005
2006 zeroit:
2007         if (printk_ratelimit()) {
2008                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2009                        "private %llu\n", page->mapping->host->i_ino,
2010                        (unsigned long long)start, csum,
2011                        (unsigned long long)private);
2012         }
2013         memset(kaddr + offset, 1, end - start + 1);
2014         flush_dcache_page(page);
2015         kunmap_atomic(kaddr, KM_USER0);
2016         if (private == 0)
2017                 return 0;
2018         return -EIO;
2019 }
2020
2021 struct delayed_iput {
2022         struct list_head list;
2023         struct inode *inode;
2024 };
2025
2026 void btrfs_add_delayed_iput(struct inode *inode)
2027 {
2028         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2029         struct delayed_iput *delayed;
2030
2031         if (atomic_add_unless(&inode->i_count, -1, 1))
2032                 return;
2033
2034         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2035         delayed->inode = inode;
2036
2037         spin_lock(&fs_info->delayed_iput_lock);
2038         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2039         spin_unlock(&fs_info->delayed_iput_lock);
2040 }
2041
2042 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2043 {
2044         LIST_HEAD(list);
2045         struct btrfs_fs_info *fs_info = root->fs_info;
2046         struct delayed_iput *delayed;
2047         int empty;
2048
2049         spin_lock(&fs_info->delayed_iput_lock);
2050         empty = list_empty(&fs_info->delayed_iputs);
2051         spin_unlock(&fs_info->delayed_iput_lock);
2052         if (empty)
2053                 return;
2054
2055         down_read(&root->fs_info->cleanup_work_sem);
2056         spin_lock(&fs_info->delayed_iput_lock);
2057         list_splice_init(&fs_info->delayed_iputs, &list);
2058         spin_unlock(&fs_info->delayed_iput_lock);
2059
2060         while (!list_empty(&list)) {
2061                 delayed = list_entry(list.next, struct delayed_iput, list);
2062                 list_del(&delayed->list);
2063                 iput(delayed->inode);
2064                 kfree(delayed);
2065         }
2066         up_read(&root->fs_info->cleanup_work_sem);
2067 }
2068
2069 /*
2070  * calculate extra metadata reservation when snapshotting a subvolume
2071  * contains orphan files.
2072  */
2073 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2074                                 struct btrfs_pending_snapshot *pending,
2075                                 u64 *bytes_to_reserve)
2076 {
2077         struct btrfs_root *root;
2078         struct btrfs_block_rsv *block_rsv;
2079         u64 num_bytes;
2080         int index;
2081
2082         root = pending->root;
2083         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2084                 return;
2085
2086         block_rsv = root->orphan_block_rsv;
2087
2088         /* orphan block reservation for the snapshot */
2089         num_bytes = block_rsv->size;
2090
2091         /*
2092          * after the snapshot is created, COWing tree blocks may use more
2093          * space than it frees. So we should make sure there is enough
2094          * reserved space.
2095          */
2096         index = trans->transid & 0x1;
2097         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2098                 num_bytes += block_rsv->size -
2099                              (block_rsv->reserved + block_rsv->freed[index]);
2100         }
2101
2102         *bytes_to_reserve += num_bytes;
2103 }
2104
2105 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2106                                 struct btrfs_pending_snapshot *pending)
2107 {
2108         struct btrfs_root *root = pending->root;
2109         struct btrfs_root *snap = pending->snap;
2110         struct btrfs_block_rsv *block_rsv;
2111         u64 num_bytes;
2112         int index;
2113         int ret;
2114
2115         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2116                 return;
2117
2118         /* refill source subvolume's orphan block reservation */
2119         block_rsv = root->orphan_block_rsv;
2120         index = trans->transid & 0x1;
2121         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2122                 num_bytes = block_rsv->size -
2123                             (block_rsv->reserved + block_rsv->freed[index]);
2124                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2125                                               root->orphan_block_rsv,
2126                                               num_bytes);
2127                 BUG_ON(ret);
2128         }
2129
2130         /* setup orphan block reservation for the snapshot */
2131         block_rsv = btrfs_alloc_block_rsv(snap);
2132         BUG_ON(!block_rsv);
2133
2134         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2135         snap->orphan_block_rsv = block_rsv;
2136
2137         num_bytes = root->orphan_block_rsv->size;
2138         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139                                       block_rsv, num_bytes);
2140         BUG_ON(ret);
2141
2142 #if 0
2143         /* insert orphan item for the snapshot */
2144         WARN_ON(!root->orphan_item_inserted);
2145         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2146                                        snap->root_key.objectid);
2147         BUG_ON(ret);
2148         snap->orphan_item_inserted = 1;
2149 #endif
2150 }
2151
2152 enum btrfs_orphan_cleanup_state {
2153         ORPHAN_CLEANUP_STARTED  = 1,
2154         ORPHAN_CLEANUP_DONE     = 2,
2155 };
2156
2157 /*
2158  * This is called in transaction commmit time. If there are no orphan
2159  * files in the subvolume, it removes orphan item and frees block_rsv
2160  * structure.
2161  */
2162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2163                               struct btrfs_root *root)
2164 {
2165         int ret;
2166
2167         if (!list_empty(&root->orphan_list) ||
2168             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2169                 return;
2170
2171         if (root->orphan_item_inserted &&
2172             btrfs_root_refs(&root->root_item) > 0) {
2173                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2174                                             root->root_key.objectid);
2175                 BUG_ON(ret);
2176                 root->orphan_item_inserted = 0;
2177         }
2178
2179         if (root->orphan_block_rsv) {
2180                 WARN_ON(root->orphan_block_rsv->size > 0);
2181                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2182                 root->orphan_block_rsv = NULL;
2183         }
2184 }
2185
2186 /*
2187  * This creates an orphan entry for the given inode in case something goes
2188  * wrong in the middle of an unlink/truncate.
2189  *
2190  * NOTE: caller of this function should reserve 5 units of metadata for
2191  *       this function.
2192  */
2193 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2194 {
2195         struct btrfs_root *root = BTRFS_I(inode)->root;
2196         struct btrfs_block_rsv *block_rsv = NULL;
2197         int reserve = 0;
2198         int insert = 0;
2199         int ret;
2200
2201         if (!root->orphan_block_rsv) {
2202                 block_rsv = btrfs_alloc_block_rsv(root);
2203                 BUG_ON(!block_rsv);
2204         }
2205
2206         spin_lock(&root->orphan_lock);
2207         if (!root->orphan_block_rsv) {
2208                 root->orphan_block_rsv = block_rsv;
2209         } else if (block_rsv) {
2210                 btrfs_free_block_rsv(root, block_rsv);
2211                 block_rsv = NULL;
2212         }
2213
2214         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2215                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2216 #if 0
2217                 /*
2218                  * For proper ENOSPC handling, we should do orphan
2219                  * cleanup when mounting. But this introduces backward
2220                  * compatibility issue.
2221                  */
2222                 if (!xchg(&root->orphan_item_inserted, 1))
2223                         insert = 2;
2224                 else
2225                         insert = 1;
2226 #endif
2227                 insert = 1;
2228         }
2229
2230         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2231                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2232                 reserve = 1;
2233         }
2234         spin_unlock(&root->orphan_lock);
2235
2236         if (block_rsv)
2237                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2238
2239         /* grab metadata reservation from transaction handle */
2240         if (reserve) {
2241                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2242                 BUG_ON(ret);
2243         }
2244
2245         /* insert an orphan item to track this unlinked/truncated file */
2246         if (insert >= 1) {
2247                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2248                 BUG_ON(ret);
2249         }
2250
2251         /* insert an orphan item to track subvolume contains orphan files */
2252         if (insert >= 2) {
2253                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2254                                                root->root_key.objectid);
2255                 BUG_ON(ret);
2256         }
2257         return 0;
2258 }
2259
2260 /*
2261  * We have done the truncate/delete so we can go ahead and remove the orphan
2262  * item for this particular inode.
2263  */
2264 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2265 {
2266         struct btrfs_root *root = BTRFS_I(inode)->root;
2267         int delete_item = 0;
2268         int release_rsv = 0;
2269         int ret = 0;
2270
2271         spin_lock(&root->orphan_lock);
2272         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2273                 list_del_init(&BTRFS_I(inode)->i_orphan);
2274                 delete_item = 1;
2275         }
2276
2277         if (BTRFS_I(inode)->orphan_meta_reserved) {
2278                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2279                 release_rsv = 1;
2280         }
2281         spin_unlock(&root->orphan_lock);
2282
2283         if (trans && delete_item) {
2284                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2285                 BUG_ON(ret);
2286         }
2287
2288         if (release_rsv)
2289                 btrfs_orphan_release_metadata(inode);
2290
2291         return 0;
2292 }
2293
2294 /*
2295  * this cleans up any orphans that may be left on the list from the last use
2296  * of this root.
2297  */
2298 int btrfs_orphan_cleanup(struct btrfs_root *root)
2299 {
2300         struct btrfs_path *path;
2301         struct extent_buffer *leaf;
2302         struct btrfs_key key, found_key;
2303         struct btrfs_trans_handle *trans;
2304         struct inode *inode;
2305         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2306
2307         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2308                 return 0;
2309
2310         path = btrfs_alloc_path();
2311         if (!path) {
2312                 ret = -ENOMEM;
2313                 goto out;
2314         }
2315         path->reada = -1;
2316
2317         key.objectid = BTRFS_ORPHAN_OBJECTID;
2318         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2319         key.offset = (u64)-1;
2320
2321         while (1) {
2322                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2323                 if (ret < 0)
2324                         goto out;
2325
2326                 /*
2327                  * if ret == 0 means we found what we were searching for, which
2328                  * is weird, but possible, so only screw with path if we didn't
2329                  * find the key and see if we have stuff that matches
2330                  */
2331                 if (ret > 0) {
2332                         ret = 0;
2333                         if (path->slots[0] == 0)
2334                                 break;
2335                         path->slots[0]--;
2336                 }
2337
2338                 /* pull out the item */
2339                 leaf = path->nodes[0];
2340                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2341
2342                 /* make sure the item matches what we want */
2343                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2344                         break;
2345                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2346                         break;
2347
2348                 /* release the path since we're done with it */
2349                 btrfs_release_path(path);
2350
2351                 /*
2352                  * this is where we are basically btrfs_lookup, without the
2353                  * crossing root thing.  we store the inode number in the
2354                  * offset of the orphan item.
2355                  */
2356                 found_key.objectid = found_key.offset;
2357                 found_key.type = BTRFS_INODE_ITEM_KEY;
2358                 found_key.offset = 0;
2359                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2360                 if (IS_ERR(inode)) {
2361                         ret = PTR_ERR(inode);
2362                         goto out;
2363                 }
2364
2365                 /*
2366                  * add this inode to the orphan list so btrfs_orphan_del does
2367                  * the proper thing when we hit it
2368                  */
2369                 spin_lock(&root->orphan_lock);
2370                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2371                 spin_unlock(&root->orphan_lock);
2372
2373                 /*
2374                  * if this is a bad inode, means we actually succeeded in
2375                  * removing the inode, but not the orphan record, which means
2376                  * we need to manually delete the orphan since iput will just
2377                  * do a destroy_inode
2378                  */
2379                 if (is_bad_inode(inode)) {
2380                         trans = btrfs_start_transaction(root, 0);
2381                         if (IS_ERR(trans)) {
2382                                 ret = PTR_ERR(trans);
2383                                 goto out;
2384                         }
2385                         btrfs_orphan_del(trans, inode);
2386                         btrfs_end_transaction(trans, root);
2387                         iput(inode);
2388                         continue;
2389                 }
2390
2391                 /* if we have links, this was a truncate, lets do that */
2392                 if (inode->i_nlink) {
2393                         if (!S_ISREG(inode->i_mode)) {
2394                                 WARN_ON(1);
2395                                 iput(inode);
2396                                 continue;
2397                         }
2398                         nr_truncate++;
2399                         ret = btrfs_truncate(inode);
2400                 } else {
2401                         nr_unlink++;
2402                 }
2403
2404                 /* this will do delete_inode and everything for us */
2405                 iput(inode);
2406                 if (ret)
2407                         goto out;
2408         }
2409         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2410
2411         if (root->orphan_block_rsv)
2412                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2413                                         (u64)-1);
2414
2415         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2416                 trans = btrfs_join_transaction(root, 1);
2417                 if (!IS_ERR(trans))
2418                         btrfs_end_transaction(trans, root);
2419         }
2420
2421         if (nr_unlink)
2422                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2423         if (nr_truncate)
2424                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2425
2426 out:
2427         if (ret)
2428                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2429         btrfs_free_path(path);
2430         return ret;
2431 }
2432
2433 /*
2434  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2435  * don't find any xattrs, we know there can't be any acls.
2436  *
2437  * slot is the slot the inode is in, objectid is the objectid of the inode
2438  */
2439 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2440                                           int slot, u64 objectid)
2441 {
2442         u32 nritems = btrfs_header_nritems(leaf);
2443         struct btrfs_key found_key;
2444         int scanned = 0;
2445
2446         slot++;
2447         while (slot < nritems) {
2448                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2449
2450                 /* we found a different objectid, there must not be acls */
2451                 if (found_key.objectid != objectid)
2452                         return 0;
2453
2454                 /* we found an xattr, assume we've got an acl */
2455                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2456                         return 1;
2457
2458                 /*
2459                  * we found a key greater than an xattr key, there can't
2460                  * be any acls later on
2461                  */
2462                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2463                         return 0;
2464
2465                 slot++;
2466                 scanned++;
2467
2468                 /*
2469                  * it goes inode, inode backrefs, xattrs, extents,
2470                  * so if there are a ton of hard links to an inode there can
2471                  * be a lot of backrefs.  Don't waste time searching too hard,
2472                  * this is just an optimization
2473                  */
2474                 if (scanned >= 8)
2475                         break;
2476         }
2477         /* we hit the end of the leaf before we found an xattr or
2478          * something larger than an xattr.  We have to assume the inode
2479          * has acls
2480          */
2481         return 1;
2482 }
2483
2484 /*
2485  * read an inode from the btree into the in-memory inode
2486  */
2487 static void btrfs_read_locked_inode(struct inode *inode)
2488 {
2489         struct btrfs_path *path;
2490         struct extent_buffer *leaf;
2491         struct btrfs_inode_item *inode_item;
2492         struct btrfs_timespec *tspec;
2493         struct btrfs_root *root = BTRFS_I(inode)->root;
2494         struct btrfs_key location;
2495         int maybe_acls;
2496         u64 alloc_group_block;
2497         u32 rdev;
2498         int ret;
2499
2500         path = btrfs_alloc_path();
2501         BUG_ON(!path);
2502         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2503
2504         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2505         if (ret)
2506                 goto make_bad;
2507
2508         leaf = path->nodes[0];
2509         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2510                                     struct btrfs_inode_item);
2511
2512         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2513         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2514         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2515         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2516         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2517
2518         tspec = btrfs_inode_atime(inode_item);
2519         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2520         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2521
2522         tspec = btrfs_inode_mtime(inode_item);
2523         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2524         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2525
2526         tspec = btrfs_inode_ctime(inode_item);
2527         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2528         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2529
2530         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2531         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2532         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2533         inode->i_generation = BTRFS_I(inode)->generation;
2534         inode->i_rdev = 0;
2535         rdev = btrfs_inode_rdev(leaf, inode_item);
2536
2537         BTRFS_I(inode)->index_cnt = (u64)-1;
2538         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2539
2540         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2541
2542         /*
2543          * try to precache a NULL acl entry for files that don't have
2544          * any xattrs or acls
2545          */
2546         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2547         if (!maybe_acls)
2548                 cache_no_acl(inode);
2549
2550         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2551                                                 alloc_group_block, 0);
2552         btrfs_free_path(path);
2553         inode_item = NULL;
2554
2555         switch (inode->i_mode & S_IFMT) {
2556         case S_IFREG:
2557                 inode->i_mapping->a_ops = &btrfs_aops;
2558                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2559                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2560                 inode->i_fop = &btrfs_file_operations;
2561                 inode->i_op = &btrfs_file_inode_operations;
2562                 break;
2563         case S_IFDIR:
2564                 inode->i_fop = &btrfs_dir_file_operations;
2565                 if (root == root->fs_info->tree_root)
2566                         inode->i_op = &btrfs_dir_ro_inode_operations;
2567                 else
2568                         inode->i_op = &btrfs_dir_inode_operations;
2569                 break;
2570         case S_IFLNK:
2571                 inode->i_op = &btrfs_symlink_inode_operations;
2572                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2573                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2574                 break;
2575         default:
2576                 inode->i_op = &btrfs_special_inode_operations;
2577                 init_special_inode(inode, inode->i_mode, rdev);
2578                 break;
2579         }
2580
2581         btrfs_update_iflags(inode);
2582         return;
2583
2584 make_bad:
2585         btrfs_free_path(path);
2586         make_bad_inode(inode);
2587 }
2588
2589 /*
2590  * given a leaf and an inode, copy the inode fields into the leaf
2591  */
2592 static void fill_inode_item(struct btrfs_trans_handle *trans,
2593                             struct extent_buffer *leaf,
2594                             struct btrfs_inode_item *item,
2595                             struct inode *inode)
2596 {
2597         if (!leaf->map_token)
2598                 map_private_extent_buffer(leaf, (unsigned long)item,
2599                                           sizeof(struct btrfs_inode_item),
2600                                           &leaf->map_token, &leaf->kaddr,
2601                                           &leaf->map_start, &leaf->map_len,
2602                                           KM_USER1);
2603
2604         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2605         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2606         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2607         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2608         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2609
2610         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2611                                inode->i_atime.tv_sec);
2612         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2613                                 inode->i_atime.tv_nsec);
2614
2615         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2616                                inode->i_mtime.tv_sec);
2617         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2618                                 inode->i_mtime.tv_nsec);
2619
2620         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2621                                inode->i_ctime.tv_sec);
2622         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2623                                 inode->i_ctime.tv_nsec);
2624
2625         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2626         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2627         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2628         btrfs_set_inode_transid(leaf, item, trans->transid);
2629         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2630         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2631         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2632
2633         if (leaf->map_token) {
2634                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2635                 leaf->map_token = NULL;
2636         }
2637 }
2638
2639 /*
2640  * copy everything in the in-memory inode into the btree.
2641  */
2642 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2643                                 struct btrfs_root *root, struct inode *inode)
2644 {
2645         struct btrfs_inode_item *inode_item;
2646         struct btrfs_path *path;
2647         struct extent_buffer *leaf;
2648         int ret;
2649
2650         path = btrfs_alloc_path();
2651         BUG_ON(!path);
2652         path->leave_spinning = 1;
2653         ret = btrfs_lookup_inode(trans, root, path,
2654                                  &BTRFS_I(inode)->location, 1);
2655         if (ret) {
2656                 if (ret > 0)
2657                         ret = -ENOENT;
2658                 goto failed;
2659         }
2660
2661         btrfs_unlock_up_safe(path, 1);
2662         leaf = path->nodes[0];
2663         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2664                                   struct btrfs_inode_item);
2665
2666         fill_inode_item(trans, leaf, inode_item, inode);
2667         btrfs_mark_buffer_dirty(leaf);
2668         btrfs_set_inode_last_trans(trans, inode);
2669         ret = 0;
2670 failed:
2671         btrfs_free_path(path);
2672         return ret;
2673 }
2674
2675
2676 /*
2677  * unlink helper that gets used here in inode.c and in the tree logging
2678  * recovery code.  It remove a link in a directory with a given name, and
2679  * also drops the back refs in the inode to the directory
2680  */
2681 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2682                                 struct btrfs_root *root,
2683                                 struct inode *dir, struct inode *inode,
2684                                 const char *name, int name_len)
2685 {
2686         struct btrfs_path *path;
2687         int ret = 0;
2688         struct extent_buffer *leaf;
2689         struct btrfs_dir_item *di;
2690         struct btrfs_key key;
2691         u64 index;
2692
2693         path = btrfs_alloc_path();
2694         if (!path) {
2695                 ret = -ENOMEM;
2696                 goto out;
2697         }
2698
2699         path->leave_spinning = 1;
2700         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2701                                     name, name_len, -1);
2702         if (IS_ERR(di)) {
2703                 ret = PTR_ERR(di);
2704                 goto err;
2705         }
2706         if (!di) {
2707                 ret = -ENOENT;
2708                 goto err;
2709         }
2710         leaf = path->nodes[0];
2711         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2712         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2713         if (ret)
2714                 goto err;
2715         btrfs_release_path(path);
2716
2717         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2718                                   inode->i_ino,
2719                                   dir->i_ino, &index);
2720         if (ret) {
2721                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2722                        "inode %lu parent %lu\n", name_len, name,
2723                        inode->i_ino, dir->i_ino);
2724                 goto err;
2725         }
2726
2727         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2728                                          index, name, name_len, -1);
2729         if (IS_ERR(di)) {
2730                 ret = PTR_ERR(di);
2731                 goto err;
2732         }
2733         if (!di) {
2734                 ret = -ENOENT;
2735                 goto err;
2736         }
2737         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2738         btrfs_release_path(path);
2739
2740         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2741                                          inode, dir->i_ino);
2742         BUG_ON(ret != 0 && ret != -ENOENT);
2743
2744         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2745                                            dir, index);
2746         if (ret == -ENOENT)
2747                 ret = 0;
2748 err:
2749         btrfs_free_path(path);
2750         if (ret)
2751                 goto out;
2752
2753         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2754         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2755         btrfs_update_inode(trans, root, dir);
2756 out:
2757         return ret;
2758 }
2759
2760 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2761                        struct btrfs_root *root,
2762                        struct inode *dir, struct inode *inode,
2763                        const char *name, int name_len)
2764 {
2765         int ret;
2766         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2767         if (!ret) {
2768                 btrfs_drop_nlink(inode);
2769                 ret = btrfs_update_inode(trans, root, inode);
2770         }
2771         return ret;
2772 }
2773
2774
2775 /* helper to check if there is any shared block in the path */
2776 static int check_path_shared(struct btrfs_root *root,
2777                              struct btrfs_path *path)
2778 {
2779         struct extent_buffer *eb;
2780         int level;
2781         u64 refs = 1;
2782
2783         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2784                 int ret;
2785
2786                 if (!path->nodes[level])
2787                         break;
2788                 eb = path->nodes[level];
2789                 if (!btrfs_block_can_be_shared(root, eb))
2790                         continue;
2791                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2792                                                &refs, NULL);
2793                 if (refs > 1)
2794                         return 1;
2795         }
2796         return 0;
2797 }
2798
2799 /*
2800  * helper to start transaction for unlink and rmdir.
