fs/btrfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/blkdev.h>
   7 #include <linux/module.h>
   8 #include <linux/fs.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/highmem.h>
  11 #include <linux/time.h>
  12 #include <linux/init.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/writeback.h>
  18 #include <linux/statfs.h>
  19 #include <linux/compat.h>
  20 #include <linux/parser.h>
  21 #include <linux/ctype.h>
  22 #include <linux/namei.h>
  23 #include <linux/miscdevice.h>
  24 #include <linux/magic.h>
  25 #include <linux/slab.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/crc32c.h>
  28 #include <linux/btrfs.h>
  29 #include <linux/security.h>
  30 #include <linux/fs_parser.h>
  31 #include "messages.h"
  32 #include "delayed-inode.h"
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "print-tree.h"
  38 #include "props.h"
  39 #include "xattr.h"
  40 #include "bio.h"
  41 #include "export.h"
  42 #include "compression.h"
  43 #include "rcu-string.h"
  44 #include "dev-replace.h"
  45 #include "free-space-cache.h"
  46 #include "backref.h"
  47 #include "space-info.h"
  48 #include "sysfs.h"
  49 #include "zoned.h"
  50 #include "tests/btrfs-tests.h"
  51 #include "block-group.h"
  52 #include "discard.h"
  53 #include "qgroup.h"
  54 #include "raid56.h"
  55 #include "fs.h"
  56 #include "accessors.h"
  57 #include "defrag.h"
  58 #include "dir-item.h"
  59 #include "ioctl.h"
  60 #include "scrub.h"
  61 #include "verity.h"
  62 #include "super.h"
  63 #include "extent-tree.h"
  64 #define CREATE_TRACE_POINTS
  65 #include <trace/events/btrfs.h>
  66
  67 static const struct super_operations btrfs_super_ops;
  68 static struct file_system_type btrfs_fs_type;
  69
  70 static void btrfs_put_super(struct super_block *sb)
  71 {
  72         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
  73
  74         btrfs_info(fs_info, "last unmount of filesystem %pU", fs_info->fs_devices->fsid);
  75         close_ctree(fs_info);
  76 }
  77
  78 /* Store the mount options related information. */
  79 struct btrfs_fs_context {
  80         char *subvol_name;
  81         u64 subvol_objectid;
  82         u64 max_inline;
  83         u32 commit_interval;
  84         u32 metadata_ratio;
  85         u32 thread_pool_size;
  86         unsigned long mount_opt;
  87         unsigned long compress_type:4;
  88         unsigned int compress_level;
  89         refcount_t refs;
  90 };
  91
  92 enum {
  93         Opt_acl,
  94         Opt_clear_cache,
  95         Opt_commit_interval,
  96         Opt_compress,
  97         Opt_compress_force,
  98         Opt_compress_force_type,
  99         Opt_compress_type,
 100         Opt_degraded,
 101         Opt_device,
 102         Opt_fatal_errors,
 103         Opt_flushoncommit,
 104         Opt_max_inline,
 105         Opt_barrier,
 106         Opt_datacow,
 107         Opt_datasum,
 108         Opt_defrag,
 109         Opt_discard,
 110         Opt_discard_mode,
 111         Opt_ratio,
 112         Opt_rescan_uuid_tree,
 113         Opt_skip_balance,
 114         Opt_space_cache,
 115         Opt_space_cache_version,
 116         Opt_ssd,
 117         Opt_ssd_spread,
 118         Opt_subvol,
 119         Opt_subvol_empty,
 120         Opt_subvolid,
 121         Opt_thread_pool,
 122         Opt_treelog,
 123         Opt_user_subvol_rm_allowed,
 124
 125         /* Rescue options */
 126         Opt_rescue,
 127         Opt_usebackuproot,
 128         Opt_nologreplay,
 129         Opt_ignorebadroots,
 130         Opt_ignoredatacsums,
 131         Opt_rescue_all,
 132
 133         /* Debugging options */
 134         Opt_enospc_debug,
 135 #ifdef CONFIG_BTRFS_DEBUG
 136         Opt_fragment, Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 137 #endif
 138 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 139         Opt_ref_verify,
 140 #endif
 141         Opt_err,
 142 };
 143
 144 enum {
 145         Opt_fatal_errors_panic,
 146         Opt_fatal_errors_bug,
 147 };
 148
 149 static const struct constant_table btrfs_parameter_fatal_errors[] = {
 150         { "panic", Opt_fatal_errors_panic },
 151         { "bug", Opt_fatal_errors_bug },
 152         {}
 153 };
 154
 155 enum {
 156         Opt_discard_sync,
 157         Opt_discard_async,
 158 };
 159
 160 static const struct constant_table btrfs_parameter_discard[] = {
 161         { "sync", Opt_discard_sync },
 162         { "async", Opt_discard_async },
 163         {}
 164 };
 165
 166 enum {
 167         Opt_space_cache_v1,
 168         Opt_space_cache_v2,
 169 };
 170
 171 static const struct constant_table btrfs_parameter_space_cache[] = {
 172         { "v1", Opt_space_cache_v1 },
 173         { "v2", Opt_space_cache_v2 },
 174         {}
 175 };
 176
 177 enum {
 178         Opt_rescue_usebackuproot,
 179         Opt_rescue_nologreplay,
 180         Opt_rescue_ignorebadroots,
 181         Opt_rescue_ignoredatacsums,
 182         Opt_rescue_parameter_all,
 183 };
 184
 185 static const struct constant_table btrfs_parameter_rescue[] = {
 186         { "usebackuproot", Opt_rescue_usebackuproot },
 187         { "nologreplay", Opt_rescue_nologreplay },
 188         { "ignorebadroots", Opt_rescue_ignorebadroots },
 189         { "ibadroots", Opt_rescue_ignorebadroots },
 190         { "ignoredatacsums", Opt_rescue_ignoredatacsums },
 191         { "idatacsums", Opt_rescue_ignoredatacsums },
 192         { "all", Opt_rescue_parameter_all },
 193         {}
 194 };
 195
 196 #ifdef CONFIG_BTRFS_DEBUG
 197 enum {
 198         Opt_fragment_parameter_data,
 199         Opt_fragment_parameter_metadata,
 200         Opt_fragment_parameter_all,
 201 };
 202
 203 static const struct constant_table btrfs_parameter_fragment[] = {
 204         { "data", Opt_fragment_parameter_data },
 205         { "metadata", Opt_fragment_parameter_metadata },
 206         { "all", Opt_fragment_parameter_all },
 207         {}
 208 };
 209 #endif
 210
 211 static const struct fs_parameter_spec btrfs_fs_parameters[] = {
 212         fsparam_flag_no("acl", Opt_acl),
 213         fsparam_flag_no("autodefrag", Opt_defrag),
 214         fsparam_flag_no("barrier", Opt_barrier),
 215         fsparam_flag("clear_cache", Opt_clear_cache),
 216         fsparam_u32("commit", Opt_commit_interval),
 217         fsparam_flag("compress", Opt_compress),
 218         fsparam_string("compress", Opt_compress_type),
 219         fsparam_flag("compress-force", Opt_compress_force),
 220         fsparam_string("compress-force", Opt_compress_force_type),
 221         fsparam_flag_no("datacow", Opt_datacow),
 222         fsparam_flag_no("datasum", Opt_datasum),
 223         fsparam_flag("degraded", Opt_degraded),
 224         fsparam_string("device", Opt_device),
 225         fsparam_flag_no("discard", Opt_discard),
 226         fsparam_enum("discard", Opt_discard_mode, btrfs_parameter_discard),
 227         fsparam_enum("fatal_errors", Opt_fatal_errors, btrfs_parameter_fatal_errors),
 228         fsparam_flag_no("flushoncommit", Opt_flushoncommit),
 229         fsparam_string("max_inline", Opt_max_inline),
 230         fsparam_u32("metadata_ratio", Opt_ratio),
 231         fsparam_flag("rescan_uuid_tree", Opt_rescan_uuid_tree),
 232         fsparam_flag("skip_balance", Opt_skip_balance),
 233         fsparam_flag_no("space_cache", Opt_space_cache),
 234         fsparam_enum("space_cache", Opt_space_cache_version, btrfs_parameter_space_cache),
 235         fsparam_flag_no("ssd", Opt_ssd),
 236         fsparam_flag_no("ssd_spread", Opt_ssd_spread),
 237         fsparam_string("subvol", Opt_subvol),
 238         fsparam_flag("subvol=", Opt_subvol_empty),
 239         fsparam_u64("subvolid", Opt_subvolid),
 240         fsparam_u32("thread_pool", Opt_thread_pool),
 241         fsparam_flag_no("treelog", Opt_treelog),
 242         fsparam_flag("user_subvol_rm_allowed", Opt_user_subvol_rm_allowed),
 243
 244         /* Rescue options. */
 245         fsparam_enum("rescue", Opt_rescue, btrfs_parameter_rescue),
 246         /* Deprecated, with alias rescue=nologreplay */
 247         __fsparam(NULL, "nologreplay", Opt_nologreplay, fs_param_deprecated, NULL),
 248         /* Deprecated, with alias rescue=usebackuproot */
 249         __fsparam(NULL, "usebackuproot", Opt_usebackuproot, fs_param_deprecated, NULL),
 250
 251         /* Debugging options. */
 252         fsparam_flag_no("enospc_debug", Opt_enospc_debug),
 253 #ifdef CONFIG_BTRFS_DEBUG
 254         fsparam_enum("fragment", Opt_fragment, btrfs_parameter_fragment),
 255 #endif
 256 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 257         fsparam_flag("ref_verify", Opt_ref_verify),
 258 #endif
 259         {}
 260 };
 261
 262 /* No support for restricting writes to btrfs devices yet... */
 263 static inline blk_mode_t btrfs_open_mode(struct fs_context *fc)
 264 {
 265         return sb_open_mode(fc->sb_flags) & ~BLK_OPEN_RESTRICT_WRITES;
 266 }
 267
 268 static int btrfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
 269 {
 270         struct btrfs_fs_context *ctx = fc->fs_private;
 271         struct fs_parse_result result;
 272         int opt;
 273
 274         opt = fs_parse(fc, btrfs_fs_parameters, param, &result);
 275         if (opt < 0)
 276                 return opt;
 277
 278         switch (opt) {
 279         case Opt_degraded:
 280                 btrfs_set_opt(ctx->mount_opt, DEGRADED);
 281                 break;
 282         case Opt_subvol_empty:
 283                 /*
 284                  * This exists because we used to allow it on accident, so we're
 285                  * keeping it to maintain ABI.  See 37becec95ac3 ("Btrfs: allow
 286                  * empty subvol= again").
