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 "props.h"
  38 #include "xattr.h"
  39 #include "bio.h"
  40 #include "export.h"
  41 #include "compression.h"
  42 #include "dev-replace.h"
  43 #include "free-space-cache.h"
  44 #include "backref.h"
  45 #include "space-info.h"
  46 #include "sysfs.h"
  47 #include "zoned.h"
  48 #include "tests/btrfs-tests.h"
  49 #include "block-group.h"
  50 #include "discard.h"
  51 #include "qgroup.h"
  52 #include "raid56.h"
  53 #include "fs.h"
  54 #include "accessors.h"
  55 #include "defrag.h"
  56 #include "dir-item.h"
  57 #include "ioctl.h"
  58 #include "scrub.h"
  59 #include "verity.h"
  60 #include "super.h"
  61 #include "extent-tree.h"
  62 #define CREATE_TRACE_POINTS
  63 #include <trace/events/btrfs.h>
  64
  65 static const struct super_operations btrfs_super_ops;
  66 static struct file_system_type btrfs_fs_type;
  67
  68 static void btrfs_put_super(struct super_block *sb)
  69 {
  70         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
  71
  72         btrfs_info(fs_info, "last unmount of filesystem %pU", fs_info->fs_devices->fsid);
  73         close_ctree(fs_info);
  74 }
  75
  76 /* Store the mount options related information. */
  77 struct btrfs_fs_context {
  78         char *subvol_name;
  79         u64 subvol_objectid;
  80         u64 max_inline;
  81         u32 commit_interval;
  82         u32 metadata_ratio;
  83         u32 thread_pool_size;
  84         unsigned long mount_opt;
  85         unsigned long compress_type:4;
  86         unsigned int compress_level;
  87         refcount_t refs;
  88 };
  89
  90 enum {
  91         Opt_acl,
  92         Opt_clear_cache,
  93         Opt_commit_interval,
  94         Opt_compress,
  95         Opt_compress_force,
  96         Opt_compress_force_type,
  97         Opt_compress_type,
  98         Opt_degraded,
  99         Opt_device,
 100         Opt_fatal_errors,
 101         Opt_flushoncommit,
 102         Opt_max_inline,
 103         Opt_barrier,
 104         Opt_datacow,
 105         Opt_datasum,
 106         Opt_defrag,
 107         Opt_discard,
 108         Opt_discard_mode,
 109         Opt_ratio,
 110         Opt_rescan_uuid_tree,
 111         Opt_skip_balance,
 112         Opt_space_cache,
 113         Opt_space_cache_version,
 114         Opt_ssd,
 115         Opt_ssd_spread,
 116         Opt_subvol,
 117         Opt_subvol_empty,
 118         Opt_subvolid,
 119         Opt_thread_pool,
 120         Opt_treelog,
 121         Opt_user_subvol_rm_allowed,
 122
 123         /* Rescue options */
 124         Opt_rescue,
 125         Opt_usebackuproot,
 126         Opt_nologreplay,
 127         Opt_ignorebadroots,
 128         Opt_ignoredatacsums,
 129         Opt_rescue_all,
 130
 131         /* Debugging options */
 132         Opt_enospc_debug,
 133 #ifdef CONFIG_BTRFS_DEBUG
 134         Opt_fragment, Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 135 #endif
 136 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 137         Opt_ref_verify,
 138 #endif
 139         Opt_err,
 140 };
 141
 142 enum {
 143         Opt_fatal_errors_panic,
 144         Opt_fatal_errors_bug,
 145 };
 146
 147 static const struct constant_table btrfs_parameter_fatal_errors[] = {
 148         { "panic", Opt_fatal_errors_panic },
 149         { "bug", Opt_fatal_errors_bug },
 150         {}
 151 };
 152
 153 enum {
 154         Opt_discard_sync,
 155         Opt_discard_async,
 156 };
 157
 158 static const struct constant_table btrfs_parameter_discard[] = {
 159         { "sync", Opt_discard_sync },
 160         { "async", Opt_discard_async },
 161         {}
 162 };
 163
 164 enum {
 165         Opt_space_cache_v1,
 166         Opt_space_cache_v2,
 167 };
 168
 169 static const struct constant_table btrfs_parameter_space_cache[] = {
 170         { "v1", Opt_space_cache_v1 },
 171         { "v2", Opt_space_cache_v2 },
 172         {}
 173 };
 174
 175 enum {
 176         Opt_rescue_usebackuproot,
 177         Opt_rescue_nologreplay,
 178         Opt_rescue_ignorebadroots,
 179         Opt_rescue_ignoredatacsums,
 180         Opt_rescue_parameter_all,
 181 };
 182
 183 static const struct constant_table btrfs_parameter_rescue[] = {
 184         { "usebackuproot", Opt_rescue_usebackuproot },
 185         { "nologreplay", Opt_rescue_nologreplay },
 186         { "ignorebadroots", Opt_rescue_ignorebadroots },
 187         { "ibadroots", Opt_rescue_ignorebadroots },
 188         { "ignoredatacsums", Opt_rescue_ignoredatacsums },
 189         { "idatacsums", Opt_rescue_ignoredatacsums },
 190         { "all", Opt_rescue_parameter_all },
 191         {}
 192 };
 193
 194 #ifdef CONFIG_BTRFS_DEBUG
 195 enum {
 196         Opt_fragment_parameter_data,
 197         Opt_fragment_parameter_metadata,
 198         Opt_fragment_parameter_all,
 199 };
 200
 201 static const struct constant_table btrfs_parameter_fragment[] = {
 202         { "data", Opt_fragment_parameter_data },
 203         { "metadata", Opt_fragment_parameter_metadata },
 204         { "all", Opt_fragment_parameter_all },
 205         {}
 206 };
 207 #endif
 208
 209 static const struct fs_parameter_spec btrfs_fs_parameters[] = {
 210         fsparam_flag_no("acl", Opt_acl),
 211         fsparam_flag_no("autodefrag", Opt_defrag),
 212         fsparam_flag_no("barrier", Opt_barrier),
 213         fsparam_flag("clear_cache", Opt_clear_cache),
 214         fsparam_u32("commit", Opt_commit_interval),
 215         fsparam_flag("compress", Opt_compress),
 216         fsparam_string("compress", Opt_compress_type),
 217         fsparam_flag("compress-force", Opt_compress_force),
 218         fsparam_string("compress-force", Opt_compress_force_type),
 219         fsparam_flag_no("datacow", Opt_datacow),
 220         fsparam_flag_no("datasum", Opt_datasum),
 221         fsparam_flag("degraded", Opt_degraded),
 222         fsparam_string("device", Opt_device),
 223         fsparam_flag_no("discard", Opt_discard),
 224         fsparam_enum("discard", Opt_discard_mode, btrfs_parameter_discard),
 225         fsparam_enum("fatal_errors", Opt_fatal_errors, btrfs_parameter_fatal_errors),
 226         fsparam_flag_no("flushoncommit", Opt_flushoncommit),
 227         fsparam_string("max_inline", Opt_max_inline),
 228         fsparam_u32("metadata_ratio", Opt_ratio),
 229         fsparam_flag("rescan_uuid_tree", Opt_rescan_uuid_tree),
 230         fsparam_flag("skip_balance", Opt_skip_balance),
 231         fsparam_flag_no("space_cache", Opt_space_cache),
 232         fsparam_enum("space_cache", Opt_space_cache_version, btrfs_parameter_space_cache),
 233         fsparam_flag_no("ssd", Opt_ssd),
 234         fsparam_flag_no("ssd_spread", Opt_ssd_spread),
 235         fsparam_string("subvol", Opt_subvol),
 236         fsparam_flag("subvol=", Opt_subvol_empty),
 237         fsparam_u64("subvolid", Opt_subvolid),
 238         fsparam_u32("thread_pool", Opt_thread_pool),
 239         fsparam_flag_no("treelog", Opt_treelog),
 240         fsparam_flag("user_subvol_rm_allowed", Opt_user_subvol_rm_allowed),
 241
 242         /* Rescue options. */
 243         fsparam_enum("rescue", Opt_rescue, btrfs_parameter_rescue),
 244         /* Deprecated, with alias rescue=nologreplay */
 245         __fsparam(NULL, "nologreplay", Opt_nologreplay, fs_param_deprecated, NULL),
 246         /* Deprecated, with alias rescue=usebackuproot */
 247         __fsparam(NULL, "usebackuproot", Opt_usebackuproot, fs_param_deprecated, NULL),
 248
 249         /* Debugging options. */
 250         fsparam_flag_no("enospc_debug", Opt_enospc_debug),
 251 #ifdef CONFIG_BTRFS_DEBUG
 252         fsparam_enum("fragment", Opt_fragment, btrfs_parameter_fragment),
 253 #endif
 254 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 255         fsparam_flag("ref_verify", Opt_ref_verify),
 256 #endif
 257         {}
 258 };
 259
 260 /* No support for restricting writes to btrfs devices yet... */
 261 static inline blk_mode_t btrfs_open_mode(struct fs_context *fc)
 262 {
 263         return sb_open_mode(fc->sb_flags) & ~BLK_OPEN_RESTRICT_WRITES;
 264 }
 265
 266 static int btrfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
 267 {
 268         struct btrfs_fs_context *ctx = fc->fs_private;
 269         struct fs_parse_result result;
 270         int opt;
 271
 272         opt = fs_parse(fc, btrfs_fs_parameters, param, &result);
 273         if (opt < 0)
 274                 return opt;
 275
 276         switch (opt) {
 277         case Opt_degraded:
 278                 btrfs_set_opt(ctx->mount_opt, DEGRADED);
 279                 break;
 280         case Opt_subvol_empty:
 281                 /*
 282                  * This exists because we used to allow it on accident, so we're
 283                  * keeping it to maintain ABI.  See 37becec95ac3 ("Btrfs: allow
 284                  * empty subvol= again").
