1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/super.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Big-endian to little-endian byte-swapping/bitmaps by 17 * David S. Miller (davem@caip.rutgers.edu), 1995 18 */ 19 20 #include <linux/module.h> 21 #include <linux/string.h> 22 #include <linux/fs.h> 23 #include <linux/time.h> 24 #include <linux/vmalloc.h> 25 #include <linux/slab.h> 26 #include <linux/init.h> 27 #include <linux/blkdev.h> 28 #include <linux/backing-dev.h> 29 #include <linux/parser.h> 30 #include <linux/buffer_head.h> 31 #include <linux/exportfs.h> 32 #include <linux/vfs.h> 33 #include <linux/random.h> 34 #include <linux/mount.h> 35 #include <linux/namei.h> 36 #include <linux/quotaops.h> 37 #include <linux/seq_file.h> 38 #include <linux/ctype.h> 39 #include <linux/log2.h> 40 #include <linux/crc16.h> 41 #include <linux/dax.h> 42 #include <linux/uaccess.h> 43 #include <linux/iversion.h> 44 #include <linux/unicode.h> 45 #include <linux/part_stat.h> 46 #include <linux/kthread.h> 47 #include <linux/freezer.h> 48 #include <linux/fsnotify.h> 49 #include <linux/fs_context.h> 50 #include <linux/fs_parser.h> 51 #include <linux/fserror.h> 52 53 #include "ext4.h" 54 #include "ext4_extents.h" /* Needed for trace points definition */ 55 #include "ext4_jbd2.h" 56 #include "xattr.h" 57 #include "acl.h" 58 #include "mballoc.h" 59 #include "fsmap.h" 60 61 #define CREATE_TRACE_POINTS 62 #include <trace/events/ext4.h> 63 64 static struct ext4_lazy_init *ext4_li_info; 65 static DEFINE_MUTEX(ext4_li_mtx); 66 static struct ratelimit_state ext4_mount_msg_ratelimit; 67 68 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 69 unsigned long journal_devnum); 70 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 71 static void ext4_update_super(struct super_block *sb); 72 static int ext4_commit_super(struct super_block *sb); 73 static int ext4_mark_recovery_complete(struct super_block *sb, 74 struct ext4_super_block *es); 75 static int ext4_clear_journal_err(struct super_block *sb, 76 struct ext4_super_block *es); 77 static int ext4_sync_fs(struct super_block *sb, int wait); 78 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 79 static int ext4_unfreeze(struct super_block *sb); 80 static int ext4_freeze(struct super_block *sb); 81 static inline int ext2_feature_set_ok(struct super_block *sb); 82 static inline int ext3_feature_set_ok(struct super_block *sb); 83 static void ext4_unregister_li_request(struct super_block *sb); 84 static void ext4_clear_request_list(void); 85 static struct inode *ext4_get_journal_inode(struct super_block *sb, 86 unsigned int journal_inum); 87 static int ext4_validate_options(struct fs_context *fc); 88 static int ext4_check_opt_consistency(struct fs_context *fc, 89 struct super_block *sb); 90 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb); 91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param); 92 static int ext4_get_tree(struct fs_context *fc); 93 static int ext4_reconfigure(struct fs_context *fc); 94 static void ext4_fc_free(struct fs_context *fc); 95 static int ext4_init_fs_context(struct fs_context *fc); 96 static void ext4_kill_sb(struct super_block *sb); 97 static const struct fs_parameter_spec ext4_param_specs[]; 98 99 /* 100 * Lock ordering 101 * 102 * page fault path: 103 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start 104 * -> page lock -> i_data_sem (rw) 105 * 106 * buffered write path: 107 * sb_start_write -> i_mutex -> mmap_lock 108 * sb_start_write -> i_mutex -> transaction start -> page lock -> 109 * i_data_sem (rw) 110 * 111 * truncate: 112 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) -> 113 * page lock 114 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start -> 115 * i_data_sem (rw) 116 * 117 * direct IO: 118 * sb_start_write -> i_mutex -> mmap_lock 119 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 120 * 121 * writepages: 122 * transaction start -> page lock(s) -> i_data_sem (rw) 123 */ 124 125 static const struct fs_context_operations ext4_context_ops = { 126 .parse_param = ext4_parse_param, 127 .get_tree = ext4_get_tree, 128 .reconfigure = ext4_reconfigure, 129 .free = ext4_fc_free, 130 }; 131 132 133 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 134 static struct file_system_type ext2_fs_type = { 135 .owner = THIS_MODULE, 136 .name = "ext2", 137 .init_fs_context = ext4_init_fs_context, 138 .parameters = ext4_param_specs, 139 .kill_sb = ext4_kill_sb, 140 .fs_flags = FS_REQUIRES_DEV, 141 }; 142 MODULE_ALIAS_FS("ext2"); 143 MODULE_ALIAS("ext2"); 144 #define IS_EXT2_SB(sb) ((sb)->s_type == &ext2_fs_type) 145 #else 146 #define IS_EXT2_SB(sb) (0) 147 #endif 148 149 150 static struct file_system_type ext3_fs_type = { 151 .owner = THIS_MODULE, 152 .name = "ext3", 153 .init_fs_context = ext4_init_fs_context, 154 .parameters = ext4_param_specs, 155 .kill_sb = ext4_kill_sb, 156 .fs_flags = FS_REQUIRES_DEV, 157 }; 158 MODULE_ALIAS_FS("ext3"); 159 MODULE_ALIAS("ext3"); 160 #define IS_EXT3_SB(sb) ((sb)->s_type == &ext3_fs_type) 161 162 163 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 164 bio_end_io_t end_io, bool simu_fail) 165 { 166 if (simu_fail) { 167 clear_buffer_uptodate(bh); 168 unlock_buffer(bh); 169 return; 170 } 171 172 /* 173 * buffer's verified bit is no longer valid after reading from 174 * disk again due to write out error, clear it to make sure we 175 * recheck the buffer contents. 176 */ 177 clear_buffer_verified(bh); 178 179 if (!end_io) 180 end_io = bh_end_read; 181 bh_submit(bh, REQ_OP_READ | op_flags, end_io); 182 } 183 184 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags, 185 bio_end_io_t end_io, bool simu_fail) 186 { 187 BUG_ON(!buffer_locked(bh)); 188 189 if (ext4_buffer_uptodate(bh)) { 190 unlock_buffer(bh); 191 return; 192 } 193 __ext4_read_bh(bh, op_flags, end_io, simu_fail); 194 } 195 196 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 197 bio_end_io_t end_io, bool simu_fail) 198 { 199 BUG_ON(!buffer_locked(bh)); 200 201 if (ext4_buffer_uptodate(bh)) { 202 unlock_buffer(bh); 203 return 0; 204 } 205 206 __ext4_read_bh(bh, op_flags, end_io, simu_fail); 207 208 wait_on_buffer(bh); 209 if (buffer_uptodate(bh)) 210 return 0; 211 return -EIO; 212 } 213 214 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 215 { 216 lock_buffer(bh); 217 if (!wait) { 218 ext4_read_bh_nowait(bh, op_flags, NULL, false); 219 return 0; 220 } 221 return ext4_read_bh(bh, op_flags, NULL, false); 222 } 223 224 /* 225 * This works like __bread_gfp() except it uses ERR_PTR for error 226 * returns. Currently with sb_bread it's impossible to distinguish 227 * between ENOMEM and EIO situations (since both result in a NULL 228 * return. 229 */ 230 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 231 sector_t block, 232 blk_opf_t op_flags, gfp_t gfp) 233 { 234 struct buffer_head *bh; 235 int ret; 236 237 bh = sb_getblk_gfp(sb, block, gfp); 238 if (bh == NULL) 239 return ERR_PTR(-ENOMEM); 240 if (ext4_buffer_uptodate(bh)) 241 return bh; 242 243 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 244 if (ret) { 245 put_bh(bh); 246 return ERR_PTR(ret); 247 } 248 return bh; 249 } 250 251 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 252 blk_opf_t op_flags) 253 { 254 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping, 255 ~__GFP_FS) | __GFP_MOVABLE; 256 257 return __ext4_sb_bread_gfp(sb, block, op_flags, gfp); 258 } 259 260 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 261 sector_t block) 262 { 263 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping, 264 ~__GFP_FS); 265 266 return __ext4_sb_bread_gfp(sb, block, 0, gfp); 267 } 268 269 struct buffer_head *ext4_sb_bread_nofail(struct super_block *sb, 270 sector_t block) 271 { 272 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping, 273 ~__GFP_FS) | __GFP_MOVABLE | __GFP_NOFAIL; 274 275 return __ext4_sb_bread_gfp(sb, block, 0, gfp); 276 } 277 278 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 279 { 280 struct buffer_head *bh = bdev_getblk(sb->s_bdev, block, 281 sb->s_blocksize, GFP_NOWAIT); 282 283 if (likely(bh)) { 284 if (trylock_buffer(bh)) 285 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL, false); 286 brelse(bh); 287 } 288 } 289 290 static int ext4_verify_csum_type(struct super_block *sb, 291 struct ext4_super_block *es) 292 { 293 if (!ext4_has_feature_metadata_csum(sb)) 294 return 1; 295 296 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 297 } 298 299 __le32 ext4_superblock_csum(struct ext4_super_block *es) 300 { 301 int offset = offsetof(struct ext4_super_block, s_checksum); 302 __u32 csum; 303 304 csum = ext4_chksum(~0, (char *)es, offset); 305 306 return cpu_to_le32(csum); 307 } 308 309 static int ext4_superblock_csum_verify(struct super_block *sb, 310 struct ext4_super_block *es) 311 { 312 if (!ext4_has_feature_metadata_csum(sb)) 313 return 1; 314 315 return es->s_checksum == ext4_superblock_csum(es); 316 } 317 318 void ext4_superblock_csum_set(struct super_block *sb) 319 { 320 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 321 322 if (!ext4_has_feature_metadata_csum(sb)) 323 return; 324 325 es->s_checksum = ext4_superblock_csum(es); 326 } 327 328 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 329 struct ext4_group_desc *bg) 330 { 331 return le32_to_cpu(bg->bg_block_bitmap_lo) | 332 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 333 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 334 } 335 336 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 337 struct ext4_group_desc *bg) 338 { 339 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 340 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 341 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 342 } 343 344 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 345 struct ext4_group_desc *bg) 346 { 347 return le32_to_cpu(bg->bg_inode_table_lo) | 348 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 349 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 350 } 351 352 __u32 ext4_free_group_clusters(struct super_block *sb, 353 struct ext4_group_desc *bg) 354 { 355 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 356 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 357 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 358 } 359 360 __u32 ext4_free_inodes_count(struct super_block *sb, 361 struct ext4_group_desc *bg) 362 { 363 return le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_lo)) | 364 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 365 (__u32)le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_hi)) << 16 : 0); 366 } 367 368 __u32 ext4_used_dirs_count(struct super_block *sb, 369 struct ext4_group_desc *bg) 370 { 371 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 372 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 373 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 374 } 375 376 __u32 ext4_itable_unused_count(struct super_block *sb, 377 struct ext4_group_desc *bg) 378 { 379 return le16_to_cpu(bg->bg_itable_unused_lo) | 380 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 381 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 382 } 383 384 void ext4_block_bitmap_set(struct super_block *sb, 385 struct ext4_group_desc *bg, ext4_fsblk_t blk) 386 { 387 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 388 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 389 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 390 } 391 392 void ext4_inode_bitmap_set(struct super_block *sb, 393 struct ext4_group_desc *bg, ext4_fsblk_t blk) 394 { 395 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 396 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 397 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 398 } 399 400 void ext4_inode_table_set(struct super_block *sb, 401 struct ext4_group_desc *bg, ext4_fsblk_t blk) 402 { 403 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 404 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 405 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 406 } 407 408 void ext4_free_group_clusters_set(struct super_block *sb, 409 struct ext4_group_desc *bg, __u32 count) 410 { 411 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 412 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 413 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 414 } 415 416 void ext4_free_inodes_set(struct super_block *sb, 417 struct ext4_group_desc *bg, __u32 count) 418 { 419 WRITE_ONCE(bg->bg_free_inodes_count_lo, cpu_to_le16((__u16)count)); 420 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 421 WRITE_ONCE(bg->bg_free_inodes_count_hi, cpu_to_le16(count >> 16)); 422 } 423 424 void ext4_used_dirs_set(struct super_block *sb, 425 struct ext4_group_desc *bg, __u32 count) 426 { 427 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 428 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 429 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 430 } 431 432 void ext4_itable_unused_set(struct super_block *sb, 433 struct ext4_group_desc *bg, __u32 count) 434 { 435 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 436 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 437 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 438 } 439 440 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 441 { 442 now = clamp_val(now, 0, (1ull << 40) - 1); 443 444 *lo = cpu_to_le32(lower_32_bits(now)); 445 *hi = upper_32_bits(now); 446 } 447 448 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 449 { 450 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 451 } 452 #define ext4_update_tstamp(es, tstamp) \ 453 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 454 ktime_get_real_seconds()) 455 #define ext4_get_tstamp(es, tstamp) \ 456 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 457 458 /* 459 * The ext4_maybe_update_superblock() function checks and updates the 460 * superblock if needed. 461 * 462 * This function is designed to update the on-disk superblock only under 463 * certain conditions to prevent excessive disk writes and unnecessary 464 * waking of the disk from sleep. The superblock will be updated if: 465 * 1. More than sbi->s_sb_update_sec (def: 1 hour) has passed since the last 466 * superblock update 467 * 2. More than sbi->s_sb_update_kb (def: 16MB) kbs have been written since the 468 * last superblock update. 469 * 470 * @sb: The superblock 471 */ 472 static void ext4_maybe_update_superblock(struct super_block *sb) 473 { 474 struct ext4_sb_info *sbi = EXT4_SB(sb); 475 struct ext4_super_block *es = sbi->s_es; 476 journal_t *journal = sbi->s_journal; 477 time64_t now; 478 __u64 last_update; 479 __u64 lifetime_write_kbytes; 480 __u64 diff_size; 481 482 if (ext4_emergency_state(sb) || sb_rdonly(sb) || 483 !(sb->s_flags & SB_ACTIVE) || !journal || 484 journal->j_flags & JBD2_UNMOUNT) 485 return; 486 487 now = ktime_get_real_seconds(); 488 last_update = ext4_get_tstamp(es, s_wtime); 489 490 if (likely(now - last_update < sbi->s_sb_update_sec)) 491 return; 492 493 lifetime_write_kbytes = sbi->s_kbytes_written + 494 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 495 sbi->s_sectors_written_start) >> 1); 496 497 /* Get the number of kilobytes not written to disk to account 498 * for statistics and compare with a multiple of 16 MB. This 499 * is used to determine when the next superblock commit should 500 * occur (i.e. not more often than once per 16MB if there was 501 * less written in an hour). 502 */ 503 diff_size = lifetime_write_kbytes - le64_to_cpu(es->s_kbytes_written); 504 505 if (diff_size > sbi->s_sb_update_kb) 506 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 507 } 508 509 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 510 { 511 struct super_block *sb = journal->j_private; 512 513 BUG_ON(txn->t_state == T_FINISHED); 514 515 ext4_process_freed_data(sb, txn->t_tid); 516 ext4_maybe_update_superblock(sb); 517 } 518 519 static bool ext4_journalled_writepage_needs_redirty(struct jbd2_inode *jinode, 520 struct folio *folio) 521 { 522 struct buffer_head *bh, *head; 523 struct journal_head *jh; 524 transaction_t *trans = READ_ONCE(jinode->i_transaction); 525 526 bh = head = folio_buffers(folio); 527 do { 528 /* 529 * We have to redirty a page in these cases: 530 * 1) If buffer is dirty, it means the page was dirty because it 531 * contains a buffer that needs checkpointing. So the dirty bit 532 * needs to be preserved so that checkpointing writes the buffer 533 * properly. 534 * 2) If buffer is not part of the committing transaction 535 * (we may have just accidentally come across this buffer because 536 * inode range tracking is not exact) or if the currently running 537 * transaction already contains this buffer as well, dirty bit 538 * needs to be preserved so that the buffer gets writeprotected 539 * properly on running transaction's commit. 540 */ 541 jh = bh2jh(bh); 542 if (buffer_dirty(bh) || 543 (jh && (jh->b_transaction != trans || 544 jh->b_next_transaction))) 545 return true; 546 } while ((bh = bh->b_this_page) != head); 547 548 return false; 549 } 550 551 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 552 { 553 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 554 loff_t range_start, range_end; 555 struct writeback_control wbc = { 556 .sync_mode = WB_SYNC_ALL, 557 .nr_to_write = LONG_MAX, 558 }; 559 struct folio *folio = NULL; 560 int error; 561 562 if (!jbd2_jinode_get_dirty_range(jinode, &range_start, &range_end)) 563 return 0; 564 565 wbc.range_start = range_start; 566 wbc.range_end = range_end; 567 568 /* 569 * writeback_iter() already checks for dirty pages and calls 570 * folio_clear_dirty_for_io(), which we want to write protect the 571 * folios. 572 * 573 * However, we may have to redirty a folio sometimes. 574 */ 575 while ((folio = writeback_iter(mapping, &wbc, folio, &error))) { 576 if (ext4_journalled_writepage_needs_redirty(jinode, folio)) 577 folio_redirty_for_writepage(&wbc, folio); 578 folio_unlock(folio); 579 } 580 581 return error; 582 } 583 584 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 585 { 586 int ret; 587 588 if (ext4_should_journal_data(jinode->i_vfs_inode)) 589 ret = ext4_journalled_submit_inode_data_buffers(jinode); 590 else 591 ret = ext4_normal_submit_inode_data_buffers(jinode); 592 return ret; 593 } 594 595 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 596 { 597 int ret = 0; 598 599 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 600 ret = jbd2_journal_finish_inode_data_buffers(jinode); 601 602 return ret; 603 } 604 605 static bool system_going_down(void) 606 { 607 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 608 || system_state == SYSTEM_RESTART; 609 } 610 611 struct ext4_err_translation { 612 int code; 613 int errno; 614 }; 615 616 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 617 618 static struct ext4_err_translation err_translation[] = { 619 EXT4_ERR_TRANSLATE(EIO), 620 EXT4_ERR_TRANSLATE(ENOMEM), 621 EXT4_ERR_TRANSLATE(EFSBADCRC), 622 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 623 EXT4_ERR_TRANSLATE(ENOSPC), 624 EXT4_ERR_TRANSLATE(ENOKEY), 625 EXT4_ERR_TRANSLATE(EROFS), 626 EXT4_ERR_TRANSLATE(EFBIG), 627 EXT4_ERR_TRANSLATE(EEXIST), 628 EXT4_ERR_TRANSLATE(ERANGE), 629 EXT4_ERR_TRANSLATE(EOVERFLOW), 630 EXT4_ERR_TRANSLATE(EBUSY), 631 EXT4_ERR_TRANSLATE(ENOTDIR), 632 EXT4_ERR_TRANSLATE(ENOTEMPTY), 633 EXT4_ERR_TRANSLATE(ESHUTDOWN), 634 EXT4_ERR_TRANSLATE(EFAULT), 635 }; 636 637 static int ext4_errno_to_code(int errno) 638 { 639 int i; 640 641 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 642 if (err_translation[i].errno == errno) 643 return err_translation[i].code; 644 return EXT4_ERR_UNKNOWN; 645 } 646 647 static void save_error_info(struct super_block *sb, int error, 648 __u32 ino, __u64 block, 649 const char *func, unsigned int line) 650 { 651 struct ext4_sb_info *sbi = EXT4_SB(sb); 652 653 /* We default to EFSCORRUPTED error... */ 654 if (error == 0) 655 error = EFSCORRUPTED; 656 657 spin_lock(&sbi->s_error_lock); 658 sbi->s_add_error_count++; 659 sbi->s_last_error_code = error; 660 sbi->s_last_error_line = line; 661 sbi->s_last_error_ino = ino; 662 sbi->s_last_error_block = block; 663 sbi->s_last_error_func = func; 664 sbi->s_last_error_time = ktime_get_real_seconds(); 665 if (!sbi->s_first_error_time) { 666 sbi->s_first_error_code = error; 667 sbi->s_first_error_line = line; 668 sbi->s_first_error_ino = ino; 669 sbi->s_first_error_block = block; 670 sbi->s_first_error_func = func; 671 sbi->s_first_error_time = sbi->s_last_error_time; 672 } 673 spin_unlock(&sbi->s_error_lock); 674 } 675 676 /* Deal with the reporting of failure conditions on a filesystem such as 677 * inconsistencies detected or read IO failures. 678 * 679 * On ext2, we can store the error state of the filesystem in the 680 * superblock. That is not possible on ext4, because we may have other 681 * write ordering constraints on the superblock which prevent us from 682 * writing it out straight away; and given that the journal is about to 683 * be aborted, we can't rely on the current, or future, transactions to 684 * write out the superblock safely. 685 * 686 * We'll just use the jbd2_journal_abort() error code to record an error in 687 * the journal instead. On recovery, the journal will complain about 688 * that error until we've noted it down and cleared it. 689 * 690 * If force_ro is set, we unconditionally force the filesystem into an 691 * ABORT|READONLY state, unless the error response on the fs has been set to 692 * panic in which case we take the easy way out and panic immediately. This is 693 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 694 * at a critical moment in log management. 695 */ 696 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 697 __u32 ino, __u64 block, 698 const char *func, unsigned int line) 699 { 700 journal_t *journal = EXT4_SB(sb)->s_journal; 701 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 702 703 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 704 if (test_opt(sb, WARN_ON_ERROR)) 705 WARN_ON_ONCE(1); 706 707 if (!continue_fs && !ext4_emergency_ro(sb) && journal) 708 jbd2_journal_abort(journal, -error); 709 710 if (!bdev_read_only(sb->s_bdev)) { 711 save_error_info(sb, error, ino, block, func, line); 712 /* 713 * In case the fs should keep running, we need to writeout 714 * superblock through the journal. Due to lock ordering 715 * constraints, it may not be safe to do it right here so we 716 * defer superblock flushing to a workqueue. We just need to be 717 * careful when the journal is already shutting down. If we get 718 * here in that case, just update the sb directly as the last 719 * transaction won't commit anyway. 720 */ 721 if (continue_fs && journal && 722 !ext4_test_mount_flag(sb, EXT4_MF_JOURNAL_DESTROY)) 723 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 724 else 725 ext4_commit_super(sb); 726 } 727 728 /* 729 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 730 * could panic during 'reboot -f' as the underlying device got already 731 * disabled. 732 */ 733 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 734 panic("EXT4-fs (device %s): panic forced after error\n", 735 sb->s_id); 736 } 737 738 if (ext4_emergency_ro(sb) || continue_fs) 739 return; 740 741 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 742 /* 743 * We don't set SB_RDONLY because that requires sb->s_umount 744 * semaphore and setting it without proper remount procedure is 745 * confusing code such as freeze_super() leading to deadlocks 746 * and other problems. 747 */ 748 set_bit(EXT4_FLAGS_EMERGENCY_RO, &EXT4_SB(sb)->s_ext4_flags); 749 } 750 751 static void update_super_work(struct work_struct *work) 752 { 753 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 754 s_sb_upd_work); 755 journal_t *journal = sbi->s_journal; 756 handle_t *handle; 757 758 /* 759 * If the journal is still running, we have to write out superblock 760 * through the journal to avoid collisions of other journalled sb 761 * updates. 762 * 763 * We use directly jbd2 functions here to avoid recursing back into 764 * ext4 error handling code during handling of previous errors. 765 */ 766 if (!ext4_emergency_state(sbi->s_sb) && 767 !sb_rdonly(sbi->s_sb) && journal) { 768 struct buffer_head *sbh = sbi->s_sbh; 769 bool call_notify_err = false; 770 771 handle = jbd2_journal_start(journal, 1); 772 if (IS_ERR(handle)) 773 goto write_directly; 774 if (jbd2_journal_get_write_access(handle, sbh)) { 775 jbd2_journal_stop(handle); 776 goto write_directly; 777 } 778 779 if (sbi->s_add_error_count > 0) 780 call_notify_err = true; 781 782 ext4_update_super(sbi->s_sb); 783 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 784 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 785 "superblock detected"); 786 clear_buffer_write_io_error(sbh); 787 set_buffer_uptodate(sbh); 788 } 789 790 if (jbd2_journal_dirty_metadata(handle, sbh)) { 791 jbd2_journal_stop(handle); 792 goto write_directly; 793 } 794 jbd2_journal_stop(handle); 795 796 if (call_notify_err) 797 ext4_notify_error_sysfs(sbi); 798 799 return; 800 } 801 write_directly: 802 /* 803 * Write through journal failed. Write sb directly to get error info 804 * out and hope for the best. 805 */ 806 ext4_commit_super(sbi->s_sb); 807 ext4_notify_error_sysfs(sbi); 808 } 809 810 #define ext4_error_ratelimit(sb) \ 811 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 812 "EXT4-fs error") 813 814 void __ext4_error(struct super_block *sb, const char *function, 815 unsigned int line, bool force_ro, int error, __u64 block, 816 const char *fmt, ...) 817 { 818 struct va_format vaf; 819 va_list args; 820 821 if (unlikely(ext4_emergency_state(sb))) 822 return; 823 824 trace_ext4_error(sb, function, line); 825 if (ext4_error_ratelimit(sb)) { 826 va_start(args, fmt); 827 vaf.fmt = fmt; 828 vaf.va = &args; 829 printk(KERN_CRIT 830 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 831 sb->s_id, function, line, current->comm, &vaf); 832 va_end(args); 833 } 834 fserror_report_metadata(sb, error ? -abs(error) : -EFSCORRUPTED, 835 GFP_ATOMIC); 836 837 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 838 } 839 840 void __ext4_error_inode(struct inode *inode, const char *function, 841 unsigned int line, ext4_fsblk_t block, int error, 842 const char *fmt, ...) 843 { 844 va_list args; 845 struct va_format vaf; 846 847 if (unlikely(ext4_emergency_state(inode->i_sb))) 848 return; 849 850 trace_ext4_error(inode->i_sb, function, line); 851 if (ext4_error_ratelimit(inode->i_sb)) { 852 va_start(args, fmt); 853 vaf.fmt = fmt; 854 vaf.va = &args; 855 if (block) 856 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 857 "inode #%llu: block %llu: comm %s: %pV\n", 858 inode->i_sb->s_id, function, line, inode->i_ino, 859 block, current->comm, &vaf); 860 else 861 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 862 "inode #%llu: comm %s: %pV\n", 863 inode->i_sb->s_id, function, line, inode->i_ino, 864 current->comm, &vaf); 865 va_end(args); 866 } 867 fserror_report_file_metadata(inode, 868 error ? -abs(error) : -EFSCORRUPTED, 869 GFP_ATOMIC); 870 871 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 872 function, line); 873 } 874 875 void __ext4_error_file(struct file *file, const char *function, 876 unsigned int line, ext4_fsblk_t block, 877 const char *fmt, ...) 