 287                  */
 288                 break;
 289         case Opt_subvol:
 290                 kfree(ctx->subvol_name);
 291                 ctx->subvol_name = kstrdup(param->string, GFP_KERNEL);
 292                 if (!ctx->subvol_name)
 293                         return -ENOMEM;
 294                 break;
 295         case Opt_subvolid:
 296                 ctx->subvol_objectid = result.uint_64;
 297
 298                 /* subvolid=0 means give me the original fs_tree. */
 299                 if (!ctx->subvol_objectid)
 300                         ctx->subvol_objectid = BTRFS_FS_TREE_OBJECTID;
 301                 break;
 302         case Opt_device: {
 303                 struct btrfs_device *device;
 304                 blk_mode_t mode = btrfs_open_mode(fc);
 305
 306                 mutex_lock(&uuid_mutex);
 307                 device = btrfs_scan_one_device(param->string, mode, false);
 308                 mutex_unlock(&uuid_mutex);
 309                 if (IS_ERR(device))
 310                         return PTR_ERR(device);
 311                 break;
 312         }
 313         case Opt_datasum:
 314                 if (result.negated) {
 315                         btrfs_set_opt(ctx->mount_opt, NODATASUM);
 316                 } else {
 317                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 318                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 319                 }
 320                 break;
 321         case Opt_datacow:
 322                 if (result.negated) {
 323                         btrfs_clear_opt(ctx->mount_opt, COMPRESS);
 324                         btrfs_clear_opt(ctx->mount_opt, FORCE_COMPRESS);
 325                         btrfs_set_opt(ctx->mount_opt, NODATACOW);
 326                         btrfs_set_opt(ctx->mount_opt, NODATASUM);
 327                 } else {
 328                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 329                 }
 330                 break;
 331         case Opt_compress_force:
 332         case Opt_compress_force_type:
 333                 btrfs_set_opt(ctx->mount_opt, FORCE_COMPRESS);
 334                 fallthrough;
 335         case Opt_compress:
 336         case Opt_compress_type:
 337                 if (opt == Opt_compress || opt == Opt_compress_force) {
 338                         ctx->compress_type = BTRFS_COMPRESS_ZLIB;
 339                         ctx->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 340                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 341                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 342                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 343                 } else if (strncmp(param->string, "zlib", 4) == 0) {
 344                         ctx->compress_type = BTRFS_COMPRESS_ZLIB;
 345                         ctx->compress_level =
 346                                 btrfs_compress_str2level(BTRFS_COMPRESS_ZLIB,
 347                                                          param->string + 4);
 348                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 349                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 350                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 351                 } else if (strncmp(param->string, "lzo", 3) == 0) {
 352                         ctx->compress_type = BTRFS_COMPRESS_LZO;
 353                         ctx->compress_level = 0;
 354                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 355                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 356                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 357                 } else if (strncmp(param->string, "zstd", 4) == 0) {
 358                         ctx->compress_type = BTRFS_COMPRESS_ZSTD;
 359                         ctx->compress_level =
 360                                 btrfs_compress_str2level(BTRFS_COMPRESS_ZSTD,
 361                                                          param->string + 4);
 362                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 363                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 364                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 365                 } else if (strncmp(param->string, "no", 2) == 0) {
 366                         ctx->compress_level = 0;
 367                         ctx->compress_type = 0;
 368                         btrfs_clear_opt(ctx->mount_opt, COMPRESS);
 369                         btrfs_clear_opt(ctx->mount_opt, FORCE_COMPRESS);
 370                 } else {
 371                         btrfs_err(NULL, "unrecognized compression value %s",
 372                                   param->string);
 373                         return -EINVAL;
 374                 }
 375                 break;
 376         case Opt_ssd:
 377                 if (result.negated) {
 378                         btrfs_set_opt(ctx->mount_opt, NOSSD);
 379                         btrfs_clear_opt(ctx->mount_opt, SSD);
 380                         btrfs_clear_opt(ctx->mount_opt, SSD_SPREAD);
 381                 } else {
 382                         btrfs_set_opt(ctx->mount_opt, SSD);
 383                         btrfs_clear_opt(ctx->mount_opt, NOSSD);
 384                 }
 385                 break;
 386         case Opt_ssd_spread:
 387                 if (result.negated) {
 388                         btrfs_clear_opt(ctx->mount_opt, SSD_SPREAD);
 389                 } else {
 390                         btrfs_set_opt(ctx->mount_opt, SSD);
 391                         btrfs_set_opt(ctx->mount_opt, SSD_SPREAD);
 392                         btrfs_clear_opt(ctx->mount_opt, NOSSD);
 393                 }
 394                 break;
 395         case Opt_barrier:
 396                 if (result.negated)
 397                         btrfs_set_opt(ctx->mount_opt, NOBARRIER);
 398                 else
 399                         btrfs_clear_opt(ctx->mount_opt, NOBARRIER);
 400                 break;
 401         case Opt_thread_pool:
 402                 if (result.uint_32 == 0) {
 403                         btrfs_err(NULL, "invalid value 0 for thread_pool");
 404                         return -EINVAL;
 405                 }
 406                 ctx->thread_pool_size = result.uint_32;
 407                 break;
 408         case Opt_max_inline:
 409                 ctx->max_inline = memparse(param->string, NULL);
 410                 break;
 411         case Opt_acl:
 412                 if (result.negated) {
 413                         fc->sb_flags &= ~SB_POSIXACL;
 414                 } else {
 415 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 416                         fc->sb_flags |= SB_POSIXACL;
 417 #else
 418                         btrfs_err(NULL, "support for ACL not compiled in");
 419                         return -EINVAL;
 420 #endif
 421                 }
 422                 /*
 423                  * VFS limits the ability to toggle ACL on and off via remount,
 424                  * despite every file system allowing this.  This seems to be
 425                  * an oversight since we all do, but it'll fail if we're
 426                  * remounting.  So don't set the mask here, we'll check it in
 427                  * btrfs_reconfigure and do the toggling ourselves.
 428                  */
 429                 if (fc->purpose != FS_CONTEXT_FOR_RECONFIGURE)
 430                         fc->sb_flags_mask |= SB_POSIXACL;
 431                 break;
 432         case Opt_treelog:
 433                 if (result.negated)
 434                         btrfs_set_opt(ctx->mount_opt, NOTREELOG);
 435                 else
 436                         btrfs_clear_opt(ctx->mount_opt, NOTREELOG);
 437                 break;
 438         case Opt_nologreplay:
 439                 btrfs_warn(NULL,
 440                 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
 441                 btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 442                 break;
 443         case Opt_flushoncommit:
 444                 if (result.negated)
 445                         btrfs_clear_opt(ctx->mount_opt, FLUSHONCOMMIT);
 446                 else
 447                         btrfs_set_opt(ctx->mount_opt, FLUSHONCOMMIT);
 448                 break;
 449         case Opt_ratio:
 450                 ctx->metadata_ratio = result.uint_32;
 451                 break;
 452         case Opt_discard:
 453                 if (result.negated) {
 454                         btrfs_clear_opt(ctx->mount_opt, DISCARD_SYNC);
 455                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 456                         btrfs_set_opt(ctx->mount_opt, NODISCARD);
 457                 } else {
 458                         btrfs_set_opt(ctx->mount_opt, DISCARD_SYNC);
 459                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 460                 }
 461                 break;
 462         case Opt_discard_mode:
 463                 switch (result.uint_32) {
 464                 case Opt_discard_sync:
 465                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 466                         btrfs_set_opt(ctx->mount_opt, DISCARD_SYNC);
 467                         break;
 468                 case Opt_discard_async:
 469                         btrfs_clear_opt(ctx->mount_opt, DISCARD_SYNC);
 470                         btrfs_set_opt(ctx->mount_opt, DISCARD_ASYNC);
 471                         break;
 472                 default:
 473                         btrfs_err(NULL, "unrecognized discard mode value %s",
 474                                   param->key);
 475                         return -EINVAL;
 476                 }
 477                 btrfs_clear_opt(ctx->mount_opt, NODISCARD);
 478                 break;
 479         case Opt_space_cache:
 480                 if (result.negated) {
 481                         btrfs_set_opt(ctx->mount_opt, NOSPACECACHE);
 482                         btrfs_clear_opt(ctx->mount_opt, SPACE_CACHE);
 483                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 484                 } else {
 485                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 486                         btrfs_set_opt(ctx->mount_opt, SPACE_CACHE);
 487                 }
 488                 break;
 489         case Opt_space_cache_version:
 490                 switch (result.uint_32) {
 491                 case Opt_space_cache_v1:
 492                         btrfs_set_opt(ctx->mount_opt, SPACE_CACHE);
 493                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 494                         break;
 495                 case Opt_space_cache_v2:
 496                         btrfs_clear_opt(ctx->mount_opt, SPACE_CACHE);
 497                         btrfs_set_opt(ctx->mount_opt, FREE_SPACE_TREE);
 498                         break;
 499                 default:
 500                         btrfs_err(NULL, "unrecognized space_cache value %s",
 501                                   param->key);
 502                         return -EINVAL;
 503                 }
 504                 break;
 505         case Opt_rescan_uuid_tree:
 506                 btrfs_set_opt(ctx->mount_opt, RESCAN_UUID_TREE);
 507                 break;
 508         case Opt_clear_cache:
 509                 btrfs_set_opt(ctx->mount_opt, CLEAR_CACHE);
 510                 break;
 511         case Opt_user_subvol_rm_allowed:
 512                 btrfs_set_opt(ctx->mount_opt, USER_SUBVOL_RM_ALLOWED);
 513                 break;
 514         case Opt_enospc_debug:
 515                 if (result.negated)
 516                         btrfs_clear_opt(ctx->mount_opt, ENOSPC_DEBUG);
 517                 else
 518                         btrfs_set_opt(ctx->mount_opt, ENOSPC_DEBUG);
 519                 break;
 520         case Opt_defrag:
 521                 if (result.negated)
 522                         btrfs_clear_opt(ctx->mount_opt, AUTO_DEFRAG);
 523                 else
 524                         btrfs_set_opt(ctx->mount_opt, AUTO_DEFRAG);
 525                 break;
 526         case Opt_usebackuproot:
 527                 btrfs_warn(NULL,
 528                            "'usebackuproot' is deprecated, use 'rescue=usebackuproot' instead");
 529                 btrfs_set_opt(ctx->mount_opt, USEBACKUPROOT);
 530
 531                 /* If we're loading the backup roots we can't trust the space cache. */
 532                 btrfs_set_opt(ctx->mount_opt, CLEAR_CACHE);
 533                 break;
 534         case Opt_skip_balance:
 535                 btrfs_set_opt(ctx->mount_opt, SKIP_BALANCE);
 536                 break;
 537         case Opt_fatal_errors:
 538                 switch (result.uint_32) {
 539                 case Opt_fatal_errors_panic:
 540                         btrfs_set_opt(ctx->mount_opt, PANIC_ON_FATAL_ERROR);
 541                         break;
 542                 case Opt_fatal_errors_bug:
 543                         btrfs_clear_opt(ctx->mount_opt, PANIC_ON_FATAL_ERROR);
 544                         break;
 545                 default:
 546                         btrfs_err(NULL, "unrecognized fatal_errors value %s",
 547                                   param->key);
 548                         return -EINVAL;
 549                 }
 550                 break;
 551         case Opt_commit_interval:
 552                 ctx->commit_interval = result.uint_32;
 553                 if (ctx->commit_interval == 0)
 554                         ctx->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 555                 break;
 556         case Opt_rescue:
 557                 switch (result.uint_32) {
 558                 case Opt_rescue_usebackuproot:
 559                         btrfs_set_opt(ctx->mount_opt, USEBACKUPROOT);
 560                         break;
 561                 case Opt_rescue_nologreplay:
 562                         btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 563                         break;
 564                 case Opt_rescue_ignorebadroots:
 565                         btrfs_set_opt(ctx->mount_opt, IGNOREBADROOTS);
 566                         break;
 567                 case Opt_rescue_ignoredatacsums:
 568                         btrfs_set_opt(ctx->mount_opt, IGNOREDATACSUMS);
 569                         break;
 570                 case Opt_rescue_parameter_all:
 571                         btrfs_set_opt(ctx->mount_opt, IGNOREDATACSUMS);
 572                         btrfs_set_opt(ctx->mount_opt, IGNOREBADROOTS);
 573                         btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 574                         break;
 575                 default:
 576                         btrfs_info(NULL, "unrecognized rescue option '%s'",
 577                                    param->key);
 578                         return -EINVAL;
 579                 }
 580                 break;
 581 #ifdef CONFIG_BTRFS_DEBUG
 582         case Opt_fragment:
 583                 switch (result.uint_32) {
 584                 case Opt_fragment_parameter_all:
 585                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_DATA);
 586                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_METADATA);
 587                         break;
 588                 case Opt_fragment_parameter_metadata:
 589                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_METADATA);
 590                         break;
 591                 case Opt_fragment_parameter_data:
 592                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_DATA);
 593                         break;
 594                 default:
 595                         btrfs_info(NULL, "unrecognized fragment option '%s'",
 596                                    param->key);
 597                         return -EINVAL;
 598                 }
 599                 break;
 600 #endif
 601 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 602         case Opt_ref_verify:
 603                 btrfs_set_opt(ctx->mount_opt, REF_VERIFY);
 604                 break;
 605 #endif
 606         default:
 607                 btrfs_err(NULL, "unrecognized mount option '%s'", param->key);
 608                 return -EINVAL;
 609         }
 610
 611         return 0;
 612 }
 613
 614 /*
 615  * Some options only have meaning at mount time and shouldn't persist across
 616  * remounts, or be displayed. Clear these at the end of mount and remount code
 617  * paths.
