fs/btrfs/ioctl.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/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/compat.h>
  37 #include <linux/bit_spinlock.h>
  38 #include <linux/security.h>
  39 #include <linux/xattr.h>
  40 #include <linux/mm.h>
  41 #include <linux/slab.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/uuid.h>
  44 #include <linux/btrfs.h>
  45 #include <linux/uaccess.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "volumes.h"
  52 #include "locking.h"
  53 #include "inode-map.h"
  54 #include "backref.h"
  55 #include "rcu-string.h"
  56 #include "send.h"
  57 #include "dev-replace.h"
  58 #include "props.h"
  59 #include "sysfs.h"
  60 #include "qgroup.h"
  61 #include "tree-log.h"
  62 #include "compression.h"
  63
  64 #ifdef CONFIG_64BIT
  65 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  66  * structures are incorrect, as the timespec structure from userspace
  67  * is 4 bytes too small. We define these alternatives here to teach
  68  * the kernel about the 32-bit struct packing.
  69  */
  70 struct btrfs_ioctl_timespec_32 {
  71         __u64 sec;
  72         __u32 nsec;
  73 } __attribute__ ((__packed__));
  74
  75 struct btrfs_ioctl_received_subvol_args_32 {
  76         char    uuid[BTRFS_UUID_SIZE];  /* in */
  77         __u64   stransid;               /* in */
  78         __u64   rtransid;               /* out */
  79         struct btrfs_ioctl_timespec_32 stime; /* in */
  80         struct btrfs_ioctl_timespec_32 rtime; /* out */
  81         __u64   flags;                  /* in */
  82         __u64   reserved[16];           /* in */
  83 } __attribute__ ((__packed__));
  84
  85 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  86                                 struct btrfs_ioctl_received_subvol_args_32)
  87 #endif
  88
  89 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  90 struct btrfs_ioctl_send_args_32 {
  91         __s64 send_fd;                  /* in */
  92         __u64 clone_sources_count;      /* in */
  93         compat_uptr_t clone_sources;    /* in */
  94         __u64 parent_root;              /* in */
  95         __u64 flags;                    /* in */
  96         __u64 reserved[4];              /* in */
  97 } __attribute__ ((__packed__));
  98
  99 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
 100                                struct btrfs_ioctl_send_args_32)
 101 #endif
 102
 103 static int btrfs_clone(struct inode *src, struct inode *inode,
 104                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
 105                        int no_time_update);
 106
 107 /* Mask out flags that are inappropriate for the given type of inode. */
 108 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
 109 {
 110         if (S_ISDIR(mode))
 111                 return flags;
 112         else if (S_ISREG(mode))
 113                 return flags & ~FS_DIRSYNC_FL;
 114         else
 115                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 116 }
 117
 118 /*
 119  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 120  */
 121 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 122 {
 123         unsigned int iflags = 0;
 124
 125         if (flags & BTRFS_INODE_SYNC)
 126                 iflags |= FS_SYNC_FL;
 127         if (flags & BTRFS_INODE_IMMUTABLE)
 128                 iflags |= FS_IMMUTABLE_FL;
 129         if (flags & BTRFS_INODE_APPEND)
 130                 iflags |= FS_APPEND_FL;
 131         if (flags & BTRFS_INODE_NODUMP)
 132                 iflags |= FS_NODUMP_FL;
 133         if (flags & BTRFS_INODE_NOATIME)
 134                 iflags |= FS_NOATIME_FL;
 135         if (flags & BTRFS_INODE_DIRSYNC)
 136                 iflags |= FS_DIRSYNC_FL;
 137         if (flags & BTRFS_INODE_NODATACOW)
 138                 iflags |= FS_NOCOW_FL;
 139
 140         if (flags & BTRFS_INODE_NOCOMPRESS)
 141                 iflags |= FS_NOCOMP_FL;
 142         else if (flags & BTRFS_INODE_COMPRESS)
 143                 iflags |= FS_COMPR_FL;
 144
 145         return iflags;
 146 }
 147
 148 /*
 149  * Update inode->i_flags based on the btrfs internal flags.
 150  */
 151 void btrfs_update_iflags(struct inode *inode)
 152 {
 153         struct btrfs_inode *ip = BTRFS_I(inode);
 154         unsigned int new_fl = 0;
 155
 156         if (ip->flags & BTRFS_INODE_SYNC)
 157                 new_fl |= S_SYNC;
 158         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 159                 new_fl |= S_IMMUTABLE;
 160         if (ip->flags & BTRFS_INODE_APPEND)
 161                 new_fl |= S_APPEND;
 162         if (ip->flags & BTRFS_INODE_NOATIME)
 163                 new_fl |= S_NOATIME;
 164         if (ip->flags & BTRFS_INODE_DIRSYNC)
 165                 new_fl |= S_DIRSYNC;
 166
 167         set_mask_bits(&inode->i_flags,
 168                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 169                       new_fl);
 170 }
 171
 172 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 173 {
 174         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 175         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 176
 177         if (copy_to_user(arg, &flags, sizeof(flags)))
 178                 return -EFAULT;
 179         return 0;
 180 }
 181
 182 static int check_flags(unsigned int flags)
 183 {
 184         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 185                       FS_NOATIME_FL | FS_NODUMP_FL | \
 186                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 187                       FS_NOCOMP_FL | FS_COMPR_FL |
 188                       FS_NOCOW_FL))
 189                 return -EOPNOTSUPP;
 190
 191         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 192                 return -EINVAL;
 193
 194         return 0;
 195 }
 196
 197 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 198 {
 199         struct inode *inode = file_inode(file);
 200         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 201         struct btrfs_inode *ip = BTRFS_I(inode);
 202         struct btrfs_root *root = ip->root;
 203         struct btrfs_trans_handle *trans;
 204         unsigned int flags, oldflags;
 205         int ret;
 206         u64 ip_oldflags;
 207         unsigned int i_oldflags;
 208         umode_t mode;
 209
 210         if (!inode_owner_or_capable(inode))
 211                 return -EPERM;
 212
 213         if (btrfs_root_readonly(root))
 214                 return -EROFS;
 215
 216         if (copy_from_user(&flags, arg, sizeof(flags)))
 217                 return -EFAULT;
 218
 219         ret = check_flags(flags);
 220         if (ret)
 221                 return ret;
 222
 223         ret = mnt_want_write_file(file);
 224         if (ret)
 225                 return ret;
 226
 227         inode_lock(inode);
 228
 229         ip_oldflags = ip->flags;
 230         i_oldflags = inode->i_flags;
 231         mode = inode->i_mode;
 232
 233         flags = btrfs_mask_flags(inode->i_mode, flags);
 234         oldflags = btrfs_flags_to_ioctl(ip->flags);
 235         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 236                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 237                         ret = -EPERM;
 238                         goto out_unlock;
 239                 }
 240         }
 241
 242         if (flags & FS_SYNC_FL)
 243                 ip->flags |= BTRFS_INODE_SYNC;
 244         else
 245                 ip->flags &= ~BTRFS_INODE_SYNC;
 246         if (flags & FS_IMMUTABLE_FL)
 247                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 248         else
 249                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 250         if (flags & FS_APPEND_FL)
 251                 ip->flags |= BTRFS_INODE_APPEND;
 252         else
 253                 ip->flags &= ~BTRFS_INODE_APPEND;
 254         if (flags & FS_NODUMP_FL)
 255                 ip->flags |= BTRFS_INODE_NODUMP;
 256         else
 257                 ip->flags &= ~BTRFS_INODE_NODUMP;
 258         if (flags & FS_NOATIME_FL)
 259                 ip->flags |= BTRFS_INODE_NOATIME;
 260         else
 261                 ip->flags &= ~BTRFS_INODE_NOATIME;
 262         if (flags & FS_DIRSYNC_FL)
 263                 ip->flags |= BTRFS_INODE_DIRSYNC;
 264         else
 265                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 266         if (flags & FS_NOCOW_FL) {
 267                 if (S_ISREG(mode)) {
 268                         /*
 269                          * It's safe to turn csums off here, no extents exist.
 270                          * Otherwise we want the flag to reflect the real COW
 271                          * status of the file and will not set it.
 272                          */
 273                         if (inode->i_size == 0)
 274                                 ip->flags |= BTRFS_INODE_NODATACOW
 275                                            | BTRFS_INODE_NODATASUM;
 276                 } else {
 277                         ip->flags |= BTRFS_INODE_NODATACOW;
 278                 }
 279         } else {
 280                 /*
 281                  * Revert back under same assumptions as above
 282                  */
 283                 if (S_ISREG(mode)) {
 284                         if (inode->i_size == 0)
 285                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 286                                              | BTRFS_INODE_NODATASUM);
 287                 } else {
 288                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 289                 }
 290         }
 291
 292         /*
 293          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 294          * flag may be changed automatically if compression code won't make
 295          * things smaller.
 296          */
 297         if (flags & FS_NOCOMP_FL) {
 298                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 299                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 300
 301                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 302                 if (ret && ret != -ENODATA)
 303                         goto out_drop;
 304         } else if (flags & FS_COMPR_FL) {
 305                 const char *comp;
 306
 307                 ip->flags |= BTRFS_INODE_COMPRESS;
 308                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 309
 310                 comp = btrfs_compress_type2str(fs_info->compress_type);
 311                 if (!comp || comp[0] == 0)
 312                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 313
 314                 ret = btrfs_set_prop(inode, "btrfs.compression",
 315                                      comp, strlen(comp), 0);
 316                 if (ret)
 317                         goto out_drop;
 318
 319         } else {
 320                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 321                 if (ret && ret != -ENODATA)
 322                         goto out_drop;
 323                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 324         }
 325
 326         trans = btrfs_start_transaction(root, 1);
 327         if (IS_ERR(trans)) {
 328                 ret = PTR_ERR(trans);
 329                 goto out_drop;
 330         }
 331
 332         btrfs_update_iflags(inode);
 333         inode_inc_iversion(inode);
 334         inode->i_ctime = current_time(inode);
 335         ret = btrfs_update_inode(trans, root, inode);
 336
 337         btrfs_end_transaction(trans);
 338  out_drop:
 339         if (ret) {
 340                 ip->flags = ip_oldflags;
 341                 inode->i_flags = i_oldflags;
 342         }
 343
 344  out_unlock:
 345         inode_unlock(inode);
 346         mnt_drop_write_file(file);
 347         return ret;
 348 }
 349
 350 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 351 {
 352         struct inode *inode = file_inode(file);
 353
 354         return put_user(inode->i_generation, arg);
 355 }
 356
 357 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 358 {
 359         struct inode *inode = file_inode(file);
 360         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 361         struct btrfs_device *device;
 362         struct request_queue *q;
 363         struct fstrim_range range;
 364         u64 minlen = ULLONG_MAX;
 365         u64 num_devices = 0;
 366         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 367         int ret;
 368
 369         if (!capable(CAP_SYS_ADMIN))
 370                 return -EPERM;
 371
 372         rcu_read_lock();
 373         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 374                                 dev_list) {
 375                 if (!device->bdev)
 376                         continue;
 377                 q = bdev_get_queue(device->bdev);
 378                 if (blk_queue_discard(q)) {
 379                         num_devices++;
 380                         minlen = min_t(u64, q->limits.discard_granularity,
 381                                      minlen);
 382                 }
 383         }
 384         rcu_read_unlock();
 385
 386         if (!num_devices)
 387                 return -EOPNOTSUPP;
 388         if (copy_from_user(&range, arg, sizeof(range)))
 389                 return -EFAULT;
 390         if (range.start > total_bytes ||
 391             range.len < fs_info->sb->s_blocksize)
 392                 return -EINVAL;
 393
 394         range.len = min(range.len, total_bytes - range.start);
 395         range.minlen = max(range.minlen, minlen);
 396         ret = btrfs_trim_fs(fs_info, &range);
 397         if (ret < 0)
 398                 return ret;
 399
 400         if (copy_to_user(arg, &range, sizeof(range)))
 401                 return -EFAULT;
 402
 403         return 0;
 404 }
 405
 406 int btrfs_is_empty_uuid(u8 *uuid)
 407 {
 408         int i;
 409
 410         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 411                 if (uuid[i])
 412                         return 0;
 413         }
 414         return 1;
 415 }
 416
 417 static noinline int create_subvol(struct inode *dir,
 418                                   struct dentry *dentry,
 419                                   const char *name, int namelen,
 420                                   u64 *async_transid,
 421                                   struct btrfs_qgroup_inherit *inherit)
 422 {
 423         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 424         struct btrfs_trans_handle *trans;
 425         struct btrfs_key key;
 426         struct btrfs_root_item *root_item;
 427         struct btrfs_inode_item *inode_item;
 428         struct extent_buffer *leaf;
 429         struct btrfs_root *root = BTRFS_I(dir)->root;
 430         struct btrfs_root *new_root;
 431         struct btrfs_block_rsv block_rsv;
 432         struct timespec cur_time = current_time(dir);
 433         struct inode *inode;
 434         int ret;
 435         int err;
 436         u64 objectid;
 437         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 438         u64 index = 0;
 439         u64 qgroup_reserved;
 440         uuid_le new_uuid;
 441
 442         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 443         if (!root_item)
 444                 return -ENOMEM;
 445
 446         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 447         if (ret)
 448                 goto fail_free;
 449
 450         /*
 451          * Don't create subvolume whose level is not zero. Or qgroup will be
 452          * screwed up since it assumes subvolume qgroup's level to be 0.
 453          */
 454         if (btrfs_qgroup_level(objectid)) {
 455                 ret = -ENOSPC;
 456                 goto fail_free;
 457         }
 458
 459         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 460         /*
 461          * The same as the snapshot creation, please see the comment
 462          * of create_snapshot().
 463          */
 464         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 465                                                8, &qgroup_reserved, false);
 466         if (ret)
 467                 goto fail_free;
 468
 469         trans = btrfs_start_transaction(root, 0);
 470         if (IS_ERR(trans)) {
 471                 ret = PTR_ERR(trans);
 472                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 473                 goto fail_free;
 474         }
 475         trans->block_rsv = &block_rsv;
 476         trans->bytes_reserved = block_rsv.size;
 477
 478         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 479         if (ret)
 480                 goto fail;
 481
 482         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 483         if (IS_ERR(leaf)) {
 484                 ret = PTR_ERR(leaf);
 485                 goto fail;
 486         }
 487
 488         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 489         btrfs_set_header_bytenr(leaf, leaf->start);
 490         btrfs_set_header_generation(leaf, trans->transid);
 491         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 492         btrfs_set_header_owner(leaf, objectid);
 493
 494         write_extent_buffer_fsid(leaf, fs_info->fsid);
 495         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 496         btrfs_mark_buffer_dirty(leaf);
 497
 498         inode_item = &root_item->inode;
 499         btrfs_set_stack_inode_generation(inode_item, 1);
 500         btrfs_set_stack_inode_size(inode_item, 3);
 501         btrfs_set_stack_inode_nlink(inode_item, 1);
 502         btrfs_set_stack_inode_nbytes(inode_item,
 503                                      fs_info->nodesize);
 504         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 505
 506         btrfs_set_root_flags(root_item, 0);
 507         btrfs_set_root_limit(root_item, 0);
 508         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 509
 510         btrfs_set_root_bytenr(root_item, leaf->start);
 511         btrfs_set_root_generation(root_item, trans->transid);
 512         btrfs_set_root_level(root_item, 0);
 513         btrfs_set_root_refs(root_item, 1);
 514         btrfs_set_root_used(root_item, leaf->len);
 515         btrfs_set_root_last_snapshot(root_item, 0);
 516
 517         btrfs_set_root_generation_v2(root_item,
 518                         btrfs_root_generation(root_item));
 519         uuid_le_gen(&new_uuid);
 520         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 521         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 522         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 523         root_item->ctime = root_item->otime;
 524         btrfs_set_root_ctransid(root_item, trans->transid);
 525         btrfs_set_root_otransid(root_item, trans->transid);
 526
 527         btrfs_tree_unlock(leaf);
 528         free_extent_buffer(leaf);
 529         leaf = NULL;
 530
 531         btrfs_set_root_dirid(root_item, new_dirid);
 532
 533         key.objectid = objectid;
 534         key.offset = 0;
 535         key.type = BTRFS_ROOT_ITEM_KEY;
 536         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 537                                 root_item);
 538         if (ret)
 539                 goto fail;
 540
 541         key.offset = (u64)-1;
 542         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 543         if (IS_ERR(new_root)) {
 544                 ret = PTR_ERR(new_root);
 545                 btrfs_abort_transaction(trans, ret);
 546                 goto fail;
 547         }
 548
 549         btrfs_record_root_in_trans(trans, new_root);
 550
 551         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 552         if (ret) {
 553                 /* We potentially lose an unused inode item here */
 554                 btrfs_abort_transaction(trans, ret);
 555                 goto fail;
 556         }
 557
 558         mutex_lock(&new_root->objectid_mutex);
 559         new_root->highest_objectid = new_dirid;
 560         mutex_unlock(&new_root->objectid_mutex);
 561
 562         /*
 563          * insert the directory item
 564          */
 565         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 566         if (ret) {
 567                 btrfs_abort_transaction(trans, ret);
 568                 goto fail;
 569         }
 570
 571         ret = btrfs_insert_dir_item(trans, root,
 572                                     name, namelen, BTRFS_I(dir), &key,
 573                                     BTRFS_FT_DIR, index);
 574         if (ret) {
 575                 btrfs_abort_transaction(trans, ret);
 576                 goto fail;
 577         }
 578
 579         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 580         ret = btrfs_update_inode(trans, root, dir);
 581         BUG_ON(ret);
 582
 583         ret = btrfs_add_root_ref(trans, fs_info,
 584                                  objectid, root->root_key.objectid,
 585                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 586         BUG_ON(ret);
 587
 588         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
 589                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 590         if (ret)
 591                 btrfs_abort_transaction(trans, ret);
 592
 593 fail:
 594         kfree(root_item);
 595         trans->block_rsv = NULL;
 596         trans->bytes_reserved = 0;
 597         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 598
 599         if (async_transid) {
 600                 *async_transid = trans->transid;
 601                 err = btrfs_commit_transaction_async(trans, 1);
 602                 if (err)
 603                         err = btrfs_commit_transaction(trans);
 604         } else {
 605                 err = btrfs_commit_transaction(trans);
 606         }
 607         if (err && !ret)
 608                 ret = err;
 609
 610         if (!ret) {
 611                 inode = btrfs_lookup_dentry(dir, dentry);
 612                 if (IS_ERR(inode))
 613                         return PTR_ERR(inode);
 614                 d_instantiate(dentry, inode);
 615         }
 616         return ret;
 617
 618 fail_free:
 619         kfree(root_item);
 620         return ret;
 621 }
 622
 623 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 624                            struct dentry *dentry,
 625                            u64 *async_transid, bool readonly,
 626                            struct btrfs_qgroup_inherit *inherit)
 627 {
 628         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 629         struct inode *inode;
 630         struct btrfs_pending_snapshot *pending_snapshot;
 631         struct btrfs_trans_handle *trans;
 632         int ret;
 633
 634         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 635                 return -EINVAL;
 636
 637         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 638         if (!pending_snapshot)
 639                 return -ENOMEM;
 640
 641         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 642                         GFP_KERNEL);
 643         pending_snapshot->path = btrfs_alloc_path();
 644         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 645                 ret = -ENOMEM;
 646                 goto free_pending;
 647         }
 648
 649         atomic_inc(&root->will_be_snapshotted);
 650         smp_mb__after_atomic();
 651         /* wait for no snapshot writes */
 652         wait_event(root->subv_writers->wait,
 653                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 654
 655         ret = btrfs_start_delalloc_inodes(root, 0);
 656         if (ret)
 657                 goto dec_and_free;
 658
 659         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 660
 661         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 662                              BTRFS_BLOCK_RSV_TEMP);
 663         /*
 664          * 1 - parent dir inode
 665          * 2 - dir entries
 666          * 1 - root item
 667          * 2 - root ref/backref
 668          * 1 - root of snapshot
 669          * 1 - UUID item
 670          */
 671         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 672                                         &pending_snapshot->block_rsv, 8,
 673                                         &pending_snapshot->qgroup_reserved,
 674                                         false);
 675         if (ret)
 676                 goto dec_and_free;
 677
 678         pending_snapshot->dentry = dentry;
 679         pending_snapshot->root = root;
 680         pending_snapshot->readonly = readonly;
 681         pending_snapshot->dir = dir;
 682         pending_snapshot->inherit = inherit;
 683
 684         trans = btrfs_start_transaction(root, 0);
 685         if (IS_ERR(trans)) {
 686                 ret = PTR_ERR(trans);
 687                 goto fail;
 688         }
 689
 690         spin_lock(&fs_info->trans_lock);
 691         list_add(&pending_snapshot->list,
 692                  &trans->transaction->pending_snapshots);
 693         spin_unlock(&fs_info->trans_lock);
 694         if (async_transid) {
 695                 *async_transid = trans->transid;
 696                 ret = btrfs_commit_transaction_async(trans, 1);
 697                 if (ret)
 698                         ret = btrfs_commit_transaction(trans);
 699         } else {
 700                 ret = btrfs_commit_transaction(trans);
 701         }
 702         if (ret)
 703                 goto fail;
 704
 705         ret = pending_snapshot->error;
 706         if (ret)
 707                 goto fail;
 708
 709         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 710         if (ret)
 711                 goto fail;
 712
 713         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 714         if (IS_ERR(inode)) {
 715                 ret = PTR_ERR(inode);
 716                 goto fail;
 717         }
 718
 719         d_instantiate(dentry, inode);
 720         ret = 0;
 721 fail:
 722         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 723 dec_and_free:
 724         if (atomic_dec_and_test(&root->will_be_snapshotted))
 725                 wake_up_atomic_t(&root->will_be_snapshotted);
 726 free_pending:
 727         kfree(pending_snapshot->root_item);
 728         btrfs_free_path(pending_snapshot->path);
 729         kfree(pending_snapshot);
 730
 731         return ret;
 732 }
 733
 734 /*  copy of may_delete in fs/namei.c()
 735  *      Check whether we can remove a link victim from directory dir, check
 736  *  whether the type of victim is right.