2801  *
2802  * unlink and rmdir are special in btrfs, they do not always free space.
2803  * so in enospc case, we should make sure they will free space before
2804  * allowing them to use the global metadata reservation.
2805  */
2806 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2807                                                        struct dentry *dentry)
2808 {
2809         struct btrfs_trans_handle *trans;
2810         struct btrfs_root *root = BTRFS_I(dir)->root;
2811         struct btrfs_path *path;
2812         struct btrfs_inode_ref *ref;
2813         struct btrfs_dir_item *di;
2814         struct inode *inode = dentry->d_inode;
2815         u64 index;
2816         int check_link = 1;
2817         int err = -ENOSPC;
2818         int ret;
2819
2820         trans = btrfs_start_transaction(root, 10);
2821         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2822                 return trans;
2823
2824         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2825                 return ERR_PTR(-ENOSPC);
2826
2827         /* check if there is someone else holds reference */
2828         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2829                 return ERR_PTR(-ENOSPC);
2830
2831         if (atomic_read(&inode->i_count) > 2)
2832                 return ERR_PTR(-ENOSPC);
2833
2834         if (xchg(&root->fs_info->enospc_unlink, 1))
2835                 return ERR_PTR(-ENOSPC);
2836
2837         path = btrfs_alloc_path();
2838         if (!path) {
2839                 root->fs_info->enospc_unlink = 0;
2840                 return ERR_PTR(-ENOMEM);
2841         }
2842
2843         trans = btrfs_start_transaction(root, 0);
2844         if (IS_ERR(trans)) {
2845                 btrfs_free_path(path);
2846                 root->fs_info->enospc_unlink = 0;
2847                 return trans;
2848         }
2849
2850         path->skip_locking = 1;
2851         path->search_commit_root = 1;
2852
2853         ret = btrfs_lookup_inode(trans, root, path,
2854                                 &BTRFS_I(dir)->location, 0);
2855         if (ret < 0) {
2856                 err = ret;
2857                 goto out;
2858         }
2859         if (ret == 0) {
2860                 if (check_path_shared(root, path))
2861                         goto out;
2862         } else {
2863                 check_link = 0;
2864         }
2865         btrfs_release_path(path);
2866
2867         ret = btrfs_lookup_inode(trans, root, path,
2868                                 &BTRFS_I(inode)->location, 0);
2869         if (ret < 0) {
2870                 err = ret;
2871                 goto out;
2872         }
2873         if (ret == 0) {
2874                 if (check_path_shared(root, path))
2875                         goto out;
2876         } else {
2877                 check_link = 0;
2878         }
2879         btrfs_release_path(path);
2880
2881         if (ret == 0 && S_ISREG(inode->i_mode)) {
2882                 ret = btrfs_lookup_file_extent(trans, root, path,
2883                                                inode->i_ino, (u64)-1, 0);
2884                 if (ret < 0) {
2885                         err = ret;
2886                         goto out;
2887                 }
2888                 BUG_ON(ret == 0);
2889                 if (check_path_shared(root, path))
2890                         goto out;
2891                 btrfs_release_path(path);
2892         }
2893
2894         if (!check_link) {
2895                 err = 0;
2896                 goto out;
2897         }
2898
2899         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2900                                 dentry->d_name.name, dentry->d_name.len, 0);
2901         if (IS_ERR(di)) {
2902                 err = PTR_ERR(di);
2903                 goto out;
2904         }
2905         if (di) {
2906                 if (check_path_shared(root, path))
2907                         goto out;
2908         } else {
2909                 err = 0;
2910                 goto out;
2911         }
2912         btrfs_release_path(path);
2913
2914         ref = btrfs_lookup_inode_ref(trans, root, path,
2915                                 dentry->d_name.name, dentry->d_name.len,
2916                                 inode->i_ino, dir->i_ino, 0);
2917         if (IS_ERR(ref)) {
2918                 err = PTR_ERR(ref);
2919                 goto out;
2920         }
2921         BUG_ON(!ref);
2922         if (check_path_shared(root, path))
2923                 goto out;
2924         index = btrfs_inode_ref_index(path->nodes[0], ref);
2925         btrfs_release_path(path);
2926
2927         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2928                                 dentry->d_name.name, dentry->d_name.len, 0);
2929         if (IS_ERR(di)) {
2930                 err = PTR_ERR(di);
2931                 goto out;
2932         }
2933         BUG_ON(ret == -ENOENT);
2934         if (check_path_shared(root, path))
2935                 goto out;
2936
2937         err = 0;
2938 out:
2939         btrfs_free_path(path);
2940         if (err) {
2941                 btrfs_end_transaction(trans, root);
2942                 root->fs_info->enospc_unlink = 0;
2943                 return ERR_PTR(err);
2944         }
2945
2946         trans->block_rsv = &root->fs_info->global_block_rsv;
2947         return trans;
2948 }
2949
2950 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2951                                struct btrfs_root *root)
2952 {
2953         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2954                 BUG_ON(!root->fs_info->enospc_unlink);
2955                 root->fs_info->enospc_unlink = 0;
2956         }
2957         btrfs_end_transaction_throttle(trans, root);
2958 }
2959
2960 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2961 {
2962         struct btrfs_root *root = BTRFS_I(dir)->root;
2963         struct btrfs_trans_handle *trans;
2964         struct inode *inode = dentry->d_inode;
2965         int ret;
2966         unsigned long nr = 0;
2967
2968         trans = __unlink_start_trans(dir, dentry);
2969         if (IS_ERR(trans))
2970                 return PTR_ERR(trans);
2971
2972         btrfs_set_trans_block_group(trans, dir);
2973
2974         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2975
2976         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2977                                  dentry->d_name.name, dentry->d_name.len);
2978         BUG_ON(ret);
2979
2980         if (inode->i_nlink == 0) {
2981                 ret = btrfs_orphan_add(trans, inode);
2982                 BUG_ON(ret);
2983         }
2984
2985         nr = trans->blocks_used;
2986         __unlink_end_trans(trans, root);
2987         btrfs_btree_balance_dirty(root, nr);
2988         return ret;
2989 }
2990
2991 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2992                         struct btrfs_root *root,
2993                         struct inode *dir, u64 objectid,
2994                         const char *name, int name_len)
2995 {
2996         struct btrfs_path *path;
2997         struct extent_buffer *leaf;
2998         struct btrfs_dir_item *di;
2999         struct btrfs_key key;
3000         u64 index;
3001         int ret;
3002
3003         path = btrfs_alloc_path();
3004         if (!path)
3005                 return -ENOMEM;
3006
3007         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3008                                    name, name_len, -1);
3009         BUG_ON(IS_ERR_OR_NULL(di));
3010
3011         leaf = path->nodes[0];
3012         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3013         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3014         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3015         BUG_ON(ret);
3016         btrfs_release_path(path);
3017
3018         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3019                                  objectid, root->root_key.objectid,
3020                                  dir->i_ino, &index, name, name_len);
3021         if (ret < 0) {
3022                 BUG_ON(ret != -ENOENT);
3023                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3024                                                  name, name_len);
3025                 BUG_ON(IS_ERR_OR_NULL(di));
3026
3027                 leaf = path->nodes[0];
3028                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3029                 btrfs_release_path(path);
3030                 index = key.offset;
3031         }
3032
3033         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3034                                          index, name, name_len, -1);
3035         BUG_ON(IS_ERR_OR_NULL(di));
3036
3037         leaf = path->nodes[0];
3038         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3039         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3040         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3041         BUG_ON(ret);
3042         btrfs_release_path(path);
3043
3044         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3045         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3046         ret = btrfs_update_inode(trans, root, dir);
3047         BUG_ON(ret);
3048
3049         btrfs_free_path(path);
3050         return 0;
3051 }
3052
3053 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3054 {
3055         struct inode *inode = dentry->d_inode;
3056         int err = 0;
3057         struct btrfs_root *root = BTRFS_I(dir)->root;
3058         struct btrfs_trans_handle *trans;
3059         unsigned long nr = 0;
3060
3061         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3062             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3063                 return -ENOTEMPTY;
3064
3065         trans = __unlink_start_trans(dir, dentry);
3066         if (IS_ERR(trans))
3067                 return PTR_ERR(trans);
3068
3069         btrfs_set_trans_block_group(trans, dir);
3070
3071         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3072                 err = btrfs_unlink_subvol(trans, root, dir,
3073                                           BTRFS_I(inode)->location.objectid,
3074                                           dentry->d_name.name,
3075                                           dentry->d_name.len);
3076                 goto out;
3077         }
3078
3079         err = btrfs_orphan_add(trans, inode);
3080         if (err)
3081                 goto out;
3082
3083         /* now the directory is empty */
3084         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3085                                  dentry->d_name.name, dentry->d_name.len);
3086         if (!err)
3087                 btrfs_i_size_write(inode, 0);
3088 out:
3089         nr = trans->blocks_used;
3090         __unlink_end_trans(trans, root);
3091         btrfs_btree_balance_dirty(root, nr);
3092
3093         return err;
3094 }
3095
3096 /*
3097  * this can truncate away extent items, csum items and directory items.
3098  * It starts at a high offset and removes keys until it can't find
3099  * any higher than new_size
3100  *
3101  * csum items that cross the new i_size are truncated to the new size
3102  * as well.
3103  *
3104  * min_type is the minimum key type to truncate down to.  If set to 0, this
3105  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3106  */
3107 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3108                                struct btrfs_root *root,
3109                                struct inode *inode,
3110                                u64 new_size, u32 min_type)
3111 {
3112         struct btrfs_path *path;
3113         struct extent_buffer *leaf;
3114         struct btrfs_file_extent_item *fi;
3115         struct btrfs_key key;
3116         struct btrfs_key found_key;
3117         u64 extent_start = 0;
3118         u64 extent_num_bytes = 0;
3119         u64 extent_offset = 0;
3120         u64 item_end = 0;
3121         u64 mask = root->sectorsize - 1;
3122         u32 found_type = (u8)-1;
3123         int found_extent;
3124         int del_item;
3125         int pending_del_nr = 0;
3126         int pending_del_slot = 0;
3127         int extent_type = -1;
3128         int encoding;
3129         int ret;
3130         int err = 0;
3131
3132         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3133
3134         if (root->ref_cows || root == root->fs_info->tree_root)
3135                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3136
3137         path = btrfs_alloc_path();
3138         BUG_ON(!path);
3139         path->reada = -1;
3140
3141         key.objectid = inode->i_ino;
3142         key.offset = (u64)-1;
3143         key.type = (u8)-1;
3144
3145 search_again:
3146         path->leave_spinning = 1;
3147         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3148         if (ret < 0) {
3149                 err = ret;
3150                 goto out;
3151         }
3152
3153         if (ret > 0) {
3154                 /* there are no items in the tree for us to truncate, we're
3155                  * done
3156                  */
3157                 if (path->slots[0] == 0)
3158                         goto out;
3159                 path->slots[0]--;
3160         }
3161
3162         while (1) {
3163                 fi = NULL;
3164                 leaf = path->nodes[0];
3165                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3166                 found_type = btrfs_key_type(&found_key);
3167                 encoding = 0;
3168
3169                 if (found_key.objectid != inode->i_ino)
3170                         break;
3171
3172                 if (found_type < min_type)
3173                         break;
3174
3175                 item_end = found_key.offset;
3176                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3177                         fi = btrfs_item_ptr(leaf, path->slots[0],
3178                                             struct btrfs_file_extent_item);
3179                         extent_type = btrfs_file_extent_type(leaf, fi);
3180                         encoding = btrfs_file_extent_compression(leaf, fi);
3181                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3182                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3183
3184                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3185                                 item_end +=
3186                                     btrfs_file_extent_num_bytes(leaf, fi);
3187                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3188                                 item_end += btrfs_file_extent_inline_len(leaf,
3189                                                                          fi);
3190                         }
3191                         item_end--;
3192                 }
3193                 if (found_type > min_type) {
3194                         del_item = 1;
3195                 } else {
3196                         if (item_end < new_size)
3197                                 break;
3198                         if (found_key.offset >= new_size)
3199                                 del_item = 1;
3200                         else
3201                                 del_item = 0;
3202                 }
3203                 found_extent = 0;
3204                 /* FIXME, shrink the extent if the ref count is only 1 */
3205                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3206                         goto delete;
3207
3208                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3209                         u64 num_dec;
3210                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3211                         if (!del_item && !encoding) {
3212                                 u64 orig_num_bytes =
3213                                         btrfs_file_extent_num_bytes(leaf, fi);
3214                                 extent_num_bytes = new_size -
3215                                         found_key.offset + root->sectorsize - 1;
3216                                 extent_num_bytes = extent_num_bytes &
3217                                         ~((u64)root->sectorsize - 1);
3218                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3219                                                          extent_num_bytes);
3220                                 num_dec = (orig_num_bytes -
3221                                            extent_num_bytes);
3222                                 if (root->ref_cows && extent_start != 0)
3223                                         inode_sub_bytes(inode, num_dec);
3224                                 btrfs_mark_buffer_dirty(leaf);
3225                         } else {
3226                                 extent_num_bytes =
3227                                         btrfs_file_extent_disk_num_bytes(leaf,
3228                                                                          fi);
3229                                 extent_offset = found_key.offset -
3230                                         btrfs_file_extent_offset(leaf, fi);
3231
3232                                 /* FIXME blocksize != 4096 */
3233                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3234                                 if (extent_start != 0) {
3235                                         found_extent = 1;
3236                                         if (root->ref_cows)
3237                                                 inode_sub_bytes(inode, num_dec);
3238                                 }
3239                         }
3240                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3241                         /*
3242                          * we can't truncate inline items that have had
3243                          * special encodings
3244                          */
3245                         if (!del_item &&
3246                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3247                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3248                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3249                                 u32 size = new_size - found_key.offset;
3250
3251                                 if (root->ref_cows) {
3252                                         inode_sub_bytes(inode, item_end + 1 -
3253                                                         new_size);
3254                                 }
3255                                 size =
3256                                     btrfs_file_extent_calc_inline_size(size);
3257                                 ret = btrfs_truncate_item(trans, root, path,
3258                                                           size, 1);
3259                                 BUG_ON(ret);
3260                         } else if (root->ref_cows) {
3261                                 inode_sub_bytes(inode, item_end + 1 -
3262                                                 found_key.offset);
3263                         }
3264                 }
3265 delete:
3266                 if (del_item) {
3267                         if (!pending_del_nr) {
3268                                 /* no pending yet, add ourselves */
3269                                 pending_del_slot = path->slots[0];
3270                                 pending_del_nr = 1;
3271                         } else if (pending_del_nr &&
3272                                    path->slots[0] + 1 == pending_del_slot) {
3273                                 /* hop on the pending chunk */
3274                                 pending_del_nr++;
3275                                 pending_del_slot = path->slots[0];
3276                         } else {
3277                                 BUG();
3278                         }
3279                 } else {
3280                         break;
3281                 }
3282                 if (found_extent && (root->ref_cows ||
3283                                      root == root->fs_info->tree_root)) {
3284                         btrfs_set_path_blocking(path);
3285                         ret = btrfs_free_extent(trans, root, extent_start,
3286                                                 extent_num_bytes, 0,
3287                                                 btrfs_header_owner(leaf),
3288                                                 inode->i_ino, extent_offset);
3289                         BUG_ON(ret);
3290                 }
3291
3292                 if (found_type == BTRFS_INODE_ITEM_KEY)
3293                         break;
3294
3295                 if (path->slots[0] == 0 ||
3296                     path->slots[0] != pending_del_slot) {
3297                         if (root->ref_cows) {
3298                                 err = -EAGAIN;
3299                                 goto out;
3300                         }
3301                         if (pending_del_nr) {
3302                                 ret = btrfs_del_items(trans, root, path,
3303                                                 pending_del_slot,
3304                                                 pending_del_nr);
3305                                 BUG_ON(ret);
3306                                 pending_del_nr = 0;
3307                         }
3308                         btrfs_release_path(path);
3309                         goto search_again;
3310                 } else {
3311                         path->slots[0]--;
3312                 }
3313         }
3314 out:
3315         if (pending_del_nr) {
3316                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3317                                       pending_del_nr);
3318                 BUG_ON(ret);
3319         }
3320         btrfs_free_path(path);
3321         return err;
3322 }
3323
3324 /*
3325  * taken from block_truncate_page, but does cow as it zeros out
3326  * any bytes left in the last page in the file.