 618  */
 619 static void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
 620 {
 621         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
 622         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
 623         btrfs_clear_opt(fs_info->mount_opt, NOSPACECACHE);
 624 }
 625
 626 static bool check_ro_option(struct btrfs_fs_info *fs_info,
 627                             unsigned long mount_opt, unsigned long opt,
 628                             const char *opt_name)
 629 {
 630         if (mount_opt & opt) {
 631                 btrfs_err(fs_info, "%s must be used with ro mount option",
 632                           opt_name);
 633                 return true;
 634         }
 635         return false;
 636 }
 637
 638 bool btrfs_check_options(struct btrfs_fs_info *info, unsigned long *mount_opt,
 639                          unsigned long flags)
 640 {
 641         bool ret = true;
 642
 643         if (!(flags & SB_RDONLY) &&
 644             (check_ro_option(info, *mount_opt, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
 645              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
 646              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums")))
 647                 ret = false;
 648
 649         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 650             !btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE) &&
 651             !btrfs_raw_test_opt(*mount_opt, CLEAR_CACHE)) {
 652                 btrfs_err(info, "cannot disable free-space-tree");
 653                 ret = false;
 654         }
 655         if (btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE) &&
 656              !btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE)) {
 657                 btrfs_err(info, "cannot disable free-space-tree with block-group-tree feature");
 658                 ret = false;
 659         }
 660
 661         if (btrfs_check_mountopts_zoned(info, mount_opt))
 662                 ret = false;
 663
 664         if (!test_bit(BTRFS_FS_STATE_REMOUNTING, &info->fs_state)) {
 665                 if (btrfs_raw_test_opt(*mount_opt, SPACE_CACHE))
 666                         btrfs_info(info, "disk space caching is enabled");
 667                 if (btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE))
 668                         btrfs_info(info, "using free-space-tree");
 669         }
 670
 671         return ret;
 672 }
 673
 674 /*
 675  * This is subtle, we only call this during open_ctree().  We need to pre-load
 676  * the mount options with the on-disk settings.  Before the new mount API took
 677  * effect we would do this on mount and remount.  With the new mount API we'll
 678  * only do this on the initial mount.
 679  *
 680  * This isn't a change in behavior, because we're using the current state of the
 681  * file system to set the current mount options.  If you mounted with special
 682  * options to disable these features and then remounted we wouldn't revert the
 683  * settings, because mounting without these features cleared the on-disk
 684  * settings, so this being called on re-mount is not needed.
 685  */
 686 void btrfs_set_free_space_cache_settings(struct btrfs_fs_info *fs_info)
 687 {
 688         if (fs_info->sectorsize < PAGE_SIZE) {
 689                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
 690                 if (!btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
 691                         btrfs_info(fs_info,
 692                                    "forcing free space tree for sector size %u with page size %lu",
 693                                    fs_info->sectorsize, PAGE_SIZE);
 694                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
 695                 }
 696         }
 697
 698         /*
 699          * At this point our mount options are populated, so we only mess with
 700          * these settings if we don't have any settings already.
 701          */
 702         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
 703                 return;
 704
 705         if (btrfs_is_zoned(fs_info) &&
 706             btrfs_free_space_cache_v1_active(fs_info)) {
 707                 btrfs_info(fs_info, "zoned: clearing existing space cache");
 708                 btrfs_set_super_cache_generation(fs_info->super_copy, 0);
 709                 return;
 710         }
 711
 712         if (btrfs_test_opt(fs_info, SPACE_CACHE))
 713                 return;
 714
 715         if (btrfs_test_opt(fs_info, NOSPACECACHE))
 716                 return;
 717
 718         /*
 719          * At this point we don't have explicit options set by the user, set
 720          * them ourselves based on the state of the file system.
 721          */
 722         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
 723                 btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
 724         else if (btrfs_free_space_cache_v1_active(fs_info))
 725                 btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
 726 }
 727
 728 static void set_device_specific_options(struct btrfs_fs_info *fs_info)
 729 {
 730         if (!btrfs_test_opt(fs_info, NOSSD) &&
 731             !fs_info->fs_devices->rotating)
 732                 btrfs_set_opt(fs_info->mount_opt, SSD);
 733
 734         /*
 735          * For devices supporting discard turn on discard=async automatically,
 736          * unless it's already set or disabled. This could be turned off by
 737          * nodiscard for the same mount.
 738          *
 739          * The zoned mode piggy backs on the discard functionality for
 740          * resetting a zone. There is no reason to delay the zone reset as it is
 741          * fast enough. So, do not enable async discard for zoned mode.
 742          */
 743         if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
 744               btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
 745               btrfs_test_opt(fs_info, NODISCARD)) &&
 746             fs_info->fs_devices->discardable &&
 747             !btrfs_is_zoned(fs_info))
 748                 btrfs_set_opt(fs_info->mount_opt, DISCARD_ASYNC);
 749 }
 750
 751 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 752                                           u64 subvol_objectid)
 753 {
 754         struct btrfs_root *root = fs_info->tree_root;
 755         struct btrfs_root *fs_root = NULL;
 756         struct btrfs_root_ref *root_ref;
 757         struct btrfs_inode_ref *inode_ref;
 758         struct btrfs_key key;
 759         struct btrfs_path *path = NULL;
 760         char *name = NULL, *ptr;
 761         u64 dirid;
 762         int len;
 763         int ret;
 764
 765         path = btrfs_alloc_path();
 766         if (!path) {
 767                 ret = -ENOMEM;
 768                 goto err;
 769         }
 770
 771         name = kmalloc(PATH_MAX, GFP_KERNEL);
 772         if (!name) {
 773                 ret = -ENOMEM;
 774                 goto err;
 775         }
 776         ptr = name + PATH_MAX - 1;
 777         ptr[0] = '\0';
 778
 779         /*
 780          * Walk up the subvolume trees in the tree of tree roots by root
 781          * backrefs until we hit the top-level subvolume.
 782          */
 783         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 784                 key.objectid = subvol_objectid;
 785                 key.type = BTRFS_ROOT_BACKREF_KEY;
 786                 key.offset = (u64)-1;
 787
 788                 ret = btrfs_search_backwards(root, &key, path);
 789                 if (ret < 0) {
 790                         goto err;
 791                 } else if (ret > 0) {
 792                         ret = -ENOENT;
 793                         goto err;
 794                 }
 795
 796                 subvol_objectid = key.offset;
 797
 798                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 799                                           struct btrfs_root_ref);
 800                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 801                 ptr -= len + 1;
 802                 if (ptr < name) {
 803                         ret = -ENAMETOOLONG;
 804                         goto err;
 805                 }
 806                 read_extent_buffer(path->nodes[0], ptr + 1,
 807                                    (unsigned long)(root_ref + 1), len);
 808                 ptr[0] = '/';
 809                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 810                 btrfs_release_path(path);
 811
 812                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
 813                 if (IS_ERR(fs_root)) {
 814                         ret = PTR_ERR(fs_root);
 815                         fs_root = NULL;
 816                         goto err;
 817                 }
 818
 819                 /*
 820                  * Walk up the filesystem tree by inode refs until we hit the
 821                  * root directory.
 822                  */
 823                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
 824                         key.objectid = dirid;
 825                         key.type = BTRFS_INODE_REF_KEY;
 826                         key.offset = (u64)-1;
 827
 828                         ret = btrfs_search_backwards(fs_root, &key, path);
 829                         if (ret < 0) {
 830                                 goto err;
 831                         } else if (ret > 0) {
 832                                 ret = -ENOENT;
 833                                 goto err;
 834                         }
 835
 836                         dirid = key.offset;
 837
 838                         inode_ref = btrfs_item_ptr(path->nodes[0],
 839                                                    path->slots[0],
 840                                                    struct btrfs_inode_ref);
 841                         len = btrfs_inode_ref_name_len(path->nodes[0],
 842                                                        inode_ref);
 843                         ptr -= len + 1;
 844                         if (ptr < name) {
 845                                 ret = -ENAMETOOLONG;
 846                                 goto err;
 847                         }
 848                         read_extent_buffer(path->nodes[0], ptr + 1,
 849                                            (unsigned long)(inode_ref + 1), len);
 850                         ptr[0] = '/';
 851                         btrfs_release_path(path);
 852                 }
 853                 btrfs_put_root(fs_root);
 854                 fs_root = NULL;
 855         }
 856
 857         btrfs_free_path(path);
 858         if (ptr == name + PATH_MAX - 1) {
 859                 name[0] = '/';
 860                 name[1] = '\0';
 861         } else {
 862                 memmove(name, ptr, name + PATH_MAX - ptr);
 863         }
 864         return name;
 865
 866 err:
 867         btrfs_put_root(fs_root);
 868         btrfs_free_path(path);
 869         kfree(name);
 870         return ERR_PTR(ret);
 871 }
 872
 873 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
 874 {
 875         struct btrfs_root *root = fs_info->tree_root;
 876         struct btrfs_dir_item *di;
 877         struct btrfs_path *path;
 878         struct btrfs_key location;
 879         struct fscrypt_str name = FSTR_INIT("default", 7);
 880         u64 dir_id;
 881
 882         path = btrfs_alloc_path();
 883         if (!path)
 884                 return -ENOMEM;
 885
 886         /*
 887          * Find the "default" dir item which points to the root item that we
 888          * will mount by default if we haven't been given a specific subvolume
 889          * to mount.
 890          */
 891         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 892         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, &name, 0);
 893         if (IS_ERR(di)) {
 894                 btrfs_free_path(path);
 895                 return PTR_ERR(di);
 896         }
 897         if (!di) {
 898                 /*
 899                  * Ok the default dir item isn't there.  This is weird since
 900                  * it's always been there, but don't freak out, just try and
 901                  * mount the top-level subvolume.
 902                  */
 903                 btrfs_free_path(path);
 904                 *objectid = BTRFS_FS_TREE_OBJECTID;
 905                 return 0;
 906         }
 907
 908         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 909         btrfs_free_path(path);
 910         *objectid = location.objectid;
 911         return 0;
 912 }
 913
 914 static int btrfs_fill_super(struct super_block *sb,
 915                             struct btrfs_fs_devices *fs_devices,
 916                             void *data)
 917 {
 918         struct inode *inode;
 919         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 920         int err;
 921
 922         sb->s_maxbytes = MAX_LFS_FILESIZE;
 923         sb->s_magic = BTRFS_SUPER_MAGIC;
 924         sb->s_op = &btrfs_super_ops;
 925         sb->s_d_op = &btrfs_dentry_operations;
 926         sb->s_export_op = &btrfs_export_ops;
 927 #ifdef CONFIG_FS_VERITY
 928         sb->s_vop = &btrfs_verityops;
 929 #endif
 930         sb->s_xattr = btrfs_xattr_handlers;
 931         sb->s_time_gran = 1;
 932         sb->s_iflags |= SB_I_CGROUPWB;
 933
 934         err = super_setup_bdi(sb);
 935         if (err) {
 936                 btrfs_err(fs_info, "super_setup_bdi failed");
 937                 return err;
 938         }
 939
 940         err = open_ctree(sb, fs_devices, (char *)data);
 941         if (err) {
 942                 btrfs_err(fs_info, "open_ctree failed");
 943                 return err;
 944         }
 945
 946         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
 947         if (IS_ERR(inode)) {
 948                 err = PTR_ERR(inode);
 949                 btrfs_handle_fs_error(fs_info, err, NULL);
 950                 goto fail_close;
 951         }
 952
 953         sb->s_root = d_make_root(inode);
 954         if (!sb->s_root) {
 955                 err = -ENOMEM;
 956                 goto fail_close;
 957         }
 958
 959         sb->s_flags |= SB_ACTIVE;
 960         return 0;
 961
 962 fail_close:
 963         close_ctree(fs_info);
 964         return err;
 965 }
 966
 967 int btrfs_sync_fs(struct super_block *sb, int wait)
 968 {
 969         struct btrfs_trans_handle *trans;
 970         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 971         struct btrfs_root *root = fs_info->tree_root;
 972
 973         trace_btrfs_sync_fs(fs_info, wait);
 974
 975         if (!wait) {
 976                 filemap_flush(fs_info->btree_inode->i_mapping);
 977                 return 0;
 978         }
 979
 980         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 981
 982         trans = btrfs_attach_transaction_barrier(root);
 983         if (IS_ERR(trans)) {
 984                 /* no transaction, don't bother */
 985                 if (PTR_ERR(trans) == -ENOENT) {
 986                         /*
 987                          * Exit unless we have some pending changes
 988                          * that need to go through commit
 989                          */
 990                         if (!test_bit(BTRFS_FS_NEED_TRANS_COMMIT,
 991                                       &fs_info->flags))
 992                                 return 0;
 993                         /*
 994                          * A non-blocking test if the fs is frozen. We must not
 995                          * start a new transaction here otherwise a deadlock
 996                          * happens. The pending operations are delayed to the
 997                          * next commit after thawing.