 737  *  1. We can't do it if dir is read-only (done in permission())
 738  *  2. We should have write and exec permissions on dir
 739  *  3. We can't remove anything from append-only dir
 740  *  4. We can't do anything with immutable dir (done in permission())
 741  *  5. If the sticky bit on dir is set we should either
 742  *      a. be owner of dir, or
 743  *      b. be owner of victim, or
 744  *      c. have CAP_FOWNER capability
 745  *  6. If the victim is append-only or immutable we can't do anything with
 746  *     links pointing to it.
 747  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 748  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 749  *  9. We can't remove a root or mountpoint.
 750  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 751  *     nfs_async_unlink().
 752  */
 753
 754 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 755 {
 756         int error;
 757
 758         if (d_really_is_negative(victim))
 759                 return -ENOENT;
 760
 761         BUG_ON(d_inode(victim->d_parent) != dir);
 762         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 763
 764         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 765         if (error)
 766                 return error;
 767         if (IS_APPEND(dir))
 768                 return -EPERM;
 769         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 770             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 771                 return -EPERM;
 772         if (isdir) {
 773                 if (!d_is_dir(victim))
 774                         return -ENOTDIR;
 775                 if (IS_ROOT(victim))
 776                         return -EBUSY;
 777         } else if (d_is_dir(victim))
 778                 return -EISDIR;
 779         if (IS_DEADDIR(dir))
 780                 return -ENOENT;
 781         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 782                 return -EBUSY;
 783         return 0;
 784 }
 785
 786 /* copy of may_create in fs/namei.c() */
 787 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 788 {
 789         if (d_really_is_positive(child))
 790                 return -EEXIST;
 791         if (IS_DEADDIR(dir))
 792                 return -ENOENT;
 793         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 794 }
 795
 796 /*
 797  * Create a new subvolume below @parent.  This is largely modeled after
 798  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 799  * inside this filesystem so it's quite a bit simpler.
 800  */
 801 static noinline int btrfs_mksubvol(const struct path *parent,
 802                                    const char *name, int namelen,
 803                                    struct btrfs_root *snap_src,
 804                                    u64 *async_transid, bool readonly,
 805                                    struct btrfs_qgroup_inherit *inherit)
 806 {
 807         struct inode *dir = d_inode(parent->dentry);
 808         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 809         struct dentry *dentry;
 810         int error;
 811
 812         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 813         if (error == -EINTR)
 814                 return error;
 815
 816         dentry = lookup_one_len(name, parent->dentry, namelen);
 817         error = PTR_ERR(dentry);
 818         if (IS_ERR(dentry))
 819                 goto out_unlock;
 820
 821         error = btrfs_may_create(dir, dentry);
 822         if (error)
 823                 goto out_dput;
 824
 825         /*
 826          * even if this name doesn't exist, we may get hash collisions.
 827          * check for them now when we can safely fail
 828          */
 829         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 830                                                dir->i_ino, name,
 831                                                namelen);
 832         if (error)
 833                 goto out_dput;
 834
 835         down_read(&fs_info->subvol_sem);
 836
 837         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 838                 goto out_up_read;
 839
 840         if (snap_src) {
 841                 error = create_snapshot(snap_src, dir, dentry,
 842                                         async_transid, readonly, inherit);
 843         } else {
 844                 error = create_subvol(dir, dentry, name, namelen,
 845                                       async_transid, inherit);
 846         }
 847         if (!error)
 848                 fsnotify_mkdir(dir, dentry);
 849 out_up_read:
 850         up_read(&fs_info->subvol_sem);
 851 out_dput:
 852         dput(dentry);
 853 out_unlock:
 854         inode_unlock(dir);
 855         return error;
 856 }
 857
 858 /*
 859  * When we're defragging a range, we don't want to kick it off again
 860  * if it is really just waiting for delalloc to send it down.
 861  * If we find a nice big extent or delalloc range for the bytes in the
 862  * file you want to defrag, we return 0 to let you know to skip this
 863  * part of the file
 864  */
 865 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 866 {
 867         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 868         struct extent_map *em = NULL;
 869         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 870         u64 end;
 871
 872         read_lock(&em_tree->lock);
 873         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 874         read_unlock(&em_tree->lock);
 875
 876         if (em) {
 877                 end = extent_map_end(em);
 878                 free_extent_map(em);
 879                 if (end - offset > thresh)
 880                         return 0;
 881         }
 882         /* if we already have a nice delalloc here, just stop */
 883         thresh /= 2;
 884         end = count_range_bits(io_tree, &offset, offset + thresh,
 885                                thresh, EXTENT_DELALLOC, 1);
 886         if (end >= thresh)
 887                 return 0;
 888         return 1;
 889 }
 890
 891 /*
 892  * helper function to walk through a file and find extents
 893  * newer than a specific transid, and smaller than thresh.
 894  *
 895  * This is used by the defragging code to find new and small
 896  * extents
 897  */
 898 static int find_new_extents(struct btrfs_root *root,
 899                             struct inode *inode, u64 newer_than,
 900                             u64 *off, u32 thresh)
 901 {
 902         struct btrfs_path *path;
 903         struct btrfs_key min_key;
 904         struct extent_buffer *leaf;
 905         struct btrfs_file_extent_item *extent;
 906         int type;
 907         int ret;
 908         u64 ino = btrfs_ino(BTRFS_I(inode));
 909
 910         path = btrfs_alloc_path();
 911         if (!path)
 912                 return -ENOMEM;
 913
 914         min_key.objectid = ino;
 915         min_key.type = BTRFS_EXTENT_DATA_KEY;
 916         min_key.offset = *off;
 917
 918         while (1) {
 919                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 920                 if (ret != 0)
 921                         goto none;
 922 process_slot:
 923                 if (min_key.objectid != ino)
 924                         goto none;
 925                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 926                         goto none;
 927
 928                 leaf = path->nodes[0];
 929                 extent = btrfs_item_ptr(leaf, path->slots[0],
 930                                         struct btrfs_file_extent_item);
 931
 932                 type = btrfs_file_extent_type(leaf, extent);
 933                 if (type == BTRFS_FILE_EXTENT_REG &&
 934                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 935                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 936                         *off = min_key.offset;
 937                         btrfs_free_path(path);
 938                         return 0;
 939                 }
 940
 941                 path->slots[0]++;
 942                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 943                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 944                         goto process_slot;
 945                 }
 946
 947                 if (min_key.offset == (u64)-1)
 948                         goto none;
 949
 950                 min_key.offset++;
 951                 btrfs_release_path(path);
 952         }
 953 none:
 954         btrfs_free_path(path);
 955         return -ENOENT;
 956 }
 957
 958 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 959 {
 960         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 961         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 962         struct extent_map *em;
 963         u64 len = PAGE_SIZE;
 964
 965         /*
 966          * hopefully we have this extent in the tree already, try without
 967          * the full extent lock
 968          */
 969         read_lock(&em_tree->lock);
 970         em = lookup_extent_mapping(em_tree, start, len);
 971         read_unlock(&em_tree->lock);
 972
 973         if (!em) {
 974                 struct extent_state *cached = NULL;
 975                 u64 end = start + len - 1;
 976
 977                 /* get the big lock and read metadata off disk */
 978                 lock_extent_bits(io_tree, start, end, &cached);
 979                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 980                 unlock_extent_cached(io_tree, start, end, &cached);
 981
 982                 if (IS_ERR(em))
 983                         return NULL;
 984         }
 985
 986         return em;
 987 }
 988
 989 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 990 {
 991         struct extent_map *next;
 992         bool ret = true;
 993
 994         /* this is the last extent */
 995         if (em->start + em->len >= i_size_read(inode))
 996                 return false;
 997
 998         next = defrag_lookup_extent(inode, em->start + em->len);
 999         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1000                 ret = false;
1001         else if ((em->block_start + em->block_len == next->block_start) &&
1002                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1003                 ret = false;
1004
1005         free_extent_map(next);
1006         return ret;
1007 }
1008
1009 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1010                                u64 *last_len, u64 *skip, u64 *defrag_end,
1011                                int compress)
1012 {
1013         struct extent_map *em;
1014         int ret = 1;
1015         bool next_mergeable = true;
1016         bool prev_mergeable = true;
1017
1018         /*
1019          * make sure that once we start defragging an extent, we keep on
1020          * defragging it
1021          */
1022         if (start < *defrag_end)
1023                 return 1;
1024
1025         *skip = 0;
1026
1027         em = defrag_lookup_extent(inode, start);
1028         if (!em)
1029                 return 0;
1030
1031         /* this will cover holes, and inline extents */
1032         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1033                 ret = 0;
1034                 goto out;
1035         }
1036
1037         if (!*defrag_end)
1038                 prev_mergeable = false;
1039
1040         next_mergeable = defrag_check_next_extent(inode, em);
1041         /*
1042          * we hit a real extent, if it is big or the next extent is not a
1043          * real extent, don't bother defragging it
1044          */
1045         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1046             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1047                 ret = 0;
1048 out:
1049         /*
1050          * last_len ends up being a counter of how many bytes we've defragged.
1051          * every time we choose not to defrag an extent, we reset *last_len
1052          * so that the next tiny extent will force a defrag.
1053          *
1054          * The end result of this is that tiny extents before a single big
1055          * extent will force at least part of that big extent to be defragged.
1056          */
1057         if (ret) {
1058                 *defrag_end = extent_map_end(em);
1059         } else {
1060                 *last_len = 0;
1061                 *skip = extent_map_end(em);
1062                 *defrag_end = 0;
1063         }
1064
1065         free_extent_map(em);
1066         return ret;
1067 }
1068
1069 /*
1070  * it doesn't do much good to defrag one or two pages
1071  * at a time.  This pulls in a nice chunk of pages
1072  * to COW and defrag.
1073  *
1074  * It also makes sure the delalloc code has enough
1075  * dirty data to avoid making new small extents as part
1076  * of the defrag
1077  *
1078  * It's a good idea to start RA on this range
1079  * before calling this.
1080  */
1081 static int cluster_pages_for_defrag(struct inode *inode,
1082                                     struct page **pages,
1083                                     unsigned long start_index,
1084                                     unsigned long num_pages)
1085 {
1086         unsigned long file_end;
1087         u64 isize = i_size_read(inode);
1088         u64 page_start;
1089         u64 page_end;
1090         u64 page_cnt;
1091         int ret;
1092         int i;
1093         int i_done;
1094         struct btrfs_ordered_extent *ordered;
1095         struct extent_state *cached_state = NULL;
1096         struct extent_io_tree *tree;
1097         struct extent_changeset *data_reserved = NULL;
1098         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1099
1100         file_end = (isize - 1) >> PAGE_SHIFT;
1101         if (!isize || start_index > file_end)
1102                 return 0;
1103
1104         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1105
1106         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1107                         start_index << PAGE_SHIFT,
1108                         page_cnt << PAGE_SHIFT);
1109         if (ret)
1110                 return ret;
1111         i_done = 0;
1112         tree = &BTRFS_I(inode)->io_tree;
1113
1114         /* step one, lock all the pages */
1115         for (i = 0; i < page_cnt; i++) {
1116                 struct page *page;
1117 again:
1118                 page = find_or_create_page(inode->i_mapping,
1119                                            start_index + i, mask);
1120                 if (!page)
1121                         break;
1122
1123                 page_start = page_offset(page);
1124                 page_end = page_start + PAGE_SIZE - 1;
1125                 while (1) {
1126                         lock_extent_bits(tree, page_start, page_end,
1127                                          &cached_state);
1128                         ordered = btrfs_lookup_ordered_extent(inode,
1129                                                               page_start);
1130                         unlock_extent_cached(tree, page_start, page_end,
1131                                              &cached_state);
1132                         if (!ordered)
1133                                 break;
1134
1135                         unlock_page(page);
1136                         btrfs_start_ordered_extent(inode, ordered, 1);
1137                         btrfs_put_ordered_extent(ordered);
1138                         lock_page(page);
1139                         /*
1140                          * we unlocked the page above, so we need check if
1141                          * it was released or not.
1142                          */
1143                         if (page->mapping != inode->i_mapping) {
1144                                 unlock_page(page);
1145                                 put_page(page);
1146                                 goto again;
1147                         }
1148                 }
1149
1150                 if (!PageUptodate(page)) {
1151                         btrfs_readpage(NULL, page);
1152                         lock_page(page);
1153                         if (!PageUptodate(page)) {
1154                                 unlock_page(page);
1155                                 put_page(page);
1156                                 ret = -EIO;
1157                                 break;
1158                         }
1159                 }
1160
1161                 if (page->mapping != inode->i_mapping) {
1162                         unlock_page(page);
1163                         put_page(page);
1164                         goto again;
1165                 }
1166
1167                 pages[i] = page;
1168                 i_done++;
1169         }
1170         if (!i_done || ret)
1171                 goto out;
1172
1173         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1174                 goto out;
1175
1176         /*
1177          * so now we have a nice long stream of locked
1178          * and up to date pages, lets wait on them
1179          */
1180         for (i = 0; i < i_done; i++)
1181                 wait_on_page_writeback(pages[i]);
1182
1183         page_start = page_offset(pages[0]);
1184         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1185
1186         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1187                          page_start, page_end - 1, &cached_state);
1188         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1189                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1190                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1191                           &cached_state);
1192
1193         if (i_done != page_cnt) {
1194                 spin_lock(&BTRFS_I(inode)->lock);
1195                 BTRFS_I(inode)->outstanding_extents++;
1196                 spin_unlock(&BTRFS_I(inode)->lock);
1197                 btrfs_delalloc_release_space(inode, data_reserved,
1198                                 start_index << PAGE_SHIFT,
1199                                 (page_cnt - i_done) << PAGE_SHIFT);
1200         }
1201
1202
1203         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1204                           &cached_state);
1205
1206         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1207                              page_start, page_end - 1, &cached_state);
1208
1209         for (i = 0; i < i_done; i++) {
1210                 clear_page_dirty_for_io(pages[i]);
1211                 ClearPageChecked(pages[i]);
1212                 set_page_extent_mapped(pages[i]);
1213                 set_page_dirty(pages[i]);
1214                 unlock_page(pages[i]);
1215                 put_page(pages[i]);
1216         }
1217         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1218         extent_changeset_free(data_reserved);
1219         return i_done;
1220 out:
1221         for (i = 0; i < i_done; i++) {
1222                 unlock_page(pages[i]);
1223                 put_page(pages[i]);
1224         }
1225         btrfs_delalloc_release_space(inode, data_reserved,
1226                         start_index << PAGE_SHIFT,
1227                         page_cnt << PAGE_SHIFT);
1228         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1229         extent_changeset_free(data_reserved);
1230         return ret;
1231
1232 }
1233
1234 int btrfs_defrag_file(struct inode *inode, struct file *file,
1235                       struct btrfs_ioctl_defrag_range_args *range,
1236                       u64 newer_than, unsigned long max_to_defrag)
1237 {
1238         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1239         struct btrfs_root *root = BTRFS_I(inode)->root;
1240         struct file_ra_state *ra = NULL;
1241         unsigned long last_index;
1242         u64 isize = i_size_read(inode);
1243         u64 last_len = 0;
1244         u64 skip = 0;
1245         u64 defrag_end = 0;
1246         u64 newer_off = range->start;
1247         unsigned long i;
1248         unsigned long ra_index = 0;
1249         int ret;
1250         int defrag_count = 0;
1251         int compress_type = BTRFS_COMPRESS_ZLIB;
1252         u32 extent_thresh = range->extent_thresh;
1253         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1254         unsigned long cluster = max_cluster;
1255         u64 new_align = ~((u64)SZ_128K - 1);
1256         struct page **pages = NULL;
1257         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1258
1259         if (isize == 0)
1260                 return 0;
1261
1262         if (range->start >= isize)
1263                 return -EINVAL;
1264
1265         if (do_compress) {
1266                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1267                         return -EINVAL;
1268                 if (range->compress_type)
1269                         compress_type = range->compress_type;
1270         }
1271
1272         if (extent_thresh == 0)
1273                 extent_thresh = SZ_256K;
1274
1275         /*
1276          * If we were not given a file, allocate a readahead context. As
1277          * readahead is just an optimization, defrag will work without it so
1278          * we don't error out.
1279          */
1280         if (!file) {
1281                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1282                 if (ra)
1283                         file_ra_state_init(ra, inode->i_mapping);
1284         } else {
1285                 ra = &file->f_ra;
1286         }
1287
1288         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1289         if (!pages) {
1290                 ret = -ENOMEM;
1291                 goto out_ra;
1292         }
1293
1294         /* find the last page to defrag */
1295         if (range->start + range->len > range->start) {
1296                 last_index = min_t(u64, isize - 1,
1297                          range->start + range->len - 1) >> PAGE_SHIFT;
1298         } else {
1299                 last_index = (isize - 1) >> PAGE_SHIFT;
1300         }
1301
1302         if (newer_than) {
1303                 ret = find_new_extents(root, inode, newer_than,
1304                                        &newer_off, SZ_64K);
1305                 if (!ret) {
1306                         range->start = newer_off;
1307                         /*
1308                          * we always align our defrag to help keep
1309                          * the extents in the file evenly spaced
1310                          */
1311                         i = (newer_off & new_align) >> PAGE_SHIFT;
1312                 } else
1313                         goto out_ra;
1314         } else {
1315                 i = range->start >> PAGE_SHIFT;
1316         }
1317         if (!max_to_defrag)
1318                 max_to_defrag = last_index - i + 1;
1319
1320         /*
1321          * make writeback starts from i, so the defrag range can be
1322          * written sequentially.