 285                  */
 286                 break;
 287         case Opt_subvol:
 288                 kfree(ctx->subvol_name);
 289                 ctx->subvol_name = kstrdup(param->string, GFP_KERNEL);
 290                 if (!ctx->subvol_name)
 291                         return -ENOMEM;
 292                 break;
 293         case Opt_subvolid:
 294                 ctx->subvol_objectid = result.uint_64;
 295
 296                 /* subvolid=0 means give me the original fs_tree. */
 297                 if (!ctx->subvol_objectid)
 298                         ctx->subvol_objectid = BTRFS_FS_TREE_OBJECTID;
 299                 break;
 300         case Opt_device: {
 301                 struct btrfs_device *device;
 302                 blk_mode_t mode = btrfs_open_mode(fc);
 303
 304                 mutex_lock(&uuid_mutex);
 305                 device = btrfs_scan_one_device(param->string, mode, false);
 306                 mutex_unlock(&uuid_mutex);
 307                 if (IS_ERR(device))
 308                         return PTR_ERR(device);
 309                 break;
 310         }
 311         case Opt_datasum:
 312                 if (result.negated) {
 313                         btrfs_set_opt(ctx->mount_opt, NODATASUM);
 314                 } else {
 315                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 316                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 317                 }
 318                 break;
 319         case Opt_datacow:
 320                 if (result.negated) {
 321                         btrfs_clear_opt(ctx->mount_opt, COMPRESS);
 322                         btrfs_clear_opt(ctx->mount_opt, FORCE_COMPRESS);
 323                         btrfs_set_opt(ctx->mount_opt, NODATACOW);
 324                         btrfs_set_opt(ctx->mount_opt, NODATASUM);
 325                 } else {
 326                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 327                 }
 328                 break;
 329         case Opt_compress_force:
 330         case Opt_compress_force_type:
 331                 btrfs_set_opt(ctx->mount_opt, FORCE_COMPRESS);
 332                 fallthrough;
 333         case Opt_compress:
 334         case Opt_compress_type:
 335                 if (opt == Opt_compress || opt == Opt_compress_force) {
 336                         ctx->compress_type = BTRFS_COMPRESS_ZLIB;
 337                         ctx->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 338                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 339                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 340                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 341                 } else if (strncmp(param->string, "zlib", 4) == 0) {
 342                         ctx->compress_type = BTRFS_COMPRESS_ZLIB;
 343                         ctx->compress_level =
 344                                 btrfs_compress_str2level(BTRFS_COMPRESS_ZLIB,
 345                                                          param->string + 4);
 346                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 347                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 348                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 349                 } else if (strncmp(param->string, "lzo", 3) == 0) {
 350                         ctx->compress_type = BTRFS_COMPRESS_LZO;
 351                         ctx->compress_level = 0;
 352                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 353                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 354                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 355                 } else if (strncmp(param->string, "zstd", 4) == 0) {
 356                         ctx->compress_type = BTRFS_COMPRESS_ZSTD;
 357                         ctx->compress_level =
 358                                 btrfs_compress_str2level(BTRFS_COMPRESS_ZSTD,
 359                                                          param->string + 4);
 360                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 361                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 362                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 363                 } else if (strncmp(param->string, "no", 2) == 0) {
 364                         ctx->compress_level = 0;
 365                         ctx->compress_type = 0;
 366                         btrfs_clear_opt(ctx->mount_opt, COMPRESS);
 367                         btrfs_clear_opt(ctx->mount_opt, FORCE_COMPRESS);
 368                 } else {
 369                         btrfs_err(NULL, "unrecognized compression value %s",
 370                                   param->string);
 371                         return -EINVAL;
 372                 }
 373                 break;
 374         case Opt_ssd:
 375                 if (result.negated) {
 376                         btrfs_set_opt(ctx->mount_opt, NOSSD);
 377                         btrfs_clear_opt(ctx->mount_opt, SSD);
 378                         btrfs_clear_opt(ctx->mount_opt, SSD_SPREAD);
 379                 } else {
 380                         btrfs_set_opt(ctx->mount_opt, SSD);
 381                         btrfs_clear_opt(ctx->mount_opt, NOSSD);
 382                 }
 383                 break;
 384         case Opt_ssd_spread:
 385                 if (result.negated) {
 386                         btrfs_clear_opt(ctx->mount_opt, SSD_SPREAD);
 387                 } else {
 388                         btrfs_set_opt(ctx->mount_opt, SSD);
 389                         btrfs_set_opt(ctx->mount_opt, SSD_SPREAD);
 390                         btrfs_clear_opt(ctx->mount_opt, NOSSD);
 391                 }
 392                 break;
 393         case Opt_barrier:
 394                 if (result.negated)
 395                         btrfs_set_opt(ctx->mount_opt, NOBARRIER);
 396                 else
 397                         btrfs_clear_opt(ctx->mount_opt, NOBARRIER);
 398                 break;
 399         case Opt_thread_pool:
 400                 if (result.uint_32 == 0) {
 401                         btrfs_err(NULL, "invalid value 0 for thread_pool");
 402                         return -EINVAL;
 403                 }
 404                 ctx->thread_pool_size = result.uint_32;
 405                 break;
 406         case Opt_max_inline:
 407                 ctx->max_inline = memparse(param->string, NULL);
 408                 break;
 409         case Opt_acl:
 410                 if (result.negated) {
 411                         fc->sb_flags &= ~SB_POSIXACL;
 412                 } else {
 413 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 414                         fc->sb_flags |= SB_POSIXACL;
 415 #else
 416                         btrfs_err(NULL, "support for ACL not compiled in");
 417                         return -EINVAL;
 418 #endif
 419                 }
 420                 /*
 421                  * VFS limits the ability to toggle ACL on and off via remount,
 422                  * despite every file system allowing this.  This seems to be
 423                  * an oversight since we all do, but it'll fail if we're
 424                  * remounting.  So don't set the mask here, we'll check it in
 425                  * btrfs_reconfigure and do the toggling ourselves.
 426                  */
 427                 if (fc->purpose != FS_CONTEXT_FOR_RECONFIGURE)
 428                         fc->sb_flags_mask |= SB_POSIXACL;
 429                 break;
 430         case Opt_treelog:
 431                 if (result.negated)
 432                         btrfs_set_opt(ctx->mount_opt, NOTREELOG);
 433                 else
 434                         btrfs_clear_opt(ctx->mount_opt, NOTREELOG);
 435                 break;
 436         case Opt_nologreplay:
 437                 btrfs_warn(NULL,
 438                 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
 439                 btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 440                 break;
 441         case Opt_flushoncommit:
 442                 if (result.negated)
 443                         btrfs_clear_opt(ctx->mount_opt, FLUSHONCOMMIT);
 444                 else
 445                         btrfs_set_opt(ctx->mount_opt, FLUSHONCOMMIT);
 446                 break;
 447         case Opt_ratio:
 448                 ctx->metadata_ratio = result.uint_32;
 449                 break;
 450         case Opt_discard:
 451                 if (result.negated) {
 452                         btrfs_clear_opt(ctx->mount_opt, DISCARD_SYNC);
 453                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 454                         btrfs_set_opt(ctx->mount_opt, NODISCARD);
 455                 } else {
 456                         btrfs_set_opt(ctx->mount_opt, DISCARD_SYNC);
 457                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 458                 }
 459                 break;
 460         case Opt_discard_mode:
 461                 switch (result.uint_32) {
 462                 case Opt_discard_sync:
 463                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 464                         btrfs_set_opt(ctx->mount_opt, DISCARD_SYNC);
 465                         break;
 466                 case Opt_discard_async:
 467                         btrfs_clear_opt(ctx->mount_opt, DISCARD_SYNC);
 468                         btrfs_set_opt(ctx->mount_opt, DISCARD_ASYNC);
 469                         break;
 470                 default:
 471                         btrfs_err(NULL, "unrecognized discard mode value %s",
 472                                   param->key);
 473                         return -EINVAL;
 474                 }
 475                 btrfs_clear_opt(ctx->mount_opt, NODISCARD);
 476                 break;
 477         case Opt_space_cache:
 478                 if (result.negated) {
 479                         btrfs_set_opt(ctx->mount_opt, NOSPACECACHE);
 480                         btrfs_clear_opt(ctx->mount_opt, SPACE_CACHE);
 481                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 482                 } else {
 483                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 484                         btrfs_set_opt(ctx->mount_opt, SPACE_CACHE);
 485                 }
 486                 break;
 487         case Opt_space_cache_version:
 488                 switch (result.uint_32) {
 489                 case Opt_space_cache_v1:
 490                         btrfs_set_opt(ctx->mount_opt, SPACE_CACHE);
 491                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 492                         break;
 493                 case Opt_space_cache_v2:
 494                         btrfs_clear_opt(ctx->mount_opt, SPACE_CACHE);
 495                         btrfs_set_opt(ctx->mount_opt, FREE_SPACE_TREE);
 496                         break;
 497                 default:
 498                         btrfs_err(NULL, "unrecognized space_cache value %s",
 499                                   param->key);
 500                         return -EINVAL;
 501                 }
 502                 break;
 503         case Opt_rescan_uuid_tree:
 504                 btrfs_set_opt(ctx->mount_opt, RESCAN_UUID_TREE);
 505                 break;
 506         case Opt_clear_cache:
 507                 btrfs_set_opt(ctx->mount_opt, CLEAR_CACHE);
 508                 break;
 509         case Opt_user_subvol_rm_allowed:
 510                 btrfs_set_opt(ctx->mount_opt, USER_SUBVOL_RM_ALLOWED);
 511                 break;
 512         case Opt_enospc_debug:
 513                 if (result.negated)
 514                         btrfs_clear_opt(ctx->mount_opt, ENOSPC_DEBUG);
 515                 else
 516                         btrfs_set_opt(ctx->mount_opt, ENOSPC_DEBUG);
 517                 break;
 518         case Opt_defrag:
 519                 if (result.negated)
 520                         btrfs_clear_opt(ctx->mount_opt, AUTO_DEFRAG);
 521                 else
 522                         btrfs_set_opt(ctx->mount_opt, AUTO_DEFRAG);
 523                 break;
 524         case Opt_usebackuproot:
 525                 btrfs_warn(NULL,
 526                            "'usebackuproot' is deprecated, use 'rescue=usebackuproot' instead");
 527                 btrfs_set_opt(ctx->mount_opt, USEBACKUPROOT);
 528
 529                 /* If we're loading the backup roots we can't trust the space cache. */
 530                 btrfs_set_opt(ctx->mount_opt, CLEAR_CACHE);
 531                 break;
 532         case Opt_skip_balance:
 533                 btrfs_set_opt(ctx->mount_opt, SKIP_BALANCE);
 534                 break;
 535         case Opt_fatal_errors:
 536                 switch (result.uint_32) {
 537                 case Opt_fatal_errors_panic:
 538                         btrfs_set_opt(ctx->mount_opt, PANIC_ON_FATAL_ERROR);
 539                         break;
 540                 case Opt_fatal_errors_bug:
 541                         btrfs_clear_opt(ctx->mount_opt, PANIC_ON_FATAL_ERROR);
 542                         break;
 543                 default:
 544                         btrfs_err(NULL, "unrecognized fatal_errors value %s",
 545                                   param->key);
 546                         return -EINVAL;
 547                 }
 548                 break;
 549         case Opt_commit_interval:
 550                 ctx->commit_interval = result.uint_32;
 551                 if (ctx->commit_interval == 0)
 552                         ctx->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 553                 break;
 554         case Opt_rescue:
 555                 switch (result.uint_32) {
 556                 case Opt_rescue_usebackuproot:
 557                         btrfs_set_opt(ctx->mount_opt, USEBACKUPROOT);
 558                         break;
 559                 case Opt_rescue_nologreplay:
 560                         btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 561                         break;
 562                 case Opt_rescue_ignorebadroots:
 563                         btrfs_set_opt(ctx->mount_opt, IGNOREBADROOTS);
 564                         break;
 565                 case Opt_rescue_ignoredatacsums:
 566                         btrfs_set_opt(ctx->mount_opt, IGNOREDATACSUMS);
 567                         break;
 568                 case Opt_rescue_parameter_all:
 569                         btrfs_set_opt(ctx->mount_opt, IGNOREDATACSUMS);
 570                         btrfs_set_opt(ctx->mount_opt, IGNOREBADROOTS);
 571                         btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 572                         break;
 573                 default:
 574                         btrfs_info(NULL, "unrecognized rescue option '%s'",
 575                                    param->key);
 576                         return -EINVAL;
 577                 }
 578                 break;
 579 #ifdef CONFIG_BTRFS_DEBUG
 580         case Opt_fragment:
 581                 switch (result.uint_32) {
 582                 case Opt_fragment_parameter_all:
 583                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_DATA);
 584                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_METADATA);
 585                         break;
 586                 case Opt_fragment_parameter_metadata:
 587                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_METADATA);
 588                         break;
 589                 case Opt_fragment_parameter_data:
 590                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_DATA);
 591                         break;
 592                 default:
 593                         btrfs_info(NULL, "unrecognized fragment option '%s'",
 594                                    param->key);
 595                         return -EINVAL;
 596                 }
 597                 break;
 598 #endif
 599 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 600         case Opt_ref_verify:
 601                 btrfs_set_opt(ctx->mount_opt, REF_VERIFY);
 602                 break;
 603 #endif
 604         default:
 605                 btrfs_err(NULL, "unrecognized mount option '%s'", param->key);
 606                 return -EINVAL;
 607         }
 608
 609         return 0;
 610 }
 611
 612 /*
 613  * Some options only have meaning at mount time and shouldn't persist across
 614  * remounts, or be displayed. Clear these at the end of mount and remount code
 615  * paths.