 998                          */
 999                         if (sb_start_write_trylock(sb))
1000                                 sb_end_write(sb);
1001                         else
1002                                 return 0;
1003                         trans = btrfs_start_transaction(root, 0);
1004                 }
1005                 if (IS_ERR(trans))
1006                         return PTR_ERR(trans);
1007         }
1008         return btrfs_commit_transaction(trans);
1009 }
1010
1011 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1012 {
1013         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1014         *printed = true;
1015 }
1016
1017 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1018 {
1019         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1020         const char *compress_type;
1021         const char *subvol_name;
1022         bool printed = false;
1023
1024         if (btrfs_test_opt(info, DEGRADED))
1025                 seq_puts(seq, ",degraded");
1026         if (btrfs_test_opt(info, NODATASUM))
1027                 seq_puts(seq, ",nodatasum");
1028         if (btrfs_test_opt(info, NODATACOW))
1029                 seq_puts(seq, ",nodatacow");
1030         if (btrfs_test_opt(info, NOBARRIER))
1031                 seq_puts(seq, ",nobarrier");
1032         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1033                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1034         if (info->thread_pool_size !=  min_t(unsigned long,
1035                                              num_online_cpus() + 2, 8))
1036                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1037         if (btrfs_test_opt(info, COMPRESS)) {
1038                 compress_type = btrfs_compress_type2str(info->compress_type);
1039                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1040                         seq_printf(seq, ",compress-force=%s", compress_type);
1041                 else
1042                         seq_printf(seq, ",compress=%s", compress_type);
1043                 if (info->compress_level)
1044                         seq_printf(seq, ":%d", info->compress_level);
1045         }
1046         if (btrfs_test_opt(info, NOSSD))
1047                 seq_puts(seq, ",nossd");
1048         if (btrfs_test_opt(info, SSD_SPREAD))
1049                 seq_puts(seq, ",ssd_spread");
1050         else if (btrfs_test_opt(info, SSD))
1051                 seq_puts(seq, ",ssd");
1052         if (btrfs_test_opt(info, NOTREELOG))
1053                 seq_puts(seq, ",notreelog");
1054         if (btrfs_test_opt(info, NOLOGREPLAY))
1055                 print_rescue_option(seq, "nologreplay", &printed);
1056         if (btrfs_test_opt(info, USEBACKUPROOT))
1057                 print_rescue_option(seq, "usebackuproot", &printed);
1058         if (btrfs_test_opt(info, IGNOREBADROOTS))
1059                 print_rescue_option(seq, "ignorebadroots", &printed);
1060         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1061                 print_rescue_option(seq, "ignoredatacsums", &printed);
1062         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1063                 seq_puts(seq, ",flushoncommit");
1064         if (btrfs_test_opt(info, DISCARD_SYNC))
1065                 seq_puts(seq, ",discard");
1066         if (btrfs_test_opt(info, DISCARD_ASYNC))
1067                 seq_puts(seq, ",discard=async");
1068         if (!(info->sb->s_flags & SB_POSIXACL))
1069                 seq_puts(seq, ",noacl");
1070         if (btrfs_free_space_cache_v1_active(info))
1071                 seq_puts(seq, ",space_cache");
1072         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1073                 seq_puts(seq, ",space_cache=v2");
1074         else
1075                 seq_puts(seq, ",nospace_cache");
1076         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1077                 seq_puts(seq, ",rescan_uuid_tree");
1078         if (btrfs_test_opt(info, CLEAR_CACHE))
1079                 seq_puts(seq, ",clear_cache");
1080         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1081                 seq_puts(seq, ",user_subvol_rm_allowed");
1082         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1083                 seq_puts(seq, ",enospc_debug");
1084         if (btrfs_test_opt(info, AUTO_DEFRAG))
1085                 seq_puts(seq, ",autodefrag");
1086         if (btrfs_test_opt(info, SKIP_BALANCE))
1087                 seq_puts(seq, ",skip_balance");
1088         if (info->metadata_ratio)
1089                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1090         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1091                 seq_puts(seq, ",fatal_errors=panic");
1092         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1093                 seq_printf(seq, ",commit=%u", info->commit_interval);
1094 #ifdef CONFIG_BTRFS_DEBUG
1095         if (btrfs_test_opt(info, FRAGMENT_DATA))
1096                 seq_puts(seq, ",fragment=data");
1097         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1098                 seq_puts(seq, ",fragment=metadata");
1099 #endif
1100         if (btrfs_test_opt(info, REF_VERIFY))
1101                 seq_puts(seq, ",ref_verify");
1102         seq_printf(seq, ",subvolid=%llu",
1103                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1104         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1105                         BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1106         if (!IS_ERR(subvol_name)) {
1107                 seq_puts(seq, ",subvol=");
1108                 seq_escape(seq, subvol_name, " \t\n\\");
1109                 kfree(subvol_name);
1110         }
1111         return 0;
1112 }
1113
1114 /*
1115  * subvolumes are identified by ino 256
1116  */
1117 static inline int is_subvolume_inode(struct inode *inode)
1118 {
1119         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1120                 return 1;
1121         return 0;
1122 }
1123
1124 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1125                                    struct vfsmount *mnt)
1126 {
1127         struct dentry *root;
1128         int ret;
1129
1130         if (!subvol_name) {
1131                 if (!subvol_objectid) {
1132                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1133                                                           &subvol_objectid);
1134                         if (ret) {
1135                                 root = ERR_PTR(ret);
1136                                 goto out;
1137                         }
1138                 }
1139                 subvol_name = btrfs_get_subvol_name_from_objectid(
1140                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1141                 if (IS_ERR(subvol_name)) {
1142                         root = ERR_CAST(subvol_name);
1143                         subvol_name = NULL;
1144                         goto out;
1145                 }
1146
1147         }
1148
1149         root = mount_subtree(mnt, subvol_name);
1150         /* mount_subtree() drops our reference on the vfsmount. */
1151         mnt = NULL;
1152
1153         if (!IS_ERR(root)) {
1154                 struct super_block *s = root->d_sb;
1155                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1156                 struct inode *root_inode = d_inode(root);
1157                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1158
1159                 ret = 0;
1160                 if (!is_subvolume_inode(root_inode)) {
1161                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1162                                subvol_name);
1163                         ret = -EINVAL;
1164                 }
1165                 if (subvol_objectid && root_objectid != subvol_objectid) {
1166                         /*
1167                          * This will also catch a race condition where a
1168                          * subvolume which was passed by ID is renamed and
1169                          * another subvolume is renamed over the old location.
1170                          */
1171                         btrfs_err(fs_info,
1172                                   "subvol '%s' does not match subvolid %llu",
1173                                   subvol_name, subvol_objectid);
1174                         ret = -EINVAL;
1175                 }
1176                 if (ret) {
1177                         dput(root);
1178                         root = ERR_PTR(ret);
1179                         deactivate_locked_super(s);
1180                 }
1181         }
1182
1183 out:
1184         mntput(mnt);
1185         kfree(subvol_name);
1186         return root;
1187 }
1188
1189 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1190                                      u32 new_pool_size, u32 old_pool_size)
1191 {
1192         if (new_pool_size == old_pool_size)
1193                 return;
1194
1195         fs_info->thread_pool_size = new_pool_size;
1196
1197         btrfs_info(fs_info, "resize thread pool %d -> %d",
1198                old_pool_size, new_pool_size);
1199
1200         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1201         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1202         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1203         workqueue_set_max_active(fs_info->endio_workers, new_pool_size);
1204         workqueue_set_max_active(fs_info->endio_meta_workers, new_pool_size);
1205         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1206         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1207         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1208 }
1209
1210 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1211                                        unsigned long old_opts, int flags)
1212 {
1213         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1214             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1215              (flags & SB_RDONLY))) {
1216                 /* wait for any defraggers to finish */
1217                 wait_event(fs_info->transaction_wait,
1218                            (atomic_read(&fs_info->defrag_running) == 0));
1219                 if (flags & SB_RDONLY)
1220                         sync_filesystem(fs_info->sb);
1221         }
1222 }
1223
1224 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1225                                          unsigned long old_opts)
1226 {
1227         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1228
1229         /*
1230          * We need to cleanup all defragable inodes if the autodefragment is
1231          * close or the filesystem is read only.
1232          */
1233         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1234             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1235                 btrfs_cleanup_defrag_inodes(fs_info);
1236         }
1237
1238         /* If we toggled discard async */
1239         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1240             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1241                 btrfs_discard_resume(fs_info);
1242         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1243                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1244                 btrfs_discard_cleanup(fs_info);
1245
1246         /* If we toggled space cache */
1247         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1248                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1249 }
1250
1251 static int btrfs_remount_rw(struct btrfs_fs_info *fs_info)
1252 {
1253         int ret;
1254
1255         if (BTRFS_FS_ERROR(fs_info)) {
1256                 btrfs_err(fs_info,
1257                           "remounting read-write after error is not allowed");
1258                 return -EINVAL;
1259         }
1260
1261         if (fs_info->fs_devices->rw_devices == 0)
1262                 return -EACCES;
1263
1264         if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1265                 btrfs_warn(fs_info,
1266                            "too many missing devices, writable remount is not allowed");
1267                 return -EACCES;
1268         }
1269
1270         if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1271                 btrfs_warn(fs_info,
1272                            "mount required to replay tree-log, cannot remount read-write");
1273                 return -EINVAL;
1274         }
1275
1276         /*
1277          * NOTE: when remounting with a change that does writes, don't put it
1278          * anywhere above this point, as we are not sure to be safe to write
1279          * until we pass the above checks.
1280          */
1281         ret = btrfs_start_pre_rw_mount(fs_info);
1282         if (ret)
1283                 return ret;
1284
1285         btrfs_clear_sb_rdonly(fs_info->sb);
1286
1287         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1288
1289         /*
1290          * If we've gone from readonly -> read-write, we need to get our
1291          * sync/async discard lists in the right state.
1292          */
1293         btrfs_discard_resume(fs_info);
1294
1295         return 0;
1296 }
1297
1298 static int btrfs_remount_ro(struct btrfs_fs_info *fs_info)
1299 {
1300         /*
1301          * This also happens on 'umount -rf' or on shutdown, when the
1302          * filesystem is busy.
1303          */
1304         cancel_work_sync(&fs_info->async_reclaim_work);
1305         cancel_work_sync(&fs_info->async_data_reclaim_work);
1306
1307         btrfs_discard_cleanup(fs_info);
1308
1309         /* Wait for the uuid_scan task to finish */
1310         down(&fs_info->uuid_tree_rescan_sem);
1311         /* Avoid complains from lockdep et al. */
1312         up(&fs_info->uuid_tree_rescan_sem);
1313
1314         btrfs_set_sb_rdonly(fs_info->sb);
1315
1316         /*
1317          * Setting SB_RDONLY will put the cleaner thread to sleep at the next
1318          * loop if it's already active.  If it's already asleep, we'll leave
1319          * unused block groups on disk until we're mounted read-write again
1320          * unless we clean them up here.
1321          */
1322         btrfs_delete_unused_bgs(fs_info);
1323
1324         /*
1325          * The cleaner task could be already running before we set the flag
1326          * BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).  We must make
1327          * sure that after we finish the remount, i.e. after we call
1328          * btrfs_commit_super(), the cleaner can no longer start a transaction
1329          * - either because it was dropping a dead root, running delayed iputs
1330          *   or deleting an unused block group (the cleaner picked a block
1331          *   group from the list of unused block groups before we were able to
1332          *   in the previous call to btrfs_delete_unused_bgs()).
