fs/f2fs/data.c

   1 /*
   2  * fs/f2fs/data.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/mpage.h>
  15 #include <linux/writeback.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21 #include <linux/uio.h>
  22 #include <linux/mm.h>
  23 #include <linux/memcontrol.h>
  24 #include <linux/cleancache.h>
  25
  26 #include "f2fs.h"
  27 #include "node.h"
  28 #include "segment.h"
  29 #include "trace.h"
  30 #include <trace/events/f2fs.h>
  31
  32 static void f2fs_read_end_io(struct bio *bio)
  33 {
  34         struct bio_vec *bvec;
  35         int i;
  36
  37 #ifdef CONFIG_F2FS_FAULT_INJECTION
  38         if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO))
  39                 bio->bi_error = -EIO;
  40 #endif
  41
  42         if (f2fs_bio_encrypted(bio)) {
  43                 if (bio->bi_error) {
  44                         fscrypt_release_ctx(bio->bi_private);
  45                 } else {
  46                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  47                         return;
  48                 }
  49         }
  50
  51         bio_for_each_segment_all(bvec, bio, i) {
  52                 struct page *page = bvec->bv_page;
  53
  54                 if (!bio->bi_error) {
  55                         if (!PageUptodate(page))
  56                                 SetPageUptodate(page);
  57                 } else {
  58                         ClearPageUptodate(page);
  59                         SetPageError(page);
  60                 }
  61                 unlock_page(page);
  62         }
  63         bio_put(bio);
  64 }
  65
  66 static void f2fs_write_end_io(struct bio *bio)
  67 {
  68         struct f2fs_sb_info *sbi = bio->bi_private;
  69         struct bio_vec *bvec;
  70         int i;
  71
  72         bio_for_each_segment_all(bvec, bio, i) {
  73                 struct page *page = bvec->bv_page;
  74
  75                 fscrypt_pullback_bio_page(&page, true);
  76
  77                 if (unlikely(bio->bi_error)) {
  78                         mapping_set_error(page->mapping, -EIO);
  79                         f2fs_stop_checkpoint(sbi, true);
  80                 }
  81                 end_page_writeback(page);
  82         }
  83         if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
  84                                 wq_has_sleeper(&sbi->cp_wait))
  85                 wake_up(&sbi->cp_wait);
  86
  87         bio_put(bio);
  88 }
  89
  90 /*
  91  * Low-level block read/write IO operations.
  92  */
  93 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  94                                 int npages, bool is_read)
  95 {
  96         struct bio *bio;
  97
  98         bio = f2fs_bio_alloc(npages);
  99
 100         bio->bi_bdev = sbi->sb->s_bdev;
 101         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 102         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
 103         bio->bi_private = is_read ? NULL : sbi;
 104
 105         return bio;
 106 }
 107
 108 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 109                                 struct bio *bio, enum page_type type)
 110 {
 111         if (!is_read_io(bio_op(bio))) {
 112                 atomic_inc(&sbi->nr_wb_bios);
 113                 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
 114                         current->plug && (type == DATA || type == NODE))
 115                         blk_finish_plug(current->plug);
 116         }
 117         submit_bio(bio);
 118 }
 119
 120 static void __submit_merged_bio(struct f2fs_bio_info *io)
 121 {
 122         struct f2fs_io_info *fio = &io->fio;
 123
 124         if (!io->bio)
 125                 return;
 126
 127         if (is_read_io(fio->op))
 128                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 129         else
 130                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 131
 132         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 133
 134         __submit_bio(io->sbi, io->bio, fio->type);
 135         io->bio = NULL;
 136 }
 137
 138 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 139                                                 struct page *page, nid_t ino)
 140 {
 141         struct bio_vec *bvec;
 142         struct page *target;
 143         int i;
 144
 145         if (!io->bio)
 146                 return false;
 147
 148         if (!inode && !page && !ino)
 149                 return true;
 150
 151         bio_for_each_segment_all(bvec, io->bio, i) {
 152
 153                 if (bvec->bv_page->mapping)
 154                         target = bvec->bv_page;
 155                 else
 156                         target = fscrypt_control_page(bvec->bv_page);
 157
 158                 if (inode && inode == target->mapping->host)
 159                         return true;
 160                 if (page && page == target)
 161                         return true;
 162                 if (ino && ino == ino_of_node(target))
 163                         return true;
 164         }
 165
 166         return false;
 167 }
 168
 169 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 170                                                 struct page *page, nid_t ino,
 171                                                 enum page_type type)
 172 {
 173         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 174         struct f2fs_bio_info *io = &sbi->write_io[btype];
 175         bool ret;
 176
 177         down_read(&io->io_rwsem);
 178         ret = __has_merged_page(io, inode, page, ino);
 179         up_read(&io->io_rwsem);
 180         return ret;
 181 }
 182
 183 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 184                                 struct inode *inode, struct page *page,
 185                                 nid_t ino, enum page_type type, int rw)
 186 {
 187         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 188         struct f2fs_bio_info *io;
 189
 190         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 191
 192         down_write(&io->io_rwsem);
 193
 194         if (!__has_merged_page(io, inode, page, ino))
 195                 goto out;
 196
 197         /* change META to META_FLUSH in the checkpoint procedure */
 198         if (type >= META_FLUSH) {
 199                 io->fio.type = META_FLUSH;
 200                 io->fio.op = REQ_OP_WRITE;
 201                 io->fio.op_flags = REQ_PREFLUSH | REQ_META | REQ_PRIO;
 202                 if (!test_opt(sbi, NOBARRIER))
 203                         io->fio.op_flags |= REQ_FUA;
 204         }
 205         __submit_merged_bio(io);
 206 out:
 207         up_write(&io->io_rwsem);
 208 }
 209
 210 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 211                                                                         int rw)
 212 {
 213         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 214 }
 215
 216 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 217                                 struct inode *inode, struct page *page,
 218                                 nid_t ino, enum page_type type, int rw)
 219 {
 220         if (has_merged_page(sbi, inode, page, ino, type))
 221                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 222 }
 223
 224 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 225 {
 226         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 227         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 228         f2fs_submit_merged_bio(sbi, META, WRITE);
 229 }
 230
 231 /*
 232  * Fill the locked page with data located in the block address.
 233  * Return unlocked page.
 234  */
 235 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 236 {
 237         struct bio *bio;
 238         struct page *page = fio->encrypted_page ?