 616  */
 617 static void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
 618 {
 619         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
 620         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
 621         btrfs_clear_opt(fs_info->mount_opt, NOSPACECACHE);
 622 }
 623
 624 static bool check_ro_option(struct btrfs_fs_info *fs_info,
 625                             unsigned long mount_opt, unsigned long opt,
 626                             const char *opt_name)
 627 {
 628         if (mount_opt & opt) {
 629                 btrfs_err(fs_info, "%s must be used with ro mount option",
 630                           opt_name);
 631                 return true;
 632         }
 633         return false;
 634 }
 635
 636 bool btrfs_check_options(struct btrfs_fs_info *info, unsigned long *mount_opt,
 637                          unsigned long flags)
 638 {
 639         bool ret = true;
 640
 641         if (!(flags & SB_RDONLY) &&
 642             (check_ro_option(info, *mount_opt, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
 643              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
 644              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums")))
 645                 ret = false;
 646
 647         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 648             !btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE) &&
 649             !btrfs_raw_test_opt(*mount_opt, CLEAR_CACHE)) {
 650                 btrfs_err(info, "cannot disable free-space-tree");
 651                 ret = false;
 652         }
 653         if (btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE) &&
 654              !btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE)) {
 655                 btrfs_err(info, "cannot disable free-space-tree with block-group-tree feature");
 656                 ret = false;
 657         }
 658
 659         if (btrfs_check_mountopts_zoned(info, mount_opt))
 660                 ret = false;
 661
 662         if (!test_bit(BTRFS_FS_STATE_REMOUNTING, &info->fs_state)) {
 663                 if (btrfs_raw_test_opt(*mount_opt, SPACE_CACHE))
 664                         btrfs_info(info, "disk space caching is enabled");
 665                 if (btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE))
 666                         btrfs_info(info, "using free-space-tree");
 667         }
 668
 669         return ret;
 670 }
 671
 672 /*
 673  * This is subtle, we only call this during open_ctree().  We need to pre-load
 674  * the mount options with the on-disk settings.  Before the new mount API took
 675  * effect we would do this on mount and remount.  With the new mount API we'll
 676  * only do this on the initial mount.
 677  *
 678  * This isn't a change in behavior, because we're using the current state of the
 679  * file system to set the current mount options.  If you mounted with special
 680  * options to disable these features and then remounted we wouldn't revert the
 681  * settings, because mounting without these features cleared the on-disk
 682  * settings, so this being called on re-mount is not needed.
 683  */
 684 void btrfs_set_free_space_cache_settings(struct btrfs_fs_info *fs_info)
 685 {
 686         if (fs_info->sectorsize < PAGE_SIZE) {
 687                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
 688                 if (!btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
 689                         btrfs_info(fs_info,
 690                                    "forcing free space tree for sector size %u with page size %lu",
 691                                    fs_info->sectorsize, PAGE_SIZE);
 692                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
 693                 }
 694         }
 695
 696         /*
 697          * At this point our mount options are populated, so we only mess with
 698          * these settings if we don't have any settings already.
 699          */
 700         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
 701                 return;
 702
 703         if (btrfs_is_zoned(fs_info) &&
 704             btrfs_free_space_cache_v1_active(fs_info)) {
 705                 btrfs_info(fs_info, "zoned: clearing existing space cache");
 706                 btrfs_set_super_cache_generation(fs_info->super_copy, 0);
 707                 return;
 708         }
 709
 710         if (btrfs_test_opt(fs_info, SPACE_CACHE))
 711                 return;
 712
 713         if (btrfs_test_opt(fs_info, NOSPACECACHE))
 714                 return;
 715
 716         /*
 717          * At this point we don't have explicit options set by the user, set
 718          * them ourselves based on the state of the file system.
 719          */
 720         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
 721                 btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
 722         else if (btrfs_free_space_cache_v1_active(fs_info))
 723                 btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
 724 }
 725
 726 static void set_device_specific_options(struct btrfs_fs_info *fs_info)
 727 {
 728         if (!btrfs_test_opt(fs_info, NOSSD) &&
 729             !fs_info->fs_devices->rotating)
 730                 btrfs_set_opt(fs_info->mount_opt, SSD);
 731
 732         /*
 733          * For devices supporting discard turn on discard=async automatically,
 734          * unless it's already set or disabled. This could be turned off by
 735          * nodiscard for the same mount.
 736          *
 737          * The zoned mode piggy backs on the discard functionality for
 738          * resetting a zone. There is no reason to delay the zone reset as it is
 739          * fast enough. So, do not enable async discard for zoned mode.
 740          */
 741         if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
 742               btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
 743               btrfs_test_opt(fs_info, NODISCARD)) &&
 744             fs_info->fs_devices->discardable &&
 745             !btrfs_is_zoned(fs_info))
 746                 btrfs_set_opt(fs_info->mount_opt, DISCARD_ASYNC);
 747 }
 748
 749 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 750                                           u64 subvol_objectid)
 751 {
 752         struct btrfs_root *root = fs_info->tree_root;
 753         struct btrfs_root *fs_root = NULL;
 754         struct btrfs_root_ref *root_ref;
 755         struct btrfs_inode_ref *inode_ref;
 756         struct btrfs_key key;
 757         struct btrfs_path *path = NULL;
 758         char *name = NULL, *ptr;
 759         u64 dirid;
 760         int len;
 761         int ret;
 762
 763         path = btrfs_alloc_path();
 764         if (!path) {
 765                 ret = -ENOMEM;
 766                 goto err;
 767         }
 768
 769         name = kmalloc(PATH_MAX, GFP_KERNEL);
 770         if (!name) {
 771                 ret = -ENOMEM;
 772                 goto err;
 773         }
 774         ptr = name + PATH_MAX - 1;
 775         ptr[0] = '\0';
 776
 777         /*
 778          * Walk up the subvolume trees in the tree of tree roots by root
 779          * backrefs until we hit the top-level subvolume.
 780          */
 781         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 782                 key.objectid = subvol_objectid;
 783                 key.type = BTRFS_ROOT_BACKREF_KEY;
 784                 key.offset = (u64)-1;
 785
 786                 ret = btrfs_search_backwards(root, &key, path);
 787                 if (ret < 0) {
 788                         goto err;
 789                 } else if (ret > 0) {
 790                         ret = -ENOENT;
 791                         goto err;
 792                 }
 793
 794                 subvol_objectid = key.offset;
 795
 796                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 797                                           struct btrfs_root_ref);
 798                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 799                 ptr -= len + 1;
 800                 if (ptr < name) {
 801                         ret = -ENAMETOOLONG;
 802                         goto err;
 803                 }
 804                 read_extent_buffer(path->nodes[0], ptr + 1,
 805                                    (unsigned long)(root_ref + 1), len);
 806                 ptr[0] = '/';
 807                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 808                 btrfs_release_path(path);
 809
 810                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
 811                 if (IS_ERR(fs_root)) {
 812                         ret = PTR_ERR(fs_root);
 813                         fs_root = NULL;
 814                         goto err;
 815                 }
 816
 817                 /*
 818                  * Walk up the filesystem tree by inode refs until we hit the
 819                  * root directory.
 820                  */
 821                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
 822                         key.objectid = dirid;
 823                         key.type = BTRFS_INODE_REF_KEY;
 824                         key.offset = (u64)-1;
 825
 826                         ret = btrfs_search_backwards(fs_root, &key, path);
 827                         if (ret < 0) {
 828                                 goto err;
 829                         } else if (ret > 0) {
 830                                 ret = -ENOENT;
 831                                 goto err;
 832                         }
 833
 834                         dirid = key.offset;
 835
 836                         inode_ref = btrfs_item_ptr(path->nodes[0],
 837                                                    path->slots[0],
 838                                                    struct btrfs_inode_ref);
 839                         len = btrfs_inode_ref_name_len(path->nodes[0],
 840                                                        inode_ref);
 841                         ptr -= len + 1;
 842                         if (ptr < name) {
 843                                 ret = -ENAMETOOLONG;
 844                                 goto err;
 845                         }
 846                         read_extent_buffer(path->nodes[0], ptr + 1,
 847                                            (unsigned long)(inode_ref + 1), len);
 848                         ptr[0] = '/';
 849                         btrfs_release_path(path);
 850                 }
 851                 btrfs_put_root(fs_root);
 852                 fs_root = NULL;
 853         }
 854
 855         btrfs_free_path(path);
 856         if (ptr == name + PATH_MAX - 1) {
 857                 name[0] = '/';
 858                 name[1] = '\0';
 859         } else {
 860                 memmove(name, ptr, name + PATH_MAX - ptr);
 861         }
 862         return name;
 863
 864 err:
 865         btrfs_put_root(fs_root);
 866         btrfs_free_path(path);
 867         kfree(name);
 868         return ERR_PTR(ret);
 869 }
 870
 871 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
 872 {
 873         struct btrfs_root *root = fs_info->tree_root;
 874         struct btrfs_dir_item *di;
 875         struct btrfs_path *path;
 876         struct btrfs_key location;
 877         struct fscrypt_str name = FSTR_INIT("default", 7);
 878         u64 dir_id;
 879
 880         path = btrfs_alloc_path();
 881         if (!path)
 882                 return -ENOMEM;
 883
 884         /*
 885          * Find the "default" dir item which points to the root item that we
 886          * will mount by default if we haven't been given a specific subvolume
 887          * to mount.
 888          */
 889         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 890         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, &name, 0);
 891         if (IS_ERR(di)) {
 892                 btrfs_free_path(path);
 893                 return PTR_ERR(di);
 894         }
 895         if (!di) {
 896                 /*
 897                  * Ok the default dir item isn't there.  This is weird since
 898                  * it's always been there, but don't freak out, just try and
 899                  * mount the top-level subvolume.
 900                  */
 901                 btrfs_free_path(path);
 902                 *objectid = BTRFS_FS_TREE_OBJECTID;
 903                 return 0;
 904         }
 905
 906         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 907         btrfs_free_path(path);
 908         *objectid = location.objectid;
 909         return 0;
 910 }
 911
 912 static int btrfs_fill_super(struct super_block *sb,
 913                             struct btrfs_fs_devices *fs_devices,
 914                             void *data)
 915 {
 916         struct inode *inode;
 917         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 918         int err;
 919
 920         sb->s_maxbytes = MAX_LFS_FILESIZE;
 921         sb->s_magic = BTRFS_SUPER_MAGIC;
 922         sb->s_op = &btrfs_super_ops;
 923         sb->s_d_op = &btrfs_dentry_operations;
 924         sb->s_export_op = &btrfs_export_ops;
 925 #ifdef CONFIG_FS_VERITY
 926         sb->s_vop = &btrfs_verityops;
 927 #endif
 928         sb->s_xattr = btrfs_xattr_handlers;
 929         sb->s_time_gran = 1;
 930         sb->s_iflags |= SB_I_CGROUPWB;
 931
 932         err = super_setup_bdi(sb);
 933         if (err) {
 934                 btrfs_err(fs_info, "super_setup_bdi failed");
 935                 return err;
 936         }
 937
 938         err = open_ctree(sb, fs_devices, (char *)data);
 939         if (err) {
 940                 btrfs_err(fs_info, "open_ctree failed");
 941                 return err;
 942         }
 943
 944         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
 945         if (IS_ERR(inode)) {
 946                 err = PTR_ERR(inode);
 947                 btrfs_handle_fs_error(fs_info, err, NULL);
 948                 goto fail_close;
 949         }
 950
 951         sb->s_root = d_make_root(inode);
 952         if (!sb->s_root) {
 953                 err = -ENOMEM;
 954                 goto fail_close;
 955         }
 956
 957         sb->s_flags |= SB_ACTIVE;
 958         return 0;
 959
 960 fail_close:
 961         close_ctree(fs_info);
 962         return err;
 963 }
 964
 965 int btrfs_sync_fs(struct super_block *sb, int wait)
 966 {
 967         struct btrfs_trans_handle *trans;
 968         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 969         struct btrfs_root *root = fs_info->tree_root;
 970
 971         trace_btrfs_sync_fs(fs_info, wait);
 972
 973         if (!wait) {
 974                 filemap_flush(fs_info->btree_inode->i_mapping);
 975                 return 0;
 976         }
 977
 978         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 979
 980         trans = btrfs_attach_transaction_barrier(root);
 981         if (IS_ERR(trans)) {
 982                 /* no transaction, don't bother */
 983                 if (PTR_ERR(trans) == -ENOENT) {
 984                         /*
 985                          * Exit unless we have some pending changes
 986                          * that need to go through commit
 987                          */
 988                         if (!test_bit(BTRFS_FS_NEED_TRANS_COMMIT,
 989                                       &fs_info->flags))
 990                                 return 0;
 991                         /*
 992                          * A non-blocking test if the fs is frozen. We must not
 993                          * start a new transaction here otherwise a deadlock
 994                          * happens. The pending operations are delayed to the
 995                          * next commit after thawing.