878 { 879 va_list args; 880 struct va_format vaf; 881 struct inode *inode = file_inode(file); 882 char pathname[80], *path; 883 884 if (unlikely(ext4_emergency_state(inode->i_sb))) 885 return; 886 887 trace_ext4_error(inode->i_sb, function, line); 888 if (ext4_error_ratelimit(inode->i_sb)) { 889 path = file_path(file, pathname, sizeof(pathname)); 890 if (IS_ERR(path)) 891 path = "(unknown)"; 892 va_start(args, fmt); 893 vaf.fmt = fmt; 894 vaf.va = &args; 895 if (block) 896 printk(KERN_CRIT 897 "EXT4-fs error (device %s): %s:%d: inode #%llu: " 898 "block %llu: comm %s: path %s: %pV\n", 899 inode->i_sb->s_id, function, line, inode->i_ino, 900 block, current->comm, path, &vaf); 901 else 902 printk(KERN_CRIT 903 "EXT4-fs error (device %s): %s:%d: inode #%llu: " 904 "comm %s: path %s: %pV\n", 905 inode->i_sb->s_id, function, line, inode->i_ino, 906 current->comm, path, &vaf); 907 va_end(args); 908 } 909 fserror_report_file_metadata(inode, -EFSCORRUPTED, GFP_ATOMIC); 910 911 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 912 function, line); 913 } 914 915 const char *ext4_decode_error(struct super_block *sb, int errno, 916 char nbuf[16]) 917 { 918 char *errstr = NULL; 919 920 switch (errno) { 921 case -EFSCORRUPTED: 922 errstr = "Corrupt filesystem"; 923 break; 924 case -EFSBADCRC: 925 errstr = "Filesystem failed CRC"; 926 break; 927 case -EIO: 928 errstr = "IO failure"; 929 break; 930 case -ENOMEM: 931 errstr = "Out of memory"; 932 break; 933 case -EROFS: 934 if (!sb || (EXT4_SB(sb)->s_journal && 935 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 936 errstr = "Journal has aborted"; 937 else 938 errstr = "Readonly filesystem"; 939 break; 940 default: 941 /* If the caller passed in an extra buffer for unknown 942 * errors, textualise them now. Else we just return 943 * NULL. */ 944 if (nbuf) { 945 /* Check for truncated error codes... */ 946 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 947 errstr = nbuf; 948 } 949 break; 950 } 951 952 return errstr; 953 } 954 955 /* __ext4_std_error decodes expected errors from journaling functions 956 * automatically and invokes the appropriate error response. */ 957 958 void __ext4_std_error(struct super_block *sb, const char *function, 959 unsigned int line, int errno) 960 { 961 char nbuf[16]; 962 const char *errstr; 963 964 if (unlikely(ext4_emergency_state(sb))) 965 return; 966 967 /* Special case: if the error is EROFS, and we're not already 968 * inside a transaction, then there's really no point in logging 969 * an error. */ 970 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 971 return; 972 973 if (ext4_error_ratelimit(sb)) { 974 errstr = ext4_decode_error(sb, errno, nbuf); 975 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 976 sb->s_id, function, line, errstr); 977 } 978 fserror_report_metadata(sb, errno ? -abs(errno) : -EFSCORRUPTED, 979 GFP_ATOMIC); 980 981 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 982 } 983 984 void __ext4_msg(struct super_block *sb, 985 const char *prefix, const char *fmt, ...) 986 { 987 struct va_format vaf; 988 va_list args; 989 990 if (sb) { 991 atomic_inc(&EXT4_SB(sb)->s_msg_count); 992 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 993 "EXT4-fs")) 994 return; 995 } 996 997 va_start(args, fmt); 998 vaf.fmt = fmt; 999 vaf.va = &args; 1000 if (sb) 1001 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 1002 else 1003 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 1004 va_end(args); 1005 } 1006 1007 static int ext4_warning_ratelimit(struct super_block *sb) 1008 { 1009 atomic_inc(&EXT4_SB(sb)->s_warning_count); 1010 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 1011 "EXT4-fs warning"); 1012 } 1013 1014 void __ext4_warning(struct super_block *sb, const char *function, 1015 unsigned int line, const char *fmt, ...) 1016 { 1017 struct va_format vaf; 1018 va_list args; 1019 1020 if (!ext4_warning_ratelimit(sb)) 1021 return; 1022 1023 va_start(args, fmt); 1024 vaf.fmt = fmt; 1025 vaf.va = &args; 1026 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 1027 sb->s_id, function, line, &vaf); 1028 va_end(args); 1029 } 1030 1031 void __ext4_warning_inode(const struct inode *inode, const char *function, 1032 unsigned int line, const char *fmt, ...) 1033 { 1034 struct va_format vaf; 1035 va_list args; 1036 1037 if (!ext4_warning_ratelimit(inode->i_sb)) 1038 return; 1039 1040 va_start(args, fmt); 1041 vaf.fmt = fmt; 1042 vaf.va = &args; 1043 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 1044 "inode #%llu: comm %s: %pV\n", inode->i_sb->s_id, 1045 function, line, inode->i_ino, current->comm, &vaf); 1046 va_end(args); 1047 } 1048 1049 void __ext4_grp_locked_error(const char *function, unsigned int line, 1050 struct super_block *sb, ext4_group_t grp, 1051 u64 ino, ext4_fsblk_t block, 1052 const char *fmt, ...) 1053 __releases(bitlock) 1054 __acquires(bitlock) 1055 { 1056 struct va_format vaf; 1057 va_list args; 1058 1059 if (unlikely(ext4_emergency_state(sb))) 1060 return; 1061 1062 trace_ext4_error(sb, function, line); 1063 if (ext4_error_ratelimit(sb)) { 1064 va_start(args, fmt); 1065 vaf.fmt = fmt; 1066 vaf.va = &args; 1067 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1068 sb->s_id, function, line, grp); 1069 if (ino) 1070 printk(KERN_CONT "inode %llu: ", ino); 1071 if (block) 1072 printk(KERN_CONT "block %llu:", 1073 (unsigned long long) block); 1074 printk(KERN_CONT "%pV\n", &vaf); 1075 va_end(args); 1076 } 1077 1078 if (test_opt(sb, ERRORS_CONT)) { 1079 if (test_opt(sb, WARN_ON_ERROR)) 1080 WARN_ON_ONCE(1); 1081 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1082 if (!bdev_read_only(sb->s_bdev)) { 1083 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1084 line); 1085 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 1086 } 1087 return; 1088 } 1089 ext4_unlock_group(sb, grp); 1090 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1091 /* 1092 * We only get here in the ERRORS_RO case; relocking the group 1093 * may be dangerous, but nothing bad will happen since the 1094 * filesystem will have already been marked read/only and the 1095 * journal has been aborted. We return 1 as a hint to callers 1096 * who might what to use the return value from 1097 * ext4_grp_locked_error() to distinguish between the 1098 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1099 * aggressively from the ext4 function in question, with a 1100 * more appropriate error code. 1101 */ 1102 ext4_lock_group(sb, grp); 1103 return; 1104 } 1105 1106 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1107 ext4_group_t group, 1108 unsigned int flags) 1109 { 1110 struct ext4_sb_info *sbi = EXT4_SB(sb); 1111 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1112 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1113 int ret; 1114 1115 if (!grp || !gdp) 1116 return; 1117 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1118 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1119 &grp->bb_state); 1120 if (!ret) 1121 percpu_counter_sub(&sbi->s_freeclusters_counter, 1122 grp->bb_free); 1123 } 1124 1125 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1126 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1127 &grp->bb_state); 1128 if (!ret && gdp) { 1129 int count; 1130 1131 count = ext4_free_inodes_count(sb, gdp); 1132 percpu_counter_sub(&sbi->s_freeinodes_counter, 1133 count); 1134 } 1135 } 1136 } 1137 1138 void ext4_update_dynamic_rev(struct super_block *sb) 1139 { 1140 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1141 1142 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1143 return; 1144 1145 ext4_warning(sb, 1146 "updating to rev %d because of new feature flag, " 1147 "running e2fsck is recommended", 1148 EXT4_DYNAMIC_REV); 1149 1150 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1151 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1152 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1153 /* leave es->s_feature_*compat flags alone */ 1154 /* es->s_uuid will be set by e2fsck if empty */ 1155 1156 /* 1157 * The rest of the superblock fields should be zero, and if not it 1158 * means they are likely already in use, so leave them alone. We 1159 * can leave it up to e2fsck to clean up any inconsistencies there. 1160 */ 1161 } 1162 1163 static inline struct inode *orphan_list_entry(struct list_head *l) 1164 { 1165 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1166 } 1167 1168 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1169 { 1170 struct list_head *l; 1171 1172 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1173 le32_to_cpu(sbi->s_es->s_last_orphan)); 1174 1175 printk(KERN_ERR "sb_info orphan list:\n"); 1176 list_for_each(l, &sbi->s_orphan) { 1177 struct inode *inode = orphan_list_entry(l); 1178 printk(KERN_ERR " " 1179 "inode %s:%llu at %p: mode %o, nlink %d, next %d\n", 1180 inode->i_sb->s_id, inode->i_ino, inode, 1181 inode->i_mode, inode->i_nlink, 1182 NEXT_ORPHAN(inode)); 1183 } 1184 } 1185 1186 #ifdef CONFIG_QUOTA 1187 static int ext4_quota_off(struct super_block *sb, int type); 1188 1189 static inline void ext4_quotas_off(struct super_block *sb, int type) 1190 { 1191 BUG_ON(type > EXT4_MAXQUOTAS); 1192 1193 /* Use our quota_off function to clear inode flags etc. */ 1194 for (type--; type >= 0; type--) 1195 ext4_quota_off(sb, type); 1196 } 1197 1198 /* 1199 * This is a helper function which is used in the mount/remount 1200 * codepaths (which holds s_umount) to fetch the quota file name. 1201 */ 1202 static inline char *get_qf_name(struct super_block *sb, 1203 struct ext4_sb_info *sbi, 1204 int type) 1205 { 1206 return rcu_dereference_protected(sbi->s_qf_names[type], 1207 lockdep_is_held(&sb->s_umount)); 1208 } 1209 #else 1210 static inline void ext4_quotas_off(struct super_block *sb, int type) 1211 { 1212 } 1213 #endif 1214 1215 static int ext4_percpu_param_init(struct ext4_sb_info *sbi) 1216 { 1217 ext4_fsblk_t block; 1218 int err; 1219 1220 block = ext4_count_free_clusters(sbi->s_sb); 1221 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block)); 1222 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 1223 GFP_KERNEL); 1224 if (!err) { 1225 unsigned long freei = ext4_count_free_inodes(sbi->s_sb); 1226 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 1227 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 1228 GFP_KERNEL); 1229 } 1230 if (!err) 1231 err = percpu_counter_init(&sbi->s_dirs_counter, 1232 ext4_count_dirs(sbi->s_sb), GFP_KERNEL); 1233 if (!err) 1234 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 1235 GFP_KERNEL); 1236 if (!err) 1237 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 1238 GFP_KERNEL); 1239 if (!err) 1240 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 1241 1242 if (err) 1243 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory"); 1244 1245 return err; 1246 } 1247 1248 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi) 1249 { 1250 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1251 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1252 percpu_counter_destroy(&sbi->s_dirs_counter); 1253 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1254 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1255 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1256 } 1257 1258 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 1259 { 1260 struct buffer_head **group_desc; 1261 int i; 1262 1263 group_desc = rcu_access_pointer(sbi->s_group_desc); 1264 for (i = 0; i < sbi->s_gdb_count; i++) 1265 brelse(group_desc[i]); 1266 kvfree(group_desc); 1267 } 1268 1269 static void ext4_flex_groups_free(struct ext4_sb_info *sbi) 1270 { 1271 struct flex_groups **flex_groups; 1272 int i; 1273 1274 flex_groups = rcu_access_pointer(sbi->s_flex_groups); 1275 if (flex_groups) { 1276 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1277 kvfree(flex_groups[i]); 1278 kvfree(flex_groups); 1279 } 1280 } 1281 1282 static void ext4_put_super(struct super_block *sb) 1283 { 1284 struct ext4_sb_info *sbi = EXT4_SB(sb); 1285 struct ext4_super_block *es = sbi->s_es; 1286 int aborted = 0; 1287 int err; 1288 1289 /* 1290 * Unregister sysfs before destroying jbd2 journal. 1291 * Since we could still access attr_journal_task attribute via sysfs 1292 * path which could have sbi->s_journal->j_task as NULL 1293 * Unregister sysfs before flush sbi->s_sb_upd_work. 1294 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1295 * read metadata verify failed then will queue error work. 1296 * update_super_work will call start_this_handle may trigger 1297 * BUG_ON. 1298 */ 1299 ext4_unregister_sysfs(sb); 1300 1301 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1302 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.", 1303 &sb->s_uuid); 1304 1305 ext4_unregister_li_request(sb); 1306 /* Drain deferred EA inode iputs while quota is still active. */ 1307 flush_delayed_work(&sbi->s_ea_inode_work); 1308 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 1309 1310 destroy_workqueue(sbi->rsv_conversion_wq); 1311 ext4_release_orphan_info(sb); 1312 1313 if (sbi->s_journal) { 1314 aborted = is_journal_aborted(sbi->s_journal); 1315 err = ext4_journal_destroy(sbi, sbi->s_journal); 1316 if ((err < 0) && !aborted) { 1317 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1318 } 1319 } else 1320 flush_work(&sbi->s_sb_upd_work); 1321 1322 ext4_es_unregister_shrinker(sbi); 1323 timer_shutdown_sync(&sbi->s_err_report); 1324 ext4_release_system_zone(sb); 1325 ext4_mb_release(sb); 1326 ext4_ext_release(sb); 1327 1328 if (!ext4_emergency_state(sb) && !sb_rdonly(sb)) { 1329 if (!aborted) { 1330 ext4_clear_feature_journal_needs_recovery(sb); 1331 ext4_clear_feature_orphan_present(sb); 1332 es->s_state = cpu_to_le16(sbi->s_mount_state); 1333 } 1334 ext4_commit_super(sb); 1335 } 1336 1337 ext4_group_desc_free(sbi); 1338 ext4_flex_groups_free(sbi); 1339 1340 WARN_ON_ONCE(!(sbi->s_mount_state & EXT4_ERROR_FS) && 1341 percpu_counter_sum(&sbi->s_dirtyclusters_counter)); 1342 ext4_percpu_param_destroy(sbi); 1343 #ifdef CONFIG_QUOTA 1344 for (int i = 0; i < EXT4_MAXQUOTAS; i++) 1345 kfree(get_qf_name(sb, sbi, i)); 1346 #endif 1347 1348 /* Debugging code just in case the in-memory inode orphan list 1349 * isn't empty. The on-disk one can be non-empty if we've 1350 * detected an error and taken the fs readonly, but the 1351 * in-memory list had better be clean by this point. */ 1352 if (!list_empty(&sbi->s_orphan)) 1353 dump_orphan_list(sb, sbi); 1354 ASSERT(list_empty(&sbi->s_orphan)); 1355 1356 sync_blockdev(sb->s_bdev); 1357 invalidate_bdev(sb->s_bdev); 1358 if (sbi->s_journal_bdev_file) { 1359 /* 1360 * Invalidate the journal device's buffers. We don't want them 1361 * floating about in memory - the physical journal device may 1362 * hotswapped, and it breaks the `ro-after' testing code. 1363 */ 1364 sync_blockdev(file_bdev(sbi->s_journal_bdev_file)); 1365 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 1366 } 1367 1368 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1369 sbi->s_ea_inode_cache = NULL; 1370 1371 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1372 sbi->s_ea_block_cache = NULL; 1373 1374 ext4_stop_mmpd(sbi); 1375 1376 brelse(sbi->s_sbh); 1377 sb->s_fs_info = NULL; 1378 /* 1379 * Now that we are completely done shutting down the 1380 * superblock, we need to actually destroy the kobject. 1381 */ 1382 kobject_put(&sbi->s_kobj); 1383 wait_for_completion(&sbi->s_kobj_unregister); 1384 kfree(sbi->s_blockgroup_lock); 1385 fs_put_dax(sbi->s_daxdev, NULL); 1386 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1387 #if IS_ENABLED(CONFIG_UNICODE) 1388 utf8_unload(sb->s_encoding); 1389 #endif 1390 kfree(sbi); 1391 } 1392 1393 static struct kmem_cache *ext4_inode_cachep; 1394 1395 /* 1396 * Called inside transaction, so use GFP_NOFS 1397 */ 1398 static struct inode *ext4_alloc_inode(struct super_block *sb) 1399 { 1400 struct ext4_inode_info *ei; 1401 1402 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1403 if (!ei) 1404 return NULL; 1405 1406 inode_set_iversion(&ei->vfs_inode, 1); 1407 ei->i_flags = 0; 1408 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ 1409 spin_lock_init(&ei->i_raw_lock); 1410 ei->i_prealloc_node = RB_ROOT; 1411 atomic_set(&ei->i_prealloc_active, 0); 1412 rwlock_init(&ei->i_prealloc_lock); 1413 ext4_es_init_tree(&ei->i_es_tree); 1414 rwlock_init(&ei->i_es_lock); 1415 INIT_LIST_HEAD(&ei->i_es_list); 1416 ei->i_es_all_nr = 0; 1417 ei->i_es_shk_nr = 0; 1418 ei->i_es_shrink_lblk = 0; 1419 ei->i_es_seq = 0; 1420 ei->i_reserved_data_blocks = 0; 1421 spin_lock_init(&(ei->i_block_reservation_lock)); 1422 ext4_init_pending_tree(&ei->i_pending_tree); 1423 #ifdef CONFIG_QUOTA 1424 ei->i_reserved_quota = 0; 1425 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1426 #endif 1427 ei->jinode = NULL; 1428 /* 1429 * Reinitialize xattr_sem every allocation because EA inodes 1430 * share this space with i_ea_iput_node (via union) which may 1431 * have overwritten the semaphore when the slab object was 1432 * previously used as an EA inode. 1433 */ 1434 init_rwsem(&ei->xattr_sem); 1435 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1436 spin_lock_init(&ei->i_completed_io_lock); 1437 ei->i_sync_tid = 0; 1438 ei->i_datasync_tid = 0; 1439 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1440 ext4_fc_init_inode(&ei->vfs_inode); 1441 spin_lock_init(&ei->i_fc_lock); 1442 ei->i_metadata_bhs = NULL; 1443 #ifdef CONFIG_LOCKDEP 1444 lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_NORMAL); 1445 #endif 1446 return &ei->vfs_inode; 1447 } 1448 1449 static int ext4_drop_inode(struct inode *inode) 1450 { 1451 int drop = inode_generic_drop(inode); 1452 1453 if (!drop) 1454 drop = fscrypt_drop_inode(inode); 1455 1456 trace_ext4_drop_inode(inode, drop); 1457 return drop; 1458 } 1459 1460 static void ext4_free_in_core_inode(struct inode *inode) 1461 { 1462 fscrypt_free_inode(inode); 1463 kfree(ext4_i_metadata_bhs(inode)); 1464 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1465 pr_warn("%s: inode %llu still in fc list", 1466 __func__, inode->i_ino); 1467 } 1468 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1469 } 1470 1471 static void ext4_destroy_inode(struct inode *inode) 1472 { 1473 if (ext4_inode_orphan_tracked(inode)) { 1474 ext4_msg(inode->i_sb, KERN_ERR, 1475 "Inode %llu (%p): inode tracked as orphan!", 1476 inode->i_ino, EXT4_I(inode)); 1477 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1478 EXT4_I(inode), sizeof(struct ext4_inode_info), 1479 true); 1480 dump_stack(); 1481 } 1482 1483 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ERROR_FS) && 1484 WARN_ON_ONCE(EXT4_I(inode)->i_reserved_data_blocks)) 1485 ext4_msg(inode->i_sb, KERN_ERR, 1486 "Inode %llu (%p): i_reserved_data_blocks (%u) not cleared!", 1487 inode->i_ino, EXT4_I(inode), 1488 EXT4_I(inode)->i_reserved_data_blocks); 1489 } 1490 1491 static void ext4_shutdown(struct super_block *sb) 1492 { 1493 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH); 1494 } 1495 1496 static void init_once(void *foo) 1497 { 1498 struct ext4_inode_info *ei = foo; 1499 1500 INIT_LIST_HEAD(&ei->i_orphan); 1501 init_rwsem(&ei->i_data_sem); 1502 inode_init_once(&ei->vfs_inode); 1503 ext4_fc_init_inode(&ei->vfs_inode); 1504 #ifdef CONFIG_FS_ENCRYPTION 1505 ei->i_crypt_info = NULL; 1506 #endif 1507 } 1508 1509 static int __init init_inodecache(void) 1510 { 1511 struct kmem_cache_args args = { 1512 .useroffset = offsetof(struct ext4_inode_info, i_data), 1513 .usersize = sizeof_field(struct ext4_inode_info, i_data), 1514 .use_freeptr_offset = true, 1515 .freeptr_offset = offsetof(struct ext4_inode_info, i_flags), 1516 .ctor = init_once, 1517 }; 1518 1519 ext4_inode_cachep = kmem_cache_create("ext4_inode_cache", 1520 sizeof(struct ext4_inode_info), 1521 &args, 1522 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT); 1523 1524 if (ext4_inode_cachep == NULL) 1525 return -ENOMEM; 1526 return 0; 1527 } 1528 1529 static void destroy_inodecache(void) 1530 { 1531 /* 1532 * Make sure all delayed rcu free inodes are flushed before we 1533 * destroy cache. 1534 */ 1535 rcu_barrier(); 1536 kmem_cache_destroy(ext4_inode_cachep); 1537 } 1538 1539 void ext4_clear_inode(struct inode *inode) 1540 { 1541 struct mapping_metadata_bhs *mmb = ext4_i_metadata_bhs(inode); 1542 1543 ext4_fc_del(inode); 1544 if (mmb) 1545 mmb_invalidate(mmb); 1546 clear_inode(inode); 1547 ext4_discard_preallocations(inode); 1548 /* 1549 * We must remove the inode from the hash before ext4_free_inode() 1550 * clears the bit in inode bitmap as otherwise another process reusing 1551 * the inode will block in insert_inode_hash() waiting for inode 1552 * eviction to complete while holding transaction handle open, but 1553 * ext4_evict_inode() still running for that inode could block waiting 1554 * for transaction commit if the inode is marked as IS_SYNC => deadlock. 1555 * 1556 * Removing the inode from the hash here is safe. There are two cases 1557 * to consider: 1558 * 1) The inode still has references to it (i_nlink > 0). In that case 1559 * we are keeping the inode and once we remove the inode from the hash, 1560 * iget() can create the new inode structure for the same inode number 1561 * and we are fine with that as all IO on behalf of the inode is 1562 * finished. 1563 * 2) We are deleting the inode (i_nlink == 0). In that case inode 1564 * number cannot be reused until ext4_free_inode() clears the bit in 1565 * the inode bitmap, at which point all IO is done and reuse is fine 1566 * again. 1567 */ 1568 remove_inode_hash(inode); 1569 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1570 dquot_drop(inode); 1571 if (EXT4_I(inode)->jinode) { 1572 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1573 EXT4_I(inode)->jinode); 1574 jbd2_free_inode(EXT4_I(inode)->jinode); 1575 EXT4_I(inode)->jinode = NULL; 1576 } 1577 fscrypt_put_encryption_info(inode); 1578 } 1579 1580 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1581 u64 ino, u32 generation) 1582 { 1583 struct inode *inode; 1584 1585 /* 1586 * Currently we don't know the generation for parent directory, so 1587 * a generation of 0 means "accept any" 1588 */ 1589 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1590 if (IS_ERR(inode)) 1591 return ERR_CAST(inode); 1592 if (generation && inode->i_generation != generation) { 1593 iput(inode); 1594 return ERR_PTR(-ESTALE); 1595 } 1596 1597 return inode; 1598 } 1599 1600 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1601 int fh_len, int fh_type) 1602 { 1603 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1604 ext4_nfs_get_inode); 1605 } 1606 1607 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1608 int fh_len, int fh_type) 1609 { 1610 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1611 ext4_nfs_get_inode); 1612 } 1613 1614 static int ext4_nfs_commit_metadata(struct inode *inode) 1615 { 1616 struct writeback_control wbc = { 1617 .sync_mode = WB_SYNC_ALL 1618 }; 1619 int ret; 1620 1621 trace_ext4_nfs_commit_metadata(inode); 1622 ret = ext4_write_inode(inode, &wbc); 1623 if (!ret && inode_state_read_once(inode) & I_METADATA_WRITEBACK) 1624 ret = ext4_sync_inode_metadata(inode, &wbc); 1625 return ret; 1626 } 1627 1628 #ifdef CONFIG_QUOTA 1629 static const char * const quotatypes[] = INITQFNAMES; 1630 #define QTYPE2NAME(t) (quotatypes[t]) 1631 1632 static int ext4_write_dquot(struct dquot *dquot); 1633 static int ext4_acquire_dquot(struct dquot *dquot); 1634 static int ext4_release_dquot(struct dquot *dquot); 1635 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1636 static int ext4_write_info(struct super_block *sb, int type); 1637 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1638 const struct path *path); 1639 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1640 size_t len, loff_t off); 1641 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1642 const char *data, size_t len, loff_t off); 1643 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1644 unsigned int flags); 1645 1646 static struct dquot __rcu **ext4_get_dquots(struct inode *inode) 1647 { 1648 return EXT4_I(inode)->i_dquot; 1649 } 1650 1651 static const struct dquot_operations ext4_quota_operations = { 1652 .get_reserved_space = ext4_get_reserved_space, 1653 .write_dquot = ext4_write_dquot, 1654 .acquire_dquot = ext4_acquire_dquot, 1655 .release_dquot = ext4_release_dquot, 1656 .mark_dirty = ext4_mark_dquot_dirty, 1657 .write_info = ext4_write_info, 1658 .alloc_dquot = dquot_alloc, 1659 .destroy_dquot = dquot_destroy, 1660 .get_projid = ext4_get_projid, 1661 .get_inode_usage = ext4_get_inode_usage, 1662 .get_next_id = dquot_get_next_id, 1663 }; 1664 1665 static const struct quotactl_ops ext4_qctl_operations = { 1666 .quota_on = ext4_quota_on, 1667 .quota_off = ext4_quota_off, 1668 .quota_sync = dquot_quota_sync, 1669 .get_state = dquot_get_state, 1670 .set_info = dquot_set_dqinfo, 1671 .get_dqblk = dquot_get_dqblk, 1672 .set_dqblk = dquot_set_dqblk, 1673 .get_nextdqblk = dquot_get_next_dqblk, 1674 }; 1675 #endif 1676 1677 static const struct super_operations ext4_sops = { 1678 .alloc_inode = ext4_alloc_inode, 1679 .free_inode = ext4_free_in_core_inode, 1680 .destroy_inode = ext4_destroy_inode, 1681 .write_inode = ext4_write_inode, 1682 .sync_inode_metadata = ext4_sync_inode_metadata, 1683 .dirty_inode = ext4_dirty_inode, 1684 .drop_inode = ext4_drop_inode, 1685 .evict_inode = ext4_evict_inode, 1686 .put_super = ext4_put_super, 1687 .sync_fs = ext4_sync_fs, 1688 .freeze_fs = ext4_freeze, 1689 .unfreeze_fs = ext4_unfreeze, 1690 .statfs = ext4_statfs, 1691 .show_options = ext4_show_options, 1692 .shutdown = ext4_shutdown, 1693 #ifdef CONFIG_QUOTA 1694 .quota_read = ext4_quota_read, 1695 .quota_write = ext4_quota_write, 1696 .get_dquots = ext4_get_dquots, 1697 #endif 1698 }; 1699 1700 static const struct export_operations ext4_export_ops = { 1701 .encode_fh = generic_encode_ino32_fh, 1702 .fh_to_dentry = ext4_fh_to_dentry, 1703 .fh_to_parent = ext4_fh_to_parent, 1704 .get_parent = ext4_get_parent, 1705 .commit_metadata = ext4_nfs_commit_metadata, 1706 }; 1707 1708 enum { 1709 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1710 Opt_resgid, Opt_resuid, Opt_sb, 1711 Opt_nouid32, Opt_debug, Opt_removed, 1712 Opt_user_xattr, Opt_acl, 1713 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1714 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1715 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1716 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1717 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1718 Opt_inlinecrypt, 1719 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1720 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1721 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1722 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1723 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1724 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1725 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1726 Opt_inode_readahead_blks, Opt_journal_ioprio, 1727 Opt_dioread_nolock, Opt_dioread_lock, 1728 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1729 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1730 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1731 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1732 #ifdef CONFIG_EXT4_DEBUG 1733 Opt_fc_debug_max_replay, Opt_fc_debug_force 1734 #endif 1735 }; 1736 1737 static const struct constant_table ext4_param_errors[] = { 1738 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1739 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1740 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1741 {} 1742 }; 1743 1744 static const struct constant_table ext4_param_data[] = { 1745 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1746 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1747 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1748 {} 1749 }; 1750 1751 static const struct constant_table ext4_param_data_err[] = { 1752 {"abort", Opt_data_err_abort}, 1753 {"ignore", Opt_data_err_ignore}, 1754 {} 1755 }; 1756 1757 static const struct constant_table ext4_param_jqfmt[] = { 1758 {"vfsold", QFMT_VFS_OLD}, 1759 {"vfsv0", QFMT_VFS_V0}, 1760 {"vfsv1", QFMT_VFS_V1}, 1761 {} 1762 }; 1763 1764 static const struct constant_table ext4_param_dax[] = { 1765 {"always", Opt_dax_always}, 1766 {"inode", Opt_dax_inode}, 1767 {"never", Opt_dax_never}, 1768 {} 1769 }; 1770 1771 /* 1772 * Mount option specification 1773 * We don't use fsparam_flag_no because of the way we set the 1774 * options and the way we show them in _ext4_show_options(). To 1775 * keep the changes to a minimum, let's keep the negative options 1776 * separate for now. 1777 */ 1778 static const struct fs_parameter_spec ext4_param_specs[] = { 1779 fsparam_flag ("bsddf", Opt_bsd_df), 1780 fsparam_flag ("minixdf", Opt_minix_df), 1781 fsparam_flag ("grpid", Opt_grpid), 1782 fsparam_flag ("bsdgroups", Opt_grpid), 1783 fsparam_flag ("nogrpid", Opt_nogrpid), 1784 fsparam_flag ("sysvgroups", Opt_nogrpid), 1785 fsparam_gid ("resgid", Opt_resgid), 1786 fsparam_uid ("resuid", Opt_resuid), 1787 fsparam_u32 ("sb", Opt_sb), 1788 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1789 fsparam_flag ("nouid32", Opt_nouid32), 1790 fsparam_flag ("debug", Opt_debug), 1791 fsparam_flag ("oldalloc", Opt_removed), 1792 fsparam_flag ("orlov", Opt_removed), 1793 fsparam_flag ("user_xattr", Opt_user_xattr), 1794 fsparam_flag ("acl", Opt_acl), 1795 fsparam_flag ("norecovery", Opt_noload), 1796 fsparam_flag ("noload", Opt_noload), 1797 fsparam_flag ("bh", Opt_removed), 1798 fsparam_flag ("nobh", Opt_removed), 1799 fsparam_u32 ("commit", Opt_commit), 1800 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1801 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1802 fsparam_u32 ("journal_dev", Opt_journal_dev), 1803 fsparam_bdev ("journal_path", Opt_journal_path), 1804 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1805 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1806 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1807 fsparam_flag ("abort", Opt_abort), 1808 fsparam_enum ("data", Opt_data, ext4_param_data), 1809 fsparam_enum ("data_err", Opt_data_err, 1810 ext4_param_data_err), 1811 fsparam_string_empty 1812 ("usrjquota", Opt_usrjquota), 1813 fsparam_string_empty 1814 ("grpjquota", Opt_grpjquota), 1815 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1816 fsparam_flag ("grpquota", Opt_grpquota), 1817 fsparam_flag ("quota", Opt_quota), 1818 fsparam_flag ("noquota", Opt_noquota), 1819 fsparam_flag ("usrquota", Opt_usrquota), 1820 fsparam_flag ("prjquota", Opt_prjquota), 1821 fsparam_flag ("barrier", Opt_barrier), 1822 fsparam_u32 ("barrier", Opt_barrier), 1823 fsparam_flag ("nobarrier", Opt_nobarrier), 1824 fsparam_flag ("i_version", Opt_removed), 1825 fsparam_flag ("dax", Opt_dax), 1826 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1827 fsparam_u32 ("stripe", Opt_stripe), 1828 fsparam_flag ("delalloc", Opt_delalloc), 1829 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1830 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1831 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1832 fsparam_u32 ("debug_want_extra_isize", 1833 Opt_debug_want_extra_isize), 1834 fsparam_flag ("mblk_io_submit", Opt_removed), 1835 fsparam_flag ("nomblk_io_submit", Opt_removed), 1836 fsparam_flag ("block_validity", Opt_block_validity), 1837 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1838 fsparam_u32 ("inode_readahead_blks", 1839 Opt_inode_readahead_blks), 1840 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1841 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1842 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1843 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1844 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1845 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1846 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1847 fsparam_flag ("discard", Opt_discard), 1848 fsparam_flag ("nodiscard", Opt_nodiscard), 1849 fsparam_u32 ("init_itable", Opt_init_itable), 1850 fsparam_flag ("init_itable", Opt_init_itable), 1851 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1852 #ifdef CONFIG_EXT4_DEBUG 1853 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1854 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1855 #endif 1856 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1857 fsparam_flag ("test_dummy_encryption", 1858 Opt_test_dummy_encryption), 1859 fsparam_string ("test_dummy_encryption", 1860 Opt_test_dummy_encryption), 1861 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1862 fsparam_flag ("nombcache", Opt_nombcache), 1863 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1864 fsparam_flag ("prefetch_block_bitmaps", 1865 Opt_removed), 1866 fsparam_flag ("no_prefetch_block_bitmaps", 1867 Opt_no_prefetch_block_bitmaps), 1868 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1869 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1870 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1871 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1872 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1873 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1874 {} 1875 }; 1876 1877 1878 #define MOPT_SET 0x0001 1879 #define MOPT_CLEAR 0x0002 1880 #define MOPT_NOSUPPORT 0x0004 1881 #define MOPT_EXPLICIT 0x0008 1882 #ifdef CONFIG_QUOTA 1883 #define MOPT_Q 0 1884 #define MOPT_QFMT 0x0010 1885 #else 1886 #define MOPT_Q MOPT_NOSUPPORT 1887 #define MOPT_QFMT MOPT_NOSUPPORT 1888 #endif 1889 #define MOPT_NO_EXT2 0x0020 1890 #define MOPT_NO_EXT3 0x0040 1891 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1892 #define MOPT_SKIP 0x0080 1893 #define MOPT_2 0x0100 1894 1895 static const struct mount_opts { 1896 int token; 1897 int mount_opt; 1898 int flags; 1899 } ext4_mount_opts[] = { 1900 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1901 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1902 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1903 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1904 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1905 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1906 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1907 MOPT_EXT4_ONLY | MOPT_SET}, 1908 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1909 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1910 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1911 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1912 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1913 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1914 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1915 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1916 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1917 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1918 {Opt_commit, 0, MOPT_NO_EXT2}, 1919 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1920 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1921 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1922 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1923 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1924 EXT4_MOUNT_JOURNAL_CHECKSUM), 1925 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1926 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1927 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1928 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1929 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1930 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1931 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1932 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1933 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1934 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1935 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1936 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1937 {Opt_data, 0, MOPT_NO_EXT2}, 1938 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1939 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1940 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1941 #else 1942 {Opt_acl, 0, MOPT_NOSUPPORT}, 1943 #endif 1944 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1945 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1946 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1947 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1948 MOPT_SET | MOPT_Q}, 1949 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1950 MOPT_SET | MOPT_Q}, 1951 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1952 MOPT_SET | MOPT_Q}, 1953 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1954 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1955 MOPT_CLEAR | MOPT_Q}, 1956 {Opt_usrjquota, 0, MOPT_Q}, 1957 {Opt_grpjquota, 0, MOPT_Q}, 1958 {Opt_jqfmt, 0, MOPT_QFMT}, 1959 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1960 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1961 MOPT_SET}, 1962 #ifdef CONFIG_EXT4_DEBUG 1963 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1964 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1965 #endif 1966 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2}, 1967 {Opt_err, 0, 0} 1968 }; 1969 1970 #if IS_ENABLED(CONFIG_UNICODE) 1971 static const struct ext4_sb_encodings { 1972 __u16 magic; 1973 char *name; 1974 unsigned int version; 1975 } ext4_sb_encoding_map[] = { 1976 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1977 }; 1978 1979 static const struct ext4_sb_encodings * 1980 ext4_sb_read_encoding(const struct ext4_super_block *es) 1981 { 1982 __u16 magic = le16_to_cpu(es->s_encoding); 1983 int i; 1984 1985 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1986 if (magic == ext4_sb_encoding_map[i].magic) 1987 return &ext4_sb_encoding_map[i]; 1988 1989 return NULL; 1990 } 1991 #endif 1992 1993 #define EXT4_SPEC_JQUOTA (1 << 0) 1994 #define EXT4_SPEC_JQFMT (1 << 1) 1995 #define EXT4_SPEC_DATAJ (1 << 2) 1996 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1997 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1998 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1999 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 2000 #define EXT4_SPEC_s_max_batch_time (1 << 8) 2001 #define EXT4_SPEC_s_min_batch_time (1 << 9) 2002 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 2003 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 2004 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 2005 #define EXT4_SPEC_s_stripe (1 << 13) 2006 #define EXT4_SPEC_s_resuid (1 << 14) 2007 #define EXT4_SPEC_s_resgid (1 << 15) 2008 #define EXT4_SPEC_s_commit_interval (1 << 16) 2009 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 2010 #define EXT4_SPEC_s_sb_block (1 << 18) 2011 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 2012 2013 struct ext4_fs_context { 2014 char *s_qf_names[EXT4_MAXQUOTAS]; 2015 struct fscrypt_dummy_policy dummy_enc_policy; 2016 int s_jquota_fmt; /* Format of quota to use */ 2017 #ifdef CONFIG_EXT4_DEBUG 2018 int s_fc_debug_max_replay; 2019 #endif 2020 unsigned short qname_spec; 2021 unsigned long vals_s_flags; /* Bits to set in s_flags */ 2022 unsigned long mask_s_flags; /* Bits changed in s_flags */ 2023 unsigned long journal_devnum; 2024 unsigned long s_commit_interval; 2025 unsigned long s_stripe; 2026 unsigned int s_inode_readahead_blks; 2027 unsigned int s_want_extra_isize; 2028 unsigned int s_li_wait_mult; 2029 unsigned int s_max_dir_size_kb; 2030 unsigned int journal_ioprio; 2031 unsigned int vals_s_mount_opt; 2032 unsigned int mask_s_mount_opt; 2033 unsigned int vals_s_mount_opt2; 2034 unsigned int mask_s_mount_opt2; 2035 unsigned int opt_flags; /* MOPT flags */ 2036 unsigned int spec; 2037 u32 s_max_batch_time; 2038 u32 s_min_batch_time; 2039 kuid_t s_resuid; 2040 kgid_t s_resgid; 2041 ext4_fsblk_t s_sb_block; 2042 }; 2043 2044 static void ext4_fc_free(struct fs_context *fc) 2045 { 2046 struct ext4_fs_context *ctx = fc->fs_private; 2047 int i; 2048 2049 if (!ctx) 2050 return; 2051 2052 for (i = 0; i < EXT4_MAXQUOTAS; i++) 2053 kfree(ctx->s_qf_names[i]); 2054 2055 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 2056 kfree(ctx); 2057 } 2058 2059 int ext4_init_fs_context(struct fs_context *fc) 2060 { 2061 struct ext4_fs_context *ctx; 2062 2063 ctx = kzalloc_obj(struct ext4_fs_context); 2064 if (!ctx) 2065 return -ENOMEM; 2066 2067 fc->fs_private = ctx; 2068 fc->ops = &ext4_context_ops; 2069 2070 /* i_version is always enabled now */ 2071 fc->sb_flags |= SB_I_VERSION; 2072 2073 return 0; 2074 } 2075 2076 #ifdef CONFIG_QUOTA 2077 /* 2078 * Note the name of the specified quota file. 2079 */ 2080 static int note_qf_name(struct fs_context *fc, int qtype, 2081 struct fs_parameter *param) 2082 { 2083 struct ext4_fs_context *ctx = fc->fs_private; 2084 char *qname; 2085 2086 if (param->size < 1) { 2087 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 2088 return -EINVAL; 2089 } 2090 if (strchr(param->string, '/')) { 2091 ext4_msg(NULL, KERN_ERR, 2092 "quotafile must be on filesystem root"); 2093 return -EINVAL; 2094 } 2095 if (ctx->s_qf_names[qtype]) { 2096 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 2097 ext4_msg(NULL, KERN_ERR, 2098 "%s quota file already specified", 2099 QTYPE2NAME(qtype)); 2100 return -EINVAL; 2101 } 2102 return 0; 2103 } 2104 2105 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 2106 if (!qname) { 2107 ext4_msg(NULL, KERN_ERR, 2108 "Not enough memory for storing quotafile name"); 2109 return -ENOMEM; 2110 } 2111 ctx->s_qf_names[qtype] = qname; 2112 ctx->qname_spec |= 1 << qtype; 2113 ctx->spec |= EXT4_SPEC_JQUOTA; 2114 return 0; 2115 } 2116 2117 /* 2118 * Clear the name of the specified quota file. 2119 */ 2120 static int unnote_qf_name(struct fs_context *fc, int qtype) 2121 { 2122 struct ext4_fs_context *ctx = fc->fs_private; 2123 2124 kfree(ctx->s_qf_names[qtype]); 2125 2126 ctx->s_qf_names[qtype] = NULL; 2127 ctx->qname_spec |= 1 << qtype; 2128 ctx->spec |= EXT4_SPEC_JQUOTA; 2129 return 0; 2130 } 2131 #endif 2132 2133 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2134 struct ext4_fs_context *ctx) 2135 { 2136 int err; 2137 2138 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2139 ext4_msg(NULL, KERN_WARNING, 2140 "test_dummy_encryption option not supported"); 2141 return -EINVAL; 2142 } 2143 err = fscrypt_parse_test_dummy_encryption(param, 2144 &ctx->dummy_enc_policy); 2145 if (err == -EINVAL) { 2146 ext4_msg(NULL, KERN_WARNING, 2147 "Value of option \"%s\" is unrecognized", param->key); 2148 } else if (err == -EEXIST) { 2149 ext4_msg(NULL, KERN_WARNING, 2150 "Conflicting test_dummy_encryption options"); 2151 return -EINVAL; 2152 } 2153 return err; 2154 } 2155 2156 #define EXT4_SET_CTX(name) \ 2157 static inline __maybe_unused \ 2158 void ctx_set_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2159 { \ 2160 ctx->mask_s_##name |= flag; \ 2161 ctx->vals_s_##name |= flag; \ 2162 } 2163 2164 #define EXT4_CLEAR_CTX(name) \ 2165 static inline __maybe_unused \ 2166 void ctx_clear_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2167 { \ 2168 ctx->mask_s_##name |= flag; \ 2169 ctx->vals_s_##name &= ~flag; \ 2170 } 2171 2172 #define EXT4_TEST_CTX(name) \ 2173 static inline unsigned long \ 2174 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2175 { \ 2176 return (ctx->vals_s_##name & flag); \ 2177 } 2178 2179 EXT4_SET_CTX(flags); /* set only */ 2180 EXT4_SET_CTX(mount_opt); 2181 EXT4_CLEAR_CTX(mount_opt); 2182 EXT4_TEST_CTX(mount_opt); 2183 EXT4_SET_CTX(mount_opt2); 2184 EXT4_CLEAR_CTX(mount_opt2); 2185 EXT4_TEST_CTX(mount_opt2); 2186 2187 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2188 { 2189 struct ext4_fs_context *ctx = fc->fs_private; 2190 struct fs_parse_result result; 2191 const struct mount_opts *m; 2192 int is_remount; 2193 int token; 2194 2195 token = fs_parse(fc, ext4_param_specs, param, &result); 2196 if (token < 0) 2197 return token; 2198 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2199 2200 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2201 if (token == m->token) 2202 break; 2203 2204 ctx->opt_flags |= m->flags; 2205 2206 if (m->flags & MOPT_EXPLICIT) { 2207 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2208 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2209 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2210 ctx_set_mount_opt2(ctx, 2211 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2212 } else 2213 return -EINVAL; 2214 } 2215 2216 if (m->flags & MOPT_NOSUPPORT) { 2217 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2218 param->key); 2219 return 0; 2220 } 2221 2222 switch (token) { 2223 #ifdef CONFIG_QUOTA 2224 case Opt_usrjquota: 2225 if (!*param->string) 2226 return unnote_qf_name(fc, USRQUOTA); 2227 else 2228 return note_qf_name(fc, USRQUOTA, param); 2229 case Opt_grpjquota: 2230 if (!*param->string) 2231 return unnote_qf_name(fc, GRPQUOTA); 2232 else 2233 return note_qf_name(fc, GRPQUOTA, param); 2234 #endif 2235 case Opt_sb: 2236 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2237 ext4_msg(NULL, KERN_WARNING, 2238 "Ignoring %s option on remount", param->key); 2239 } else { 2240 ctx->s_sb_block = result.uint_32; 2241 ctx->spec |= EXT4_SPEC_s_sb_block; 2242 } 2243 return 0; 2244 case Opt_removed: 2245 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2246 param->key); 2247 return 0; 2248 case Opt_inlinecrypt: 2249 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2250 ctx_set_flags(ctx, SB_INLINECRYPT); 2251 #else 2252 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2253 #endif 2254 return 0; 2255 case Opt_errors: 2256 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2257 ctx_set_mount_opt(ctx, result.uint_32); 2258 return 0; 2259 #ifdef CONFIG_QUOTA 2260 case Opt_jqfmt: 2261 ctx->s_jquota_fmt = result.uint_32; 2262 ctx->spec |= EXT4_SPEC_JQFMT; 2263 return 0; 2264 #endif 2265 case Opt_data: 2266 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2267 ctx_set_mount_opt(ctx, result.uint_32); 2268 ctx->spec |= EXT4_SPEC_DATAJ; 2269 return 0; 2270 case Opt_commit: 2271 if (result.uint_32 == 0) 2272 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE; 2273 else if (result.uint_32 > INT_MAX / HZ) { 2274 ext4_msg(NULL, KERN_ERR, 2275 "Invalid commit interval %d, " 2276 "must be smaller than %d", 2277 result.uint_32, INT_MAX / HZ); 2278 return -EINVAL; 2279 } 2280 ctx->s_commit_interval = HZ * result.uint_32; 2281 ctx->spec |= EXT4_SPEC_s_commit_interval; 2282 return 0; 2283 case Opt_debug_want_extra_isize: 2284 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2285 ext4_msg(NULL, KERN_ERR, 2286 "Invalid want_extra_isize %d", result.uint_32); 2287 return -EINVAL; 2288 } 2289 ctx->s_want_extra_isize = result.uint_32; 2290 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2291 return 0; 2292 case Opt_max_batch_time: 2293 ctx->s_max_batch_time = result.uint_32; 2294 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2295 return 0; 2296 case Opt_min_batch_time: 2297 ctx->s_min_batch_time = result.uint_32; 2298 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2299 return 0; 2300 case Opt_inode_readahead_blks: 2301 if (result.uint_32 && 2302 (result.uint_32 > (1 << 30) || 2303 !is_power_of_2(result.uint_32))) { 2304 ext4_msg(NULL, KERN_ERR, 2305 "EXT4-fs: inode_readahead_blks must be " 2306 "0 or a power of 2 smaller than 2^31"); 2307 return -EINVAL; 2308 } 2309 ctx->s_inode_readahead_blks = result.uint_32; 2310 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2311 return 0; 2312 case Opt_init_itable: 2313 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2314 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2315 if (param->type == fs_value_is_string) 2316 ctx->s_li_wait_mult = result.uint_32; 2317 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2318 return 0; 2319 case Opt_max_dir_size_kb: 2320 ctx->s_max_dir_size_kb = result.uint_32; 2321 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2322 return 0; 2323 #ifdef CONFIG_EXT4_DEBUG 2324 case Opt_fc_debug_max_replay: 2325 ctx->s_fc_debug_max_replay = result.uint_32; 2326 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2327 return 0; 2328 #endif 2329 case Opt_stripe: 2330 ctx->s_stripe = result.uint_32; 2331 ctx->spec |= EXT4_SPEC_s_stripe; 2332 return 0; 2333 case Opt_resuid: 2334 ctx->s_resuid = result.uid; 2335 ctx->spec |= EXT4_SPEC_s_resuid; 2336 return 0; 2337 case Opt_resgid: 2338 ctx->s_resgid = result.gid; 2339 ctx->spec |= EXT4_SPEC_s_resgid; 2340 return 0; 2341 case Opt_journal_dev: 2342 if (is_remount) { 2343 ext4_msg(NULL, KERN_ERR, 2344 "Cannot specify journal on remount"); 2345 return -EINVAL; 2346 } 2347 ctx->journal_devnum = result.uint_32; 2348 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2349 return 0; 2350 case Opt_journal_path: 2351 { 2352 struct inode *journal_inode; 2353 struct path path; 2354 int error; 2355 2356 if (is_remount) { 2357 ext4_msg(NULL, KERN_ERR, 2358 "Cannot specify journal on remount"); 2359 return -EINVAL; 2360 } 2361 2362 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path); 2363 if (error) { 2364 ext4_msg(NULL, KERN_ERR, "error: could not find " 2365 "journal device path"); 2366 return -EINVAL; 2367 } 2368 2369 journal_inode = d_inode(path.dentry); 2370 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2371 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2372 path_put(&path); 2373 return 0; 2374 } 2375 case Opt_journal_ioprio: 2376 if (result.uint_32 > 7) { 2377 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2378 " (must be 0-7)"); 2379 return -EINVAL; 2380 } 2381 ctx->journal_ioprio = 2382 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2383 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2384 return 0; 2385 case Opt_test_dummy_encryption: 2386 return ext4_parse_test_dummy_encryption(param, ctx); 2387 case Opt_dax: 2388 case Opt_dax_type: 2389 #ifdef CONFIG_FS_DAX 2390 { 2391 int type = (token == Opt_dax) ? 2392 Opt_dax : result.uint_32; 2393 2394 switch (type) { 2395 case Opt_dax: 2396 case Opt_dax_always: 2397 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2398 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2399 break; 2400 case Opt_dax_never: 2401 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2402 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2403 break; 2404 case Opt_dax_inode: 2405 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2406 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2407 /* Strictly for printing options */ 2408 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2409 break; 2410 } 2411 return 0; 2412 } 2413 #else 2414 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2415 return -EINVAL; 2416 #endif 2417 case Opt_data_err: 2418 if (result.uint_32 == Opt_data_err_abort) 2419 ctx_set_mount_opt(ctx, m->mount_opt); 2420 else if (result.uint_32 == Opt_data_err_ignore) 2421 ctx_clear_mount_opt(ctx, m->mount_opt); 2422 return 0; 2423 case Opt_mb_optimize_scan: 2424 if (result.int_32 == 1) { 2425 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2426 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2427 } else if (result.int_32 == 0) { 2428 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2429 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2430 } else { 2431 ext4_msg(NULL, KERN_WARNING, 2432 "mb_optimize_scan should be set to 0 or 1."); 2433 return -EINVAL; 2434 } 2435 return 0; 2436 } 2437 2438 /* 2439 * At this point we should only be getting options requiring MOPT_SET, 2440 * or MOPT_CLEAR. Anything else is a bug 2441 */ 2442 if (m->token == Opt_err) { 2443 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2444 param->key); 2445 WARN_ON(1); 2446 return -EINVAL; 2447 } 2448 2449 else { 2450 unsigned int set = 0; 2451 2452 if ((param->type == fs_value_is_flag) || 2453 result.uint_32 > 0) 2454 set = 1; 2455 2456 if (m->flags & MOPT_CLEAR) 2457 set = !set; 2458 else if (unlikely(!(m->flags & MOPT_SET))) { 2459 ext4_msg(NULL, KERN_WARNING, 2460 "buggy handling of option %s", 2461 param->key); 2462 WARN_ON(1); 2463 return -EINVAL; 2464 } 2465 if (m->flags & MOPT_2) { 2466 if (set != 0) 2467 ctx_set_mount_opt2(ctx, m->mount_opt); 2468 else 2469 ctx_clear_mount_opt2(ctx, m->mount_opt); 2470 } else { 2471 if (set != 0) 2472 ctx_set_mount_opt(ctx, m->mount_opt); 2473 else 2474 ctx_clear_mount_opt(ctx, m->mount_opt); 2475 } 2476 } 2477 2478 return 0; 2479 } 2480 2481 static int parse_options(struct fs_context *fc, char *options) 2482 { 2483 struct fs_parameter param; 2484 int ret; 2485 char *key; 2486 2487 if (!options) 2488 return 0; 2489 2490 while ((key = strsep(&options, ",")) != NULL) { 2491 if (*key) { 2492 size_t v_len = 0; 2493 char *value = strchr(key, '='); 2494 2495 param.type = fs_value_is_flag; 2496 param.string = NULL; 2497 2498 if (value) { 2499 if (value == key) 2500 continue; 2501 2502 *value++ = 0; 2503 v_len = strlen(value); 2504 param.string = kmemdup_nul(value, v_len, 2505 GFP_KERNEL); 2506 if (!param.string) 2507 return -ENOMEM; 2508 param.type = fs_value_is_string; 2509 } 2510 2511 param.key = key; 2512 param.size = v_len; 2513 2514 ret = ext4_parse_param(fc, ¶m); 2515 kfree(param.string); 2516 if (ret < 0) 2517 return ret; 2518 } 2519 } 2520 2521 ret = ext4_validate_options(fc); 2522 if (ret < 0) 2523 return ret; 2524 2525 return 0; 2526 } 2527 2528 static int parse_apply_sb_mount_options(struct super_block *sb, 2529 struct ext4_fs_context *m_ctx) 2530 { 2531 struct ext4_sb_info *sbi = EXT4_SB(sb); 2532 char s_mount_opts[64]; 2533 struct ext4_fs_context *s_ctx = NULL; 2534 struct fs_context *fc = NULL; 2535 int ret = -ENOMEM; 2536 2537 if (!sbi->s_es->s_mount_opts[0]) 2538 return 0; 2539 2540 if (strscpy_pad(s_mount_opts, sbi->s_es->s_mount_opts) < 0) 2541 return -E2BIG; 2542 2543 fc = kzalloc_obj(struct fs_context); 2544 if (!fc) 2545 return -ENOMEM; 2546 2547 s_ctx = kzalloc_obj(struct ext4_fs_context); 2548 if (!s_ctx) 2549 goto out_free; 2550 2551 fc->fs_private = s_ctx; 2552 fc->s_fs_info = sbi; 2553 2554 ret = parse_options(fc, s_mount_opts); 2555 if (ret < 0) 2556 goto parse_failed; 2557 2558 ret = ext4_check_opt_consistency(fc, sb); 2559 if (ret < 0) { 2560 parse_failed: 2561 ext4_msg(sb, KERN_WARNING, 2562 "failed to parse options in superblock: %s", 2563 s_mount_opts); 2564 ret = 0; 2565 goto out_free; 2566 } 2567 2568 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2569 m_ctx->journal_devnum = s_ctx->journal_devnum; 2570 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2571 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2572 2573 ext4_apply_options(fc, sb); 2574 ret = 0; 2575 2576 out_free: 2577 ext4_fc_free(fc); 2578 kfree(fc); 2579 return ret; 2580 } 2581 2582 static void ext4_apply_quota_options(struct fs_context *fc, 2583 struct super_block *sb) 2584 { 2585 #ifdef CONFIG_QUOTA 2586 bool quota_feature = ext4_has_feature_quota(sb); 2587 struct ext4_fs_context *ctx = fc->fs_private; 2588 struct ext4_sb_info *sbi = EXT4_SB(sb); 2589 char *qname; 2590 int i; 2591 2592 if (quota_feature) 2593 return; 2594 2595 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2596 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2597 if (!(ctx->qname_spec & (1 << i))) 2598 continue; 2599 2600 qname = ctx->s_qf_names[i]; /* May be NULL */ 2601 if (qname) 2602 set_opt(sb, QUOTA); 2603 ctx->s_qf_names[i] = NULL; 2604 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2605 lockdep_is_held(&sb->s_umount)); 2606 if (qname) 2607 kfree_rcu_mightsleep(qname); 2608 } 2609 } 2610 2611 if (ctx->spec & EXT4_SPEC_JQFMT) 2612 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2613 #endif 2614 } 2615 2616 /* 2617 * Check quota settings consistency. 2618 */ 2619 static int ext4_check_quota_consistency(struct fs_context *fc, 2620 struct super_block *sb) 2621 { 2622 #ifdef CONFIG_QUOTA 2623 struct ext4_fs_context *ctx = fc->fs_private; 2624 struct ext4_sb_info *sbi = EXT4_SB(sb); 2625 bool quota_feature = ext4_has_feature_quota(sb); 2626 bool quota_loaded = sb_any_quota_loaded(sb); 2627 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2628 int quota_flags, i; 2629 2630 /* 2631 * We do the test below only for project quotas. 'usrquota' and 2632 * 'grpquota' mount options are allowed even without quota feature 2633 * to support legacy quotas in quota files. 2634 */ 2635 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2636 !ext4_has_feature_project(sb)) { 2637 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2638 "Cannot enable project quota enforcement."); 2639 return -EINVAL; 2640 } 2641 2642 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2643 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2644 if (quota_loaded && 2645 ctx->mask_s_mount_opt & quota_flags && 2646 !ctx_test_mount_opt(ctx, quota_flags)) 2647 goto err_quota_change; 2648 2649 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2650 2651 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2652 if (!