 239                         fio->encrypted_page : fio->page;
 240
 241         trace_f2fs_submit_page_bio(page, fio);
 242         f2fs_trace_ios(fio, 0);
 243
 244         /* Allocate a new bio */
 245         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
 246
 247         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 248                 bio_put(bio);
 249                 return -EFAULT;
 250         }
 251         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 252
 253         __submit_bio(fio->sbi, bio, fio->type);
 254         return 0;
 255 }
 256
 257 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 258 {
 259         struct f2fs_sb_info *sbi = fio->sbi;
 260         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 261         struct f2fs_bio_info *io;
 262         bool is_read = is_read_io(fio->op);
 263         struct page *bio_page;
 264
 265         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 266
 267         if (fio->old_blkaddr != NEW_ADDR)
 268                 verify_block_addr(sbi, fio->old_blkaddr);
 269         verify_block_addr(sbi, fio->new_blkaddr);
 270
 271         down_write(&io->io_rwsem);
 272
 273         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 274             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags)))
 275                 __submit_merged_bio(io);
 276 alloc_new:
 277         if (io->bio == NULL) {
 278                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 279
 280                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 281                                                 bio_blocks, is_read);
 282                 io->fio = *fio;
 283         }
 284
 285         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 286
 287         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 288                                                         PAGE_SIZE) {
 289                 __submit_merged_bio(io);
 290                 goto alloc_new;
 291         }
 292
 293         io->last_block_in_bio = fio->new_blkaddr;
 294         f2fs_trace_ios(fio, 0);
 295
 296         up_write(&io->io_rwsem);
 297         trace_f2fs_submit_page_mbio(fio->page, fio);
 298 }
 299
 300 static void __set_data_blkaddr(struct dnode_of_data *dn)
 301 {
 302         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 303         __le32 *addr_array;
 304
 305         /* Get physical address of data block */
 306         addr_array = blkaddr_in_node(rn);
 307         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 308 }
 309
 310 /*
 311  * Lock ordering for the change of data block address:
 312  * ->data_page
 313  *  ->node_page
 314  *    update block addresses in the node page
 315  */
 316 void set_data_blkaddr(struct dnode_of_data *dn)
 317 {
 318         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 319         __set_data_blkaddr(dn);
 320         if (set_page_dirty(dn->node_page))
 321                 dn->node_changed = true;
 322 }
 323
 324 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 325 {
 326         dn->data_blkaddr = blkaddr;
 327         set_data_blkaddr(dn);
 328         f2fs_update_extent_cache(dn);
 329 }
 330
 331 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 332 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 333 {
 334         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 335
 336         if (!count)
 337                 return 0;
 338
 339         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 340                 return -EPERM;
 341         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 342                 return -ENOSPC;
 343
 344         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 345                                                 dn->ofs_in_node, count);
 346
 347         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 348
 349         for (; count > 0; dn->ofs_in_node++) {
 350                 block_t blkaddr =
 351                         datablock_addr(dn->node_page, dn->ofs_in_node);
 352                 if (blkaddr == NULL_ADDR) {
 353                         dn->data_blkaddr = NEW_ADDR;
 354                         __set_data_blkaddr(dn);
 355                         count--;
 356                 }
 357         }
 358
 359         if (set_page_dirty(dn->node_page))
 360                 dn->node_changed = true;
 361         return 0;
 362 }
 363
 364 /* Should keep dn->ofs_in_node unchanged */
 365 int reserve_new_block(struct dnode_of_data *dn)
 366 {
 367         unsigned int ofs_in_node = dn->ofs_in_node;
 368         int ret;
 369
 370         ret = reserve_new_blocks(dn, 1);
 371         dn->ofs_in_node = ofs_in_node;
 372         return ret;
 373 }
 374
 375 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 376 {
 377         bool need_put = dn->inode_page ? false : true;
 378         int err;
 379
 380         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 381         if (err)
 382                 return err;
 383
 384         if (dn->data_blkaddr == NULL_ADDR)
 385                 err = reserve_new_block(dn);
 386         if (err || need_put)
 387                 f2fs_put_dnode(dn);
 388         return err;
 389 }
 390
 391 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 392 {
 393         struct extent_info ei;
 394         struct inode *inode = dn->inode;
 395
 396         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 397                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 398                 return 0;
 399         }
 400
 401         return f2fs_reserve_block(dn, index);
 402 }
 403
 404 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 405                                                 int op_flags, bool for_write)
 406 {
 407         struct address_space *mapping = inode->i_mapping;
 408         struct dnode_of_data dn;
 409         struct page *page;
 410         struct extent_info ei;
 411         int err;
 412         struct f2fs_io_info fio = {
 413                 .sbi = F2FS_I_SB(inode),
 414                 .type = DATA,
 415                 .op = REQ_OP_READ,
 416                 .op_flags = op_flags,
 417                 .encrypted_page = NULL,
 418         };
 419
 420         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 421                 return read_mapping_page(mapping, index, NULL);
 422
 423         page = f2fs_grab_cache_page(mapping, index, for_write);
 424         if (!page)
 425                 return ERR_PTR(-ENOMEM);
 426
 427         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 428                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 429                 goto got_it;
 430         }
 431
 432         set_new_dnode(&dn, inode, NULL, NULL, 0);
 433         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 434         if (err)
 435                 goto put_err;
 436         f2fs_put_dnode(&dn);
 437
 438         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 439                 err = -ENOENT;
 440                 goto put_err;
 441         }
 442 got_it:
 443         if (PageUptodate(page)) {
 444                 unlock_page(page);
 445                 return page;
 446         }
 447
 448         /*
 449          * A new dentry page is allocated but not able to be written, since its
 450          * new inode page couldn't be allocated due to -ENOSPC.
 451          * In such the case, its blkaddr can be remained as NEW_ADDR.
 452          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 453          */
 454         if (dn.data_blkaddr == NEW_ADDR) {
 455                 zero_user_segment(page, 0, PAGE_SIZE);
 456                 if (!PageUptodate(page))
 457                         SetPageUptodate(page);
 458                 unlock_page(page);
 459                 return page;
 460         }
 461
 462         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 463         fio.page = page;
 464         err = f2fs_submit_page_bio(&fio);
 465         if (err)
 466                 goto put_err;
 467         return page;
 468
 469 put_err:
 470         f2fs_put_page(page, 1);
 471         return ERR_PTR(err);
 472 }
 473
 474 struct page *find_data_page(struct inode *inode, pgoff_t index)
 475 {
 476         struct address_space *mapping = inode->i_mapping;
 477         struct page *page;
 478
 479         page = find_get_page(mapping, index);
 480         if (page && PageUptodate(page))
 481                 return page;
 482         f2fs_put_page(page, 0);
 483
 484         page = get_read_data_page(inode, index, 0, false);
 485         if (IS_ERR(page))
 486                 return page;
 487
 488         if (PageUptodate(page))
 489                 return page;
 490
 491         wait_on_page_locked(page);
 492         if (unlikely(!PageUptodate(page))) {
 493                 f2fs_put_page(page, 0);
 494                 return ERR_PTR(-EIO);
 495         }
 496         return page;
 497 }
 498
 499 /*
 500  * If it tries to access a hole, return an error.
 501  * Because, the callers, functions in dir.c and GC, should be able to know
 502  * whether this page exists or not.
 503  */
 504 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 505                                                         bool for_write)
 506 {
 507         struct address_space *mapping = inode->i_mapping;
 508         struct page *page;
 509 repeat:
 510         page = get_read_data_page(inode, index, 0, for_write);
 511         if (IS_ERR(page))
 512                 return page;
 513
 514         /* wait for read completion */
 515         lock_page(page);
 516         if (unlikely(page->mapping != mapping)) {
 517                 f2fs_put_page(page, 1);
 518                 goto repeat;
 519         }
 520         if (unlikely(!PageUptodate(page))) {
 521                 f2fs_put_page(page, 1);
 522                 return ERR_PTR(-EIO);
 523         }
 524         return page;
 525 }
 526
 527 /*
 528  * Caller ensures that this data page is never allocated.
 529  * A new zero-filled data page is allocated in the page cache.
 530  *
 531  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 532  * f2fs_unlock_op().
 533  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 534  * ipage should be released by this function.