1323          */
1324         if (i < inode->i_mapping->writeback_index)
1325                 inode->i_mapping->writeback_index = i;
1326
1327         while (i <= last_index && defrag_count < max_to_defrag &&
1328                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1329                 /*
1330                  * make sure we stop running if someone unmounts
1331                  * the FS
1332                  */
1333                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1334                         break;
1335
1336                 if (btrfs_defrag_cancelled(fs_info)) {
1337                         btrfs_debug(fs_info, "defrag_file cancelled");
1338                         ret = -EAGAIN;
1339                         break;
1340                 }
1341
1342                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1343                                          extent_thresh, &last_len, &skip,
1344                                          &defrag_end, do_compress)){
1345                         unsigned long next;
1346                         /*
1347                          * the should_defrag function tells us how much to skip
1348                          * bump our counter by the suggested amount
1349                          */
1350                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1351                         i = max(i + 1, next);
1352                         continue;
1353                 }
1354
1355                 if (!newer_than) {
1356                         cluster = (PAGE_ALIGN(defrag_end) >>
1357                                    PAGE_SHIFT) - i;
1358                         cluster = min(cluster, max_cluster);
1359                 } else {
1360                         cluster = max_cluster;
1361                 }
1362
1363                 if (i + cluster > ra_index) {
1364                         ra_index = max(i, ra_index);
1365                         if (ra)
1366                                 page_cache_sync_readahead(inode->i_mapping, ra,
1367                                                 file, ra_index, cluster);
1368                         ra_index += cluster;
1369                 }
1370
1371                 inode_lock(inode);
1372                 if (do_compress)
1373                         BTRFS_I(inode)->defrag_compress = compress_type;
1374                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1375                 if (ret < 0) {
1376                         inode_unlock(inode);
1377                         goto out_ra;
1378                 }
1379
1380                 defrag_count += ret;
1381                 balance_dirty_pages_ratelimited(inode->i_mapping);
1382                 inode_unlock(inode);
1383
1384                 if (newer_than) {
1385                         if (newer_off == (u64)-1)
1386                                 break;
1387
1388                         if (ret > 0)
1389                                 i += ret;
1390
1391                         newer_off = max(newer_off + 1,
1392                                         (u64)i << PAGE_SHIFT);
1393
1394                         ret = find_new_extents(root, inode, newer_than,
1395                                                &newer_off, SZ_64K);
1396                         if (!ret) {
1397                                 range->start = newer_off;
1398                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1399                         } else {
1400                                 break;
1401                         }
1402                 } else {
1403                         if (ret > 0) {
1404                                 i += ret;
1405                                 last_len += ret << PAGE_SHIFT;
1406                         } else {
1407                                 i++;
1408                                 last_len = 0;
1409                         }
1410                 }
1411         }
1412
1413         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1414                 filemap_flush(inode->i_mapping);
1415                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1416                              &BTRFS_I(inode)->runtime_flags))
1417                         filemap_flush(inode->i_mapping);
1418         }
1419
1420         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1421                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1422         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1423                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1424         }
1425
1426         ret = defrag_count;
1427
1428 out_ra:
1429         if (do_compress) {
1430                 inode_lock(inode);
1431                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1432                 inode_unlock(inode);
1433         }
1434         if (!file)
1435                 kfree(ra);
1436         kfree(pages);
1437         return ret;
1438 }
1439
1440 static noinline int btrfs_ioctl_resize(struct file *file,
1441                                         void __user *arg)
1442 {
1443         struct inode *inode = file_inode(file);
1444         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1445         u64 new_size;
1446         u64 old_size;
1447         u64 devid = 1;
1448         struct btrfs_root *root = BTRFS_I(inode)->root;
1449         struct btrfs_ioctl_vol_args *vol_args;
1450         struct btrfs_trans_handle *trans;
1451         struct btrfs_device *device = NULL;
1452         char *sizestr;
1453         char *retptr;
1454         char *devstr = NULL;
1455         int ret = 0;
1456         int mod = 0;
1457
1458         if (!capable(CAP_SYS_ADMIN))
1459                 return -EPERM;
1460
1461         ret = mnt_want_write_file(file);
1462         if (ret)
1463                 return ret;
1464
1465         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1466                 mnt_drop_write_file(file);
1467                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1468         }
1469
1470         mutex_lock(&fs_info->volume_mutex);
1471         vol_args = memdup_user(arg, sizeof(*vol_args));
1472         if (IS_ERR(vol_args)) {
1473                 ret = PTR_ERR(vol_args);
1474                 goto out;
1475         }
1476
1477         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1478
1479         sizestr = vol_args->name;
1480         devstr = strchr(sizestr, ':');
1481         if (devstr) {
1482                 sizestr = devstr + 1;
1483                 *devstr = '\0';
1484                 devstr = vol_args->name;
1485                 ret = kstrtoull(devstr, 10, &devid);
1486                 if (ret)
1487                         goto out_free;
1488                 if (!devid) {
1489                         ret = -EINVAL;
1490                         goto out_free;
1491                 }
1492                 btrfs_info(fs_info, "resizing devid %llu", devid);
1493         }
1494
1495         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1496         if (!device) {
1497                 btrfs_info(fs_info, "resizer unable to find device %llu",
1498                            devid);
1499                 ret = -ENODEV;
1500                 goto out_free;
1501         }
1502
1503         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1504                 btrfs_info(fs_info,
1505                            "resizer unable to apply on readonly device %llu",
1506                        devid);
1507                 ret = -EPERM;
1508                 goto out_free;
1509         }
1510
1511         if (!strcmp(sizestr, "max"))
1512                 new_size = device->bdev->bd_inode->i_size;
1513         else {
1514                 if (sizestr[0] == '-') {
1515                         mod = -1;
1516                         sizestr++;
1517                 } else if (sizestr[0] == '+') {
1518                         mod = 1;
1519                         sizestr++;
1520                 }
1521                 new_size = memparse(sizestr, &retptr);
1522                 if (*retptr != '\0' || new_size == 0) {
1523                         ret = -EINVAL;
1524                         goto out_free;
1525                 }
1526         }
1527
1528         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1529                 ret = -EPERM;
1530                 goto out_free;
1531         }
1532
1533         old_size = btrfs_device_get_total_bytes(device);
1534
1535         if (mod < 0) {
1536                 if (new_size > old_size) {
1537                         ret = -EINVAL;
1538                         goto out_free;
1539                 }
1540                 new_size = old_size - new_size;
1541         } else if (mod > 0) {
1542                 if (new_size > ULLONG_MAX - old_size) {
1543                         ret = -ERANGE;
1544                         goto out_free;
1545                 }
1546                 new_size = old_size + new_size;
1547         }
1548
1549         if (new_size < SZ_256M) {
1550                 ret = -EINVAL;
1551                 goto out_free;
1552         }
1553         if (new_size > device->bdev->bd_inode->i_size) {
1554                 ret = -EFBIG;
1555                 goto out_free;
1556         }
1557
1558         new_size = round_down(new_size, fs_info->sectorsize);
1559
1560         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1561                           rcu_str_deref(device->name), new_size);
1562
1563         if (new_size > old_size) {
1564                 trans = btrfs_start_transaction(root, 0);
1565                 if (IS_ERR(trans)) {
1566                         ret = PTR_ERR(trans);
1567                         goto out_free;
1568                 }
1569                 ret = btrfs_grow_device(trans, device, new_size);
1570                 btrfs_commit_transaction(trans);
1571         } else if (new_size < old_size) {
1572                 ret = btrfs_shrink_device(device, new_size);
1573         } /* equal, nothing need to do */
1574
1575 out_free:
1576         kfree(vol_args);
1577 out:
1578         mutex_unlock(&fs_info->volume_mutex);
1579         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1580         mnt_drop_write_file(file);
1581         return ret;
1582 }
1583
1584 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1585                                 const char *name, unsigned long fd, int subvol,
1586                                 u64 *transid, bool readonly,
1587                                 struct btrfs_qgroup_inherit *inherit)
1588 {
1589         int namelen;
1590         int ret = 0;
1591
1592         if (!S_ISDIR(file_inode(file)->i_mode))
1593                 return -ENOTDIR;
1594
1595         ret = mnt_want_write_file(file);
1596         if (ret)
1597                 goto out;
1598
1599         namelen = strlen(name);
1600         if (strchr(name, '/')) {
1601                 ret = -EINVAL;
1602                 goto out_drop_write;
1603         }
1604
1605         if (name[0] == '.' &&
1606            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1607                 ret = -EEXIST;
1608                 goto out_drop_write;
1609         }
1610
1611         if (subvol) {
1612                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1613                                      NULL, transid, readonly, inherit);
1614         } else {
1615                 struct fd src = fdget(fd);
1616                 struct inode *src_inode;
1617                 if (!src.file) {
1618                         ret = -EINVAL;
1619                         goto out_drop_write;
1620                 }
1621
1622                 src_inode = file_inode(src.file);
1623                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1624                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1625                                    "Snapshot src from another FS");
1626                         ret = -EXDEV;
1627                 } else if (!inode_owner_or_capable(src_inode)) {
1628                         /*
1629                          * Subvolume creation is not restricted, but snapshots
1630                          * are limited to own subvolumes only
1631                          */
1632                         ret = -EPERM;
1633                 } else {
1634                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1635                                              BTRFS_I(src_inode)->root,
1636                                              transid, readonly, inherit);
1637                 }
1638                 fdput(src);
1639         }
1640 out_drop_write:
1641         mnt_drop_write_file(file);
1642 out:
1643         return ret;
1644 }
1645
1646 static noinline int btrfs_ioctl_snap_create(struct file *file,
1647                                             void __user *arg, int subvol)
1648 {
1649         struct btrfs_ioctl_vol_args *vol_args;
1650         int ret;
1651
1652         if (!S_ISDIR(file_inode(file)->i_mode))
1653                 return -ENOTDIR;
1654
1655         vol_args = memdup_user(arg, sizeof(*vol_args));
1656         if (IS_ERR(vol_args))
1657                 return PTR_ERR(vol_args);
1658         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1659
1660         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1661                                               vol_args->fd, subvol,
1662                                               NULL, false, NULL);
1663
1664         kfree(vol_args);
1665         return ret;
1666 }
1667
1668 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1669                                                void __user *arg, int subvol)
1670 {
1671         struct btrfs_ioctl_vol_args_v2 *vol_args;
1672         int ret;
1673         u64 transid = 0;
1674         u64 *ptr = NULL;
1675         bool readonly = false;
1676         struct btrfs_qgroup_inherit *inherit = NULL;
1677
1678         if (!S_ISDIR(file_inode(file)->i_mode))
1679                 return -ENOTDIR;
1680
1681         vol_args = memdup_user(arg, sizeof(*vol_args));
1682         if (IS_ERR(vol_args))
1683                 return PTR_ERR(vol_args);
1684         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1685
1686         if (vol_args->flags &
1687             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1688               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1689                 ret = -EOPNOTSUPP;
1690                 goto free_args;
1691         }
1692
1693         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1694                 ptr = &transid;
1695         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1696                 readonly = true;
1697         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1698                 if (vol_args->size > PAGE_SIZE) {
1699                         ret = -EINVAL;
1700                         goto free_args;
1701                 }
1702                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1703                 if (IS_ERR(inherit)) {
1704                         ret = PTR_ERR(inherit);
1705                         goto free_args;
1706                 }
1707         }
1708
1709         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1710                                               vol_args->fd, subvol, ptr,
1711                                               readonly, inherit);
1712         if (ret)
1713                 goto free_inherit;
1714
1715         if (ptr && copy_to_user(arg +
1716                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1717                                         transid),
1718                                 ptr, sizeof(*ptr)))
1719                 ret = -EFAULT;
1720
1721 free_inherit:
1722         kfree(inherit);
1723 free_args:
1724         kfree(vol_args);
1725         return ret;
1726 }
1727
1728 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1729                                                 void __user *arg)
1730 {
1731         struct inode *inode = file_inode(file);
1732         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1733         struct btrfs_root *root = BTRFS_I(inode)->root;
1734         int ret = 0;
1735         u64 flags = 0;
1736
1737         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1738                 return -EINVAL;
1739
1740         down_read(&fs_info->subvol_sem);
1741         if (btrfs_root_readonly(root))
1742                 flags |= BTRFS_SUBVOL_RDONLY;
1743         up_read(&fs_info->subvol_sem);
1744
1745         if (copy_to_user(arg, &flags, sizeof(flags)))
1746                 ret = -EFAULT;
1747
1748         return ret;
1749 }
1750
1751 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1752                                               void __user *arg)
1753 {
1754         struct inode *inode = file_inode(file);
1755         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1756         struct btrfs_root *root = BTRFS_I(inode)->root;
1757         struct btrfs_trans_handle *trans;
1758         u64 root_flags;
1759         u64 flags;
1760         int ret = 0;
1761
1762         if (!inode_owner_or_capable(inode))
1763                 return -EPERM;
1764
1765         ret = mnt_want_write_file(file);
1766         if (ret)
1767                 goto out;
1768
1769         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1770                 ret = -EINVAL;
1771                 goto out_drop_write;
1772         }
1773
1774         if (copy_from_user(&flags, arg, sizeof(flags))) {
1775                 ret = -EFAULT;
1776                 goto out_drop_write;
1777         }
1778
1779         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1780                 ret = -EINVAL;
1781                 goto out_drop_write;
1782         }
1783
1784         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1785                 ret = -EOPNOTSUPP;
1786                 goto out_drop_write;
1787         }
1788
1789         down_write(&fs_info->subvol_sem);
1790
1791         /* nothing to do */
1792         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1793                 goto out_drop_sem;
1794
1795         root_flags = btrfs_root_flags(&root->root_item);
1796         if (flags & BTRFS_SUBVOL_RDONLY) {
1797                 btrfs_set_root_flags(&root->root_item,
1798                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1799         } else {
1800                 /*
1801                  * Block RO -> RW transition if this subvolume is involved in
1802                  * send
1803                  */
1804                 spin_lock(&root->root_item_lock);
1805                 if (root->send_in_progress == 0) {
1806                         btrfs_set_root_flags(&root->root_item,
1807                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1808                         spin_unlock(&root->root_item_lock);
1809                 } else {
1810                         spin_unlock(&root->root_item_lock);
1811                         btrfs_warn(fs_info,
1812                                    "Attempt to set subvolume %llu read-write during send",
1813                                    root->root_key.objectid);
1814                         ret = -EPERM;
1815                         goto out_drop_sem;
1816                 }
1817         }
1818
1819         trans = btrfs_start_transaction(root, 1);
1820         if (IS_ERR(trans)) {
1821                 ret = PTR_ERR(trans);
1822                 goto out_reset;
1823         }
1824
1825         ret = btrfs_update_root(trans, fs_info->tree_root,
1826                                 &root->root_key, &root->root_item);
1827         if (ret < 0) {
1828                 btrfs_end_transaction(trans);
1829                 goto out_reset;
1830         }
1831
1832         ret = btrfs_commit_transaction(trans);
1833
1834 out_reset:
1835         if (ret)
1836                 btrfs_set_root_flags(&root->root_item, root_flags);
1837 out_drop_sem:
1838         up_write(&fs_info->subvol_sem);
1839 out_drop_write:
1840         mnt_drop_write_file(file);
1841 out:
1842         return ret;
1843 }
1844
1845 /*
1846  * helper to check if the subvolume references other subvolumes
1847  */
1848 static noinline int may_destroy_subvol(struct btrfs_root *root)
1849 {
1850         struct btrfs_fs_info *fs_info = root->fs_info;
1851         struct btrfs_path *path;
1852         struct btrfs_dir_item *di;
1853         struct btrfs_key key;
1854         u64 dir_id;
1855         int ret;
1856
1857         path = btrfs_alloc_path();
1858         if (!path)
1859                 return -ENOMEM;
1860
1861         /* Make sure this root isn't set as the default subvol */
1862         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1863         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1864                                    dir_id, "default", 7, 0);
1865         if (di && !IS_ERR(di)) {
1866                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1867                 if (key.objectid == root->root_key.objectid) {
1868                         ret = -EPERM;
1869                         btrfs_err(fs_info,
1870                                   "deleting default subvolume %llu is not allowed",
1871                                   key.objectid);
1872                         goto out;
1873                 }
1874                 btrfs_release_path(path);
1875         }
1876
1877         key.objectid = root->root_key.objectid;
1878         key.type = BTRFS_ROOT_REF_KEY;
1879         key.offset = (u64)-1;
1880
1881         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1882         if (ret < 0)
1883                 goto out;
1884         BUG_ON(ret == 0);
1885
1886         ret = 0;
1887         if (path->slots[0] > 0) {
1888                 path->slots[0]--;
1889                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1890                 if (key.objectid == root->root_key.objectid &&
1891                     key.type == BTRFS_ROOT_REF_KEY)
1892                         ret = -ENOTEMPTY;
1893         }
1894 out:
1895         btrfs_free_path(path);
1896         return ret;
1897 }
1898
1899 static noinline int key_in_sk(struct btrfs_key *key,
1900                               struct btrfs_ioctl_search_key *sk)
1901 {
1902         struct btrfs_key test;
1903         int ret;
1904
1905         test.objectid = sk->min_objectid;
1906         test.type = sk->min_type;
1907         test.offset = sk->min_offset;
1908
1909         ret = btrfs_comp_cpu_keys(key, &test);
1910         if (ret < 0)
1911                 return 0;
1912
1913         test.objectid = sk->max_objectid;
1914         test.type = sk->max_type;
1915         test.offset = sk->max_offset;
1916
1917         ret = btrfs_comp_cpu_keys(key, &test);
1918         if (ret > 0)
1919                 return 0;
1920         return 1;
1921 }
1922
1923 static noinline int copy_to_sk(struct btrfs_path *path,
1924                                struct btrfs_key *key,
1925                                struct btrfs_ioctl_search_key *sk,
1926                                size_t *buf_size,
1927                                char __user *ubuf,
1928                                unsigned long *sk_offset,
1929                                int *num_found)
1930 {
1931         u64 found_transid;
1932         struct extent_buffer *leaf;
1933         struct btrfs_ioctl_search_header sh;
1934         struct btrfs_key test;
1935         unsigned long item_off;
1936         unsigned long item_len;
1937         int nritems;
1938         int i;
1939         int slot;
1940         int ret = 0;
1941
1942         leaf = path->nodes[0];
1943         slot = path->slots[0];
1944         nritems = btrfs_header_nritems(leaf);
1945
1946         if (btrfs_header_generation(leaf) > sk->max_transid) {
1947                 i = nritems;
1948                 goto advance_key;
1949         }
1950         found_transid = btrfs_header_generation(leaf);
1951
1952         for (i = slot; i < nritems; i++) {
1953                 item_off = btrfs_item_ptr_offset(leaf, i);
1954                 item_len = btrfs_item_size_nr(leaf, i);
1955
1956                 btrfs_item_key_to_cpu(leaf, key, i);
1957                 if (!key_in_sk(key, sk))
1958                         continue;
1959
1960                 if (sizeof(sh) + item_len > *buf_size) {
1961                         if (*num_found) {
1962                                 ret = 1;
1963                                 goto out;
1964                         }
1965
1966                         /*
1967                          * return one empty item back for v1, which does not
1968                          * handle -EOVERFLOW
1969                          */
1970
1971                         *buf_size = sizeof(sh) + item_len;
1972                         item_len = 0;
1973                         ret = -EOVERFLOW;
1974                 }
1975
1976                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1977                         ret = 1;
1978                         goto out;
1979                 }
1980
1981                 sh.objectid = key->objectid;
1982                 sh.offset = key->offset;
1983                 sh.type = key->type;
1984                 sh.len = item_len;
1985                 sh.transid = found_transid;
1986
1987                 /* copy search result header */
1988                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1989                         ret = -EFAULT;
1990                         goto out;
1991                 }
1992
1993                 *sk_offset += sizeof(sh);
1994
1995                 if (item_len) {
1996                         char __user *up = ubuf + *sk_offset;
1997                         /* copy the item */
1998                         if (read_extent_buffer_to_user(leaf, up,
1999                                                        item_off, item_len)) {
2000                                 ret = -EFAULT;
2001                                 goto out;
2002                         }
2003
2004                         *sk_offset += item_len;
2005                 }
2006                 (*num_found)++;
2007
2008                 if (ret) /* -EOVERFLOW from above */
2009                         goto out;
2010
2011                 if (*num_found >= sk->nr_items) {
2012                         ret = 1;
2013                         goto out;
2014                 }
2015         }
2016 advance_key:
2017         ret = 0;
2018         test.objectid = sk->max_objectid;
2019         test.type = sk->max_type;
2020         test.offset = sk->max_offset;
2021         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2022                 ret = 1;
2023         else if (key->offset < (u64)-1)
2024                 key->offset++;
2025         else if (key->type < (u8)-1) {
2026                 key->offset = 0;
2027                 key->type++;
2028         } else if (key->objectid < (u64)-1) {
2029                 key->offset = 0;
2030                 key->type = 0;
2031                 key->objectid++;
2032         } else
2033                 ret = 1;
2034 out:
2035         /*
2036          *  0: all items from this leaf copied, continue with next
2037          *  1: * more items can be copied, but unused buffer is too small
2038          *     * all items were found
2039          *     Either way, it will stops the loop which iterates to the next
2040          *     leaf
2041          *  -EOVERFLOW: item was to large for buffer
2042          *  -EFAULT: could not copy extent buffer back to userspace
2043          */
2044         return ret;
2045 }
2046
2047 static noinline int search_ioctl(struct inode *inode,
2048                                  struct btrfs_ioctl_search_key *sk,
2049                                  size_t *buf_size,
2050                                  char __user *ubuf)
2051 {
2052         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2053         struct btrfs_root *root;
2054         struct btrfs_key key;
2055         struct btrfs_path *path;
2056         int ret;
2057         int num_found = 0;
2058         unsigned long sk_offset = 0;
2059
2060         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2061                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2062                 return -EOVERFLOW;
2063         }
2064
2065         path = btrfs_alloc_path();
2066         if (!path)
2067                 return -ENOMEM;
2068
2069         if (sk->tree_id == 0) {
2070                 /* search the root of the inode that was passed */
2071                 root = BTRFS_I(inode)->root;
2072         } else {
2073                 key.objectid = sk->tree_id;
2074                 key.type = BTRFS_ROOT_ITEM_KEY;
2075                 key.offset = (u64)-1;
2076                 root = btrfs_read_fs_root_no_name(info, &key);
2077                 if (IS_ERR(root)) {
2078                         btrfs_free_path(path);
2079                         return -ENOENT;
2080                 }
2081         }
2082
2083         key.objectid = sk->min_objectid;
2084         key.type = sk->min_type;
2085         key.offset = sk->min_offset;
2086
2087         while (1) {
2088                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2089                 if (ret != 0) {
2090                         if (ret > 0)
2091                                 ret = 0;
2092                         goto err;
2093                 }
2094                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2095                                  &sk_offset, &num_found);
2096                 btrfs_release_path(path);
2097                 if (ret)
2098                         break;
2099
2100         }
2101         if (ret > 0)
2102                 ret = 0;
2103 err:
2104         sk->nr_items = num_found;
2105         btrfs_free_path(path);
2106         return ret;
2107 }
2108
2109 static noinline int btrfs_ioctl_tree_search(struct file *file,
2110                                            void __user *argp)
2111 {
2112         struct btrfs_ioctl_search_args __user *uargs;
2113         struct btrfs_ioctl_search_key sk;
2114         struct inode *inode;
2115         int ret;
2116         size_t buf_size;
2117
2118         if (!capable(CAP_SYS_ADMIN))
2119                 return -EPERM;
2120
2121         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2122
2123         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2124                 return -EFAULT;
2125
2126         buf_size = sizeof(uargs->buf);
2127
2128         inode = file_inode(file);
2129         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2130
2131         /*
2132          * In the origin implementation an overflow is handled by returning a
2133          * search header with a len of zero, so reset ret.
2134          */
2135         if (ret == -EOVERFLOW)
2136                 ret = 0;
2137
2138         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2139                 ret = -EFAULT;
2140         return ret;
2141 }
2142
2143 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2144                                                void __user *argp)
2145 {
2146         struct btrfs_ioctl_search_args_v2 __user *uarg;
2147         struct btrfs_ioctl_search_args_v2 args;
2148         struct inode *inode;
2149         int ret;
2150         size_t buf_size;
2151         const size_t buf_limit = SZ_16M;
2152
2153         if (!capable(CAP_SYS_ADMIN))
2154                 return -EPERM;
2155
2156         /* copy search header and buffer size */
2157         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2158         if (copy_from_user(&args, uarg, sizeof(args)))
2159                 return -EFAULT;
2160
2161         buf_size = args.buf_size;
2162
2163         /* limit result size to 16MB */
2164         if (buf_size > buf_limit)
2165                 buf_size = buf_limit;
2166
2167         inode = file_inode(file);
2168         ret = search_ioctl(inode, &args.key, &buf_size,
2169                            (char __user *)(&uarg->buf[0]));
2170         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2171                 ret = -EFAULT;
2172         else if (ret == -EOVERFLOW &&
2173                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2174                 ret = -EFAULT;
2175
2176         return ret;
2177 }
2178
2179 /*
2180  * Search INODE_REFs to identify path name of 'dirid' directory
2181  * in a 'tree_id' tree. and sets path name to 'name'.