1333          */
1334         wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING, TASK_UNINTERRUPTIBLE);
1335
1336         /*
1337          * We've set the superblock to RO mode, so we might have made the
1338          * cleaner task sleep without running all pending delayed iputs. Go
1339          * through all the delayed iputs here, so that if an unmount happens
1340          * without remounting RW we don't end up at finishing close_ctree()
1341          * with a non-empty list of delayed iputs.
1342          */
1343         btrfs_run_delayed_iputs(fs_info);
1344
1345         btrfs_dev_replace_suspend_for_unmount(fs_info);
1346         btrfs_scrub_cancel(fs_info);
1347         btrfs_pause_balance(fs_info);
1348
1349         /*
1350          * Pause the qgroup rescan worker if it is running. We don't want it to
1351          * be still running after we are in RO mode, as after that, by the time
1352          * we unmount, it might have left a transaction open, so we would leak
1353          * the transaction and/or crash.
1354          */
1355         btrfs_qgroup_wait_for_completion(fs_info, false);
1356
1357         return btrfs_commit_super(fs_info);
1358 }
1359
1360 static void btrfs_ctx_to_info(struct btrfs_fs_info *fs_info, struct btrfs_fs_context *ctx)
1361 {
1362         fs_info->max_inline = ctx->max_inline;
1363         fs_info->commit_interval = ctx->commit_interval;
1364         fs_info->metadata_ratio = ctx->metadata_ratio;
1365         fs_info->thread_pool_size = ctx->thread_pool_size;
1366         fs_info->mount_opt = ctx->mount_opt;
1367         fs_info->compress_type = ctx->compress_type;
1368         fs_info->compress_level = ctx->compress_level;
1369 }
1370
1371 static void btrfs_info_to_ctx(struct btrfs_fs_info *fs_info, struct btrfs_fs_context *ctx)
1372 {
1373         ctx->max_inline = fs_info->max_inline;
1374         ctx->commit_interval = fs_info->commit_interval;
1375         ctx->metadata_ratio = fs_info->metadata_ratio;
1376         ctx->thread_pool_size = fs_info->thread_pool_size;
1377         ctx->mount_opt = fs_info->mount_opt;
1378         ctx->compress_type = fs_info->compress_type;
1379         ctx->compress_level = fs_info->compress_level;
1380 }
1381
1382 #define btrfs_info_if_set(fs_info, old_ctx, opt, fmt, args...)                  \
1383 do {                                                                            \
1384         if ((!old_ctx || !btrfs_raw_test_opt(old_ctx->mount_opt, opt)) &&       \
1385             btrfs_raw_test_opt(fs_info->mount_opt, opt))                        \
1386                 btrfs_info(fs_info, fmt, ##args);                               \
1387 } while (0)
1388
1389 #define btrfs_info_if_unset(fs_info, old_ctx, opt, fmt, args...)        \
1390 do {                                                                    \
1391         if ((old_ctx && btrfs_raw_test_opt(old_ctx->mount_opt, opt)) && \
1392             !btrfs_raw_test_opt(fs_info->mount_opt, opt))               \
1393                 btrfs_info(fs_info, fmt, ##args);                       \
1394 } while (0)
1395
1396 static void btrfs_emit_options(struct btrfs_fs_info *info,
1397                                struct btrfs_fs_context *old)
1398 {
1399         btrfs_info_if_set(info, old, NODATASUM, "setting nodatasum");
1400         btrfs_info_if_set(info, old, DEGRADED, "allowing degraded mounts");
1401         btrfs_info_if_set(info, old, NODATASUM, "setting nodatasum");
1402         btrfs_info_if_set(info, old, SSD, "enabling ssd optimizations");
1403         btrfs_info_if_set(info, old, SSD_SPREAD, "using spread ssd allocation scheme");
1404         btrfs_info_if_set(info, old, NOBARRIER, "turning off barriers");
1405         btrfs_info_if_set(info, old, NOTREELOG, "disabling tree log");
1406         btrfs_info_if_set(info, old, NOLOGREPLAY, "disabling log replay at mount time");
1407         btrfs_info_if_set(info, old, FLUSHONCOMMIT, "turning on flush-on-commit");
1408         btrfs_info_if_set(info, old, DISCARD_SYNC, "turning on sync discard");
1409         btrfs_info_if_set(info, old, DISCARD_ASYNC, "turning on async discard");
1410         btrfs_info_if_set(info, old, FREE_SPACE_TREE, "enabling free space tree");
1411         btrfs_info_if_set(info, old, SPACE_CACHE, "enabling disk space caching");
1412         btrfs_info_if_set(info, old, CLEAR_CACHE, "force clearing of disk cache");
1413         btrfs_info_if_set(info, old, AUTO_DEFRAG, "enabling auto defrag");
1414         btrfs_info_if_set(info, old, FRAGMENT_DATA, "fragmenting data");
1415         btrfs_info_if_set(info, old, FRAGMENT_METADATA, "fragmenting metadata");
1416         btrfs_info_if_set(info, old, REF_VERIFY, "doing ref verification");
1417         btrfs_info_if_set(info, old, USEBACKUPROOT, "trying to use backup root at mount time");
1418         btrfs_info_if_set(info, old, IGNOREBADROOTS, "ignoring bad roots");
1419         btrfs_info_if_set(info, old, IGNOREDATACSUMS, "ignoring data csums");
1420
1421         btrfs_info_if_unset(info, old, NODATACOW, "setting datacow");
1422         btrfs_info_if_unset(info, old, SSD, "not using ssd optimizations");
1423         btrfs_info_if_unset(info, old, SSD_SPREAD, "not using spread ssd allocation scheme");
1424         btrfs_info_if_unset(info, old, NOBARRIER, "turning off barriers");
1425         btrfs_info_if_unset(info, old, NOTREELOG, "enabling tree log");
1426         btrfs_info_if_unset(info, old, SPACE_CACHE, "disabling disk space caching");
1427         btrfs_info_if_unset(info, old, FREE_SPACE_TREE, "disabling free space tree");
1428         btrfs_info_if_unset(info, old, AUTO_DEFRAG, "disabling auto defrag");
1429         btrfs_info_if_unset(info, old, COMPRESS, "use no compression");
1430
1431         /* Did the compression settings change? */
1432         if (btrfs_test_opt(info, COMPRESS) &&
1433             (!old ||
1434              old->compress_type != info->compress_type ||
1435              old->compress_level != info->compress_level ||
1436              (!btrfs_raw_test_opt(old->mount_opt, FORCE_COMPRESS) &&
1437               btrfs_raw_test_opt(info->mount_opt, FORCE_COMPRESS)))) {
1438                 const char *compress_type = btrfs_compress_type2str(info->compress_type);
1439
1440                 btrfs_info(info, "%s %s compression, level %d",
1441                            btrfs_test_opt(info, FORCE_COMPRESS) ? "force" : "use",
1442                            compress_type, info->compress_level);
1443         }
1444
1445         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1446                 btrfs_info(info, "max_inline set to %llu", info->max_inline);
1447 }
1448
1449 static int btrfs_reconfigure(struct fs_context *fc)
1450 {
1451         struct super_block *sb = fc->root->d_sb;
1452         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1453         struct btrfs_fs_context *ctx = fc->fs_private;
1454         struct btrfs_fs_context old_ctx;
1455         int ret = 0;
1456         bool mount_reconfigure = (fc->s_fs_info != NULL);
1457
1458         btrfs_info_to_ctx(fs_info, &old_ctx);
1459
1460         /*
1461          * This is our "bind mount" trick, we don't want to allow the user to do
1462          * anything other than mount a different ro/rw and a different subvol,
1463          * all of the mount options should be maintained.
1464          */
1465         if (mount_reconfigure)
1466                 ctx->mount_opt = old_ctx.mount_opt;
1467
1468         sync_filesystem(sb);
1469         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1470
1471         if (!mount_reconfigure &&
1472             !btrfs_check_options(fs_info, &ctx->mount_opt, fc->sb_flags))
1473                 return -EINVAL;
1474
1475         ret = btrfs_check_features(fs_info, !(fc->sb_flags & SB_RDONLY));
1476         if (ret < 0)
1477                 return ret;
1478
1479         btrfs_ctx_to_info(fs_info, ctx);
1480         btrfs_remount_begin(fs_info, old_ctx.mount_opt, fc->sb_flags);
1481         btrfs_resize_thread_pool(fs_info, fs_info->thread_pool_size,
1482                                  old_ctx.thread_pool_size);
1483
1484         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
1485             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
1486             (!sb_rdonly(sb) || (fc->sb_flags & SB_RDONLY))) {
1487                 btrfs_warn(fs_info,
1488                 "remount supports changing free space tree only from RO to RW");
1489                 /* Make sure free space cache options match the state on disk. */
1490                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
1491                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1492                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
1493                 }
1494                 if (btrfs_free_space_cache_v1_active(fs_info)) {
1495                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1496                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
1497                 }
1498         }
1499
1500         ret = 0;
1501         if (!sb_rdonly(sb) && (fc->sb_flags & SB_RDONLY))
1502                 ret = btrfs_remount_ro(fs_info);
1503         else if (sb_rdonly(sb) && !(fc->sb_flags & SB_RDONLY))
1504                 ret = btrfs_remount_rw(fs_info);
1505         if (ret)
1506                 goto restore;
1507
1508         /*
1509          * If we set the mask during the parameter parsing VFS would reject the
1510          * remount.  Here we can set the mask and the value will be updated
1511          * appropriately.
1512          */
1513         if ((fc->sb_flags & SB_POSIXACL) != (sb->s_flags & SB_POSIXACL))
1514                 fc->sb_flags_mask |= SB_POSIXACL;
1515
1516         btrfs_emit_options(fs_info, &old_ctx);
1517         wake_up_process(fs_info->transaction_kthread);
1518         btrfs_remount_cleanup(fs_info, old_ctx.mount_opt);
1519         btrfs_clear_oneshot_options(fs_info);
1520         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1521
1522         return 0;
1523 restore:
1524         btrfs_ctx_to_info(fs_info, &old_ctx);
1525         btrfs_remount_cleanup(fs_info, old_ctx.mount_opt);
1526         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1527         return ret;
1528 }
1529
1530 /* Used to sort the devices by max_avail(descending sort) */
1531 static int btrfs_cmp_device_free_bytes(const void *a, const void *b)
1532 {
1533         const struct btrfs_device_info *dev_info1 = a;
1534         const struct btrfs_device_info *dev_info2 = b;
1535
1536         if (dev_info1->max_avail > dev_info2->max_avail)
1537                 return -1;
1538         else if (dev_info1->max_avail < dev_info2->max_avail)
1539                 return 1;
1540         return 0;
1541 }
1542
1543 /*
1544  * sort the devices by max_avail, in which max free extent size of each device
1545  * is stored.(Descending Sort)
1546  */
1547 static inline void btrfs_descending_sort_devices(
1548                                         struct btrfs_device_info *devices,
1549                                         size_t nr_devices)
1550 {
1551         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1552              btrfs_cmp_device_free_bytes, NULL);
1553 }
1554
1555 /*
1556  * The helper to calc the free space on the devices that can be used to store
1557  * file data.
1558  */
1559 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1560                                               u64 *free_bytes)
1561 {
1562         struct btrfs_device_info *devices_info;
1563         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1564         struct btrfs_device *device;
1565         u64 type;
1566         u64 avail_space;
1567         u64 min_stripe_size;
1568         int num_stripes = 1;
1569         int i = 0, nr_devices;
1570         const struct btrfs_raid_attr *rattr;
1571
1572         /*
1573          * We aren't under the device list lock, so this is racy-ish, but good
1574          * enough for our purposes.