 535  */
 536 struct page *get_new_data_page(struct inode *inode,
 537                 struct page *ipage, pgoff_t index, bool new_i_size)
 538 {
 539         struct address_space *mapping = inode->i_mapping;
 540         struct page *page;
 541         struct dnode_of_data dn;
 542         int err;
 543
 544         page = f2fs_grab_cache_page(mapping, index, true);
 545         if (!page) {
 546                 /*
 547                  * before exiting, we should make sure ipage will be released
 548                  * if any error occur.
 549                  */
 550                 f2fs_put_page(ipage, 1);
 551                 return ERR_PTR(-ENOMEM);
 552         }
 553
 554         set_new_dnode(&dn, inode, ipage, NULL, 0);
 555         err = f2fs_reserve_block(&dn, index);
 556         if (err) {
 557                 f2fs_put_page(page, 1);
 558                 return ERR_PTR(err);
 559         }
 560         if (!ipage)
 561                 f2fs_put_dnode(&dn);
 562
 563         if (PageUptodate(page))
 564                 goto got_it;
 565
 566         if (dn.data_blkaddr == NEW_ADDR) {
 567                 zero_user_segment(page, 0, PAGE_SIZE);
 568                 if (!PageUptodate(page))
 569                         SetPageUptodate(page);
 570         } else {
 571                 f2fs_put_page(page, 1);
 572
 573                 /* if ipage exists, blkaddr should be NEW_ADDR */
 574                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 575                 page = get_lock_data_page(inode, index, true);
 576                 if (IS_ERR(page))
 577                         return page;
 578         }
 579 got_it:
 580         if (new_i_size && i_size_read(inode) <
 581                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 582                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 583         return page;
 584 }
 585
 586 static int __allocate_data_block(struct dnode_of_data *dn)
 587 {
 588         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 589         struct f2fs_summary sum;
 590         struct node_info ni;
 591         int seg = CURSEG_WARM_DATA;
 592         pgoff_t fofs;
 593         blkcnt_t count = 1;
 594
 595         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 596                 return -EPERM;
 597
 598         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 599         if (dn->data_blkaddr == NEW_ADDR)
 600                 goto alloc;
 601
 602         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 603                 return -ENOSPC;
 604
 605 alloc:
 606         get_node_info(sbi, dn->nid, &ni);
 607         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 608
 609         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 610                 seg = CURSEG_DIRECT_IO;
 611
 612         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 613                                                                 &sum, seg);
 614         set_data_blkaddr(dn);
 615
 616         /* update i_size */
 617         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 618                                                         dn->ofs_in_node;
 619         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 620                 f2fs_i_size_write(dn->inode,
 621                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 622         return 0;
 623 }
 624
 625 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 626 {
 627         struct inode *inode = file_inode(iocb->ki_filp);
 628         struct f2fs_map_blocks map;
 629         ssize_t ret = 0;
 630
 631         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 632         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 633         if (map.m_len > map.m_lblk)
 634                 map.m_len -= map.m_lblk;
 635         else
 636                 map.m_len = 0;
 637
 638         map.m_next_pgofs = NULL;
 639
 640         if (iocb->ki_flags & IOCB_DIRECT) {
 641                 ret = f2fs_convert_inline_inode(inode);
 642                 if (ret)
 643                         return ret;
 644                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 645         }
 646         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 647                 ret = f2fs_convert_inline_inode(inode);
 648                 if (ret)
 649                         return ret;
 650         }
 651         if (!f2fs_has_inline_data(inode))
 652                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 653         return ret;
 654 }
 655
 656 /*
 657  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 658  * f2fs_map_blocks structure.
 659  * If original data blocks are allocated, then give them to blockdev.
 660  * Otherwise,
 661  *     a. preallocate requested block addresses
 662  *     b. do not use extent cache for better performance
 663  *     c. give the block addresses to blockdev
 664  */
 665 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 666                                                 int create, int flag)
 667 {
 668         unsigned int maxblocks = map->m_len;
 669         struct dnode_of_data dn;
 670         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 671         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 672         pgoff_t pgofs, end_offset, end;
 673         int err = 0, ofs = 1;
 674         unsigned int ofs_in_node, last_ofs_in_node;
 675         blkcnt_t prealloc;
 676         struct extent_info ei;
 677         bool allocated = false;
 678         block_t blkaddr;
 679
 680         if (!maxblocks)
 681                 return 0;
 682
 683         map->m_len = 0;
 684         map->m_flags = 0;
 685
 686         /* it only supports block size == page size */
 687         pgofs = (pgoff_t)map->m_lblk;
 688         end = pgofs + maxblocks;
 689
 690         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 691                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 692                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 693                 map->m_flags = F2FS_MAP_MAPPED;
 694                 goto out;
 695         }
 696
 697 next_dnode:
 698         if (create)
 699                 f2fs_lock_op(sbi);
 700
 701         /* When reading holes, we need its node page */
 702         set_new_dnode(&dn, inode, NULL, NULL, 0);
 703         err = get_dnode_of_data(&dn, pgofs, mode);
 704         if (err) {
 705                 if (flag == F2FS_GET_BLOCK_BMAP)
 706                         map->m_pblk = 0;
 707                 if (err == -ENOENT) {
 708                         err = 0;
 709                         if (map->m_next_pgofs)
 710                                 *map->m_next_pgofs =
 711                                         get_next_page_offset(&dn, pgofs);
 712                 }
 713                 goto unlock_out;
 714         }
 715
 716         prealloc = 0;
 717         ofs_in_node = dn.ofs_in_node;
 718         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 719
 720 next_block:
 721         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 722
 723         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 724                 if (create) {
 725                         if (unlikely(f2fs_cp_error(sbi))) {
 726                                 err = -EIO;
 727                                 goto sync_out;
 728                         }
 729                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 730                                 if (blkaddr == NULL_ADDR) {
 731                                         prealloc++;
 732                                         last_ofs_in_node = dn.ofs_in_node;
 733                                 }
 734                         } else {
 735                                 err = __allocate_data_block(&dn);
 736                                 if (!err) {
 737                                         set_inode_flag(inode, FI_APPEND_WRITE);
 738                                         allocated = true;
 739                                 }
 740                         }
 741                         if (err)
 742                                 goto sync_out;
 743                         map->m_flags = F2FS_MAP_NEW;
 744                         blkaddr = dn.data_blkaddr;
 745                 } else {
 746                         if (flag == F2FS_GET_BLOCK_BMAP) {
 747                                 map->m_pblk = 0;
 748                                 goto sync_out;
 749                         }
 750                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 751                                                 blkaddr == NULL_ADDR) {
 752                                 if (map->m_next_pgofs)
 753                                         *map->m_next_pgofs = pgofs + 1;
 754                         }
 755                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 756                                                 blkaddr != NEW_ADDR)
 757                                 goto sync_out;