(ctx->qname_spec & (1 << i))) 2653 continue; 2654 2655 if (quota_loaded && 2656 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2657 goto err_jquota_change; 2658 2659 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2660 strcmp(get_qf_name(sb, sbi, i), 2661 ctx->s_qf_names[i]) != 0) 2662 goto err_jquota_specified; 2663 } 2664 2665 if (quota_feature) { 2666 ext4_msg(NULL, KERN_INFO, 2667 "Journaled quota options ignored when " 2668 "QUOTA feature is enabled"); 2669 return 0; 2670 } 2671 } 2672 2673 if (ctx->spec & EXT4_SPEC_JQFMT) { 2674 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2675 goto err_jquota_change; 2676 if (quota_feature) { 2677 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2678 "ignored when QUOTA feature is enabled"); 2679 return 0; 2680 } 2681 } 2682 2683 /* Make sure we don't mix old and new quota format */ 2684 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2685 ctx->s_qf_names[USRQUOTA]); 2686 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2687 ctx->s_qf_names[GRPQUOTA]); 2688 2689 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2690 test_opt(sb, USRQUOTA)); 2691 2692 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2693 test_opt(sb, GRPQUOTA)); 2694 2695 if (usr_qf_name) { 2696 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2697 usrquota = false; 2698 } 2699 if (grp_qf_name) { 2700 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2701 grpquota = false; 2702 } 2703 2704 if (usr_qf_name || grp_qf_name) { 2705 if (usrquota || grpquota) { 2706 ext4_msg(NULL, KERN_ERR, "old and new quota " 2707 "format mixing"); 2708 return -EINVAL; 2709 } 2710 2711 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2712 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2713 "not specified"); 2714 return -EINVAL; 2715 } 2716 } 2717 2718 return 0; 2719 2720 err_quota_change: 2721 ext4_msg(NULL, KERN_ERR, 2722 "Cannot change quota options when quota turned on"); 2723 return -EINVAL; 2724 err_jquota_change: 2725 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2726 "options when quota turned on"); 2727 return -EINVAL; 2728 err_jquota_specified: 2729 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2730 QTYPE2NAME(i)); 2731 return -EINVAL; 2732 #else 2733 return 0; 2734 #endif 2735 } 2736 2737 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2738 struct super_block *sb) 2739 { 2740 const struct ext4_fs_context *ctx = fc->fs_private; 2741 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2742 2743 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2744 return 0; 2745 2746 if (!ext4_has_feature_encrypt(sb)) { 2747 ext4_msg(NULL, KERN_WARNING, 2748 "test_dummy_encryption requires encrypt feature"); 2749 return -EINVAL; 2750 } 2751 /* 2752 * This mount option is just for testing, and it's not worthwhile to 2753 * implement the extra complexity (e.g. RCU protection) that would be 2754 * needed to allow it to be set or changed during remount. We do allow 2755 * it to be specified during remount, but only if there is no change. 2756 */ 2757 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2758 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2759 &ctx->dummy_enc_policy)) 2760 return 0; 2761 ext4_msg(NULL, KERN_WARNING, 2762 "Can't set or change test_dummy_encryption on remount"); 2763 return -EINVAL; 2764 } 2765 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2766 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2767 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2768 &ctx->dummy_enc_policy)) 2769 return 0; 2770 ext4_msg(NULL, KERN_WARNING, 2771 "Conflicting test_dummy_encryption options"); 2772 return -EINVAL; 2773 } 2774 return 0; 2775 } 2776 2777 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2778 struct super_block *sb) 2779 { 2780 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2781 /* if already set, it was already verified to be the same */ 2782 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2783 return; 2784 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2785 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2786 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2787 } 2788 2789 static int ext4_check_opt_consistency(struct fs_context *fc, 2790 struct super_block *sb) 2791 { 2792 struct ext4_fs_context *ctx = fc->fs_private; 2793 struct ext4_sb_info *sbi = fc->s_fs_info; 2794 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2795 int err; 2796 2797 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2798 ext4_msg(NULL, KERN_ERR, 2799 "Mount option(s) incompatible with ext2"); 2800 return -EINVAL; 2801 } 2802 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2803 ext4_msg(NULL, KERN_ERR, 2804 "Mount option(s) incompatible with ext3"); 2805 return -EINVAL; 2806 } 2807 2808 if (ctx->s_want_extra_isize > 2809 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2810 ext4_msg(NULL, KERN_ERR, 2811 "Invalid want_extra_isize %d", 2812 ctx->s_want_extra_isize); 2813 return -EINVAL; 2814 } 2815 2816 err = ext4_check_test_dummy_encryption(fc, sb); 2817 if (err) 2818 return err; 2819 2820 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2821 if (!sbi->s_journal) { 2822 ext4_msg(NULL, KERN_WARNING, 2823 "Remounting file system with no journal " 2824 "so ignoring journalled data option"); 2825 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2826 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2827 test_opt(sb, DATA_FLAGS)) { 2828 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2829 "on remount"); 2830 return -EINVAL; 2831 } 2832 } 2833 2834 if (is_remount) { 2835 if (!sbi->s_journal && 2836 ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT)) { 2837 ext4_msg(NULL, KERN_WARNING, 2838 "Remounting fs w/o journal so ignoring data_err option"); 2839 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT); 2840 } 2841 2842 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2843 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2844 ext4_msg(NULL, KERN_ERR, "can't mount with " 2845 "both data=journal and dax"); 2846 return -EINVAL; 2847 } 2848 2849 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2850 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2851 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2852 fail_dax_change_remount: 2853 ext4_msg(NULL, KERN_ERR, "can't change " 2854 "dax mount option while remounting"); 2855 return -EINVAL; 2856 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2857 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2858 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2859 goto fail_dax_change_remount; 2860 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2861 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2862 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2863 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2864 goto fail_dax_change_remount; 2865 } 2866 } 2867 2868 return ext4_check_quota_consistency(fc, sb); 2869 } 2870 2871 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2872 { 2873 struct ext4_fs_context *ctx = fc->fs_private; 2874 struct ext4_sb_info *sbi = fc->s_fs_info; 2875 2876 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2877 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2878 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2879 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2880 sb->s_flags &= ~ctx->mask_s_flags; 2881 sb->s_flags |= ctx->vals_s_flags; 2882 2883 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2884 APPLY(s_commit_interval); 2885 APPLY(s_stripe); 2886 APPLY(s_max_batch_time); 2887 APPLY(s_min_batch_time); 2888 APPLY(s_want_extra_isize); 2889 APPLY(s_inode_readahead_blks); 2890 APPLY(s_max_dir_size_kb); 2891 APPLY(s_li_wait_mult); 2892 APPLY(s_resgid); 2893 APPLY(s_resuid); 2894 2895 #ifdef CONFIG_EXT4_DEBUG 2896 APPLY(s_fc_debug_max_replay); 2897 #endif 2898 2899 ext4_apply_quota_options(fc, sb); 2900 ext4_apply_test_dummy_encryption(ctx, sb); 2901 } 2902 2903 2904 static int ext4_validate_options(struct fs_context *fc) 2905 { 2906 #ifdef CONFIG_QUOTA 2907 struct ext4_fs_context *ctx = fc->fs_private; 2908 char *usr_qf_name, *grp_qf_name; 2909 2910 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2911 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2912 2913 if (usr_qf_name || grp_qf_name) { 2914 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2915 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2916 2917 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2918 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2919 2920 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2921 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2922 ext4_msg(NULL, KERN_ERR, "old and new quota " 2923 "format mixing"); 2924 return -EINVAL; 2925 } 2926 } 2927 #endif 2928 return 1; 2929 } 2930 2931 static inline void ext4_show_quota_options(struct seq_file *seq, 2932 struct super_block *sb) 2933 { 2934 #if defined(CONFIG_QUOTA) 2935 struct ext4_sb_info *sbi = EXT4_SB(sb); 2936 char *usr_qf_name, *grp_qf_name; 2937 2938 if (sbi->s_jquota_fmt) { 2939 char *fmtname = ""; 2940 2941 switch (sbi->s_jquota_fmt) { 2942 case QFMT_VFS_OLD: 2943 fmtname = "vfsold"; 2944 break; 2945 case QFMT_VFS_V0: 2946 fmtname = "vfsv0"; 2947 break; 2948 case QFMT_VFS_V1: 2949 fmtname = "vfsv1"; 2950 break; 2951 } 2952 seq_printf(seq, ",jqfmt=%s", fmtname); 2953 } 2954 2955 rcu_read_lock(); 2956 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2957 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2958 if (usr_qf_name) 2959 seq_show_option(seq, "usrjquota", usr_qf_name); 2960 if (grp_qf_name) 2961 seq_show_option(seq, "grpjquota", grp_qf_name); 2962 rcu_read_unlock(); 2963 #endif 2964 } 2965 2966 static const char *token2str(int token) 2967 { 2968 const struct fs_parameter_spec *spec; 2969 2970 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2971 if (spec->opt == token && !spec->type) 2972 break; 2973 return spec->name; 2974 } 2975 2976 /* 2977 * Show an option if 2978 * - it's set to a non-default value OR 2979 * - if the per-sb default is different from the global default 2980 */ 2981 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2982 int nodefs) 2983 { 2984 struct ext4_sb_info *sbi = EXT4_SB(sb); 2985 struct ext4_super_block *es = sbi->s_es; 2986 int def_errors; 2987 const struct mount_opts *m; 2988 char sep = nodefs ? '\n' : ','; 2989 2990 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2991 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2992 2993 if (sbi->s_sb_block != 1) 2994 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2995 2996 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2997 int want_set = m->flags & MOPT_SET; 2998 int opt_2 = m->flags & MOPT_2; 2999 unsigned int mount_opt, def_mount_opt; 3000 3001 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 3002 m->flags & MOPT_SKIP) 3003 continue; 3004 3005 if (opt_2) { 3006 mount_opt = sbi->s_mount_opt2; 3007 def_mount_opt = sbi->s_def_mount_opt2; 3008 } else { 3009 mount_opt = sbi->s_mount_opt; 3010 def_mount_opt = sbi->s_def_mount_opt; 3011 } 3012 /* skip if same as the default */ 3013 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt))) 3014 continue; 3015 /* select Opt_noFoo vs Opt_Foo */ 3016 if ((want_set && 3017 (mount_opt & m->mount_opt) != m->mount_opt) || 3018 (!want_set && (mount_opt & m->mount_opt))) 3019 continue; 3020 SEQ_OPTS_PRINT("%s", token2str(m->token)); 3021 } 3022 3023 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 3024 ext4_get_resuid(es) != EXT4_DEF_RESUID) 3025 SEQ_OPTS_PRINT("resuid=%u", 3026 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 3027 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 3028 ext4_get_resgid(es) != EXT4_DEF_RESGID) 3029 SEQ_OPTS_PRINT("resgid=%u", 3030 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 3031 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 3032 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 3033 SEQ_OPTS_PUTS("errors=remount-ro"); 3034 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 3035 SEQ_OPTS_PUTS("errors=continue"); 3036 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 3037 SEQ_OPTS_PUTS("errors=panic"); 3038 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 3039 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 3040 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 3041 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 3042 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 3043 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 3044 if (nodefs && sb->s_flags & SB_I_VERSION) 3045 SEQ_OPTS_PUTS("i_version"); 3046 if (nodefs || sbi->s_stripe) 3047 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 3048 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 3049 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 3050 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 3051 SEQ_OPTS_PUTS("data=journal"); 3052 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 3053 SEQ_OPTS_PUTS("data=ordered"); 3054 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 3055 SEQ_OPTS_PUTS("data=writeback"); 3056 } 3057 if (nodefs || 3058 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 3059 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 3060 sbi->s_inode_readahead_blks); 3061 3062 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 3063 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 3064 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 3065 if (nodefs || sbi->s_max_dir_size_kb) 3066 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 3067 if (test_opt(sb, DATA_ERR_ABORT)) 3068 SEQ_OPTS_PUTS("data_err=abort"); 3069 3070 fscrypt_show_test_dummy_encryption(seq, sep, sb); 3071 3072 if (sb->s_flags & SB_INLINECRYPT) 3073 SEQ_OPTS_PUTS("inlinecrypt"); 3074 3075 if (test_opt(sb, DAX_ALWAYS)) { 3076 if (IS_EXT2_SB(sb)) 3077 SEQ_OPTS_PUTS("dax"); 3078 else 3079 SEQ_OPTS_PUTS("dax=always"); 3080 } else if (test_opt2(sb, DAX_NEVER)) { 3081 SEQ_OPTS_PUTS("dax=never"); 3082 } else if (test_opt2(sb, DAX_INODE)) { 3083 SEQ_OPTS_PUTS("dax=inode"); 3084 } 3085 3086 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3087 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3088 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 3089 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3090 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3091 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 3092 } 3093 3094 if (nodefs && !test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) 3095 SEQ_OPTS_PUTS("prefetch_block_bitmaps"); 3096 3097 if (ext4_emergency_ro(sb)) 3098 SEQ_OPTS_PUTS("emergency_ro"); 3099 3100 if (ext4_forced_shutdown(sb)) 3101 SEQ_OPTS_PUTS("shutdown"); 3102 3103 ext4_show_quota_options(seq, sb); 3104 return 0; 3105 } 3106 3107 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 3108 { 3109 return _ext4_show_options(seq, root->d_sb, 0); 3110 } 3111 3112 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3113 { 3114 struct super_block *sb = seq->private; 3115 int rc; 3116 3117 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3118 rc = _ext4_show_options(seq, sb, 1); 3119 seq_putc(seq, '\n'); 3120 return rc; 3121 } 3122 3123 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3124 int read_only) 3125 { 3126 struct ext4_sb_info *sbi = EXT4_SB(sb); 3127 int err = 0; 3128 3129 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3130 ext4_msg(sb, KERN_ERR, "revision level too high, " 3131 "forcing read-only mode"); 3132 err = -EROFS; 3133 goto done; 3134 } 3135 if (read_only) 3136 goto done; 3137 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3138 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3139 "running e2fsck is recommended"); 3140 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3141 ext4_msg(sb, KERN_WARNING, 3142 "warning: mounting fs with errors, " 3143 "running e2fsck is recommended"); 3144 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3145 le16_to_cpu(es->s_mnt_count) >= 3146 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3147 ext4_msg(sb, KERN_WARNING, 3148 "warning: maximal mount count reached, " 3149 "running e2fsck is recommended"); 3150 else if (le32_to_cpu(es->s_checkinterval) && 3151 (ext4_get_tstamp(es, s_lastcheck) + 3152 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3153 ext4_msg(sb, KERN_WARNING, 3154 "warning: checktime reached, " 3155 "running e2fsck is recommended"); 3156 if (!sbi->s_journal) 3157 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3158 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3159 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3160 le16_add_cpu(&es->s_mnt_count, 1); 3161 ext4_update_tstamp(es, s_mtime); 3162 if (sbi->s_journal) { 3163 ext4_set_feature_journal_needs_recovery(sb); 3164 if (ext4_has_feature_orphan_file(sb)) 3165 ext4_set_feature_orphan_present(sb); 3166 } 3167 3168 err = ext4_commit_super(sb); 3169 done: 3170 if (test_opt(sb, DEBUG)) 3171 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3172 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3173 sb->s_blocksize, 3174 sbi->s_groups_count, 3175 EXT4_BLOCKS_PER_GROUP(sb), 3176 EXT4_INODES_PER_GROUP(sb), 3177 sbi->s_mount_opt, sbi->s_mount_opt2); 3178 return err; 3179 } 3180 3181 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3182 { 3183 struct ext4_sb_info *sbi = EXT4_SB(sb); 3184 struct flex_groups **old_groups, **new_groups; 3185 int size, i, j; 3186 3187 if (!sbi->s_log_groups_per_flex) 3188 return 0; 3189 3190 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3191 if (size <= sbi->s_flex_groups_allocated) 3192 return 0; 3193 3194 new_groups = kvzalloc(roundup_pow_of_two(size * 3195 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3196 if (!new_groups) { 3197 ext4_msg(sb, KERN_ERR, 3198 "not enough memory for %d flex group pointers", size); 3199 return -ENOMEM; 3200 } 3201 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3202 new_groups[i] = kvzalloc(roundup_pow_of_two( 3203 sizeof(struct flex_groups)), 3204 GFP_KERNEL); 3205 if (!new_groups[i]) { 3206 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3207 kvfree(new_groups[j]); 3208 kvfree(new_groups); 3209 ext4_msg(sb, KERN_ERR, 3210 "not enough memory for %d flex groups", size); 3211 return -ENOMEM; 3212 } 3213 } 3214 rcu_read_lock(); 3215 old_groups = rcu_dereference(sbi->s_flex_groups); 3216 if (old_groups) 3217 memcpy(new_groups, old_groups, 3218 (sbi->s_flex_groups_allocated * 3219 sizeof(struct flex_groups *))); 3220 rcu_read_unlock(); 3221 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3222 sbi->s_flex_groups_allocated = size; 3223 if (old_groups) 3224 ext4_kvfree_array_rcu(old_groups); 3225 return 0; 3226 } 3227 3228 static int ext4_fill_flex_info(struct super_block *sb) 3229 { 3230 struct ext4_sb_info *sbi = EXT4_SB(sb); 3231 struct ext4_group_desc *gdp = NULL; 3232 struct flex_groups *fg; 3233 ext4_group_t flex_group; 3234 int i, err; 3235 3236 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3237 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3238 sbi->s_log_groups_per_flex = 0; 3239 return 1; 3240 } 3241 3242 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3243 if (err) 3244 goto failed; 3245 3246 for (i = 0; i < sbi->s_groups_count; i++) { 3247 gdp = ext4_get_group_desc(sb, i, NULL); 3248 3249 flex_group = ext4_flex_group(sbi, i); 3250 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3251 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3252 atomic64_add(ext4_free_group_clusters(sb, gdp), 3253 &fg->free_clusters); 3254 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3255 } 3256 3257 return 1; 3258 failed: 3259 return 0; 3260 } 3261 3262 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3263 struct ext4_group_desc *gdp) 3264 { 3265 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3266 __u16 crc = 0; 3267 __le32 le_group = cpu_to_le32(block_group); 3268 struct ext4_sb_info *sbi = EXT4_SB(sb); 3269 3270 if (ext4_has_feature_metadata_csum(sbi->s_sb)) { 3271 /* Use new metadata_csum algorithm */ 3272 __u32 csum32; 3273 __u16 dummy_csum = 0; 3274 3275 csum32 = ext4_chksum(sbi->s_csum_seed, (__u8 *)&le_group, 3276 sizeof(le_group)); 3277 csum32 = ext4_chksum(csum32, (__u8 *)gdp, offset); 3278 csum32 = ext4_chksum(csum32, (__u8 *)&dummy_csum, 3279 sizeof(dummy_csum)); 3280 offset += sizeof(dummy_csum); 3281 if (offset < sbi->s_desc_size) 3282 csum32 = ext4_chksum(csum32, (__u8 *)gdp + offset, 3283 sbi->s_desc_size - offset); 3284 3285 crc = csum32 & 0xFFFF; 3286 goto out; 3287 } 3288 3289 /* old crc16 code */ 3290 if (!ext4_has_feature_gdt_csum(sb)) 3291 return 0; 3292 3293 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3294 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3295 crc = crc16(crc, (__u8 *)gdp, offset); 3296 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3297 /* for checksum of struct ext4_group_desc do the rest...*/ 3298 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size) 3299 crc = crc16(crc, (__u8 *)gdp + offset, 3300 sbi->s_desc_size - offset); 3301 3302 out: 3303 return cpu_to_le16(crc); 3304 } 3305 3306 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3307 struct ext4_group_desc *gdp) 3308 { 3309 if (ext4_has_group_desc_csum(sb) && 3310 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3311 return 0; 3312 3313 return 1; 3314 } 3315 3316 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3317 struct ext4_group_desc *gdp) 3318 { 3319 if (!ext4_has_group_desc_csum(sb)) 3320 return; 3321 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3322 } 3323 3324 /* Called at mount-time, super-block is locked */ 3325 static int ext4_check_descriptors(struct super_block *sb, 3326 ext4_fsblk_t sb_block, 3327 ext4_group_t *first_not_zeroed) 3328 { 3329 struct ext4_sb_info *sbi = EXT4_SB(sb); 3330 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3331 ext4_fsblk_t last_block; 3332 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3333 ext4_fsblk_t block_bitmap; 3334 ext4_fsblk_t inode_bitmap; 3335 ext4_fsblk_t inode_table; 3336 int flexbg_flag = 0; 3337 ext4_group_t i, grp = sbi->s_groups_count; 3338 3339 if (ext4_has_feature_flex_bg(sb)) 3340 flexbg_flag = 1; 3341 3342 ext4_debug("Checking group descriptors"); 3343 3344 for (i = 0; i < sbi->s_groups_count; i++) { 3345 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3346 3347 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3348 last_block = ext4_blocks_count(sbi->s_es) - 1; 3349 else 3350 last_block = first_block + 3351 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3352 3353 if ((grp == sbi->s_groups_count) && 3354 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3355 grp = i; 3356 3357 block_bitmap = ext4_block_bitmap(sb, gdp); 3358 if (block_bitmap == sb_block) { 3359 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3360 "Block bitmap for group %u overlaps " 3361 "superblock", i); 3362 if (!sb_rdonly(sb)) 3363 return 0; 3364 } 3365 if (block_bitmap >= sb_block + 1 && 3366 block_bitmap <= last_bg_block) { 3367 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3368 "Block bitmap for group %u overlaps " 3369 "block group descriptors", i); 3370 if (!sb_rdonly(sb)) 3371 return 0; 3372 } 3373 if (block_bitmap < first_block || block_bitmap > last_block) { 3374 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3375 "Block bitmap for group %u not in group " 3376 "(block %llu)!", i, block_bitmap); 3377 return 0; 3378 } 3379 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3380 if (inode_bitmap == sb_block) { 3381 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3382 "Inode bitmap for group %u overlaps " 3383 "superblock", i); 3384 if (!sb_rdonly(sb)) 3385 return 0; 3386 } 3387 if (inode_bitmap >= sb_block + 1 && 3388 inode_bitmap <= last_bg_block) { 3389 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3390 "Inode bitmap for group %u overlaps " 3391 "block group descriptors", i); 3392 if (!sb_rdonly(sb)) 3393 return 0; 3394 } 3395 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3396 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3397 "Inode bitmap for group %u not in group " 3398 "(block %llu)!", i, inode_bitmap); 3399 return 0; 3400 } 3401 inode_table = ext4_inode_table(sb, gdp); 3402 if (inode_table == sb_block) { 3403 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3404 "Inode table for group %u overlaps " 3405 "superblock", i); 3406 if (!sb_rdonly(sb)) 3407 return 0; 3408 } 3409 if (inode_table >= sb_block + 1 && 3410 inode_table <= last_bg_block) { 3411 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3412 "Inode table for group %u overlaps " 3413 "block group descriptors", i); 3414 if (!sb_rdonly(sb)) 3415 return 0; 3416 } 3417 if (inode_table < first_block || 3418 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3419 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3420 "Inode table for group %u not in group " 3421 "(block %llu)!", i, inode_table); 3422 return 0; 3423 } 3424 ext4_lock_group(sb, i); 3425 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3426 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3427 "Checksum for group %u failed (%u!=%u)", 3428 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3429 gdp)), le16_to_cpu(gdp->bg_checksum)); 3430 if (!sb_rdonly(sb)) { 3431 ext4_unlock_group(sb, i); 3432 return 0; 3433 } 3434 } 3435 ext4_unlock_group(sb, i); 3436 if (!flexbg_flag) 3437 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3438 } 3439 if (NULL != first_not_zeroed) 3440 *first_not_zeroed = grp; 3441 return 1; 3442 } 3443 3444 /* 3445 * Maximal extent format file size. 3446 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3447 * extent format containers, within a sector_t, and within i_blocks 3448 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3449 * so that won't be a limiting factor. 3450 * 3451 * However there is other limiting factor. We do store extents in the form 3452 * of starting block and length, hence the resulting length of the extent 3453 * covering maximum file size must fit into on-disk format containers as 3454 * well. Given that length is always by 1 unit bigger than max unit (because 3455 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3456 * 3457 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3458 */ 3459 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3460 { 3461 loff_t res; 3462 loff_t upper_limit = MAX_LFS_FILESIZE; 3463 3464 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3465 3466 if (!has_huge_files) { 3467 upper_limit = (1LL << 32) - 1; 3468 3469 /* total blocks in file system block size */ 3470 upper_limit >>= (blkbits - 9); 3471 upper_limit <<= blkbits; 3472 } 3473 3474 /* 3475 * 32-bit extent-start container, ee_block. We lower the maxbytes 3476 * by one fs block, so ee_len can cover the extent of maximum file 3477 * size 3478 */ 3479 res = (1LL << 32) - 1; 3480 res <<= blkbits; 3481 3482 /* Sanity check against vm- & vfs- imposed limits */ 3483 if (res > upper_limit) 3484 res = upper_limit; 3485 3486 return res; 3487 } 3488 3489 /* 3490 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3491 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3492 * We need to be 1 filesystem block less than the 2^48 sector limit. 3493 */ 3494 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3495 { 3496 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3497 int meta_blocks; 3498 unsigned int ppb = 1 << (bits - 2); 3499 3500 /* 3501 * This is calculated to be the largest file size for a dense, block 3502 * mapped file such that the file's total number of 512-byte sectors, 3503 * including data and all indirect blocks, does not exceed (2^48 - 1). 3504 * 3505 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3506 * number of 512-byte sectors of the file. 3507 */ 3508 if (!has_huge_files) { 3509 /* 3510 * !has_huge_files or implies that the inode i_block field 3511 * represents total file blocks in 2^32 512-byte sectors == 3512 * size of vfs inode i_blocks * 8 3513 */ 3514 upper_limit = (1LL << 32) - 1; 3515 3516 /* total blocks in file system block size */ 3517 upper_limit >>= (bits - 9); 3518 3519 } else { 3520 /* 3521 * We use 48 bit ext4_inode i_blocks 3522 * With EXT4_HUGE_FILE_FL set the i_blocks 3523 * represent total number of blocks in 3524 * file system block size 3525 */ 3526 upper_limit = (1LL << 48) - 1; 3527 3528 } 3529 3530 /* Compute how many blocks we can address by block tree */ 3531 res += ppb; 3532 res += ppb * ppb; 3533 res += ((loff_t)ppb) * ppb * ppb; 3534 /* Compute how many metadata blocks are needed */ 3535 meta_blocks = 1; 3536 meta_blocks += 1 + ppb; 3537 meta_blocks += 1 + ppb + ppb * ppb; 3538 /* Does block tree limit file size? */ 3539 if (res + meta_blocks <= upper_limit) 3540 goto check_lfs; 3541 3542 res = upper_limit; 3543 /* How many metadata blocks are needed for addressing upper_limit? */ 3544 upper_limit -= EXT4_NDIR_BLOCKS; 3545 /* indirect blocks */ 3546 meta_blocks = 1; 3547 upper_limit -= ppb; 3548 /* double indirect blocks */ 3549 if (upper_limit < ppb * ppb) { 3550 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3551 res -= meta_blocks; 3552 goto check_lfs; 3553 } 3554 meta_blocks += 1 + ppb; 3555 upper_limit -= ppb * ppb; 3556 /* tripple indirect blocks for the rest */ 3557 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3558 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3559 res -= meta_blocks; 3560 check_lfs: 3561 res <<= bits; 3562 if (res > MAX_LFS_FILESIZE) 3563 res = MAX_LFS_FILESIZE; 3564 3565 return res; 3566 } 3567 3568 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3569 ext4_fsblk_t logical_sb_block, int nr) 3570 { 3571 struct ext4_sb_info *sbi = EXT4_SB(sb); 3572 ext4_group_t bg, first_meta_bg; 3573 int has_super = 0; 3574 3575 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3576 3577 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3578 return logical_sb_block + nr + 1; 3579 bg = sbi->s_desc_per_block * nr; 3580 if (ext4_bg_has_super(sb, bg)) 3581 has_super = 1; 3582 3583 /* 3584 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3585 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3586 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3587 * compensate. 