 996                          */
 997                         if (sb_start_write_trylock(sb))
 998                                 sb_end_write(sb);
 999                         else
1000                                 return 0;
1001                         trans = btrfs_start_transaction(root, 0);
1002                 }
1003                 if (IS_ERR(trans))
1004                         return PTR_ERR(trans);
1005         }
1006         return btrfs_commit_transaction(trans);
1007 }
1008
1009 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1010 {
1011         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1012         *printed = true;
1013 }
1014
1015 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1016 {
1017         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1018         const char *compress_type;
1019         const char *subvol_name;
1020         bool printed = false;
1021
1022         if (btrfs_test_opt(info, DEGRADED))
1023                 seq_puts(seq, ",degraded");
1024         if (btrfs_test_opt(info, NODATASUM))
1025                 seq_puts(seq, ",nodatasum");
1026         if (btrfs_test_opt(info, NODATACOW))
1027                 seq_puts(seq, ",nodatacow");
1028         if (btrfs_test_opt(info, NOBARRIER))
1029                 seq_puts(seq, ",nobarrier");
1030         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1031                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1032         if (info->thread_pool_size !=  min_t(unsigned long,
1033                                              num_online_cpus() + 2, 8))
1034                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1035         if (btrfs_test_opt(info, COMPRESS)) {
1036                 compress_type = btrfs_compress_type2str(info->compress_type);
1037                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1038                         seq_printf(seq, ",compress-force=%s", compress_type);
1039                 else
1040                         seq_printf(seq, ",compress=%s", compress_type);
1041                 if (info->compress_level)
1042                         seq_printf(seq, ":%d", info->compress_level);
1043         }
1044         if (btrfs_test_opt(info, NOSSD))
1045                 seq_puts(seq, ",nossd");
1046         if (btrfs_test_opt(info, SSD_SPREAD))
1047                 seq_puts(seq, ",ssd_spread");
1048         else if (btrfs_test_opt(info, SSD))
1049                 seq_puts(seq, ",ssd");
1050         if (btrfs_test_opt(info, NOTREELOG))
1051                 seq_puts(seq, ",notreelog");
1052         if (btrfs_test_opt(info, NOLOGREPLAY))
1053                 print_rescue_option(seq, "nologreplay", &printed);
1054         if (btrfs_test_opt(info, USEBACKUPROOT))
1055                 print_rescue_option(seq, "usebackuproot", &printed);
1056         if (btrfs_test_opt(info, IGNOREBADROOTS))
1057                 print_rescue_option(seq, "ignorebadroots", &printed);
1058         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1059                 print_rescue_option(seq, "ignoredatacsums", &printed);
1060         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1061                 seq_puts(seq, ",flushoncommit");
1062         if (btrfs_test_opt(info, DISCARD_SYNC))
1063                 seq_puts(seq, ",discard");
1064         if (btrfs_test_opt(info, DISCARD_ASYNC))
1065                 seq_puts(seq, ",discard=async");
1066         if (!(info->sb->s_flags & SB_POSIXACL))
1067                 seq_puts(seq, ",noacl");
1068         if (btrfs_free_space_cache_v1_active(info))
1069                 seq_puts(seq, ",space_cache");
1070         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1071                 seq_puts(seq, ",space_cache=v2");
1072         else
1073                 seq_puts(seq, ",nospace_cache");
1074         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1075                 seq_puts(seq, ",rescan_uuid_tree");
1076         if (btrfs_test_opt(info, CLEAR_CACHE))
1077                 seq_puts(seq, ",clear_cache");
1078         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1079                 seq_puts(seq, ",user_subvol_rm_allowed");
1080         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1081                 seq_puts(seq, ",enospc_debug");
1082         if (btrfs_test_opt(info, AUTO_DEFRAG))
1083                 seq_puts(seq, ",autodefrag");
1084         if (btrfs_test_opt(info, SKIP_BALANCE))
1085                 seq_puts(seq, ",skip_balance");
1086         if (info->metadata_ratio)
1087                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1088         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1089                 seq_puts(seq, ",fatal_errors=panic");
1090         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1091                 seq_printf(seq, ",commit=%u", info->commit_interval);
1092 #ifdef CONFIG_BTRFS_DEBUG
1093         if (btrfs_test_opt(info, FRAGMENT_DATA))
1094                 seq_puts(seq, ",fragment=data");
1095         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1096                 seq_puts(seq, ",fragment=metadata");
1097 #endif
1098         if (btrfs_test_opt(info, REF_VERIFY))
1099                 seq_puts(seq, ",ref_verify");
1100         seq_printf(seq, ",subvolid=%llu", btrfs_root_id(BTRFS_I(d_inode(dentry))->root));
1101         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1102                         btrfs_root_id(BTRFS_I(d_inode(dentry))->root));
1103         if (!IS_ERR(subvol_name)) {
1104                 seq_puts(seq, ",subvol=");
1105                 seq_escape(seq, subvol_name, " \t\n\\");
1106                 kfree(subvol_name);
1107         }
1108         return 0;
1109 }
1110
1111 /*
1112  * subvolumes are identified by ino 256
1113  */
1114 static inline int is_subvolume_inode(struct inode *inode)
1115 {
1116         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1117                 return 1;
1118         return 0;
1119 }
1120
1121 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1122                                    struct vfsmount *mnt)
1123 {
1124         struct dentry *root;
1125         int ret;
1126
1127         if (!subvol_name) {
1128                 if (!subvol_objectid) {
1129                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1130                                                           &subvol_objectid);
1131                         if (ret) {
1132                                 root = ERR_PTR(ret);
1133                                 goto out;
1134                         }
1135                 }
1136                 subvol_name = btrfs_get_subvol_name_from_objectid(
1137                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1138                 if (IS_ERR(subvol_name)) {
1139                         root = ERR_CAST(subvol_name);
1140                         subvol_name = NULL;
1141                         goto out;
1142                 }
1143
1144         }
1145
1146         root = mount_subtree(mnt, subvol_name);
1147         /* mount_subtree() drops our reference on the vfsmount. */
1148         mnt = NULL;
1149
1150         if (!IS_ERR(root)) {
1151                 struct super_block *s = root->d_sb;
1152                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1153                 struct inode *root_inode = d_inode(root);
1154                 u64 root_objectid = btrfs_root_id(BTRFS_I(root_inode)->root);
1155
1156                 ret = 0;
1157                 if (!is_subvolume_inode(root_inode)) {
1158                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1159                                subvol_name);
1160                         ret = -EINVAL;
1161                 }
1162                 if (subvol_objectid && root_objectid != subvol_objectid) {
1163                         /*
1164                          * This will also catch a race condition where a
1165                          * subvolume which was passed by ID is renamed and
1166                          * another subvolume is renamed over the old location.
1167                          */
1168                         btrfs_err(fs_info,
1169                                   "subvol '%s' does not match subvolid %llu",
1170                                   subvol_name, subvol_objectid);
1171                         ret = -EINVAL;
1172                 }
1173                 if (ret) {
1174                         dput(root);
1175                         root = ERR_PTR(ret);
1176                         deactivate_locked_super(s);
1177                 }
1178         }
1179
1180 out:
1181         mntput(mnt);
1182         kfree(subvol_name);
1183         return root;
1184 }
1185
1186 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1187                                      u32 new_pool_size, u32 old_pool_size)
1188 {
1189         if (new_pool_size == old_pool_size)
1190                 return;
1191
1192         fs_info->thread_pool_size = new_pool_size;
1193
1194         btrfs_info(fs_info, "resize thread pool %d -> %d",
1195                old_pool_size, new_pool_size);
1196
1197         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1198         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1199         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1200         workqueue_set_max_active(fs_info->endio_workers, new_pool_size);
1201         workqueue_set_max_active(fs_info->endio_meta_workers, new_pool_size);
1202         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1203         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1204         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1205 }
1206
1207 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1208                                        unsigned long old_opts, int flags)
1209 {
1210         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1211             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1212              (flags & SB_RDONLY))) {
1213                 /* wait for any defraggers to finish */
1214                 wait_event(fs_info->transaction_wait,
1215                            (atomic_read(&fs_info->defrag_running) == 0));
1216                 if (flags & SB_RDONLY)
1217                         sync_filesystem(fs_info->sb);
1218         }
1219 }
1220
1221 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1222                                          unsigned long old_opts)
1223 {
1224         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1225
1226         /*
1227          * We need to cleanup all defragable inodes if the autodefragment is
1228          * close or the filesystem is read only.
1229          */
1230         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1231             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1232                 btrfs_cleanup_defrag_inodes(fs_info);
1233         }
1234
1235         /* If we toggled discard async */
1236         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1237             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1238                 btrfs_discard_resume(fs_info);
1239         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1240                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1241                 btrfs_discard_cleanup(fs_info);
1242
1243         /* If we toggled space cache */
1244         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1245                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1246 }
1247
1248 static int btrfs_remount_rw(struct btrfs_fs_info *fs_info)
1249 {
1250         int ret;
1251
1252         if (BTRFS_FS_ERROR(fs_info)) {
1253                 btrfs_err(fs_info,
1254                           "remounting read-write after error is not allowed");
1255                 return -EINVAL;
1256         }
1257
1258         if (fs_info->fs_devices->rw_devices == 0)
1259                 return -EACCES;
1260
1261         if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1262                 btrfs_warn(fs_info,
1263                            "too many missing devices, writable remount is not allowed");
1264                 return -EACCES;
1265         }
1266
1267         if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1268                 btrfs_warn(fs_info,
1269                            "mount required to replay tree-log, cannot remount read-write");
1270                 return -EINVAL;
1271         }
1272
1273         /*
1274          * NOTE: when remounting with a change that does writes, don't put it
1275          * anywhere above this point, as we are not sure to be safe to write
1276          * until we pass the above checks.
1277          */
1278         ret = btrfs_start_pre_rw_mount(fs_info);
1279         if (ret)
1280                 return ret;
1281
1282         btrfs_clear_sb_rdonly(fs_info->sb);
1283
1284         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1285
1286         /*
1287          * If we've gone from readonly -> read-write, we need to get our
1288          * sync/async discard lists in the right state.
1289          */
1290         btrfs_discard_resume(fs_info);
1291
1292         return 0;
1293 }
1294
1295 static int btrfs_remount_ro(struct btrfs_fs_info *fs_info)
1296 {
1297         /*
1298          * This also happens on 'umount -rf' or on shutdown, when the
1299          * filesystem is busy.
1300          */
1301         cancel_work_sync(&fs_info->async_reclaim_work);
1302         cancel_work_sync(&fs_info->async_data_reclaim_work);
1303
1304         btrfs_discard_cleanup(fs_info);
1305
1306         /* Wait for the uuid_scan task to finish */
1307         down(&fs_info->uuid_tree_rescan_sem);
1308         /* Avoid complains from lockdep et al. */
1309         up(&fs_info->uuid_tree_rescan_sem);
1310
1311         btrfs_set_sb_rdonly(fs_info->sb);
1312
1313         /*
1314          * Setting SB_RDONLY will put the cleaner thread to sleep at the next
1315          * loop if it's already active.  If it's already asleep, we'll leave
1316          * unused block groups on disk until we're mounted read-write again
1317          * unless we clean them up here.
1318          */
1319         btrfs_delete_unused_bgs(fs_info);
1320
1321         /*
1322          * The cleaner task could be already running before we set the flag
1323          * BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).  We must make
1324          * sure that after we finish the remount, i.e. after we call
1325          * btrfs_commit_super(), the cleaner can no longer start a transaction
1326          * - either because it was dropping a dead root, running delayed iputs
1327          *   or deleting an unused block group (the cleaner picked a block
1328          *   group from the list of unused block groups before we were able to
1329          *   in the previous call to btrfs_delete_unused_bgs()).
1330          */
1331         wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING, TASK_UNINTERRUPTIBLE);
1332
1333         /*
1334          * We've set the superblock to RO mode, so we might have made the
1335          * cleaner task sleep without running all pending delayed iputs. Go
1336          * through all the delayed iputs here, so that if an unmount happens
1337          * without remounting RW we don't end up at finishing close_ctree()
1338          * with a non-empty list of delayed iputs.