 758                 }
 759         }
 760
 761         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 762                 goto skip;
 763
 764         if (map->m_len == 0) {
 765                 /* preallocated unwritten block should be mapped for fiemap. */
 766                 if (blkaddr == NEW_ADDR)
 767                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 768                 map->m_flags |= F2FS_MAP_MAPPED;
 769
 770                 map->m_pblk = blkaddr;
 771                 map->m_len = 1;
 772         } else if ((map->m_pblk != NEW_ADDR &&
 773                         blkaddr == (map->m_pblk + ofs)) ||
 774                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 775                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 776                 ofs++;
 777                 map->m_len++;
 778         } else {
 779                 goto sync_out;
 780         }
 781
 782 skip:
 783         dn.ofs_in_node++;
 784         pgofs++;
 785
 786         /* preallocate blocks in batch for one dnode page */
 787         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 788                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 789
 790                 dn.ofs_in_node = ofs_in_node;
 791                 err = reserve_new_blocks(&dn, prealloc);
 792                 if (err)
 793                         goto sync_out;
 794                 allocated = dn.node_changed;
 795
 796                 map->m_len += dn.ofs_in_node - ofs_in_node;
 797                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 798                         err = -ENOSPC;
 799                         goto sync_out;
 800                 }
 801                 dn.ofs_in_node = end_offset;
 802         }
 803
 804         if (pgofs >= end)
 805                 goto sync_out;
 806         else if (dn.ofs_in_node < end_offset)
 807                 goto next_block;
 808
 809         f2fs_put_dnode(&dn);
 810
 811         if (create) {
 812                 f2fs_unlock_op(sbi);
 813                 f2fs_balance_fs(sbi, allocated);
 814         }
 815         allocated = false;
 816         goto next_dnode;
 817
 818 sync_out:
 819         f2fs_put_dnode(&dn);
 820 unlock_out:
 821         if (create) {
 822                 f2fs_unlock_op(sbi);
 823                 f2fs_balance_fs(sbi, allocated);
 824         }
 825 out:
 826         trace_f2fs_map_blocks(inode, map, err);
 827         return err;
 828 }
 829
 830 static int __get_data_block(struct inode *inode, sector_t iblock,
 831                         struct buffer_head *bh, int create, int flag,
 832                         pgoff_t *next_pgofs)
 833 {
 834         struct f2fs_map_blocks map;
 835         int ret;
 836
 837         map.m_lblk = iblock;
 838         map.m_len = bh->b_size >> inode->i_blkbits;
 839         map.m_next_pgofs = next_pgofs;
 840
 841         ret = f2fs_map_blocks(inode, &map, create, flag);
 842         if (!ret) {
 843                 map_bh(bh, inode->i_sb, map.m_pblk);
 844                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 845                 bh->b_size = map.m_len << inode->i_blkbits;
 846         }
 847         return ret;
 848 }
 849
 850 static int get_data_block(struct inode *inode, sector_t iblock,
 851                         struct buffer_head *bh_result, int create, int flag,
 852                         pgoff_t *next_pgofs)
 853 {
 854         return __get_data_block(inode, iblock, bh_result, create,
 855                                                         flag, next_pgofs);
 856 }
 857
 858 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 859                         struct buffer_head *bh_result, int create)
 860 {
 861         return __get_data_block(inode, iblock, bh_result, create,
 862                                                 F2FS_GET_BLOCK_DIO, NULL);
 863 }
 864
 865 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 866                         struct buffer_head *bh_result, int create)
 867 {
 868         /* Block number less than F2FS MAX BLOCKS */
 869         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 870                 return -EFBIG;
 871
 872         return __get_data_block(inode, iblock, bh_result, create,
 873                                                 F2FS_GET_BLOCK_BMAP, NULL);
 874 }
 875
 876 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 877 {
 878         return (offset >> inode->i_blkbits);
 879 }
 880
 881 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 882 {
 883         return (blk << inode->i_blkbits);
 884 }
 885
 886 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 887                 u64 start, u64 len)
 888 {
 889         struct buffer_head map_bh;
 890         sector_t start_blk, last_blk;
 891         pgoff_t next_pgofs;
 892         loff_t isize;
 893         u64 logical = 0, phys = 0, size = 0;
 894         u32 flags = 0;
 895         int ret = 0;
 896
 897         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 898         if (ret)
 899                 return ret;
 900
 901         if (f2fs_has_inline_data(inode)) {
 902                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 903                 if (ret != -EAGAIN)
 904                         return ret;
 905         }
 906
 907         inode_lock(inode);
 908
 909         isize = i_size_read(inode);
 910         if (start >= isize)
 911                 goto out;
 912
 913         if (start + len > isize)
 914                 len = isize - start;
 915
 916         if (logical_to_blk(inode, len) == 0)
 917                 len = blk_to_logical(inode, 1);
 918
 919         start_blk = logical_to_blk(inode, start);
 920         last_blk = logical_to_blk(inode, start + len - 1);
 921
 922 next:
 923         memset(&map_bh, 0, sizeof(struct buffer_head));
 924         map_bh.b_size = len;
 925
 926         ret = get_data_block(inode, start_blk, &map_bh, 0,
 927                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 928         if (ret)
 929                 goto out;
 930
 931         /* HOLE */
 932         if (!buffer_mapped(&map_bh)) {
 933                 start_blk = next_pgofs;
 934                 /* Go through holes util pass the EOF */
 935                 if (blk_to_logical(inode, start_blk) < isize)
 936                         goto prep_next;
 937                 /* Found a hole beyond isize means no more extents.
 938                  * Note that the premise is that filesystems don't
 939                  * punch holes beyond isize and keep size unchanged.
 940                  */
 941                 flags |= FIEMAP_EXTENT_LAST;
 942         }
 943
 944         if (size) {
 945                 if (f2fs_encrypted_inode(inode))
 946                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 947
 948                 ret = fiemap_fill_next_extent(fieinfo, logical,
 949                                 phys, size, flags);
 950         }
 951
 952         if (start_blk > last_blk || ret)
 953                 goto out;
 954
 955         logical = blk_to_logical(inode, start_blk);
 956         phys = blk_to_logical(inode, map_bh.b_blocknr);
 957         size = map_bh.b_size;
 958         flags = 0;
 959         if (buffer_unwritten(&map_bh))
 960                 flags = FIEMAP_EXTENT_UNWRITTEN;
 961
 962         start_blk += logical_to_blk(inode, size);
 963
 964 prep_next:
 965         cond_resched();
 966         if (fatal_signal_pending(current))
 967                 ret = -EINTR;
 968         else
 969                 goto next;
 970 out:
 971         if (ret == 1)
 972                 ret = 0;
 973
 974         inode_unlock(inode);
 975         return ret;
 976 }
 977
 978 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
 979                                  unsigned nr_pages)
 980 {
 981         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 982         struct fscrypt_ctx *ctx = NULL;
 983         struct block_device *bdev = sbi->sb->s_bdev;
 984         struct bio *bio;
 985
 986         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
 987                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
 988                 if (IS_ERR(ctx))
 989                         return ERR_CAST(ctx);
 990
 991                 /* wait the page to be moved by cleaning */
 992                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
 993         }
 994
 995         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
 996         if (!bio) {
 997                 if (ctx)
 998                         fscrypt_release_ctx(ctx);
 999                 return ERR_PTR(-ENOMEM);
1000         }
1001         bio->bi_bdev = bdev;
1002         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
1003         bio->bi_end_io = f2fs_read_end_io;
1004         bio->bi_private = ctx;
1005
1006         return bio;
1007 }
1008
1009 /*
1010  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1011  * Major change was from block_size == page_size in f2fs by default.