1575          */
1576         nr_devices = fs_info->fs_devices->open_devices;
1577         if (!nr_devices) {
1578                 smp_mb();
1579                 nr_devices = fs_info->fs_devices->open_devices;
1580                 ASSERT(nr_devices);
1581                 if (!nr_devices) {
1582                         *free_bytes = 0;
1583                         return 0;
1584                 }
1585         }
1586
1587         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1588                                GFP_KERNEL);
1589         if (!devices_info)
1590                 return -ENOMEM;
1591
1592         /* calc min stripe number for data space allocation */
1593         type = btrfs_data_alloc_profile(fs_info);
1594         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
1595
1596         if (type & BTRFS_BLOCK_GROUP_RAID0)
1597                 num_stripes = nr_devices;
1598         else if (type & BTRFS_BLOCK_GROUP_RAID1_MASK)
1599                 num_stripes = rattr->ncopies;
1600         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1601                 num_stripes = 4;
1602
1603         /* Adjust for more than 1 stripe per device */
1604         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
1605
1606         rcu_read_lock();
1607         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1608                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1609                                                 &device->dev_state) ||
1610                     !device->bdev ||
1611                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1612                         continue;
1613
1614                 if (i >= nr_devices)
1615                         break;
1616
1617                 avail_space = device->total_bytes - device->bytes_used;
1618
1619                 /* align with stripe_len */
1620                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
1621
1622                 /*
1623                  * Ensure we have at least min_stripe_size on top of the
1624                  * reserved space on the device.
1625                  */
1626                 if (avail_space <= BTRFS_DEVICE_RANGE_RESERVED + min_stripe_size)
1627                         continue;
1628
1629                 avail_space -= BTRFS_DEVICE_RANGE_RESERVED;
1630
1631                 devices_info[i].dev = device;
1632                 devices_info[i].max_avail = avail_space;
1633
1634                 i++;
1635         }
1636         rcu_read_unlock();
1637
1638         nr_devices = i;
1639
1640         btrfs_descending_sort_devices(devices_info, nr_devices);
1641
1642         i = nr_devices - 1;
1643         avail_space = 0;
1644         while (nr_devices >= rattr->devs_min) {
1645                 num_stripes = min(num_stripes, nr_devices);
1646
1647                 if (devices_info[i].max_avail >= min_stripe_size) {
1648                         int j;
1649                         u64 alloc_size;
1650
1651                         avail_space += devices_info[i].max_avail * num_stripes;
1652                         alloc_size = devices_info[i].max_avail;
1653                         for (j = i + 1 - num_stripes; j <= i; j++)
1654                                 devices_info[j].max_avail -= alloc_size;
1655                 }
1656                 i--;
1657                 nr_devices--;
1658         }
1659
1660         kfree(devices_info);
1661         *free_bytes = avail_space;
1662         return 0;
1663 }
1664
1665 /*
1666  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
1667  *
1668  * If there's a redundant raid level at DATA block groups, use the respective
1669  * multiplier to scale the sizes.
1670  *
1671  * Unused device space usage is based on simulating the chunk allocator
1672  * algorithm that respects the device sizes and order of allocations.  This is
1673  * a close approximation of the actual use but there are other factors that may
1674  * change the result (like a new metadata chunk).
1675  *
1676  * If metadata is exhausted, f_bavail will be 0.
1677  */
1678 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1679 {
1680         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1681         struct btrfs_super_block *disk_super = fs_info->super_copy;
1682         struct btrfs_space_info *found;
1683         u64 total_used = 0;
1684         u64 total_free_data = 0;
1685         u64 total_free_meta = 0;
1686         u32 bits = fs_info->sectorsize_bits;
1687         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
1688         unsigned factor = 1;
1689         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
1690         int ret;
1691         u64 thresh = 0;
1692         int mixed = 0;
1693
1694         list_for_each_entry(found, &fs_info->space_info, list) {
1695                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1696                         int i;
1697
1698                         total_free_data += found->disk_total - found->disk_used;
1699                         total_free_data -=
1700                                 btrfs_account_ro_block_groups_free_space(found);
1701
1702                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1703                                 if (!list_empty(&found->block_groups[i]))
1704                                         factor = btrfs_bg_type_to_factor(
1705                                                 btrfs_raid_array[i].bg_flag);
1706                         }
1707                 }
1708
1709                 /*
1710                  * Metadata in mixed block group profiles are accounted in data
1711                  */
1712                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
1713                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
1714                                 mixed = 1;
1715                         else
1716                                 total_free_meta += found->disk_total -
1717                                         found->disk_used;
1718                 }
1719
1720                 total_used += found->disk_used;
1721         }
1722
1723         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
1724         buf->f_blocks >>= bits;
1725         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
1726
1727         /* Account global block reserve as used, it's in logical size already */
1728         spin_lock(&block_rsv->lock);
1729         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
1730         if (buf->f_bfree >= block_rsv->size >> bits)
1731                 buf->f_bfree -= block_rsv->size >> bits;
1732         else
1733                 buf->f_bfree = 0;
1734         spin_unlock(&block_rsv->lock);
1735
1736         buf->f_bavail = div_u64(total_free_data, factor);
1737         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
1738         if (ret)
1739                 return ret;
1740         buf->f_bavail += div_u64(total_free_data, factor);
1741         buf->f_bavail = buf->f_bavail >> bits;
1742
1743         /*
1744          * We calculate the remaining metadata space minus global reserve. If
1745          * this is (supposedly) smaller than zero, there's no space. But this
1746          * does not hold in practice, the exhausted state happens where's still
1747          * some positive delta. So we apply some guesswork and compare the
1748          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
1749          *
1750          * We probably cannot calculate the exact threshold value because this
1751          * depends on the internal reservations requested by various
1752          * operations, so some operations that consume a few metadata will
1753          * succeed even if the Avail is zero. But this is better than the other
1754          * way around.
1755          */
1756         thresh = SZ_4M;
1757
1758         /*
1759          * We only want to claim there's no available space if we can no longer
1760          * allocate chunks for our metadata profile and our global reserve will
1761          * not fit in the free metadata space.  If we aren't ->full then we
1762          * still can allocate chunks and thus are fine using the currently
1763          * calculated f_bavail.
1764          */
1765         if (!mixed && block_rsv->space_info->full &&
1766             (total_free_meta < thresh || total_free_meta - thresh < block_rsv->size))
1767                 buf->f_bavail = 0;
1768
1769         buf->f_type = BTRFS_SUPER_MAGIC;
1770         buf->f_bsize = dentry->d_sb->s_blocksize;
1771         buf->f_namelen = BTRFS_NAME_LEN;
1772
1773         /* We treat it as constant endianness (it doesn't matter _which_)
1774            because we want the fsid to come out the same whether mounted
1775            on a big-endian or little-endian host */
1776         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1777         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1778         /* Mask in the root object ID too, to disambiguate subvols */
1779         buf->f_fsid.val[0] ^=
1780                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
1781         buf->f_fsid.val[1] ^=
1782                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
1783
1784         return 0;
1785 }
1786
1787 static int btrfs_fc_test_super(struct super_block *sb, struct fs_context *fc)
1788 {
1789         struct btrfs_fs_info *p = fc->s_fs_info;
1790         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1791
1792         return fs_info->fs_devices == p->fs_devices;
1793 }
1794
1795 static int btrfs_get_tree_super(struct fs_context *fc)
1796 {
1797         struct btrfs_fs_info *fs_info = fc->s_fs_info;
1798         struct btrfs_fs_context *ctx = fc->fs_private;
1799         struct btrfs_fs_devices *fs_devices = NULL;
1800         struct block_device *bdev;
1801         struct btrfs_device *device;
1802         struct super_block *sb;
1803         blk_mode_t mode = btrfs_open_mode(fc);
1804         int ret;
1805
1806         btrfs_ctx_to_info(fs_info, ctx);
1807         mutex_lock(&uuid_mutex);
1808
1809         /*
1810          * With 'true' passed to btrfs_scan_one_device() (mount time) we expect
1811          * either a valid device or an error.
1812          */
1813         device = btrfs_scan_one_device(fc->source, mode, true);
1814         ASSERT(device != NULL);
1815         if (IS_ERR(device)) {
1816                 mutex_unlock(&uuid_mutex);
1817                 return PTR_ERR(device);
1818         }
1819
1820         fs_devices = device->fs_devices;
1821         fs_info->fs_devices = fs_devices;
1822
1823         ret = btrfs_open_devices(fs_devices, mode, &btrfs_fs_type);
1824         mutex_unlock(&uuid_mutex);
1825         if (ret)
1826                 return ret;
1827
1828         if (!(fc->sb_flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1829                 ret = -EACCES;
1830                 goto error;
1831         }
1832
1833         bdev = fs_devices->latest_dev->bdev;
1834
1835         /*
1836          * From now on the error handling is not straightforward.
1837          *
1838          * If successful, this will transfer the fs_info into the super block,
1839          * and fc->s_fs_info will be NULL.  However if there's an existing
1840          * super, we'll still have fc->s_fs_info populated.  If we error
1841          * completely out it'll be cleaned up when we drop the fs_context,
1842          * otherwise it's tied to the lifetime of the super_block.
1843          */
1844         sb = sget_fc(fc, btrfs_fc_test_super, set_anon_super_fc);
1845         if (IS_ERR(sb)) {
1846                 ret = PTR_ERR(sb);
1847                 goto error;
1848         }
1849
1850         set_device_specific_options(fs_info);
1851
1852         if (sb->s_root) {
1853                 btrfs_close_devices(fs_devices);
1854                 if ((fc->sb_flags ^ sb->s_flags) & SB_RDONLY)
1855                         ret = -EBUSY;
1856         } else {
1857                 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1858                 shrinker_debugfs_rename(sb->s_shrink, "sb-btrfs:%s", sb->s_id);
1859                 btrfs_sb(sb)->bdev_holder = &btrfs_fs_type;
1860                 ret = btrfs_fill_super(sb, fs_devices, NULL);
1861         }
1862
1863         if (ret) {
1864                 deactivate_locked_super(sb);
1865                 return ret;
1866         }
1867
1868         btrfs_clear_oneshot_options(fs_info);
1869
1870         fc->root = dget(sb->s_root);
1871         return 0;
1872
1873 error:
1874         btrfs_close_devices(fs_devices);
1875         return ret;
1876 }
1877
1878 /*
1879  * Ever since commit 0723a0473fb4 ("btrfs: allow mounting btrfs subvolumes
1880  * with different ro/rw options") the following works:
1881  *
1882  *        (i) mount /dev/sda3 -o subvol=foo,ro /mnt/foo
1883  *       (ii) mount /dev/sda3 -o subvol=bar,rw /mnt/bar
1884  *
1885  * which looks nice and innocent but is actually pretty intricate and deserves
1886  * a long comment.
1887  *
1888  * On another filesystem a subvolume mount is close to something like:
1889  *
1890  *      (iii) # create rw superblock + initial mount
1891  *            mount -t xfs /dev/sdb /opt/
1892  *
1893  *            # create ro bind mount
1894  *            mount --bind -o ro /opt/foo /mnt/foo
1895  *
1896  *            # unmount initial mount
1897  *            umount /opt
1898  *
1899  * Of course, there's some special subvolume sauce and there's the fact that the
1900  * sb->s_root dentry is really swapped after mount_subtree(). But conceptually
1901  * it's very close and will help us understand the issue.
1902  *
1903  * The old mount API didn't cleanly distinguish between a mount being made ro
1904  * and a superblock being made ro.  The only way to change the ro state of
1905  * either object was by passing ms_rdonly. If a new mount was created via
1906  * mount(2) such as:
1907  *
1908  *      mount("/dev/sdb", "/mnt", "xfs", ms_rdonly, null);
1909  *
1910  * the MS_RDONLY flag being specified had two effects:
1911  *
1912  * (1) MNT_READONLY was raised -> the resulting mount got
1913  *     @mnt->mnt_flags |= MNT_READONLY raised.
1914  *
1915  * (2) MS_RDONLY was passed to the filesystem's mount method and the filesystems
1916  *     made the superblock ro. Note, how SB_RDONLY has the same value as
1917  *     ms_rdonly and is raised whenever MS_RDONLY is passed through mount(2).
1918  *
1919  * Creating a subtree mount via (iii) ends up leaving a rw superblock with a
1920  * subtree mounted ro.
1921  *
1922  * But consider the effect on the old mount API on btrfs subvolume mounting
1923  * which combines the distinct step in (iii) into a single step.