2182  */
2183 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2184                                 u64 tree_id, u64 dirid, char *name)
2185 {
2186         struct btrfs_root *root;
2187         struct btrfs_key key;
2188         char *ptr;
2189         int ret = -1;
2190         int slot;
2191         int len;
2192         int total_len = 0;
2193         struct btrfs_inode_ref *iref;
2194         struct extent_buffer *l;
2195         struct btrfs_path *path;
2196
2197         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2198                 name[0]='\0';
2199                 return 0;
2200         }
2201
2202         path = btrfs_alloc_path();
2203         if (!path)
2204                 return -ENOMEM;
2205
2206         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2207
2208         key.objectid = tree_id;
2209         key.type = BTRFS_ROOT_ITEM_KEY;
2210         key.offset = (u64)-1;
2211         root = btrfs_read_fs_root_no_name(info, &key);
2212         if (IS_ERR(root)) {
2213                 btrfs_err(info, "could not find root %llu", tree_id);
2214                 ret = -ENOENT;
2215                 goto out;
2216         }
2217
2218         key.objectid = dirid;
2219         key.type = BTRFS_INODE_REF_KEY;
2220         key.offset = (u64)-1;
2221
2222         while (1) {
2223                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2224                 if (ret < 0)
2225                         goto out;
2226                 else if (ret > 0) {
2227                         ret = btrfs_previous_item(root, path, dirid,
2228                                                   BTRFS_INODE_REF_KEY);
2229                         if (ret < 0)
2230                                 goto out;
2231                         else if (ret > 0) {
2232                                 ret = -ENOENT;
2233                                 goto out;
2234                         }
2235                 }
2236
2237                 l = path->nodes[0];
2238                 slot = path->slots[0];
2239                 btrfs_item_key_to_cpu(l, &key, slot);
2240
2241                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2242                 len = btrfs_inode_ref_name_len(l, iref);
2243                 ptr -= len + 1;
2244                 total_len += len + 1;
2245                 if (ptr < name) {
2246                         ret = -ENAMETOOLONG;
2247                         goto out;
2248                 }
2249
2250                 *(ptr + len) = '/';
2251                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2252
2253                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2254                         break;
2255
2256                 btrfs_release_path(path);
2257                 key.objectid = key.offset;
2258                 key.offset = (u64)-1;
2259                 dirid = key.objectid;
2260         }
2261         memmove(name, ptr, total_len);
2262         name[total_len] = '\0';
2263         ret = 0;
2264 out:
2265         btrfs_free_path(path);
2266         return ret;
2267 }
2268
2269 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2270                                            void __user *argp)
2271 {
2272          struct btrfs_ioctl_ino_lookup_args *args;
2273          struct inode *inode;
2274         int ret = 0;
2275
2276         args = memdup_user(argp, sizeof(*args));
2277         if (IS_ERR(args))
2278                 return PTR_ERR(args);
2279
2280         inode = file_inode(file);
2281
2282         /*
2283          * Unprivileged query to obtain the containing subvolume root id. The
2284          * path is reset so it's consistent with btrfs_search_path_in_tree.
2285          */
2286         if (args->treeid == 0)
2287                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2288
2289         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2290                 args->name[0] = 0;
2291                 goto out;
2292         }
2293
2294         if (!capable(CAP_SYS_ADMIN)) {
2295                 ret = -EPERM;
2296                 goto out;
2297         }
2298
2299         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2300                                         args->treeid, args->objectid,
2301                                         args->name);
2302
2303 out:
2304         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2305                 ret = -EFAULT;
2306
2307         kfree(args);
2308         return ret;
2309 }
2310
2311 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2312                                              void __user *arg)
2313 {
2314         struct dentry *parent = file->f_path.dentry;
2315         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2316         struct dentry *dentry;
2317         struct inode *dir = d_inode(parent);
2318         struct inode *inode;
2319         struct btrfs_root *root = BTRFS_I(dir)->root;
2320         struct btrfs_root *dest = NULL;
2321         struct btrfs_ioctl_vol_args *vol_args;
2322         struct btrfs_trans_handle *trans;
2323         struct btrfs_block_rsv block_rsv;
2324         u64 root_flags;
2325         u64 qgroup_reserved;
2326         int namelen;
2327         int ret;
2328         int err = 0;
2329
2330         if (!S_ISDIR(dir->i_mode))
2331                 return -ENOTDIR;
2332
2333         vol_args = memdup_user(arg, sizeof(*vol_args));
2334         if (IS_ERR(vol_args))
2335                 return PTR_ERR(vol_args);
2336
2337         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2338         namelen = strlen(vol_args->name);
2339         if (strchr(vol_args->name, '/') ||
2340             strncmp(vol_args->name, "..", namelen) == 0) {
2341                 err = -EINVAL;
2342                 goto out;
2343         }
2344
2345         err = mnt_want_write_file(file);
2346         if (err)
2347                 goto out;
2348
2349
2350         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2351         if (err == -EINTR)
2352                 goto out_drop_write;
2353         dentry = lookup_one_len(vol_args->name, parent, namelen);
2354         if (IS_ERR(dentry)) {
2355                 err = PTR_ERR(dentry);
2356                 goto out_unlock_dir;
2357         }
2358
2359         if (d_really_is_negative(dentry)) {
2360                 err = -ENOENT;
2361                 goto out_dput;
2362         }
2363
2364         inode = d_inode(dentry);
2365         dest = BTRFS_I(inode)->root;
2366         if (!capable(CAP_SYS_ADMIN)) {
2367                 /*
2368                  * Regular user.  Only allow this with a special mount
2369                  * option, when the user has write+exec access to the
2370                  * subvol root, and when rmdir(2) would have been
2371                  * allowed.
2372                  *
2373                  * Note that this is _not_ check that the subvol is
2374                  * empty or doesn't contain data that we wouldn't
2375                  * otherwise be able to delete.
2376                  *
2377                  * Users who want to delete empty subvols should try
2378                  * rmdir(2).
2379                  */
2380                 err = -EPERM;
2381                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2382                         goto out_dput;
2383
2384                 /*
2385                  * Do not allow deletion if the parent dir is the same
2386                  * as the dir to be deleted.  That means the ioctl
2387                  * must be called on the dentry referencing the root
2388                  * of the subvol, not a random directory contained
2389                  * within it.
2390                  */
2391                 err = -EINVAL;
2392                 if (root == dest)
2393                         goto out_dput;
2394
2395                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2396                 if (err)
2397                         goto out_dput;
2398         }
2399
2400         /* check if subvolume may be deleted by a user */
2401         err = btrfs_may_delete(dir, dentry, 1);
2402         if (err)
2403                 goto out_dput;
2404
2405         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2406                 err = -EINVAL;
2407                 goto out_dput;
2408         }
2409
2410         inode_lock(inode);
2411
2412         /*
2413          * Don't allow to delete a subvolume with send in progress. This is
2414          * inside the i_mutex so the error handling that has to drop the bit
2415          * again is not run concurrently.
2416          */
2417         spin_lock(&dest->root_item_lock);
2418         root_flags = btrfs_root_flags(&dest->root_item);
2419         if (dest->send_in_progress == 0) {
2420                 btrfs_set_root_flags(&dest->root_item,
2421                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2422                 spin_unlock(&dest->root_item_lock);
2423         } else {
2424                 spin_unlock(&dest->root_item_lock);
2425                 btrfs_warn(fs_info,
2426                            "Attempt to delete subvolume %llu during send",
2427                            dest->root_key.objectid);
2428                 err = -EPERM;
2429                 goto out_unlock_inode;
2430         }
2431
2432         down_write(&fs_info->subvol_sem);
2433
2434         err = may_destroy_subvol(dest);
2435         if (err)
2436                 goto out_up_write;
2437
2438         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2439         /*
2440          * One for dir inode, two for dir entries, two for root
2441          * ref/backref.
2442          */
2443         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2444                                                5, &qgroup_reserved, true);
2445         if (err)
2446                 goto out_up_write;
2447
2448         trans = btrfs_start_transaction(root, 0);
2449         if (IS_ERR(trans)) {
2450                 err = PTR_ERR(trans);
2451                 goto out_release;
2452         }
2453         trans->block_rsv = &block_rsv;
2454         trans->bytes_reserved = block_rsv.size;
2455
2456         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2457
2458         ret = btrfs_unlink_subvol(trans, root, dir,
2459                                 dest->root_key.objectid,
2460                                 dentry->d_name.name,
2461                                 dentry->d_name.len);
2462         if (ret) {
2463                 err = ret;
2464                 btrfs_abort_transaction(trans, ret);
2465                 goto out_end_trans;
2466         }
2467
2468         btrfs_record_root_in_trans(trans, dest);
2469
2470         memset(&dest->root_item.drop_progress, 0,
2471                 sizeof(dest->root_item.drop_progress));
2472         dest->root_item.drop_level = 0;
2473         btrfs_set_root_refs(&dest->root_item, 0);
2474
2475         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2476                 ret = btrfs_insert_orphan_item(trans,
2477                                         fs_info->tree_root,
2478                                         dest->root_key.objectid);
2479                 if (ret) {
2480                         btrfs_abort_transaction(trans, ret);
2481                         err = ret;
2482                         goto out_end_trans;
2483                 }
2484         }
2485
2486         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2487                                   BTRFS_UUID_KEY_SUBVOL,
2488                                   dest->root_key.objectid);
2489         if (ret && ret != -ENOENT) {
2490                 btrfs_abort_transaction(trans, ret);
2491                 err = ret;
2492                 goto out_end_trans;
2493         }
2494         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2495                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2496                                           dest->root_item.received_uuid,
2497                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2498                                           dest->root_key.objectid);
2499                 if (ret && ret != -ENOENT) {
2500                         btrfs_abort_transaction(trans, ret);
2501                         err = ret;
2502                         goto out_end_trans;
2503                 }
2504         }
2505
2506 out_end_trans:
2507         trans->block_rsv = NULL;
2508         trans->bytes_reserved = 0;
2509         ret = btrfs_end_transaction(trans);
2510         if (ret && !err)
2511                 err = ret;
2512         inode->i_flags |= S_DEAD;
2513 out_release:
2514         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2515 out_up_write:
2516         up_write(&fs_info->subvol_sem);
2517         if (err) {
2518                 spin_lock(&dest->root_item_lock);
2519                 root_flags = btrfs_root_flags(&dest->root_item);
2520                 btrfs_set_root_flags(&dest->root_item,
2521                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2522                 spin_unlock(&dest->root_item_lock);
2523         }
2524 out_unlock_inode:
2525         inode_unlock(inode);
2526         if (!err) {
2527                 d_invalidate(dentry);
2528                 btrfs_invalidate_inodes(dest);
2529                 d_delete(dentry);
2530                 ASSERT(dest->send_in_progress == 0);
2531
2532                 /* the last ref */
2533                 if (dest->ino_cache_inode) {
2534                         iput(dest->ino_cache_inode);
2535                         dest->ino_cache_inode = NULL;
2536                 }
2537         }
2538 out_dput:
2539         dput(dentry);
2540 out_unlock_dir:
2541         inode_unlock(dir);
2542 out_drop_write:
2543         mnt_drop_write_file(file);
2544 out:
2545         kfree(vol_args);
2546         return err;
2547 }
2548
2549 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2550 {
2551         struct inode *inode = file_inode(file);
2552         struct btrfs_root *root = BTRFS_I(inode)->root;
2553         struct btrfs_ioctl_defrag_range_args *range;
2554         int ret;
2555
2556         ret = mnt_want_write_file(file);
2557         if (ret)
2558                 return ret;
2559
2560         if (btrfs_root_readonly(root)) {
2561                 ret = -EROFS;
2562                 goto out;
2563         }
2564
2565         switch (inode->i_mode & S_IFMT) {
2566         case S_IFDIR:
2567                 if (!capable(CAP_SYS_ADMIN)) {
2568                         ret = -EPERM;
2569                         goto out;
2570                 }
2571                 ret = btrfs_defrag_root(root);
2572                 break;
2573         case S_IFREG:
2574                 if (!(file->f_mode & FMODE_WRITE)) {
2575                         ret = -EINVAL;
2576                         goto out;
2577                 }
2578
2579                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2580                 if (!range) {
2581                         ret = -ENOMEM;
2582                         goto out;
2583                 }
2584
2585                 if (argp) {
2586                         if (copy_from_user(range, argp,
2587                                            sizeof(*range))) {
2588                                 ret = -EFAULT;
2589                                 kfree(range);
2590                                 goto out;
2591                         }
2592                         /* compression requires us to start the IO */
2593                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2594                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2595                                 range->extent_thresh = (u32)-1;
2596                         }
2597                 } else {
2598                         /* the rest are all set to zero by kzalloc */
2599                         range->len = (u64)-1;
2600                 }
2601                 ret = btrfs_defrag_file(file_inode(file), file,
2602                                         range, 0, 0);
2603                 if (ret > 0)
2604                         ret = 0;
2605                 kfree(range);
2606                 break;
2607         default:
2608                 ret = -EINVAL;
2609         }
2610 out:
2611         mnt_drop_write_file(file);
2612         return ret;
2613 }
2614
2615 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2616 {
2617         struct btrfs_ioctl_vol_args *vol_args;
2618         int ret;
2619
2620         if (!capable(CAP_SYS_ADMIN))
2621                 return -EPERM;
2622
2623         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2624                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2625
2626         mutex_lock(&fs_info->volume_mutex);
2627         vol_args = memdup_user(arg, sizeof(*vol_args));
2628         if (IS_ERR(vol_args)) {
2629                 ret = PTR_ERR(vol_args);
2630                 goto out;
2631         }
2632
2633         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2634         ret = btrfs_init_new_device(fs_info, vol_args->name);
2635
2636         if (!ret)
2637                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2638
2639         kfree(vol_args);
2640 out:
2641         mutex_unlock(&fs_info->volume_mutex);
2642         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2643         return ret;
2644 }
2645
2646 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2647 {
2648         struct inode *inode = file_inode(file);
2649         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2650         struct btrfs_ioctl_vol_args_v2 *vol_args;
2651         int ret;
2652
2653         if (!capable(CAP_SYS_ADMIN))
2654                 return -EPERM;
2655
2656         ret = mnt_want_write_file(file);
2657         if (ret)
2658                 return ret;
2659
2660         vol_args = memdup_user(arg, sizeof(*vol_args));
2661         if (IS_ERR(vol_args)) {
2662                 ret = PTR_ERR(vol_args);
2663                 goto err_drop;
2664         }
2665
2666         /* Check for compatibility reject unknown flags */
2667         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2668                 return -EOPNOTSUPP;
2669
2670         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2671                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2672                 goto out;
2673         }
2674
2675         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2676                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2677         } else {
2678                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2679                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2680         }
2681         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2682
2683         if (!ret) {
2684                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2685                         btrfs_info(fs_info, "device deleted: id %llu",
2686                                         vol_args->devid);
2687                 else
2688                         btrfs_info(fs_info, "device deleted: %s",
2689                                         vol_args->name);
2690         }
2691 out:
2692         kfree(vol_args);
2693 err_drop:
2694         mnt_drop_write_file(file);
2695         return ret;
2696 }
2697
2698 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2699 {
2700         struct inode *inode = file_inode(file);
2701         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2702         struct btrfs_ioctl_vol_args *vol_args;
2703         int ret;
2704
2705         if (!capable(CAP_SYS_ADMIN))
2706                 return -EPERM;
2707
2708         ret = mnt_want_write_file(file);
2709         if (ret)
2710                 return ret;
2711
2712         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2713                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2714                 goto out_drop_write;
2715         }
2716
2717         vol_args = memdup_user(arg, sizeof(*vol_args));
2718         if (IS_ERR(vol_args)) {
2719                 ret = PTR_ERR(vol_args);
2720                 goto out;
2721         }
2722
2723         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2724         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2725
2726         if (!ret)
2727                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2728         kfree(vol_args);
2729 out:
2730         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2731 out_drop_write:
2732         mnt_drop_write_file(file);
2733
2734         return ret;
2735 }
2736
2737 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2738                                 void __user *arg)
2739 {
2740         struct btrfs_ioctl_fs_info_args *fi_args;
2741         struct btrfs_device *device;
2742         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2743         int ret = 0;
2744
2745         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2746         if (!fi_args)
2747                 return -ENOMEM;
2748
2749         rcu_read_lock();
2750         fi_args->num_devices = fs_devices->num_devices;
2751
2752         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2753                 if (device->devid > fi_args->max_id)
2754                         fi_args->max_id = device->devid;
2755         }
2756         rcu_read_unlock();
2757
2758         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2759         fi_args->nodesize = fs_info->nodesize;
2760         fi_args->sectorsize = fs_info->sectorsize;
2761         fi_args->clone_alignment = fs_info->sectorsize;
2762
2763         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2764                 ret = -EFAULT;
2765
2766         kfree(fi_args);
2767         return ret;
2768 }
2769
2770 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2771                                  void __user *arg)
2772 {
2773         struct btrfs_ioctl_dev_info_args *di_args;
2774         struct btrfs_device *dev;
2775         int ret = 0;
2776         char *s_uuid = NULL;
2777
2778         di_args = memdup_user(arg, sizeof(*di_args));
2779         if (IS_ERR(di_args))
2780                 return PTR_ERR(di_args);
2781
2782         if (!btrfs_is_empty_uuid(di_args->uuid))
2783                 s_uuid = di_args->uuid;
2784
2785         rcu_read_lock();
2786         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2787
2788         if (!dev) {
2789                 ret = -ENODEV;
2790                 goto out;
2791         }
2792
2793         di_args->devid = dev->devid;
2794         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2795         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2796         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2797         if (dev->name) {
2798                 struct rcu_string *name;
2799
2800                 name = rcu_dereference(dev->name);
2801                 strncpy(di_args->path, name->str, sizeof(di_args->path) - 1);
2802                 di_args->path[sizeof(di_args->path) - 1] = 0;
2803         } else {
2804                 di_args->path[0] = '\0';
2805         }
2806
2807 out:
2808         rcu_read_unlock();
2809         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2810                 ret = -EFAULT;
2811
2812         kfree(di_args);
2813         return ret;
2814 }
2815
2816 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2817 {
2818         struct page *page;
2819
2820         page = grab_cache_page(inode->i_mapping, index);
2821         if (!page)
2822                 return ERR_PTR(-ENOMEM);
2823
2824         if (!PageUptodate(page)) {
2825                 int ret;
2826
2827                 ret = btrfs_readpage(NULL, page);
2828                 if (ret)
2829                         return ERR_PTR(ret);
2830                 lock_page(page);
2831                 if (!PageUptodate(page)) {
2832                         unlock_page(page);
2833                         put_page(page);
2834                         return ERR_PTR(-EIO);
2835                 }
2836                 if (page->mapping != inode->i_mapping) {
2837                         unlock_page(page);
2838                         put_page(page);
2839                         return ERR_PTR(-EAGAIN);
2840                 }
2841         }
2842
2843         return page;
2844 }
2845
2846 static int gather_extent_pages(struct inode *inode, struct page **pages,
2847                                int num_pages, u64 off)
2848 {
2849         int i;
2850         pgoff_t index = off >> PAGE_SHIFT;
2851
2852         for (i = 0; i < num_pages; i++) {
2853 again:
2854                 pages[i] = extent_same_get_page(inode, index + i);
2855                 if (IS_ERR(pages[i])) {
2856                         int err = PTR_ERR(pages[i]);
2857
2858                         if (err == -EAGAIN)
2859                                 goto again;
2860                         pages[i] = NULL;
2861                         return err;
2862                 }
2863         }
2864         return 0;
2865 }
2866
2867 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2868                              bool retry_range_locking)
2869 {
2870         /*
2871          * Do any pending delalloc/csum calculations on inode, one way or
2872          * another, and lock file content.