1012  */
1013 static int f2fs_mpage_readpages(struct address_space *mapping,
1014                         struct list_head *pages, struct page *page,
1015                         unsigned nr_pages)
1016 {
1017         struct bio *bio = NULL;
1018         unsigned page_idx;
1019         sector_t last_block_in_bio = 0;
1020         struct inode *inode = mapping->host;
1021         const unsigned blkbits = inode->i_blkbits;
1022         const unsigned blocksize = 1 << blkbits;
1023         sector_t block_in_file;
1024         sector_t last_block;
1025         sector_t last_block_in_file;
1026         sector_t block_nr;
1027         struct f2fs_map_blocks map;
1028
1029         map.m_pblk = 0;
1030         map.m_lblk = 0;
1031         map.m_len = 0;
1032         map.m_flags = 0;
1033         map.m_next_pgofs = NULL;
1034
1035         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1036
1037                 prefetchw(&page->flags);
1038                 if (pages) {
1039                         page = list_entry(pages->prev, struct page, lru);
1040                         list_del(&page->lru);
1041                         if (add_to_page_cache_lru(page, mapping,
1042                                                   page->index,
1043                                                   readahead_gfp_mask(mapping)))
1044                                 goto next_page;
1045                 }
1046
1047                 block_in_file = (sector_t)page->index;
1048                 last_block = block_in_file + nr_pages;
1049                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1050                                                                 blkbits;
1051                 if (last_block > last_block_in_file)
1052                         last_block = last_block_in_file;
1053
1054                 /*
1055                  * Map blocks using the previous result first.
1056                  */
1057                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1058                                 block_in_file > map.m_lblk &&
1059                                 block_in_file < (map.m_lblk + map.m_len))
1060                         goto got_it;
1061
1062                 /*
1063                  * Then do more f2fs_map_blocks() calls until we are
1064                  * done with this page.
1065                  */
1066                 map.m_flags = 0;
1067
1068                 if (block_in_file < last_block) {
1069                         map.m_lblk = block_in_file;
1070                         map.m_len = last_block - block_in_file;
1071
1072                         if (f2fs_map_blocks(inode, &map, 0,
1073                                                 F2FS_GET_BLOCK_READ))
1074                                 goto set_error_page;
1075                 }
1076 got_it:
1077                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1078                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1079                         SetPageMappedToDisk(page);
1080
1081                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1082                                 SetPageUptodate(page);
1083                                 goto confused;
1084                         }
1085                 } else {
1086                         zero_user_segment(page, 0, PAGE_SIZE);
1087                         if (!PageUptodate(page))
1088                                 SetPageUptodate(page);
1089                         unlock_page(page);
1090                         goto next_page;
1091                 }
1092
1093                 /*
1094                  * This page will go to BIO.  Do we need to send this
1095                  * BIO off first?
1096                  */
1097                 if (bio && (last_block_in_bio != block_nr - 1)) {
1098 submit_and_realloc:
1099                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1100                         bio = NULL;
1101                 }
1102                 if (bio == NULL) {
1103                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1104                         if (IS_ERR(bio)) {
1105                                 bio = NULL;
1106                                 goto set_error_page;
1107                         }
1108                         bio_set_op_attrs(bio, REQ_OP_READ, 0);
1109                 }
1110
1111                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1112                         goto submit_and_realloc;
1113
1114                 last_block_in_bio = block_nr;
1115                 goto next_page;
1116 set_error_page:
1117                 SetPageError(page);
1118                 zero_user_segment(page, 0, PAGE_SIZE);
1119                 unlock_page(page);
1120                 goto next_page;
1121 confused:
1122                 if (bio) {
1123                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1124                         bio = NULL;
1125                 }
1126                 unlock_page(page);
1127 next_page:
1128                 if (pages)
1129                         put_page(page);
1130         }
1131         BUG_ON(pages && !list_empty(pages));
1132         if (bio)
1133                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1134         return 0;
1135 }
1136
1137 static int f2fs_read_data_page(struct file *file, struct page *page)
1138 {
1139         struct inode *inode = page->mapping->host;
1140         int ret = -EAGAIN;
1141
1142         trace_f2fs_readpage(page, DATA);
1143
1144         /* If the file has inline data, try to read it directly */
1145         if (f2fs_has_inline_data(inode))
1146                 ret = f2fs_read_inline_data(inode, page);
1147         if (ret == -EAGAIN)
1148                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1149         return ret;
1150 }
1151
1152 static int f2fs_read_data_pages(struct file *file,
1153                         struct address_space *mapping,
1154                         struct list_head *pages, unsigned nr_pages)
1155 {
1156         struct inode *inode = file->f_mapping->host;
1157         struct page *page = list_entry(pages->prev, struct page, lru);
1158
1159         trace_f2fs_readpages(inode, page, nr_pages);
1160
1161         /* If the file has inline data, skip readpages */
1162         if (f2fs_has_inline_data(inode))
1163                 return 0;
1164
1165         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1166 }
1167
1168 int do_write_data_page(struct f2fs_io_info *fio)
1169 {
1170         struct page *page = fio->page;
1171         struct inode *inode = page->mapping->host;
1172         struct dnode_of_data dn;
1173         int err = 0;
1174
1175         set_new_dnode(&dn, inode, NULL, NULL, 0);
1176         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1177         if (err)
1178                 return err;
1179
1180         fio->old_blkaddr = dn.data_blkaddr;
1181
1182         /* This page is already truncated */
1183         if (fio->old_blkaddr == NULL_ADDR) {
1184                 ClearPageUptodate(page);
1185                 goto out_writepage;
1186         }
1187
1188         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1189                 gfp_t gfp_flags = GFP_NOFS;
1190
1191                 /* wait for GCed encrypted page writeback */
1192                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1193                                                         fio->old_blkaddr);
1194 retry_encrypt:
1195                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1196                                                                 gfp_flags);
1197                 if (IS_ERR(fio->encrypted_page)) {
1198                         err = PTR_ERR(fio->encrypted_page);
1199                         if (err == -ENOMEM) {
1200                                 /* flush pending ios and wait for a while */
1201                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1202                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1203                                 gfp_flags |= __GFP_NOFAIL;
1204                                 err = 0;
1205                                 goto retry_encrypt;
1206                         }
1207                         goto out_writepage;
1208                 }
1209         }
1210
1211         set_page_writeback(page);
1212
1213         /*
1214          * If current allocation needs SSR,
1215          * it had better in-place writes for updated data.
1216          */
1217         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1218                         !is_cold_data(page) &&
1219                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1220                         need_inplace_update(inode))) {
1221                 rewrite_data_page(fio);
1222                 set_inode_flag(inode, FI_UPDATE_WRITE);
1223                 trace_f2fs_do_write_data_page(page, IPU);
1224         } else {
1225                 write_data_page(&dn, fio);
1226                 trace_f2fs_do_write_data_page(page, OPU);
1227                 set_inode_flag(inode, FI_APPEND_WRITE);
1228                 if (page->index == 0)
1229                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1230         }
1231 out_writepage:
1232         f2fs_put_dnode(&dn);
1233         return err;
1234 }
1235
1236 static int f2fs_write_data_page(struct page *page,
1237                                         struct writeback_control *wbc)
1238 {
1239         struct inode *inode = page->mapping->host;
1240         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1241         loff_t i_size = i_size_read(inode);
1242         const pgoff_t end_index = ((unsigned long long) i_size)
1243                                                         >> PAGE_SHIFT;
1244         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1245         unsigned offset = 0;
1246         bool need_balance_fs = false;
1247         int err = 0;
1248         struct f2fs_io_info fio = {
1249                 .sbi = sbi,
1250                 .type = DATA,
1251                 .op = REQ_OP_WRITE,
1252                 .op_flags = wbc_to_write_flags(wbc),
1253                 .page = page,
1254                 .encrypted_page = NULL,
1255         };
1256
1257         trace_f2fs_writepage(page, DATA);
1258
1259         if (page->index < end_index)
1260                 goto write;
1261
1262         /*
1263          * If the offset is out-of-range of file size,
1264          * this page does not have to be written to disk.