3327  */
3328 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3329 {
3330         struct inode *inode = mapping->host;
3331         struct btrfs_root *root = BTRFS_I(inode)->root;
3332         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3333         struct btrfs_ordered_extent *ordered;
3334         struct extent_state *cached_state = NULL;
3335         char *kaddr;
3336         u32 blocksize = root->sectorsize;
3337         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3338         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3339         struct page *page;
3340         int ret = 0;
3341         u64 page_start;
3342         u64 page_end;
3343
3344         if ((offset & (blocksize - 1)) == 0)
3345                 goto out;
3346         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3347         if (ret)
3348                 goto out;
3349
3350         ret = -ENOMEM;
3351 again:
3352         page = grab_cache_page(mapping, index);
3353         if (!page) {
3354                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3355                 goto out;
3356         }
3357
3358         page_start = page_offset(page);
3359         page_end = page_start + PAGE_CACHE_SIZE - 1;
3360
3361         if (!PageUptodate(page)) {
3362                 ret = btrfs_readpage(NULL, page);
3363                 lock_page(page);
3364                 if (page->mapping != mapping) {
3365                         unlock_page(page);
3366                         page_cache_release(page);
3367                         goto again;
3368                 }
3369                 if (!PageUptodate(page)) {
3370                         ret = -EIO;
3371                         goto out_unlock;
3372                 }
3373         }
3374         wait_on_page_writeback(page);
3375
3376         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3377                          GFP_NOFS);
3378         set_page_extent_mapped(page);
3379
3380         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3381         if (ordered) {
3382                 unlock_extent_cached(io_tree, page_start, page_end,
3383                                      &cached_state, GFP_NOFS);
3384                 unlock_page(page);
3385                 page_cache_release(page);
3386                 btrfs_start_ordered_extent(inode, ordered, 1);
3387                 btrfs_put_ordered_extent(ordered);
3388                 goto again;
3389         }
3390
3391         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3392                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3393                           0, 0, &cached_state, GFP_NOFS);
3394
3395         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3396                                         &cached_state);
3397         if (ret) {
3398                 unlock_extent_cached(io_tree, page_start, page_end,
3399                                      &cached_state, GFP_NOFS);
3400                 goto out_unlock;
3401         }
3402
3403         ret = 0;
3404         if (offset != PAGE_CACHE_SIZE) {
3405                 kaddr = kmap(page);
3406                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3407                 flush_dcache_page(page);
3408                 kunmap(page);
3409         }
3410         ClearPageChecked(page);
3411         set_page_dirty(page);
3412         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3413                              GFP_NOFS);
3414
3415 out_unlock:
3416         if (ret)
3417                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3418         unlock_page(page);
3419         page_cache_release(page);
3420 out:
3421         return ret;
3422 }
3423
3424 /*
3425  * This function puts in dummy file extents for the area we're creating a hole
3426  * for.  So if we are truncating this file to a larger size we need to insert
3427  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3428  * the range between oldsize and size
3429  */
3430 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3431 {
3432         struct btrfs_trans_handle *trans;
3433         struct btrfs_root *root = BTRFS_I(inode)->root;
3434         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3435         struct extent_map *em = NULL;
3436         struct extent_state *cached_state = NULL;
3437         u64 mask = root->sectorsize - 1;
3438         u64 hole_start = (oldsize + mask) & ~mask;
3439         u64 block_end = (size + mask) & ~mask;
3440         u64 last_byte;
3441         u64 cur_offset;
3442         u64 hole_size;
3443         int err = 0;
3444
3445         if (size <= hole_start)
3446                 return 0;
3447
3448         while (1) {
3449                 struct btrfs_ordered_extent *ordered;
3450                 btrfs_wait_ordered_range(inode, hole_start,
3451                                          block_end - hole_start);
3452                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3453                                  &cached_state, GFP_NOFS);
3454                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3455                 if (!ordered)
3456                         break;
3457                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3458                                      &cached_state, GFP_NOFS);
3459                 btrfs_put_ordered_extent(ordered);
3460         }
3461
3462         cur_offset = hole_start;
3463         while (1) {
3464                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3465                                 block_end - cur_offset, 0);
3466                 BUG_ON(IS_ERR_OR_NULL(em));
3467                 last_byte = min(extent_map_end(em), block_end);
3468                 last_byte = (last_byte + mask) & ~mask;
3469                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3470                         u64 hint_byte = 0;
3471                         hole_size = last_byte - cur_offset;
3472
3473                         trans = btrfs_start_transaction(root, 2);
3474                         if (IS_ERR(trans)) {
3475                                 err = PTR_ERR(trans);
3476                                 break;
3477                         }
3478                         btrfs_set_trans_block_group(trans, inode);
3479
3480                         err = btrfs_drop_extents(trans, inode, cur_offset,
3481                                                  cur_offset + hole_size,
3482                                                  &hint_byte, 1);
3483                         if (err)
3484                                 break;
3485
3486                         err = btrfs_insert_file_extent(trans, root,
3487                                         inode->i_ino, cur_offset, 0,
3488                                         0, hole_size, 0, hole_size,
3489                                         0, 0, 0);
3490                         if (err)
3491                                 break;
3492
3493                         btrfs_drop_extent_cache(inode, hole_start,
3494                                         last_byte - 1, 0);
3495
3496                         btrfs_end_transaction(trans, root);
3497                 }
3498                 free_extent_map(em);
3499                 em = NULL;
3500                 cur_offset = last_byte;
3501                 if (cur_offset >= block_end)
3502                         break;
3503         }
3504
3505         free_extent_map(em);
3506         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3507                              GFP_NOFS);
3508         return err;
3509 }
3510
3511 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3512 {
3513         loff_t oldsize = i_size_read(inode);
3514         int ret;
3515
3516         if (newsize == oldsize)
3517                 return 0;
3518
3519         if (newsize > oldsize) {
3520                 i_size_write(inode, newsize);
3521                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3522                 truncate_pagecache(inode, oldsize, newsize);
3523                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3524                 if (ret) {
3525                         btrfs_setsize(inode, oldsize);
3526                         return ret;
3527                 }
3528
3529                 mark_inode_dirty(inode);
3530         } else {
3531
3532                 /*
3533                  * We're truncating a file that used to have good data down to
3534                  * zero. Make sure it gets into the ordered flush list so that
3535                  * any new writes get down to disk quickly.
3536                  */
3537                 if (newsize == 0)
3538                         BTRFS_I(inode)->ordered_data_close = 1;
3539
3540                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3541                 truncate_setsize(inode, newsize);
3542                 ret = btrfs_truncate(inode);
3543         }
3544
3545         return ret;
3546 }
3547
3548 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3549 {
3550         struct inode *inode = dentry->d_inode;
3551         struct btrfs_root *root = BTRFS_I(inode)->root;
3552         int err;
3553
3554         if (btrfs_root_readonly(root))
3555                 return -EROFS;
3556
3557         err = inode_change_ok(inode, attr);
3558         if (err)
3559                 return err;
3560
3561         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3562                 err = btrfs_setsize(inode, attr->ia_size);
3563                 if (err)
3564                         return err;
3565         }
3566
3567         if (attr->ia_valid) {
3568                 setattr_copy(inode, attr);
3569                 mark_inode_dirty(inode);
3570
3571                 if (attr->ia_valid & ATTR_MODE)
3572                         err = btrfs_acl_chmod(inode);
3573         }
3574
3575         return err;
3576 }
3577
3578 void btrfs_evict_inode(struct inode *inode)
3579 {
3580         struct btrfs_trans_handle *trans;
3581         struct btrfs_root *root = BTRFS_I(inode)->root;
3582         unsigned long nr;
3583         int ret;
3584
3585         trace_btrfs_inode_evict(inode);
3586
3587         truncate_inode_pages(&inode->i_data, 0);
3588         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3589                                root == root->fs_info->tree_root))
3590                 goto no_delete;
3591
3592         if (is_bad_inode(inode)) {
3593                 btrfs_orphan_del(NULL, inode);
3594                 goto no_delete;
3595         }
3596         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3597         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3598
3599         if (root->fs_info->log_root_recovering) {
3600                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3601                 goto no_delete;
3602         }
3603
3604         if (inode->i_nlink > 0) {
3605                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3606                 goto no_delete;
3607         }
3608
3609         btrfs_i_size_write(inode, 0);
3610
3611         while (1) {
3612                 trans = btrfs_start_transaction(root, 0);
3613                 BUG_ON(IS_ERR(trans));
3614                 btrfs_set_trans_block_group(trans, inode);
3615                 trans->block_rsv = root->orphan_block_rsv;
3616
3617                 ret = btrfs_block_rsv_check(trans, root,
3618                                             root->orphan_block_rsv, 0, 5);
3619                 if (ret) {
3620                         BUG_ON(ret != -EAGAIN);
3621                         ret = btrfs_commit_transaction(trans, root);
3622                         BUG_ON(ret);
3623                         continue;
3624                 }
3625
3626                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3627                 if (ret != -EAGAIN)
3628                         break;
3629
3630                 nr = trans->blocks_used;
3631                 btrfs_end_transaction(trans, root);
3632                 trans = NULL;
3633                 btrfs_btree_balance_dirty(root, nr);
3634
3635         }
3636
3637         if (ret == 0) {
3638                 ret = btrfs_orphan_del(trans, inode);
3639                 BUG_ON(ret);
3640         }
3641
3642         nr = trans->blocks_used;
3643         btrfs_end_transaction(trans, root);
3644         btrfs_btree_balance_dirty(root, nr);
3645 no_delete:
3646         end_writeback(inode);
3647         return;
3648 }
3649
3650 /*
3651  * this returns the key found in the dir entry in the location pointer.
3652  * If no dir entries were found, location->objectid is 0.
3653  */
3654 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3655                                struct btrfs_key *location)
3656 {
3657         const char *name = dentry->d_name.name;
3658         int namelen = dentry->d_name.len;
3659         struct btrfs_dir_item *di;
3660         struct btrfs_path *path;
3661         struct btrfs_root *root = BTRFS_I(dir)->root;
3662         int ret = 0;
3663
3664         path = btrfs_alloc_path();
3665         BUG_ON(!path);
3666
3667         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3668                                     namelen, 0);
3669         if (IS_ERR(di))
3670                 ret = PTR_ERR(di);
3671
3672         if (IS_ERR_OR_NULL(di))
3673                 goto out_err;
3674
3675         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3676 out:
3677         btrfs_free_path(path);
3678         return ret;
3679 out_err:
3680         location->objectid = 0;
3681         goto out;
3682 }
3683
3684 /*
3685  * when we hit a tree root in a directory, the btrfs part of the inode
3686  * needs to be changed to reflect the root directory of the tree root.  This
3687  * is kind of like crossing a mount point.
3688  */
3689 static int fixup_tree_root_location(struct btrfs_root *root,
3690                                     struct inode *dir,
3691                                     struct dentry *dentry,
3692                                     struct btrfs_key *location,
3693                                     struct btrfs_root **sub_root)
3694 {
3695         struct btrfs_path *path;
3696         struct btrfs_root *new_root;
3697         struct btrfs_root_ref *ref;
3698         struct extent_buffer *leaf;
3699         int ret;
3700         int err = 0;
3701
3702         path = btrfs_alloc_path();
3703         if (!path) {
3704                 err = -ENOMEM;
3705                 goto out;
3706         }
3707
3708         err = -ENOENT;
3709         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3710                                   BTRFS_I(dir)->root->root_key.objectid,
3711                                   location->objectid);
3712         if (ret) {
3713                 if (ret < 0)
3714                         err = ret;
3715                 goto out;
3716         }
3717
3718         leaf = path->nodes[0];
3719         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3720         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3721             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3722                 goto out;
3723
3724         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3725                                    (unsigned long)(ref + 1),
3726                                    dentry->d_name.len);
3727         if (ret)
3728                 goto out;
3729
3730         btrfs_release_path(path);
3731
3732         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3733         if (IS_ERR(new_root)) {
3734                 err = PTR_ERR(new_root);
3735                 goto out;
3736         }
3737
3738         if (btrfs_root_refs(&new_root->root_item) == 0) {
3739                 err = -ENOENT;
3740                 goto out;
3741         }
3742
3743         *sub_root = new_root;
3744         location->objectid = btrfs_root_dirid(&new_root->root_item);
3745         location->type = BTRFS_INODE_ITEM_KEY;
3746         location->offset = 0;
3747         err = 0;
3748 out:
3749         btrfs_free_path(path);
3750         return err;
3751 }
3752
3753 static void inode_tree_add(struct inode *inode)
3754 {
3755         struct btrfs_root *root = BTRFS_I(inode)->root;
3756         struct btrfs_inode *entry;
3757         struct rb_node **p;
3758         struct rb_node *parent;
3759 again:
3760         p = &root->inode_tree.rb_node;
3761         parent = NULL;
3762
3763         if (inode_unhashed(inode))
3764                 return;
3765
3766         spin_lock(&root->inode_lock);
3767         while (*p) {
3768                 parent = *p;
3769                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3770
3771                 if (inode->i_ino < entry->vfs_inode.i_ino)
3772                         p = &parent->rb_left;
3773                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3774                         p = &parent->rb_right;
3775                 else {
3776                         WARN_ON(!(entry->vfs_inode.i_state &
3777                                   (I_WILL_FREE | I_FREEING)));
3778                         rb_erase(parent, &root->inode_tree);
3779                         RB_CLEAR_NODE(parent);
3780                         spin_unlock(&root->inode_lock);
3781                         goto again;
3782                 }
3783         }
3784         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3785         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3786         spin_unlock(&root->inode_lock);
3787 }
3788
3789 static void inode_tree_del(struct inode *inode)
3790 {
3791         struct btrfs_root *root = BTRFS_I(inode)->root;
3792         int empty = 0;
3793
3794         spin_lock(&root->inode_lock);
3795         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3796                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3797                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3798                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3799         }
3800         spin_unlock(&root->inode_lock);
3801
3802         /*
3803          * Free space cache has inodes in the tree root, but the tree root has a
3804          * root_refs of 0, so this could end up dropping the tree root as a
3805          * snapshot, so we need the extra !root->fs_info->tree_root check to
3806          * make sure we don't drop it.
3807          */
3808         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3809             root != root->fs_info->tree_root) {
3810                 synchronize_srcu(&root->fs_info->subvol_srcu);
3811                 spin_lock(&root->inode_lock);
3812                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3813                 spin_unlock(&root->inode_lock);
3814                 if (empty)
3815                         btrfs_add_dead_root(root);
3816         }
3817 }
3818
3819 int btrfs_invalidate_inodes(struct btrfs_root *root)
3820 {
3821         struct rb_node *node;
3822         struct rb_node *prev;
3823         struct btrfs_inode *entry;
3824         struct inode *inode;
3825         u64 objectid = 0;
3826
3827         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3828
3829         spin_lock(&root->inode_lock);
3830 again:
3831         node = root->inode_tree.rb_node;
3832         prev = NULL;
3833         while (node) {
3834                 prev = node;
3835                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3836
3837                 if (objectid < entry->vfs_inode.i_ino)
3838                         node = node->rb_left;
3839                 else if (objectid > entry->vfs_inode.i_ino)
3840                         node = node->rb_right;
3841                 else
3842                         break;
3843         }
3844         if (!node) {
3845                 while (prev) {
3846                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3847                         if (objectid <= entry->vfs_inode.i_ino) {
3848                                 node = prev;
3849                                 break;
3850                         }
3851                         prev = rb_next(prev);
3852                 }
3853         }
3854         while (node) {
3855                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3856                 objectid = entry->vfs_inode.i_ino + 1;
3857                 inode = igrab(&entry->vfs_inode);
3858                 if (inode) {
3859                         spin_unlock(&root->inode_lock);
3860                         if (atomic_read(&inode->i_count) > 1)
3861                                 d_prune_aliases(inode);
3862                         /*
3863                          * btrfs_drop_inode will have it removed from
3864                          * the inode cache when its usage count
3865                          * hits zero.
3866                          */
3867                         iput(inode);
3868                         cond_resched();
3869                         spin_lock(&root->inode_lock);
3870                         goto again;
3871                 }
3872
3873                 if (cond_resched_lock(&root->inode_lock))
3874                         goto again;
3875
3876                 node = rb_next(node);
3877         }
3878         spin_unlock(&root->inode_lock);
3879         return 0;
3880 }
3881
3882 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3883 {
3884         struct btrfs_iget_args *args = p;
3885         inode->i_ino = args->ino;
3886         BTRFS_I(inode)->root = args->root;
3887         btrfs_set_inode_space_info(args->root, inode);
3888         return 0;
3889 }
3890
3891 static int btrfs_find_actor(struct inode *inode, void *opaque)
3892 {
3893         struct btrfs_iget_args *args = opaque;
3894         return args->ino == inode->i_ino &&
3895                 args->root == BTRFS_I(inode)->root;
3896 }
3897
3898 static struct inode *btrfs_iget_locked(struct super_block *s,
3899                                        u64 objectid,
3900                                        struct btrfs_root *root)
3901 {
3902         struct inode *inode;
3903         struct btrfs_iget_args args;
3904         args.ino = objectid;
3905         args.root = root;
3906
3907         inode = iget5_locked(s, objectid, btrfs_find_actor,
3908                              btrfs_init_locked_inode,
3909                              (void *)&args);
3910         return inode;
3911 }
3912
3913 /* Get an inode object given its location and corresponding root.
3914  * Returns in *is_new if the inode was read from disk
3915  */
3916 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3917                          struct btrfs_root *root, int *new)
3918 {
3919         struct inode *inode;
3920
3921         inode = btrfs_iget_locked(s, location->objectid, root);
3922         if (!inode)
3923                 return ERR_PTR(-ENOMEM);
3924
3925         if (inode->i_state & I_NEW) {
3926                 BTRFS_I(inode)->root = root;
3927                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3928                 btrfs_read_locked_inode(inode);
3929                 inode_tree_add(inode);
3930                 unlock_new_inode(inode);
3931                 if (new)
3932                         *new = 1;
3933         }
3934
3935         return inode;
3936 }
3937
3938 static struct inode *new_simple_dir(struct super_block *s,
3939                                     struct btrfs_key *key,
3940                                     struct btrfs_root *root)
3941 {
3942         struct inode *inode = new_inode(s);
3943
3944         if (!inode)
3945                 return ERR_PTR(-ENOMEM);
3946
3947         BTRFS_I(inode)->root = root;
3948         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3949         BTRFS_I(inode)->dummy_inode = 1;
3950
3951         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3952         inode->i_op = &simple_dir_inode_operations;
3953         inode->i_fop = &simple_dir_operations;
3954         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3955         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3956
3957         return inode;
3958 }
3959
3960 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3961 {
3962         struct inode *inode;
3963         struct btrfs_root *root = BTRFS_I(dir)->root;
3964         struct btrfs_root *sub_root = root;
3965         struct btrfs_key location;
3966         int index;
3967         int ret;
3968
3969         if (dentry->d_name.len > BTRFS_NAME_LEN)
3970                 return ERR_PTR(-ENAMETOOLONG);
3971
3972         ret = btrfs_inode_by_name(dir, dentry, &location);
3973
3974         if (ret < 0)
3975                 return ERR_PTR(ret);
3976
3977         if (location.objectid == 0)
3978                 return NULL;
3979
3980         if (location.type == BTRFS_INODE_ITEM_KEY) {
3981                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3982                 return inode;
3983         }
3984
3985         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3986
3987         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3988         ret = fixup_tree_root_location(root, dir, dentry,
3989                                        &location, &sub_root);
3990         if (ret < 0) {
3991                 if (ret != -ENOENT)
3992                         inode = ERR_PTR(ret);
3993                 else
3994                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3995         } else {
3996                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3997         }
3998         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3999
4000         if (!IS_ERR(inode) && root != sub_root) {
4001                 down_read(&root->fs_info->cleanup_work_sem);
4002                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4003                         ret = btrfs_orphan_cleanup(sub_root);
4004                 up_read(&root->fs_info->cleanup_work_sem);
4005                 if (ret)
4006                         inode = ERR_PTR(ret);
4007         }
4008
4009         return inode;
4010 }
4011
4012 static int btrfs_dentry_delete(const struct dentry *dentry)
4013 {
4014         struct btrfs_root *root;
4015
4016         if (!dentry->d_inode && !IS_ROOT(dentry))
4017                 dentry = dentry->d_parent;
4018
4019         if (dentry->d_inode) {
4020                 root = BTRFS_I(dentry->d_inode)->root;
4021                 if (btrfs_root_refs(&root->root_item) == 0)
4022                         return 1;
4023         }
4024         return 0;
4025 }
4026
4027 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4028                                    struct nameidata *nd)
4029 {
4030         struct inode *inode;
4031
4032         inode = btrfs_lookup_dentry(dir, dentry);
4033         if (IS_ERR(inode))
4034                 return ERR_CAST(inode);
4035
4036         return d_splice_alias(inode, dentry);
4037 }
4038
4039 static unsigned char btrfs_filetype_table[] = {
4040         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4041 };
4042
4043 static int btrfs_real_readdir(struct file *filp, void *dirent,
4044                               filldir_t filldir)
4045 {
4046         struct inode *inode = filp->f_dentry->d_inode;
4047         struct btrfs_root *root = BTRFS_I(inode)->root;
4048         struct btrfs_item *item;
4049         struct btrfs_dir_item *di;
4050         struct btrfs_key key;
4051         struct btrfs_key found_key;
4052         struct btrfs_path *path;
4053         int ret;
4054         struct extent_buffer *leaf;
4055         int slot;
4056         unsigned char d_type;
4057         int over = 0;
4058         u32 di_cur;
4059         u32 di_total;
4060         u32 di_len;
4061         int key_type = BTRFS_DIR_INDEX_KEY;
4062         char tmp_name[32];
4063         char *name_ptr;
4064         int name_len;
4065
4066         /* FIXME, use a real flag for deciding about the key type */
4067         if (root->fs_info->tree_root == root)
4068                 key_type = BTRFS_DIR_ITEM_KEY;
4069
4070         /* special case for "." */
4071         if (filp->f_pos == 0) {
4072                 over = filldir(dirent, ".", 1,
4073                                1, inode->i_ino,
4074                                DT_DIR);
4075                 if (over)
4076                         return 0;
4077                 filp->f_pos = 1;
4078         }
4079         /* special case for .., just use the back ref */
4080         if (filp->f_pos == 1) {
4081                 u64 pino = parent_ino(filp->f_path.dentry);
4082                 over = filldir(dirent, "..", 2,
4083                                2, pino, DT_DIR);
4084                 if (over)
4085                         return 0;
4086                 filp->f_pos = 2;
4087         }
4088         path = btrfs_alloc_path();
4089         path->reada = 2;
4090
4091         btrfs_set_key_type(&key, key_type);
4092         key.offset = filp->f_pos;
4093         key.objectid = inode->i_ino;
4094
4095         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4096         if (ret < 0)
4097                 goto err;
4098
4099         while (1) {
4100                 leaf = path->nodes[0];
4101                 slot = path->slots[0];
4102                 if (slot >= btrfs_header_nritems(leaf)) {
4103                         ret = btrfs_next_leaf(root, path);
4104                         if (ret < 0)
4105                                 goto err;
4106                         else if (ret > 0)
4107                                 break;
4108                         continue;
4109                 }
4110
4111                 item = btrfs_item_nr(leaf, slot);
4112                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4113
4114                 if (found_key.objectid != key.objectid)
4115                         break;
4116                 if (btrfs_key_type(&found_key) != key_type)
4117                         break;
4118                 if (found_key.offset < filp->f_pos)
4119                         goto next;
4120
4121                 filp->f_pos = found_key.offset;
4122
4123                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4124                 di_cur = 0;
4125                 di_total = btrfs_item_size(leaf, item);
4126
4127                 while (di_cur < di_total) {
4128                         struct btrfs_key location;
4129
4130                         if (verify_dir_item(root, leaf, di))
4131                                 break;
4132
4133                         name_len = btrfs_dir_name_len(leaf, di);
4134                         if (name_len <= sizeof(tmp_name)) {
4135                                 name_ptr = tmp_name;
4136                         } else {
4137                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4138                                 if (!name_ptr) {
4139                                         ret = -ENOMEM;
4140                                         goto err;
4141                                 }
4142                         }
4143                         read_extent_buffer(leaf, name_ptr,
4144                                            (unsigned long)(di + 1), name_len);
4145
4146                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4147                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4148
4149                         /* is this a reference to our own snapshot? If so
4150                          * skip it
4151                          */
4152                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4153                             location.objectid == root->root_key.objectid) {
4154                                 over = 0;
4155                                 goto skip;
4156                         }
4157                         over = filldir(dirent, name_ptr, name_len,
4158                                        found_key.offset, location.objectid,
4159                                        d_type);
4160
4161 skip:
4162                         if (name_ptr != tmp_name)
4163                                 kfree(name_ptr);
4164
4165                         if (over)
4166                                 goto nopos;
4167                         di_len = btrfs_dir_name_len(leaf, di) +
4168                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4169                         di_cur += di_len;
4170                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4171                 }
4172 next:
4173                 path->slots[0]++;
4174         }
4175
4176         /* Reached end of directory/root. Bump pos past the last item. */
4177         if (key_type == BTRFS_DIR_INDEX_KEY)
4178                 /*
4179                  * 32-bit glibc will use getdents64, but then strtol -
4180                  * so the last number we can serve is this.