3588 */ 3589 if (sb->s_blocksize == 1024 && nr == 0 && 3590 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3591 has_super++; 3592 3593 return (has_super + ext4_group_first_block_no(sb, bg)); 3594 } 3595 3596 /** 3597 * ext4_get_stripe_size: Get the stripe size. 3598 * @sbi: In memory super block info 3599 * 3600 * If we have specified it via mount option, then 3601 * use the mount option value. If the value specified at mount time is 3602 * greater than the blocks per group use the super block value. 3603 * If the super block value is greater than blocks per group return 0. 3604 * Allocator needs it be less than blocks per group. 3605 * 3606 */ 3607 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3608 { 3609 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3610 unsigned long stripe_width = 3611 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3612 int ret; 3613 3614 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3615 ret = sbi->s_stripe; 3616 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3617 ret = stripe_width; 3618 else if (stride && stride <= sbi->s_blocks_per_group) 3619 ret = stride; 3620 else 3621 ret = 0; 3622 3623 /* 3624 * If the stripe width is 1, this makes no sense and 3625 * we set it to 0 to turn off stripe handling code. 3626 */ 3627 if (ret <= 1) 3628 ret = 0; 3629 3630 return ret; 3631 } 3632 3633 /* 3634 * Check whether this filesystem can be mounted based on 3635 * the features present and the RDONLY/RDWR mount requested. 3636 * Returns 1 if this filesystem can be mounted as requested, 3637 * 0 if it cannot be. 3638 */ 3639 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3640 { 3641 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3642 ext4_msg(sb, KERN_ERR, 3643 "Couldn't mount because of " 3644 "unsupported optional features (%x)", 3645 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3646 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3647 return 0; 3648 } 3649 3650 if (!IS_ENABLED(CONFIG_UNICODE) && ext4_has_feature_casefold(sb)) { 3651 ext4_msg(sb, KERN_ERR, 3652 "Filesystem with casefold feature cannot be " 3653 "mounted without CONFIG_UNICODE"); 3654 return 0; 3655 } 3656 3657 if (readonly) 3658 return 1; 3659 3660 if (ext4_has_feature_readonly(sb)) { 3661 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3662 sb->s_flags |= SB_RDONLY; 3663 return 1; 3664 } 3665 3666 /* Check that feature set is OK for a read-write mount */ 3667 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3668 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3669 "unsupported optional features (%x)", 3670 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3671 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3672 return 0; 3673 } 3674 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3675 ext4_msg(sb, KERN_ERR, 3676 "Can't support bigalloc feature without " 3677 "extents feature\n"); 3678 return 0; 3679 } 3680 if (ext4_has_feature_bigalloc(sb) && 3681 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) { 3682 ext4_msg(sb, KERN_WARNING, 3683 "bad geometry: bigalloc file system with non-zero " 3684 "first_data_block\n"); 3685 return 0; 3686 } 3687 3688 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3689 if (!readonly && (ext4_has_feature_quota(sb) || 3690 ext4_has_feature_project(sb))) { 3691 ext4_msg(sb, KERN_ERR, 3692 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3693 return 0; 3694 } 3695 #endif /* CONFIG_QUOTA */ 3696 return 1; 3697 } 3698 3699 /* 3700 * This function is called once a day by default if we have errors logged 3701 * on the file system. 3702 * Use the err_report_sec sysfs attribute to disable or adjust its call 3703 * freequency. 3704 */ 3705 void print_daily_error_info(struct timer_list *t) 3706 { 3707 struct ext4_sb_info *sbi = timer_container_of(sbi, t, s_err_report); 3708 struct super_block *sb = sbi->s_sb; 3709 struct ext4_super_block *es = sbi->s_es; 3710 3711 if (es->s_error_count) 3712 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3713 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3714 le32_to_cpu(es->s_error_count)); 3715 if (es->s_first_error_time) { 3716 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3717 sb->s_id, 3718 ext4_get_tstamp(es, s_first_error_time), 3719 (int) sizeof(es->s_first_error_func), 3720 es->s_first_error_func, 3721 le32_to_cpu(es->s_first_error_line)); 3722 if (es->s_first_error_ino) 3723 printk(KERN_CONT ": inode %u", 3724 le32_to_cpu(es->s_first_error_ino)); 3725 if (es->s_first_error_block) 3726 printk(KERN_CONT ": block %llu", (unsigned long long) 3727 le64_to_cpu(es->s_first_error_block)); 3728 printk(KERN_CONT "\n"); 3729 } 3730 if (es->s_last_error_time) { 3731 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3732 sb->s_id, 3733 ext4_get_tstamp(es, s_last_error_time), 3734 (int) sizeof(es->s_last_error_func), 3735 es->s_last_error_func, 3736 le32_to_cpu(es->s_last_error_line)); 3737 if (es->s_last_error_ino) 3738 printk(KERN_CONT ": inode %u", 3739 le32_to_cpu(es->s_last_error_ino)); 3740 if (es->s_last_error_block) 3741 printk(KERN_CONT ": block %llu", (unsigned long long) 3742 le64_to_cpu(es->s_last_error_block)); 3743 printk(KERN_CONT "\n"); 3744 } 3745 3746 if (sbi->s_err_report_sec) 3747 mod_timer(&sbi->s_err_report, jiffies + secs_to_jiffies(sbi->s_err_report_sec)); 3748 } 3749 3750 /* Find next suitable group and run ext4_init_inode_table */ 3751 static int ext4_run_li_request(struct ext4_li_request *elr) 3752 { 3753 struct ext4_group_desc *gdp = NULL; 3754 struct super_block *sb = elr->lr_super; 3755 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3756 ext4_group_t group = elr->lr_next_group; 3757 unsigned int prefetch_ios = 0; 3758 int ret = 0; 3759 int nr = EXT4_SB(sb)->s_mb_prefetch; 3760 u64 start_time; 3761 3762 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3763 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios); 3764 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr); 3765 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr); 3766 if (group >= elr->lr_next_group) { 3767 ret = 1; 3768 if (elr->lr_first_not_zeroed != ngroups && 3769 !ext4_emergency_state(sb) && !sb_rdonly(sb) && 3770 test_opt(sb, INIT_INODE_TABLE)) { 3771 elr->lr_next_group = elr->lr_first_not_zeroed; 3772 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3773 ret = 0; 3774 } 3775 } 3776 return ret; 3777 } 3778 3779 for (; group < ngroups; group++) { 3780 gdp = ext4_get_group_desc(sb, group, NULL); 3781 if (!gdp) { 3782 ret = 1; 3783 break; 3784 } 3785 3786 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3787 break; 3788 } 3789 3790 if (group >= ngroups) 3791 ret = 1; 3792 3793 if (!ret) { 3794 start_time = ktime_get_ns(); 3795 ret = ext4_init_inode_table(sb, group, 3796 elr->lr_timeout ? 0 : 1); 3797 trace_ext4_lazy_itable_init(sb, group); 3798 if (elr->lr_timeout == 0) { 3799 elr->lr_timeout = nsecs_to_jiffies((ktime_get_ns() - start_time) * 3800 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3801 } 3802 elr->lr_next_sched = jiffies + elr->lr_timeout; 3803 elr->lr_next_group = group + 1; 3804 } 3805 return ret; 3806 } 3807 3808 /* 3809 * Remove lr_request from the list_request and free the 3810 * request structure. Should be called with li_list_mtx held 3811 */ 3812 static void ext4_remove_li_request(struct ext4_li_request *elr) 3813 { 3814 if (!elr) 3815 return; 3816 3817 list_del(&elr->lr_request); 3818 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3819 kfree(elr); 3820 } 3821 3822 static void ext4_unregister_li_request(struct super_block *sb) 3823 { 3824 mutex_lock(&ext4_li_mtx); 3825 if (!ext4_li_info) { 3826 mutex_unlock(&ext4_li_mtx); 3827 return; 3828 } 3829 3830 mutex_lock(&ext4_li_info->li_list_mtx); 3831 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3832 mutex_unlock(&ext4_li_info->li_list_mtx); 3833 mutex_unlock(&ext4_li_mtx); 3834 } 3835 3836 static struct task_struct *ext4_lazyinit_task; 3837 3838 /* 3839 * This is the function where ext4lazyinit thread lives. It walks 3840 * through the request list searching for next scheduled filesystem. 3841 * When such a fs is found, run the lazy initialization request 3842 * (ext4_rn_li_request) and keep track of the time spend in this 3843 * function. Based on that time we compute next schedule time of 3844 * the request. When walking through the list is complete, compute 3845 * next waking time and put itself into sleep. 3846 */ 3847 static int ext4_lazyinit_thread(void *arg) 3848 { 3849 struct ext4_lazy_init *eli = arg; 3850 struct list_head *pos, *n; 3851 struct ext4_li_request *elr; 3852 unsigned long next_wakeup, cur; 3853 3854 BUG_ON(NULL == eli); 3855 set_freezable(); 3856 3857 cont_thread: 3858 while (true) { 3859 bool next_wakeup_initialized = false; 3860 3861 next_wakeup = 0; 3862 mutex_lock(&eli->li_list_mtx); 3863 if (list_empty(&eli->li_request_list)) { 3864 mutex_unlock(&eli->li_list_mtx); 3865 goto exit_thread; 3866 } 3867 list_for_each_safe(pos, n, &eli->li_request_list) { 3868 int err = 0; 3869 int progress = 0; 3870 elr = list_entry(pos, struct ext4_li_request, 3871 lr_request); 3872 3873 if (time_before(jiffies, elr->lr_next_sched)) { 3874 if (!next_wakeup_initialized || 3875 time_before(elr->lr_next_sched, next_wakeup)) { 3876 next_wakeup = elr->lr_next_sched; 3877 next_wakeup_initialized = true; 3878 } 3879 continue; 3880 } 3881 if (down_read_trylock(&elr->lr_super->s_umount)) { 3882 if (sb_start_write_trylock(elr->lr_super)) { 3883 progress = 1; 3884 /* 3885 * We hold sb->s_umount, sb can not 3886 * be removed from the list, it is 3887 * now safe to drop li_list_mtx 3888 */ 3889 mutex_unlock(&eli->li_list_mtx); 3890 err = ext4_run_li_request(elr); 3891 sb_end_write(elr->lr_super); 3892 mutex_lock(&eli->li_list_mtx); 3893 n = pos->next; 3894 } 3895 up_read((&elr->lr_super->s_umount)); 3896 } 3897 /* error, remove the lazy_init job */ 3898 if (err) { 3899 ext4_remove_li_request(elr); 3900 continue; 3901 } 3902 if (!progress) { 3903 elr->lr_next_sched = jiffies + 3904 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3905 } 3906 if (!next_wakeup_initialized || 3907 time_before(elr->lr_next_sched, next_wakeup)) { 3908 next_wakeup = elr->lr_next_sched; 3909 next_wakeup_initialized = true; 3910 } 3911 } 3912 mutex_unlock(&eli->li_list_mtx); 3913 3914 try_to_freeze(); 3915 3916 cur = jiffies; 3917 if (!next_wakeup_initialized || time_after_eq(cur, next_wakeup)) { 3918 cond_resched(); 3919 continue; 3920 } 3921 3922 schedule_timeout_interruptible(next_wakeup - cur); 3923 3924 if (kthread_should_stop()) { 3925 ext4_clear_request_list(); 3926 goto exit_thread; 3927 } 3928 } 3929 3930 exit_thread: 3931 /* 3932 * It looks like the request list is empty, but we need 3933 * to check it under the li_list_mtx lock, to prevent any 3934 * additions into it, and of course we should lock ext4_li_mtx 3935 * to atomically free the list and ext4_li_info, because at 3936 * this point another ext4 filesystem could be registering 3937 * new one. 3938 */ 3939 mutex_lock(&ext4_li_mtx); 3940 mutex_lock(&eli->li_list_mtx); 3941 if (!list_empty(&eli->li_request_list)) { 3942 mutex_unlock(&eli->li_list_mtx); 3943 mutex_unlock(&ext4_li_mtx); 3944 goto cont_thread; 3945 } 3946 mutex_unlock(&eli->li_list_mtx); 3947 kfree(ext4_li_info); 3948 ext4_li_info = NULL; 3949 mutex_unlock(&ext4_li_mtx); 3950 3951 return 0; 3952 } 3953 3954 static void ext4_clear_request_list(void) 3955 { 3956 struct list_head *pos, *n; 3957 struct ext4_li_request *elr; 3958 3959 mutex_lock(&ext4_li_info->li_list_mtx); 3960 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3961 elr = list_entry(pos, struct ext4_li_request, 3962 lr_request); 3963 ext4_remove_li_request(elr); 3964 } 3965 mutex_unlock(&ext4_li_info->li_list_mtx); 3966 } 3967 3968 static int ext4_run_lazyinit_thread(void) 3969 { 3970 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3971 ext4_li_info, "ext4lazyinit"); 3972 if (IS_ERR(ext4_lazyinit_task)) { 3973 int err = PTR_ERR(ext4_lazyinit_task); 3974 ext4_clear_request_list(); 3975 kfree(ext4_li_info); 3976 ext4_li_info = NULL; 3977 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3978 "initialization thread\n", 3979 err); 3980 return err; 3981 } 3982 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3983 return 0; 3984 } 3985 3986 /* 3987 * Check whether it make sense to run itable init. thread or not. 3988 * If there is at least one uninitialized inode table, return 3989 * corresponding group number, else the loop goes through all 3990 * groups and return total number of groups. 3991 */ 3992 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3993 { 3994 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3995 struct ext4_group_desc *gdp = NULL; 3996 3997 if (!ext4_has_group_desc_csum(sb)) 3998 return ngroups; 3999 4000 for (group = 0; group < ngroups; group++) { 4001 gdp = ext4_get_group_desc(sb, group, NULL); 4002 if (!gdp) 4003 continue; 4004 4005 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 4006 break; 4007 } 4008 4009 return group; 4010 } 4011 4012 static int ext4_li_info_new(void) 4013 { 4014 struct ext4_lazy_init *eli = NULL; 4015 4016 eli = kzalloc_obj(*eli); 4017 if (!eli) 4018 return -ENOMEM; 4019 4020 INIT_LIST_HEAD(&eli->li_request_list); 4021 mutex_init(&eli->li_list_mtx); 4022 4023 eli->li_state |= EXT4_LAZYINIT_QUIT; 4024 4025 ext4_li_info = eli; 4026 4027 return 0; 4028 } 4029 4030 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 4031 ext4_group_t start) 4032 { 4033 struct ext4_li_request *elr; 4034 4035 elr = kzalloc_obj(*elr); 4036 if (!elr) 4037 return NULL; 4038 4039 elr->lr_super = sb; 4040 elr->lr_first_not_zeroed = start; 4041 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 4042 elr->lr_mode = EXT4_LI_MODE_ITABLE; 4043 elr->lr_next_group = start; 4044 } else { 4045 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 4046 } 4047 4048 /* 4049 * Randomize first schedule time of the request to 4050 * spread the inode table initialization requests 4051 * better. 4052 */ 4053 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 4054 return elr; 4055 } 4056 4057 int ext4_register_li_request(struct super_block *sb, 4058 ext4_group_t first_not_zeroed) 4059 { 4060 struct ext4_sb_info *sbi = EXT4_SB(sb); 4061 struct ext4_li_request *elr = NULL; 4062 ext4_group_t ngroups = sbi->s_groups_count; 4063 int ret = 0; 4064 4065 mutex_lock(&ext4_li_mtx); 4066 if (sbi->s_li_request != NULL) { 4067 /* 4068 * Reset timeout so it can be computed again, because 4069 * s_li_wait_mult might have changed. 4070 */ 4071 sbi->s_li_request->lr_timeout = 0; 4072 goto out; 4073 } 4074 4075 if (ext4_emergency_state(sb) || sb_rdonly(sb) || 4076 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 4077 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 4078 goto out; 4079 4080 elr = ext4_li_request_new(sb, first_not_zeroed); 4081 if (!elr) { 4082 ret = -ENOMEM; 4083 goto out; 4084 } 4085 4086 if (NULL == ext4_li_info) { 4087 ret = ext4_li_info_new(); 4088 if (ret) 4089 goto out; 4090 } 4091 4092 mutex_lock(&ext4_li_info->li_list_mtx); 4093 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 4094 mutex_unlock(&ext4_li_info->li_list_mtx); 4095 4096 sbi->s_li_request = elr; 4097 /* 4098 * set elr to NULL here since it has been inserted to 4099 * the request_list and the removal and free of it is 4100 * handled by ext4_clear_request_list from now on. 4101 */ 4102 elr = NULL; 4103 4104 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 4105 ret = ext4_run_lazyinit_thread(); 4106 if (ret) 4107 goto out; 4108 } 4109 out: 4110 mutex_unlock(&ext4_li_mtx); 4111 if (ret) 4112 kfree(elr); 4113 return ret; 4114 } 4115 4116 /* 4117 * We do not need to lock anything since this is called on 4118 * module unload. 4119 */ 4120 static void ext4_destroy_lazyinit_thread(void) 4121 { 4122 /* 4123 * If thread exited earlier 4124 * there's nothing to be done. 4125 */ 4126 if (!ext4_li_info || !ext4_lazyinit_task) 4127 return; 4128 4129 kthread_stop(ext4_lazyinit_task); 4130 } 4131 4132 static int set_journal_csum_feature_set(struct super_block *sb) 4133 { 4134 int ret = 1; 4135 int compat, incompat; 4136 struct ext4_sb_info *sbi = EXT4_SB(sb); 4137 4138 if (ext4_has_feature_metadata_csum(sb)) { 4139 /* journal checksum v3 */ 4140 compat = 0; 4141 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4142 } else { 4143 /* journal checksum v1 */ 4144 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4145 incompat = 0; 4146 } 4147 4148 jbd2_journal_clear_features(sbi->s_journal, 4149 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4150 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4151 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4152 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4153 ret = jbd2_journal_set_features(sbi->s_journal, 4154 compat, 0, 4155 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4156 incompat); 4157 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4158 ret = jbd2_journal_set_features(sbi->s_journal, 4159 compat, 0, 4160 incompat); 4161 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4162 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4163 } else { 4164 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4165 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4166 } 4167 4168 return ret; 4169 } 4170 4171 /* 4172 * Note: calculating the overhead so we can be compatible with 4173 * historical BSD practice is quite difficult in the face of 4174 * clusters/bigalloc. This is because multiple metadata blocks from 4175 * different block group can end up in the same allocation cluster. 4176 * Calculating the exact overhead in the face of clustered allocation 4177 * requires either O(all block bitmaps) in memory or O(number of block 4178 * groups**2) in time. We will still calculate the superblock for 4179 * older file systems --- and if we come across with a bigalloc file 4180 * system with zero in s_overhead_clusters the estimate will be close to 4181 * correct especially for very large cluster sizes --- but for newer 4182 * file systems, it's better to calculate this figure once at mkfs 4183 * time, and store it in the superblock. If the superblock value is 4184 * present (even for non-bigalloc file systems), we will use it. 4185 */ 4186 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4187 char *buf) 4188 { 4189 struct ext4_sb_info *sbi = EXT4_SB(sb); 4190 struct ext4_group_desc *gdp; 4191 ext4_fsblk_t first_block, last_block, b; 4192 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4193 int s, j, count = 0; 4194 int has_super = ext4_bg_has_super(sb, grp); 4195 4196 if (!ext4_has_feature_bigalloc(sb)) 4197 return (has_super + ext4_bg_num_gdb(sb, grp) + 4198 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4199 sbi->s_itb_per_group + 2); 4200 4201 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4202 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4203 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4204 for (i = 0; i < ngroups; i++) { 4205 gdp = ext4_get_group_desc(sb, i, NULL); 4206 b = ext4_block_bitmap(sb, gdp); 4207 if (b >= first_block && b <= last_block) { 4208 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4209 count++; 4210 } 4211 b = ext4_inode_bitmap(sb, gdp); 4212 if (b >= first_block && b <= last_block) { 4213 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4214 count++; 4215 } 4216 b = ext4_inode_table(sb, gdp); 4217 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4218 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4219 int c = EXT4_B2C(sbi, b - first_block); 4220 ext4_set_bit(c, buf); 4221 count++; 4222 } 4223 if (i != grp) 4224 continue; 4225 s = 0; 4226 if (ext4_bg_has_super(sb, grp)) { 4227 ext4_set_bit(s++, buf); 4228 count++; 4229 } 4230 j = ext4_bg_num_gdb(sb, grp); 4231 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4232 ext4_error(sb, "Invalid number of block group " 4233 "descriptor blocks: %d", j); 4234 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4235 } 4236 count += j; 4237 for (; j > 0; j--) 4238 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4239 } 4240 if (!count) 4241 return 0; 4242 return EXT4_CLUSTERS_PER_GROUP(sb) - 4243 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4244 } 4245 4246 /* 4247 * Compute the overhead and stash it in sbi->s_overhead 4248 */ 4249 int ext4_calculate_overhead(struct super_block *sb) 4250 { 4251 struct ext4_sb_info *sbi = EXT4_SB(sb); 4252 struct ext4_super_block *es = sbi->s_es; 4253 struct inode *j_inode; 4254 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4255 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4256 ext4_fsblk_t overhead = 0; 4257 char *buf = kvmalloc(sb->s_blocksize, GFP_NOFS | __GFP_ZERO); 4258 4259 if (!buf) 4260 return -ENOMEM; 4261 4262 /* 4263 * Compute the overhead (FS structures). This is constant 4264 * for a given filesystem unless the number of block groups 4265 * changes so we cache the previous value until it does. 4266 */ 4267 4268 /* 4269 * All of the blocks before first_data_block are overhead 4270 */ 4271 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4272 4273 /* 4274 * Add the overhead found in each block group 4275 */ 4276 for (i = 0; i < ngroups; i++) { 4277 int blks; 4278 4279 blks = count_overhead(sb, i, buf); 4280 overhead += blks; 4281 if (blks) 4282 memset(buf, 0, sb->s_blocksize); 4283 cond_resched(); 4284 } 4285 4286 /* 4287 * Add the internal journal blocks whether the journal has been 4288 * loaded or not 4289 */ 4290 if (sbi->s_journal && !sbi->s_journal_bdev_file) 4291 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4292 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4293 /* j_inum for internal journal is non-zero */ 4294 j_inode = ext4_get_journal_inode(sb, j_inum); 4295 if (!IS_ERR(j_inode)) { 4296 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4297 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4298 iput(j_inode); 4299 } else { 4300 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4301 } 4302 } 4303 sbi->s_overhead = overhead; 4304 smp_wmb(); 4305 kvfree(buf); 4306 return 0; 4307 } 4308 4309 static void ext4_set_resv_clusters(struct super_block *sb) 4310 { 4311 ext4_fsblk_t resv_clusters; 4312 struct ext4_sb_info *sbi = EXT4_SB(sb); 4313 4314 /* 4315 * There's no need to reserve anything when we aren't using extents. 4316 * The space estimates are exact, there are no unwritten extents, 4317 * hole punching doesn't need new metadata... This is needed especially 4318 * to keep ext2/3 backward compatibility. 4319 */ 4320 if (!ext4_has_feature_extents(sb)) 4321 return; 4322 /* 4323 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4324 * This should cover the situations where we can not afford to run 4325 * out of space like for example punch hole, or converting 4326 * unwritten extents in delalloc path. In most cases such 4327 * allocation would require 1, or 2 blocks, higher numbers are 4328 * very rare. 4329 */ 4330 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4331 sbi->s_cluster_bits); 4332 4333 do_div(resv_clusters, 50); 4334 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4335 4336 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4337 } 4338 4339 static const char *ext4_quota_mode(struct super_block *sb) 4340 { 4341 #ifdef CONFIG_QUOTA 4342 if (!ext4_quota_capable(sb)) 4343 return "none"; 4344 4345 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4346 return "journalled"; 4347 else 4348 return "writeback"; 4349 #else 4350 return "disabled"; 4351 #endif 4352 } 4353 4354 static void ext4_setup_csum_trigger(struct super_block *sb, 4355 enum ext4_journal_trigger_type type, 4356 void (*trigger)( 4357 struct jbd2_buffer_trigger_type *type, 4358 struct buffer_head *bh, 4359 void *mapped_data, 4360 size_t size)) 4361 { 4362 struct ext4_sb_info *sbi = EXT4_SB(sb); 4363 4364 sbi->s_journal_triggers[type].sb = sb; 4365 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4366 } 4367 4368 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4369 { 4370 if (!sbi) 4371 return; 4372 4373 kfree(sbi->s_blockgroup_lock); 4374 fs_put_dax(sbi->s_daxdev, NULL); 4375 kfree(sbi); 4376 } 4377 4378 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4379 { 4380 struct ext4_sb_info *sbi; 4381 4382 sbi = kzalloc_obj(*sbi); 4383 if (!sbi) 4384 return NULL; 4385 4386 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4387 NULL, NULL); 4388 4389 sbi->s_blockgroup_lock = 4390 kzalloc_obj(struct blockgroup_lock); 4391 4392 if (!sbi->s_blockgroup_lock) 4393 goto err_out; 4394 4395 sb->s_fs_info = sbi; 4396 sbi->s_sb = sb; 4397 return sbi; 4398 err_out: 4399 fs_put_dax(sbi->s_daxdev, NULL); 4400 kfree(sbi); 4401 return NULL; 4402 } 4403 4404 static void ext4_set_def_opts(struct super_block *sb, 4405 struct ext4_super_block *es) 4406 { 4407 unsigned long def_mount_opts; 4408 4409 /* Set defaults before we parse the mount options */ 4410 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4411 set_opt(sb, INIT_INODE_TABLE); 4412 if (def_mount_opts & EXT4_DEFM_DEBUG) 4413 set_opt(sb, DEBUG); 4414 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4415 set_opt(sb, GRPID); 4416 if (def_mount_opts & EXT4_DEFM_UID16) 4417 set_opt(sb, NO_UID32); 4418 /* xattr user namespace & acls are now defaulted on */ 4419 set_opt(sb, XATTR_USER); 4420 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4421 set_opt(sb, POSIX_ACL); 4422 #endif 4423 if (ext4_has_feature_fast_commit(sb)) 4424 set_opt2(sb, JOURNAL_FAST_COMMIT); 4425 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4426 if (ext4_has_feature_metadata_csum(sb)) 4427 set_opt(sb, JOURNAL_CHECKSUM); 4428 4429 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4430 set_opt(sb, JOURNAL_DATA); 4431 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4432 set_opt(sb, ORDERED_DATA); 4433 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4434 set_opt(sb, WRITEBACK_DATA); 4435 4436 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4437 set_opt(sb, ERRORS_PANIC); 4438 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4439 set_opt(sb, ERRORS_CONT); 4440 else 4441 set_opt(sb, ERRORS_RO); 4442 /* block_validity enabled by default; disable with noblock_validity */ 4443 set_opt(sb, BLOCK_VALIDITY); 4444 if (def_mount_opts & EXT4_DEFM_DISCARD) 4445 set_opt(sb, DISCARD); 4446 4447 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4448 set_opt(sb, BARRIER); 4449 4450 /* 4451 * enable delayed allocation by default 4452 * Use -o nodelalloc to turn it off 4453 */ 4454 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4455 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4456 set_opt(sb, DELALLOC); 4457 4458 set_opt(sb, DIOREAD_NOLOCK); 4459 } 4460 4461 static int ext4_handle_clustersize(struct super_block *sb) 4462 { 4463 struct ext4_sb_info *sbi = EXT4_SB(sb); 4464 struct ext4_super_block *es = sbi->s_es; 4465 int clustersize; 4466 4467 /* Handle clustersize */ 4468 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4469 if (ext4_has_feature_bigalloc(sb)) { 4470 if (clustersize < sb->s_blocksize) { 4471 ext4_msg(sb, KERN_ERR, 4472 "cluster size (%d) smaller than " 4473 "block size (%lu)", clustersize, sb->s_blocksize); 4474 return -EINVAL; 4475 } 4476 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4477 le32_to_cpu(es->s_log_block_size); 4478 } else { 4479 if (clustersize != sb->s_blocksize) { 4480 ext4_msg(sb, KERN_ERR, 4481 "fragment/cluster size (%d) != " 4482 "block size (%lu)", clustersize, sb->s_blocksize); 4483 return -EINVAL; 4484 } 4485 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4486 ext4_msg(sb, KERN_ERR, 4487 "#blocks per group too big: %lu", 4488 sbi->s_blocks_per_group); 4489 return -EINVAL; 4490 } 4491 sbi->s_cluster_bits = 0; 4492 } 4493 sbi->s_clusters_per_group = le32_to_cpu(es->s_clusters_per_group); 4494 if (sbi->s_clusters_per_group > sb->s_blocksize * 8 || 4495 sbi->s_clusters_per_group & 7) { 4496 ext4_msg(sb, KERN_ERR, "invalid #clusters per group: %lu", 4497 sbi->s_clusters_per_group); 4498 return -EINVAL; 4499 } 4500 if (sbi->s_blocks_per_group != 4501 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4502 ext4_msg(sb, KERN_ERR, 4503 "blocks per group (%lu) and clusters per group (%lu) inconsistent", 4504 sbi->s_blocks_per_group, sbi->s_clusters_per_group); 4505 return -EINVAL; 4506 } 4507 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4508 4509 /* Do we have standard group size of clustersize * 8 blocks ? */ 4510 if (sbi->s_blocks_per_group == clustersize << 3) 4511 set_opt2(sb, STD_GROUP_SIZE); 4512 4513 return 0; 4514 } 4515 4516 /* 4517 * ext4_atomic_write_init: Initializes filesystem min & max atomic write units. 