1339          */
1340         btrfs_run_delayed_iputs(fs_info);
1341
1342         btrfs_dev_replace_suspend_for_unmount(fs_info);
1343         btrfs_scrub_cancel(fs_info);
1344         btrfs_pause_balance(fs_info);
1345
1346         /*
1347          * Pause the qgroup rescan worker if it is running. We don't want it to
1348          * be still running after we are in RO mode, as after that, by the time
1349          * we unmount, it might have left a transaction open, so we would leak
1350          * the transaction and/or crash.
1351          */
1352         btrfs_qgroup_wait_for_completion(fs_info, false);
1353
1354         return btrfs_commit_super(fs_info);
1355 }
1356
1357 static void btrfs_ctx_to_info(struct btrfs_fs_info *fs_info, struct btrfs_fs_context *ctx)
1358 {
1359         fs_info->max_inline = ctx->max_inline;
1360         fs_info->commit_interval = ctx->commit_interval;
1361         fs_info->metadata_ratio = ctx->metadata_ratio;
1362         fs_info->thread_pool_size = ctx->thread_pool_size;
1363         fs_info->mount_opt = ctx->mount_opt;
1364         fs_info->compress_type = ctx->compress_type;
1365         fs_info->compress_level = ctx->compress_level;
1366 }
1367
1368 static void btrfs_info_to_ctx(struct btrfs_fs_info *fs_info, struct btrfs_fs_context *ctx)
1369 {
1370         ctx->max_inline = fs_info->max_inline;
1371         ctx->commit_interval = fs_info->commit_interval;
1372         ctx->metadata_ratio = fs_info->metadata_ratio;
1373         ctx->thread_pool_size = fs_info->thread_pool_size;
1374         ctx->mount_opt = fs_info->mount_opt;
1375         ctx->compress_type = fs_info->compress_type;
1376         ctx->compress_level = fs_info->compress_level;
1377 }
1378
1379 #define btrfs_info_if_set(fs_info, old_ctx, opt, fmt, args...)                  \
1380 do {                                                                            \
1381         if ((!old_ctx || !btrfs_raw_test_opt(old_ctx->mount_opt, opt)) &&       \
1382             btrfs_raw_test_opt(fs_info->mount_opt, opt))                        \
1383                 btrfs_info(fs_info, fmt, ##args);                               \
1384 } while (0)
1385
1386 #define btrfs_info_if_unset(fs_info, old_ctx, opt, fmt, args...)        \
1387 do {                                                                    \
1388         if ((old_ctx && btrfs_raw_test_opt(old_ctx->mount_opt, opt)) && \
1389             !btrfs_raw_test_opt(fs_info->mount_opt, opt))               \
1390                 btrfs_info(fs_info, fmt, ##args);                       \
1391 } while (0)
1392
1393 static void btrfs_emit_options(struct btrfs_fs_info *info,
1394                                struct btrfs_fs_context *old)
1395 {
1396         btrfs_info_if_set(info, old, NODATASUM, "setting nodatasum");
1397         btrfs_info_if_set(info, old, DEGRADED, "allowing degraded mounts");
1398         btrfs_info_if_set(info, old, NODATASUM, "setting nodatasum");
1399         btrfs_info_if_set(info, old, SSD, "enabling ssd optimizations");
1400         btrfs_info_if_set(info, old, SSD_SPREAD, "using spread ssd allocation scheme");
1401         btrfs_info_if_set(info, old, NOBARRIER, "turning off barriers");
1402         btrfs_info_if_set(info, old, NOTREELOG, "disabling tree log");
1403         btrfs_info_if_set(info, old, NOLOGREPLAY, "disabling log replay at mount time");
1404         btrfs_info_if_set(info, old, FLUSHONCOMMIT, "turning on flush-on-commit");
1405         btrfs_info_if_set(info, old, DISCARD_SYNC, "turning on sync discard");
1406         btrfs_info_if_set(info, old, DISCARD_ASYNC, "turning on async discard");
1407         btrfs_info_if_set(info, old, FREE_SPACE_TREE, "enabling free space tree");
1408         btrfs_info_if_set(info, old, SPACE_CACHE, "enabling disk space caching");
1409         btrfs_info_if_set(info, old, CLEAR_CACHE, "force clearing of disk cache");
1410         btrfs_info_if_set(info, old, AUTO_DEFRAG, "enabling auto defrag");
1411         btrfs_info_if_set(info, old, FRAGMENT_DATA, "fragmenting data");
1412         btrfs_info_if_set(info, old, FRAGMENT_METADATA, "fragmenting metadata");
1413         btrfs_info_if_set(info, old, REF_VERIFY, "doing ref verification");
1414         btrfs_info_if_set(info, old, USEBACKUPROOT, "trying to use backup root at mount time");
1415         btrfs_info_if_set(info, old, IGNOREBADROOTS, "ignoring bad roots");
1416         btrfs_info_if_set(info, old, IGNOREDATACSUMS, "ignoring data csums");
1417
1418         btrfs_info_if_unset(info, old, NODATACOW, "setting datacow");
1419         btrfs_info_if_unset(info, old, SSD, "not using ssd optimizations");
1420         btrfs_info_if_unset(info, old, SSD_SPREAD, "not using spread ssd allocation scheme");
1421         btrfs_info_if_unset(info, old, NOBARRIER, "turning off barriers");
1422         btrfs_info_if_unset(info, old, NOTREELOG, "enabling tree log");
1423         btrfs_info_if_unset(info, old, SPACE_CACHE, "disabling disk space caching");
1424         btrfs_info_if_unset(info, old, FREE_SPACE_TREE, "disabling free space tree");
1425         btrfs_info_if_unset(info, old, AUTO_DEFRAG, "disabling auto defrag");
1426         btrfs_info_if_unset(info, old, COMPRESS, "use no compression");
1427
1428         /* Did the compression settings change? */
1429         if (btrfs_test_opt(info, COMPRESS) &&
1430             (!old ||
1431              old->compress_type != info->compress_type ||
1432              old->compress_level != info->compress_level ||
1433              (!btrfs_raw_test_opt(old->mount_opt, FORCE_COMPRESS) &&
1434               btrfs_raw_test_opt(info->mount_opt, FORCE_COMPRESS)))) {
1435                 const char *compress_type = btrfs_compress_type2str(info->compress_type);
1436
1437                 btrfs_info(info, "%s %s compression, level %d",
1438                            btrfs_test_opt(info, FORCE_COMPRESS) ? "force" : "use",
1439                            compress_type, info->compress_level);
1440         }
1441
1442         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1443                 btrfs_info(info, "max_inline set to %llu", info->max_inline);
1444 }
1445
1446 static int btrfs_reconfigure(struct fs_context *fc)
1447 {
1448         struct super_block *sb = fc->root->d_sb;
1449         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1450         struct btrfs_fs_context *ctx = fc->fs_private;
1451         struct btrfs_fs_context old_ctx;
1452         int ret = 0;
1453         bool mount_reconfigure = (fc->s_fs_info != NULL);
1454
1455         btrfs_info_to_ctx(fs_info, &old_ctx);
1456
1457         /*
1458          * This is our "bind mount" trick, we don't want to allow the user to do
1459          * anything other than mount a different ro/rw and a different subvol,
1460          * all of the mount options should be maintained.
1461          */
1462         if (mount_reconfigure)
1463                 ctx->mount_opt = old_ctx.mount_opt;
1464
1465         sync_filesystem(sb);
1466         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1467
1468         if (!mount_reconfigure &&
1469             !btrfs_check_options(fs_info, &ctx->mount_opt, fc->sb_flags))
1470                 return -EINVAL;
1471
1472         ret = btrfs_check_features(fs_info, !(fc->sb_flags & SB_RDONLY));
1473         if (ret < 0)
1474                 return ret;
1475
1476         btrfs_ctx_to_info(fs_info, ctx);
1477         btrfs_remount_begin(fs_info, old_ctx.mount_opt, fc->sb_flags);
1478         btrfs_resize_thread_pool(fs_info, fs_info->thread_pool_size,
1479                                  old_ctx.thread_pool_size);
1480
1481         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
1482             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
1483             (!sb_rdonly(sb) || (fc->sb_flags & SB_RDONLY))) {
1484                 btrfs_warn(fs_info,
1485                 "remount supports changing free space tree only from RO to RW");
1486                 /* Make sure free space cache options match the state on disk. */
1487                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
1488                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1489                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
1490                 }
1491                 if (btrfs_free_space_cache_v1_active(fs_info)) {
1492                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1493                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
1494                 }
1495         }
1496
1497         ret = 0;
1498         if (!sb_rdonly(sb) && (fc->sb_flags & SB_RDONLY))
1499                 ret = btrfs_remount_ro(fs_info);
1500         else if (sb_rdonly(sb) && !(fc->sb_flags & SB_RDONLY))
1501                 ret = btrfs_remount_rw(fs_info);
1502         if (ret)
1503                 goto restore;
1504
1505         /*
1506          * If we set the mask during the parameter parsing VFS would reject the
1507          * remount.  Here we can set the mask and the value will be updated
1508          * appropriately.
1509          */
1510         if ((fc->sb_flags & SB_POSIXACL) != (sb->s_flags & SB_POSIXACL))
1511                 fc->sb_flags_mask |= SB_POSIXACL;
1512
1513         btrfs_emit_options(fs_info, &old_ctx);
1514         wake_up_process(fs_info->transaction_kthread);
1515         btrfs_remount_cleanup(fs_info, old_ctx.mount_opt);
1516         btrfs_clear_oneshot_options(fs_info);
1517         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1518
1519         return 0;
1520 restore:
1521         btrfs_ctx_to_info(fs_info, &old_ctx);
1522         btrfs_remount_cleanup(fs_info, old_ctx.mount_opt);
1523         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1524         return ret;
1525 }
1526
1527 /* Used to sort the devices by max_avail(descending sort) */
1528 static int btrfs_cmp_device_free_bytes(const void *a, const void *b)
1529 {
1530         const struct btrfs_device_info *dev_info1 = a;
1531         const struct btrfs_device_info *dev_info2 = b;
1532
1533         if (dev_info1->max_avail > dev_info2->max_avail)
1534                 return -1;
1535         else if (dev_info1->max_avail < dev_info2->max_avail)
1536                 return 1;
1537         return 0;
1538 }
1539
1540 /*
1541  * sort the devices by max_avail, in which max free extent size of each device
1542  * is stored.(Descending Sort)
1543  */
1544 static inline void btrfs_descending_sort_devices(
1545                                         struct btrfs_device_info *devices,
1546                                         size_t nr_devices)
1547 {
1548         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1549              btrfs_cmp_device_free_bytes, NULL);
1550 }
1551
1552 /*
1553  * The helper to calc the free space on the devices that can be used to store
1554  * file data.
1555  */
1556 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1557                                               u64 *free_bytes)
1558 {
1559         struct btrfs_device_info *devices_info;
1560         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1561         struct btrfs_device *device;
1562         u64 type;
1563         u64 avail_space;
1564         u64 min_stripe_size;
1565         int num_stripes = 1;
1566         int i = 0, nr_devices;
1567         const struct btrfs_raid_attr *rattr;
1568
1569         /*
1570          * We aren't under the device list lock, so this is racy-ish, but good
1571          * enough for our purposes.
1572          */
1573         nr_devices = fs_info->fs_devices->open_devices;
1574         if (!nr_devices) {
1575                 smp_mb();
1576                 nr_devices = fs_info->fs_devices->open_devices;
1577                 ASSERT(nr_devices);
1578                 if (!nr_devices) {
1579                         *free_bytes = 0;
1580                         return 0;
1581                 }
1582         }
1583
1584         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1585                                GFP_KERNEL);
1586         if (!devices_info)
1587                 return -ENOMEM;
1588
1589         /* calc min stripe number for data space allocation */
1590         type = btrfs_data_alloc_profile(fs_info);
1591         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
1592
1593         if (type & BTRFS_BLOCK_GROUP_RAID0)
1594                 num_stripes = nr_devices;
1595         else if (type & BTRFS_BLOCK_GROUP_RAID1_MASK)
1596                 num_stripes = rattr->ncopies;
1597         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1598                 num_stripes = 4;
1599
1600         /* Adjust for more than 1 stripe per device */
1601         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
1602
1603         rcu_read_lock();
1604         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1605                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1606                                                 &device->dev_state) ||
1607                     !device->bdev ||
1608                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1609                         continue;
1610
1611                 if (i >= nr_devices)
1612                         break;
1613
1614                 avail_space = device->total_bytes - device->bytes_used;
1615
1616                 /* align with stripe_len */
1617                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
1618
1619                 /*
1620                  * Ensure we have at least min_stripe_size on top of the
1621                  * reserved space on the device.