4181                  */
4182                 filp->f_pos = 0x7fffffff;
4183         else
4184                 filp->f_pos++;
4185 nopos:
4186         ret = 0;
4187 err:
4188         btrfs_free_path(path);
4189         return ret;
4190 }
4191
4192 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4193 {
4194         struct btrfs_root *root = BTRFS_I(inode)->root;
4195         struct btrfs_trans_handle *trans;
4196         int ret = 0;
4197         bool nolock = false;
4198
4199         if (BTRFS_I(inode)->dummy_inode)
4200                 return 0;
4201
4202         smp_mb();
4203         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4204
4205         if (wbc->sync_mode == WB_SYNC_ALL) {
4206                 if (nolock)
4207                         trans = btrfs_join_transaction_nolock(root, 1);
4208                 else
4209                         trans = btrfs_join_transaction(root, 1);
4210                 if (IS_ERR(trans))
4211                         return PTR_ERR(trans);
4212                 btrfs_set_trans_block_group(trans, inode);
4213                 if (nolock)
4214                         ret = btrfs_end_transaction_nolock(trans, root);
4215                 else
4216                         ret = btrfs_commit_transaction(trans, root);
4217         }
4218         return ret;
4219 }
4220
4221 /*
4222  * This is somewhat expensive, updating the tree every time the
4223  * inode changes.  But, it is most likely to find the inode in cache.
4224  * FIXME, needs more benchmarking...there are no reasons other than performance
4225  * to keep or drop this code.
4226  */
4227 void btrfs_dirty_inode(struct inode *inode)
4228 {
4229         struct btrfs_root *root = BTRFS_I(inode)->root;
4230         struct btrfs_trans_handle *trans;
4231         int ret;
4232
4233         if (BTRFS_I(inode)->dummy_inode)
4234                 return;
4235
4236         trans = btrfs_join_transaction(root, 1);
4237         BUG_ON(IS_ERR(trans));
4238         btrfs_set_trans_block_group(trans, inode);
4239
4240         ret = btrfs_update_inode(trans, root, inode);
4241         if (ret && ret == -ENOSPC) {
4242                 /* whoops, lets try again with the full transaction */
4243                 btrfs_end_transaction(trans, root);
4244                 trans = btrfs_start_transaction(root, 1);
4245                 if (IS_ERR(trans)) {
4246                         if (printk_ratelimit()) {
4247                                 printk(KERN_ERR "btrfs: fail to "
4248                                        "dirty  inode %lu error %ld\n",
4249                                        inode->i_ino, PTR_ERR(trans));
4250                         }
4251                         return;
4252                 }
4253                 btrfs_set_trans_block_group(trans, inode);
4254
4255                 ret = btrfs_update_inode(trans, root, inode);
4256                 if (ret) {
4257                         if (printk_ratelimit()) {
4258                                 printk(KERN_ERR "btrfs: fail to "
4259                                        "dirty  inode %lu error %d\n",
4260                                        inode->i_ino, ret);
4261                         }
4262                 }
4263         }
4264         btrfs_end_transaction(trans, root);
4265 }
4266
4267 /*
4268  * find the highest existing sequence number in a directory
4269  * and then set the in-memory index_cnt variable to reflect
4270  * free sequence numbers
4271  */
4272 static int btrfs_set_inode_index_count(struct inode *inode)
4273 {
4274         struct btrfs_root *root = BTRFS_I(inode)->root;
4275         struct btrfs_key key, found_key;
4276         struct btrfs_path *path;
4277         struct extent_buffer *leaf;
4278         int ret;
4279
4280         key.objectid = inode->i_ino;
4281         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4282         key.offset = (u64)-1;
4283
4284         path = btrfs_alloc_path();
4285         if (!path)
4286                 return -ENOMEM;
4287
4288         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4289         if (ret < 0)
4290                 goto out;
4291         /* FIXME: we should be able to handle this */
4292         if (ret == 0)
4293                 goto out;
4294         ret = 0;
4295
4296         /*
4297          * MAGIC NUMBER EXPLANATION:
4298          * since we search a directory based on f_pos we have to start at 2
4299          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4300          * else has to start at 2
4301          */
4302         if (path->slots[0] == 0) {
4303                 BTRFS_I(inode)->index_cnt = 2;
4304                 goto out;
4305         }
4306
4307         path->slots[0]--;
4308
4309         leaf = path->nodes[0];
4310         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4311
4312         if (found_key.objectid != inode->i_ino ||
4313             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4314                 BTRFS_I(inode)->index_cnt = 2;
4315                 goto out;
4316         }
4317
4318         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4319 out:
4320         btrfs_free_path(path);
4321         return ret;
4322 }
4323
4324 /*
4325  * helper to find a free sequence number in a given directory.  This current
4326  * code is very simple, later versions will do smarter things in the btree
4327  */
4328 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4329 {
4330         int ret = 0;
4331
4332         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4333                 ret = btrfs_set_inode_index_count(dir);
4334                 if (ret)
4335                         return ret;
4336         }
4337
4338         *index = BTRFS_I(dir)->index_cnt;
4339         BTRFS_I(dir)->index_cnt++;
4340
4341         return ret;
4342 }
4343
4344 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4345                                      struct btrfs_root *root,
4346                                      struct inode *dir,
4347                                      const char *name, int name_len,
4348                                      u64 ref_objectid, u64 objectid,
4349                                      u64 alloc_hint, int mode, u64 *index)
4350 {
4351         struct inode *inode;
4352         struct btrfs_inode_item *inode_item;
4353         struct btrfs_key *location;
4354         struct btrfs_path *path;
4355         struct btrfs_inode_ref *ref;
4356         struct btrfs_key key[2];
4357         u32 sizes[2];
4358         unsigned long ptr;
4359         int ret;
4360         int owner;
4361
4362         path = btrfs_alloc_path();
4363         BUG_ON(!path);
4364
4365         inode = new_inode(root->fs_info->sb);
4366         if (!inode) {
4367                 btrfs_free_path(path);
4368                 return ERR_PTR(-ENOMEM);
4369         }
4370
4371         if (dir) {
4372                 trace_btrfs_inode_request(dir);
4373
4374                 ret = btrfs_set_inode_index(dir, index);
4375                 if (ret) {
4376                         btrfs_free_path(path);
4377                         iput(inode);
4378                         return ERR_PTR(ret);
4379                 }
4380         }
4381         /*
4382          * index_cnt is ignored for everything but a dir,
4383          * btrfs_get_inode_index_count has an explanation for the magic
4384          * number
4385          */
4386         BTRFS_I(inode)->index_cnt = 2;
4387         BTRFS_I(inode)->root = root;
4388         BTRFS_I(inode)->generation = trans->transid;
4389         inode->i_generation = BTRFS_I(inode)->generation;
4390         btrfs_set_inode_space_info(root, inode);
4391
4392         if (mode & S_IFDIR)
4393                 owner = 0;
4394         else
4395                 owner = 1;
4396         BTRFS_I(inode)->block_group =
4397                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4398
4399         key[0].objectid = objectid;
4400         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4401         key[0].offset = 0;
4402
4403         key[1].objectid = objectid;
4404         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4405         key[1].offset = ref_objectid;
4406
4407         sizes[0] = sizeof(struct btrfs_inode_item);
4408         sizes[1] = name_len + sizeof(*ref);
4409
4410         path->leave_spinning = 1;
4411         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4412         if (ret != 0)
4413                 goto fail;
4414
4415         inode_init_owner(inode, dir, mode);
4416         inode->i_ino = objectid;
4417         inode_set_bytes(inode, 0);
4418         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4419         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4420                                   struct btrfs_inode_item);
4421         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4422
4423         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4424                              struct btrfs_inode_ref);
4425         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4426         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4427         ptr = (unsigned long)(ref + 1);
4428         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4429
4430         btrfs_mark_buffer_dirty(path->nodes[0]);
4431         btrfs_free_path(path);
4432
4433         location = &BTRFS_I(inode)->location;
4434         location->objectid = objectid;
4435         location->offset = 0;
4436         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4437
4438         btrfs_inherit_iflags(inode, dir);
4439
4440         if ((mode & S_IFREG)) {
4441                 if (btrfs_test_opt(root, NODATASUM))
4442                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4443                 if (btrfs_test_opt(root, NODATACOW) ||
4444                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4445                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4446         }
4447
4448         insert_inode_hash(inode);
4449         inode_tree_add(inode);
4450
4451         trace_btrfs_inode_new(inode);
4452
4453         return inode;
4454 fail:
4455         if (dir)
4456                 BTRFS_I(dir)->index_cnt--;
4457         btrfs_free_path(path);
4458         iput(inode);
4459         return ERR_PTR(ret);
4460 }
4461
4462 static inline u8 btrfs_inode_type(struct inode *inode)
4463 {
4464         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4465 }
4466
4467 /*
4468  * utility function to add 'inode' into 'parent_inode' with
4469  * a give name and a given sequence number.
4470  * if 'add_backref' is true, also insert a backref from the
4471  * inode to the parent directory.
4472  */
4473 int btrfs_add_link(struct btrfs_trans_handle *trans,
4474                    struct inode *parent_inode, struct inode *inode,
4475                    const char *name, int name_len, int add_backref, u64 index)
4476 {
4477         int ret = 0;
4478         struct btrfs_key key;
4479         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4480
4481         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4482                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4483         } else {
4484                 key.objectid = inode->i_ino;
4485                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4486                 key.offset = 0;
4487         }
4488
4489         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4490                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4491                                          key.objectid, root->root_key.objectid,
4492                                          parent_inode->i_ino,
4493                                          index, name, name_len);
4494         } else if (add_backref) {
4495                 ret = btrfs_insert_inode_ref(trans, root,
4496                                              name, name_len, inode->i_ino,
4497                                              parent_inode->i_ino, index);
4498         }
4499
4500         if (ret == 0) {
4501                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4502                                             parent_inode->i_ino, &key,
4503                                             btrfs_inode_type(inode), index);
4504                 BUG_ON(ret);
4505
4506                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4507                                    name_len * 2);
4508                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4509                 ret = btrfs_update_inode(trans, root, parent_inode);
4510         }
4511         return ret;
4512 }
4513
4514 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4515                             struct inode *dir, struct dentry *dentry,
4516                             struct inode *inode, int backref, u64 index)
4517 {
4518         int err = btrfs_add_link(trans, dir, inode,
4519                                  dentry->d_name.name, dentry->d_name.len,
4520                                  backref, index);
4521         if (!err) {
4522                 d_instantiate(dentry, inode);
4523                 return 0;
4524         }
4525         if (err > 0)
4526                 err = -EEXIST;
4527         return err;
4528 }
4529
4530 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4531                         int mode, dev_t rdev)
4532 {
4533         struct btrfs_trans_handle *trans;
4534         struct btrfs_root *root = BTRFS_I(dir)->root;
4535         struct inode *inode = NULL;
4536         int err;
4537         int drop_inode = 0;
4538         u64 objectid;
4539         unsigned long nr = 0;
4540         u64 index = 0;
4541
4542         if (!new_valid_dev(rdev))
4543                 return -EINVAL;
4544
4545         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4546         if (err)
4547                 return err;
4548
4549         /*
4550          * 2 for inode item and ref
4551          * 2 for dir items
4552          * 1 for xattr if selinux is on
4553          */
4554         trans = btrfs_start_transaction(root, 5);
4555         if (IS_ERR(trans))
4556                 return PTR_ERR(trans);
4557
4558         btrfs_set_trans_block_group(trans, dir);
4559
4560         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4561                                 dentry->d_name.len, dir->i_ino, objectid,
4562                                 BTRFS_I(dir)->block_group, mode, &index);
4563         if (IS_ERR(inode)) {
4564                 err = PTR_ERR(inode);
4565                 goto out_unlock;
4566         }
4567
4568         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4569         if (err) {
4570                 drop_inode = 1;
4571                 goto out_unlock;
4572         }
4573
4574         btrfs_set_trans_block_group(trans, inode);
4575         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4576         if (err)
4577                 drop_inode = 1;
4578         else {
4579                 inode->i_op = &btrfs_special_inode_operations;
4580                 init_special_inode(inode, inode->i_mode, rdev);
4581                 btrfs_update_inode(trans, root, inode);
4582         }
4583         btrfs_update_inode_block_group(trans, inode);
4584         btrfs_update_inode_block_group(trans, dir);
4585 out_unlock:
4586         nr = trans->blocks_used;
4587         btrfs_end_transaction_throttle(trans, root);
4588         btrfs_btree_balance_dirty(root, nr);
4589         if (drop_inode) {
4590                 inode_dec_link_count(inode);
4591                 iput(inode);
4592         }
4593         return err;
4594 }
4595
4596 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4597                         int mode, struct nameidata *nd)
4598 {
4599         struct btrfs_trans_handle *trans;
4600         struct btrfs_root *root = BTRFS_I(dir)->root;
4601         struct inode *inode = NULL;
4602         int drop_inode = 0;
4603         int err;
4604         unsigned long nr = 0;
4605         u64 objectid;
4606         u64 index = 0;
4607
4608         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4609         if (err)
4610                 return err;
4611         /*
4612          * 2 for inode item and ref
4613          * 2 for dir items
4614          * 1 for xattr if selinux is on
4615          */
4616         trans = btrfs_start_transaction(root, 5);
4617         if (IS_ERR(trans))
4618                 return PTR_ERR(trans);
4619
4620         btrfs_set_trans_block_group(trans, dir);
4621
4622         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4623                                 dentry->d_name.len, dir->i_ino, objectid,
4624                                 BTRFS_I(dir)->block_group, mode, &index);
4625         if (IS_ERR(inode)) {
4626                 err = PTR_ERR(inode);
4627                 goto out_unlock;
4628         }
4629
4630         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4631         if (err) {
4632                 drop_inode = 1;
4633                 goto out_unlock;
4634         }
4635
4636         btrfs_set_trans_block_group(trans, inode);
4637         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4638         if (err)
4639                 drop_inode = 1;
4640         else {
4641                 inode->i_mapping->a_ops = &btrfs_aops;
4642                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4643                 inode->i_fop = &btrfs_file_operations;
4644                 inode->i_op = &btrfs_file_inode_operations;
4645                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4646         }
4647         btrfs_update_inode_block_group(trans, inode);
4648         btrfs_update_inode_block_group(trans, dir);
4649 out_unlock:
4650         nr = trans->blocks_used;
4651         btrfs_end_transaction_throttle(trans, root);
4652         if (drop_inode) {
4653                 inode_dec_link_count(inode);
4654                 iput(inode);
4655         }
4656         btrfs_btree_balance_dirty(root, nr);
4657         return err;
4658 }
4659
4660 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4661                       struct dentry *dentry)
4662 {
4663         struct btrfs_trans_handle *trans;
4664         struct btrfs_root *root = BTRFS_I(dir)->root;
4665         struct inode *inode = old_dentry->d_inode;
4666         u64 index;
4667         unsigned long nr = 0;
4668         int err;
4669         int drop_inode = 0;
4670
4671         /* do not allow sys_link's with other subvols of the same device */
4672         if (root->objectid != BTRFS_I(inode)->root->objectid)
4673                 return -EXDEV;
4674
4675         if (inode->i_nlink == ~0U)
4676                 return -EMLINK;
4677
4678         err = btrfs_set_inode_index(dir, &index);
4679         if (err)
4680                 goto fail;
4681
4682         /*
4683          * 2 items for inode and inode ref
4684          * 2 items for dir items
4685          * 1 item for parent inode
4686          */
4687         trans = btrfs_start_transaction(root, 5);
4688         if (IS_ERR(trans)) {
4689                 err = PTR_ERR(trans);
4690                 goto fail;
4691         }
4692
4693         btrfs_inc_nlink(inode);
4694         inode->i_ctime = CURRENT_TIME;
4695
4696         btrfs_set_trans_block_group(trans, dir);
4697         ihold(inode);
4698
4699         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4700
4701         if (err) {
4702                 drop_inode = 1;
4703         } else {
4704                 struct dentry *parent = dget_parent(dentry);
4705                 btrfs_update_inode_block_group(trans, dir);
4706                 err = btrfs_update_inode(trans, root, inode);
4707                 BUG_ON(err);
4708                 btrfs_log_new_name(trans, inode, NULL, parent);
4709                 dput(parent);
4710         }
4711
4712         nr = trans->blocks_used;
4713         btrfs_end_transaction_throttle(trans, root);
4714 fail:
4715         if (drop_inode) {
4716                 inode_dec_link_count(inode);
4717                 iput(inode);
4718         }
4719         btrfs_btree_balance_dirty(root, nr);
4720         return err;
4721 }
4722
4723 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4724 {
4725         struct inode *inode = NULL;
4726         struct btrfs_trans_handle *trans;
4727         struct btrfs_root *root = BTRFS_I(dir)->root;
4728         int err = 0;
4729         int drop_on_err = 0;
4730         u64 objectid = 0;
4731         u64 index = 0;
4732         unsigned long nr = 1;
4733
4734         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4735         if (err)
4736                 return err;
4737
4738         /*
4739          * 2 items for inode and ref
4740          * 2 items for dir items
4741          * 1 for xattr if selinux is on
4742          */
4743         trans = btrfs_start_transaction(root, 5);
4744         if (IS_ERR(trans))
4745                 return PTR_ERR(trans);
4746         btrfs_set_trans_block_group(trans, dir);
4747
4748         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4749                                 dentry->d_name.len, dir->i_ino, objectid,
4750                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4751                                 &index);
4752         if (IS_ERR(inode)) {
4753                 err = PTR_ERR(inode);
4754                 goto out_fail;
4755         }
4756
4757         drop_on_err = 1;
4758
4759         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4760         if (err)
4761                 goto out_fail;
4762
4763         inode->i_op = &btrfs_dir_inode_operations;
4764         inode->i_fop = &btrfs_dir_file_operations;
4765         btrfs_set_trans_block_group(trans, inode);
4766
4767         btrfs_i_size_write(inode, 0);
4768         err = btrfs_update_inode(trans, root, inode);
4769         if (err)
4770                 goto out_fail;
4771
4772         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4773                              dentry->d_name.len, 0, index);
4774         if (err)
4775                 goto out_fail;
4776
4777         d_instantiate(dentry, inode);
4778         drop_on_err = 0;
4779         btrfs_update_inode_block_group(trans, inode);
4780         btrfs_update_inode_block_group(trans, dir);
4781
4782 out_fail:
4783         nr = trans->blocks_used;
4784         btrfs_end_transaction_throttle(trans, root);
4785         if (drop_on_err)
4786                 iput(inode);
4787         btrfs_btree_balance_dirty(root, nr);
4788         return err;
4789 }
4790
4791 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4792  * and an extent that you want to insert, deal with overlap and insert
4793  * the new extent into the tree.