2873          * The locking order is:
2874          *
2875          *   1) pages
2876          *   2) range in the inode's io tree
2877          */
2878         while (1) {
2879                 struct btrfs_ordered_extent *ordered;
2880                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2881                 ordered = btrfs_lookup_first_ordered_extent(inode,
2882                                                             off + len - 1);
2883                 if ((!ordered ||
2884                      ordered->file_offset + ordered->len <= off ||
2885                      ordered->file_offset >= off + len) &&
2886                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2887                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2888                         if (ordered)
2889                                 btrfs_put_ordered_extent(ordered);
2890                         break;
2891                 }
2892                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2893                 if (ordered)
2894                         btrfs_put_ordered_extent(ordered);
2895                 if (!retry_range_locking)
2896                         return -EAGAIN;
2897                 btrfs_wait_ordered_range(inode, off, len);
2898         }
2899         return 0;
2900 }
2901
2902 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2903 {
2904         inode_unlock(inode1);
2905         inode_unlock(inode2);
2906 }
2907
2908 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2909 {
2910         if (inode1 < inode2)
2911                 swap(inode1, inode2);
2912
2913         inode_lock_nested(inode1, I_MUTEX_PARENT);
2914         inode_lock_nested(inode2, I_MUTEX_CHILD);
2915 }
2916
2917 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2918                                       struct inode *inode2, u64 loff2, u64 len)
2919 {
2920         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2921         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2922 }
2923
2924 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2925                                     struct inode *inode2, u64 loff2, u64 len,
2926                                     bool retry_range_locking)
2927 {
2928         int ret;
2929
2930         if (inode1 < inode2) {
2931                 swap(inode1, inode2);
2932                 swap(loff1, loff2);
2933         }
2934         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2935         if (ret)
2936                 return ret;
2937         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2938         if (ret)
2939                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2940                               loff1 + len - 1);
2941         return ret;
2942 }
2943
2944 struct cmp_pages {
2945         int             num_pages;
2946         struct page     **src_pages;
2947         struct page     **dst_pages;
2948 };
2949
2950 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2951 {
2952         int i;
2953         struct page *pg;
2954
2955         for (i = 0; i < cmp->num_pages; i++) {
2956                 pg = cmp->src_pages[i];
2957                 if (pg) {
2958                         unlock_page(pg);
2959                         put_page(pg);
2960                 }
2961                 pg = cmp->dst_pages[i];
2962                 if (pg) {
2963                         unlock_page(pg);
2964                         put_page(pg);
2965                 }
2966         }
2967         kfree(cmp->src_pages);
2968         kfree(cmp->dst_pages);
2969 }
2970
2971 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2972                                   struct inode *dst, u64 dst_loff,
2973                                   u64 len, struct cmp_pages *cmp)
2974 {
2975         int ret;
2976         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2977         struct page **src_pgarr, **dst_pgarr;
2978
2979         /*
2980          * We must gather up all the pages before we initiate our
2981          * extent locking. We use an array for the page pointers. Size
2982          * of the array is bounded by len, which is in turn bounded by
2983          * BTRFS_MAX_DEDUPE_LEN.
2984          */
2985         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2986         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2987         if (!src_pgarr || !dst_pgarr) {
2988                 kfree(src_pgarr);
2989                 kfree(dst_pgarr);
2990                 return -ENOMEM;
2991         }
2992         cmp->num_pages = num_pages;
2993         cmp->src_pages = src_pgarr;
2994         cmp->dst_pages = dst_pgarr;
2995
2996         /*
2997          * If deduping ranges in the same inode, locking rules make it mandatory
2998          * to always lock pages in ascending order to avoid deadlocks with
2999          * concurrent tasks (such as starting writeback/delalloc).
3000          */
3001         if (src == dst && dst_loff < loff) {
3002                 swap(src_pgarr, dst_pgarr);
3003                 swap(loff, dst_loff);
3004         }
3005
3006         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3007         if (ret)
3008                 goto out;
3009
3010         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3011
3012 out:
3013         if (ret)
3014                 btrfs_cmp_data_free(cmp);
3015         return ret;
3016 }
3017
3018 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3019 {
3020         int ret = 0;
3021         int i;
3022         struct page *src_page, *dst_page;
3023         unsigned int cmp_len = PAGE_SIZE;
3024         void *addr, *dst_addr;
3025
3026         i = 0;
3027         while (len) {
3028                 if (len < PAGE_SIZE)
3029                         cmp_len = len;
3030
3031                 BUG_ON(i >= cmp->num_pages);
3032
3033                 src_page = cmp->src_pages[i];
3034                 dst_page = cmp->dst_pages[i];
3035                 ASSERT(PageLocked(src_page));
3036                 ASSERT(PageLocked(dst_page));
3037
3038                 addr = kmap_atomic(src_page);
3039                 dst_addr = kmap_atomic(dst_page);
3040
3041                 flush_dcache_page(src_page);
3042                 flush_dcache_page(dst_page);
3043
3044                 if (memcmp(addr, dst_addr, cmp_len))
3045                         ret = -EBADE;
3046
3047                 kunmap_atomic(addr);
3048                 kunmap_atomic(dst_addr);
3049
3050                 if (ret)
3051                         break;
3052
3053                 len -= cmp_len;
3054                 i++;
3055         }
3056
3057         return ret;
3058 }
3059
3060 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3061                                      u64 olen)
3062 {
3063         u64 len = *plen;
3064         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3065
3066         if (off + olen > inode->i_size || off + olen < off)
3067                 return -EINVAL;
3068
3069         /* if we extend to eof, continue to block boundary */
3070         if (off + len == inode->i_size)
3071                 *plen = len = ALIGN(inode->i_size, bs) - off;
3072
3073         /* Check that we are block aligned - btrfs_clone() requires this */
3074         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3075                 return -EINVAL;
3076
3077         return 0;
3078 }
3079
3080 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3081                              struct inode *dst, u64 dst_loff)
3082 {
3083         int ret;
3084         u64 len = olen;
3085         struct cmp_pages cmp;
3086         bool same_inode = (src == dst);
3087         u64 same_lock_start = 0;
3088         u64 same_lock_len = 0;
3089
3090         if (len == 0)
3091                 return 0;
3092
3093         if (same_inode)
3094                 inode_lock(src);
3095         else
3096                 btrfs_double_inode_lock(src, dst);
3097
3098         ret = extent_same_check_offsets(src, loff, &len, olen);
3099         if (ret)
3100                 goto out_unlock;
3101
3102         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3103         if (ret)
3104                 goto out_unlock;
3105
3106         if (same_inode) {
3107                 /*
3108                  * Single inode case wants the same checks, except we
3109                  * don't want our length pushed out past i_size as
3110                  * comparing that data range makes no sense.
3111                  *
3112                  * extent_same_check_offsets() will do this for an
3113                  * unaligned length at i_size, so catch it here and
3114                  * reject the request.
3115                  *
3116                  * This effectively means we require aligned extents
3117                  * for the single-inode case, whereas the other cases
3118                  * allow an unaligned length so long as it ends at
3119                  * i_size.
3120                  */
3121                 if (len != olen) {
3122                         ret = -EINVAL;
3123                         goto out_unlock;
3124                 }
3125
3126                 /* Check for overlapping ranges */
3127                 if (dst_loff + len > loff && dst_loff < loff + len) {
3128                         ret = -EINVAL;
3129                         goto out_unlock;
3130                 }
3131
3132                 same_lock_start = min_t(u64, loff, dst_loff);
3133                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3134         }
3135
3136         /* don't make the dst file partly checksummed */
3137         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3138             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3139                 ret = -EINVAL;
3140                 goto out_unlock;
3141         }
3142
3143 again:
3144         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3145         if (ret)
3146                 goto out_unlock;
3147
3148         if (same_inode)
3149                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3150                                         false);
3151         else
3152                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3153                                                false);
3154         /*
3155          * If one of the inodes has dirty pages in the respective range or
3156          * ordered extents, we need to flush dellaloc and wait for all ordered
3157          * extents in the range. We must unlock the pages and the ranges in the
3158          * io trees to avoid deadlocks when flushing delalloc (requires locking
3159          * pages) and when waiting for ordered extents to complete (they require
3160          * range locking).
3161          */
3162         if (ret == -EAGAIN) {
3163                 /*
3164                  * Ranges in the io trees already unlocked. Now unlock all
3165                  * pages before waiting for all IO to complete.
3166                  */
3167                 btrfs_cmp_data_free(&cmp);
3168                 if (same_inode) {
3169                         btrfs_wait_ordered_range(src, same_lock_start,
3170                                                  same_lock_len);
3171                 } else {
3172                         btrfs_wait_ordered_range(src, loff, len);
3173                         btrfs_wait_ordered_range(dst, dst_loff, len);
3174                 }
3175                 goto again;
3176         }
3177         ASSERT(ret == 0);
3178         if (WARN_ON(ret)) {
3179                 /* ranges in the io trees already unlocked */
3180                 btrfs_cmp_data_free(&cmp);
3181                 return ret;
3182         }
3183
3184         /* pass original length for comparison so we stay within i_size */
3185         ret = btrfs_cmp_data(olen, &cmp);
3186         if (ret == 0)
3187                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3188
3189         if (same_inode)
3190                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3191                               same_lock_start + same_lock_len - 1);
3192         else
3193                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3194
3195         btrfs_cmp_data_free(&cmp);
3196 out_unlock:
3197         if (same_inode)
3198                 inode_unlock(src);
3199         else
3200                 btrfs_double_inode_unlock(src, dst);
3201
3202         return ret;
3203 }
3204
3205 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3206
3207 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3208                                 struct file *dst_file, u64 dst_loff)
3209 {
3210         struct inode *src = file_inode(src_file);
3211         struct inode *dst = file_inode(dst_file);
3212         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3213         ssize_t res;
3214
3215         if (olen > BTRFS_MAX_DEDUPE_LEN)
3216                 olen = BTRFS_MAX_DEDUPE_LEN;
3217
3218         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3219                 /*
3220                  * Btrfs does not support blocksize < page_size. As a
3221                  * result, btrfs_cmp_data() won't correctly handle
3222                  * this situation without an update.
3223                  */
3224                 return -EINVAL;
3225         }
3226
3227         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3228         if (res)
3229                 return res;
3230         return olen;
3231 }
3232
3233 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3234                                      struct inode *inode,
3235                                      u64 endoff,
3236                                      const u64 destoff,
3237                                      const u64 olen,
3238                                      int no_time_update)
3239 {
3240         struct btrfs_root *root = BTRFS_I(inode)->root;
3241         int ret;
3242
3243         inode_inc_iversion(inode);
3244         if (!no_time_update)
3245                 inode->i_mtime = inode->i_ctime = current_time(inode);
3246         /*
3247          * We round up to the block size at eof when determining which
3248          * extents to clone above, but shouldn't round up the file size.
3249          */
3250         if (endoff > destoff + olen)
3251                 endoff = destoff + olen;
3252         if (endoff > inode->i_size)
3253                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3254
3255         ret = btrfs_update_inode(trans, root, inode);
3256         if (ret) {
3257                 btrfs_abort_transaction(trans, ret);
3258                 btrfs_end_transaction(trans);
3259                 goto out;
3260         }
3261         ret = btrfs_end_transaction(trans);
3262 out:
3263         return ret;
3264 }
3265
3266 static void clone_update_extent_map(struct btrfs_inode *inode,
3267                                     const struct btrfs_trans_handle *trans,
3268                                     const struct btrfs_path *path,
3269                                     const u64 hole_offset,
3270                                     const u64 hole_len)
3271 {
3272         struct extent_map_tree *em_tree = &inode->extent_tree;
3273         struct extent_map *em;
3274         int ret;
3275
3276         em = alloc_extent_map();
3277         if (!em) {
3278                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3279                 return;
3280         }
3281
3282         if (path) {
3283                 struct btrfs_file_extent_item *fi;
3284
3285                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3286                                     struct btrfs_file_extent_item);
3287                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3288                 em->generation = -1;
3289                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3290                     BTRFS_FILE_EXTENT_INLINE)
3291                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3292                                         &inode->runtime_flags);
3293         } else {
3294                 em->start = hole_offset;
3295                 em->len = hole_len;
3296                 em->ram_bytes = em->len;
3297                 em->orig_start = hole_offset;
3298                 em->block_start = EXTENT_MAP_HOLE;
3299                 em->block_len = 0;
3300                 em->orig_block_len = 0;
3301                 em->compress_type = BTRFS_COMPRESS_NONE;
3302                 em->generation = trans->transid;
3303         }
3304
3305         while (1) {
3306                 write_lock(&em_tree->lock);
3307                 ret = add_extent_mapping(em_tree, em, 1);
3308                 write_unlock(&em_tree->lock);
3309                 if (ret != -EEXIST) {
3310                         free_extent_map(em);
3311                         break;
3312                 }
3313                 btrfs_drop_extent_cache(inode, em->start,
3314                                         em->start + em->len - 1, 0);
3315         }
3316
3317         if (ret)
3318                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3319 }
3320
3321 /*
3322  * Make sure we do not end up inserting an inline extent into a file that has
3323  * already other (non-inline) extents. If a file has an inline extent it can
3324  * not have any other extents and the (single) inline extent must start at the
3325  * file offset 0. Failing to respect these rules will lead to file corruption,
3326  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3327  *
3328  * We can have extents that have been already written to disk or we can have
3329  * dirty ranges still in delalloc, in which case the extent maps and items are
3330  * created only when we run delalloc, and the delalloc ranges might fall outside
3331  * the range we are currently locking in the inode's io tree. So we check the
3332  * inode's i_size because of that (i_size updates are done while holding the
3333  * i_mutex, which we are holding here).
3334  * We also check to see if the inode has a size not greater than "datal" but has
3335  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3336  * protected against such concurrent fallocate calls by the i_mutex).
3337  *
3338  * If the file has no extents but a size greater than datal, do not allow the
3339  * copy because we would need turn the inline extent into a non-inline one (even
3340  * with NO_HOLES enabled). If we find our destination inode only has one inline
3341  * extent, just overwrite it with the source inline extent if its size is less
3342  * than the source extent's size, or we could copy the source inline extent's
3343  * data into the destination inode's inline extent if the later is greater then
3344  * the former.
3345  */
3346 static int clone_copy_inline_extent(struct inode *dst,
3347                                     struct btrfs_trans_handle *trans,
3348                                     struct btrfs_path *path,
3349                                     struct btrfs_key *new_key,
3350                                     const u64 drop_start,
3351                                     const u64 datal,
3352                                     const u64 skip,
3353                                     const u64 size,
3354                                     char *inline_data)
3355 {
3356         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3357         struct btrfs_root *root = BTRFS_I(dst)->root;
3358         const u64 aligned_end = ALIGN(new_key->offset + datal,
3359                                       fs_info->sectorsize);
3360         int ret;
3361         struct btrfs_key key;
3362
3363         if (new_key->offset > 0)
3364                 return -EOPNOTSUPP;
3365
3366         key.objectid = btrfs_ino(BTRFS_I(dst));
3367         key.type = BTRFS_EXTENT_DATA_KEY;
3368         key.offset = 0;
3369         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3370         if (ret < 0) {
3371                 return ret;
3372         } else if (ret > 0) {
3373                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3374                         ret = btrfs_next_leaf(root, path);
3375                         if (ret < 0)
3376                                 return ret;
3377                         else if (ret > 0)
3378                                 goto copy_inline_extent;
3379                 }
3380                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3381                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3382                     key.type == BTRFS_EXTENT_DATA_KEY) {
3383                         ASSERT(key.offset > 0);
3384                         return -EOPNOTSUPP;
3385                 }
3386         } else if (i_size_read(dst) <= datal) {
3387                 struct btrfs_file_extent_item *ei;
3388                 u64 ext_len;
3389
3390                 /*
3391                  * If the file size is <= datal, make sure there are no other
3392                  * extents following (can happen do to an fallocate call with
3393                  * the flag FALLOC_FL_KEEP_SIZE).
3394                  */
3395                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3396                                     struct btrfs_file_extent_item);
3397                 /*
3398                  * If it's an inline extent, it can not have other extents
3399                  * following it.
3400                  */
3401                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3402                     BTRFS_FILE_EXTENT_INLINE)
3403                         goto copy_inline_extent;
3404
3405                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3406                 if (ext_len > aligned_end)
3407                         return -EOPNOTSUPP;
3408
3409                 ret = btrfs_next_item(root, path);
3410                 if (ret < 0) {
3411                         return ret;
3412                 } else if (ret == 0) {
3413                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3414                                               path->slots[0]);
3415                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3416                             key.type == BTRFS_EXTENT_DATA_KEY)
3417                                 return -EOPNOTSUPP;
3418                 }
3419         }
3420
3421 copy_inline_extent:
3422         /*
3423          * We have no extent items, or we have an extent at offset 0 which may
3424          * or may not be inlined. All these cases are dealt the same way.
3425          */
3426         if (i_size_read(dst) > datal) {
3427                 /*
3428                  * If the destination inode has an inline extent...
3429                  * This would require copying the data from the source inline
3430                  * extent into the beginning of the destination's inline extent.
3431                  * But this is really complex, both extents can be compressed
3432                  * or just one of them, which would require decompressing and
3433                  * re-compressing data (which could increase the new compressed
3434                  * size, not allowing the compressed data to fit anymore in an
3435                  * inline extent).
3436                  * So just don't support this case for now (it should be rare,
3437                  * we are not really saving space when cloning inline extents).
3438                  */
3439                 return -EOPNOTSUPP;
3440         }
3441
3442         btrfs_release_path(path);
3443         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3444         if (ret)
3445                 return ret;
3446         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3447         if (ret)
3448                 return ret;
3449
3450         if (skip) {
3451                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3452
3453                 memmove(inline_data + start, inline_data + start + skip, datal);
3454         }
3455
3456         write_extent_buffer(path->nodes[0], inline_data,
3457                             btrfs_item_ptr_offset(path->nodes[0],
3458                                                   path->slots[0]),
3459                             size);
3460         inode_add_bytes(dst, datal);
3461
3462         return 0;
3463 }
3464
3465 /**
3466  * btrfs_clone() - clone a range from inode file to another
3467  *
3468  * @src: Inode to clone from
3469  * @inode: Inode to clone to
3470  * @off: Offset within source to start clone from
3471  * @olen: Original length, passed by user, of range to clone
3472  * @olen_aligned: Block-aligned value of olen
3473  * @destoff: Offset within @inode to start clone
3474  * @no_time_update: Whether to update mtime/ctime on the target inode
3475  */
3476 static int btrfs_clone(struct inode *src, struct inode *inode,
3477                        const u64 off, const u64 olen, const u64 olen_aligned,
3478                        const u64 destoff, int no_time_update)
3479 {
3480         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3481         struct btrfs_root *root = BTRFS_I(inode)->root;
3482         struct btrfs_path *path = NULL;
3483         struct extent_buffer *leaf;
3484         struct btrfs_trans_handle *trans;
3485         char *buf = NULL;
3486         struct btrfs_key key;
3487         u32 nritems;
3488         int slot;
3489         int ret;
3490         const u64 len = olen_aligned;
3491         u64 last_dest_end = destoff;
3492
3493         ret = -ENOMEM;
3494         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3495         if (!buf)
3496                 return ret;
3497
3498         path = btrfs_alloc_path();
3499         if (!path) {
3500                 kvfree(buf);
3501                 return ret;
3502         }
3503
3504         path->reada = READA_FORWARD;
3505         /* clone data */
3506         key.objectid = btrfs_ino(BTRFS_I(src));
3507         key.type = BTRFS_EXTENT_DATA_KEY;
3508         key.offset = off;
3509
3510         while (1) {
3511                 u64 next_key_min_offset = key.offset + 1;
3512
3513                 /*
3514                  * note the key will change type as we walk through the
3515                  * tree.
3516                  */
3517                 path->leave_spinning = 1;
3518                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3519                                 0, 0);
3520                 if (ret < 0)
3521                         goto out;
3522                 /*
3523                  * First search, if no extent item that starts at offset off was
3524                  * found but the previous item is an extent item, it's possible
3525                  * it might overlap our target range, therefore process it.
3526                  */
3527                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3528                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3529                                               path->slots[0] - 1);
3530                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3531                                 path->slots[0]--;
3532                 }
3533
3534                 nritems = btrfs_header_nritems(path->nodes[0]);
3535 process_slot:
3536                 if (path->slots[0] >= nritems) {
3537                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3538                         if (ret < 0)
3539                                 goto out;
3540                         if (ret > 0)
3541                                 break;
3542                         nritems = btrfs_header_nritems(path->nodes[0]);
3543                 }
3544                 leaf = path->nodes[0];
3545                 slot = path->slots[0];
3546
3547                 btrfs_item_key_to_cpu(leaf, &key, slot);
3548                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3549                     key.objectid != btrfs_ino(BTRFS_I(src)))
3550                         break;
3551
3552                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3553                         struct btrfs_file_extent_item *extent;
3554                         int type;
3555                         u32 size;
3556                         struct btrfs_key new_key;
3557                         u64 disko = 0, diskl = 0;
3558                         u64 datao = 0, datal = 0;
3559                         u8 comp;
3560                         u64 drop_start;
3561
3562                         extent = btrfs_item_ptr(leaf, slot,
3563                                                 struct btrfs_file_extent_item);
3564                         comp = btrfs_file_extent_compression(leaf, extent);
3565                         type = btrfs_file_extent_type(leaf, extent);
3566                         if (type == BTRFS_FILE_EXTENT_REG ||
3567                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3568                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3569                                                                       extent);
3570                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3571                                                                  extent);
3572                                 datao = btrfs_file_extent_offset(leaf, extent);
3573                                 datal = btrfs_file_extent_num_bytes(leaf,
3574                                                                     extent);
3575                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3576                                 /* take upper bound, may be compressed */
3577                                 datal = btrfs_file_extent_ram_bytes(leaf,
3578                                                                     extent);
3579                         }
3580
3581                         /*
3582                          * The first search might have left us at an extent
3583                          * item that ends before our target range's start, can
3584                          * happen if we have holes and NO_HOLES feature enabled.
3585                          */
3586                         if (key.offset + datal <= off) {
3587                                 path->slots[0]++;
3588                                 goto process_slot;
3589                         } else if (key.offset >= off + len) {
3590                                 break;
3591                         }
3592                         next_key_min_offset = key.offset + datal;
3593                         size = btrfs_item_size_nr(leaf, slot);
3594                         read_extent_buffer(leaf, buf,
3595                                            btrfs_item_ptr_offset(leaf, slot),
3596                                            size);
3597
3598                         btrfs_release_path(path);
3599                         path->leave_spinning = 0;
3600
3601                         memcpy(&new_key, &key, sizeof(new_key));
3602                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3603                         if (off <= key.offset)
3604                                 new_key.offset = key.offset + destoff - off;
3605                         else
3606                                 new_key.offset = destoff;
3607
3608                         /*
3609                          * Deal with a hole that doesn't have an extent item
3610                          * that represents it (NO_HOLES feature enabled).