1622                  */
1623                 if (avail_space <= BTRFS_DEVICE_RANGE_RESERVED + min_stripe_size)
1624                         continue;
1625
1626                 avail_space -= BTRFS_DEVICE_RANGE_RESERVED;
1627
1628                 devices_info[i].dev = device;
1629                 devices_info[i].max_avail = avail_space;
1630
1631                 i++;
1632         }
1633         rcu_read_unlock();
1634
1635         nr_devices = i;
1636
1637         btrfs_descending_sort_devices(devices_info, nr_devices);
1638
1639         i = nr_devices - 1;
1640         avail_space = 0;
1641         while (nr_devices >= rattr->devs_min) {
1642                 num_stripes = min(num_stripes, nr_devices);
1643
1644                 if (devices_info[i].max_avail >= min_stripe_size) {
1645                         int j;
1646                         u64 alloc_size;
1647
1648                         avail_space += devices_info[i].max_avail * num_stripes;
1649                         alloc_size = devices_info[i].max_avail;
1650                         for (j = i + 1 - num_stripes; j <= i; j++)
1651                                 devices_info[j].max_avail -= alloc_size;
1652                 }
1653                 i--;
1654                 nr_devices--;
1655         }
1656
1657         kfree(devices_info);
1658         *free_bytes = avail_space;
1659         return 0;
1660 }
1661
1662 /*
1663  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
1664  *
1665  * If there's a redundant raid level at DATA block groups, use the respective
1666  * multiplier to scale the sizes.
1667  *
1668  * Unused device space usage is based on simulating the chunk allocator
1669  * algorithm that respects the device sizes and order of allocations.  This is
1670  * a close approximation of the actual use but there are other factors that may
1671  * change the result (like a new metadata chunk).
1672  *
1673  * If metadata is exhausted, f_bavail will be 0.
1674  */
1675 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1676 {
1677         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1678         struct btrfs_super_block *disk_super = fs_info->super_copy;
1679         struct btrfs_space_info *found;
1680         u64 total_used = 0;
1681         u64 total_free_data = 0;
1682         u64 total_free_meta = 0;
1683         u32 bits = fs_info->sectorsize_bits;
1684         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
1685         unsigned factor = 1;
1686         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
1687         int ret;
1688         u64 thresh = 0;
1689         int mixed = 0;
1690
1691         list_for_each_entry(found, &fs_info->space_info, list) {
1692                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1693                         int i;
1694
1695                         total_free_data += found->disk_total - found->disk_used;
1696                         total_free_data -=
1697                                 btrfs_account_ro_block_groups_free_space(found);
1698
1699                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1700                                 if (!list_empty(&found->block_groups[i]))
1701                                         factor = btrfs_bg_type_to_factor(
1702                                                 btrfs_raid_array[i].bg_flag);
1703                         }
1704                 }
1705
1706                 /*
1707                  * Metadata in mixed block group profiles are accounted in data
1708                  */
1709                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
1710                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
1711                                 mixed = 1;
1712                         else
1713                                 total_free_meta += found->disk_total -
1714                                         found->disk_used;
1715                 }
1716
1717                 total_used += found->disk_used;
1718         }
1719
1720         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
1721         buf->f_blocks >>= bits;
1722         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
1723
1724         /* Account global block reserve as used, it's in logical size already */
1725         spin_lock(&block_rsv->lock);
1726         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
1727         if (buf->f_bfree >= block_rsv->size >> bits)
1728                 buf->f_bfree -= block_rsv->size >> bits;
1729         else
1730                 buf->f_bfree = 0;
1731         spin_unlock(&block_rsv->lock);
1732
1733         buf->f_bavail = div_u64(total_free_data, factor);
1734         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
1735         if (ret)
1736                 return ret;
1737         buf->f_bavail += div_u64(total_free_data, factor);
1738         buf->f_bavail = buf->f_bavail >> bits;
1739
1740         /*
1741          * We calculate the remaining metadata space minus global reserve. If
1742          * this is (supposedly) smaller than zero, there's no space. But this
1743          * does not hold in practice, the exhausted state happens where's still
1744          * some positive delta. So we apply some guesswork and compare the
1745          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
1746          *
1747          * We probably cannot calculate the exact threshold value because this
1748          * depends on the internal reservations requested by various
1749          * operations, so some operations that consume a few metadata will
1750          * succeed even if the Avail is zero. But this is better than the other
1751          * way around.
1752          */
1753         thresh = SZ_4M;
1754
1755         /*
1756          * We only want to claim there's no available space if we can no longer
1757          * allocate chunks for our metadata profile and our global reserve will
1758          * not fit in the free metadata space.  If we aren't ->full then we
1759          * still can allocate chunks and thus are fine using the currently
1760          * calculated f_bavail.
1761          */
1762         if (!mixed && block_rsv->space_info->full &&
1763             (total_free_meta < thresh || total_free_meta - thresh < block_rsv->size))
1764                 buf->f_bavail = 0;
1765
1766         buf->f_type = BTRFS_SUPER_MAGIC;
1767         buf->f_bsize = fs_info->sectorsize;
1768         buf->f_namelen = BTRFS_NAME_LEN;
1769
1770         /* We treat it as constant endianness (it doesn't matter _which_)
1771            because we want the fsid to come out the same whether mounted
1772            on a big-endian or little-endian host */
1773         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1774         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1775         /* Mask in the root object ID too, to disambiguate subvols */
1776         buf->f_fsid.val[0] ^= btrfs_root_id(BTRFS_I(d_inode(dentry))->root) >> 32;
1777         buf->f_fsid.val[1] ^= btrfs_root_id(BTRFS_I(d_inode(dentry))->root);
1778
1779         return 0;
1780 }
1781
1782 static int btrfs_fc_test_super(struct super_block *sb, struct fs_context *fc)
1783 {
1784         struct btrfs_fs_info *p = fc->s_fs_info;
1785         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1786
1787         return fs_info->fs_devices == p->fs_devices;
1788 }
1789
1790 static int btrfs_get_tree_super(struct fs_context *fc)
1791 {
1792         struct btrfs_fs_info *fs_info = fc->s_fs_info;
1793         struct btrfs_fs_context *ctx = fc->fs_private;
1794         struct btrfs_fs_devices *fs_devices = NULL;
1795         struct block_device *bdev;
1796         struct btrfs_device *device;
1797         struct super_block *sb;
1798         blk_mode_t mode = btrfs_open_mode(fc);
1799         int ret;
1800
1801         btrfs_ctx_to_info(fs_info, ctx);
1802         mutex_lock(&uuid_mutex);
1803
1804         /*
1805          * With 'true' passed to btrfs_scan_one_device() (mount time) we expect
1806          * either a valid device or an error.
1807          */
1808         device = btrfs_scan_one_device(fc->source, mode, true);
1809         ASSERT(device != NULL);
1810         if (IS_ERR(device)) {
1811                 mutex_unlock(&uuid_mutex);
1812                 return PTR_ERR(device);
1813         }
1814
1815         fs_devices = device->fs_devices;
1816         fs_info->fs_devices = fs_devices;
1817
1818         ret = btrfs_open_devices(fs_devices, mode, &btrfs_fs_type);
1819         mutex_unlock(&uuid_mutex);
1820         if (ret)
1821                 return ret;
1822
1823         if (!(fc->sb_flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1824                 ret = -EACCES;
1825                 goto error;
1826         }
1827
1828         bdev = fs_devices->latest_dev->bdev;
1829
1830         /*
1831          * From now on the error handling is not straightforward.
1832          *
1833          * If successful, this will transfer the fs_info into the super block,
1834          * and fc->s_fs_info will be NULL.  However if there's an existing
1835          * super, we'll still have fc->s_fs_info populated.  If we error
1836          * completely out it'll be cleaned up when we drop the fs_context,
1837          * otherwise it's tied to the lifetime of the super_block.
1838          */
1839         sb = sget_fc(fc, btrfs_fc_test_super, set_anon_super_fc);
1840         if (IS_ERR(sb)) {
1841                 ret = PTR_ERR(sb);
1842                 goto error;
1843         }
1844
1845         set_device_specific_options(fs_info);
1846
1847         if (sb->s_root) {
1848                 btrfs_close_devices(fs_devices);
1849                 if ((fc->sb_flags ^ sb->s_flags) & SB_RDONLY)
1850                         ret = -EBUSY;
1851         } else {
1852                 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1853                 shrinker_debugfs_rename(sb->s_shrink, "sb-btrfs:%s", sb->s_id);
1854                 btrfs_sb(sb)->bdev_holder = &btrfs_fs_type;
1855                 ret = btrfs_fill_super(sb, fs_devices, NULL);
1856         }
1857
1858         if (ret) {
1859                 deactivate_locked_super(sb);
1860                 return ret;
1861         }
1862
1863         btrfs_clear_oneshot_options(fs_info);
1864
1865         fc->root = dget(sb->s_root);
1866         return 0;
1867
1868 error:
1869         btrfs_close_devices(fs_devices);
1870         return ret;
1871 }
1872
1873 /*
1874  * Ever since commit 0723a0473fb4 ("btrfs: allow mounting btrfs subvolumes
1875  * with different ro/rw options") the following works:
1876  *
1877  *        (i) mount /dev/sda3 -o subvol=foo,ro /mnt/foo
1878  *       (ii) mount /dev/sda3 -o subvol=bar,rw /mnt/bar
1879  *
1880  * which looks nice and innocent but is actually pretty intricate and deserves
1881  * a long comment.
1882  *
1883  * On another filesystem a subvolume mount is close to something like:
1884  *
1885  *      (iii) # create rw superblock + initial mount
1886  *            mount -t xfs /dev/sdb /opt/
1887  *
1888  *            # create ro bind mount
1889  *            mount --bind -o ro /opt/foo /mnt/foo
1890  *
1891  *            # unmount initial mount
1892  *            umount /opt
1893  *
1894  * Of course, there's some special subvolume sauce and there's the fact that the
1895  * sb->s_root dentry is really swapped after mount_subtree(). But conceptually
1896  * it's very close and will help us understand the issue.
1897  *
1898  * The old mount API didn't cleanly distinguish between a mount being made ro
1899  * and a superblock being made ro.  The only way to change the ro state of
1900  * either object was by passing ms_rdonly. If a new mount was created via
1901  * mount(2) such as:
1902  *
1903  *      mount("/dev/sdb", "/mnt", "xfs", ms_rdonly, null);
1904  *
1905  * the MS_RDONLY flag being specified had two effects:
1906  *
1907  * (1) MNT_READONLY was raised -> the resulting mount got
1908  *     @mnt->mnt_flags |= MNT_READONLY raised.
1909  *
1910  * (2) MS_RDONLY was passed to the filesystem's mount method and the filesystems
1911  *     made the superblock ro. Note, how SB_RDONLY has the same value as
1912  *     ms_rdonly and is raised whenever MS_RDONLY is passed through mount(2).
1913  *
1914  * Creating a subtree mount via (iii) ends up leaving a rw superblock with a
1915  * subtree mounted ro.
1916  *
1917  * But consider the effect on the old mount API on btrfs subvolume mounting
1918  * which combines the distinct step in (iii) into a single step.
1919  *
1920  * By issuing (i) both the mount and the superblock are turned ro. Now when (ii)
1921  * is issued the superblock is ro and thus even if the mount created for (ii) is
1922  * rw it wouldn't help. Hence, btrfs needed to transition the superblock from ro
1923  * to rw for (ii) which it did using an internal remount call.
1924  *
1925  * IOW, subvolume mounting was inherently complicated due to the ambiguity of
1926  * MS_RDONLY in mount(2). Note, this ambiguity has mount(8) always translate
1927  * "ro" to MS_RDONLY. IOW, in both (i) and (ii) "ro" becomes MS_RDONLY when
1928  * passed by mount(8) to mount(2).
1929  *
1930  * Enter the new mount API. The new mount API disambiguates making a mount ro
1931  * and making a superblock ro.