4794  */
4795 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4796                                 struct extent_map *existing,
4797                                 struct extent_map *em,
4798                                 u64 map_start, u64 map_len)
4799 {
4800         u64 start_diff;
4801
4802         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4803         start_diff = map_start - em->start;
4804         em->start = map_start;
4805         em->len = map_len;
4806         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4807             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4808                 em->block_start += start_diff;
4809                 em->block_len -= start_diff;
4810         }
4811         return add_extent_mapping(em_tree, em);
4812 }
4813
4814 static noinline int uncompress_inline(struct btrfs_path *path,
4815                                       struct inode *inode, struct page *page,
4816                                       size_t pg_offset, u64 extent_offset,
4817                                       struct btrfs_file_extent_item *item)
4818 {
4819         int ret;
4820         struct extent_buffer *leaf = path->nodes[0];
4821         char *tmp;
4822         size_t max_size;
4823         unsigned long inline_size;
4824         unsigned long ptr;
4825         int compress_type;
4826
4827         WARN_ON(pg_offset != 0);
4828         compress_type = btrfs_file_extent_compression(leaf, item);
4829         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4830         inline_size = btrfs_file_extent_inline_item_len(leaf,
4831                                         btrfs_item_nr(leaf, path->slots[0]));
4832         tmp = kmalloc(inline_size, GFP_NOFS);
4833         if (!tmp)
4834                 return -ENOMEM;
4835         ptr = btrfs_file_extent_inline_start(item);
4836
4837         read_extent_buffer(leaf, tmp, ptr, inline_size);
4838
4839         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4840         ret = btrfs_decompress(compress_type, tmp, page,
4841                                extent_offset, inline_size, max_size);
4842         if (ret) {
4843                 char *kaddr = kmap_atomic(page, KM_USER0);
4844                 unsigned long copy_size = min_t(u64,
4845                                   PAGE_CACHE_SIZE - pg_offset,
4846                                   max_size - extent_offset);
4847                 memset(kaddr + pg_offset, 0, copy_size);
4848                 kunmap_atomic(kaddr, KM_USER0);
4849         }
4850         kfree(tmp);
4851         return 0;
4852 }
4853
4854 /*
4855  * a bit scary, this does extent mapping from logical file offset to the disk.
4856  * the ugly parts come from merging extents from the disk with the in-ram
4857  * representation.  This gets more complex because of the data=ordered code,
4858  * where the in-ram extents might be locked pending data=ordered completion.
4859  *
4860  * This also copies inline extents directly into the page.
4861  */
4862
4863 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4864                                     size_t pg_offset, u64 start, u64 len,
4865                                     int create)
4866 {
4867         int ret;
4868         int err = 0;
4869         u64 bytenr;
4870         u64 extent_start = 0;
4871         u64 extent_end = 0;
4872         u64 objectid = inode->i_ino;
4873         u32 found_type;
4874         struct btrfs_path *path = NULL;
4875         struct btrfs_root *root = BTRFS_I(inode)->root;
4876         struct btrfs_file_extent_item *item;
4877         struct extent_buffer *leaf;
4878         struct btrfs_key found_key;
4879         struct extent_map *em = NULL;
4880         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4881         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4882         struct btrfs_trans_handle *trans = NULL;
4883         int compress_type;
4884
4885 again:
4886         read_lock(&em_tree->lock);
4887         em = lookup_extent_mapping(em_tree, start, len);
4888         if (em)
4889                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4890         read_unlock(&em_tree->lock);
4891
4892         if (em) {
4893                 if (em->start > start || em->start + em->len <= start)
4894                         free_extent_map(em);
4895                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4896                         free_extent_map(em);
4897                 else
4898                         goto out;
4899         }
4900         em = alloc_extent_map();
4901         if (!em) {
4902                 err = -ENOMEM;
4903                 goto out;
4904         }
4905         em->bdev = root->fs_info->fs_devices->latest_bdev;
4906         em->start = EXTENT_MAP_HOLE;
4907         em->orig_start = EXTENT_MAP_HOLE;
4908         em->len = (u64)-1;
4909         em->block_len = (u64)-1;
4910
4911         if (!path) {
4912                 path = btrfs_alloc_path();
4913                 BUG_ON(!path);
4914         }
4915
4916         ret = btrfs_lookup_file_extent(trans, root, path,
4917                                        objectid, start, trans != NULL);
4918         if (ret < 0) {
4919                 err = ret;
4920                 goto out;
4921         }
4922
4923         if (ret != 0) {
4924                 if (path->slots[0] == 0)
4925                         goto not_found;
4926                 path->slots[0]--;
4927         }
4928
4929         leaf = path->nodes[0];
4930         item = btrfs_item_ptr(leaf, path->slots[0],
4931                               struct btrfs_file_extent_item);
4932         /* are we inside the extent that was found? */
4933         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4934         found_type = btrfs_key_type(&found_key);
4935         if (found_key.objectid != objectid ||
4936             found_type != BTRFS_EXTENT_DATA_KEY) {
4937                 goto not_found;
4938         }
4939
4940         found_type = btrfs_file_extent_type(leaf, item);
4941         extent_start = found_key.offset;
4942         compress_type = btrfs_file_extent_compression(leaf, item);
4943         if (found_type == BTRFS_FILE_EXTENT_REG ||
4944             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4945                 extent_end = extent_start +
4946                        btrfs_file_extent_num_bytes(leaf, item);
4947         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4948                 size_t size;
4949                 size = btrfs_file_extent_inline_len(leaf, item);
4950                 extent_end = (extent_start + size + root->sectorsize - 1) &
4951                         ~((u64)root->sectorsize - 1);
4952         }
4953
4954         if (start >= extent_end) {
4955                 path->slots[0]++;
4956                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4957                         ret = btrfs_next_leaf(root, path);
4958                         if (ret < 0) {
4959                                 err = ret;
4960                                 goto out;
4961                         }
4962                         if (ret > 0)
4963                                 goto not_found;
4964                         leaf = path->nodes[0];
4965                 }
4966                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4967                 if (found_key.objectid != objectid ||
4968                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4969                         goto not_found;
4970                 if (start + len <= found_key.offset)
4971                         goto not_found;
4972                 em->start = start;
4973                 em->len = found_key.offset - start;
4974                 goto not_found_em;
4975         }
4976
4977         if (found_type == BTRFS_FILE_EXTENT_REG ||
4978             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4979                 em->start = extent_start;
4980                 em->len = extent_end - extent_start;
4981                 em->orig_start = extent_start -
4982                                  btrfs_file_extent_offset(leaf, item);
4983                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4984                 if (bytenr == 0) {
4985                         em->block_start = EXTENT_MAP_HOLE;
4986                         goto insert;
4987                 }
4988                 if (compress_type != BTRFS_COMPRESS_NONE) {
4989                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4990                         em->compress_type = compress_type;
4991                         em->block_start = bytenr;
4992                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4993                                                                          item);
4994                 } else {
4995                         bytenr += btrfs_file_extent_offset(leaf, item);
4996                         em->block_start = bytenr;
4997                         em->block_len = em->len;
4998                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4999                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5000                 }
5001                 goto insert;
5002         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5003                 unsigned long ptr;
5004                 char *map;
5005                 size_t size;
5006                 size_t extent_offset;
5007                 size_t copy_size;
5008
5009                 em->block_start = EXTENT_MAP_INLINE;
5010                 if (!page || create) {
5011                         em->start = extent_start;
5012                         em->len = extent_end - extent_start;
5013                         goto out;
5014                 }
5015
5016                 size = btrfs_file_extent_inline_len(leaf, item);
5017                 extent_offset = page_offset(page) + pg_offset - extent_start;
5018                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5019                                 size - extent_offset);
5020                 em->start = extent_start + extent_offset;
5021                 em->len = (copy_size + root->sectorsize - 1) &
5022                         ~((u64)root->sectorsize - 1);
5023                 em->orig_start = EXTENT_MAP_INLINE;
5024                 if (compress_type) {
5025                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5026                         em->compress_type = compress_type;
5027                 }
5028                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5029                 if (create == 0 && !PageUptodate(page)) {
5030                         if (btrfs_file_extent_compression(leaf, item) !=
5031                             BTRFS_COMPRESS_NONE) {
5032                                 ret = uncompress_inline(path, inode, page,
5033                                                         pg_offset,
5034                                                         extent_offset, item);
5035                                 BUG_ON(ret);
5036                         } else {
5037                                 map = kmap(page);
5038                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5039                                                    copy_size);
5040                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5041                                         memset(map + pg_offset + copy_size, 0,
5042                                                PAGE_CACHE_SIZE - pg_offset -
5043                                                copy_size);
5044                                 }
5045                                 kunmap(page);
5046                         }
5047                         flush_dcache_page(page);
5048                 } else if (create && PageUptodate(page)) {
5049                         WARN_ON(1);
5050                         if (!trans) {
5051                                 kunmap(page);
5052                                 free_extent_map(em);
5053                                 em = NULL;
5054                                 btrfs_release_path(path);
5055                                 trans = btrfs_join_transaction(root, 1);
5056                                 if (IS_ERR(trans))
5057                                         return ERR_CAST(trans);
5058                                 goto again;
5059                         }
5060                         map = kmap(page);
5061                         write_extent_buffer(leaf, map + pg_offset, ptr,
5062                                             copy_size);
5063                         kunmap(page);
5064                         btrfs_mark_buffer_dirty(leaf);
5065                 }
5066                 set_extent_uptodate(io_tree, em->start,
5067                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5068                 goto insert;
5069         } else {
5070                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5071                 WARN_ON(1);
5072         }
5073 not_found:
5074         em->start = start;
5075         em->len = len;
5076 not_found_em:
5077         em->block_start = EXTENT_MAP_HOLE;
5078         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5079 insert:
5080         btrfs_release_path(path);
5081         if (em->start > start || extent_map_end(em) <= start) {
5082                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5083                        "[%llu %llu]\n", (unsigned long long)em->start,
5084                        (unsigned long long)em->len,
5085                        (unsigned long long)start,
5086                        (unsigned long long)len);
5087                 err = -EIO;
5088                 goto out;
5089         }
5090
5091         err = 0;
5092         write_lock(&em_tree->lock);
5093         ret = add_extent_mapping(em_tree, em);
5094         /* it is possible that someone inserted the extent into the tree
5095          * while we had the lock dropped.  It is also possible that
5096          * an overlapping map exists in the tree
5097          */
5098         if (ret == -EEXIST) {
5099                 struct extent_map *existing;
5100
5101                 ret = 0;
5102
5103                 existing = lookup_extent_mapping(em_tree, start, len);
5104                 if (existing && (existing->start > start ||
5105                     existing->start + existing->len <= start)) {
5106                         free_extent_map(existing);
5107                         existing = NULL;
5108                 }
5109                 if (!existing) {
5110                         existing = lookup_extent_mapping(em_tree, em->start,
5111                                                          em->len);
5112                         if (existing) {
5113                                 err = merge_extent_mapping(em_tree, existing,
5114                                                            em, start,
5115                                                            root->sectorsize);
5116                                 free_extent_map(existing);
5117                                 if (err) {
5118                                         free_extent_map(em);
5119                                         em = NULL;
5120                                 }
5121                         } else {
5122                                 err = -EIO;
5123                                 free_extent_map(em);
5124                                 em = NULL;
5125                         }
5126                 } else {
5127                         free_extent_map(em);
5128                         em = existing;
5129                         err = 0;
5130                 }
5131         }
5132         write_unlock(&em_tree->lock);
5133 out:
5134
5135         trace_btrfs_get_extent(root, em);
5136
5137         if (path)
5138                 btrfs_free_path(path);
5139         if (trans) {
5140                 ret = btrfs_end_transaction(trans, root);
5141                 if (!err)
5142                         err = ret;
5143         }
5144         if (err) {
5145                 free_extent_map(em);
5146                 return ERR_PTR(err);
5147         }
5148         return em;
5149 }
5150
5151 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5152                                            size_t pg_offset, u64 start, u64 len,
5153                                            int create)
5154 {
5155         struct extent_map *em;
5156         struct extent_map *hole_em = NULL;
5157         u64 range_start = start;
5158         u64 end;
5159         u64 found;
5160         u64 found_end;
5161         int err = 0;
5162
5163         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5164         if (IS_ERR(em))
5165                 return em;
5166         if (em) {
5167                 /*
5168                  * if our em maps to a hole, there might
5169                  * actually be delalloc bytes behind it
5170                  */
5171                 if (em->block_start != EXTENT_MAP_HOLE)
5172                         return em;
5173                 else
5174                         hole_em = em;
5175         }
5176
5177         /* check to see if we've wrapped (len == -1 or similar) */
5178         end = start + len;
5179         if (end < start)
5180                 end = (u64)-1;
5181         else
5182                 end -= 1;
5183
5184         em = NULL;
5185
5186         /* ok, we didn't find anything, lets look for delalloc */
5187         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5188                                  end, len, EXTENT_DELALLOC, 1);
5189         found_end = range_start + found;
5190         if (found_end < range_start)
5191                 found_end = (u64)-1;
5192
5193         /*
5194          * we didn't find anything useful, return
5195          * the original results from get_extent()
5196          */
5197         if (range_start > end || found_end <= start) {
5198                 em = hole_em;
5199                 hole_em = NULL;
5200                 goto out;
5201         }
5202
5203         /* adjust the range_start to make sure it doesn't
5204          * go backwards from the start they passed in
5205          */
5206         range_start = max(start,range_start);
5207         found = found_end - range_start;
5208
5209         if (found > 0) {
5210                 u64 hole_start = start;
5211                 u64 hole_len = len;
5212
5213                 em = alloc_extent_map();
5214                 if (!em) {
5215                         err = -ENOMEM;
5216                         goto out;
5217                 }
5218                 /*
5219                  * when btrfs_get_extent can't find anything it
5220                  * returns one huge hole
5221                  *
5222                  * make sure what it found really fits our range, and
5223                  * adjust to make sure it is based on the start from
5224                  * the caller
5225                  */
5226                 if (hole_em) {
5227                         u64 calc_end = extent_map_end(hole_em);
5228
5229                         if (calc_end <= start || (hole_em->start > end)) {
5230                                 free_extent_map(hole_em);
5231                                 hole_em = NULL;
5232                         } else {
5233                                 hole_start = max(hole_em->start, start);
5234                                 hole_len = calc_end - hole_start;
5235                         }
5236                 }
5237                 em->bdev = NULL;
5238                 if (hole_em && range_start > hole_start) {
5239                         /* our hole starts before our delalloc, so we
5240                          * have to return just the parts of the hole
5241                          * that go until  the delalloc starts
5242                          */
5243                         em->len = min(hole_len,
5244                                       range_start - hole_start);
5245                         em->start = hole_start;
5246                         em->orig_start = hole_start;
5247                         /*
5248                          * don't adjust block start at all,
5249                          * it is fixed at EXTENT_MAP_HOLE
5250                          */
5251                         em->block_start = hole_em->block_start;
5252                         em->block_len = hole_len;
5253                 } else {
5254                         em->start = range_start;
5255                         em->len = found;
5256                         em->orig_start = range_start;
5257                         em->block_start = EXTENT_MAP_DELALLOC;
5258                         em->block_len = found;
5259                 }
5260         } else if (hole_em) {
5261                 return hole_em;
5262         }
5263 out:
5264
5265         free_extent_map(hole_em);
5266         if (err) {
5267                 free_extent_map(em);
5268                 return ERR_PTR(err);
5269         }
5270         return em;
5271 }
5272
5273 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5274                                                   struct extent_map *em,
5275                                                   u64 start, u64 len)
5276 {
5277         struct btrfs_root *root = BTRFS_I(inode)->root;
5278         struct btrfs_trans_handle *trans;
5279         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5280         struct btrfs_key ins;
5281         u64 alloc_hint;
5282         int ret;
5283         bool insert = false;
5284
5285         /*
5286          * Ok if the extent map we looked up is a hole and is for the exact
5287          * range we want, there is no reason to allocate a new one, however if
5288          * it is not right then we need to free this one and drop the cache for
5289          * our range.
5290          */
5291         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5292             em->len != len) {
5293                 free_extent_map(em);
5294                 em = NULL;
5295                 insert = true;
5296                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5297         }
5298
5299         trans = btrfs_join_transaction(root, 0);
5300         if (IS_ERR(trans))
5301                 return ERR_CAST(trans);
5302
5303         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5304
5305         alloc_hint = get_extent_allocation_hint(inode, start, len);
5306         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5307                                    alloc_hint, (u64)-1, &ins, 1);
5308         if (ret) {
5309                 em = ERR_PTR(ret);
5310                 goto out;
5311         }
5312
5313         if (!em) {
5314                 em = alloc_extent_map();
5315                 if (!em) {
5316                         em = ERR_PTR(-ENOMEM);
5317                         goto out;
5318                 }
5319         }
5320
5321         em->start = start;
5322         em->orig_start = em->start;
5323         em->len = ins.offset;
5324
5325         em->block_start = ins.objectid;
5326         em->block_len = ins.offset;
5327         em->bdev = root->fs_info->fs_devices->latest_bdev;
5328
5329         /*
5330          * We need to do this because if we're using the original em we searched
5331          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5332          */
5333         em->flags = 0;
5334         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5335
5336         while (insert) {
5337                 write_lock(&em_tree->lock);
5338                 ret = add_extent_mapping(em_tree, em);
5339                 write_unlock(&em_tree->lock);
5340                 if (ret != -EEXIST)
5341                         break;
5342                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5343         }
5344
5345         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5346                                            ins.offset, ins.offset, 0);
5347         if (ret) {
5348                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5349                 em = ERR_PTR(ret);
5350         }
5351 out:
5352         btrfs_end_transaction(trans, root);
5353         return em;
5354 }
5355
5356 /*
5357  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5358  * block must be cow'd
5359  */
5360 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5361                                       struct inode *inode, u64 offset, u64 len)
5362 {
5363         struct btrfs_path *path;
5364         int ret;
5365         struct extent_buffer *leaf;
5366         struct btrfs_root *root = BTRFS_I(inode)->root;
5367         struct btrfs_file_extent_item *fi;
5368         struct btrfs_key key;
5369         u64 disk_bytenr;
5370         u64 backref_offset;
5371         u64 extent_end;
5372         u64 num_bytes;
5373         int slot;
5374         int found_type;
5375
5376         path = btrfs_alloc_path();
5377         if (!path)
5378                 return -ENOMEM;
5379
5380         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5381                                        offset, 0);
5382         if (ret < 0)
5383                 goto out;
5384
5385         slot = path->slots[0];
5386         if (ret == 1) {
5387                 if (slot == 0) {
5388                         /* can't find the item, must cow */
5389                         ret = 0;
5390                         goto out;
5391                 }
5392                 slot--;
5393         }
5394         ret = 0;
5395         leaf = path->nodes[0];
5396         btrfs_item_key_to_cpu(leaf, &key, slot);
5397         if (key.objectid != inode->i_ino ||
5398             key.type != BTRFS_EXTENT_DATA_KEY) {
5399                 /* not our file or wrong item type, must cow */
5400                 goto out;
5401         }
5402
5403         if (key.offset > offset) {
5404                 /* Wrong offset, must cow */
5405                 goto out;
5406         }
5407
5408         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5409         found_type = btrfs_file_extent_type(leaf, fi);
5410         if (found_type != BTRFS_FILE_EXTENT_REG &&
5411             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5412                 /* not a regular extent, must cow */
5413                 goto out;
5414         }
5415         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5416         backref_offset = btrfs_file_extent_offset(leaf, fi);
5417
5418         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5419         if (extent_end < offset + len) {
5420                 /* extent doesn't include our full range, must cow */
5421                 goto out;
5422         }
5423
5424         if (btrfs_extent_readonly(root, disk_bytenr))
5425                 goto out;
5426
5427         /*
5428          * look for other files referencing this extent, if we
5429          * find any we must cow
5430          */
5431         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5432                                   key.offset - backref_offset, disk_bytenr))
5433                 goto out;
5434
5435         /*
5436          * adjust disk_bytenr and num_bytes to cover just the bytes
5437          * in this extent we are about to write.  If there
5438          * are any csums in that range we have to cow in order
5439          * to keep the csums correct
5440          */
5441         disk_bytenr += backref_offset;
5442         disk_bytenr += offset - key.offset;
5443         num_bytes = min(offset + len, extent_end) - offset;
5444         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5445                                 goto out;
5446         /*
5447          * all of the above have passed, it is safe to overwrite this extent
5448          * without cow
5449          */
5450         ret = 1;
5451 out:
5452         btrfs_free_path(path);
5453         return ret;
5454 }
5455
5456 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5457                                    struct buffer_head *bh_result, int create)
5458 {
5459         struct extent_map *em;
5460         struct btrfs_root *root = BTRFS_I(inode)->root;
5461         u64 start = iblock << inode->i_blkbits;
5462         u64 len = bh_result->b_size;
5463         struct btrfs_trans_handle *trans;
5464
5465         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5466         if (IS_ERR(em))
5467                 return PTR_ERR(em);
5468
5469         /*
5470          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5471          * io.  INLINE is special, and we could probably kludge it in here, but
5472          * it's still buffered so for safety lets just fall back to the generic
5473          * buffered path.