1924  *
1925  * By issuing (i) both the mount and the superblock are turned ro. Now when (ii)
1926  * is issued the superblock is ro and thus even if the mount created for (ii) is
1927  * rw it wouldn't help. Hence, btrfs needed to transition the superblock from ro
1928  * to rw for (ii) which it did using an internal remount call.
1929  *
1930  * IOW, subvolume mounting was inherently complicated due to the ambiguity of
1931  * MS_RDONLY in mount(2). Note, this ambiguity has mount(8) always translate
1932  * "ro" to MS_RDONLY. IOW, in both (i) and (ii) "ro" becomes MS_RDONLY when
1933  * passed by mount(8) to mount(2).
1934  *
1935  * Enter the new mount API. The new mount API disambiguates making a mount ro
1936  * and making a superblock ro.
1937  *
1938  * (3) To turn a mount ro the MOUNT_ATTR_ONLY flag can be used with either
1939  *     fsmount() or mount_setattr() this is a pure VFS level change for a
1940  *     specific mount or mount tree that is never seen by the filesystem itself.
1941  *
1942  * (4) To turn a superblock ro the "ro" flag must be used with
1943  *     fsconfig(FSCONFIG_SET_FLAG, "ro"). This option is seen by the filesystem
1944  *     in fc->sb_flags.
1945  *
1946  * This disambiguation has rather positive consequences.  Mounting a subvolume
1947  * ro will not also turn the superblock ro. Only the mount for the subvolume
1948  * will become ro.
1949  *
1950  * So, if the superblock creation request comes from the new mount API the
1951  * caller must have explicitly done:
1952  *
1953  *      fsconfig(FSCONFIG_SET_FLAG, "ro")
1954  *      fsmount/mount_setattr(MOUNT_ATTR_RDONLY)
1955  *
1956  * IOW, at some point the caller must have explicitly turned the whole
1957  * superblock ro and we shouldn't just undo it like we did for the old mount
1958  * API. In any case, it lets us avoid the hack in the new mount API.
1959  *
1960  * Consequently, the remounting hack must only be used for requests originating
1961  * from the old mount API and should be marked for full deprecation so it can be
1962  * turned off in a couple of years.
1963  *
1964  * The new mount API has no reason to support this hack.
1965  */
1966 static struct vfsmount *btrfs_reconfigure_for_mount(struct fs_context *fc)
1967 {
1968         struct vfsmount *mnt;
1969         int ret;
1970         const bool ro2rw = !(fc->sb_flags & SB_RDONLY);
1971
1972         /*
1973          * We got an EBUSY because our SB_RDONLY flag didn't match the existing
1974          * super block, so invert our setting here and retry the mount so we
1975          * can get our vfsmount.
1976          */
1977         if (ro2rw)
1978                 fc->sb_flags |= SB_RDONLY;
1979         else
1980                 fc->sb_flags &= ~SB_RDONLY;
1981
1982         mnt = fc_mount(fc);
1983         if (IS_ERR(mnt))
1984                 return mnt;
1985
1986         if (!fc->oldapi || !ro2rw)
1987                 return mnt;
1988
1989         /* We need to convert to rw, call reconfigure. */
1990         fc->sb_flags &= ~SB_RDONLY;
1991         down_write(&mnt->mnt_sb->s_umount);
1992         ret = btrfs_reconfigure(fc);
1993         up_write(&mnt->mnt_sb->s_umount);
1994         if (ret) {
1995                 mntput(mnt);
1996                 return ERR_PTR(ret);
1997         }
1998         return mnt;
1999 }
2000
2001 static int btrfs_get_tree_subvol(struct fs_context *fc)
2002 {
2003         struct btrfs_fs_info *fs_info = NULL;
2004         struct btrfs_fs_context *ctx = fc->fs_private;
2005         struct fs_context *dup_fc;
2006         struct dentry *dentry;
2007         struct vfsmount *mnt;
2008
2009         /*
2010          * Setup a dummy root and fs_info for test/set super.  This is because
2011          * we don't actually fill this stuff out until open_ctree, but we need
2012          * then open_ctree will properly initialize the file system specific
2013          * settings later.  btrfs_init_fs_info initializes the static elements
2014          * of the fs_info (locks and such) to make cleanup easier if we find a
2015          * superblock with our given fs_devices later on at sget() time.
2016          */
2017         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
2018         if (!fs_info)
2019                 return -ENOMEM;
2020
2021         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
2022         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
2023         if (!fs_info->super_copy || !fs_info->super_for_commit) {
2024                 btrfs_free_fs_info(fs_info);
2025                 return -ENOMEM;
2026         }
2027         btrfs_init_fs_info(fs_info);
2028
2029         dup_fc = vfs_dup_fs_context(fc);
2030         if (IS_ERR(dup_fc)) {
2031                 btrfs_free_fs_info(fs_info);
2032                 return PTR_ERR(dup_fc);
2033         }
2034
2035         /*
2036          * When we do the sget_fc this gets transferred to the sb, so we only
2037          * need to set it on the dup_fc as this is what creates the super block.
2038          */
2039         dup_fc->s_fs_info = fs_info;
2040
2041         /*
2042          * We'll do the security settings in our btrfs_get_tree_super() mount
2043          * loop, they were duplicated into dup_fc, we can drop the originals
2044          * here.
2045          */
2046         security_free_mnt_opts(&fc->security);
2047         fc->security = NULL;
2048
2049         mnt = fc_mount(dup_fc);
2050         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY)
2051                 mnt = btrfs_reconfigure_for_mount(dup_fc);
2052         put_fs_context(dup_fc);
2053         if (IS_ERR(mnt))
2054                 return PTR_ERR(mnt);
2055
2056         /*
2057          * This free's ->subvol_name, because if it isn't set we have to
2058          * allocate a buffer to hold the subvol_name, so we just drop our
2059          * reference to it here.
2060          */
2061         dentry = mount_subvol(ctx->subvol_name, ctx->subvol_objectid, mnt);
2062         ctx->subvol_name = NULL;
2063         if (IS_ERR(dentry))
2064                 return PTR_ERR(dentry);
2065
2066         fc->root = dentry;
2067         return 0;
2068 }
2069
2070 static int btrfs_get_tree(struct fs_context *fc)
2071 {
2072         /*
2073          * Since we use mount_subtree to mount the default/specified subvol, we
2074          * have to do mounts in two steps.
2075          *
2076          * First pass through we call btrfs_get_tree_subvol(), this is just a
2077          * wrapper around fc_mount() to call back into here again, and this time
2078          * we'll call btrfs_get_tree_super().  This will do the open_ctree() and
2079          * everything to open the devices and file system.  Then we return back
2080          * with a fully constructed vfsmount in btrfs_get_tree_subvol(), and
2081          * from there we can do our mount_subvol() call, which will lookup
2082          * whichever subvol we're mounting and setup this fc with the
2083          * appropriate dentry for the subvol.
2084          */
2085         if (fc->s_fs_info)
2086                 return btrfs_get_tree_super(fc);
2087         return btrfs_get_tree_subvol(fc);
2088 }
2089
2090 static void btrfs_kill_super(struct super_block *sb)
2091 {
2092         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2093         kill_anon_super(sb);
2094         btrfs_free_fs_info(fs_info);
2095 }
2096
2097 static void btrfs_free_fs_context(struct fs_context *fc)
2098 {
2099         struct btrfs_fs_context *ctx = fc->fs_private;
2100         struct btrfs_fs_info *fs_info = fc->s_fs_info;
2101
2102         if (fs_info)
2103                 btrfs_free_fs_info(fs_info);
2104
2105         if (ctx && refcount_dec_and_test(&ctx->refs)) {
2106                 kfree(ctx->subvol_name);
2107                 kfree(ctx);
2108         }
2109 }
2110
2111 static int btrfs_dup_fs_context(struct fs_context *fc, struct fs_context *src_fc)
2112 {
2113         struct btrfs_fs_context *ctx = src_fc->fs_private;
2114
2115         /*
2116          * Give a ref to our ctx to this dup, as we want to keep it around for
2117          * our original fc so we can have the subvolume name or objectid.
2118          *
2119          * We unset ->source in the original fc because the dup needs it for
2120          * mounting, and then once we free the dup it'll free ->source, so we
2121          * need to make sure we're only pointing to it in one fc.
2122          */
2123         refcount_inc(&ctx->refs);
2124         fc->fs_private = ctx;
2125         fc->source = src_fc->source;
2126         src_fc->source = NULL;
2127         return 0;
2128 }
2129
2130 static const struct fs_context_operations btrfs_fs_context_ops = {
2131         .parse_param    = btrfs_parse_param,
2132         .reconfigure    = btrfs_reconfigure,
2133         .get_tree       = btrfs_get_tree,
2134         .dup            = btrfs_dup_fs_context,
2135         .free           = btrfs_free_fs_context,
2136 };
2137
2138 static int btrfs_init_fs_context(struct fs_context *fc)
2139 {
2140         struct btrfs_fs_context *ctx;
2141
2142         ctx = kzalloc(sizeof(struct btrfs_fs_context), GFP_KERNEL);
2143         if (!ctx)
2144                 return -ENOMEM;
2145
2146         refcount_set(&ctx->refs, 1);
2147         fc->fs_private = ctx;
2148         fc->ops = &btrfs_fs_context_ops;
2149
2150         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2151                 btrfs_info_to_ctx(btrfs_sb(fc->root->d_sb), ctx);
2152         } else {
2153                 ctx->thread_pool_size =
2154                         min_t(unsigned long, num_online_cpus() + 2, 8);
2155                 ctx->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2156                 ctx->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2157         }
2158
2159 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
2160         fc->sb_flags |= SB_POSIXACL;
2161 #endif
2162         fc->sb_flags |= SB_I_VERSION;
2163
2164         return 0;
2165 }
2166
2167 static struct file_system_type btrfs_fs_type = {
2168         .owner                  = THIS_MODULE,
2169         .name                   = "btrfs",
2170         .init_fs_context        = btrfs_init_fs_context,
2171         .parameters             = btrfs_fs_parameters,
2172         .kill_sb                = btrfs_kill_super,
2173         .fs_flags               = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA | FS_ALLOW_IDMAP,
2174  };
2175
2176 MODULE_ALIAS_FS("btrfs");
2177
2178 static int btrfs_control_open(struct inode *inode, struct file *file)
2179 {
2180         /*
2181          * The control file's private_data is used to hold the
2182          * transaction when it is started and is used to keep
2183          * track of whether a transaction is already in progress.
2184          */
2185         file->private_data = NULL;
2186         return 0;
2187 }
2188
2189 /*
2190  * Used by /dev/btrfs-control for devices ioctls.
2191  */
2192 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2193                                 unsigned long arg)
2194 {
2195         struct btrfs_ioctl_vol_args *vol;
2196         struct btrfs_device *device = NULL;
2197         dev_t devt = 0;
2198         int ret = -ENOTTY;
2199
2200         if (!capable(CAP_SYS_ADMIN))
2201                 return -EPERM;
2202
2203         vol = memdup_user((void __user *)arg, sizeof(*vol));
2204         if (IS_ERR(vol))
2205                 return PTR_ERR(vol);
2206         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2207
2208         switch (cmd) {
2209         case BTRFS_IOC_SCAN_DEV:
2210                 mutex_lock(&uuid_mutex);
2211                 /*
2212                  * Scanning outside of mount can return NULL which would turn
2213                  * into 0 error code.
2214                  */
2215                 device = btrfs_scan_one_device(vol->name, BLK_OPEN_READ, false);
2216                 ret = PTR_ERR_OR_ZERO(device);
2217                 mutex_unlock(&uuid_mutex);
2218                 break;
2219         case BTRFS_IOC_FORGET_DEV:
2220                 if (vol->name[0] != 0) {
2221                         ret = lookup_bdev(vol->name, &devt);
2222                         if (ret)
2223                                 break;
2224                 }
2225                 ret = btrfs_forget_devices(devt);
2226                 break;
2227         case BTRFS_IOC_DEVICES_READY:
2228                 mutex_lock(&uuid_mutex);
2229                 /*
2230                  * Scanning outside of mount can return NULL which would turn
2231                  * into 0 error code.
