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