5474          *
5475          * For COMPRESSED we _have_ to read the entire extent in so we can
5476          * decompress it, so there will be buffering required no matter what we
5477          * do, so go ahead and fallback to buffered.
5478          *
5479          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5480          * to buffered IO.  Don't blame me, this is the price we pay for using
5481          * the generic code.
5482          */
5483         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5484             em->block_start == EXTENT_MAP_INLINE) {
5485                 free_extent_map(em);
5486                 return -ENOTBLK;
5487         }
5488
5489         /* Just a good old fashioned hole, return */
5490         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5491                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5492                 free_extent_map(em);
5493                 /* DIO will do one hole at a time, so just unlock a sector */
5494                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5495                               start + root->sectorsize - 1, GFP_NOFS);
5496                 return 0;
5497         }
5498
5499         /*
5500          * We don't allocate a new extent in the following cases
5501          *
5502          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5503          * existing extent.
5504          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5505          * just use the extent.
5506          *
5507          */
5508         if (!create) {
5509                 len = em->len - (start - em->start);
5510                 goto map;
5511         }
5512
5513         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5514             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5515              em->block_start != EXTENT_MAP_HOLE)) {
5516                 int type;
5517                 int ret;
5518                 u64 block_start;
5519
5520                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5521                         type = BTRFS_ORDERED_PREALLOC;
5522                 else
5523                         type = BTRFS_ORDERED_NOCOW;
5524                 len = min(len, em->len - (start - em->start));
5525                 block_start = em->block_start + (start - em->start);
5526
5527                 /*
5528                  * we're not going to log anything, but we do need
5529                  * to make sure the current transaction stays open
5530                  * while we look for nocow cross refs
5531                  */
5532                 trans = btrfs_join_transaction(root, 0);
5533                 if (IS_ERR(trans))
5534                         goto must_cow;
5535
5536                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5537                         ret = btrfs_add_ordered_extent_dio(inode, start,
5538                                            block_start, len, len, type);
5539                         btrfs_end_transaction(trans, root);
5540                         if (ret) {
5541                                 free_extent_map(em);
5542                                 return ret;
5543                         }
5544                         goto unlock;
5545                 }
5546                 btrfs_end_transaction(trans, root);
5547         }
5548 must_cow:
5549         /*
5550          * this will cow the extent, reset the len in case we changed
5551          * it above
5552          */
5553         len = bh_result->b_size;
5554         em = btrfs_new_extent_direct(inode, em, start, len);
5555         if (IS_ERR(em))
5556                 return PTR_ERR(em);
5557         len = min(len, em->len - (start - em->start));
5558 unlock:
5559         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5560                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5561                           0, NULL, GFP_NOFS);
5562 map:
5563         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5564                 inode->i_blkbits;
5565         bh_result->b_size = len;
5566         bh_result->b_bdev = em->bdev;
5567         set_buffer_mapped(bh_result);
5568         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5569                 set_buffer_new(bh_result);
5570
5571         free_extent_map(em);
5572
5573         return 0;
5574 }
5575
5576 struct btrfs_dio_private {
5577         struct inode *inode;
5578         u64 logical_offset;
5579         u64 disk_bytenr;
5580         u64 bytes;
5581         u32 *csums;
5582         void *private;
5583
5584         /* number of bios pending for this dio */
5585         atomic_t pending_bios;
5586
5587         /* IO errors */
5588         int errors;
5589
5590         struct bio *orig_bio;
5591 };
5592
5593 static void btrfs_endio_direct_read(struct bio *bio, int err)
5594 {
5595         struct btrfs_dio_private *dip = bio->bi_private;
5596         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5597         struct bio_vec *bvec = bio->bi_io_vec;
5598         struct inode *inode = dip->inode;
5599         struct btrfs_root *root = BTRFS_I(inode)->root;
5600         u64 start;
5601         u32 *private = dip->csums;
5602
5603         start = dip->logical_offset;
5604         do {
5605                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5606                         struct page *page = bvec->bv_page;
5607                         char *kaddr;
5608                         u32 csum = ~(u32)0;
5609                         unsigned long flags;
5610
5611                         local_irq_save(flags);
5612                         kaddr = kmap_atomic(page, KM_IRQ0);
5613                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5614                                                csum, bvec->bv_len);
5615                         btrfs_csum_final(csum, (char *)&csum);
5616                         kunmap_atomic(kaddr, KM_IRQ0);
5617                         local_irq_restore(flags);
5618
5619                         flush_dcache_page(bvec->bv_page);
5620                         if (csum != *private) {
5621                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5622                                       " %llu csum %u private %u\n",
5623                                       inode->i_ino, (unsigned long long)start,
5624                                       csum, *private);
5625                                 err = -EIO;
5626                         }
5627                 }
5628
5629                 start += bvec->bv_len;
5630                 private++;
5631                 bvec++;
5632         } while (bvec <= bvec_end);
5633
5634         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5635                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5636         bio->bi_private = dip->private;
5637
5638         kfree(dip->csums);
5639         kfree(dip);
5640
5641         /* If we had a csum failure make sure to clear the uptodate flag */
5642         if (err)
5643                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5644         dio_end_io(bio, err);
5645 }
5646
5647 static void btrfs_endio_direct_write(struct bio *bio, int err)
5648 {
5649         struct btrfs_dio_private *dip = bio->bi_private;
5650         struct inode *inode = dip->inode;
5651         struct btrfs_root *root = BTRFS_I(inode)->root;
5652         struct btrfs_trans_handle *trans;
5653         struct btrfs_ordered_extent *ordered = NULL;
5654         struct extent_state *cached_state = NULL;
5655         u64 ordered_offset = dip->logical_offset;
5656         u64 ordered_bytes = dip->bytes;
5657         int ret;
5658
5659         if (err)
5660                 goto out_done;
5661 again:
5662         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5663                                                    &ordered_offset,
5664                                                    ordered_bytes);
5665         if (!ret)
5666                 goto out_test;
5667
5668         BUG_ON(!ordered);
5669
5670         trans = btrfs_join_transaction(root, 1);
5671         if (IS_ERR(trans)) {
5672                 err = -ENOMEM;
5673                 goto out;
5674         }
5675         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5676
5677         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5678                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5679                 if (!ret)
5680                         ret = btrfs_update_inode(trans, root, inode);
5681                 err = ret;
5682                 goto out;
5683         }
5684
5685         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5686                          ordered->file_offset + ordered->len - 1, 0,
5687                          &cached_state, GFP_NOFS);
5688
5689         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5690                 ret = btrfs_mark_extent_written(trans, inode,
5691                                                 ordered->file_offset,
5692                                                 ordered->file_offset +
5693                                                 ordered->len);
5694                 if (ret) {
5695                         err = ret;
5696                         goto out_unlock;
5697                 }
5698         } else {
5699                 ret = insert_reserved_file_extent(trans, inode,
5700                                                   ordered->file_offset,
5701                                                   ordered->start,
5702                                                   ordered->disk_len,
5703                                                   ordered->len,
5704                                                   ordered->len,
5705                                                   0, 0, 0,
5706                                                   BTRFS_FILE_EXTENT_REG);
5707                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5708                                    ordered->file_offset, ordered->len);
5709                 if (ret) {
5710                         err = ret;
5711                         WARN_ON(1);
5712                         goto out_unlock;
5713                 }
5714         }
5715
5716         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5717         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5718         if (!ret)
5719                 btrfs_update_inode(trans, root, inode);
5720         ret = 0;
5721 out_unlock:
5722         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5723                              ordered->file_offset + ordered->len - 1,
5724                              &cached_state, GFP_NOFS);
5725 out:
5726         btrfs_delalloc_release_metadata(inode, ordered->len);
5727         btrfs_end_transaction(trans, root);
5728         ordered_offset = ordered->file_offset + ordered->len;
5729         btrfs_put_ordered_extent(ordered);
5730         btrfs_put_ordered_extent(ordered);
5731
5732 out_test:
5733         /*
5734          * our bio might span multiple ordered extents.  If we haven't
5735          * completed the accounting for the whole dio, go back and try again
5736          */
5737         if (ordered_offset < dip->logical_offset + dip->bytes) {
5738                 ordered_bytes = dip->logical_offset + dip->bytes -
5739                         ordered_offset;
5740                 goto again;
5741         }
5742 out_done:
5743         bio->bi_private = dip->private;
5744
5745         kfree(dip->csums);
5746         kfree(dip);
5747
5748         /* If we had an error make sure to clear the uptodate flag */
5749         if (err)
5750                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5751         dio_end_io(bio, err);
5752 }
5753
5754 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5755                                     struct bio *bio, int mirror_num,
5756                                     unsigned long bio_flags, u64 offset)
5757 {
5758         int ret;
5759         struct btrfs_root *root = BTRFS_I(inode)->root;
5760         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5761         BUG_ON(ret);
5762         return 0;
5763 }
5764
5765 static void btrfs_end_dio_bio(struct bio *bio, int err)
5766 {
5767         struct btrfs_dio_private *dip = bio->bi_private;
5768
5769         if (err) {
5770                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5771                       "sector %#Lx len %u err no %d\n",
5772                       dip->inode->i_ino, bio->bi_rw,
5773                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5774                 dip->errors = 1;
5775
5776                 /*
5777                  * before atomic variable goto zero, we must make sure
5778                  * dip->errors is perceived to be set.
5779                  */
5780                 smp_mb__before_atomic_dec();
5781         }
5782
5783         /* if there are more bios still pending for this dio, just exit */
5784         if (!atomic_dec_and_test(&dip->pending_bios))
5785                 goto out;
5786
5787         if (dip->errors)
5788                 bio_io_error(dip->orig_bio);
5789         else {
5790                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5791                 bio_endio(dip->orig_bio, 0);
5792         }
5793 out:
5794         bio_put(bio);
5795 }
5796
5797 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5798                                        u64 first_sector, gfp_t gfp_flags)
5799 {
5800         int nr_vecs = bio_get_nr_vecs(bdev);
5801         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5802 }
5803
5804 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5805                                          int rw, u64 file_offset, int skip_sum,
5806                                          u32 *csums, int async_submit)
5807 {
5808         int write = rw & REQ_WRITE;
5809         struct btrfs_root *root = BTRFS_I(inode)->root;
5810         int ret;
5811
5812         bio_get(bio);
5813         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5814         if (ret)
5815                 goto err;
5816
5817         if (skip_sum)
5818                 goto map;
5819
5820         if (write && async_submit) {
5821                 ret = btrfs_wq_submit_bio(root->fs_info,
5822                                    inode, rw, bio, 0, 0,
5823                                    file_offset,
5824                                    __btrfs_submit_bio_start_direct_io,
5825                                    __btrfs_submit_bio_done);
5826                 goto err;
5827         } else if (write) {
5828                 /*
5829                  * If we aren't doing async submit, calculate the csum of the
5830                  * bio now.
5831                  */
5832                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5833                 if (ret)
5834                         goto err;
5835         } else if (!skip_sum) {
5836                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5837                                           file_offset, csums);
5838                 if (ret)
5839                         goto err;
5840         }
5841
5842 map:
5843         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5844 err:
5845         bio_put(bio);
5846         return ret;
5847 }
5848
5849 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5850                                     int skip_sum)
5851 {
5852         struct inode *inode = dip->inode;
5853         struct btrfs_root *root = BTRFS_I(inode)->root;
5854         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5855         struct bio *bio;
5856         struct bio *orig_bio = dip->orig_bio;
5857         struct bio_vec *bvec = orig_bio->bi_io_vec;
5858         u64 start_sector = orig_bio->bi_sector;
5859         u64 file_offset = dip->logical_offset;
5860         u64 submit_len = 0;
5861         u64 map_length;
5862         int nr_pages = 0;
5863         u32 *csums = dip->csums;
5864         int ret = 0;
5865         int async_submit = 0;
5866         int write = rw & REQ_WRITE;
5867
5868         map_length = orig_bio->bi_size;
5869         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5870                               &map_length, NULL, 0);
5871         if (ret) {
5872                 bio_put(orig_bio);
5873                 return -EIO;
5874         }
5875
5876         if (map_length >= orig_bio->bi_size) {
5877                 bio = orig_bio;
5878                 goto submit;
5879         }
5880
5881         async_submit = 1;
5882         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5883         if (!bio)
5884                 return -ENOMEM;
5885         bio->bi_private = dip;
5886         bio->bi_end_io = btrfs_end_dio_bio;
5887         atomic_inc(&dip->pending_bios);
5888
5889         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5890                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5891                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5892                                  bvec->bv_offset) < bvec->bv_len)) {
5893                         /*
5894                          * inc the count before we submit the bio so
5895                          * we know the end IO handler won't happen before
5896                          * we inc the count. Otherwise, the dip might get freed
5897                          * before we're done setting it up
5898                          */
5899                         atomic_inc(&dip->pending_bios);
5900                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5901                                                      file_offset, skip_sum,
5902                                                      csums, async_submit);
5903                         if (ret) {
5904                                 bio_put(bio);
5905                                 atomic_dec(&dip->pending_bios);
5906                                 goto out_err;
5907                         }
5908
5909                         /* Write's use the ordered csums */
5910                         if (!write && !skip_sum)
5911                                 csums = csums + nr_pages;
5912                         start_sector += submit_len >> 9;
5913                         file_offset += submit_len;
5914
5915                         submit_len = 0;
5916                         nr_pages = 0;
5917
5918                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5919                                                   start_sector, GFP_NOFS);
5920                         if (!bio)
5921                                 goto out_err;
5922                         bio->bi_private = dip;
5923                         bio->bi_end_io = btrfs_end_dio_bio;
5924
5925                         map_length = orig_bio->bi_size;
5926                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5927                                               &map_length, NULL, 0);
5928                         if (ret) {
5929                                 bio_put(bio);
5930                                 goto out_err;
5931                         }
5932                 } else {
5933                         submit_len += bvec->bv_len;
5934                         nr_pages ++;
5935                         bvec++;
5936                 }
5937         }
5938
5939 submit:
5940         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5941                                      csums, async_submit);
5942         if (!ret)
5943                 return 0;
5944
5945         bio_put(bio);
5946 out_err:
5947         dip->errors = 1;
5948         /*
5949          * before atomic variable goto zero, we must
5950          * make sure dip->errors is perceived to be set.
5951          */
5952         smp_mb__before_atomic_dec();
5953         if (atomic_dec_and_test(&dip->pending_bios))
5954                 bio_io_error(dip->orig_bio);
5955
5956         /* bio_end_io() will handle error, so we needn't return it */
5957         return 0;
5958 }
5959
5960 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5961                                 loff_t file_offset)
5962 {
5963         struct btrfs_root *root = BTRFS_I(inode)->root;
5964         struct btrfs_dio_private *dip;
5965         struct bio_vec *bvec = bio->bi_io_vec;
5966         int skip_sum;
5967         int write = rw & REQ_WRITE;
5968         int ret = 0;
5969
5970         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5971
5972         dip = kmalloc(sizeof(*dip), GFP_NOFS);
5973         if (!dip) {
5974                 ret = -ENOMEM;
5975                 goto free_ordered;
5976         }
5977         dip->csums = NULL;
5978
5979         /* Write's use the ordered csum stuff, so we don't need dip->csums */
5980         if (!write && !skip_sum) {
5981                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5982                 if (!dip->csums) {
5983                         kfree(dip);
5984                         ret = -ENOMEM;
5985                         goto free_ordered;
5986                 }
5987         }
5988
5989         dip->private = bio->bi_private;
5990         dip->inode = inode;
5991         dip->logical_offset = file_offset;
5992
5993         dip->bytes = 0;
5994         do {
5995                 dip->bytes += bvec->bv_len;
5996                 bvec++;
5997         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5998
5999         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6000         bio->bi_private = dip;
6001         dip->errors = 0;
6002         dip->orig_bio = bio;
6003         atomic_set(&dip->pending_bios, 0);
6004
6005         if (write)
6006                 bio->bi_end_io = btrfs_endio_direct_write;
6007         else
6008                 bio->bi_end_io = btrfs_endio_direct_read;
6009
6010         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6011         if (!ret)
6012                 return;
6013 free_ordered:
6014         /*
6015          * If this is a write, we need to clean up the reserved space and kill
6016          * the ordered extent.
6017          */
6018         if (write) {
6019                 struct btrfs_ordered_extent *ordered;
6020                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6021                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6022                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6023                         btrfs_free_reserved_extent(root, ordered->start,
6024                                                    ordered->disk_len);
6025                 btrfs_put_ordered_extent(ordered);
6026                 btrfs_put_ordered_extent(ordered);
6027         }
6028         bio_endio(bio, ret);
6029 }
6030
6031 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6032                         const struct iovec *iov, loff_t offset,
6033                         unsigned long nr_segs)
6034 {
6035         int seg;
6036         int i;
6037         size_t size;
6038         unsigned long addr;
6039         unsigned blocksize_mask = root->sectorsize - 1;
6040         ssize_t retval = -EINVAL;
6041         loff_t end = offset;
6042
6043         if (offset & blocksize_mask)
6044                 goto out;
6045
6046         /* Check the memory alignment.  Blocks cannot straddle pages */
6047         for (seg = 0; seg < nr_segs; seg++) {
6048                 addr = (unsigned long)iov[seg].iov_base;
6049                 size = iov[seg].iov_len;
6050                 end += size;
6051                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6052                         goto out;
6053
6054                 /* If this is a write we don't need to check anymore */
6055                 if (rw & WRITE)
6056                         continue;
6057
6058                 /*
6059                  * Check to make sure we don't have duplicate iov_base's in this
6060                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6061                  * when reading back.
6062                  */
6063                 for (i = seg + 1; i < nr_segs; i++) {
6064                         if (iov[seg].iov_base == iov[i].iov_base)
6065                                 goto out;
6066                 }
6067         }
6068         retval = 0;
6069 out:
6070         return retval;
6071 }
6072 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6073                         const struct iovec *iov, loff_t offset,
6074                         unsigned long nr_segs)
6075 {
6076         struct file *file = iocb->ki_filp;
6077         struct inode *inode = file->f_mapping->host;
6078         struct btrfs_ordered_extent *ordered;
6079         struct extent_state *cached_state = NULL;
6080         u64 lockstart, lockend;
6081         ssize_t ret;
6082         int writing = rw & WRITE;
6083         int write_bits = 0;
6084         size_t count = iov_length(iov, nr_segs);
6085
6086         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6087                             offset, nr_segs)) {
6088                 return 0;
6089         }
6090
6091         lockstart = offset;
6092         lockend = offset + count - 1;
6093
6094         if (writing) {
6095                 ret = btrfs_delalloc_reserve_space(inode, count);
6096                 if (ret)
6097                         goto out;
6098         }
6099
6100         while (1) {
6101                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6102                                  0, &cached_state, GFP_NOFS);
6103                 /*
6104                  * We're concerned with the entire range that we're going to be
6105                  * doing DIO to, so we need to make sure theres no ordered
6106                  * extents in this range.