4518 * With non-bigalloc filesystem awu will be based upon filesystem blocksize 4519 * & bdev awu units. 4520 * With bigalloc it will be based upon bigalloc cluster size & bdev awu units. 4521 * @sb: super block 4522 */ 4523 static void ext4_atomic_write_init(struct super_block *sb) 4524 { 4525 struct ext4_sb_info *sbi = EXT4_SB(sb); 4526 struct block_device *bdev = sb->s_bdev; 4527 unsigned int clustersize = EXT4_CLUSTER_SIZE(sb); 4528 4529 if (!bdev_can_atomic_write(bdev)) 4530 return; 4531 4532 if (!ext4_has_feature_extents(sb)) 4533 return; 4534 4535 sbi->s_awu_min = max(sb->s_blocksize, 4536 bdev_atomic_write_unit_min_bytes(bdev)); 4537 sbi->s_awu_max = min(clustersize, 4538 bdev_atomic_write_unit_max_bytes(bdev)); 4539 if (sbi->s_awu_min && sbi->s_awu_max && 4540 sbi->s_awu_min <= sbi->s_awu_max) { 4541 ext4_msg(sb, KERN_NOTICE, "Supports (experimental) DIO atomic writes awu_min: %u, awu_max: %u", 4542 sbi->s_awu_min, sbi->s_awu_max); 4543 } else { 4544 sbi->s_awu_min = 0; 4545 sbi->s_awu_max = 0; 4546 } 4547 } 4548 4549 static void ext4_fast_commit_init(struct super_block *sb) 4550 { 4551 struct ext4_sb_info *sbi = EXT4_SB(sb); 4552 4553 /* Initialize fast commit stuff */ 4554 atomic_set(&sbi->s_fc_subtid, 0); 4555 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4556 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4557 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4558 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4559 sbi->s_fc_bytes = 0; 4560 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4561 sbi->s_fc_ineligible_tid = 0; 4562 mutex_init(&sbi->s_fc_lock); 4563 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4564 memset(&sbi->s_fc_snap_stats, 0, sizeof(sbi->s_fc_snap_stats)); 4565 sbi->s_fc_replay_state.fc_regions = NULL; 4566 sbi->s_fc_replay_state.fc_regions_size = 0; 4567 sbi->s_fc_replay_state.fc_regions_used = 0; 4568 sbi->s_fc_replay_state.fc_regions_valid = 0; 4569 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4570 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4571 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4572 } 4573 4574 static int ext4_inode_info_init(struct super_block *sb, 4575 struct ext4_super_block *es) 4576 { 4577 struct ext4_sb_info *sbi = EXT4_SB(sb); 4578 4579 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4580 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4581 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4582 } else { 4583 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4584 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4585 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4586 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4587 sbi->s_first_ino); 4588 return -EINVAL; 4589 } 4590 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4591 (!is_power_of_2(sbi->s_inode_size)) || 4592 (sbi->s_inode_size > sb->s_blocksize)) { 4593 ext4_msg(sb, KERN_ERR, 4594 "unsupported inode size: %d", 4595 sbi->s_inode_size); 4596 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4597 return -EINVAL; 4598 } 4599 /* 4600 * i_atime_extra is the last extra field available for 4601 * [acm]times in struct ext4_inode. Checking for that 4602 * field should suffice to ensure we have extra space 4603 * for all three. 4604 */ 4605 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4606 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4607 sb->s_time_gran = 1; 4608 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4609 } else { 4610 sb->s_time_gran = NSEC_PER_SEC; 4611 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4612 } 4613 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4614 } 4615 4616 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4617 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4618 EXT4_GOOD_OLD_INODE_SIZE; 4619 if (ext4_has_feature_extra_isize(sb)) { 4620 unsigned v, max = (sbi->s_inode_size - 4621 EXT4_GOOD_OLD_INODE_SIZE); 4622 4623 v = le16_to_cpu(es->s_want_extra_isize); 4624 if (v > max) { 4625 ext4_msg(sb, KERN_ERR, 4626 "bad s_want_extra_isize: %d", v); 4627 return -EINVAL; 4628 } 4629 if (sbi->s_want_extra_isize < v) 4630 sbi->s_want_extra_isize = v; 4631 4632 v = le16_to_cpu(es->s_min_extra_isize); 4633 if (v > max) { 4634 ext4_msg(sb, KERN_ERR, 4635 "bad s_min_extra_isize: %d", v); 4636 return -EINVAL; 4637 } 4638 if (sbi->s_want_extra_isize < v) 4639 sbi->s_want_extra_isize = v; 4640 } 4641 } 4642 4643 return 0; 4644 } 4645 4646 #if IS_ENABLED(CONFIG_UNICODE) 4647 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4648 { 4649 const struct ext4_sb_encodings *encoding_info; 4650 struct unicode_map *encoding; 4651 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4652 4653 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4654 return 0; 4655 4656 encoding_info = ext4_sb_read_encoding(es); 4657 if (!encoding_info) { 4658 ext4_msg(sb, KERN_ERR, 4659 "Encoding requested by superblock is unknown"); 4660 return -EINVAL; 4661 } 4662 4663 encoding = utf8_load(encoding_info->version); 4664 if (IS_ERR(encoding)) { 4665 ext4_msg(sb, KERN_ERR, 4666 "can't mount with superblock charset: %s-%u.%u.%u " 4667 "not supported by the kernel. flags: 0x%x.", 4668 encoding_info->name, 4669 unicode_major(encoding_info->version), 4670 unicode_minor(encoding_info->version), 4671 unicode_rev(encoding_info->version), 4672 encoding_flags); 4673 return -EINVAL; 4674 } 4675 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4676 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4677 unicode_major(encoding_info->version), 4678 unicode_minor(encoding_info->version), 4679 unicode_rev(encoding_info->version), 4680 encoding_flags); 4681 4682 sb->s_encoding = encoding; 4683 sb->s_encoding_flags = encoding_flags; 4684 4685 return 0; 4686 } 4687 #else 4688 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4689 { 4690 return 0; 4691 } 4692 #endif 4693 4694 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4695 { 4696 struct ext4_sb_info *sbi = EXT4_SB(sb); 4697 4698 /* Warn if metadata_csum and gdt_csum are both set. */ 4699 if (ext4_has_feature_metadata_csum(sb) && 4700 ext4_has_feature_gdt_csum(sb)) 4701 ext4_warning(sb, "metadata_csum and uninit_bg are " 4702 "redundant flags; please run fsck."); 4703 4704 /* Check for a known checksum algorithm */ 4705 if (!ext4_verify_csum_type(sb, es)) { 4706 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4707 "unknown checksum algorithm."); 4708 return -EINVAL; 4709 } 4710 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4711 ext4_orphan_file_block_trigger); 4712 4713 /* Check superblock checksum */ 4714 if (!ext4_superblock_csum_verify(sb, es)) { 4715 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4716 "invalid superblock checksum. Run e2fsck?"); 4717 return -EFSBADCRC; 4718 } 4719 4720 /* Precompute checksum seed for all metadata */ 4721 if (ext4_has_feature_csum_seed(sb)) 4722 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4723 else if (ext4_has_feature_metadata_csum(sb) || 4724 ext4_has_feature_ea_inode(sb)) 4725 sbi->s_csum_seed = ext4_chksum(~0, es->s_uuid, 4726 sizeof(es->s_uuid)); 4727 return 0; 4728 } 4729 4730 static int ext4_check_feature_compatibility(struct super_block *sb, 4731 struct ext4_super_block *es, 4732 int silent) 4733 { 4734 struct ext4_sb_info *sbi = EXT4_SB(sb); 4735 4736 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4737 (ext4_has_compat_features(sb) || 4738 ext4_has_ro_compat_features(sb) || 4739 ext4_has_incompat_features(sb))) 4740 ext4_msg(sb, KERN_WARNING, 4741 "feature flags set on rev 0 fs, " 4742 "running e2fsck is recommended"); 4743 4744 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4745 set_opt2(sb, HURD_COMPAT); 4746 if (ext4_has_feature_64bit(sb)) { 4747 ext4_msg(sb, KERN_ERR, 4748 "The Hurd can't support 64-bit file systems"); 4749 return -EINVAL; 4750 } 4751 4752 /* 4753 * ea_inode feature uses l_i_version field which is not 4754 * available in HURD_COMPAT mode. 4755 */ 4756 if (ext4_has_feature_ea_inode(sb)) { 4757 ext4_msg(sb, KERN_ERR, 4758 "ea_inode feature is not supported for Hurd"); 4759 return -EINVAL; 4760 } 4761 } 4762 4763 if (IS_EXT2_SB(sb)) { 4764 if (ext2_feature_set_ok(sb)) 4765 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4766 "using the ext4 subsystem"); 4767 else { 4768 /* 4769 * If we're probing be silent, if this looks like 4770 * it's actually an ext[34] filesystem. 4771 */ 4772 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4773 return -EINVAL; 4774 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4775 "to feature incompatibilities"); 4776 return -EINVAL; 4777 } 4778 } 4779 4780 if (IS_EXT3_SB(sb)) { 4781 if (ext3_feature_set_ok(sb)) 4782 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4783 "using the ext4 subsystem"); 4784 else { 4785 /* 4786 * If we're probing be silent, if this looks like 4787 * it's actually an ext4 filesystem. 4788 */ 4789 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4790 return -EINVAL; 4791 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4792 "to feature incompatibilities"); 4793 return -EINVAL; 4794 } 4795 } 4796 4797 /* 4798 * Check feature flags regardless of the revision level, since we 4799 * previously didn't change the revision level when setting the flags, 4800 * so there is a chance incompat flags are set on a rev 0 filesystem. 4801 */ 4802 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4803 return -EINVAL; 4804 4805 if (sbi->s_daxdev) { 4806 if (sb->s_blocksize == PAGE_SIZE) 4807 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4808 else 4809 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 4810 } 4811 4812 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4813 if (ext4_has_feature_inline_data(sb)) { 4814 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4815 " that may contain inline data"); 4816 return -EINVAL; 4817 } 4818 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4819 ext4_msg(sb, KERN_ERR, 4820 "DAX unsupported by block device."); 4821 return -EINVAL; 4822 } 4823 } 4824 4825 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4826 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4827 es->s_encryption_level); 4828 return -EINVAL; 4829 } 4830 4831 return 0; 4832 } 4833 4834 static int ext4_check_geometry(struct super_block *sb, 4835 struct ext4_super_block *es) 4836 { 4837 struct ext4_sb_info *sbi = EXT4_SB(sb); 4838 __u64 blocks_count; 4839 int err; 4840 4841 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 4842 ext4_msg(sb, KERN_ERR, 4843 "Number of reserved GDT blocks insanely large: %d", 4844 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4845 return -EINVAL; 4846 } 4847 /* 4848 * Test whether we have more sectors than will fit in sector_t, 4849 * and whether the max offset is addressable by the page cache. 4850 */ 4851 err = generic_check_addressable(sb->s_blocksize_bits, 4852 ext4_blocks_count(es)); 4853 if (err) { 4854 ext4_msg(sb, KERN_ERR, "filesystem" 4855 " too large to mount safely on this system"); 4856 return err; 4857 } 4858 4859 /* check blocks count against device size */ 4860 blocks_count = sb_bdev_nr_blocks(sb); 4861 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4862 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4863 "exceeds size of device (%llu blocks)", 4864 ext4_blocks_count(es), blocks_count); 4865 return -EINVAL; 4866 } 4867 4868 /* 4869 * It makes no sense for the first data block to be beyond the end 4870 * of the filesystem. 4871 */ 4872 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4873 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4874 "block %u is beyond end of filesystem (%llu)", 4875 le32_to_cpu(es->s_first_data_block), 4876 ext4_blocks_count(es)); 4877 return -EINVAL; 4878 } 4879 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4880 (sbi->s_cluster_ratio == 1)) { 4881 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4882 "block is 0 with a 1k block and cluster size"); 4883 return -EINVAL; 4884 } 4885 4886 blocks_count = (ext4_blocks_count(es) - 4887 le32_to_cpu(es->s_first_data_block) + 4888 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4889 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4890 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4891 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4892 "(block count %llu, first data block %u, " 4893 "blocks per group %lu)", blocks_count, 4894 ext4_blocks_count(es), 4895 le32_to_cpu(es->s_first_data_block), 4896 EXT4_BLOCKS_PER_GROUP(sb)); 4897 return -EINVAL; 4898 } 4899 sbi->s_groups_count = blocks_count; 4900 sbi->s_blockfile_groups = min(sbi->s_groups_count, 4901 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4902 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4903 le32_to_cpu(es->s_inodes_count)) { 4904 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4905 le32_to_cpu(es->s_inodes_count), 4906 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4907 return -EINVAL; 4908 } 4909 4910 return 0; 4911 } 4912 4913 static int ext4_group_desc_init(struct super_block *sb, 4914 struct ext4_super_block *es, 4915 ext4_fsblk_t logical_sb_block, 4916 ext4_group_t *first_not_zeroed) 4917 { 4918 struct ext4_sb_info *sbi = EXT4_SB(sb); 4919 unsigned int db_count; 4920 ext4_fsblk_t block; 4921 int i; 4922 4923 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4924 EXT4_DESC_PER_BLOCK(sb); 4925 if (ext4_has_feature_meta_bg(sb)) { 4926 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4927 ext4_msg(sb, KERN_WARNING, 4928 "first meta block group too large: %u " 4929 "(group descriptor block count %u)", 4930 le32_to_cpu(es->s_first_meta_bg), db_count); 4931 return -EINVAL; 4932 } 4933 } 4934 rcu_assign_pointer(sbi->s_group_desc, 4935 kvmalloc_objs(struct buffer_head *, db_count)); 4936 if (sbi->s_group_desc == NULL) { 4937 ext4_msg(sb, KERN_ERR, "not enough memory"); 4938 return -ENOMEM; 4939 } 4940 4941 bgl_lock_init(sbi->s_blockgroup_lock); 4942 4943 /* Pre-read the descriptors into the buffer cache */ 4944 for (i = 0; i < db_count; i++) { 4945 block = descriptor_loc(sb, logical_sb_block, i); 4946 ext4_sb_breadahead_unmovable(sb, block); 4947 } 4948 4949 for (i = 0; i < db_count; i++) { 4950 struct buffer_head *bh; 4951 4952 block = descriptor_loc(sb, logical_sb_block, i); 4953 bh = ext4_sb_bread_unmovable(sb, block); 4954 if (IS_ERR(bh)) { 4955 ext4_msg(sb, KERN_ERR, 4956 "can't read group descriptor %d", i); 4957 sbi->s_gdb_count = i; 4958 return PTR_ERR(bh); 4959 } 4960 rcu_read_lock(); 4961 rcu_dereference(sbi->s_group_desc)[i] = bh; 4962 rcu_read_unlock(); 4963 } 4964 sbi->s_gdb_count = db_count; 4965 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4966 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4967 return -EFSCORRUPTED; 4968 } 4969 4970 return 0; 4971 } 4972 4973 static int ext4_load_and_init_journal(struct super_block *sb, 4974 struct ext4_super_block *es, 4975 struct ext4_fs_context *ctx) 4976 { 4977 struct ext4_sb_info *sbi = EXT4_SB(sb); 4978 int err; 4979 4980 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4981 if (err) 4982 return err; 4983 4984 if (ext4_has_feature_64bit(sb) && 4985 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4986 JBD2_FEATURE_INCOMPAT_64BIT)) { 4987 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4988 goto out; 4989 } 4990 4991 if (!set_journal_csum_feature_set(sb)) { 4992 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4993 "feature set"); 4994 goto out; 4995 } 4996 4997 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4998 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4999 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 5000 ext4_msg(sb, KERN_ERR, 5001 "Failed to set fast commit journal feature"); 5002 goto out; 5003 } 5004 5005 /* We have now updated the journal if required, so we can 5006 * validate the data journaling mode. */ 5007 switch (test_opt(sb, DATA_FLAGS)) { 5008 case 0: 5009 /* No mode set, assume a default based on the journal 5010 * capabilities: ORDERED_DATA if the journal can 5011 * cope, else JOURNAL_DATA 5012 */ 5013 if (jbd2_journal_check_available_features 5014 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 5015 set_opt(sb, ORDERED_DATA); 5016 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 5017 } else { 5018 set_opt(sb, JOURNAL_DATA); 5019 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 5020 } 5021 break; 5022 5023 case EXT4_MOUNT_ORDERED_DATA: 5024 case EXT4_MOUNT_WRITEBACK_DATA: 5025 if (!jbd2_journal_check_available_features 5026 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 5027 ext4_msg(sb, KERN_ERR, "Journal does not support " 5028 "requested data journaling mode"); 5029 goto out; 5030 } 5031 break; 5032 default: 5033 break; 5034 } 5035 5036 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 5037 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5038 ext4_msg(sb, KERN_ERR, "can't mount with " 5039 "journal_async_commit in data=ordered mode"); 5040 goto out; 5041 } 5042 5043 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 5044 5045 sbi->s_journal->j_submit_inode_data_buffers = 5046 ext4_journal_submit_inode_data_buffers; 5047 sbi->s_journal->j_finish_inode_data_buffers = 5048 ext4_journal_finish_inode_data_buffers; 5049 5050 return 0; 5051 5052 out: 5053 ext4_journal_destroy(sbi, sbi->s_journal); 5054 return -EINVAL; 5055 } 5056 5057 static int ext4_check_journal_data_mode(struct super_block *sb) 5058 { 5059 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5060 printk(KERN_WARNING "EXT4-fs: The data=journal mount option is deprecated and support will be removed in January 2028\n"); 5061 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 5062 "data=journal disables delayed allocation, " 5063 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 5064 /* can't mount with both data=journal and dioread_nolock. */ 5065 clear_opt(sb, DIOREAD_NOLOCK); 5066 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5067 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5068 ext4_msg(sb, KERN_ERR, "can't mount with " 5069 "both data=journal and delalloc"); 5070 return -EINVAL; 5071 } 5072 if (test_opt(sb, DAX_ALWAYS)) { 5073 ext4_msg(sb, KERN_ERR, "can't mount with " 5074 "both data=journal and dax"); 5075 return -EINVAL; 5076 } 5077 if (ext4_has_feature_encrypt(sb)) { 5078 ext4_msg(sb, KERN_WARNING, 5079 "encrypted files will use data=ordered " 5080 "instead of data journaling mode"); 5081 } 5082 if (test_opt(sb, DELALLOC)) 5083 clear_opt(sb, DELALLOC); 5084 } else { 5085 sb->s_iflags |= SB_I_CGROUPWB; 5086 } 5087 5088 return 0; 5089 } 5090 5091 static const char *ext4_has_journal_option(struct super_block *sb) 5092 { 5093 struct ext4_sb_info *sbi = EXT4_SB(sb); 5094 5095 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) 5096 return "journal_async_commit"; 5097 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) 5098 return "journal_checksum"; 5099 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 5100 return "commit="; 5101 if (EXT4_MOUNT_DATA_FLAGS & 5102 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) 5103 return "data="; 5104 if (test_opt(sb, DATA_ERR_ABORT)) 5105 return "data_err=abort"; 5106 return NULL; 5107 } 5108 5109 /* 5110 * Limit the maximum folio order to 2048 blocks to prevent overestimation 5111 * of reserve handle credits during the folio writeback in environments 5112 * where the PAGE_SIZE exceeds 4KB. 5113 */ 5114 #define EXT4_MAX_PAGECACHE_ORDER(sb) \ 5115 umin(MAX_PAGECACHE_ORDER, (11 + (sb)->s_blocksize_bits - PAGE_SHIFT)) 5116 static void ext4_set_max_mapping_order(struct super_block *sb) 5117 { 5118 struct ext4_sb_info *sbi = EXT4_SB(sb); 5119 5120 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5121 sbi->s_max_folio_order = sbi->s_min_folio_order; 5122 else 5123 sbi->s_max_folio_order = EXT4_MAX_PAGECACHE_ORDER(sb); 5124 } 5125 5126 static int ext4_check_large_folio(struct super_block *sb) 5127 { 5128 const char *err_str = NULL; 5129 5130 if (ext4_has_feature_encrypt(sb)) 5131 err_str = "encrypt"; 5132 5133 if (!err_str) { 5134 ext4_set_max_mapping_order(sb); 5135 } else if (sb->s_blocksize > PAGE_SIZE) { 5136 ext4_msg(sb, KERN_ERR, "bs(%lu) > ps(%lu) unsupported for %s", 5137 sb->s_blocksize, PAGE_SIZE, err_str); 5138 return -EINVAL; 5139 } 5140 5141 return 0; 5142 } 5143 5144 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 5145 int silent) 5146 { 5147 struct ext4_sb_info *sbi = EXT4_SB(sb); 5148 struct ext4_super_block *es; 5149 ext4_fsblk_t logical_sb_block; 5150 unsigned long offset = 0; 5151 struct buffer_head *bh; 5152 int ret = -EINVAL; 5153 int blocksize; 5154 5155 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 5156 if (!blocksize) { 5157 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 5158 return -EINVAL; 5159 } 5160 5161 /* 5162 * The ext4 superblock will not be buffer aligned for other than 1kB 5163 * block sizes. We need to calculate the offset from buffer start. 5164 */ 5165 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 5166 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5167 offset = do_div(logical_sb_block, blocksize); 5168 } else { 5169 logical_sb_block = sbi->s_sb_block; 5170 } 5171 5172 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5173 if (IS_ERR(bh)) { 5174 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 5175 return PTR_ERR(bh); 5176 } 5177 /* 5178 * Note: s_es must be initialized as soon as possible because 5179 * some ext4 macro-instructions depend on its value 5180 */ 5181 es = (struct ext4_super_block *) (bh->b_data + offset); 5182 sbi->s_es = es; 5183 sb->s_magic = le16_to_cpu(es->s_magic); 5184 if (sb->s_magic != EXT4_SUPER_MAGIC) { 5185 if (!silent) 5186 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5187 goto out; 5188 } 5189 5190 if (le32_to_cpu(es->s_log_block_size) > 5191 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5192 ext4_msg(sb, KERN_ERR, 5193 "Invalid log block size: %u", 5194 le32_to_cpu(es->s_log_block_size)); 5195 goto out; 5196 } 5197 if (le32_to_cpu(es->s_log_cluster_size) > 5198 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5199 ext4_msg(sb, KERN_ERR, 5200 "Invalid log cluster size: %u", 5201 le32_to_cpu(es->s_log_cluster_size)); 5202 goto out; 5203 } 5204 5205 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 5206 5207 /* 5208 * If the default block size is not the same as the real block size, 5209 * we need to reload it. 5210 */ 5211 if (sb->s_blocksize == blocksize) 5212 goto success; 5213 5214 /* 5215 * bh must be released before kill_bdev(), otherwise 5216 * it won't be freed and its page also. kill_bdev() 5217 * is called by sb_set_blocksize(). 5218 */ 5219 brelse(bh); 5220 /* Validate the filesystem blocksize */ 5221 if (!sb_set_blocksize(sb, blocksize)) { 5222 ext4_msg(sb, KERN_ERR, "bad block size %d", 5223 blocksize); 5224 bh = NULL; 5225 goto out; 5226 } 5227 5228 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5229 offset = do_div(logical_sb_block, blocksize); 5230 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5231 if (IS_ERR(bh)) { 5232 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5233 ret = PTR_ERR(bh); 5234 bh = NULL; 5235 goto out; 5236 } 5237 es = (struct ext4_super_block *)(bh->b_data + offset); 5238 sbi->s_es = es; 5239 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5240 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5241 goto out; 5242 } 5243 5244 success: 5245 sbi->s_min_folio_order = get_order(blocksize); 5246 *lsb = logical_sb_block; 5247 sbi->s_sbh = bh; 5248 return 0; 5249 out: 5250 brelse(bh); 5251 return ret; 5252 } 5253 5254 static int ext4_hash_info_init(struct super_block *sb) 5255 { 5256 struct ext4_sb_info *sbi = EXT4_SB(sb); 5257 struct ext4_super_block *es = sbi->s_es; 5258 unsigned int i; 5259 5260 sbi->s_def_hash_version = es->s_def_hash_version; 5261 5262 if (sbi->s_def_hash_version > DX_HASH_LAST) { 5263 ext4_msg(sb, KERN_ERR, 5264 "Invalid default hash set in the superblock"); 5265 return -EINVAL; 5266 } else if (sbi->s_def_hash_version == DX_HASH_SIPHASH) { 5267 ext4_msg(sb, KERN_ERR, 5268 "SIPHASH is not a valid default hash value"); 5269 return -EINVAL; 5270 } 5271 5272 for (i = 0; i < 4; i++) 5273 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5274 5275 if (ext4_has_feature_dir_index(sb)) { 5276 i = le32_to_cpu(es->s_flags); 5277 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5278 sbi->s_hash_unsigned = 3; 5279 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5280 #ifdef __CHAR_UNSIGNED__ 5281 if (!sb_rdonly(sb)) 5282 es->s_flags |= 5283 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5284 sbi->s_hash_unsigned = 3; 5285 #else 5286 if (!sb_rdonly(sb)) 5287 es->s_flags |= 5288 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5289 #endif 5290 } 5291 } 5292 return 0; 5293 } 5294 5295 static int ext4_block_group_meta_init(struct super_block *sb, int silent) 5296 { 5297 struct ext4_sb_info *sbi = EXT4_SB(sb); 5298 struct ext4_super_block *es = sbi->s_es; 5299 int has_huge_files; 5300 5301 has_huge_files = ext4_has_feature_huge_file(sb); 5302 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5303 has_huge_files); 5304 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5305 5306 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5307 if (ext4_has_feature_64bit(sb)) { 5308 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5309 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5310 !is_power_of_2(sbi->s_desc_size)) { 5311 ext4_msg(sb, KERN_ERR, 5312 "unsupported descriptor size %lu", 5313 sbi->s_desc_size); 5314 return -EINVAL; 5315 } 5316 } else 5317 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5318 5319 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5320 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5321 5322 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5323 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5324 if (!silent) 5325 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5326 return -EINVAL; 5327 } 5328 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5329 sbi->s_inodes_per_group > sb->s_blocksize * 8 || 5330 sbi->s_inodes_per_group & 7 || 5331 sbi->s_inodes_per_group % sbi->s_inodes_per_block) { 5332 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu", 5333 sbi->s_inodes_per_group); 5334 return -EINVAL; 5335 } 5336 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5337 sbi->s_inodes_per_block; 5338 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5339 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5340 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5341 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5342 5343 return 0; 5344 } 5345 5346 /* 5347 * It's hard to get stripe aligned blocks if stripe is not aligned with 5348 * cluster, just disable stripe and alert user to simplify code and avoid 5349 * stripe aligned allocation which will rarely succeed. 