1265          */
1266         offset = i_size & (PAGE_SIZE - 1);
1267         if ((page->index >= end_index + 1) || !offset)
1268                 goto out;
1269
1270         zero_user_segment(page, offset, PAGE_SIZE);
1271 write:
1272         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1273                 goto redirty_out;
1274         if (f2fs_is_drop_cache(inode))
1275                 goto out;
1276         /* we should not write 0'th page having journal header */
1277         if (f2fs_is_volatile_file(inode) && (!page->index ||
1278                         (!wbc->for_reclaim &&
1279                         available_free_memory(sbi, BASE_CHECK))))
1280                 goto redirty_out;
1281
1282         /* we should bypass data pages to proceed the kworkder jobs */
1283         if (unlikely(f2fs_cp_error(sbi))) {
1284                 mapping_set_error(page->mapping, -EIO);
1285                 goto out;
1286         }
1287
1288         /* Dentry blocks are controlled by checkpoint */
1289         if (S_ISDIR(inode->i_mode)) {
1290                 err = do_write_data_page(&fio);
1291                 goto done;
1292         }
1293
1294         if (!wbc->for_reclaim)
1295                 need_balance_fs = true;
1296         else if (has_not_enough_free_secs(sbi, 0, 0))
1297                 goto redirty_out;
1298
1299         err = -EAGAIN;
1300         f2fs_lock_op(sbi);
1301         if (f2fs_has_inline_data(inode))
1302                 err = f2fs_write_inline_data(inode, page);
1303         if (err == -EAGAIN)
1304                 err = do_write_data_page(&fio);
1305         if (F2FS_I(inode)->last_disk_size < psize)
1306                 F2FS_I(inode)->last_disk_size = psize;
1307         f2fs_unlock_op(sbi);
1308 done:
1309         if (err && err != -ENOENT)
1310                 goto redirty_out;
1311
1312         clear_cold_data(page);
1313 out:
1314         inode_dec_dirty_pages(inode);
1315         if (err)
1316                 ClearPageUptodate(page);
1317
1318         if (wbc->for_reclaim) {
1319                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1320                 remove_dirty_inode(inode);
1321         }
1322
1323         unlock_page(page);
1324         f2fs_balance_fs(sbi, need_balance_fs);
1325
1326         if (unlikely(f2fs_cp_error(sbi)))
1327                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1328
1329         return 0;
1330
1331 redirty_out:
1332         redirty_page_for_writepage(wbc, page);
1333         unlock_page(page);
1334         return err;
1335 }
1336
1337 /*
1338  * This function was copied from write_cche_pages from mm/page-writeback.c.
1339  * The major change is making write step of cold data page separately from
1340  * warm/hot data page.
1341  */
1342 static int f2fs_write_cache_pages(struct address_space *mapping,
1343                                         struct writeback_control *wbc)
1344 {
1345         int ret = 0;
1346         int done = 0;
1347         struct pagevec pvec;
1348         int nr_pages;
1349         pgoff_t uninitialized_var(writeback_index);
1350         pgoff_t index;
1351         pgoff_t end;            /* Inclusive */
1352         pgoff_t done_index;
1353         int cycled;
1354         int range_whole = 0;
1355         int tag;
1356         int nwritten = 0;
1357
1358         pagevec_init(&pvec, 0);
1359
1360         if (wbc->range_cyclic) {
1361                 writeback_index = mapping->writeback_index; /* prev offset */
1362                 index = writeback_index;
1363                 if (index == 0)
1364                         cycled = 1;
1365                 else
1366                         cycled = 0;
1367                 end = -1;
1368         } else {
1369                 index = wbc->range_start >> PAGE_SHIFT;
1370                 end = wbc->range_end >> PAGE_SHIFT;
1371                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1372                         range_whole = 1;
1373                 cycled = 1; /* ignore range_cyclic tests */
1374         }
1375         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1376                 tag = PAGECACHE_TAG_TOWRITE;
1377         else
1378                 tag = PAGECACHE_TAG_DIRTY;
1379 retry:
1380         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1381                 tag_pages_for_writeback(mapping, index, end);
1382         done_index = index;
1383         while (!done && (index <= end)) {
1384                 int i;
1385
1386                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1387                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1388                 if (nr_pages == 0)
1389                         break;
1390
1391                 for (i = 0; i < nr_pages; i++) {
1392                         struct page *page = pvec.pages[i];
1393
1394                         if (page->index > end) {
1395                                 done = 1;
1396                                 break;
1397                         }
1398
1399                         done_index = page->index;
1400
1401                         lock_page(page);
1402
1403                         if (unlikely(page->mapping != mapping)) {
1404 continue_unlock:
1405                                 unlock_page(page);
1406                                 continue;
1407                         }
1408
1409                         if (!PageDirty(page)) {
1410                                 /* someone wrote it for us */
1411                                 goto continue_unlock;
1412                         }
1413
1414                         if (PageWriteback(page)) {
1415                                 if (wbc->sync_mode != WB_SYNC_NONE)
1416                                         f2fs_wait_on_page_writeback(page,
1417                                                                 DATA, true);
1418                                 else
1419                                         goto continue_unlock;
1420                         }
1421
1422                         BUG_ON(PageWriteback(page));
1423                         if (!clear_page_dirty_for_io(page))
1424                                 goto continue_unlock;
1425
1426                         ret = mapping->a_ops->writepage(page, wbc);
1427                         if (unlikely(ret)) {
1428                                 done_index = page->index + 1;
1429                                 done = 1;
1430                                 break;
1431                         } else {
1432                                 nwritten++;
1433                         }
1434
1435                         if (--wbc->nr_to_write <= 0 &&
1436                             wbc->sync_mode == WB_SYNC_NONE) {
1437                                 done = 1;
1438                                 break;
1439                         }
1440                 }
1441                 pagevec_release(&pvec);
1442                 cond_resched();
1443         }
1444
1445         if (!cycled && !done) {
1446                 cycled = 1;
1447                 index = 0;
1448                 end = writeback_index - 1;
1449                 goto retry;
1450         }
1451         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1452                 mapping->writeback_index = done_index;
1453
1454         if (nwritten)
1455                 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1456                                                         NULL, 0, DATA, WRITE);
1457
1458         return ret;
1459 }
1460
1461 static int f2fs_write_data_pages(struct address_space *mapping,
1462                             struct writeback_control *wbc)
1463 {
1464         struct inode *inode = mapping->host;
1465         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1466         struct blk_plug plug;
1467         int ret;
1468
1469         /* deal with chardevs and other special file */
1470         if (!mapping->a_ops->writepage)
1471                 return 0;
1472
1473         /* skip writing if there is no dirty page in this inode */
1474         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1475                 return 0;
1476
1477         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1478                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1479                         available_free_memory(sbi, DIRTY_DENTS))
1480                 goto skip_write;
1481
1482         /* skip writing during file defragment */
1483         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1484                 goto skip_write;
1485
1486         /* during POR, we don't need to trigger writepage at all. */
1487         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1488                 goto skip_write;
1489
1490         trace_f2fs_writepages(mapping->host, wbc, DATA);
1491
1492         blk_start_plug(&plug);
1493         ret = f2fs_write_cache_pages(mapping, wbc);
1494         blk_finish_plug(&plug);
1495         /*
1496          * if some pages were truncated, we cannot guarantee its mapping->host
1497          * to detect pending bios.