1932  *
1933  * (3) To turn a mount ro the MOUNT_ATTR_ONLY flag can be used with either
1934  *     fsmount() or mount_setattr() this is a pure VFS level change for a
1935  *     specific mount or mount tree that is never seen by the filesystem itself.
1936  *
1937  * (4) To turn a superblock ro the "ro" flag must be used with
1938  *     fsconfig(FSCONFIG_SET_FLAG, "ro"). This option is seen by the filesystem
1939  *     in fc->sb_flags.
1940  *
1941  * This disambiguation has rather positive consequences.  Mounting a subvolume
1942  * ro will not also turn the superblock ro. Only the mount for the subvolume
1943  * will become ro.
1944  *
1945  * So, if the superblock creation request comes from the new mount API the
1946  * caller must have explicitly done:
1947  *
1948  *      fsconfig(FSCONFIG_SET_FLAG, "ro")
1949  *      fsmount/mount_setattr(MOUNT_ATTR_RDONLY)
1950  *
1951  * IOW, at some point the caller must have explicitly turned the whole
1952  * superblock ro and we shouldn't just undo it like we did for the old mount
1953  * API. In any case, it lets us avoid the hack in the new mount API.
1954  *
1955  * Consequently, the remounting hack must only be used for requests originating
1956  * from the old mount API and should be marked for full deprecation so it can be
1957  * turned off in a couple of years.
1958  *
1959  * The new mount API has no reason to support this hack.
1960  */
1961 static struct vfsmount *btrfs_reconfigure_for_mount(struct fs_context *fc)
1962 {
1963         struct vfsmount *mnt;
1964         int ret;
1965         const bool ro2rw = !(fc->sb_flags & SB_RDONLY);
1966
1967         /*
1968          * We got an EBUSY because our SB_RDONLY flag didn't match the existing
1969          * super block, so invert our setting here and retry the mount so we
1970          * can get our vfsmount.
1971          */
1972         if (ro2rw)
1973                 fc->sb_flags |= SB_RDONLY;
1974         else
1975                 fc->sb_flags &= ~SB_RDONLY;
1976
1977         mnt = fc_mount(fc);
1978         if (IS_ERR(mnt))
1979                 return mnt;
1980
1981         if (!fc->oldapi || !ro2rw)
1982                 return mnt;
1983
1984         /* We need to convert to rw, call reconfigure. */
1985         fc->sb_flags &= ~SB_RDONLY;
1986         down_write(&mnt->mnt_sb->s_umount);
1987         ret = btrfs_reconfigure(fc);
1988         up_write(&mnt->mnt_sb->s_umount);
1989         if (ret) {
1990                 mntput(mnt);
1991                 return ERR_PTR(ret);
1992         }
1993         return mnt;
1994 }
1995
1996 static int btrfs_get_tree_subvol(struct fs_context *fc)
1997 {
1998         struct btrfs_fs_info *fs_info = NULL;
1999         struct btrfs_fs_context *ctx = fc->fs_private;
2000         struct fs_context *dup_fc;
2001         struct dentry *dentry;
2002         struct vfsmount *mnt;
2003
2004         /*
2005          * Setup a dummy root and fs_info for test/set super.  This is because
2006          * we don't actually fill this stuff out until open_ctree, but we need
2007          * then open_ctree will properly initialize the file system specific
2008          * settings later.  btrfs_init_fs_info initializes the static elements
2009          * of the fs_info (locks and such) to make cleanup easier if we find a
2010          * superblock with our given fs_devices later on at sget() time.
2011          */
2012         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
2013         if (!fs_info)
2014                 return -ENOMEM;
2015
2016         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
2017         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
2018         if (!fs_info->super_copy || !fs_info->super_for_commit) {
2019                 btrfs_free_fs_info(fs_info);
2020                 return -ENOMEM;
2021         }
2022         btrfs_init_fs_info(fs_info);
2023
2024         dup_fc = vfs_dup_fs_context(fc);
2025         if (IS_ERR(dup_fc)) {
2026                 btrfs_free_fs_info(fs_info);
2027                 return PTR_ERR(dup_fc);
2028         }
2029
2030         /*
2031          * When we do the sget_fc this gets transferred to the sb, so we only
2032          * need to set it on the dup_fc as this is what creates the super block.
2033          */
2034         dup_fc->s_fs_info = fs_info;
2035
2036         /*
2037          * We'll do the security settings in our btrfs_get_tree_super() mount
2038          * loop, they were duplicated into dup_fc, we can drop the originals
2039          * here.
2040          */
2041         security_free_mnt_opts(&fc->security);
2042         fc->security = NULL;
2043
2044         mnt = fc_mount(dup_fc);
2045         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY)
2046                 mnt = btrfs_reconfigure_for_mount(dup_fc);
2047         put_fs_context(dup_fc);
2048         if (IS_ERR(mnt))
2049                 return PTR_ERR(mnt);
2050
2051         /*
2052          * This free's ->subvol_name, because if it isn't set we have to
2053          * allocate a buffer to hold the subvol_name, so we just drop our
2054          * reference to it here.
2055          */
2056         dentry = mount_subvol(ctx->subvol_name, ctx->subvol_objectid, mnt);
2057         ctx->subvol_name = NULL;
2058         if (IS_ERR(dentry))
2059                 return PTR_ERR(dentry);
2060
2061         fc->root = dentry;
2062         return 0;
2063 }
2064
2065 static int btrfs_get_tree(struct fs_context *fc)
2066 {
2067         /*
2068          * Since we use mount_subtree to mount the default/specified subvol, we
2069          * have to do mounts in two steps.
2070          *
2071          * First pass through we call btrfs_get_tree_subvol(), this is just a
2072          * wrapper around fc_mount() to call back into here again, and this time
2073          * we'll call btrfs_get_tree_super().  This will do the open_ctree() and
2074          * everything to open the devices and file system.  Then we return back
2075          * with a fully constructed vfsmount in btrfs_get_tree_subvol(), and
2076          * from there we can do our mount_subvol() call, which will lookup
2077          * whichever subvol we're mounting and setup this fc with the
2078          * appropriate dentry for the subvol.
2079          */
2080         if (fc->s_fs_info)
2081                 return btrfs_get_tree_super(fc);
2082         return btrfs_get_tree_subvol(fc);
2083 }
2084
2085 static void btrfs_kill_super(struct super_block *sb)
2086 {
2087         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2088         kill_anon_super(sb);
2089         btrfs_free_fs_info(fs_info);
2090 }
2091
2092 static void btrfs_free_fs_context(struct fs_context *fc)
2093 {
2094         struct btrfs_fs_context *ctx = fc->fs_private;
2095         struct btrfs_fs_info *fs_info = fc->s_fs_info;
2096
2097         if (fs_info)
2098                 btrfs_free_fs_info(fs_info);
2099
2100         if (ctx && refcount_dec_and_test(&ctx->refs)) {
2101                 kfree(ctx->subvol_name);
2102                 kfree(ctx);
2103         }
2104 }
2105
2106 static int btrfs_dup_fs_context(struct fs_context *fc, struct fs_context *src_fc)
2107 {
2108         struct btrfs_fs_context *ctx = src_fc->fs_private;
2109
2110         /*
2111          * Give a ref to our ctx to this dup, as we want to keep it around for
2112          * our original fc so we can have the subvolume name or objectid.
2113          *
2114          * We unset ->source in the original fc because the dup needs it for
2115          * mounting, and then once we free the dup it'll free ->source, so we
2116          * need to make sure we're only pointing to it in one fc.
2117          */
2118         refcount_inc(&ctx->refs);
2119         fc->fs_private = ctx;
2120         fc->source = src_fc->source;
2121         src_fc->source = NULL;
2122         return 0;
2123 }
2124
2125 static const struct fs_context_operations btrfs_fs_context_ops = {
2126         .parse_param    = btrfs_parse_param,
2127         .reconfigure    = btrfs_reconfigure,
2128         .get_tree       = btrfs_get_tree,
2129         .dup            = btrfs_dup_fs_context,
2130         .free           = btrfs_free_fs_context,
2131 };
2132
2133 static int btrfs_init_fs_context(struct fs_context *fc)
2134 {
2135         struct btrfs_fs_context *ctx;
2136
2137         ctx = kzalloc(sizeof(struct btrfs_fs_context), GFP_KERNEL);
2138         if (!ctx)
2139                 return -ENOMEM;
2140
2141         refcount_set(&ctx->refs, 1);
2142         fc->fs_private = ctx;
2143         fc->ops = &btrfs_fs_context_ops;
2144
2145         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2146                 btrfs_info_to_ctx(btrfs_sb(fc->root->d_sb), ctx);
2147         } else {
2148                 ctx->thread_pool_size =
2149                         min_t(unsigned long, num_online_cpus() + 2, 8);
2150                 ctx->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2151                 ctx->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2152         }
2153
2154 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
2155         fc->sb_flags |= SB_POSIXACL;
2156 #endif
2157         fc->sb_flags |= SB_I_VERSION;
2158
2159         return 0;
2160 }
2161
2162 static struct file_system_type btrfs_fs_type = {
2163         .owner                  = THIS_MODULE,
2164         .name                   = "btrfs",
2165         .init_fs_context        = btrfs_init_fs_context,
2166         .parameters             = btrfs_fs_parameters,
2167         .kill_sb                = btrfs_kill_super,
2168         .fs_flags               = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA | FS_ALLOW_IDMAP,
2169  };
2170
2171 MODULE_ALIAS_FS("btrfs");
2172
2173 static int btrfs_control_open(struct inode *inode, struct file *file)
2174 {
2175         /*
2176          * The control file's private_data is used to hold the
2177          * transaction when it is started and is used to keep
2178          * track of whether a transaction is already in progress.
2179          */
2180         file->private_data = NULL;
2181         return 0;
2182 }
2183
2184 /*
2185  * Used by /dev/btrfs-control for devices ioctls.
2186  */
2187 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2188                                 unsigned long arg)
2189 {
2190         struct btrfs_ioctl_vol_args *vol;
2191         struct btrfs_device *device = NULL;
2192         dev_t devt = 0;
2193         int ret = -ENOTTY;
2194
2195         if (!capable(CAP_SYS_ADMIN))
2196                 return -EPERM;
2197
2198         vol = memdup_user((void __user *)arg, sizeof(*vol));
2199         if (IS_ERR(vol))
2200                 return PTR_ERR(vol);
2201         ret = btrfs_check_ioctl_vol_args_path(vol);
2202         if (ret < 0)
2203                 goto out;
2204
2205         switch (cmd) {
2206         case BTRFS_IOC_SCAN_DEV:
2207                 mutex_lock(&uuid_mutex);
2208                 /*
2209                  * Scanning outside of mount can return NULL which would turn
2210                  * into 0 error code.
2211                  */
2212                 device = btrfs_scan_one_device(vol->name, BLK_OPEN_READ, false);
2213                 ret = PTR_ERR_OR_ZERO(device);
2214                 mutex_unlock(&uuid_mutex);
2215                 break;
2216         case BTRFS_IOC_FORGET_DEV:
2217                 if (vol->name[0] != 0) {
2218                         ret = lookup_bdev(vol->name, &devt);
2219                         if (ret)
2220                                 break;
2221                 }
2222                 ret = btrfs_forget_devices(devt);
2223                 break;
2224         case BTRFS_IOC_DEVICES_READY:
2225                 mutex_lock(&uuid_mutex);
2226                 /*
2227                  * Scanning outside of mount can return NULL which would turn
2228                  * into 0 error code.
2229                  */
2230                 device = btrfs_scan_one_device(vol->name, BLK_OPEN_READ, false);
2231                 if (IS_ERR_OR_NULL(device)) {
2232                         mutex_unlock(&uuid_mutex);
2233                         ret = PTR_ERR(device);
2234                         break;
2235                 }
2236                 ret = !(device->fs_devices->num_devices ==
2237                         device->fs_devices->total_devices);
2238                 mutex_unlock(&uuid_mutex);
2239                 break;
2240         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2241                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2242                 break;
2243         }
2244
2245 out:
2246         kfree(vol);
2247         return ret;
2248 }
2249
2250 static int btrfs_freeze(struct super_block *sb)
2251 {
2252         struct btrfs_trans_handle *trans;
2253         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2254         struct btrfs_root *root = fs_info->tree_root;
2255
2256         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2257         /*
2258          * We don't need a barrier here, we'll wait for any transaction that
2259          * could be in progress on other threads (and do delayed iputs that
2260          * we want to avoid on a frozen filesystem), or do the commit
2261          * ourselves.