5350 */ 5351 static bool ext4_is_stripe_incompatible(struct super_block *sb, unsigned long stripe) 5352 { 5353 struct ext4_sb_info *sbi = EXT4_SB(sb); 5354 return (stripe > 0 && sbi->s_cluster_ratio > 1 && 5355 stripe % sbi->s_cluster_ratio != 0); 5356 } 5357 5358 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5359 { 5360 struct ext4_super_block *es = NULL; 5361 struct ext4_sb_info *sbi = EXT4_SB(sb); 5362 ext4_fsblk_t logical_sb_block; 5363 struct inode *root; 5364 int needs_recovery; 5365 int err; 5366 ext4_group_t first_not_zeroed; 5367 struct ext4_fs_context *ctx = fc->fs_private; 5368 int silent = fc->sb_flags & SB_SILENT; 5369 5370 /* Set defaults for the variables that will be set during parsing */ 5371 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5372 ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO; 5373 5374 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5375 sbi->s_sectors_written_start = 5376 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5377 5378 err = ext4_load_super(sb, &logical_sb_block, silent); 5379 if (err) 5380 goto out_fail; 5381 5382 es = sbi->s_es; 5383 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5384 5385 err = ext4_init_metadata_csum(sb, es); 5386 if (err) 5387 goto failed_mount; 5388 5389 ext4_set_def_opts(sb, es); 5390 5391 sbi->s_resuid = make_kuid(&init_user_ns, ext4_get_resuid(es)); 5392 sbi->s_resgid = make_kgid(&init_user_ns, ext4_get_resgid(es)); 5393 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5394 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5395 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5396 sbi->s_sb_update_kb = EXT4_DEF_SB_UPDATE_INTERVAL_KB; 5397 sbi->s_sb_update_sec = EXT4_DEF_SB_UPDATE_INTERVAL_SEC; 5398 5399 /* 5400 * set default s_li_wait_mult for lazyinit, for the case there is 5401 * no mount option specified. 5402 */ 5403 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5404 5405 err = ext4_inode_info_init(sb, es); 5406 if (err) 5407 goto failed_mount; 5408 5409 err = parse_apply_sb_mount_options(sb, ctx); 5410 if (err < 0) 5411 goto failed_mount; 5412 5413 sbi->s_def_mount_opt = sbi->s_mount_opt; 5414 sbi->s_def_mount_opt2 = sbi->s_mount_opt2; 5415 5416 err = ext4_check_opt_consistency(fc, sb); 5417 if (err < 0) 5418 goto failed_mount; 5419 5420 ext4_apply_options(fc, sb); 5421 5422 err = ext4_check_large_folio(sb); 5423 if (err < 0) 5424 goto failed_mount; 5425 5426 err = ext4_encoding_init(sb, es); 5427 if (err) 5428 goto failed_mount; 5429 5430 err = ext4_check_journal_data_mode(sb); 5431 if (err) 5432 goto failed_mount; 5433 5434 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5435 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5436 5437 /* HSM events are allowed by default. */ 5438 sb->s_iflags |= SB_I_ALLOW_HSM; 5439 5440 err = ext4_check_feature_compatibility(sb, es, silent); 5441 if (err) 5442 goto failed_mount; 5443 5444 err = ext4_block_group_meta_init(sb, silent); 5445 if (err) 5446 goto failed_mount; 5447 5448 err = ext4_hash_info_init(sb); 5449 if (err) 5450 goto failed_mount; 5451 5452 err = ext4_handle_clustersize(sb); 5453 if (err) 5454 goto failed_mount; 5455 5456 err = ext4_check_geometry(sb, es); 5457 if (err) 5458 goto failed_mount; 5459 5460 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5461 spin_lock_init(&sbi->s_error_lock); 5462 mutex_init(&sbi->s_error_notify_mutex); 5463 INIT_WORK(&sbi->s_sb_upd_work, update_super_work); 5464 5465 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5466 if (err) 5467 goto failed_mount3; 5468 5469 err = ext4_es_register_shrinker(sbi); 5470 if (err) 5471 goto failed_mount3; 5472 5473 sbi->s_stripe = ext4_get_stripe_size(sbi); 5474 if (ext4_is_stripe_incompatible(sb, sbi->s_stripe)) { 5475 ext4_msg(sb, KERN_WARNING, 5476 "stripe (%lu) is not aligned with cluster size (%u), " 5477 "stripe is disabled", 5478 sbi->s_stripe, sbi->s_cluster_ratio); 5479 sbi->s_stripe = 0; 5480 } 5481 sbi->s_extent_max_zeroout_kb = 32; 5482 5483 /* 5484 * set up enough so that it can read an inode 5485 */ 5486 sb->s_op = &ext4_sops; 5487 sb->s_export_op = &ext4_export_ops; 5488 sb->s_xattr = ext4_xattr_handlers; 5489 #ifdef CONFIG_FS_ENCRYPTION 5490 sb->s_cop = &ext4_cryptops; 5491 #endif 5492 #ifdef CONFIG_FS_VERITY 5493 sb->s_vop = &ext4_verityops; 5494 #endif 5495 #ifdef CONFIG_QUOTA 5496 sb->dq_op = &ext4_quota_operations; 5497 if (ext4_has_feature_quota(sb)) 5498 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5499 else 5500 sb->s_qcop = &ext4_qctl_operations; 5501 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5502 #endif 5503 super_set_uuid(sb, es->s_uuid, sizeof(es->s_uuid)); 5504 super_set_sysfs_name_bdev(sb); 5505 5506 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5507 mutex_init(&sbi->s_orphan_lock); 5508 5509 spin_lock_init(&sbi->s_bdev_wb_lock); 5510 5511 ext4_atomic_write_init(sb); 5512 ext4_fast_commit_init(sb); 5513 5514 sb->s_root = NULL; 5515 5516 needs_recovery = (es->s_last_orphan != 0 || 5517 ext4_has_feature_orphan_present(sb) || 5518 ext4_has_feature_journal_needs_recovery(sb)); 5519 5520 ext4_init_ea_inode_work(sbi); 5521 5522 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) { 5523 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)); 5524 if (err) 5525 goto failed_mount3a; 5526 } 5527 5528 err = -EINVAL; 5529 /* 5530 * The first inode we look at is the journal inode. Don't try 5531 * root first: it may be modified in the journal! 5532 */ 5533 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5534 err = ext4_load_and_init_journal(sb, es, ctx); 5535 if (err) 5536 goto failed_mount3a; 5537 if (bdev_read_only(sb->s_bdev)) 5538 needs_recovery = 0; 5539 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5540 ext4_has_feature_journal_needs_recovery(sb)) { 5541 ext4_msg(sb, KERN_ERR, "required journal recovery " 5542 "suppressed and not mounted read-only"); 5543 goto failed_mount3a; 5544 } else { 5545 const char *journal_option; 5546 5547 /* Nojournal mode, all journal mount options are illegal */ 5548 journal_option = ext4_has_journal_option(sb); 5549 if (journal_option != NULL) { 5550 ext4_msg(sb, KERN_ERR, 5551 "can't mount with %s, fs mounted w/o journal", 5552 journal_option); 5553 goto failed_mount3a; 5554 } 5555 5556 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5557 clear_opt(sb, JOURNAL_CHECKSUM); 5558 clear_opt(sb, DATA_FLAGS); 5559 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5560 sbi->s_journal = NULL; 5561 needs_recovery = 0; 5562 } 5563 5564 if (!test_opt(sb, NO_MBCACHE)) { 5565 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5566 if (!sbi->s_ea_block_cache) { 5567 ext4_msg(sb, KERN_ERR, 5568 "Failed to create ea_block_cache"); 5569 err = -EINVAL; 5570 goto failed_mount_wq; 5571 } 5572 5573 if (ext4_has_feature_ea_inode(sb)) { 5574 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5575 if (!sbi->s_ea_inode_cache) { 5576 ext4_msg(sb, KERN_ERR, 5577 "Failed to create ea_inode_cache"); 5578 err = -EINVAL; 5579 goto failed_mount_wq; 5580 } 5581 } 5582 } 5583 5584 /* 5585 * Get the # of file system overhead blocks from the 5586 * superblock if present. 5587 */ 5588 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5589 /* ignore the precalculated value if it is ridiculous */ 5590 if (sbi->s_overhead > ext4_blocks_count(es)) 5591 sbi->s_overhead = 0; 5592 /* 5593 * If the bigalloc feature is not enabled recalculating the 5594 * overhead doesn't take long, so we might as well just redo 5595 * it to make sure we are using the correct value. 5596 */ 5597 if (!ext4_has_feature_bigalloc(sb)) 5598 sbi->s_overhead = 0; 5599 if (sbi->s_overhead == 0) { 5600 err = ext4_calculate_overhead(sb); 5601 if (err) 5602 goto failed_mount_wq; 5603 } 5604 5605 /* 5606 * The maximum number of concurrent works can be high and 5607 * concurrency isn't really necessary. Limit it to 1. 5608 */ 5609 EXT4_SB(sb)->rsv_conversion_wq = 5610 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5611 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5612 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5613 err = -ENOMEM; 5614 goto failed_mount4; 5615 } 5616 5617 /* 5618 * The jbd2_journal_load will have done any necessary log recovery, 5619 * so we can safely mount the rest of the filesystem now. 5620 */ 5621 5622 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5623 if (IS_ERR(root)) { 5624 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5625 err = PTR_ERR(root); 5626 root = NULL; 5627 goto failed_mount4; 5628 } 5629 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5630 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5631 iput(root); 5632 err = -EFSCORRUPTED; 5633 goto failed_mount4; 5634 } 5635 5636 generic_set_sb_d_ops(sb); 5637 sb->s_root = d_make_root(root); 5638 if (!sb->s_root) { 5639 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5640 err = -ENOMEM; 5641 goto failed_mount4; 5642 } 5643 5644 err = ext4_setup_super(sb, es, sb_rdonly(sb)); 5645 if (err == -EROFS) { 5646 sb->s_flags |= SB_RDONLY; 5647 } else if (err) 5648 goto failed_mount4a; 5649 5650 ext4_set_resv_clusters(sb); 5651 5652 if (test_opt(sb, BLOCK_VALIDITY)) { 5653 err = ext4_setup_system_zone(sb); 5654 if (err) { 5655 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5656 "zone (%d)", err); 5657 goto failed_mount4a; 5658 } 5659 } 5660 ext4_fc_replay_cleanup(sb); 5661 5662 ext4_ext_init(sb); 5663 5664 /* 5665 * Enable optimize_scan if number of groups is > threshold. This can be 5666 * turned off by passing "mb_optimize_scan=0". This can also be 5667 * turned on forcefully by passing "mb_optimize_scan=1". 5668 */ 5669 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5670 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5671 set_opt2(sb, MB_OPTIMIZE_SCAN); 5672 else 5673 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5674 } 5675 5676 err = ext4_percpu_param_init(sbi); 5677 if (err) 5678 goto failed_mount5; 5679 5680 err = ext4_mb_init(sb); 5681 if (err) { 5682 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5683 err); 5684 goto failed_mount5; 5685 } 5686 5687 /* 5688 * We can only set up the journal commit callback once 5689 * mballoc is initialized 5690 */ 5691 if (sbi->s_journal) 5692 sbi->s_journal->j_commit_callback = 5693 ext4_journal_commit_callback; 5694 5695 if (ext4_has_feature_flex_bg(sb)) 5696 if (!ext4_fill_flex_info(sb)) { 5697 ext4_msg(sb, KERN_ERR, 5698 "unable to initialize " 5699 "flex_bg meta info!"); 5700 err = -ENOMEM; 5701 goto failed_mount6; 5702 } 5703 5704 err = ext4_register_li_request(sb, first_not_zeroed); 5705 if (err) 5706 goto failed_mount6; 5707 5708 err = ext4_init_orphan_info(sb); 5709 if (err) 5710 goto failed_mount7; 5711 #ifdef CONFIG_QUOTA 5712 /* Enable quota usage during mount. */ 5713 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5714 err = ext4_enable_quotas(sb); 5715 if (err) 5716 goto failed_mount8; 5717 } 5718 #endif /* CONFIG_QUOTA */ 5719 5720 /* 5721 * Save the original bdev mapping's wb_err value which could be 5722 * used to detect the metadata async write error. 5723 */ 5724 errseq_check_and_advance(&sb->s_bdev->bd_mapping->wb_err, 5725 &sbi->s_bdev_wb_err); 5726 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5727 ext4_orphan_cleanup(sb, es); 5728 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5729 /* 5730 * Update the checksum after updating free space/inode counters and 5731 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5732 * checksum in the buffer cache until it is written out and 5733 * e2fsprogs programs trying to open a file system immediately 5734 * after it is mounted can fail. 5735 */ 5736 ext4_superblock_csum_set(sb); 5737 if (needs_recovery) { 5738 ext4_msg(sb, KERN_INFO, "recovery complete"); 5739 err = ext4_mark_recovery_complete(sb, es); 5740 if (err) 5741 goto failed_mount9; 5742 } 5743 5744 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) { 5745 ext4_msg(sb, KERN_WARNING, 5746 "mounting with \"discard\" option, but the device does not support discard"); 5747 clear_opt(sb, DISCARD); 5748 } 5749 5750 if (es->s_error_count) { 5751 sbi->s_err_report_sec = 5*60; /* first time 5 minutes */ 5752 mod_timer(&sbi->s_err_report, 5753 jiffies + secs_to_jiffies(sbi->s_err_report_sec)); 5754 } 5755 sbi->s_err_report_sec = 24*60*60; /* Once a day */ 5756 5757 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5758 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5759 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5760 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5761 atomic_set(&sbi->s_warning_count, 0); 5762 atomic_set(&sbi->s_msg_count, 0); 5763 5764 /* Register sysfs after all initializations are complete. */ 5765 err = ext4_register_sysfs(sb); 5766 if (err) 5767 goto failed_mount9; 5768 5769 return 0; 5770 5771 failed_mount9: 5772 /* Drain deferred EA inode iputs before quota shutdown */ 5773 flush_delayed_work(&sbi->s_ea_inode_work); 5774 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 5775 failed_mount8: __maybe_unused 5776 ext4_release_orphan_info(sb); 5777 failed_mount7: 5778 ext4_unregister_li_request(sb); 5779 failed_mount6: 5780 ext4_mb_release(sb); 5781 ext4_flex_groups_free(sbi); 5782 failed_mount5: 5783 ext4_percpu_param_destroy(sbi); 5784 ext4_ext_release(sb); 5785 ext4_release_system_zone(sb); 5786 failed_mount4a: 5787 dput(sb->s_root); 5788 sb->s_root = NULL; 5789 failed_mount4: 5790 ext4_msg(sb, KERN_ERR, "mount failed"); 5791 if (EXT4_SB(sb)->rsv_conversion_wq) 5792 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5793 failed_mount_wq: 5794 /* Drain deferred EA inode iputs before freeing structures */ 5795 flush_delayed_work(&sbi->s_ea_inode_work); 5796 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5797 sbi->s_ea_inode_cache = NULL; 5798 5799 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5800 sbi->s_ea_block_cache = NULL; 5801 5802 if (sbi->s_journal) { 5803 ext4_journal_destroy(sbi, sbi->s_journal); 5804 } 5805 failed_mount3a: 5806 /* Drain deferred EA inode iputs from journal replay */ 5807 flush_delayed_work(&sbi->s_ea_inode_work); 5808 ext4_es_unregister_shrinker(sbi); 5809 failed_mount3: 5810 /* flush s_sb_upd_work before sbi destroy */ 5811 flush_work(&sbi->s_sb_upd_work); 5812 ext4_stop_mmpd(sbi); 5813 timer_delete_sync(&sbi->s_err_report); 5814 ext4_group_desc_free(sbi); 5815 failed_mount: 5816 #if IS_ENABLED(CONFIG_UNICODE) 5817 utf8_unload(sb->s_encoding); 5818 #endif 5819 5820 #ifdef CONFIG_QUOTA 5821 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++) 5822 kfree(get_qf_name(sb, sbi, i)); 5823 #endif 5824 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5825 brelse(sbi->s_sbh); 5826 if (sbi->s_journal_bdev_file) { 5827 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 5828 fs_bdev_file_release(sbi->s_journal_bdev_file, sb); 5829 } 5830 out_fail: 5831 invalidate_bdev(sb->s_bdev); 5832 sb->s_fs_info = NULL; 5833 return err; 5834 } 5835 5836 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5837 { 5838 struct ext4_fs_context *ctx = fc->fs_private; 5839 struct ext4_sb_info *sbi; 5840 const char *descr; 5841 int ret; 5842 5843 sbi = ext4_alloc_sbi(sb); 5844 if (!sbi) 5845 return -ENOMEM; 5846 5847 fc->s_fs_info = sbi; 5848 5849 /* Cleanup superblock name */ 5850 strreplace(sb->s_id, '/', '!'); 5851 5852 sbi->s_sb_block = 1; /* Default super block location */ 5853 if (ctx->spec & EXT4_SPEC_s_sb_block) 5854 sbi->s_sb_block = ctx->s_sb_block; 5855 5856 ret = __ext4_fill_super(fc, sb); 5857 if (ret < 0) 5858 goto free_sbi; 5859 5860 if (sbi->s_journal) { 5861 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5862 descr = " journalled data mode"; 5863 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5864 descr = " ordered data mode"; 5865 else 5866 descr = " writeback data mode"; 5867 } else 5868 descr = "out journal"; 5869 5870 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5871 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. " 5872 "Quota mode: %s.", &sb->s_uuid, 5873 sb_rdonly(sb) ? "ro" : "r/w", descr, 5874 ext4_quota_mode(sb)); 5875 5876 /* Update the s_overhead_clusters if necessary */ 5877 ext4_update_overhead(sb, false); 5878 return 0; 5879 5880 free_sbi: 5881 ext4_free_sbi(sbi); 5882 fc->s_fs_info = NULL; 5883 return ret; 5884 } 5885 5886 static int ext4_get_tree(struct fs_context *fc) 5887 { 5888 return get_tree_bdev(fc, ext4_fill_super); 5889 } 5890 5891 /* 5892 * Setup any per-fs journal parameters now. We'll do this both on 5893 * initial mount, once the journal has been initialised but before we've 5894 * done any recovery; and again on any subsequent remount. 5895 */ 5896 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5897 { 5898 struct ext4_sb_info *sbi = EXT4_SB(sb); 5899 5900 journal->j_commit_interval = sbi->s_commit_interval; 5901 journal->j_min_batch_time = sbi->s_min_batch_time; 5902 journal->j_max_batch_time = sbi->s_max_batch_time; 5903 ext4_fc_init(sb, journal); 5904 5905 write_lock(&journal->j_state_lock); 5906 if (test_opt(sb, BARRIER)) 5907 journal->j_flags |= JBD2_BARRIER; 5908 else 5909 journal->j_flags &= ~JBD2_BARRIER; 5910 /* 5911 * Always enable journal cycle record option, letting the journal 5912 * records log transactions continuously between each mount. 5913 */ 5914 journal->j_flags |= JBD2_CYCLE_RECORD; 5915 write_unlock(&journal->j_state_lock); 5916 } 5917 5918 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5919 unsigned int journal_inum) 5920 { 5921 struct inode *journal_inode; 5922 5923 /* 5924 * Test for the existence of a valid inode on disk. Bad things 5925 * happen if we iget() an unused inode, as the subsequent iput() 5926 * will try to delete it. 5927 */ 5928 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5929 if (IS_ERR(journal_inode)) { 5930 ext4_msg(sb, KERN_ERR, "no journal found"); 5931 return ERR_CAST(journal_inode); 5932 } 5933 if (!journal_inode->i_nlink) { 5934 make_bad_inode(journal_inode); 5935 iput(journal_inode); 5936 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5937 return ERR_PTR(-EFSCORRUPTED); 5938 } 5939 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) { 5940 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5941 iput(journal_inode); 5942 return ERR_PTR(-EFSCORRUPTED); 5943 } 5944 5945 #ifdef CONFIG_LOCKDEP 5946 lockdep_set_subclass(&EXT4_I(journal_inode)->i_data_sem, 5947 I_DATA_SEM_JOURNAL); 5948 #endif 5949 5950 ext4_debug("Journal inode found at %p: %lld bytes\n", 5951 journal_inode, journal_inode->i_size); 5952 return journal_inode; 5953 } 5954 5955 static int ext4_journal_bmap(journal_t *journal, sector_t *block) 5956 { 5957 struct ext4_map_blocks map; 5958 int ret; 5959 5960 if (journal->j_inode == NULL) 5961 return 0; 5962 5963 map.m_lblk = *block; 5964 map.m_len = 1; 5965 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0); 5966 if (ret <= 0) { 5967 ext4_msg(journal->j_inode->i_sb, KERN_CRIT, 5968 "journal bmap failed: block %llu ret %d\n", 5969 *block, ret); 5970 jbd2_journal_abort(journal, ret ? ret : -EFSCORRUPTED); 5971 return ret; 5972 } 5973 *block = map.m_pblk; 5974 return 0; 5975 } 5976 5977 static journal_t *ext4_open_inode_journal(struct super_block *sb, 5978 unsigned int journal_inum) 5979 { 5980 struct inode *journal_inode; 5981 journal_t *journal; 5982 5983 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5984 if (IS_ERR(journal_inode)) 5985 return ERR_CAST(journal_inode); 5986 5987 journal = jbd2_journal_init_inode(journal_inode); 5988 if (IS_ERR(journal)) { 5989 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5990 iput(journal_inode); 5991 return ERR_CAST(journal); 5992 } 5993 journal->j_private = sb; 5994 journal->j_bmap = ext4_journal_bmap; 5995 ext4_init_journal_params(sb, journal); 5996 return journal; 5997 } 5998 5999 static struct file *ext4_get_journal_blkdev(struct super_block *sb, 6000 dev_t j_dev, ext4_fsblk_t *j_start, 6001 ext4_fsblk_t *j_len) 6002 { 6003 struct buffer_head *bh; 6004 struct block_device *bdev; 6005 struct file *bdev_file; 6006 int hblock, blocksize; 6007 ext4_fsblk_t sb_block; 6008 unsigned long offset; 6009 struct ext4_super_block *es; 6010 int errno; 6011 6012 bdev_file = fs_bdev_file_open_by_dev(j_dev, 6013 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES, 6014 sb, sb); 6015 if (IS_ERR(bdev_file)) { 6016 ext4_msg(sb, KERN_ERR, 6017 "failed to open journal device unknown-block(%u,%u) %pe", 6018 MAJOR(j_dev), MINOR(j_dev), bdev_file); 6019 return bdev_file; 6020 } 6021 6022 bdev = file_bdev(bdev_file); 6023 blocksize = sb->s_blocksize; 6024 hblock = bdev_logical_block_size(bdev); 6025 if (blocksize < hblock) { 6026 ext4_msg(sb, KERN_ERR, 6027 "blocksize too small for journal device"); 6028 errno = -EINVAL; 6029 goto out_bdev; 6030 } 6031 6032 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 6033 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 6034 set_blocksize(bdev_file, blocksize); 6035 bh = __bread(bdev, sb_block, blocksize); 6036 if (!bh) { 6037 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 6038 "external journal"); 6039 errno = -EINVAL; 6040 goto out_bdev; 6041 } 6042 6043 es = (struct ext4_super_block *) (bh->b_data + offset); 6044 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 6045 !(le32_to_cpu(es->s_feature_incompat) & 6046 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 6047 ext4_msg(sb, KERN_ERR, "external journal has bad superblock"); 6048 errno = -EFSCORRUPTED; 6049 goto out_bh; 6050 } 6051 6052 if ((le32_to_cpu(es->s_feature_ro_compat) & 6053 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 6054 es->s_checksum != ext4_superblock_csum(es)) { 6055 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock"); 6056 errno = -EFSCORRUPTED; 6057 goto out_bh; 6058 } 6059 6060 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 6061 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 6062 errno = -EFSCORRUPTED; 6063 goto out_bh; 6064 } 6065 6066 *j_start = sb_block + 1; 6067 *j_len = ext4_blocks_count(es); 6068 brelse(bh); 6069 return bdev_file; 6070 6071 out_bh: 6072 brelse(bh); 6073 out_bdev: 6074 fs_bdev_file_release(bdev_file, sb); 6075 return ERR_PTR(errno); 6076 } 6077 6078 static journal_t *ext4_open_dev_journal(struct super_block *sb, 6079 dev_t j_dev) 6080 { 6081 journal_t *journal; 6082 ext4_fsblk_t j_start; 6083 ext4_fsblk_t j_len; 6084 struct file *bdev_file; 6085 int errno = 0; 6086 6087 bdev_file = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len); 6088 if (IS_ERR(bdev_file)) 6089 return ERR_CAST(bdev_file); 6090 6091 journal = jbd2_journal_init_dev(file_bdev(bdev_file), sb->s_bdev, j_start, 6092 j_len, sb->s_blocksize); 6093 if (IS_ERR(journal)) { 6094 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 6095 errno = PTR_ERR(journal); 6096 goto out_bdev; 6097 } 6098 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 6099 ext4_msg(sb, KERN_ERR, "External journal has more than one " 6100 "user (unsupported) - %d", 6101 be32_to_cpu(journal->j_superblock->s_nr_users)); 6102 errno = -EINVAL; 6103 goto out_journal; 6104 } 6105 journal->j_private = sb; 6106 EXT4_SB(sb)->s_journal_bdev_file = bdev_file; 6107 ext4_init_journal_params(sb, journal); 6108 return journal; 6109 6110 out_journal: 6111 ext4_journal_destroy(EXT4_SB(sb), journal); 6112 out_bdev: 6113 fs_bdev_file_release(bdev_file, sb); 6114 return ERR_PTR(errno); 6115 } 6116 6117 static int ext4_load_journal(struct super_block *sb, 6118 struct ext4_super_block *es, 6119 unsigned long journal_devnum) 6120 { 6121 journal_t *journal; 6122 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 6123 dev_t journal_dev; 6124 int err = 0; 6125 int really_read_only; 6126 int journal_dev_ro; 6127 6128 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 6129 return -EFSCORRUPTED; 6130 6131 if (journal_devnum && 6132 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6133 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 6134 "numbers have changed"); 6135 journal_dev = new_decode_dev(journal_devnum); 6136 } else 6137 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 6138 6139 if (journal_inum && journal_dev) { 6140 ext4_msg(sb, KERN_ERR, 6141 "filesystem has both journal inode and journal device!"); 6142 return -EINVAL; 6143 } 6144 6145 if (journal_inum) { 6146 journal = ext4_open_inode_journal(sb, journal_inum); 6147 if (IS_ERR(journal)) 6148 return PTR_ERR(journal); 6149 } else { 6150 journal = ext4_open_dev_journal(sb, journal_dev); 6151 if (IS_ERR(journal)) 6152 return PTR_ERR(journal); 6153 } 6154 6155 journal_dev_ro = bdev_read_only(journal->j_dev); 6156 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 6157 6158 if (journal_dev_ro && !sb_rdonly(sb)) { 6159 ext4_msg(sb, KERN_ERR, 6160 "journal device read-only, try mounting with '-o ro'"); 6161 err = -EROFS; 6162 goto err_out; 6163 } 6164 6165 /* 6166 * Are we loading a blank journal or performing recovery after a 6167 * crash? For recovery, we need to check in advance whether we 6168 * can get read-write access to the device. 6169 */ 6170 if (ext4_has_feature_journal_needs_recovery(sb)) { 6171 if (sb_rdonly(sb)) { 6172 ext4_msg(sb, KERN_INFO, "INFO: recovery " 6173 "required on readonly filesystem"); 6174 if (really_read_only) { 6175 ext4_msg(sb, KERN_ERR, "write access " 6176 "unavailable, cannot proceed " 6177 "(try mounting with noload)"); 6178 err = -EROFS; 6179 goto err_out; 6180 } 6181 ext4_msg(sb, KERN_INFO, "write access will " 6182 "be enabled during recovery"); 6183 } 6184 } 6185 6186 if (!(journal->j_flags & JBD2_BARRIER)) 6187 ext4_msg(sb, KERN_INFO, "barriers disabled"); 6188 6189 if (!ext4_has_feature_journal_needs_recovery(sb)) 6190 err = jbd2_journal_wipe(journal, !really_read_only); 6191 if (!err) { 6192 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 6193 __le16 orig_state; 6194 bool changed = false; 6195 6196 if (save) 6197 memcpy(save, ((char *) es) + 6198 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 6199 err = jbd2_journal_load(journal); 6200 if (save && memcmp(((char *) es) + EXT4_S_ERR_START, 6201 save, EXT4_S_ERR_LEN)) { 6202 memcpy(((char *) es) + EXT4_S_ERR_START, 6203 save, EXT4_S_ERR_LEN); 6204 changed = true; 6205 } 6206 kfree(save); 6207 orig_state = es->s_state; 6208 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state & 6209 EXT4_ERROR_FS); 6210 if (orig_state != es->s_state) 6211 changed = true; 6212 /* Write out restored error information to the superblock */ 6213 if (changed && !really_read_only) { 6214 int err2; 6215 err2 = ext4_commit_super(sb); 6216 err = err ? : err2; 6217 } 6218 } 6219 6220 if (err) { 6221 ext4_msg(sb, KERN_ERR, "error loading journal"); 6222 goto err_out; 6223 } 6224 6225 EXT4_SB(sb)->s_journal = journal; 6226 err = ext4_clear_journal_err(sb, es); 6227 if (err) { 6228 ext4_journal_destroy(EXT4_SB(sb), journal); 6229 return err; 6230 } 6231 6232 if (!really_read_only && journal_devnum && 6233 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6234 es->s_journal_dev = cpu_to_le32(journal_devnum); 6235 ext4_commit_super(sb); 6236 } 6237 if (!really_read_only && journal_inum && 6238 journal_inum != le32_to_cpu(es->s_journal_inum)) { 6239 es->s_journal_inum = cpu_to_le32(journal_inum); 6240 ext4_commit_super(sb); 6241 } 6242 6243 return 0; 6244 6245 err_out: 6246 ext4_journal_destroy(EXT4_SB(sb), journal); 6247 return err; 6248 } 6249 6250 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 6251 static void ext4_update_super(struct super_block *sb) 6252 { 6253 struct ext4_sb_info *sbi = EXT4_SB(sb); 6254 struct ext4_super_block *es = sbi->s_es; 6255 struct buffer_head *sbh = sbi->s_sbh; 6256 6257 lock_buffer(sbh); 6258 /* 6259 * If the file system is mounted read-only, don't update the 6260 * superblock write time. This avoids updating the superblock 6261 * write time when we are mounting the root file system 6262 * read/only but we need to replay the journal; at that point, 6263 * for people who are east of GMT and who make their clock 6264 * tick in localtime for Windows bug-for-bug compatibility, 6265 * the clock is set in the future, and this will cause e2fsck 6266 * to complain and force a full file system check. 