1498          */
1499
1500         remove_dirty_inode(inode);
1501         return ret;
1502
1503 skip_write:
1504         wbc->pages_skipped += get_dirty_pages(inode);
1505         trace_f2fs_writepages(mapping->host, wbc, DATA);
1506         return 0;
1507 }
1508
1509 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1510 {
1511         struct inode *inode = mapping->host;
1512         loff_t i_size = i_size_read(inode);
1513
1514         if (to > i_size) {
1515                 truncate_pagecache(inode, i_size);
1516                 truncate_blocks(inode, i_size, true);
1517         }
1518 }
1519
1520 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1521                         struct page *page, loff_t pos, unsigned len,
1522                         block_t *blk_addr, bool *node_changed)
1523 {
1524         struct inode *inode = page->mapping->host;
1525         pgoff_t index = page->index;
1526         struct dnode_of_data dn;
1527         struct page *ipage;
1528         bool locked = false;
1529         struct extent_info ei;
1530         int err = 0;
1531
1532         /*
1533          * we already allocated all the blocks, so we don't need to get
1534          * the block addresses when there is no need to fill the page.
1535          */
1536         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE)
1537                 return 0;
1538
1539         if (f2fs_has_inline_data(inode) ||
1540                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1541                 f2fs_lock_op(sbi);
1542                 locked = true;
1543         }
1544 restart:
1545         /* check inline_data */
1546         ipage = get_node_page(sbi, inode->i_ino);
1547         if (IS_ERR(ipage)) {
1548                 err = PTR_ERR(ipage);
1549                 goto unlock_out;
1550         }
1551
1552         set_new_dnode(&dn, inode, ipage, ipage, 0);
1553
1554         if (f2fs_has_inline_data(inode)) {
1555                 if (pos + len <= MAX_INLINE_DATA) {
1556                         read_inline_data(page, ipage);
1557                         set_inode_flag(inode, FI_DATA_EXIST);
1558                         if (inode->i_nlink)
1559                                 set_inline_node(ipage);
1560                 } else {
1561                         err = f2fs_convert_inline_page(&dn, page);
1562                         if (err)
1563                                 goto out;
1564                         if (dn.data_blkaddr == NULL_ADDR)
1565                                 err = f2fs_get_block(&dn, index);
1566                 }
1567         } else if (locked) {
1568                 err = f2fs_get_block(&dn, index);
1569         } else {
1570                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1571                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1572                 } else {
1573                         /* hole case */
1574                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1575                         if (err || dn.data_blkaddr == NULL_ADDR) {
1576                                 f2fs_put_dnode(&dn);
1577                                 f2fs_lock_op(sbi);
1578                                 locked = true;
1579                                 goto restart;
1580                         }
1581                 }
1582         }
1583
1584         /* convert_inline_page can make node_changed */
1585         *blk_addr = dn.data_blkaddr;
1586         *node_changed = dn.node_changed;
1587 out:
1588         f2fs_put_dnode(&dn);
1589 unlock_out:
1590         if (locked)
1591                 f2fs_unlock_op(sbi);
1592         return err;
1593 }
1594
1595 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1596                 loff_t pos, unsigned len, unsigned flags,
1597                 struct page **pagep, void **fsdata)
1598 {
1599         struct inode *inode = mapping->host;
1600         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1601         struct page *page = NULL;
1602         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1603         bool need_balance = false;
1604         block_t blkaddr = NULL_ADDR;
1605         int err = 0;
1606
1607         trace_f2fs_write_begin(inode, pos, len, flags);
1608
1609         /*
1610          * We should check this at this moment to avoid deadlock on inode page
1611          * and #0 page. The locking rule for inline_data conversion should be:
1612          * lock_page(page #0) -> lock_page(inode_page)
1613          */
1614         if (index != 0) {
1615                 err = f2fs_convert_inline_inode(inode);
1616                 if (err)
1617                         goto fail;
1618         }
1619 repeat:
1620         page = grab_cache_page_write_begin(mapping, index, flags);
1621         if (!page) {
1622                 err = -ENOMEM;
1623                 goto fail;
1624         }
1625
1626         *pagep = page;
1627
1628         err = prepare_write_begin(sbi, page, pos, len,
1629                                         &blkaddr, &need_balance);
1630         if (err)
1631                 goto fail;
1632
1633         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1634                 unlock_page(page);
1635                 f2fs_balance_fs(sbi, true);
1636                 lock_page(page);
1637                 if (page->mapping != mapping) {
1638                         /* The page got truncated from under us */
1639                         f2fs_put_page(page, 1);
1640                         goto repeat;
1641                 }
1642         }
1643
1644         f2fs_wait_on_page_writeback(page, DATA, false);
1645
1646         /* wait for GCed encrypted page writeback */
1647         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1648                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1649
1650         if (len == PAGE_SIZE || PageUptodate(page))
1651                 return 0;
1652
1653         if (blkaddr == NEW_ADDR) {
1654                 zero_user_segment(page, 0, PAGE_SIZE);
1655                 SetPageUptodate(page);
1656         } else {
1657                 struct bio *bio;
1658
1659                 bio = f2fs_grab_bio(inode, blkaddr, 1);
1660                 if (IS_ERR(bio)) {
1661                         err = PTR_ERR(bio);
1662                         goto fail;
1663                 }
1664                 bio->bi_opf = REQ_OP_READ;
1665                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1666                         bio_put(bio);
1667                         err = -EFAULT;
1668                         goto fail;
1669                 }
1670
1671                 __submit_bio(sbi, bio, DATA);
1672
1673                 lock_page(page);
1674                 if (unlikely(page->mapping != mapping)) {
1675                         f2fs_put_page(page, 1);
1676                         goto repeat;
1677                 }
1678                 if (unlikely(!PageUptodate(page))) {
1679                         err = -EIO;
1680                         goto fail;
1681                 }
1682         }
1683         return 0;
1684
1685 fail:
1686         f2fs_put_page(page, 1);
1687         f2fs_write_failed(mapping, pos + len);
1688         return err;
1689 }
1690
1691 static int f2fs_write_end(struct file *file,
1692                         struct address_space *mapping,
1693                         loff_t pos, unsigned len, unsigned copied,
1694                         struct page *page, void *fsdata)
1695 {
1696         struct inode *inode = page->mapping->host;
1697
1698         trace_f2fs_write_end(inode, pos, len, copied);
1699
1700         /*
1701          * This should be come from len == PAGE_SIZE, and we expect copied
1702          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1703          * let generic_perform_write() try to copy data again through copied=0.