3611                          * This hole is either in the middle of the cloning
3612                          * range or at the beginning (fully overlaps it or
3613                          * partially overlaps it).
3614                          */
3615                         if (new_key.offset != last_dest_end)
3616                                 drop_start = last_dest_end;
3617                         else
3618                                 drop_start = new_key.offset;
3619
3620                         /*
3621                          * 1 - adjusting old extent (we may have to split it)
3622                          * 1 - add new extent
3623                          * 1 - inode update
3624                          */
3625                         trans = btrfs_start_transaction(root, 3);
3626                         if (IS_ERR(trans)) {
3627                                 ret = PTR_ERR(trans);
3628                                 goto out;
3629                         }
3630
3631                         if (type == BTRFS_FILE_EXTENT_REG ||
3632                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3633                                 /*
3634                                  *    a  | --- range to clone ---|  b
3635                                  * | ------------- extent ------------- |
3636                                  */
3637
3638                                 /* subtract range b */
3639                                 if (key.offset + datal > off + len)
3640                                         datal = off + len - key.offset;
3641
3642                                 /* subtract range a */
3643                                 if (off > key.offset) {
3644                                         datao += off - key.offset;
3645                                         datal -= off - key.offset;
3646                                 }
3647
3648                                 ret = btrfs_drop_extents(trans, root, inode,
3649                                                          drop_start,
3650                                                          new_key.offset + datal,
3651                                                          1);
3652                                 if (ret) {
3653                                         if (ret != -EOPNOTSUPP)
3654                                                 btrfs_abort_transaction(trans,
3655                                                                         ret);
3656                                         btrfs_end_transaction(trans);
3657                                         goto out;
3658                                 }
3659
3660                                 ret = btrfs_insert_empty_item(trans, root, path,
3661                                                               &new_key, size);
3662                                 if (ret) {
3663                                         btrfs_abort_transaction(trans, ret);
3664                                         btrfs_end_transaction(trans);
3665                                         goto out;
3666                                 }
3667
3668                                 leaf = path->nodes[0];
3669                                 slot = path->slots[0];
3670                                 write_extent_buffer(leaf, buf,
3671                                             btrfs_item_ptr_offset(leaf, slot),
3672                                             size);
3673
3674                                 extent = btrfs_item_ptr(leaf, slot,
3675                                                 struct btrfs_file_extent_item);
3676
3677                                 /* disko == 0 means it's a hole */
3678                                 if (!disko)
3679                                         datao = 0;
3680
3681                                 btrfs_set_file_extent_offset(leaf, extent,
3682                                                              datao);
3683                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3684                                                                 datal);
3685
3686                                 if (disko) {
3687                                         inode_add_bytes(inode, datal);
3688                                         ret = btrfs_inc_extent_ref(trans,
3689                                                         root,
3690                                                         disko, diskl, 0,
3691                                                         root->root_key.objectid,
3692                                                         btrfs_ino(BTRFS_I(inode)),
3693                                                         new_key.offset - datao);
3694                                         if (ret) {
3695                                                 btrfs_abort_transaction(trans,
3696                                                                         ret);
3697                                                 btrfs_end_transaction(trans);
3698                                                 goto out;
3699
3700                                         }
3701                                 }
3702                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3703                                 u64 skip = 0;
3704                                 u64 trim = 0;
3705
3706                                 if (off > key.offset) {
3707                                         skip = off - key.offset;
3708                                         new_key.offset += skip;
3709                                 }
3710
3711                                 if (key.offset + datal > off + len)
3712                                         trim = key.offset + datal - (off + len);
3713
3714                                 if (comp && (skip || trim)) {
3715                                         ret = -EINVAL;
3716                                         btrfs_end_transaction(trans);
3717                                         goto out;
3718                                 }
3719                                 size -= skip + trim;
3720                                 datal -= skip + trim;
3721
3722                                 ret = clone_copy_inline_extent(inode,
3723                                                                trans, path,
3724                                                                &new_key,
3725                                                                drop_start,
3726                                                                datal,
3727                                                                skip, size, buf);
3728                                 if (ret) {
3729                                         if (ret != -EOPNOTSUPP)
3730                                                 btrfs_abort_transaction(trans,
3731                                                                         ret);
3732                                         btrfs_end_transaction(trans);
3733                                         goto out;
3734                                 }
3735                                 leaf = path->nodes[0];
3736                                 slot = path->slots[0];
3737                         }
3738
3739                         /* If we have an implicit hole (NO_HOLES feature). */
3740                         if (drop_start < new_key.offset)
3741                                 clone_update_extent_map(BTRFS_I(inode), trans,
3742                                                 NULL, drop_start,
3743                                                 new_key.offset - drop_start);
3744
3745                         clone_update_extent_map(BTRFS_I(inode), trans,
3746                                         path, 0, 0);
3747
3748                         btrfs_mark_buffer_dirty(leaf);
3749                         btrfs_release_path(path);
3750
3751                         last_dest_end = ALIGN(new_key.offset + datal,
3752                                               fs_info->sectorsize);
3753                         ret = clone_finish_inode_update(trans, inode,
3754                                                         last_dest_end,
3755                                                         destoff, olen,
3756                                                         no_time_update);
3757                         if (ret)
3758                                 goto out;
3759                         if (new_key.offset + datal >= destoff + len)
3760                                 break;
3761                 }
3762                 btrfs_release_path(path);
3763                 key.offset = next_key_min_offset;
3764
3765                 if (fatal_signal_pending(current)) {
3766                         ret = -EINTR;
3767                         goto out;
3768                 }
3769         }
3770         ret = 0;
3771
3772         if (last_dest_end < destoff + len) {
3773                 /*
3774                  * We have an implicit hole (NO_HOLES feature is enabled) that
3775                  * fully or partially overlaps our cloning range at its end.
3776                  */
3777                 btrfs_release_path(path);
3778
3779                 /*
3780                  * 1 - remove extent(s)
3781                  * 1 - inode update
3782                  */
3783                 trans = btrfs_start_transaction(root, 2);
3784                 if (IS_ERR(trans)) {
3785                         ret = PTR_ERR(trans);
3786                         goto out;
3787                 }
3788                 ret = btrfs_drop_extents(trans, root, inode,
3789                                          last_dest_end, destoff + len, 1);
3790                 if (ret) {
3791                         if (ret != -EOPNOTSUPP)
3792                                 btrfs_abort_transaction(trans, ret);
3793                         btrfs_end_transaction(trans);
3794                         goto out;
3795                 }
3796                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3797                                 last_dest_end,
3798                                 destoff + len - last_dest_end);
3799                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3800                                                 destoff, olen, no_time_update);
3801         }
3802
3803 out:
3804         btrfs_free_path(path);
3805         kvfree(buf);
3806         return ret;
3807 }
3808
3809 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3810                                         u64 off, u64 olen, u64 destoff)
3811 {
3812         struct inode *inode = file_inode(file);
3813         struct inode *src = file_inode(file_src);
3814         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3815         struct btrfs_root *root = BTRFS_I(inode)->root;
3816         int ret;
3817         u64 len = olen;
3818         u64 bs = fs_info->sb->s_blocksize;
3819         int same_inode = src == inode;
3820
3821         /*
3822          * TODO:
3823          * - split compressed inline extents.  annoying: we need to
3824          *   decompress into destination's address_space (the file offset
3825          *   may change, so source mapping won't do), then recompress (or
3826          *   otherwise reinsert) a subrange.
3827          *
3828          * - split destination inode's inline extents.  The inline extents can
3829          *   be either compressed or non-compressed.
3830          */
3831
3832         if (btrfs_root_readonly(root))
3833                 return -EROFS;
3834
3835         if (file_src->f_path.mnt != file->f_path.mnt ||
3836             src->i_sb != inode->i_sb)
3837                 return -EXDEV;
3838
3839         /* don't make the dst file partly checksummed */
3840         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3841             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3842                 return -EINVAL;
3843
3844         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3845                 return -EISDIR;
3846
3847         if (!same_inode) {
3848                 btrfs_double_inode_lock(src, inode);
3849         } else {
3850                 inode_lock(src);
3851         }
3852
3853         /* determine range to clone */
3854         ret = -EINVAL;
3855         if (off + len > src->i_size || off + len < off)
3856                 goto out_unlock;
3857         if (len == 0)
3858                 olen = len = src->i_size - off;
3859         /* if we extend to eof, continue to block boundary */
3860         if (off + len == src->i_size)
3861                 len = ALIGN(src->i_size, bs) - off;
3862
3863         if (len == 0) {
3864                 ret = 0;
3865                 goto out_unlock;
3866         }
3867
3868         /* verify the end result is block aligned */
3869         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3870             !IS_ALIGNED(destoff, bs))
3871                 goto out_unlock;
3872
3873         /* verify if ranges are overlapped within the same file */
3874         if (same_inode) {
3875                 if (destoff + len > off && destoff < off + len)
3876                         goto out_unlock;
3877         }
3878
3879         if (destoff > inode->i_size) {
3880                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3881                 if (ret)
3882                         goto out_unlock;
3883         }
3884
3885         /*
3886          * Lock the target range too. Right after we replace the file extent
3887          * items in the fs tree (which now point to the cloned data), we might
3888          * have a worker replace them with extent items relative to a write
3889          * operation that was issued before this clone operation (i.e. confront
3890          * with inode.c:btrfs_finish_ordered_io).
3891          */
3892         if (same_inode) {
3893                 u64 lock_start = min_t(u64, off, destoff);
3894                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3895
3896                 ret = lock_extent_range(src, lock_start, lock_len, true);
3897         } else {
3898                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3899                                                true);
3900         }
3901         ASSERT(ret == 0);
3902         if (WARN_ON(ret)) {
3903                 /* ranges in the io trees already unlocked */
3904                 goto out_unlock;
3905         }
3906
3907         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3908
3909         if (same_inode) {
3910                 u64 lock_start = min_t(u64, off, destoff);
3911                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3912
3913                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3914         } else {
3915                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3916         }
3917         /*
3918          * Truncate page cache pages so that future reads will see the cloned
3919          * data immediately and not the previous data.
3920          */
3921         truncate_inode_pages_range(&inode->i_data,
3922                                 round_down(destoff, PAGE_SIZE),
3923                                 round_up(destoff + len, PAGE_SIZE) - 1);
3924 out_unlock:
3925         if (!same_inode)
3926                 btrfs_double_inode_unlock(src, inode);
3927         else
3928                 inode_unlock(src);
3929         return ret;
3930 }
3931
3932 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3933                 struct file *dst_file, loff_t destoff, u64 len)
3934 {
3935         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3936 }
3937
3938 /*
3939  * there are many ways the trans_start and trans_end ioctls can lead
3940  * to deadlocks.  They should only be used by applications that
3941  * basically own the machine, and have a very in depth understanding
3942  * of all the possible deadlocks and enospc problems.
3943  */
3944 static long btrfs_ioctl_trans_start(struct file *file)
3945 {
3946         struct inode *inode = file_inode(file);
3947         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3948         struct btrfs_root *root = BTRFS_I(inode)->root;
3949         struct btrfs_trans_handle *trans;
3950         struct btrfs_file_private *private;
3951         int ret;
3952         static bool warned = false;
3953
3954         ret = -EPERM;
3955         if (!capable(CAP_SYS_ADMIN))
3956                 goto out;
3957
3958         if (!warned) {
3959                 btrfs_warn(fs_info,
3960                         "Userspace transaction mechanism is considered "
3961                         "deprecated and slated to be removed in 4.17. "
3962                         "If you have a valid use case please "
3963                         "speak up on the mailing list");
3964                 WARN_ON(1);
3965                 warned = true;
3966         }
3967
3968         ret = -EINPROGRESS;
3969         private = file->private_data;
3970         if (private && private->trans)
3971                 goto out;
3972         if (!private) {
3973                 private = kzalloc(sizeof(struct btrfs_file_private),
3974                                   GFP_KERNEL);
3975                 if (!private)
3976                         return -ENOMEM;
3977                 file->private_data = private;
3978         }
3979
3980         ret = -EROFS;
3981         if (btrfs_root_readonly(root))
3982                 goto out;
3983
3984         ret = mnt_want_write_file(file);
3985         if (ret)
3986                 goto out;
3987
3988         atomic_inc(&fs_info->open_ioctl_trans);
3989
3990         ret = -ENOMEM;
3991         trans = btrfs_start_ioctl_transaction(root);
3992         if (IS_ERR(trans))
3993                 goto out_drop;
3994
3995         private->trans = trans;
3996         return 0;
3997
3998 out_drop:
3999         atomic_dec(&fs_info->open_ioctl_trans);
4000         mnt_drop_write_file(file);
4001 out:
4002         return ret;
4003 }
4004
4005 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4006 {
4007         struct inode *inode = file_inode(file);
4008         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4009         struct btrfs_root *root = BTRFS_I(inode)->root;
4010         struct btrfs_root *new_root;
4011         struct btrfs_dir_item *di;
4012         struct btrfs_trans_handle *trans;
4013         struct btrfs_path *path;
4014         struct btrfs_key location;
4015         struct btrfs_disk_key disk_key;
4016         u64 objectid = 0;
4017         u64 dir_id;
4018         int ret;
4019
4020         if (!capable(CAP_SYS_ADMIN))
4021                 return -EPERM;
4022
4023         ret = mnt_want_write_file(file);
4024         if (ret)
4025                 return ret;
4026
4027         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4028                 ret = -EFAULT;
4029                 goto out;
4030         }
4031
4032         if (!objectid)
4033                 objectid = BTRFS_FS_TREE_OBJECTID;
4034
4035         location.objectid = objectid;
4036         location.type = BTRFS_ROOT_ITEM_KEY;
4037         location.offset = (u64)-1;
4038
4039         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4040         if (IS_ERR(new_root)) {
4041                 ret = PTR_ERR(new_root);
4042                 goto out;
4043         }
4044         if (!is_fstree(new_root->objectid)) {
4045                 ret = -ENOENT;
4046                 goto out;
4047         }
4048
4049         path = btrfs_alloc_path();
4050         if (!path) {
4051                 ret = -ENOMEM;
4052                 goto out;
4053         }
4054         path->leave_spinning = 1;
4055
4056         trans = btrfs_start_transaction(root, 1);
4057         if (IS_ERR(trans)) {
4058                 btrfs_free_path(path);
4059                 ret = PTR_ERR(trans);
4060                 goto out;
4061         }
4062
4063         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4064         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4065                                    dir_id, "default", 7, 1);
4066         if (IS_ERR_OR_NULL(di)) {
4067                 btrfs_free_path(path);
4068                 btrfs_end_transaction(trans);
4069                 btrfs_err(fs_info,
4070                           "Umm, you don't have the default diritem, this isn't going to work");
4071                 ret = -ENOENT;
4072                 goto out;
4073         }
4074
4075         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4076         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4077         btrfs_mark_buffer_dirty(path->nodes[0]);
4078         btrfs_free_path(path);
4079
4080         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4081         btrfs_end_transaction(trans);
4082 out:
4083         mnt_drop_write_file(file);
4084         return ret;
4085 }
4086
4087 void btrfs_get_block_group_info(struct list_head *groups_list,
4088                                 struct btrfs_ioctl_space_info *space)
4089 {
4090         struct btrfs_block_group_cache *block_group;
4091
4092         space->total_bytes = 0;
4093         space->used_bytes = 0;
4094         space->flags = 0;
4095         list_for_each_entry(block_group, groups_list, list) {
4096                 space->flags = block_group->flags;
4097                 space->total_bytes += block_group->key.offset;
4098                 space->used_bytes +=
4099                         btrfs_block_group_used(&block_group->item);
4100         }
4101 }
4102
4103 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4104                                    void __user *arg)
4105 {
4106         struct btrfs_ioctl_space_args space_args;
4107         struct btrfs_ioctl_space_info space;
4108         struct btrfs_ioctl_space_info *dest;
4109         struct btrfs_ioctl_space_info *dest_orig;
4110         struct btrfs_ioctl_space_info __user *user_dest;
4111         struct btrfs_space_info *info;
4112         static const u64 types[] = {
4113                 BTRFS_BLOCK_GROUP_DATA,
4114                 BTRFS_BLOCK_GROUP_SYSTEM,
4115                 BTRFS_BLOCK_GROUP_METADATA,
4116                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4117         };
4118         int num_types = 4;
4119         int alloc_size;
4120         int ret = 0;
4121         u64 slot_count = 0;
4122         int i, c;
4123
4124         if (copy_from_user(&space_args,
4125                            (struct btrfs_ioctl_space_args __user *)arg,
4126                            sizeof(space_args)))
4127                 return -EFAULT;
4128
4129         for (i = 0; i < num_types; i++) {
4130                 struct btrfs_space_info *tmp;
4131
4132                 info = NULL;
4133                 rcu_read_lock();
4134                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4135                                         list) {
4136                         if (tmp->flags == types[i]) {
4137                                 info = tmp;
4138                                 break;
4139                         }
4140                 }
4141                 rcu_read_unlock();
4142
4143                 if (!info)
4144                         continue;
4145
4146                 down_read(&info->groups_sem);
4147                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4148                         if (!list_empty(&info->block_groups[c]))
4149                                 slot_count++;
4150                 }
4151                 up_read(&info->groups_sem);
4152         }
4153
4154         /*
4155          * Global block reserve, exported as a space_info
4156          */
4157         slot_count++;
4158
4159         /* space_slots == 0 means they are asking for a count */
4160         if (space_args.space_slots == 0) {
4161                 space_args.total_spaces = slot_count;
4162                 goto out;
4163         }
4164
4165         slot_count = min_t(u64, space_args.space_slots, slot_count);
4166
4167         alloc_size = sizeof(*dest) * slot_count;
4168
4169         /* we generally have at most 6 or so space infos, one for each raid
4170          * level.  So, a whole page should be more than enough for everyone
4171          */
4172         if (alloc_size > PAGE_SIZE)
4173                 return -ENOMEM;
4174
4175         space_args.total_spaces = 0;
4176         dest = kmalloc(alloc_size, GFP_KERNEL);
4177         if (!dest)
4178                 return -ENOMEM;
4179         dest_orig = dest;
4180
4181         /* now we have a buffer to copy into */
4182         for (i = 0; i < num_types; i++) {
4183                 struct btrfs_space_info *tmp;
4184
4185                 if (!slot_count)
4186                         break;
4187
4188                 info = NULL;
4189                 rcu_read_lock();
4190                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4191                                         list) {
4192                         if (tmp->flags == types[i]) {
4193                                 info = tmp;
4194                                 break;
4195                         }
4196                 }
4197                 rcu_read_unlock();
4198
4199                 if (!info)
4200                         continue;
4201                 down_read(&info->groups_sem);
4202                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4203                         if (!list_empty(&info->block_groups[c])) {
4204                                 btrfs_get_block_group_info(
4205                                         &info->block_groups[c], &space);
4206                                 memcpy(dest, &space, sizeof(space));
4207                                 dest++;
4208                                 space_args.total_spaces++;
4209                                 slot_count--;
4210                         }
4211                         if (!slot_count)
4212                                 break;
4213                 }
4214                 up_read(&info->groups_sem);
4215         }
4216
4217         /*
4218          * Add global block reserve
4219          */
4220         if (slot_count) {
4221                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4222
4223                 spin_lock(&block_rsv->lock);
4224                 space.total_bytes = block_rsv->size;
4225                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4226                 spin_unlock(&block_rsv->lock);
4227                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4228                 memcpy(dest, &space, sizeof(space));
4229                 space_args.total_spaces++;
4230         }
4231
4232         user_dest = (struct btrfs_ioctl_space_info __user *)
4233                 (arg + sizeof(struct btrfs_ioctl_space_args));
4234
4235         if (copy_to_user(user_dest, dest_orig, alloc_size))
4236                 ret = -EFAULT;
4237
4238         kfree(dest_orig);
4239 out:
4240         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4241                 ret = -EFAULT;
4242
4243         return ret;
4244 }
4245
4246 /*
4247  * there are many ways the trans_start and trans_end ioctls can lead
4248  * to deadlocks.  They should only be used by applications that
4249  * basically own the machine, and have a very in depth understanding
4250  * of all the possible deadlocks and enospc problems.