6107                  */
6108                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6109                                                      lockend - lockstart + 1);
6110                 if (!ordered)
6111                         break;
6112                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6113                                      &cached_state, GFP_NOFS);
6114                 btrfs_start_ordered_extent(inode, ordered, 1);
6115                 btrfs_put_ordered_extent(ordered);
6116                 cond_resched();
6117         }
6118
6119         /*
6120          * we don't use btrfs_set_extent_delalloc because we don't want
6121          * the dirty or uptodate bits
6122          */
6123         if (writing) {
6124                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6125                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6126                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6127                                      GFP_NOFS);
6128                 if (ret) {
6129                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6130                                          lockend, EXTENT_LOCKED | write_bits,
6131                                          1, 0, &cached_state, GFP_NOFS);
6132                         goto out;
6133                 }
6134         }
6135
6136         free_extent_state(cached_state);
6137         cached_state = NULL;
6138
6139         ret = __blockdev_direct_IO(rw, iocb, inode,
6140                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6141                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6142                    btrfs_submit_direct, 0);
6143
6144         if (ret < 0 && ret != -EIOCBQUEUED) {
6145                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6146                               offset + iov_length(iov, nr_segs) - 1,
6147                               EXTENT_LOCKED | write_bits, 1, 0,
6148                               &cached_state, GFP_NOFS);
6149         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6150                 /*
6151                  * We're falling back to buffered, unlock the section we didn't
6152                  * do IO on.
6153                  */
6154                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6155                               offset + iov_length(iov, nr_segs) - 1,
6156                               EXTENT_LOCKED | write_bits, 1, 0,
6157                               &cached_state, GFP_NOFS);
6158         }
6159 out:
6160         free_extent_state(cached_state);
6161         return ret;
6162 }
6163
6164 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6165                 __u64 start, __u64 len)
6166 {
6167         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6168 }
6169
6170 int btrfs_readpage(struct file *file, struct page *page)
6171 {
6172         struct extent_io_tree *tree;
6173         tree = &BTRFS_I(page->mapping->host)->io_tree;
6174         return extent_read_full_page(tree, page, btrfs_get_extent);
6175 }
6176
6177 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6178 {
6179         struct extent_io_tree *tree;
6180
6181
6182         if (current->flags & PF_MEMALLOC) {
6183                 redirty_page_for_writepage(wbc, page);
6184                 unlock_page(page);
6185                 return 0;
6186         }
6187         tree = &BTRFS_I(page->mapping->host)->io_tree;
6188         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6189 }
6190
6191 int btrfs_writepages(struct address_space *mapping,
6192                      struct writeback_control *wbc)
6193 {
6194         struct extent_io_tree *tree;
6195
6196         tree = &BTRFS_I(mapping->host)->io_tree;
6197         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6198 }
6199
6200 static int
6201 btrfs_readpages(struct file *file, struct address_space *mapping,
6202                 struct list_head *pages, unsigned nr_pages)
6203 {
6204         struct extent_io_tree *tree;
6205         tree = &BTRFS_I(mapping->host)->io_tree;
6206         return extent_readpages(tree, mapping, pages, nr_pages,
6207                                 btrfs_get_extent);
6208 }
6209 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6210 {
6211         struct extent_io_tree *tree;
6212         struct extent_map_tree *map;
6213         int ret;
6214
6215         tree = &BTRFS_I(page->mapping->host)->io_tree;
6216         map = &BTRFS_I(page->mapping->host)->extent_tree;
6217         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6218         if (ret == 1) {
6219                 ClearPagePrivate(page);
6220                 set_page_private(page, 0);
6221                 page_cache_release(page);
6222         }
6223         return ret;
6224 }
6225
6226 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6227 {
6228         if (PageWriteback(page) || PageDirty(page))
6229                 return 0;
6230         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6231 }
6232
6233 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6234 {
6235         struct extent_io_tree *tree;
6236         struct btrfs_ordered_extent *ordered;
6237         struct extent_state *cached_state = NULL;
6238         u64 page_start = page_offset(page);
6239         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6240
6241
6242         /*
6243          * we have the page locked, so new writeback can't start,
6244          * and the dirty bit won't be cleared while we are here.
6245          *
6246          * Wait for IO on this page so that we can safely clear
6247          * the PagePrivate2 bit and do ordered accounting
6248          */
6249         wait_on_page_writeback(page);
6250
6251         tree = &BTRFS_I(page->mapping->host)->io_tree;
6252         if (offset) {
6253                 btrfs_releasepage(page, GFP_NOFS);
6254                 return;
6255         }
6256         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6257                          GFP_NOFS);
6258         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6259                                            page_offset(page));
6260         if (ordered) {
6261                 /*
6262                  * IO on this page will never be started, so we need
6263                  * to account for any ordered extents now
6264                  */
6265                 clear_extent_bit(tree, page_start, page_end,
6266                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6267                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6268                                  &cached_state, GFP_NOFS);
6269                 /*
6270                  * whoever cleared the private bit is responsible
6271                  * for the finish_ordered_io
6272                  */
6273                 if (TestClearPagePrivate2(page)) {
6274                         btrfs_finish_ordered_io(page->mapping->host,
6275                                                 page_start, page_end);
6276                 }
6277                 btrfs_put_ordered_extent(ordered);
6278                 cached_state = NULL;
6279                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6280                                  GFP_NOFS);
6281         }
6282         clear_extent_bit(tree, page_start, page_end,
6283                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6284                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6285         __btrfs_releasepage(page, GFP_NOFS);
6286
6287         ClearPageChecked(page);
6288         if (PagePrivate(page)) {
6289                 ClearPagePrivate(page);
6290                 set_page_private(page, 0);
6291                 page_cache_release(page);
6292         }
6293 }
6294
6295 /*
6296  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6297  * called from a page fault handler when a page is first dirtied. Hence we must
6298  * be careful to check for EOF conditions here. We set the page up correctly
6299  * for a written page which means we get ENOSPC checking when writing into
6300  * holes and correct delalloc and unwritten extent mapping on filesystems that
6301  * support these features.
6302  *
6303  * We are not allowed to take the i_mutex here so we have to play games to
6304  * protect against truncate races as the page could now be beyond EOF.  Because
6305  * vmtruncate() writes the inode size before removing pages, once we have the
6306  * page lock we can determine safely if the page is beyond EOF. If it is not
6307  * beyond EOF, then the page is guaranteed safe against truncation until we
6308  * unlock the page.
6309  */
6310 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6311 {
6312         struct page *page = vmf->page;
6313         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6314         struct btrfs_root *root = BTRFS_I(inode)->root;
6315         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6316         struct btrfs_ordered_extent *ordered;
6317         struct extent_state *cached_state = NULL;
6318         char *kaddr;
6319         unsigned long zero_start;
6320         loff_t size;
6321         int ret;
6322         u64 page_start;
6323         u64 page_end;
6324
6325         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6326         if (ret) {
6327                 if (ret == -ENOMEM)
6328                         ret = VM_FAULT_OOM;
6329                 else /* -ENOSPC, -EIO, etc */
6330                         ret = VM_FAULT_SIGBUS;
6331                 goto out;
6332         }
6333
6334         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6335 again:
6336         lock_page(page);
6337         size = i_size_read(inode);
6338         page_start = page_offset(page);
6339         page_end = page_start + PAGE_CACHE_SIZE - 1;
6340
6341         if ((page->mapping != inode->i_mapping) ||
6342             (page_start >= size)) {
6343                 /* page got truncated out from underneath us */
6344                 goto out_unlock;
6345         }
6346         wait_on_page_writeback(page);
6347
6348         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6349                          GFP_NOFS);
6350         set_page_extent_mapped(page);
6351
6352         /*
6353          * we can't set the delalloc bits if there are pending ordered
6354          * extents.  Drop our locks and wait for them to finish
6355          */
6356         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6357         if (ordered) {
6358                 unlock_extent_cached(io_tree, page_start, page_end,
6359                                      &cached_state, GFP_NOFS);
6360                 unlock_page(page);
6361                 btrfs_start_ordered_extent(inode, ordered, 1);
6362                 btrfs_put_ordered_extent(ordered);
6363                 goto again;
6364         }
6365
6366         /*
6367          * XXX - page_mkwrite gets called every time the page is dirtied, even
6368          * if it was already dirty, so for space accounting reasons we need to
6369          * clear any delalloc bits for the range we are fixing to save.  There
6370          * is probably a better way to do this, but for now keep consistent with
6371          * prepare_pages in the normal write path.
6372          */
6373         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6374                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6375                           0, 0, &cached_state, GFP_NOFS);
6376
6377         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6378                                         &cached_state);
6379         if (ret) {
6380                 unlock_extent_cached(io_tree, page_start, page_end,
6381                                      &cached_state, GFP_NOFS);
6382                 ret = VM_FAULT_SIGBUS;
6383                 goto out_unlock;
6384         }
6385         ret = 0;
6386
6387         /* page is wholly or partially inside EOF */
6388         if (page_start + PAGE_CACHE_SIZE > size)
6389                 zero_start = size & ~PAGE_CACHE_MASK;
6390         else
6391                 zero_start = PAGE_CACHE_SIZE;
6392
6393         if (zero_start != PAGE_CACHE_SIZE) {
6394                 kaddr = kmap(page);
6395                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6396                 flush_dcache_page(page);
6397                 kunmap(page);
6398         }
6399         ClearPageChecked(page);
6400         set_page_dirty(page);
6401         SetPageUptodate(page);
6402
6403         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6404         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6405
6406         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6407
6408 out_unlock:
6409         if (!ret)
6410                 return VM_FAULT_LOCKED;
6411         unlock_page(page);
6412         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6413 out:
6414         return ret;
6415 }
6416
6417 static int btrfs_truncate(struct inode *inode)
6418 {
6419         struct btrfs_root *root = BTRFS_I(inode)->root;
6420         int ret;
6421         int err = 0;
6422         struct btrfs_trans_handle *trans;
6423         unsigned long nr;
6424         u64 mask = root->sectorsize - 1;
6425
6426         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6427         if (ret)
6428                 return ret;
6429
6430         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6431         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6432
6433         trans = btrfs_start_transaction(root, 5);
6434         if (IS_ERR(trans))
6435                 return PTR_ERR(trans);
6436
6437         btrfs_set_trans_block_group(trans, inode);
6438
6439         ret = btrfs_orphan_add(trans, inode);
6440         if (ret) {
6441                 btrfs_end_transaction(trans, root);
6442                 return ret;
6443         }
6444
6445         nr = trans->blocks_used;
6446         btrfs_end_transaction(trans, root);
6447         btrfs_btree_balance_dirty(root, nr);
6448
6449         /* Now start a transaction for the truncate */
6450         trans = btrfs_start_transaction(root, 0);
6451         if (IS_ERR(trans))
6452                 return PTR_ERR(trans);
6453         btrfs_set_trans_block_group(trans, inode);
6454         trans->block_rsv = root->orphan_block_rsv;
6455
6456         /*
6457          * setattr is responsible for setting the ordered_data_close flag,
6458          * but that is only tested during the last file release.  That
6459          * could happen well after the next commit, leaving a great big
6460          * window where new writes may get lost if someone chooses to write
6461          * to this file after truncating to zero
6462          *
6463          * The inode doesn't have any dirty data here, and so if we commit
6464          * this is a noop.  If someone immediately starts writing to the inode
6465          * it is very likely we'll catch some of their writes in this
6466          * transaction, and the commit will find this file on the ordered
6467          * data list with good things to send down.
6468          *
6469          * This is a best effort solution, there is still a window where
6470          * using truncate to replace the contents of the file will
6471          * end up with a zero length file after a crash.
6472          */
6473         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6474                 btrfs_add_ordered_operation(trans, root, inode);
6475
6476         while (1) {
6477                 if (!trans) {
6478                         trans = btrfs_start_transaction(root, 0);
6479                         if (IS_ERR(trans))
6480                                 return PTR_ERR(trans);
6481                         btrfs_set_trans_block_group(trans, inode);
6482                         trans->block_rsv = root->orphan_block_rsv;
6483                 }
6484
6485                 ret = btrfs_block_rsv_check(trans, root,
6486                                             root->orphan_block_rsv, 0, 5);
6487                 if (ret == -EAGAIN) {
6488                         ret = btrfs_commit_transaction(trans, root);
6489                         if (ret)
6490                                 return ret;
6491                         trans = NULL;
6492                         continue;
6493                 } else if (ret) {
6494                         err = ret;
6495                         break;
6496                 }
6497
6498                 ret = btrfs_truncate_inode_items(trans, root, inode,
6499                                                  inode->i_size,
6500                                                  BTRFS_EXTENT_DATA_KEY);
6501                 if (ret != -EAGAIN) {
6502                         err = ret;
6503                         break;
6504                 }
6505
6506                 ret = btrfs_update_inode(trans, root, inode);
6507                 if (ret) {
6508                         err = ret;
6509                         break;
6510                 }
6511
6512                 nr = trans->blocks_used;
6513                 btrfs_end_transaction(trans, root);
6514                 trans = NULL;
6515                 btrfs_btree_balance_dirty(root, nr);
6516         }
6517
6518         if (ret == 0 && inode->i_nlink > 0) {
6519                 ret = btrfs_orphan_del(trans, inode);
6520                 if (ret)
6521                         err = ret;
6522         } else if (ret && inode->i_nlink > 0) {
6523                 /*
6524                  * Failed to do the truncate, remove us from the in memory
6525                  * orphan list.
6526                  */
6527                 ret = btrfs_orphan_del(NULL, inode);
6528         }
6529
6530         ret = btrfs_update_inode(trans, root, inode);
6531         if (ret && !err)
6532                 err = ret;
6533
6534         nr = trans->blocks_used;
6535         ret = btrfs_end_transaction_throttle(trans, root);
6536         if (ret && !err)
6537                 err = ret;
6538         btrfs_btree_balance_dirty(root, nr);
6539
6540         return err;
6541 }
6542
6543 /*
6544  * create a new subvolume directory/inode (helper for the ioctl).
6545  */
6546 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6547                              struct btrfs_root *new_root,
6548                              u64 new_dirid, u64 alloc_hint)
6549 {
6550         struct inode *inode;
6551         int err;
6552         u64 index = 0;
6553
6554         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6555                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6556         if (IS_ERR(inode))
6557                 return PTR_ERR(inode);
6558         inode->i_op = &btrfs_dir_inode_operations;
6559         inode->i_fop = &btrfs_dir_file_operations;
6560
6561         inode->i_nlink = 1;
6562         btrfs_i_size_write(inode, 0);
6563
6564         err = btrfs_update_inode(trans, new_root, inode);
6565         BUG_ON(err);
6566
6567         iput(inode);
6568         return 0;
6569 }
6570
6571 /* helper function for file defrag and space balancing.  This
6572  * forces readahead on a given range of bytes in an inode
6573  */
6574 unsigned long btrfs_force_ra(struct address_space *mapping,
6575                               struct file_ra_state *ra, struct file *file,
6576                               pgoff_t offset, pgoff_t last_index)
6577 {
6578         pgoff_t req_size = last_index - offset + 1;
6579
6580         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6581         return offset + req_size;
6582 }
6583
6584 struct inode *btrfs_alloc_inode(struct super_block *sb)
6585 {
6586         struct btrfs_inode *ei;
6587         struct inode *inode;
6588
6589         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6590         if (!ei)
6591                 return NULL;
6592
6593         ei->root = NULL;
6594         ei->space_info = NULL;
6595         ei->generation = 0;
6596         ei->sequence = 0;
6597         ei->last_trans = 0;
6598         ei->last_sub_trans = 0;
6599         ei->logged_trans = 0;
6600         ei->delalloc_bytes = 0;
6601         ei->reserved_bytes = 0;
6602         ei->disk_i_size = 0;
6603         ei->flags = 0;
6604         ei->index_cnt = (u64)-1;
6605         ei->last_unlink_trans = 0;
6606
6607         atomic_set(&ei->outstanding_extents, 0);
6608         atomic_set(&ei->reserved_extents, 0);
6609
6610         ei->ordered_data_close = 0;
6611         ei->orphan_meta_reserved = 0;
6612         ei->dummy_inode = 0;
6613         ei->force_compress = BTRFS_COMPRESS_NONE;
6614
6615         inode = &ei->vfs_inode;
6616         extent_map_tree_init(&ei->extent_tree);
6617         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6618         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6619         mutex_init(&ei->log_mutex);
6620         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6621         INIT_LIST_HEAD(&ei->i_orphan);
6622         INIT_LIST_HEAD(&ei->delalloc_inodes);
6623         INIT_LIST_HEAD(&ei->ordered_operations);
6624         RB_CLEAR_NODE(&ei->rb_node);
6625
6626         return inode;
6627 }
6628
6629 static void btrfs_i_callback(struct rcu_head *head)
6630 {
6631         struct inode *inode = container_of(head, struct inode, i_rcu);
6632         INIT_LIST_HEAD(&inode->i_dentry);
6633         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6634 }
6635
6636 void btrfs_destroy_inode(struct inode *inode)
6637 {
6638         struct btrfs_ordered_extent *ordered;
6639         struct btrfs_root *root = BTRFS_I(inode)->root;
6640
6641         WARN_ON(!list_empty(&inode->i_dentry));
6642         WARN_ON(inode->i_data.nrpages);
6643         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6644         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6645
6646         /*
6647          * This can happen where we create an inode, but somebody else also
6648          * created the same inode and we need to destroy the one we already
6649          * created.
6650          */
6651         if (!root)
6652                 goto free;
6653
6654         /*
6655          * Make sure we're properly removed from the ordered operation
6656          * lists.
6657          */
6658         smp_mb();
6659         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6660                 spin_lock(&root->fs_info->ordered_extent_lock);
6661                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6662                 spin_unlock(&root->fs_info->ordered_extent_lock);
6663         }
6664
6665         if (root == root->fs_info->tree_root) {
6666                 struct btrfs_block_group_cache *block_group;
6667
6668                 block_group = btrfs_lookup_block_group(root->fs_info,
6669                                                 BTRFS_I(inode)->block_group);
6670                 if (block_group && block_group->inode == inode) {
6671                         spin_lock(&block_group->lock);
6672                         block_group->inode = NULL;
6673                         spin_unlock(&block_group->lock);
6674                         btrfs_put_block_group(block_group);
6675                 } else if (block_group) {
6676                         btrfs_put_block_group(block_group);
6677                 }
6678         }
6679
6680         spin_lock(&root->orphan_lock);
6681         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6682                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6683                        inode->i_ino);
6684                 list_del_init(&BTRFS_I(inode)->i_orphan);
6685         }
6686         spin_unlock(&root->orphan_lock);
6687
6688         while (1) {
6689                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6690                 if (!ordered)
6691                         break;
6692                 else {
6693                         printk(KERN_ERR "btrfs found ordered "
6694                                "extent %llu %llu on inode cleanup\n",
6695                                (unsigned long long)ordered->file_offset,
6696                                (unsigned long long)ordered->len);
6697                         btrfs_remove_ordered_extent(inode, ordered);
6698                         btrfs_put_ordered_extent(ordered);
6699                         btrfs_put_ordered_extent(ordered);
6700                 }
6701         }
6702         inode_tree_del(inode);
6703         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6704 free:
6705         call_rcu(&inode->i_rcu, btrfs_i_callback);
6706 }
6707
6708 int btrfs_drop_inode(struct inode *inode)
6709 {
6710         struct btrfs_root *root = BTRFS_I(inode)->root;
6711
6712         if (btrfs_root_refs(&root->root_item) == 0 &&
6713             root != root->fs_info->tree_root)
6714                 return 1;
6715         else
6716                 return generic_drop_inode(inode);
6717 }
6718
6719 static void init_once(void *foo)
6720 {
6721         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6722
6723         inode_init_once(&ei->vfs_inode);
6724 }
6725
6726 void btrfs_destroy_cachep(void)
6727 {
6728         if (btrfs_inode_cachep)
6729                 kmem_cache_destroy(btrfs_inode_cachep);
6730         if (btrfs_trans_handle_cachep)
6731                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6732         if (btrfs_transaction_cachep)
6733                 kmem_cache_destroy(btrfs_transaction_cachep);
6734         if (btrfs_path_cachep)
6735                 kmem_cache_destroy(btrfs_path_cachep);
6736         if (btrfs_free_space_cachep)
6737                 kmem_cache_destroy(btrfs_free_space_cachep);
6738 }
6739
6740 int btrfs_init_cachep(void)
6741 {
6742         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6743                         sizeof(struct btrfs_inode), 0,
6744                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6745         if (!btrfs_inode_cachep)
6746                 goto fail;
6747
6748         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6749                         sizeof(struct btrfs_trans_handle), 0,
6750                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6751         if (!btrfs_trans_handle_cachep)
6752                 goto fail;
6753
6754         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6755                         sizeof(struct btrfs_transaction), 0,
6756                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6757         if (!btrfs_transaction_cachep)
6758                 goto fail;
6759
6760         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6761                         sizeof(struct btrfs_path), 0,
6762                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6763         if (!btrfs_path_cachep)
6764                 goto fail;
6765
6766         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6767                         sizeof(struct btrfs_free_space), 0,
6768                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6769         if (!btrfs_free_space_cachep)
6770                 goto fail;
6771
6772         return 0;
6773 fail:
6774         btrfs_destroy_cachep();
6775         return -ENOMEM;
6776 }
6777
6778 static int btrfs_getattr(struct vfsmount *mnt,
6779                          struct dentry *dentry, struct kstat *stat)
6780 {
6781         struct inode *inode = dentry->d_inode;
6782         generic_fillattr(inode, stat);
6783         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6784         stat->blksize = PAGE_CACHE_SIZE;
6785         stat->blocks = (inode_get_bytes(inode) +
6786                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6787         return 0;
6788 }
6789
6790 /*
6791  * If a file is moved, it will inherit the cow and compression flags of the new
6792  * directory.