2262          */
2263         trans = btrfs_attach_transaction_barrier(root);
2264         if (IS_ERR(trans)) {
2265                 /* no transaction, don't bother */
2266                 if (PTR_ERR(trans) == -ENOENT)
2267                         return 0;
2268                 return PTR_ERR(trans);
2269         }
2270         return btrfs_commit_transaction(trans);
2271 }
2272
2273 static int check_dev_super(struct btrfs_device *dev)
2274 {
2275         struct btrfs_fs_info *fs_info = dev->fs_info;
2276         struct btrfs_super_block *sb;
2277         u64 last_trans;
2278         u16 csum_type;
2279         int ret = 0;
2280
2281         /* This should be called with fs still frozen. */
2282         ASSERT(test_bit(BTRFS_FS_FROZEN, &fs_info->flags));
2283
2284         /* Missing dev, no need to check. */
2285         if (!dev->bdev)
2286                 return 0;
2287
2288         /* Only need to check the primary super block. */
2289         sb = btrfs_read_dev_one_super(dev->bdev, 0, true);
2290         if (IS_ERR(sb))
2291                 return PTR_ERR(sb);
2292
2293         /* Verify the checksum. */
2294         csum_type = btrfs_super_csum_type(sb);
2295         if (csum_type != btrfs_super_csum_type(fs_info->super_copy)) {
2296                 btrfs_err(fs_info, "csum type changed, has %u expect %u",
2297                           csum_type, btrfs_super_csum_type(fs_info->super_copy));
2298                 ret = -EUCLEAN;
2299                 goto out;
2300         }
2301
2302         if (btrfs_check_super_csum(fs_info, sb)) {
2303                 btrfs_err(fs_info, "csum for on-disk super block no longer matches");
2304                 ret = -EUCLEAN;
2305                 goto out;
2306         }
2307
2308         /* Btrfs_validate_super() includes fsid check against super->fsid. */
2309         ret = btrfs_validate_super(fs_info, sb, 0);
2310         if (ret < 0)
2311                 goto out;
2312
2313         last_trans = btrfs_get_last_trans_committed(fs_info);
2314         if (btrfs_super_generation(sb) != last_trans) {
2315                 btrfs_err(fs_info, "transid mismatch, has %llu expect %llu",
2316                           btrfs_super_generation(sb), last_trans);
2317                 ret = -EUCLEAN;
2318                 goto out;
2319         }
2320 out:
2321         btrfs_release_disk_super(sb);
2322         return ret;
2323 }
2324
2325 static int btrfs_unfreeze(struct super_block *sb)
2326 {
2327         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2328         struct btrfs_device *device;
2329         int ret = 0;
2330
2331         /*
2332          * Make sure the fs is not changed by accident (like hibernation then
2333          * modified by other OS).
2334          * If we found anything wrong, we mark the fs error immediately.
2335          *
2336          * And since the fs is frozen, no one can modify the fs yet, thus
2337          * we don't need to hold device_list_mutex.
2338          */
2339         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
2340                 ret = check_dev_super(device);
2341                 if (ret < 0) {
2342                         btrfs_handle_fs_error(fs_info, ret,
2343                                 "super block on devid %llu got modified unexpectedly",
2344                                 device->devid);
2345                         break;
2346                 }
2347         }
2348         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2349
2350         /*
2351          * We still return 0, to allow VFS layer to unfreeze the fs even the
2352          * above checks failed. Since the fs is either fine or read-only, we're
2353          * safe to continue, without causing further damage.
2354          */
2355         return 0;
2356 }
2357
2358 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2359 {
2360         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2361
2362         /*
2363          * There should be always a valid pointer in latest_dev, it may be stale
2364          * for a short moment in case it's being deleted but still valid until
2365          * the end of RCU grace period.
2366          */
2367         rcu_read_lock();
2368         seq_escape(m, btrfs_dev_name(fs_info->fs_devices->latest_dev), " \t\n\\");
2369         rcu_read_unlock();
2370
2371         return 0;
2372 }
2373
2374 static long btrfs_nr_cached_objects(struct super_block *sb, struct shrink_control *sc)
2375 {
2376         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2377         const s64 nr = percpu_counter_sum_positive(&fs_info->evictable_extent_maps);
2378
2379         trace_btrfs_extent_map_shrinker_count(fs_info, nr);
2380
2381         return nr;
2382 }
2383
2384 static long btrfs_free_cached_objects(struct super_block *sb, struct shrink_control *sc)
2385 {
2386         const long nr_to_scan = min_t(unsigned long, LONG_MAX, sc->nr_to_scan);
2387         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2388
2389         return btrfs_free_extent_maps(fs_info, nr_to_scan);
2390 }
2391
2392 static const struct super_operations btrfs_super_ops = {
2393         .drop_inode     = btrfs_drop_inode,
2394         .evict_inode    = btrfs_evict_inode,
2395         .put_super      = btrfs_put_super,
2396         .sync_fs        = btrfs_sync_fs,
2397         .show_options   = btrfs_show_options,
2398         .show_devname   = btrfs_show_devname,
2399         .alloc_inode    = btrfs_alloc_inode,
2400         .destroy_inode  = btrfs_destroy_inode,
2401         .free_inode     = btrfs_free_inode,
2402         .statfs         = btrfs_statfs,
2403         .freeze_fs      = btrfs_freeze,
2404         .unfreeze_fs    = btrfs_unfreeze,
2405         .nr_cached_objects = btrfs_nr_cached_objects,
2406         .free_cached_objects = btrfs_free_cached_objects,
2407 };
2408
2409 static const struct file_operations btrfs_ctl_fops = {
2410         .open = btrfs_control_open,
2411         .unlocked_ioctl  = btrfs_control_ioctl,
2412         .compat_ioctl = compat_ptr_ioctl,
2413         .owner   = THIS_MODULE,
2414         .llseek = noop_llseek,
2415 };
2416
2417 static struct miscdevice btrfs_misc = {
2418         .minor          = BTRFS_MINOR,
2419         .name           = "btrfs-control",
2420         .fops           = &btrfs_ctl_fops
2421 };
2422
2423 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2424 MODULE_ALIAS("devname:btrfs-control");
2425
2426 static int __init btrfs_interface_init(void)
2427 {
2428         return misc_register(&btrfs_misc);
2429 }
2430
2431 static __cold void btrfs_interface_exit(void)
2432 {
2433         misc_deregister(&btrfs_misc);
2434 }
2435
2436 static int __init btrfs_print_mod_info(void)
2437 {
2438         static const char options[] = ""
2439 #ifdef CONFIG_BTRFS_DEBUG
2440                         ", debug=on"
2441 #endif
2442 #ifdef CONFIG_BTRFS_ASSERT
2443                         ", assert=on"
2444 #endif
2445 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2446                         ", ref-verify=on"
2447 #endif
2448 #ifdef CONFIG_BLK_DEV_ZONED
2449                         ", zoned=yes"
2450 #else
2451                         ", zoned=no"
2452 #endif
2453 #ifdef CONFIG_FS_VERITY
2454                         ", fsverity=yes"
2455 #else
2456                         ", fsverity=no"
2457 #endif
2458                         ;
2459         pr_info("Btrfs loaded%s\n", options);
2460         return 0;
2461 }
2462
2463 static int register_btrfs(void)
2464 {
2465         return register_filesystem(&btrfs_fs_type);
2466 }
2467
2468 static void unregister_btrfs(void)
2469 {
2470         unregister_filesystem(&btrfs_fs_type);
2471 }
2472
2473 /* Helper structure for long init/exit functions. */
2474 struct init_sequence {
2475         int (*init_func)(void);
2476         /* Can be NULL if the init_func doesn't need cleanup. */
2477         void (*exit_func)(void);
2478 };
2479
2480 static const struct init_sequence mod_init_seq[] = {
2481         {
2482                 .init_func = btrfs_props_init,
2483                 .exit_func = NULL,
2484         }, {
2485                 .init_func = btrfs_init_sysfs,
2486                 .exit_func = btrfs_exit_sysfs,
2487         }, {
2488                 .init_func = btrfs_init_compress,
2489                 .exit_func = btrfs_exit_compress,
2490         }, {
2491                 .init_func = btrfs_init_cachep,
2492                 .exit_func = btrfs_destroy_cachep,
2493         }, {
2494                 .init_func = btrfs_transaction_init,
2495                 .exit_func = btrfs_transaction_exit,
2496         }, {
2497                 .init_func = btrfs_ctree_init,
2498                 .exit_func = btrfs_ctree_exit,
2499         }, {
2500                 .init_func = btrfs_free_space_init,
2501                 .exit_func = btrfs_free_space_exit,
2502         }, {
2503                 .init_func = extent_state_init_cachep,
2504                 .exit_func = extent_state_free_cachep,
2505         }, {
2506                 .init_func = extent_buffer_init_cachep,
2507                 .exit_func = extent_buffer_free_cachep,
2508         }, {
2509                 .init_func = btrfs_bioset_init,
2510                 .exit_func = btrfs_bioset_exit,
2511         }, {
2512                 .init_func = extent_map_init,
2513                 .exit_func = extent_map_exit,
2514         }, {
2515                 .init_func = ordered_data_init,
2516                 .exit_func = ordered_data_exit,
2517         }, {
2518                 .init_func = btrfs_delayed_inode_init,
2519                 .exit_func = btrfs_delayed_inode_exit,
2520         }, {
2521                 .init_func = btrfs_auto_defrag_init,
2522                 .exit_func = btrfs_auto_defrag_exit,
2523         }, {
2524                 .init_func = btrfs_delayed_ref_init,
2525                 .exit_func = btrfs_delayed_ref_exit,
2526         }, {
2527                 .init_func = btrfs_prelim_ref_init,
2528                 .exit_func = btrfs_prelim_ref_exit,
2529         }, {
2530                 .init_func = btrfs_interface_init,
2531                 .exit_func = btrfs_interface_exit,
2532         }, {
2533                 .init_func = btrfs_print_mod_info,
2534                 .exit_func = NULL,
2535         }, {
2536                 .init_func = btrfs_run_sanity_tests,
2537                 .exit_func = NULL,
2538         }, {
2539                 .init_func = register_btrfs,
2540                 .exit_func = unregister_btrfs,
2541         }
2542 };
2543
2544 static bool mod_init_result[ARRAY_SIZE(mod_init_seq)];
2545
2546 static __always_inline void btrfs_exit_btrfs_fs(void)
2547 {
2548         int i;
2549
2550         for (i = ARRAY_SIZE(mod_init_seq) - 1; i >= 0; i--) {
2551                 if (!mod_init_result[i])
2552                         continue;
2553                 if (mod_init_seq[i].exit_func)
2554                         mod_init_seq[i].exit_func();
2555                 mod_init_result[i] = false;
2556         }
2557 }
2558
2559 static void __exit exit_btrfs_fs(void)
2560 {
2561         btrfs_exit_btrfs_fs();
2562         btrfs_cleanup_fs_uuids();
2563 }
2564
2565 static int __init init_btrfs_fs(void)
2566 {
2567         int ret;
2568         int i;
2569
2570         for (i = 0; i < ARRAY_SIZE(mod_init_seq); i++) {
2571                 ASSERT(!mod_init_result[i]);
2572                 ret = mod_init_seq[i].init_func();
2573                 if (ret < 0) {
2574                         btrfs_exit_btrfs_fs();
2575                         return ret;
2576                 }
2577                 mod_init_result[i] = true;
2578         }
2579         return 0;
2580 }
2581
2582 late_initcall(init_btrfs_fs);
2583 module_exit(exit_btrfs_fs)
2584
2585 MODULE_LICENSE("GPL");
2586 MODULE_SOFTDEP("pre: crc32c");
2587 MODULE_SOFTDEP("pre: xxhash64");
2588 MODULE_SOFTDEP("pre: sha256");
2589 MODULE_SOFTDEP("pre: blake2b-256");