4251  */
4252 long btrfs_ioctl_trans_end(struct file *file)
4253 {
4254         struct inode *inode = file_inode(file);
4255         struct btrfs_root *root = BTRFS_I(inode)->root;
4256         struct btrfs_file_private *private = file->private_data;
4257
4258         if (!private || !private->trans)
4259                 return -EINVAL;
4260
4261         btrfs_end_transaction(private->trans);
4262         private->trans = NULL;
4263
4264         atomic_dec(&root->fs_info->open_ioctl_trans);
4265
4266         mnt_drop_write_file(file);
4267         return 0;
4268 }
4269
4270 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4271                                             void __user *argp)
4272 {
4273         struct btrfs_trans_handle *trans;
4274         u64 transid;
4275         int ret;
4276
4277         trans = btrfs_attach_transaction_barrier(root);
4278         if (IS_ERR(trans)) {
4279                 if (PTR_ERR(trans) != -ENOENT)
4280                         return PTR_ERR(trans);
4281
4282                 /* No running transaction, don't bother */
4283                 transid = root->fs_info->last_trans_committed;
4284                 goto out;
4285         }
4286         transid = trans->transid;
4287         ret = btrfs_commit_transaction_async(trans, 0);
4288         if (ret) {
4289                 btrfs_end_transaction(trans);
4290                 return ret;
4291         }
4292 out:
4293         if (argp)
4294                 if (copy_to_user(argp, &transid, sizeof(transid)))
4295                         return -EFAULT;
4296         return 0;
4297 }
4298
4299 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4300                                            void __user *argp)
4301 {
4302         u64 transid;
4303
4304         if (argp) {
4305                 if (copy_from_user(&transid, argp, sizeof(transid)))
4306                         return -EFAULT;
4307         } else {
4308                 transid = 0;  /* current trans */
4309         }
4310         return btrfs_wait_for_commit(fs_info, transid);
4311 }
4312
4313 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4314 {
4315         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4316         struct btrfs_ioctl_scrub_args *sa;
4317         int ret;
4318
4319         if (!capable(CAP_SYS_ADMIN))
4320                 return -EPERM;
4321
4322         sa = memdup_user(arg, sizeof(*sa));
4323         if (IS_ERR(sa))
4324                 return PTR_ERR(sa);
4325
4326         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4327                 ret = mnt_want_write_file(file);
4328                 if (ret)
4329                         goto out;
4330         }
4331
4332         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4333                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4334                               0);
4335
4336         if (copy_to_user(arg, sa, sizeof(*sa)))
4337                 ret = -EFAULT;
4338
4339         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4340                 mnt_drop_write_file(file);
4341 out:
4342         kfree(sa);
4343         return ret;
4344 }
4345
4346 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4347 {
4348         if (!capable(CAP_SYS_ADMIN))
4349                 return -EPERM;
4350
4351         return btrfs_scrub_cancel(fs_info);
4352 }
4353
4354 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4355                                        void __user *arg)
4356 {
4357         struct btrfs_ioctl_scrub_args *sa;
4358         int ret;
4359
4360         if (!capable(CAP_SYS_ADMIN))
4361                 return -EPERM;
4362
4363         sa = memdup_user(arg, sizeof(*sa));
4364         if (IS_ERR(sa))
4365                 return PTR_ERR(sa);
4366
4367         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4368
4369         if (copy_to_user(arg, sa, sizeof(*sa)))
4370                 ret = -EFAULT;
4371
4372         kfree(sa);
4373         return ret;
4374 }
4375
4376 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4377                                       void __user *arg)
4378 {
4379         struct btrfs_ioctl_get_dev_stats *sa;
4380         int ret;
4381
4382         sa = memdup_user(arg, sizeof(*sa));
4383         if (IS_ERR(sa))
4384                 return PTR_ERR(sa);
4385
4386         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4387                 kfree(sa);
4388                 return -EPERM;
4389         }
4390
4391         ret = btrfs_get_dev_stats(fs_info, sa);
4392
4393         if (copy_to_user(arg, sa, sizeof(*sa)))
4394                 ret = -EFAULT;
4395
4396         kfree(sa);
4397         return ret;
4398 }
4399
4400 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4401                                     void __user *arg)
4402 {
4403         struct btrfs_ioctl_dev_replace_args *p;
4404         int ret;
4405
4406         if (!capable(CAP_SYS_ADMIN))
4407                 return -EPERM;
4408
4409         p = memdup_user(arg, sizeof(*p));
4410         if (IS_ERR(p))
4411                 return PTR_ERR(p);
4412
4413         switch (p->cmd) {
4414         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4415                 if (sb_rdonly(fs_info->sb)) {
4416                         ret = -EROFS;
4417                         goto out;
4418                 }
4419                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4420                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4421                 } else {
4422                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4423                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4424                 }
4425                 break;
4426         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4427                 btrfs_dev_replace_status(fs_info, p);
4428                 ret = 0;
4429                 break;
4430         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4431                 ret = btrfs_dev_replace_cancel(fs_info, p);
4432                 break;
4433         default:
4434                 ret = -EINVAL;
4435                 break;
4436         }
4437
4438         if (copy_to_user(arg, p, sizeof(*p)))
4439                 ret = -EFAULT;
4440 out:
4441         kfree(p);
4442         return ret;
4443 }
4444
4445 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4446 {
4447         int ret = 0;
4448         int i;
4449         u64 rel_ptr;
4450         int size;
4451         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4452         struct inode_fs_paths *ipath = NULL;
4453         struct btrfs_path *path;
4454
4455         if (!capable(CAP_DAC_READ_SEARCH))
4456                 return -EPERM;
4457
4458         path = btrfs_alloc_path();
4459         if (!path) {
4460                 ret = -ENOMEM;
4461                 goto out;
4462         }
4463
4464         ipa = memdup_user(arg, sizeof(*ipa));
4465         if (IS_ERR(ipa)) {
4466                 ret = PTR_ERR(ipa);
4467                 ipa = NULL;
4468                 goto out;
4469         }
4470
4471         size = min_t(u32, ipa->size, 4096);
4472         ipath = init_ipath(size, root, path);
4473         if (IS_ERR(ipath)) {
4474                 ret = PTR_ERR(ipath);
4475                 ipath = NULL;
4476                 goto out;
4477         }
4478
4479         ret = paths_from_inode(ipa->inum, ipath);
4480         if (ret < 0)
4481                 goto out;
4482
4483         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4484                 rel_ptr = ipath->fspath->val[i] -
4485                           (u64)(unsigned long)ipath->fspath->val;
4486                 ipath->fspath->val[i] = rel_ptr;
4487         }
4488
4489         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4490                            ipath->fspath, size);
4491         if (ret) {
4492                 ret = -EFAULT;
4493                 goto out;
4494         }
4495
4496 out:
4497         btrfs_free_path(path);
4498         free_ipath(ipath);
4499         kfree(ipa);
4500
4501         return ret;
4502 }
4503
4504 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4505 {
4506         struct btrfs_data_container *inodes = ctx;
4507         const size_t c = 3 * sizeof(u64);
4508
4509         if (inodes->bytes_left >= c) {
4510                 inodes->bytes_left -= c;
4511                 inodes->val[inodes->elem_cnt] = inum;
4512                 inodes->val[inodes->elem_cnt + 1] = offset;
4513                 inodes->val[inodes->elem_cnt + 2] = root;
4514                 inodes->elem_cnt += 3;
4515         } else {
4516                 inodes->bytes_missing += c - inodes->bytes_left;
4517                 inodes->bytes_left = 0;
4518                 inodes->elem_missed += 3;
4519         }
4520
4521         return 0;
4522 }
4523
4524 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4525                                         void __user *arg, int version)
4526 {
4527         int ret = 0;
4528         int size;
4529         struct btrfs_ioctl_logical_ino_args *loi;
4530         struct btrfs_data_container *inodes = NULL;
4531         struct btrfs_path *path = NULL;
4532         bool ignore_offset;
4533
4534         if (!capable(CAP_SYS_ADMIN))
4535                 return -EPERM;
4536
4537         loi = memdup_user(arg, sizeof(*loi));
4538         if (IS_ERR(loi))
4539                 return PTR_ERR(loi);
4540
4541         if (version == 1) {
4542                 ignore_offset = false;
4543                 size = min_t(u32, loi->size, SZ_64K);
4544         } else {
4545                 /* All reserved bits must be 0 for now */
4546                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4547                         ret = -EINVAL;
4548                         goto out_loi;
4549                 }
4550                 /* Only accept flags we have defined so far */
4551                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4552                         ret = -EINVAL;
4553                         goto out_loi;
4554                 }
4555                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4556                 size = min_t(u32, loi->size, SZ_16M);
4557         }
4558
4559         path = btrfs_alloc_path();
4560         if (!path) {
4561                 ret = -ENOMEM;
4562                 goto out;
4563         }
4564
4565         inodes = init_data_container(size);
4566         if (IS_ERR(inodes)) {
4567                 ret = PTR_ERR(inodes);
4568                 inodes = NULL;
4569                 goto out;
4570         }
4571
4572         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4573                                           build_ino_list, inodes, ignore_offset);
4574         if (ret == -EINVAL)
4575                 ret = -ENOENT;
4576         if (ret < 0)
4577                 goto out;
4578
4579         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4580                            size);
4581         if (ret)
4582                 ret = -EFAULT;
4583
4584 out:
4585         btrfs_free_path(path);
4586         kvfree(inodes);
4587 out_loi:
4588         kfree(loi);
4589
4590         return ret;
4591 }
4592
4593 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4594                                struct btrfs_ioctl_balance_args *bargs)
4595 {
4596         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4597
4598         bargs->flags = bctl->flags;
4599
4600         if (atomic_read(&fs_info->balance_running))
4601                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4602         if (atomic_read(&fs_info->balance_pause_req))
4603                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4604         if (atomic_read(&fs_info->balance_cancel_req))
4605                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4606
4607         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4608         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4609         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4610
4611         if (lock) {
4612                 spin_lock(&fs_info->balance_lock);
4613                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4614                 spin_unlock(&fs_info->balance_lock);
4615         } else {
4616                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4617         }
4618 }
4619
4620 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4621 {
4622         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4623         struct btrfs_fs_info *fs_info = root->fs_info;
4624         struct btrfs_ioctl_balance_args *bargs;
4625         struct btrfs_balance_control *bctl;
4626         bool need_unlock; /* for mut. excl. ops lock */
4627         int ret;
4628
4629         if (!capable(CAP_SYS_ADMIN))
4630                 return -EPERM;
4631
4632         ret = mnt_want_write_file(file);
4633         if (ret)
4634                 return ret;
4635
4636 again:
4637         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4638                 mutex_lock(&fs_info->volume_mutex);
4639                 mutex_lock(&fs_info->balance_mutex);
4640                 need_unlock = true;
4641                 goto locked;
4642         }
4643
4644         /*
4645          * mut. excl. ops lock is locked.  Three possibilities:
4646          *   (1) some other op is running
4647          *   (2) balance is running
4648          *   (3) balance is paused -- special case (think resume)
4649          */
4650         mutex_lock(&fs_info->balance_mutex);
4651         if (fs_info->balance_ctl) {
4652                 /* this is either (2) or (3) */
4653                 if (!atomic_read(&fs_info->balance_running)) {
4654                         mutex_unlock(&fs_info->balance_mutex);
4655                         if (!mutex_trylock(&fs_info->volume_mutex))
4656                                 goto again;
4657                         mutex_lock(&fs_info->balance_mutex);
4658
4659                         if (fs_info->balance_ctl &&
4660                             !atomic_read(&fs_info->balance_running)) {
4661                                 /* this is (3) */
4662                                 need_unlock = false;
4663                                 goto locked;
4664                         }
4665
4666                         mutex_unlock(&fs_info->balance_mutex);
4667                         mutex_unlock(&fs_info->volume_mutex);
4668                         goto again;
4669                 } else {
4670                         /* this is (2) */
4671                         mutex_unlock(&fs_info->balance_mutex);
4672                         ret = -EINPROGRESS;
4673                         goto out;
4674                 }
4675         } else {
4676                 /* this is (1) */
4677                 mutex_unlock(&fs_info->balance_mutex);
4678                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4679                 goto out;
4680         }
4681
4682 locked:
4683         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4684
4685         if (arg) {
4686                 bargs = memdup_user(arg, sizeof(*bargs));
4687                 if (IS_ERR(bargs)) {
4688                         ret = PTR_ERR(bargs);
4689                         goto out_unlock;
4690                 }
4691
4692                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4693                         if (!fs_info->balance_ctl) {
4694                                 ret = -ENOTCONN;
4695                                 goto out_bargs;
4696                         }
4697
4698                         bctl = fs_info->balance_ctl;
4699                         spin_lock(&fs_info->balance_lock);
4700                         bctl->flags |= BTRFS_BALANCE_RESUME;
4701                         spin_unlock(&fs_info->balance_lock);
4702
4703                         goto do_balance;
4704                 }
4705         } else {
4706                 bargs = NULL;
4707         }
4708
4709         if (fs_info->balance_ctl) {
4710                 ret = -EINPROGRESS;
4711                 goto out_bargs;
4712         }
4713
4714         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4715         if (!bctl) {
4716                 ret = -ENOMEM;
4717                 goto out_bargs;
4718         }
4719
4720         bctl->fs_info = fs_info;
4721         if (arg) {
4722                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4723                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4724                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4725
4726                 bctl->flags = bargs->flags;
4727         } else {
4728                 /* balance everything - no filters */
4729                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4730         }
4731
4732         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4733                 ret = -EINVAL;
4734                 goto out_bctl;
4735         }
4736
4737 do_balance:
4738         /*
4739          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4740          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4741          * or, if restriper was paused all the way until unmount, in
4742          * free_fs_info.  The flag is cleared in __cancel_balance.
4743          */
4744         need_unlock = false;
4745
4746         ret = btrfs_balance(bctl, bargs);
4747         bctl = NULL;
4748
4749         if (arg) {
4750                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4751                         ret = -EFAULT;
4752         }
4753
4754 out_bctl:
4755         kfree(bctl);
4756 out_bargs:
4757         kfree(bargs);
4758 out_unlock:
4759         mutex_unlock(&fs_info->balance_mutex);
4760         mutex_unlock(&fs_info->volume_mutex);
4761         if (need_unlock)
4762                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4763 out:
4764         mnt_drop_write_file(file);
4765         return ret;
4766 }
4767
4768 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4769 {
4770         if (!capable(CAP_SYS_ADMIN))
4771                 return -EPERM;
4772
4773         switch (cmd) {
4774         case BTRFS_BALANCE_CTL_PAUSE:
4775                 return btrfs_pause_balance(fs_info);
4776         case BTRFS_BALANCE_CTL_CANCEL:
4777                 return btrfs_cancel_balance(fs_info);
4778         }
4779
4780         return -EINVAL;
4781 }
4782
4783 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4784                                          void __user *arg)
4785 {
4786         struct btrfs_ioctl_balance_args *bargs;
4787         int ret = 0;
4788
4789         if (!capable(CAP_SYS_ADMIN))
4790                 return -EPERM;
4791
4792         mutex_lock(&fs_info->balance_mutex);
4793         if (!fs_info->balance_ctl) {
4794                 ret = -ENOTCONN;
4795                 goto out;
4796         }
4797
4798         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4799         if (!bargs) {
4800                 ret = -ENOMEM;
4801                 goto out;
4802         }
4803
4804         update_ioctl_balance_args(fs_info, 1, bargs);
4805
4806         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4807                 ret = -EFAULT;
4808
4809         kfree(bargs);
4810 out:
4811         mutex_unlock(&fs_info->balance_mutex);
4812         return ret;
4813 }
4814
4815 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4816 {
4817         struct inode *inode = file_inode(file);
4818         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4819         struct btrfs_ioctl_quota_ctl_args *sa;
4820         struct btrfs_trans_handle *trans = NULL;
4821         int ret;
4822         int err;
4823
4824         if (!capable(CAP_SYS_ADMIN))
4825                 return -EPERM;
4826
4827         ret = mnt_want_write_file(file);
4828         if (ret)
4829                 return ret;
4830
4831         sa = memdup_user(arg, sizeof(*sa));
4832         if (IS_ERR(sa)) {
4833                 ret = PTR_ERR(sa);
4834                 goto drop_write;
4835         }
4836
4837         down_write(&fs_info->subvol_sem);
4838         trans = btrfs_start_transaction(fs_info->tree_root, 2);
4839         if (IS_ERR(trans)) {
4840                 ret = PTR_ERR(trans);
4841                 goto out;
4842         }
4843
4844         switch (sa->cmd) {
4845         case BTRFS_QUOTA_CTL_ENABLE:
4846                 ret = btrfs_quota_enable(trans, fs_info);
4847                 break;
4848         case BTRFS_QUOTA_CTL_DISABLE:
4849                 ret = btrfs_quota_disable(trans, fs_info);
4850                 break;
4851         default:
4852                 ret = -EINVAL;
4853                 break;
4854         }
4855
4856         err = btrfs_commit_transaction(trans);
4857         if (err && !ret)
4858                 ret = err;
4859 out:
4860         kfree(sa);
4861         up_write(&fs_info->subvol_sem);
4862 drop_write:
4863         mnt_drop_write_file(file);
4864         return ret;
4865 }
4866
4867 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4868 {
4869         struct inode *inode = file_inode(file);
4870         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4871         struct btrfs_root *root = BTRFS_I(inode)->root;
4872         struct btrfs_ioctl_qgroup_assign_args *sa;
4873         struct btrfs_trans_handle *trans;
4874         int ret;
4875         int err;
4876
4877         if (!capable(CAP_SYS_ADMIN))
4878                 return -EPERM;
4879
4880         ret = mnt_want_write_file(file);
4881         if (ret)
4882                 return ret;
4883
4884         sa = memdup_user(arg, sizeof(*sa));
4885         if (IS_ERR(sa)) {
4886                 ret = PTR_ERR(sa);
4887                 goto drop_write;
4888         }
4889
4890         trans = btrfs_join_transaction(root);
4891         if (IS_ERR(trans)) {
4892                 ret = PTR_ERR(trans);
4893                 goto out;
4894         }
4895
4896         if (sa->assign) {
4897                 ret = btrfs_add_qgroup_relation(trans, fs_info,
4898                                                 sa->src, sa->dst);
4899         } else {
4900                 ret = btrfs_del_qgroup_relation(trans, fs_info,
4901                                                 sa->src, sa->dst);
4902         }
4903
4904         /* update qgroup status and info */
4905         err = btrfs_run_qgroups(trans, fs_info);
4906         if (err < 0)
4907                 btrfs_handle_fs_error(fs_info, err,
4908                                       "failed to update qgroup status and info");
4909         err = btrfs_end_transaction(trans);
4910         if (err && !ret)
4911                 ret = err;
4912
4913 out:
4914         kfree(sa);
4915 drop_write:
4916         mnt_drop_write_file(file);
4917         return ret;
4918 }
4919
4920 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4921 {
4922         struct inode *inode = file_inode(file);
4923         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4924         struct btrfs_root *root = BTRFS_I(inode)->root;
4925         struct btrfs_ioctl_qgroup_create_args *sa;
4926         struct btrfs_trans_handle *trans;
4927         int ret;
4928         int err;
4929
4930         if (!capable(CAP_SYS_ADMIN))
4931                 return -EPERM;
4932
4933         ret = mnt_want_write_file(file);
4934         if (ret)
4935                 return ret;
4936
4937         sa = memdup_user(arg, sizeof(*sa));
4938         if (IS_ERR(sa)) {
4939                 ret = PTR_ERR(sa);
4940                 goto drop_write;
4941         }
4942
4943         if (!sa->qgroupid) {
4944                 ret = -EINVAL;
4945                 goto out;
4946         }
4947
4948         trans = btrfs_join_transaction(root);
4949         if (IS_ERR(trans)) {
4950                 ret = PTR_ERR(trans);
4951                 goto out;
4952         }
4953
4954         if (sa->create) {
4955                 ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4956         } else {
4957                 ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4958         }
4959
4960         err = btrfs_end_transaction(trans);
4961         if (err && !ret)
4962                 ret = err;
4963
4964 out:
4965         kfree(sa);
4966 drop_write:
4967         mnt_drop_write_file(file);
4968         return ret;
4969 }
4970
4971 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4972 {
4973         struct inode *inode = file_inode(file);
4974         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4975         struct btrfs_root *root = BTRFS_I(inode)->root;
4976         struct btrfs_ioctl_qgroup_limit_args *sa;
4977         struct btrfs_trans_handle *trans;
4978         int ret;
4979         int err;
4980         u64 qgroupid;
4981
4982         if (!capable(CAP_SYS_ADMIN))
4983                 return -EPERM;
4984
4985         ret = mnt_want_write_file(file);
4986         if (ret)
4987                 return ret;
4988
4989         sa = memdup_user(arg, sizeof(*sa));
4990         if (IS_ERR(sa)) {
4991                 ret = PTR_ERR(sa);
4992                 goto drop_write;
4993         }
4994
4995         trans = btrfs_join_transaction(root);
4996         if (IS_ERR(trans)) {
4997                 ret = PTR_ERR(trans);
4998                 goto out;
4999         }
5000
5001         qgroupid = sa->qgroupid;
5002         if (!qgroupid) {
5003                 /* take the current subvol as qgroup */
5004                 qgroupid = root->root_key.objectid;
5005         }
5006
5007         ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
5008
5009         err = btrfs_end_transaction(trans);
5010         if (err && !ret)
5011                 ret = err;
5012
5013 out:
5014         kfree(sa);
5015 drop_write:
5016         mnt_drop_write_file(file);
5017         return ret;
5018 }
5019
5020 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5021 {
5022         struct inode *inode = file_inode(file);
5023         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5024         struct btrfs_ioctl_quota_rescan_args *qsa;
5025         int ret;
5026
5027         if (!capable(CAP_SYS_ADMIN))
5028                 return -EPERM;
5029
5030         ret = mnt_want_write_file(file);
5031         if (ret)
5032                 return ret;
5033
5034         qsa = memdup_user(arg, sizeof(*qsa));
5035         if (IS_ERR(qsa)) {
5036                 ret = PTR_ERR(qsa);