1704          */
1705         if (!PageUptodate(page)) {
1706                 if (unlikely(copied != PAGE_SIZE))
1707                         copied = 0;
1708                 else
1709                         SetPageUptodate(page);
1710         }
1711         if (!copied)
1712                 goto unlock_out;
1713
1714         set_page_dirty(page);
1715         clear_cold_data(page);
1716
1717         if (pos + copied > i_size_read(inode))
1718                 f2fs_i_size_write(inode, pos + copied);
1719 unlock_out:
1720         f2fs_put_page(page, 1);
1721         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1722         return copied;
1723 }
1724
1725 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1726                            loff_t offset)
1727 {
1728         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1729
1730         if (offset & blocksize_mask)
1731                 return -EINVAL;
1732
1733         if (iov_iter_alignment(iter) & blocksize_mask)
1734                 return -EINVAL;
1735
1736         return 0;
1737 }
1738
1739 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1740 {
1741         struct address_space *mapping = iocb->ki_filp->f_mapping;
1742         struct inode *inode = mapping->host;
1743         size_t count = iov_iter_count(iter);
1744         loff_t offset = iocb->ki_pos;
1745         int rw = iov_iter_rw(iter);
1746         int err;
1747
1748         err = check_direct_IO(inode, iter, offset);
1749         if (err)
1750                 return err;
1751
1752         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1753                 return 0;
1754         if (test_opt(F2FS_I_SB(inode), LFS))
1755                 return 0;
1756
1757         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1758
1759         down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1760         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1761         up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1762
1763         if (rw == WRITE) {
1764                 if (err > 0)
1765                         set_inode_flag(inode, FI_UPDATE_WRITE);
1766                 else if (err < 0)
1767                         f2fs_write_failed(mapping, offset + count);
1768         }
1769
1770         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1771
1772         return err;
1773 }
1774
1775 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1776                                                         unsigned int length)
1777 {
1778         struct inode *inode = page->mapping->host;
1779         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1780
1781         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1782                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1783                 return;
1784
1785         if (PageDirty(page)) {
1786                 if (inode->i_ino == F2FS_META_INO(sbi))
1787                         dec_page_count(sbi, F2FS_DIRTY_META);
1788                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1789                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1790                 else
1791                         inode_dec_dirty_pages(inode);
1792         }
1793
1794         /* This is atomic written page, keep Private */
1795         if (IS_ATOMIC_WRITTEN_PAGE(page))
1796                 return;
1797
1798         set_page_private(page, 0);
1799         ClearPagePrivate(page);
1800 }
1801
1802 int f2fs_release_page(struct page *page, gfp_t wait)
1803 {
1804         /* If this is dirty page, keep PagePrivate */
1805         if (PageDirty(page))
1806                 return 0;
1807
1808         /* This is atomic written page, keep Private */
1809         if (IS_ATOMIC_WRITTEN_PAGE(page))
1810                 return 0;
1811
1812         set_page_private(page, 0);
1813         ClearPagePrivate(page);
1814         return 1;
1815 }
1816
1817 /*
1818  * This was copied from __set_page_dirty_buffers which gives higher performance
1819  * in very high speed storages. (e.g., pmem)
1820  */
1821 void f2fs_set_page_dirty_nobuffers(struct page *page)
1822 {
1823         struct address_space *mapping = page->mapping;
1824         unsigned long flags;
1825
1826         if (unlikely(!mapping))
1827                 return;
1828
1829         spin_lock(&mapping->private_lock);
1830         lock_page_memcg(page);
1831         SetPageDirty(page);
1832         spin_unlock(&mapping->private_lock);
1833
1834         spin_lock_irqsave(&mapping->tree_lock, flags);
1835         WARN_ON_ONCE(!PageUptodate(page));
1836         account_page_dirtied(page, mapping);
1837         radix_tree_tag_set(&mapping->page_tree,
1838                         page_index(page), PAGECACHE_TAG_DIRTY);
1839         spin_unlock_irqrestore(&mapping->tree_lock, flags);
1840         unlock_page_memcg(page);
1841
1842         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1843         return;
1844 }
1845
1846 static int f2fs_set_data_page_dirty(struct page *page)
1847 {
1848         struct address_space *mapping = page->mapping;
1849         struct inode *inode = mapping->host;
1850
1851         trace_f2fs_set_page_dirty(page, DATA);
1852
1853         if (!PageUptodate(page))
1854                 SetPageUptodate(page);
1855
1856         if (f2fs_is_atomic_file(inode)) {
1857                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1858                         register_inmem_page(inode, page);
1859                         return 1;
1860                 }
1861                 /*
1862                  * Previously, this page has been registered, we just
1863                  * return here.
1864                  */
1865                 return 0;
1866         }
1867
1868         if (!PageDirty(page)) {
1869                 f2fs_set_page_dirty_nobuffers(page);
1870                 update_dirty_page(inode, page);
1871                 return 1;
1872         }
1873         return 0;
1874 }
1875
1876 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1877 {
1878         struct inode *inode = mapping->host;
1879
1880         if (f2fs_has_inline_data(inode))
1881                 return 0;
1882
1883         /* make sure allocating whole blocks */
1884         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1885                 filemap_write_and_wait(mapping);
1886
1887         return generic_block_bmap(mapping, block, get_data_block_bmap);
1888 }
1889
1890 #ifdef CONFIG_MIGRATION
1891 #include <linux/migrate.h>
1892
1893 int f2fs_migrate_page(struct address_space *mapping,
1894                 struct page *newpage, struct page *page, enum migrate_mode mode)
1895 {
1896         int rc, extra_count;
1897         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
1898         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
1899
1900         BUG_ON(PageWriteback(page));
1901
1902         /* migrating an atomic written page is safe with the inmem_lock hold */
1903         if (atomic_written && !mutex_trylock(&fi->inmem_lock))
1904                 return -EAGAIN;
1905
1906         /*
1907          * A reference is expected if PagePrivate set when move mapping,
1908          * however F2FS breaks this for maintaining dirty page counts when
1909          * truncating pages. So here adjusting the 'extra_count' make it work.
1910          */
1911         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
1912         rc = migrate_page_move_mapping(mapping, newpage,
1913                                 page, NULL, mode, extra_count);
1914         if (rc != MIGRATEPAGE_SUCCESS) {
1915                 if (atomic_written)
1916                         mutex_unlock(&fi->inmem_lock);
1917                 return rc;
1918         }
1919
1920         if (atomic_written) {
1921                 struct inmem_pages *cur;
1922                 list_for_each_entry(cur, &fi->inmem_pages, list)
1923                         if (cur->page == page) {
1924                                 cur->page = newpage;
1925                                 break;
1926                         }
1927                 mutex_unlock(&fi->inmem_lock);
1928                 put_page(page);
1929                 get_page(newpage);
1930         }
1931
1932         if (PagePrivate(page))
1933                 SetPagePrivate(newpage);
1934         set_page_private(newpage, page_private(page));
1935
1936         migrate_page_copy(newpage, page);
1937
1938         return MIGRATEPAGE_SUCCESS;
1939 }
1940 #endif
1941
1942 const struct address_space_operations f2fs_dblock_aops = {
1943         .readpage       = f2fs_read_data_page,
1944         .readpages      = f2fs_read_data_pages,
1945         .writepage      = f2fs_write_data_page,
1946         .writepages     = f2fs_write_data_pages,
1947         .write_begin    = f2fs_write_begin,
1948         .write_end      = f2fs_write_end,
1949         .set_page_dirty = f2fs_set_data_page_dirty,
1950         .invalidatepage = f2fs_invalidate_page,
1951         .releasepage    = f2fs_release_page,
1952         .direct_IO      = f2fs_direct_IO,
1953         .bmap           = f2fs_bmap,
1954 #ifdef CONFIG_MIGRATION
1955         .migratepage    = f2fs_migrate_page,
1956 #endif
1957 };