6793  */
6794 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6795 {
6796         struct btrfs_inode *b_dir = BTRFS_I(dir);
6797         struct btrfs_inode *b_inode = BTRFS_I(inode);
6798
6799         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6800                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6801         else
6802                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6803
6804         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6805                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6806         else
6807                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6808 }
6809
6810 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6811                            struct inode *new_dir, struct dentry *new_dentry)
6812 {
6813         struct btrfs_trans_handle *trans;
6814         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6815         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6816         struct inode *newinode = new_dentry->d_inode;
6817         struct inode *old_inode = old_dentry->d_inode;
6818         struct timespec ctime = CURRENT_TIME;
6819         u64 index = 0;
6820         u64 root_objectid;
6821         int ret;
6822
6823         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6824                 return -EPERM;
6825
6826         /* we only allow rename subvolume link between subvolumes */
6827         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6828                 return -EXDEV;
6829
6830         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6831             (newinode && newinode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6832                 return -ENOTEMPTY;
6833
6834         if (S_ISDIR(old_inode->i_mode) && newinode &&
6835             newinode->i_size > BTRFS_EMPTY_DIR_SIZE)
6836                 return -ENOTEMPTY;
6837         /*
6838          * we're using rename to replace one file with another.
6839          * and the replacement file is large.  Start IO on it now so
6840          * we don't add too much work to the end of the transaction
6841          */
6842         if (newinode && S_ISREG(old_inode->i_mode) && newinode->i_size &&
6843             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6844                 filemap_flush(old_inode->i_mapping);
6845
6846         /* close the racy window with snapshot create/destroy ioctl */
6847         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6848                 down_read(&root->fs_info->subvol_sem);
6849         /*
6850          * We want to reserve the absolute worst case amount of items.  So if
6851          * both inodes are subvols and we need to unlink them then that would
6852          * require 4 item modifications, but if they are both normal inodes it
6853          * would require 5 item modifications, so we'll assume their normal
6854          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6855          * should cover the worst case number of items we'll modify.
6856          */
6857         trans = btrfs_start_transaction(root, 20);
6858         if (IS_ERR(trans)) {
6859                 ret = PTR_ERR(trans);
6860                 goto out_notrans;
6861         }
6862
6863         btrfs_set_trans_block_group(trans, new_dir);
6864
6865         if (dest != root)
6866                 btrfs_record_root_in_trans(trans, dest);
6867
6868         ret = btrfs_set_inode_index(new_dir, &index);
6869         if (ret)
6870                 goto out_fail;
6871
6872         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6873                 /* force full log commit if subvolume involved. */
6874                 root->fs_info->last_trans_log_full_commit = trans->transid;
6875         } else {
6876                 ret = btrfs_insert_inode_ref(trans, dest,
6877                                              new_dentry->d_name.name,
6878                                              new_dentry->d_name.len,
6879                                              old_inode->i_ino,
6880                                              new_dir->i_ino, index);
6881                 if (ret)
6882                         goto out_fail;
6883                 /*
6884                  * this is an ugly little race, but the rename is required
6885                  * to make sure that if we crash, the inode is either at the
6886                  * old name or the new one.  pinning the log transaction lets
6887                  * us make sure we don't allow a log commit to come in after
6888                  * we unlink the name but before we add the new name back in.
6889                  */
6890                 btrfs_pin_log_trans(root);
6891         }
6892         /*
6893          * make sure the inode gets flushed if it is replacing
6894          * something.
6895          */
6896         if (newinode && newinode->i_size &&
6897             old_inode && S_ISREG(old_inode->i_mode)) {
6898                 btrfs_add_ordered_operation(trans, root, old_inode);
6899         }
6900
6901         old_dir->i_ctime = old_dir->i_mtime = ctime;
6902         new_dir->i_ctime = new_dir->i_mtime = ctime;
6903         old_inode->i_ctime = ctime;
6904
6905         if (old_dentry->d_parent != new_dentry->d_parent)
6906                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6907
6908         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6909                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6910                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6911                                         old_dentry->d_name.name,
6912                                         old_dentry->d_name.len);
6913         } else {
6914                 ret = __btrfs_unlink_inode(trans, root, old_dir,
6915                                         old_dentry->d_inode,
6916                                         old_dentry->d_name.name,
6917                                         old_dentry->d_name.len);
6918                 if (!ret)
6919                         ret = btrfs_update_inode(trans, root, old_inode);
6920         }
6921         BUG_ON(ret);
6922
6923         if (newinode) {
6924                 newinode->i_ctime = CURRENT_TIME;
6925                 if (unlikely(newinode->i_ino ==
6926                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6927                         root_objectid = BTRFS_I(newinode)->location.objectid;
6928                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6929                                                 root_objectid,
6930                                                 new_dentry->d_name.name,
6931                                                 new_dentry->d_name.len);
6932                         BUG_ON(newinode->i_nlink == 0);
6933                 } else {
6934                         ret = btrfs_unlink_inode(trans, dest, new_dir,
6935                                                  new_dentry->d_inode,
6936                                                  new_dentry->d_name.name,
6937                                                  new_dentry->d_name.len);
6938                 }
6939                 BUG_ON(ret);
6940                 if (newinode->i_nlink == 0) {
6941                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6942                         BUG_ON(ret);
6943                 }
6944         }
6945
6946         fixup_inode_flags(new_dir, old_inode);
6947
6948         ret = btrfs_add_link(trans, new_dir, old_inode,
6949                              new_dentry->d_name.name,
6950                              new_dentry->d_name.len, 0, index);
6951         BUG_ON(ret);
6952
6953         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6954                 struct dentry *parent = dget_parent(new_dentry);
6955                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6956                 dput(parent);
6957                 btrfs_end_log_trans(root);
6958         }
6959 out_fail:
6960         btrfs_end_transaction_throttle(trans, root);
6961 out_notrans:
6962         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6963                 up_read(&root->fs_info->subvol_sem);
6964
6965         return ret;
6966 }
6967
6968 /*
6969  * some fairly slow code that needs optimization. This walks the list
6970  * of all the inodes with pending delalloc and forces them to disk.
6971  */
6972 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6973 {
6974         struct list_head *head = &root->fs_info->delalloc_inodes;
6975         struct btrfs_inode *binode;
6976         struct inode *inode;
6977
6978         if (root->fs_info->sb->s_flags & MS_RDONLY)
6979                 return -EROFS;
6980
6981         spin_lock(&root->fs_info->delalloc_lock);
6982         while (!list_empty(head)) {
6983                 binode = list_entry(head->next, struct btrfs_inode,
6984                                     delalloc_inodes);
6985                 inode = igrab(&binode->vfs_inode);
6986                 if (!inode)
6987                         list_del_init(&binode->delalloc_inodes);
6988                 spin_unlock(&root->fs_info->delalloc_lock);
6989                 if (inode) {
6990                         filemap_flush(inode->i_mapping);
6991                         if (delay_iput)
6992                                 btrfs_add_delayed_iput(inode);
6993                         else
6994                                 iput(inode);
6995                 }
6996                 cond_resched();
6997                 spin_lock(&root->fs_info->delalloc_lock);
6998         }
6999         spin_unlock(&root->fs_info->delalloc_lock);
7000
7001         /* the filemap_flush will queue IO into the worker threads, but
7002          * we have to make sure the IO is actually started and that
7003          * ordered extents get created before we return
7004          */
7005         atomic_inc(&root->fs_info->async_submit_draining);
7006         while (atomic_read(&root->fs_info->nr_async_submits) ||
7007               atomic_read(&root->fs_info->async_delalloc_pages)) {
7008                 wait_event(root->fs_info->async_submit_wait,
7009                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7010                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7011         }
7012         atomic_dec(&root->fs_info->async_submit_draining);
7013         return 0;
7014 }
7015
7016 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7017                          const char *symname)
7018 {
7019         struct btrfs_trans_handle *trans;
7020         struct btrfs_root *root = BTRFS_I(dir)->root;
7021         struct btrfs_path *path;
7022         struct btrfs_key key;
7023         struct inode *inode = NULL;
7024         int err;
7025         int drop_inode = 0;
7026         u64 objectid;
7027         u64 index = 0 ;
7028         int name_len;
7029         int datasize;
7030         unsigned long ptr;
7031         struct btrfs_file_extent_item *ei;
7032         struct extent_buffer *leaf;
7033         unsigned long nr = 0;
7034
7035         name_len = strlen(symname) + 1;
7036         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7037                 return -ENAMETOOLONG;
7038
7039         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7040         if (err)
7041                 return err;
7042         /*
7043          * 2 items for inode item and ref
7044          * 2 items for dir items
7045          * 1 item for xattr if selinux is on
7046          */
7047         trans = btrfs_start_transaction(root, 5);
7048         if (IS_ERR(trans))
7049                 return PTR_ERR(trans);
7050
7051         btrfs_set_trans_block_group(trans, dir);
7052
7053         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7054                                 dentry->d_name.len, dir->i_ino, objectid,
7055                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7056                                 &index);
7057         if (IS_ERR(inode)) {
7058                 err = PTR_ERR(inode);
7059                 goto out_unlock;
7060         }
7061
7062         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7063         if (err) {
7064                 drop_inode = 1;
7065                 goto out_unlock;
7066         }
7067
7068         btrfs_set_trans_block_group(trans, inode);
7069         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7070         if (err)
7071                 drop_inode = 1;
7072         else {
7073                 inode->i_mapping->a_ops = &btrfs_aops;
7074                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7075                 inode->i_fop = &btrfs_file_operations;
7076                 inode->i_op = &btrfs_file_inode_operations;
7077                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7078         }
7079         btrfs_update_inode_block_group(trans, inode);
7080         btrfs_update_inode_block_group(trans, dir);
7081         if (drop_inode)
7082                 goto out_unlock;
7083
7084         path = btrfs_alloc_path();
7085         BUG_ON(!path);
7086         key.objectid = inode->i_ino;
7087         key.offset = 0;
7088         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7089         datasize = btrfs_file_extent_calc_inline_size(name_len);
7090         err = btrfs_insert_empty_item(trans, root, path, &key,
7091                                       datasize);
7092         if (err) {
7093                 drop_inode = 1;
7094                 goto out_unlock;
7095         }
7096         leaf = path->nodes[0];
7097         ei = btrfs_item_ptr(leaf, path->slots[0],
7098                             struct btrfs_file_extent_item);
7099         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7100         btrfs_set_file_extent_type(leaf, ei,
7101                                    BTRFS_FILE_EXTENT_INLINE);
7102         btrfs_set_file_extent_encryption(leaf, ei, 0);
7103         btrfs_set_file_extent_compression(leaf, ei, 0);
7104         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7105         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7106
7107         ptr = btrfs_file_extent_inline_start(ei);
7108         write_extent_buffer(leaf, symname, ptr, name_len);
7109         btrfs_mark_buffer_dirty(leaf);
7110         btrfs_free_path(path);
7111
7112         inode->i_op = &btrfs_symlink_inode_operations;
7113         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7114         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7115         inode_set_bytes(inode, name_len);
7116         btrfs_i_size_write(inode, name_len - 1);
7117         err = btrfs_update_inode(trans, root, inode);
7118         if (err)
7119                 drop_inode = 1;
7120
7121 out_unlock:
7122         nr = trans->blocks_used;
7123         btrfs_end_transaction_throttle(trans, root);
7124         if (drop_inode) {
7125                 inode_dec_link_count(inode);
7126                 iput(inode);
7127         }
7128         btrfs_btree_balance_dirty(root, nr);
7129         return err;
7130 }
7131
7132 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7133                                        u64 start, u64 num_bytes, u64 min_size,
7134                                        loff_t actual_len, u64 *alloc_hint,
7135                                        struct btrfs_trans_handle *trans)
7136 {
7137         struct btrfs_root *root = BTRFS_I(inode)->root;
7138         struct btrfs_key ins;
7139         u64 cur_offset = start;
7140         u64 i_size;
7141         int ret = 0;
7142         bool own_trans = true;
7143
7144         if (trans)
7145                 own_trans = false;
7146         while (num_bytes > 0) {
7147                 if (own_trans) {
7148                         trans = btrfs_start_transaction(root, 3);
7149                         if (IS_ERR(trans)) {
7150                                 ret = PTR_ERR(trans);
7151                                 break;
7152                         }
7153                 }
7154
7155                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7156                                            0, *alloc_hint, (u64)-1, &ins, 1);
7157                 if (ret) {
7158                         if (own_trans)
7159                                 btrfs_end_transaction(trans, root);
7160                         break;
7161                 }
7162
7163                 ret = insert_reserved_file_extent(trans, inode,
7164                                                   cur_offset, ins.objectid,
7165                                                   ins.offset, ins.offset,
7166                                                   ins.offset, 0, 0, 0,
7167                                                   BTRFS_FILE_EXTENT_PREALLOC);
7168                 BUG_ON(ret);
7169                 btrfs_drop_extent_cache(inode, cur_offset,
7170                                         cur_offset + ins.offset -1, 0);
7171
7172                 num_bytes -= ins.offset;
7173                 cur_offset += ins.offset;
7174                 *alloc_hint = ins.objectid + ins.offset;
7175
7176                 inode->i_ctime = CURRENT_TIME;
7177                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7178                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7179                     (actual_len > inode->i_size) &&
7180                     (cur_offset > inode->i_size)) {
7181                         if (cur_offset > actual_len)
7182                                 i_size = actual_len;
7183                         else
7184                                 i_size = cur_offset;
7185                         i_size_write(inode, i_size);
7186                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7187                 }
7188
7189                 ret = btrfs_update_inode(trans, root, inode);
7190                 BUG_ON(ret);
7191
7192                 if (own_trans)
7193                         btrfs_end_transaction(trans, root);
7194         }
7195         return ret;
7196 }
7197
7198 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7199                               u64 start, u64 num_bytes, u64 min_size,
7200                               loff_t actual_len, u64 *alloc_hint)
7201 {
7202         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7203                                            min_size, actual_len, alloc_hint,
7204                                            NULL);
7205 }
7206
7207 int btrfs_prealloc_file_range_trans(struct inode *inode,
7208                                     struct btrfs_trans_handle *trans, int mode,
7209                                     u64 start, u64 num_bytes, u64 min_size,
7210                                     loff_t actual_len, u64 *alloc_hint)
7211 {
7212         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7213                                            min_size, actual_len, alloc_hint, trans);
7214 }
7215
7216 static int btrfs_set_page_dirty(struct page *page)
7217 {
7218         return __set_page_dirty_nobuffers(page);
7219 }
7220
7221 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7222 {
7223         struct btrfs_root *root = BTRFS_I(inode)->root;
7224
7225         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7226                 return -EROFS;
7227         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7228                 return -EACCES;
7229         return generic_permission(inode, mask, flags, btrfs_check_acl);
7230 }
7231
7232 static const struct inode_operations btrfs_dir_inode_operations = {
7233         .getattr        = btrfs_getattr,
7234         .lookup         = btrfs_lookup,
7235         .create         = btrfs_create,
7236         .unlink         = btrfs_unlink,
7237         .link           = btrfs_link,
7238         .mkdir          = btrfs_mkdir,
7239         .rmdir          = btrfs_rmdir,
7240         .rename         = btrfs_rename,
7241         .symlink        = btrfs_symlink,
7242         .setattr        = btrfs_setattr,
7243         .mknod          = btrfs_mknod,
7244         .setxattr       = btrfs_setxattr,
7245         .getxattr       = btrfs_getxattr,
7246         .listxattr      = btrfs_listxattr,
7247         .removexattr    = btrfs_removexattr,
7248         .permission     = btrfs_permission,
7249 };
7250 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7251         .lookup         = btrfs_lookup,
7252         .permission     = btrfs_permission,
7253 };
7254
7255 static const struct file_operations btrfs_dir_file_operations = {
7256         .llseek         = generic_file_llseek,
7257         .read           = generic_read_dir,
7258         .readdir        = btrfs_real_readdir,
7259         .unlocked_ioctl = btrfs_ioctl,
7260 #ifdef CONFIG_COMPAT
7261         .compat_ioctl   = btrfs_ioctl,
7262 #endif
7263         .release        = btrfs_release_file,
7264         .fsync          = btrfs_sync_file,
7265 };
7266
7267 static struct extent_io_ops btrfs_extent_io_ops = {
7268         .fill_delalloc = run_delalloc_range,
7269         .submit_bio_hook = btrfs_submit_bio_hook,
7270         .merge_bio_hook = btrfs_merge_bio_hook,
7271         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7272         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7273         .writepage_start_hook = btrfs_writepage_start_hook,
7274         .readpage_io_failed_hook = btrfs_io_failed_hook,
7275         .set_bit_hook = btrfs_set_bit_hook,
7276         .clear_bit_hook = btrfs_clear_bit_hook,
7277         .merge_extent_hook = btrfs_merge_extent_hook,
7278         .split_extent_hook = btrfs_split_extent_hook,
7279 };
7280
7281 /*
7282  * btrfs doesn't support the bmap operation because swapfiles
7283  * use bmap to make a mapping of extents in the file.  They assume
7284  * these extents won't change over the life of the file and they
7285  * use the bmap result to do IO directly to the drive.
7286  *
7287  * the btrfs bmap call would return logical addresses that aren't
7288  * suitable for IO and they also will change frequently as COW
7289  * operations happen.  So, swapfile + btrfs == corruption.
7290  *
7291  * For now we're avoiding this by dropping bmap.
7292  */
7293 static const struct address_space_operations btrfs_aops = {
7294         .readpage       = btrfs_readpage,
7295         .writepage      = btrfs_writepage,
7296         .writepages     = btrfs_writepages,
7297         .readpages      = btrfs_readpages,
7298         .direct_IO      = btrfs_direct_IO,
7299         .invalidatepage = btrfs_invalidatepage,
7300         .releasepage    = btrfs_releasepage,
7301         .set_page_dirty = btrfs_set_page_dirty,
7302         .error_remove_page = generic_error_remove_page,
7303 };
7304
7305 static const struct address_space_operations btrfs_symlink_aops = {
7306         .readpage       = btrfs_readpage,
7307         .writepage      = btrfs_writepage,
7308         .invalidatepage = btrfs_invalidatepage,
7309         .releasepage    = btrfs_releasepage,
7310 };
7311
7312 static const struct inode_operations btrfs_file_inode_operations = {
7313         .getattr        = btrfs_getattr,
7314         .setattr        = btrfs_setattr,
7315         .setxattr       = btrfs_setxattr,
7316         .getxattr       = btrfs_getxattr,
7317         .listxattr      = btrfs_listxattr,
7318         .removexattr    = btrfs_removexattr,
7319         .permission     = btrfs_permission,
7320         .fiemap         = btrfs_fiemap,
7321 };
7322 static const struct inode_operations btrfs_special_inode_operations = {
7323         .getattr        = btrfs_getattr,
7324         .setattr        = btrfs_setattr,
7325         .permission     = btrfs_permission,
7326         .setxattr       = btrfs_setxattr,
7327         .getxattr       = btrfs_getxattr,
7328         .listxattr      = btrfs_listxattr,
7329         .removexattr    = btrfs_removexattr,
7330 };
7331 static const struct inode_operations btrfs_symlink_inode_operations = {
7332         .readlink       = generic_readlink,
7333         .follow_link    = page_follow_link_light,
7334         .put_link       = page_put_link,
7335         .getattr        = btrfs_getattr,
7336         .permission     = btrfs_permission,
7337         .setxattr       = btrfs_setxattr,
7338         .getxattr       = btrfs_getxattr,
7339         .listxattr      = btrfs_listxattr,
7340         .removexattr    = btrfs_removexattr,
7341 };
7342
7343 const struct dentry_operations btrfs_dentry_operations = {
7344         .d_delete       = btrfs_dentry_delete,
7345 };