5037                 goto drop_write;
5038         }
5039
5040         if (qsa->flags) {
5041                 ret = -EINVAL;
5042                 goto out;
5043         }
5044
5045         ret = btrfs_qgroup_rescan(fs_info);
5046
5047 out:
5048         kfree(qsa);
5049 drop_write:
5050         mnt_drop_write_file(file);
5051         return ret;
5052 }
5053
5054 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5055 {
5056         struct inode *inode = file_inode(file);
5057         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5058         struct btrfs_ioctl_quota_rescan_args *qsa;
5059         int ret = 0;
5060
5061         if (!capable(CAP_SYS_ADMIN))
5062                 return -EPERM;
5063
5064         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5065         if (!qsa)
5066                 return -ENOMEM;
5067
5068         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5069                 qsa->flags = 1;
5070                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5071         }
5072
5073         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5074                 ret = -EFAULT;
5075
5076         kfree(qsa);
5077         return ret;
5078 }
5079
5080 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5081 {
5082         struct inode *inode = file_inode(file);
5083         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5084
5085         if (!capable(CAP_SYS_ADMIN))
5086                 return -EPERM;
5087
5088         return btrfs_qgroup_wait_for_completion(fs_info, true);
5089 }
5090
5091 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5092                                             struct btrfs_ioctl_received_subvol_args *sa)
5093 {
5094         struct inode *inode = file_inode(file);
5095         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5096         struct btrfs_root *root = BTRFS_I(inode)->root;
5097         struct btrfs_root_item *root_item = &root->root_item;
5098         struct btrfs_trans_handle *trans;
5099         struct timespec ct = current_time(inode);
5100         int ret = 0;
5101         int received_uuid_changed;
5102
5103         if (!inode_owner_or_capable(inode))
5104                 return -EPERM;
5105
5106         ret = mnt_want_write_file(file);
5107         if (ret < 0)
5108                 return ret;
5109
5110         down_write(&fs_info->subvol_sem);
5111
5112         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5113                 ret = -EINVAL;
5114                 goto out;
5115         }
5116
5117         if (btrfs_root_readonly(root)) {
5118                 ret = -EROFS;
5119                 goto out;
5120         }
5121
5122         /*
5123          * 1 - root item
5124          * 2 - uuid items (received uuid + subvol uuid)
5125          */
5126         trans = btrfs_start_transaction(root, 3);
5127         if (IS_ERR(trans)) {
5128                 ret = PTR_ERR(trans);
5129                 trans = NULL;
5130                 goto out;
5131         }
5132
5133         sa->rtransid = trans->transid;
5134         sa->rtime.sec = ct.tv_sec;
5135         sa->rtime.nsec = ct.tv_nsec;
5136
5137         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5138                                        BTRFS_UUID_SIZE);
5139         if (received_uuid_changed &&
5140             !btrfs_is_empty_uuid(root_item->received_uuid))
5141                 btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5142                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5143                                     root->root_key.objectid);
5144         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5145         btrfs_set_root_stransid(root_item, sa->stransid);
5146         btrfs_set_root_rtransid(root_item, sa->rtransid);
5147         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5148         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5149         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5150         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5151
5152         ret = btrfs_update_root(trans, fs_info->tree_root,
5153                                 &root->root_key, &root->root_item);
5154         if (ret < 0) {
5155                 btrfs_end_transaction(trans);
5156                 goto out;
5157         }
5158         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5159                 ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5160                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5161                                           root->root_key.objectid);
5162                 if (ret < 0 && ret != -EEXIST) {
5163                         btrfs_abort_transaction(trans, ret);
5164                         btrfs_end_transaction(trans);
5165                         goto out;
5166                 }
5167         }
5168         ret = btrfs_commit_transaction(trans);
5169 out:
5170         up_write(&fs_info->subvol_sem);
5171         mnt_drop_write_file(file);
5172         return ret;
5173 }
5174
5175 #ifdef CONFIG_64BIT
5176 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5177                                                 void __user *arg)
5178 {
5179         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5180         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5181         int ret = 0;
5182
5183         args32 = memdup_user(arg, sizeof(*args32));
5184         if (IS_ERR(args32))
5185                 return PTR_ERR(args32);
5186
5187         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5188         if (!args64) {
5189                 ret = -ENOMEM;
5190                 goto out;
5191         }
5192
5193         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5194         args64->stransid = args32->stransid;
5195         args64->rtransid = args32->rtransid;
5196         args64->stime.sec = args32->stime.sec;
5197         args64->stime.nsec = args32->stime.nsec;
5198         args64->rtime.sec = args32->rtime.sec;
5199         args64->rtime.nsec = args32->rtime.nsec;
5200         args64->flags = args32->flags;
5201
5202         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5203         if (ret)
5204                 goto out;
5205
5206         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5207         args32->stransid = args64->stransid;
5208         args32->rtransid = args64->rtransid;
5209         args32->stime.sec = args64->stime.sec;
5210         args32->stime.nsec = args64->stime.nsec;
5211         args32->rtime.sec = args64->rtime.sec;
5212         args32->rtime.nsec = args64->rtime.nsec;
5213         args32->flags = args64->flags;
5214
5215         ret = copy_to_user(arg, args32, sizeof(*args32));
5216         if (ret)
5217                 ret = -EFAULT;
5218
5219 out:
5220         kfree(args32);
5221         kfree(args64);
5222         return ret;
5223 }
5224 #endif
5225
5226 static long btrfs_ioctl_set_received_subvol(struct file *file,
5227                                             void __user *arg)
5228 {
5229         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5230         int ret = 0;
5231
5232         sa = memdup_user(arg, sizeof(*sa));
5233         if (IS_ERR(sa))
5234                 return PTR_ERR(sa);
5235
5236         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5237
5238         if (ret)
5239                 goto out;
5240
5241         ret = copy_to_user(arg, sa, sizeof(*sa));
5242         if (ret)
5243                 ret = -EFAULT;
5244
5245 out:
5246         kfree(sa);
5247         return ret;
5248 }
5249
5250 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5251 {
5252         struct inode *inode = file_inode(file);
5253         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5254         size_t len;
5255         int ret;
5256         char label[BTRFS_LABEL_SIZE];
5257
5258         spin_lock(&fs_info->super_lock);
5259         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5260         spin_unlock(&fs_info->super_lock);
5261
5262         len = strnlen(label, BTRFS_LABEL_SIZE);
5263
5264         if (len == BTRFS_LABEL_SIZE) {
5265                 btrfs_warn(fs_info,
5266                            "label is too long, return the first %zu bytes",
5267                            --len);
5268         }
5269
5270         ret = copy_to_user(arg, label, len);
5271
5272         return ret ? -EFAULT : 0;
5273 }
5274
5275 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5276 {
5277         struct inode *inode = file_inode(file);
5278         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5279         struct btrfs_root *root = BTRFS_I(inode)->root;
5280         struct btrfs_super_block *super_block = fs_info->super_copy;
5281         struct btrfs_trans_handle *trans;
5282         char label[BTRFS_LABEL_SIZE];
5283         int ret;
5284
5285         if (!capable(CAP_SYS_ADMIN))
5286                 return -EPERM;
5287
5288         if (copy_from_user(label, arg, sizeof(label)))
5289                 return -EFAULT;
5290
5291         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5292                 btrfs_err(fs_info,
5293                           "unable to set label with more than %d bytes",
5294                           BTRFS_LABEL_SIZE - 1);
5295                 return -EINVAL;
5296         }
5297
5298         ret = mnt_want_write_file(file);
5299         if (ret)
5300                 return ret;
5301
5302         trans = btrfs_start_transaction(root, 0);
5303         if (IS_ERR(trans)) {
5304                 ret = PTR_ERR(trans);
5305                 goto out_unlock;
5306         }
5307
5308         spin_lock(&fs_info->super_lock);
5309         strcpy(super_block->label, label);
5310         spin_unlock(&fs_info->super_lock);
5311         ret = btrfs_commit_transaction(trans);
5312
5313 out_unlock:
5314         mnt_drop_write_file(file);
5315         return ret;
5316 }
5317
5318 #define INIT_FEATURE_FLAGS(suffix) \
5319         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5320           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5321           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5322
5323 int btrfs_ioctl_get_supported_features(void __user *arg)
5324 {
5325         static const struct btrfs_ioctl_feature_flags features[3] = {
5326                 INIT_FEATURE_FLAGS(SUPP),
5327                 INIT_FEATURE_FLAGS(SAFE_SET),
5328                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5329         };
5330
5331         if (copy_to_user(arg, &features, sizeof(features)))
5332                 return -EFAULT;
5333
5334         return 0;
5335 }
5336
5337 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5338 {
5339         struct inode *inode = file_inode(file);
5340         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5341         struct btrfs_super_block *super_block = fs_info->super_copy;
5342         struct btrfs_ioctl_feature_flags features;
5343
5344         features.compat_flags = btrfs_super_compat_flags(super_block);
5345         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5346         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5347
5348         if (copy_to_user(arg, &features, sizeof(features)))
5349                 return -EFAULT;
5350
5351         return 0;
5352 }
5353
5354 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5355                               enum btrfs_feature_set set,
5356                               u64 change_mask, u64 flags, u64 supported_flags,
5357                               u64 safe_set, u64 safe_clear)
5358 {
5359         const char *type = btrfs_feature_set_names[set];
5360         char *names;
5361         u64 disallowed, unsupported;
5362         u64 set_mask = flags & change_mask;
5363         u64 clear_mask = ~flags & change_mask;
5364
5365         unsupported = set_mask & ~supported_flags;
5366         if (unsupported) {
5367                 names = btrfs_printable_features(set, unsupported);
5368                 if (names) {
5369                         btrfs_warn(fs_info,
5370                                    "this kernel does not support the %s feature bit%s",
5371                                    names, strchr(names, ',') ? "s" : "");
5372                         kfree(names);
5373                 } else
5374                         btrfs_warn(fs_info,
5375                                    "this kernel does not support %s bits 0x%llx",
5376                                    type, unsupported);
5377                 return -EOPNOTSUPP;
5378         }
5379
5380         disallowed = set_mask & ~safe_set;
5381         if (disallowed) {
5382                 names = btrfs_printable_features(set, disallowed);
5383                 if (names) {
5384                         btrfs_warn(fs_info,
5385                                    "can't set the %s feature bit%s while mounted",
5386                                    names, strchr(names, ',') ? "s" : "");
5387                         kfree(names);
5388                 } else
5389                         btrfs_warn(fs_info,
5390                                    "can't set %s bits 0x%llx while mounted",
5391                                    type, disallowed);
5392                 return -EPERM;
5393         }
5394
5395         disallowed = clear_mask & ~safe_clear;
5396         if (disallowed) {
5397                 names = btrfs_printable_features(set, disallowed);
5398                 if (names) {
5399                         btrfs_warn(fs_info,
5400                                    "can't clear the %s feature bit%s while mounted",
5401                                    names, strchr(names, ',') ? "s" : "");
5402                         kfree(names);
5403                 } else
5404                         btrfs_warn(fs_info,
5405                                    "can't clear %s bits 0x%llx while mounted",
5406                                    type, disallowed);
5407                 return -EPERM;
5408         }
5409
5410         return 0;
5411 }
5412
5413 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5414 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5415                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5416                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5417                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5418
5419 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5420 {
5421         struct inode *inode = file_inode(file);
5422         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5423         struct btrfs_root *root = BTRFS_I(inode)->root;
5424         struct btrfs_super_block *super_block = fs_info->super_copy;
5425         struct btrfs_ioctl_feature_flags flags[2];
5426         struct btrfs_trans_handle *trans;
5427         u64 newflags;
5428         int ret;
5429
5430         if (!capable(CAP_SYS_ADMIN))
5431                 return -EPERM;
5432
5433         if (copy_from_user(flags, arg, sizeof(flags)))
5434                 return -EFAULT;
5435
5436         /* Nothing to do */
5437         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5438             !flags[0].incompat_flags)
5439                 return 0;
5440
5441         ret = check_feature(fs_info, flags[0].compat_flags,
5442                             flags[1].compat_flags, COMPAT);
5443         if (ret)
5444                 return ret;
5445
5446         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5447                             flags[1].compat_ro_flags, COMPAT_RO);
5448         if (ret)
5449                 return ret;
5450
5451         ret = check_feature(fs_info, flags[0].incompat_flags,
5452                             flags[1].incompat_flags, INCOMPAT);
5453         if (ret)
5454                 return ret;
5455
5456         ret = mnt_want_write_file(file);
5457         if (ret)
5458                 return ret;
5459
5460         trans = btrfs_start_transaction(root, 0);
5461         if (IS_ERR(trans)) {
5462                 ret = PTR_ERR(trans);
5463                 goto out_drop_write;
5464         }
5465
5466         spin_lock(&fs_info->super_lock);
5467         newflags = btrfs_super_compat_flags(super_block);
5468         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5469         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5470         btrfs_set_super_compat_flags(super_block, newflags);
5471
5472         newflags = btrfs_super_compat_ro_flags(super_block);
5473         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5474         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5475         btrfs_set_super_compat_ro_flags(super_block, newflags);
5476
5477         newflags = btrfs_super_incompat_flags(super_block);
5478         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5479         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5480         btrfs_set_super_incompat_flags(super_block, newflags);
5481         spin_unlock(&fs_info->super_lock);
5482
5483         ret = btrfs_commit_transaction(trans);
5484 out_drop_write:
5485         mnt_drop_write_file(file);
5486
5487         return ret;
5488 }
5489
5490 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5491 {
5492         struct btrfs_ioctl_send_args *arg;
5493         int ret;
5494
5495         if (compat) {
5496 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5497                 struct btrfs_ioctl_send_args_32 args32;
5498
5499                 ret = copy_from_user(&args32, argp, sizeof(args32));
5500                 if (ret)
5501                         return -EFAULT;
5502                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5503                 if (!arg)
5504                         return -ENOMEM;
5505                 arg->send_fd = args32.send_fd;
5506                 arg->clone_sources_count = args32.clone_sources_count;
5507                 arg->clone_sources = compat_ptr(args32.clone_sources);
5508                 arg->parent_root = args32.parent_root;
5509                 arg->flags = args32.flags;
5510                 memcpy(arg->reserved, args32.reserved,
5511                        sizeof(args32.reserved));
5512 #else
5513                 return -ENOTTY;
5514 #endif
5515         } else {
5516                 arg = memdup_user(argp, sizeof(*arg));
5517                 if (IS_ERR(arg))
5518                         return PTR_ERR(arg);
5519         }
5520         ret = btrfs_ioctl_send(file, arg);
5521         kfree(arg);
5522         return ret;
5523 }
5524
5525 long btrfs_ioctl(struct file *file, unsigned int
5526                 cmd, unsigned long arg)
5527 {
5528         struct inode *inode = file_inode(file);
5529         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5530         struct btrfs_root *root = BTRFS_I(inode)->root;
5531         void __user *argp = (void __user *)arg;
5532
5533         switch (cmd) {
5534         case FS_IOC_GETFLAGS:
5535                 return btrfs_ioctl_getflags(file, argp);
5536         case FS_IOC_SETFLAGS:
5537                 return btrfs_ioctl_setflags(file, argp);
5538         case FS_IOC_GETVERSION:
5539                 return btrfs_ioctl_getversion(file, argp);
5540         case FITRIM:
5541                 return btrfs_ioctl_fitrim(file, argp);
5542         case BTRFS_IOC_SNAP_CREATE:
5543                 return btrfs_ioctl_snap_create(file, argp, 0);
5544         case BTRFS_IOC_SNAP_CREATE_V2:
5545                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5546         case BTRFS_IOC_SUBVOL_CREATE:
5547                 return btrfs_ioctl_snap_create(file, argp, 1);
5548         case BTRFS_IOC_SUBVOL_CREATE_V2:
5549                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5550         case BTRFS_IOC_SNAP_DESTROY:
5551                 return btrfs_ioctl_snap_destroy(file, argp);
5552         case BTRFS_IOC_SUBVOL_GETFLAGS:
5553                 return btrfs_ioctl_subvol_getflags(file, argp);
5554         case BTRFS_IOC_SUBVOL_SETFLAGS:
5555                 return btrfs_ioctl_subvol_setflags(file, argp);
5556         case BTRFS_IOC_DEFAULT_SUBVOL:
5557                 return btrfs_ioctl_default_subvol(file, argp);
5558         case BTRFS_IOC_DEFRAG:
5559                 return btrfs_ioctl_defrag(file, NULL);
5560         case BTRFS_IOC_DEFRAG_RANGE:
5561                 return btrfs_ioctl_defrag(file, argp);
5562         case BTRFS_IOC_RESIZE:
5563                 return btrfs_ioctl_resize(file, argp);
5564         case BTRFS_IOC_ADD_DEV:
5565                 return btrfs_ioctl_add_dev(fs_info, argp);
5566         case BTRFS_IOC_RM_DEV:
5567                 return btrfs_ioctl_rm_dev(file, argp);
5568         case BTRFS_IOC_RM_DEV_V2:
5569                 return btrfs_ioctl_rm_dev_v2(file, argp);
5570         case BTRFS_IOC_FS_INFO:
5571                 return btrfs_ioctl_fs_info(fs_info, argp);
5572         case BTRFS_IOC_DEV_INFO:
5573                 return btrfs_ioctl_dev_info(fs_info, argp);
5574         case BTRFS_IOC_BALANCE:
5575                 return btrfs_ioctl_balance(file, NULL);
5576         case BTRFS_IOC_TRANS_START:
5577                 return btrfs_ioctl_trans_start(file);
5578         case BTRFS_IOC_TRANS_END:
5579                 return btrfs_ioctl_trans_end(file);
5580         case BTRFS_IOC_TREE_SEARCH:
5581                 return btrfs_ioctl_tree_search(file, argp);
5582         case BTRFS_IOC_TREE_SEARCH_V2:
5583                 return btrfs_ioctl_tree_search_v2(file, argp);
5584         case BTRFS_IOC_INO_LOOKUP:
5585                 return btrfs_ioctl_ino_lookup(file, argp);
5586         case BTRFS_IOC_INO_PATHS:
5587                 return btrfs_ioctl_ino_to_path(root, argp);
5588         case BTRFS_IOC_LOGICAL_INO:
5589                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5590         case BTRFS_IOC_LOGICAL_INO_V2:
5591                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5592         case BTRFS_IOC_SPACE_INFO:
5593                 return btrfs_ioctl_space_info(fs_info, argp);
5594         case BTRFS_IOC_SYNC: {
5595                 int ret;
5596
5597                 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5598                 if (ret)
5599                         return ret;
5600                 ret = btrfs_sync_fs(inode->i_sb, 1);
5601                 /*
5602                  * The transaction thread may want to do more work,
5603                  * namely it pokes the cleaner kthread that will start
5604                  * processing uncleaned subvols.
5605                  */
5606                 wake_up_process(fs_info->transaction_kthread);
5607                 return ret;
5608         }
5609         case BTRFS_IOC_START_SYNC:
5610                 return btrfs_ioctl_start_sync(root, argp);
5611         case BTRFS_IOC_WAIT_SYNC:
5612                 return btrfs_ioctl_wait_sync(fs_info, argp);
5613         case BTRFS_IOC_SCRUB:
5614                 return btrfs_ioctl_scrub(file, argp);
5615         case BTRFS_IOC_SCRUB_CANCEL:
5616                 return btrfs_ioctl_scrub_cancel(fs_info);
5617         case BTRFS_IOC_SCRUB_PROGRESS:
5618                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5619         case BTRFS_IOC_BALANCE_V2:
5620                 return btrfs_ioctl_balance(file, argp);
5621         case BTRFS_IOC_BALANCE_CTL:
5622                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5623         case BTRFS_IOC_BALANCE_PROGRESS:
5624                 return btrfs_ioctl_balance_progress(fs_info, argp);
5625         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5626                 return btrfs_ioctl_set_received_subvol(file, argp);
5627 #ifdef CONFIG_64BIT
5628         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5629                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5630 #endif
5631         case BTRFS_IOC_SEND:
5632                 return _btrfs_ioctl_send(file, argp, false);
5633 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5634         case BTRFS_IOC_SEND_32:
5635                 return _btrfs_ioctl_send(file, argp, true);
5636 #endif
5637         case BTRFS_IOC_GET_DEV_STATS:
5638                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5639         case BTRFS_IOC_QUOTA_CTL:
5640                 return btrfs_ioctl_quota_ctl(file, argp);
5641         case BTRFS_IOC_QGROUP_ASSIGN:
5642                 return btrfs_ioctl_qgroup_assign(file, argp);
5643         case BTRFS_IOC_QGROUP_CREATE:
5644                 return btrfs_ioctl_qgroup_create(file, argp);
5645         case BTRFS_IOC_QGROUP_LIMIT:
5646                 return btrfs_ioctl_qgroup_limit(file, argp);
5647         case BTRFS_IOC_QUOTA_RESCAN:
5648                 return btrfs_ioctl_quota_rescan(file, argp);
5649         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5650                 return btrfs_ioctl_quota_rescan_status(file, argp);
5651         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5652                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5653         case BTRFS_IOC_DEV_REPLACE:
5654                 return btrfs_ioctl_dev_replace(fs_info, argp);
5655         case BTRFS_IOC_GET_FSLABEL:
5656                 return btrfs_ioctl_get_fslabel(file, argp);
5657         case BTRFS_IOC_SET_FSLABEL:
5658                 return btrfs_ioctl_set_fslabel(file, argp);
5659         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5660                 return btrfs_ioctl_get_supported_features(argp);
5661         case BTRFS_IOC_GET_FEATURES:
5662                 return btrfs_ioctl_get_features(file, argp);
5663         case BTRFS_IOC_SET_FEATURES:
5664                 return btrfs_ioctl_set_features(file, argp);
5665         }
5666
5667         return -ENOTTY;
5668 }
5669
5670 #ifdef CONFIG_COMPAT
5671 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5672 {
5673         /*
5674          * These all access 32-bit values anyway so no further
5675          * handling is necessary.
5676          */
5677         switch (cmd) {
5678         case FS_IOC32_GETFLAGS:
5679                 cmd = FS_IOC_GETFLAGS;
5680                 break;
5681         case FS_IOC32_SETFLAGS:
5682                 cmd = FS_IOC_SETFLAGS;
5683                 break;
5684         case FS_IOC32_GETVERSION:
5685                 cmd = FS_IOC_GETVERSION;
5686                 break;
5687         }
5688
5689         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5690 }
5691 #endif