2232                  */
2233                 device = btrfs_scan_one_device(vol->name, BLK_OPEN_READ, false);
2234                 if (IS_ERR_OR_NULL(device)) {
2235                         mutex_unlock(&uuid_mutex);
2236                         ret = PTR_ERR(device);
2237                         break;
2238                 }
2239                 ret = !(device->fs_devices->num_devices ==
2240                         device->fs_devices->total_devices);
2241                 mutex_unlock(&uuid_mutex);
2242                 break;
2243         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2244                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2245                 break;
2246         }
2247
2248         kfree(vol);
2249         return ret;
2250 }
2251
2252 static int btrfs_freeze(struct super_block *sb)
2253 {
2254         struct btrfs_trans_handle *trans;
2255         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2256         struct btrfs_root *root = fs_info->tree_root;
2257
2258         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2259         /*
2260          * We don't need a barrier here, we'll wait for any transaction that
2261          * could be in progress on other threads (and do delayed iputs that
2262          * we want to avoid on a frozen filesystem), or do the commit
2263          * ourselves.
2264          */
2265         trans = btrfs_attach_transaction_barrier(root);
2266         if (IS_ERR(trans)) {
2267                 /* no transaction, don't bother */
2268                 if (PTR_ERR(trans) == -ENOENT)
2269                         return 0;
2270                 return PTR_ERR(trans);
2271         }
2272         return btrfs_commit_transaction(trans);
2273 }
2274
2275 static int check_dev_super(struct btrfs_device *dev)
2276 {
2277         struct btrfs_fs_info *fs_info = dev->fs_info;
2278         struct btrfs_super_block *sb;
2279         u64 last_trans;
2280         u16 csum_type;
2281         int ret = 0;
2282
2283         /* This should be called with fs still frozen. */
2284         ASSERT(test_bit(BTRFS_FS_FROZEN, &fs_info->flags));
2285
2286         /* Missing dev, no need to check. */
2287         if (!dev->bdev)
2288                 return 0;
2289
2290         /* Only need to check the primary super block. */
2291         sb = btrfs_read_dev_one_super(dev->bdev, 0, true);
2292         if (IS_ERR(sb))
2293                 return PTR_ERR(sb);
2294
2295         /* Verify the checksum. */
2296         csum_type = btrfs_super_csum_type(sb);
2297         if (csum_type != btrfs_super_csum_type(fs_info->super_copy)) {
2298                 btrfs_err(fs_info, "csum type changed, has %u expect %u",
2299                           csum_type, btrfs_super_csum_type(fs_info->super_copy));
2300                 ret = -EUCLEAN;
2301                 goto out;
2302         }
2303
2304         if (btrfs_check_super_csum(fs_info, sb)) {
2305                 btrfs_err(fs_info, "csum for on-disk super block no longer matches");
2306                 ret = -EUCLEAN;
2307                 goto out;
2308         }
2309
2310         /* Btrfs_validate_super() includes fsid check against super->fsid. */
2311         ret = btrfs_validate_super(fs_info, sb, 0);
2312         if (ret < 0)
2313                 goto out;
2314
2315         last_trans = btrfs_get_last_trans_committed(fs_info);
2316         if (btrfs_super_generation(sb) != last_trans) {
2317                 btrfs_err(fs_info, "transid mismatch, has %llu expect %llu",
2318                           btrfs_super_generation(sb), last_trans);
2319                 ret = -EUCLEAN;
2320                 goto out;
2321         }
2322 out:
2323         btrfs_release_disk_super(sb);
2324         return ret;
2325 }
2326
2327 static int btrfs_unfreeze(struct super_block *sb)
2328 {
2329         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2330         struct btrfs_device *device;
2331         int ret = 0;
2332
2333         /*
2334          * Make sure the fs is not changed by accident (like hibernation then
2335          * modified by other OS).
2336          * If we found anything wrong, we mark the fs error immediately.
2337          *
2338          * And since the fs is frozen, no one can modify the fs yet, thus
2339          * we don't need to hold device_list_mutex.
2340          */
2341         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
2342                 ret = check_dev_super(device);
2343                 if (ret < 0) {
2344                         btrfs_handle_fs_error(fs_info, ret,
2345                                 "super block on devid %llu got modified unexpectedly",
2346                                 device->devid);
2347                         break;
2348                 }
2349         }
2350         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2351
2352         /*
2353          * We still return 0, to allow VFS layer to unfreeze the fs even the
2354          * above checks failed. Since the fs is either fine or read-only, we're
2355          * safe to continue, without causing further damage.
2356          */
2357         return 0;
2358 }
2359
2360 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2361 {
2362         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2363
2364         /*
2365          * There should be always a valid pointer in latest_dev, it may be stale
2366          * for a short moment in case it's being deleted but still valid until
2367          * the end of RCU grace period.
2368          */
2369         rcu_read_lock();
2370         seq_escape(m, btrfs_dev_name(fs_info->fs_devices->latest_dev), " \t\n\\");
2371         rcu_read_unlock();
2372
2373         return 0;
2374 }
2375
2376 static const struct super_operations btrfs_super_ops = {
2377         .drop_inode     = btrfs_drop_inode,
2378         .evict_inode    = btrfs_evict_inode,
2379         .put_super      = btrfs_put_super,
2380         .sync_fs        = btrfs_sync_fs,
2381         .show_options   = btrfs_show_options,
2382         .show_devname   = btrfs_show_devname,
2383         .alloc_inode    = btrfs_alloc_inode,
2384         .destroy_inode  = btrfs_destroy_inode,
2385         .free_inode     = btrfs_free_inode,
2386         .statfs         = btrfs_statfs,
2387         .freeze_fs      = btrfs_freeze,
2388         .unfreeze_fs    = btrfs_unfreeze,
2389 };
2390
2391 static const struct file_operations btrfs_ctl_fops = {
2392         .open = btrfs_control_open,
2393         .unlocked_ioctl  = btrfs_control_ioctl,
2394         .compat_ioctl = compat_ptr_ioctl,
2395         .owner   = THIS_MODULE,
2396         .llseek = noop_llseek,
2397 };
2398
2399 static struct miscdevice btrfs_misc = {
2400         .minor          = BTRFS_MINOR,
2401         .name           = "btrfs-control",
2402         .fops           = &btrfs_ctl_fops
2403 };
2404
2405 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2406 MODULE_ALIAS("devname:btrfs-control");
2407
2408 static int __init btrfs_interface_init(void)
2409 {
2410         return misc_register(&btrfs_misc);
2411 }
2412
2413 static __cold void btrfs_interface_exit(void)
2414 {
2415         misc_deregister(&btrfs_misc);
2416 }
2417
2418 static int __init btrfs_print_mod_info(void)
2419 {
2420         static const char options[] = ""
2421 #ifdef CONFIG_BTRFS_DEBUG
2422                         ", debug=on"
2423 #endif
2424 #ifdef CONFIG_BTRFS_ASSERT
2425                         ", assert=on"
2426 #endif
2427 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2428                         ", ref-verify=on"
2429 #endif
2430 #ifdef CONFIG_BLK_DEV_ZONED
2431                         ", zoned=yes"
2432 #else
2433                         ", zoned=no"
2434 #endif
2435 #ifdef CONFIG_FS_VERITY
2436                         ", fsverity=yes"
2437 #else
2438                         ", fsverity=no"
2439 #endif
2440                         ;
2441         pr_info("Btrfs loaded%s\n", options);
2442         return 0;
2443 }
2444
2445 static int register_btrfs(void)
2446 {
2447         return register_filesystem(&btrfs_fs_type);
2448 }
2449
2450 static void unregister_btrfs(void)
2451 {
2452         unregister_filesystem(&btrfs_fs_type);
2453 }
2454
2455 /* Helper structure for long init/exit functions. */
2456 struct init_sequence {
2457         int (*init_func)(void);
2458         /* Can be NULL if the init_func doesn't need cleanup. */
2459         void (*exit_func)(void);
2460 };
2461
2462 static const struct init_sequence mod_init_seq[] = {
2463         {
2464                 .init_func = btrfs_props_init,
2465                 .exit_func = NULL,
2466         }, {
2467                 .init_func = btrfs_init_sysfs,
2468                 .exit_func = btrfs_exit_sysfs,
2469         }, {
2470                 .init_func = btrfs_init_compress,
2471                 .exit_func = btrfs_exit_compress,
2472         }, {
2473                 .init_func = btrfs_init_cachep,
2474                 .exit_func = btrfs_destroy_cachep,
2475         }, {
2476                 .init_func = btrfs_transaction_init,
2477                 .exit_func = btrfs_transaction_exit,
2478         }, {
2479                 .init_func = btrfs_ctree_init,
2480                 .exit_func = btrfs_ctree_exit,
2481         }, {
2482                 .init_func = btrfs_free_space_init,
2483                 .exit_func = btrfs_free_space_exit,
2484         }, {
2485                 .init_func = extent_state_init_cachep,
2486                 .exit_func = extent_state_free_cachep,
2487         }, {
2488                 .init_func = extent_buffer_init_cachep,
2489                 .exit_func = extent_buffer_free_cachep,
2490         }, {
2491                 .init_func = btrfs_bioset_init,
2492                 .exit_func = btrfs_bioset_exit,
2493         }, {
2494                 .init_func = extent_map_init,
2495                 .exit_func = extent_map_exit,
2496         }, {
2497                 .init_func = ordered_data_init,
2498                 .exit_func = ordered_data_exit,
2499         }, {
2500                 .init_func = btrfs_delayed_inode_init,
2501                 .exit_func = btrfs_delayed_inode_exit,
2502         }, {
2503                 .init_func = btrfs_auto_defrag_init,
2504                 .exit_func = btrfs_auto_defrag_exit,
2505         }, {
2506                 .init_func = btrfs_delayed_ref_init,
2507                 .exit_func = btrfs_delayed_ref_exit,
2508         }, {
2509                 .init_func = btrfs_prelim_ref_init,
2510                 .exit_func = btrfs_prelim_ref_exit,
2511         }, {
2512                 .init_func = btrfs_interface_init,
2513                 .exit_func = btrfs_interface_exit,
2514         }, {
2515                 .init_func = btrfs_print_mod_info,
2516                 .exit_func = NULL,
2517         }, {
2518                 .init_func = btrfs_run_sanity_tests,
2519                 .exit_func = NULL,
2520         }, {
2521                 .init_func = register_btrfs,
2522                 .exit_func = unregister_btrfs,
2523         }
2524 };
2525
2526 static bool mod_init_result[ARRAY_SIZE(mod_init_seq)];
2527
2528 static __always_inline void btrfs_exit_btrfs_fs(void)
2529 {
2530         int i;
2531
2532         for (i = ARRAY_SIZE(mod_init_seq) - 1; i >= 0; i--) {
2533                 if (!mod_init_result[i])
2534                         continue;
2535                 if (mod_init_seq[i].exit_func)
2536                         mod_init_seq[i].exit_func();
2537                 mod_init_result[i] = false;
2538         }
2539 }
2540
2541 static void __exit exit_btrfs_fs(void)
2542 {
2543         btrfs_exit_btrfs_fs();
2544         btrfs_cleanup_fs_uuids();
2545 }
2546
2547 static int __init init_btrfs_fs(void)
2548 {
2549         int ret;
2550         int i;
2551
2552         for (i = 0; i < ARRAY_SIZE(mod_init_seq); i++) {
2553                 ASSERT(!mod_init_result[i]);
2554                 ret = mod_init_seq[i].init_func();
2555                 if (ret < 0) {
2556                         btrfs_exit_btrfs_fs();
2557                         return ret;
2558                 }
2559                 mod_init_result[i] = true;
2560         }
2561         return 0;
2562 }
2563
2564 late_initcall(init_btrfs_fs);
2565 module_exit(exit_btrfs_fs)
2566
2567 MODULE_LICENSE("GPL");
2568 MODULE_SOFTDEP("pre: crc32c");
2569 MODULE_SOFTDEP("pre: xxhash64");
2570 MODULE_SOFTDEP("pre: sha256");
2571 MODULE_SOFTDEP("pre: blake2b-256");