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