6267 */ 6268 if (!sb_rdonly(sb)) 6269 ext4_update_tstamp(es, s_wtime); 6270 es->s_kbytes_written = 6271 cpu_to_le64(sbi->s_kbytes_written + 6272 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 6273 sbi->s_sectors_written_start) >> 1)); 6274 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 6275 ext4_free_blocks_count_set(es, 6276 EXT4_C2B(sbi, percpu_counter_sum_positive( 6277 &sbi->s_freeclusters_counter))); 6278 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 6279 es->s_free_inodes_count = 6280 cpu_to_le32(percpu_counter_sum_positive( 6281 &sbi->s_freeinodes_counter)); 6282 /* Copy error information to the on-disk superblock */ 6283 spin_lock(&sbi->s_error_lock); 6284 if (sbi->s_add_error_count > 0) { 6285 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6286 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6287 __ext4_update_tstamp(&es->s_first_error_time, 6288 &es->s_first_error_time_hi, 6289 sbi->s_first_error_time); 6290 strtomem_pad(es->s_first_error_func, 6291 sbi->s_first_error_func, 0); 6292 es->s_first_error_line = 6293 cpu_to_le32(sbi->s_first_error_line); 6294 es->s_first_error_ino = 6295 cpu_to_le32(sbi->s_first_error_ino); 6296 es->s_first_error_block = 6297 cpu_to_le64(sbi->s_first_error_block); 6298 es->s_first_error_errcode = 6299 ext4_errno_to_code(sbi->s_first_error_code); 6300 } 6301 __ext4_update_tstamp(&es->s_last_error_time, 6302 &es->s_last_error_time_hi, 6303 sbi->s_last_error_time); 6304 strtomem_pad(es->s_last_error_func, sbi->s_last_error_func, 0); 6305 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6306 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6307 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6308 es->s_last_error_errcode = 6309 ext4_errno_to_code(sbi->s_last_error_code); 6310 /* 6311 * Start the daily error reporting function if it hasn't been 6312 * started already and sbi->s_err_report_sec is not zero 6313 */ 6314 if (!es->s_error_count && !sbi->s_err_report_sec) 6315 mod_timer(&sbi->s_err_report, 6316 jiffies + secs_to_jiffies(sbi->s_err_report_sec)); 6317 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6318 sbi->s_add_error_count = 0; 6319 } 6320 spin_unlock(&sbi->s_error_lock); 6321 6322 ext4_superblock_csum_set(sb); 6323 unlock_buffer(sbh); 6324 } 6325 6326 static int ext4_commit_super(struct super_block *sb) 6327 { 6328 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6329 6330 if (!sbh) 6331 return -EINVAL; 6332 6333 ext4_update_super(sb); 6334 6335 lock_buffer(sbh); 6336 /* Buffer got discarded which means block device got invalidated */ 6337 if (!buffer_mapped(sbh)) { 6338 unlock_buffer(sbh); 6339 return -EIO; 6340 } 6341 6342 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6343 /* 6344 * Oh, dear. A previous attempt to write the 6345 * superblock failed. This could happen because the 6346 * USB device was yanked out. Or it could happen to 6347 * be a transient write error and maybe the block will 6348 * be remapped. Nothing we can do but to retry the 6349 * write and hope for the best. 6350 */ 6351 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6352 "superblock detected"); 6353 clear_buffer_write_io_error(sbh); 6354 set_buffer_uptodate(sbh); 6355 } 6356 /* Clear potential dirty bit if it was journalled update */ 6357 clear_buffer_dirty(sbh); 6358 bh_submit(sbh, REQ_OP_WRITE | REQ_SYNC | 6359 (test_opt(sb, BARRIER) ? REQ_FUA : 0), bh_end_write); 6360 wait_on_buffer(sbh); 6361 if (buffer_write_io_error(sbh)) { 6362 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6363 "superblock"); 6364 clear_buffer_write_io_error(sbh); 6365 set_buffer_uptodate(sbh); 6366 return -EIO; 6367 } 6368 return 0; 6369 } 6370 6371 /* 6372 * Have we just finished recovery? If so, and if we are mounting (or 6373 * remounting) the filesystem readonly, then we will end up with a 6374 * consistent fs on disk. Record that fact. 6375 */ 6376 static int ext4_mark_recovery_complete(struct super_block *sb, 6377 struct ext4_super_block *es) 6378 { 6379 int err; 6380 journal_t *journal = EXT4_SB(sb)->s_journal; 6381 6382 if (!ext4_has_feature_journal(sb)) { 6383 if (journal != NULL) { 6384 ext4_error(sb, "Journal got removed while the fs was " 6385 "mounted!"); 6386 return -EFSCORRUPTED; 6387 } 6388 return 0; 6389 } 6390 jbd2_journal_lock_updates(journal); 6391 err = jbd2_journal_flush(journal, 0); 6392 if (err < 0) 6393 goto out; 6394 6395 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6396 ext4_has_feature_orphan_present(sb))) { 6397 if (!ext4_orphan_file_empty(sb)) { 6398 ext4_error(sb, "Orphan file not empty on read-only fs."); 6399 err = -EFSCORRUPTED; 6400 goto out; 6401 } 6402 ext4_clear_feature_journal_needs_recovery(sb); 6403 ext4_clear_feature_orphan_present(sb); 6404 ext4_commit_super(sb); 6405 } 6406 out: 6407 jbd2_journal_unlock_updates(journal); 6408 return err; 6409 } 6410 6411 /* 6412 * If we are mounting (or read-write remounting) a filesystem whose journal 6413 * has recorded an error from a previous lifetime, move that error to the 6414 * main filesystem now. 6415 */ 6416 static int ext4_clear_journal_err(struct super_block *sb, 6417 struct ext4_super_block *es) 6418 { 6419 journal_t *journal; 6420 int j_errno; 6421 const char *errstr; 6422 6423 if (!ext4_has_feature_journal(sb)) { 6424 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6425 return -EFSCORRUPTED; 6426 } 6427 6428 journal = EXT4_SB(sb)->s_journal; 6429 6430 /* 6431 * Now check for any error status which may have been recorded in the 6432 * journal by a prior ext4_error() or ext4_abort() 6433 */ 6434 6435 j_errno = jbd2_journal_errno(journal); 6436 if (j_errno) { 6437 char nbuf[16]; 6438 6439 errstr = ext4_decode_error(sb, j_errno, nbuf); 6440 ext4_warning(sb, "Filesystem error recorded " 6441 "from previous mount: %s", errstr); 6442 6443 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6444 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6445 j_errno = ext4_commit_super(sb); 6446 if (j_errno) 6447 return j_errno; 6448 ext4_warning(sb, "Marked fs in need of filesystem check."); 6449 6450 jbd2_journal_clear_err(journal); 6451 jbd2_journal_update_sb_errno(journal); 6452 } 6453 return 0; 6454 } 6455 6456 /* 6457 * Force the running and committing transactions to commit, 6458 * and wait on the commit. 6459 */ 6460 int ext4_force_commit(struct super_block *sb) 6461 { 6462 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal); 6463 } 6464 6465 static int ext4_sync_fs(struct super_block *sb, int wait) 6466 { 6467 int ret = 0; 6468 tid_t target; 6469 bool needs_barrier = false; 6470 struct ext4_sb_info *sbi = EXT4_SB(sb); 6471 6472 ret = ext4_emergency_state(sb); 6473 if (unlikely(ret)) 6474 return ret; 6475 6476 trace_ext4_sync_fs(sb, wait); 6477 flush_workqueue(sbi->rsv_conversion_wq); 6478 flush_delayed_work(&sbi->s_ea_inode_work); 6479 /* 6480 * Writeback quota in non-journalled quota case - journalled quota has 6481 * no dirty dquots 6482 */ 6483 dquot_writeback_dquots(sb, -1); 6484 /* 6485 * Data writeback is possible w/o journal transaction, so barrier must 6486 * being sent at the end of the function. But we can skip it if 6487 * transaction_commit will do it for us. 6488 */ 6489 if (sbi->s_journal) { 6490 target = jbd2_get_latest_transaction(sbi->s_journal); 6491 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6492 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6493 needs_barrier = true; 6494 6495 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6496 if (wait) 6497 ret = jbd2_log_wait_commit(sbi->s_journal, 6498 target); 6499 } 6500 } else if (wait && test_opt(sb, BARRIER)) 6501 needs_barrier = true; 6502 if (needs_barrier) { 6503 int err; 6504 err = blkdev_issue_flush(sb->s_bdev); 6505 if (!ret) 6506 ret = err; 6507 } 6508 6509 return ret; 6510 } 6511 6512 /* 6513 * LVM calls this function before a (read-only) snapshot is created. This 6514 * gives us a chance to flush the journal completely and mark the fs clean. 6515 * 6516 * Note that only this function cannot bring a filesystem to be in a clean 6517 * state independently. It relies on upper layer to stop all data & metadata 6518 * modifications. 6519 */ 6520 static int ext4_freeze(struct super_block *sb) 6521 { 6522 int error = 0; 6523 journal_t *journal = EXT4_SB(sb)->s_journal; 6524 6525 if (journal) { 6526 /* Now we set up the journal barrier. */ 6527 jbd2_journal_lock_updates(journal); 6528 6529 /* 6530 * Don't clear the needs_recovery flag if we failed to 6531 * flush the journal. 6532 */ 6533 error = jbd2_journal_flush(journal, 0); 6534 if (error < 0) 6535 goto out; 6536 6537 /* Journal blocked and flushed, clear needs_recovery flag. */ 6538 ext4_clear_feature_journal_needs_recovery(sb); 6539 if (ext4_orphan_file_empty(sb)) 6540 ext4_clear_feature_orphan_present(sb); 6541 } 6542 6543 error = ext4_commit_super(sb); 6544 out: 6545 if (journal) 6546 /* we rely on upper layer to stop further updates */ 6547 jbd2_journal_unlock_updates(journal); 6548 return error; 6549 } 6550 6551 /* 6552 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6553 * flag here, even though the filesystem is not technically dirty yet. 6554 */ 6555 static int ext4_unfreeze(struct super_block *sb) 6556 { 6557 if (ext4_emergency_state(sb)) 6558 return 0; 6559 6560 if (EXT4_SB(sb)->s_journal) { 6561 /* Reset the needs_recovery flag before the fs is unlocked. */ 6562 ext4_set_feature_journal_needs_recovery(sb); 6563 if (ext4_has_feature_orphan_file(sb)) 6564 ext4_set_feature_orphan_present(sb); 6565 } 6566 6567 ext4_commit_super(sb); 6568 return 0; 6569 } 6570 6571 /* 6572 * Structure to save mount options for ext4_remount's benefit 6573 */ 6574 struct ext4_mount_options { 6575 unsigned long s_mount_opt; 6576 unsigned long s_mount_opt2; 6577 kuid_t s_resuid; 6578 kgid_t s_resgid; 6579 unsigned long s_commit_interval; 6580 u32 s_min_batch_time, s_max_batch_time; 6581 #ifdef CONFIG_QUOTA 6582 int s_jquota_fmt; 6583 char *s_qf_names[EXT4_MAXQUOTAS]; 6584 #endif 6585 }; 6586 6587 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6588 { 6589 struct ext4_fs_context *ctx = fc->fs_private; 6590 struct ext4_super_block *es; 6591 struct ext4_sb_info *sbi = EXT4_SB(sb); 6592 unsigned long old_sb_flags; 6593 struct ext4_mount_options old_opts; 6594 ext4_group_t g; 6595 int err = 0; 6596 int alloc_ctx; 6597 #ifdef CONFIG_QUOTA 6598 int enable_quota = 0; 6599 int i, j; 6600 char *to_free[EXT4_MAXQUOTAS]; 6601 #endif 6602 6603 6604 /* Store the original options */ 6605 old_sb_flags = sb->s_flags; 6606 old_opts.s_mount_opt = sbi->s_mount_opt; 6607 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6608 old_opts.s_resuid = sbi->s_resuid; 6609 old_opts.s_resgid = sbi->s_resgid; 6610 old_opts.s_commit_interval = sbi->s_commit_interval; 6611 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6612 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6613 #ifdef CONFIG_QUOTA 6614 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6615 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6616 if (sbi->s_qf_names[i]) { 6617 char *qf_name = get_qf_name(sb, sbi, i); 6618 6619 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6620 if (!old_opts.s_qf_names[i]) { 6621 for (j = 0; j < i; j++) 6622 kfree(old_opts.s_qf_names[j]); 6623 return -ENOMEM; 6624 } 6625 } else 6626 old_opts.s_qf_names[i] = NULL; 6627 #endif 6628 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6629 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6630 ctx->journal_ioprio = 6631 sbi->s_journal->j_task->io_context->ioprio; 6632 else 6633 ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO; 6634 6635 } 6636 6637 if ((ctx->spec & EXT4_SPEC_s_stripe) && 6638 ext4_is_stripe_incompatible(sb, ctx->s_stripe)) { 6639 ext4_msg(sb, KERN_WARNING, 6640 "stripe (%lu) is not aligned with cluster size (%u), " 6641 "stripe is disabled", 6642 ctx->s_stripe, sbi->s_cluster_ratio); 6643 ctx->s_stripe = 0; 6644 } 6645 6646 /* 6647 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause 6648 * two calls to ext4_should_dioread_nolock() to return inconsistent 6649 * values, triggering WARN_ON in ext4_add_complete_io(). we grab 6650 * here s_writepages_rwsem to avoid race between writepages ops and 6651 * remount. 6652 */ 6653 alloc_ctx = ext4_writepages_down_write(sb); 6654 ext4_apply_options(fc, sb); 6655 ext4_writepages_up_write(sb, alloc_ctx); 6656 6657 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6658 test_opt(sb, JOURNAL_CHECKSUM)) { 6659 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6660 "during remount not supported; ignoring"); 6661 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6662 } 6663 6664 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6665 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6666 ext4_msg(sb, KERN_ERR, "can't mount with " 6667 "both data=journal and delalloc"); 6668 err = -EINVAL; 6669 goto restore_opts; 6670 } 6671 if (test_opt(sb, DIOREAD_NOLOCK)) { 6672 ext4_msg(sb, KERN_ERR, "can't mount with " 6673 "both data=journal and dioread_nolock"); 6674 err = -EINVAL; 6675 goto restore_opts; 6676 } 6677 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6678 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6679 ext4_msg(sb, KERN_ERR, "can't mount with " 6680 "journal_async_commit in data=ordered mode"); 6681 err = -EINVAL; 6682 goto restore_opts; 6683 } 6684 } 6685 6686 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6687 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6688 err = -EINVAL; 6689 goto restore_opts; 6690 } 6691 6692 if ((old_opts.s_mount_opt & EXT4_MOUNT_DELALLOC) && 6693 !test_opt(sb, DELALLOC)) { 6694 ext4_msg(sb, KERN_ERR, "can't disable delalloc during remount"); 6695 err = -EINVAL; 6696 goto restore_opts; 6697 } 6698 6699 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6700 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6701 6702 es = sbi->s_es; 6703 6704 if (sbi->s_journal) { 6705 ext4_init_journal_params(sb, sbi->s_journal); 6706 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6707 } 6708 6709 /* Flush outstanding errors before changing fs state */ 6710 flush_work(&sbi->s_sb_upd_work); 6711 6712 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6713 if (ext4_emergency_state(sb)) { 6714 err = -EROFS; 6715 goto restore_opts; 6716 } 6717 6718 if (fc->sb_flags & SB_RDONLY) { 6719 err = sync_filesystem(sb); 6720 if (err < 0) 6721 goto restore_opts; 6722 err = dquot_suspend(sb, -1); 6723 if (err < 0) 6724 goto restore_opts; 6725 6726 /* 6727 * First of all, the unconditional stuff we have to do 6728 * to disable replay of the journal when we next remount 6729 */ 6730 sb->s_flags |= SB_RDONLY; 6731 6732 /* 6733 * OK, test if we are remounting a valid rw partition 6734 * readonly, and if so set the rdonly flag and then 6735 * mark the partition as valid again. 6736 */ 6737 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6738 (sbi->s_mount_state & EXT4_VALID_FS)) 6739 es->s_state = cpu_to_le16(sbi->s_mount_state); 6740 6741 if (sbi->s_journal) { 6742 /* 6743 * We let remount-ro finish even if marking fs 6744 * as clean failed... 6745 */ 6746 ext4_mark_recovery_complete(sb, es); 6747 } 6748 } else { 6749 /* Make sure we can mount this feature set readwrite */ 6750 if (ext4_has_feature_readonly(sb) || 6751 !ext4_feature_set_ok(sb, 0)) { 6752 err = -EROFS; 6753 goto restore_opts; 6754 } 6755 /* 6756 * Make sure the group descriptor checksums 6757 * are sane. If they aren't, refuse to remount r/w. 6758 */ 6759 for (g = 0; g < sbi->s_groups_count; g++) { 6760 struct ext4_group_desc *gdp = 6761 ext4_get_group_desc(sb, g, NULL); 6762 6763 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6764 ext4_msg(sb, KERN_ERR, 6765 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6766 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6767 le16_to_cpu(gdp->bg_checksum)); 6768 err = -EFSBADCRC; 6769 goto restore_opts; 6770 } 6771 } 6772 6773 /* 6774 * If we have an unprocessed orphan list hanging 6775 * around from a previously readonly bdev mount, 6776 * require a full umount/remount for now. 6777 */ 6778 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6779 ext4_msg(sb, KERN_WARNING, "Couldn't " 6780 "remount RDWR because of unprocessed " 6781 "orphan inode list. Please " 6782 "umount/remount instead"); 6783 err = -EINVAL; 6784 goto restore_opts; 6785 } 6786 6787 /* 6788 * Mounting a RDONLY partition read-write, so reread 6789 * and store the current valid flag. (It may have 6790 * been changed by e2fsck since we originally mounted 6791 * the partition.) 6792 */ 6793 if (sbi->s_journal) { 6794 err = ext4_clear_journal_err(sb, es); 6795 if (err) 6796 goto restore_opts; 6797 } 6798 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6799 ~EXT4_FC_REPLAY); 6800 6801 err = ext4_setup_super(sb, es, 0); 6802 if (err) 6803 goto restore_opts; 6804 6805 sb->s_flags &= ~SB_RDONLY; 6806 if (ext4_has_feature_mmp(sb)) { 6807 err = ext4_multi_mount_protect(sb, 6808 le64_to_cpu(es->s_mmp_block)); 6809 if (err) 6810 goto restore_opts; 6811 } 6812 #ifdef CONFIG_QUOTA 6813 enable_quota = 1; 6814 #endif 6815 } 6816 } 6817 6818 /* 6819 * Handle creation of system zone data early because it can fail. 6820 * Releasing of existing data is done when we are sure remount will 6821 * succeed. 6822 */ 6823 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6824 err = ext4_setup_system_zone(sb); 6825 if (err) 6826 goto restore_opts; 6827 } 6828 6829 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6830 err = ext4_commit_super(sb); 6831 if (err) 6832 goto restore_opts; 6833 } 6834 6835 #ifdef CONFIG_QUOTA 6836 if (enable_quota) { 6837 if (sb_any_quota_suspended(sb)) 6838 dquot_resume(sb, -1); 6839 else if (ext4_has_feature_quota(sb)) { 6840 err = ext4_enable_quotas(sb); 6841 if (err) 6842 goto restore_opts; 6843 } 6844 } 6845 /* Release old quota file names */ 6846 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6847 kfree(old_opts.s_qf_names[i]); 6848 #endif 6849 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6850 ext4_release_system_zone(sb); 6851 6852 /* 6853 * Reinitialize lazy itable initialization thread based on 6854 * current settings 6855 */ 6856 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6857 ext4_unregister_li_request(sb); 6858 else { 6859 ext4_group_t first_not_zeroed; 6860 first_not_zeroed = ext4_has_uninit_itable(sb); 6861 ext4_register_li_request(sb, first_not_zeroed); 6862 } 6863 6864 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6865 ext4_stop_mmpd(sbi); 6866 6867 /* 6868 * Handle aborting the filesystem as the last thing during remount to 6869 * avoid obsure errors during remount when some option changes fail to 6870 * apply due to shutdown filesystem. 6871 */ 6872 if (test_opt2(sb, ABORT)) 6873 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6874 6875 return 0; 6876 6877 restore_opts: 6878 /* 6879 * If there was a failing r/w to ro transition, we may need to 6880 * re-enable quota 6881 */ 6882 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) && 6883 sb_any_quota_suspended(sb)) 6884 dquot_resume(sb, -1); 6885 6886 alloc_ctx = ext4_writepages_down_write(sb); 6887 sb->s_flags = old_sb_flags; 6888 sbi->s_mount_opt = old_opts.s_mount_opt; 6889 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6890 sbi->s_resuid = old_opts.s_resuid; 6891 sbi->s_resgid = old_opts.s_resgid; 6892 sbi->s_commit_interval = old_opts.s_commit_interval; 6893 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6894 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6895 ext4_writepages_up_write(sb, alloc_ctx); 6896 6897 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6898 ext4_release_system_zone(sb); 6899 #ifdef CONFIG_QUOTA 6900 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6901 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6902 to_free[i] = get_qf_name(sb, sbi, i); 6903 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6904 } 6905 synchronize_rcu(); 6906 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6907 kfree(to_free[i]); 6908 #endif 6909 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6910 ext4_stop_mmpd(sbi); 6911 return err; 6912 } 6913 6914 static int ext4_reconfigure(struct fs_context *fc) 6915 { 6916 struct super_block *sb = fc->root->d_sb; 6917 int ret; 6918 bool old_ro = sb_rdonly(sb); 6919 6920 fc->s_fs_info = EXT4_SB(sb); 6921 6922 ret = ext4_check_opt_consistency(fc, sb); 6923 if (ret < 0) 6924 return ret; 6925 6926 ret = __ext4_remount(fc, sb); 6927 if (ret < 0) 6928 return ret; 6929 6930 ext4_msg(sb, KERN_INFO, "re-mounted %pU%s.", 6931 &sb->s_uuid, 6932 (old_ro != sb_rdonly(sb)) ? (sb_rdonly(sb) ? " ro" : " r/w") : ""); 6933 6934 return 0; 6935 } 6936 6937 #ifdef CONFIG_QUOTA 6938 static int ext4_statfs_project(struct super_block *sb, 6939 kprojid_t projid, struct kstatfs *buf) 6940 { 6941 struct kqid qid; 6942 struct dquot *dquot; 6943 u64 limit; 6944 u64 curblock; 6945 6946 qid = make_kqid_projid(projid); 6947 dquot = dqget(sb, qid); 6948 if (IS_ERR(dquot)) 6949 return PTR_ERR(dquot); 6950 spin_lock(&dquot->dq_dqb_lock); 6951 6952 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6953 dquot->dq_dqb.dqb_bhardlimit); 6954 limit >>= sb->s_blocksize_bits; 6955 6956 if (limit) { 6957 uint64_t remaining = 0; 6958 6959 curblock = (dquot->dq_dqb.dqb_curspace + 6960 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6961 if (limit > curblock) 6962 remaining = limit - curblock; 6963 6964 buf->f_blocks = min(buf->f_blocks, limit); 6965 buf->f_bfree = min(buf->f_bfree, remaining); 6966 buf->f_bavail = min(buf->f_bavail, remaining); 6967 } 6968 6969 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6970 dquot->dq_dqb.dqb_ihardlimit); 6971 if (limit) { 6972 uint64_t remaining = 0; 6973 6974 if (limit > dquot->dq_dqb.dqb_curinodes) 6975 remaining = limit - dquot->dq_dqb.dqb_curinodes; 6976 6977 buf->f_files = min(buf->f_files, limit); 6978 buf->f_ffree = min(buf->f_ffree, remaining); 6979 } 6980 6981 spin_unlock(&dquot->dq_dqb_lock); 6982 dqput(dquot); 6983 return 0; 6984 } 6985 #endif 6986 6987 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6988 { 6989 struct super_block *sb = dentry->d_sb; 6990 struct ext4_sb_info *sbi = EXT4_SB(sb); 6991 struct ext4_super_block *es = sbi->s_es; 6992 ext4_fsblk_t overhead = 0, resv_blocks; 6993 s64 bfree; 6994 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6995 6996 if (!test_opt(sb, MINIX_DF)) 6997 overhead = sbi->s_overhead; 6998 6999 buf->f_type = EXT4_SUPER_MAGIC; 7000 buf->f_bsize = sb->s_blocksize; 7001 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 7002 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 7003 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 7004 /* prevent underflow in case that few free space is available */ 7005 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 7006 buf->f_bavail = buf->f_bfree - 7007 (ext4_r_blocks_count(es) + resv_blocks); 7008 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 7009 buf->f_bavail = 0; 7010 buf->f_files = le32_to_cpu(es->s_inodes_count); 7011 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 7012 buf->f_namelen = EXT4_NAME_LEN; 7013 buf->f_fsid = uuid_to_fsid(es->s_uuid); 7014 7015 #ifdef CONFIG_QUOTA 7016 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 7017 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 7018 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 7019 #endif 7020 return 0; 7021 } 7022 7023 7024 #ifdef CONFIG_QUOTA 7025 7026 /* 7027 * Helper functions so that transaction is started before we acquire dqio_sem 7028 * to keep correct lock ordering of transaction > dqio_sem 7029 */ 7030 static inline struct inode *dquot_to_inode(struct dquot *dquot) 7031 { 7032 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 7033 } 7034 7035 static int ext4_write_dquot(struct dquot *dquot) 7036 { 7037 int ret, err; 7038 handle_t *handle; 7039 struct inode *inode; 7040 7041 inode = dquot_to_inode(dquot); 7042 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 7043 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 7044 if (IS_ERR(handle)) 7045 return PTR_ERR(handle); 7046 ret = dquot_commit(dquot); 7047 if (ret < 0) 7048 ext4_error_err(dquot->dq_sb, -ret, 7049 "Failed to commit dquot type %d", 7050 dquot->dq_id.type); 7051 err = ext4_journal_stop(handle); 7052 if (!ret) 7053 ret = err; 7054 return ret; 7055 } 7056 7057 static int ext4_acquire_dquot(struct dquot *dquot) 7058 { 7059 int ret, err; 7060 handle_t *handle; 7061 7062 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 7063 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 7064 if (IS_ERR(handle)) 7065 return PTR_ERR(handle); 7066 ret = dquot_acquire(dquot); 7067 if (ret < 0) 7068 ext4_error_err(dquot->dq_sb, -ret, 7069 "Failed to acquire dquot type %d", 7070 dquot->dq_id.type); 7071 err = ext4_journal_stop(handle); 7072 if (!ret) 7073 ret = err; 7074 return ret; 7075 } 7076 7077 static int ext4_release_dquot(struct dquot *dquot) 7078 { 7079 int ret, err; 7080 handle_t *handle; 7081 bool freeze_protected = false; 7082 7083 /* 7084 * Trying to sb_start_intwrite() in a running transaction 7085 * can result in a deadlock. Further, running transactions 7086 * are already protected from freezing. 7087 */ 7088 if (!ext4_journal_current_handle()) { 7089 sb_start_intwrite(dquot->dq_sb); 7090 freeze_protected = true; 7091 } 7092 7093 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 7094 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 7095 if (IS_ERR(handle)) { 7096 /* Release dquot anyway to avoid endless cycle in dqput() */ 7097 dquot_release(dquot); 7098 if (freeze_protected) 7099 sb_end_intwrite(dquot->dq_sb); 7100 return PTR_ERR(handle); 7101 } 7102 ret = dquot_release(dquot); 7103 if (ret < 0) 7104 ext4_error_err(dquot->dq_sb, -ret, 7105 "Failed to release dquot type %d", 7106 dquot->dq_id.type); 7107 err = ext4_journal_stop(handle); 7108 if (!ret) 7109 ret = err; 7110 7111 if (freeze_protected) 7112 sb_end_intwrite(dquot->dq_sb); 7113 7114 return ret; 7115 } 7116 7117 static int ext4_mark_dquot_dirty(struct dquot *dquot) 7118 { 7119 struct super_block *sb = dquot->dq_sb; 7120 7121 if (ext4_is_quota_journalled(sb)) { 7122 dquot_mark_dquot_dirty(dquot); 7123 return ext4_write_dquot(dquot); 7124 } else { 7125 return dquot_mark_dquot_dirty(dquot); 7126 } 7127 } 7128 7129 static int ext4_write_info(struct super_block *sb, int type) 7130 { 7131 int ret, err; 7132 handle_t *handle; 7133 7134 /* Data block + inode block */ 7135 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 7136 if (IS_ERR(handle)) 7137 return PTR_ERR(handle); 7138 ret = dquot_commit_info(sb, type); 7139 err = ext4_journal_stop(handle); 7140 if (!ret) 7141 ret = err; 7142 return ret; 7143 } 7144 7145 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 7146 { 7147 struct ext4_inode_info *ei = EXT4_I(inode); 7148 7149 /* The first argument of lockdep_set_subclass has to be 7150 * *exactly* the same as the argument to init_rwsem() --- in 7151 * this case, in init_once() --- or lockdep gets unhappy 7152 * because the name of the lock is set using the 7153 * stringification of the argument to init_rwsem(). 7154 */ 7155 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 7156 lockdep_set_subclass(&ei->i_data_sem, subclass); 7157 } 7158 7159 /* 7160 * Standard function to be called on quota_on 7161 */ 7162 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 7163 const struct path *path) 7164 { 7165 int err; 7166 7167 if (!test_opt(sb, QUOTA)) 7168 return -EINVAL; 7169 7170 /* Quotafile not on the same filesystem? */ 7171 if (path->dentry->d_sb != sb) 7172 return -EXDEV; 7173 7174 /* Quota already enabled for this file? */ 7175 if (IS_NOQUOTA(d_inode(path->dentry))) 7176 return -EBUSY; 7177 7178 /* Journaling quota? */ 7179 if (EXT4_SB(sb)->s_qf_names[type]) { 7180 /* Quotafile not in fs root? */ 7181 if (path->dentry->d_parent != sb->s_root) 7182 ext4_msg(sb, KERN_WARNING, 7183 "Quota file not on filesystem root. " 7184 "Journaled quota will not work"); 7185 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 7186 } else { 7187 /* 7188 * Clear the flag just in case mount options changed since 7189 * last time. 7190 */ 7191 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 7192 } 7193 7194 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 7195 err = dquot_quota_on(sb, type, format_id, path); 7196 if (!err) { 7197 struct inode *inode = d_inode(path->dentry); 7198 handle_t *handle; 7199 7200 /* 7201 * Set inode flags to prevent userspace from messing with quota 7202 * files. If this fails, we return success anyway since quotas 7203 * are already enabled and this is not a hard failure. 7204 */ 7205 inode_lock(inode); 7206 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7207 if (IS_ERR(handle)) 7208 goto unlock_inode; 7209 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 7210 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 7211 S_NOATIME | S_IMMUTABLE); 7212 err = ext4_mark_inode_dirty(handle, inode); 7213 ext4_journal_stop(handle); 7214 unlock_inode: 7215 inode_unlock(inode); 7216 if (err) 7217 dquot_quota_off(sb, type); 7218 } 7219 if (err) 7220 lockdep_set_quota_inode(path->dentry->d_inode, 7221 I_DATA_SEM_NORMAL); 7222 return err; 7223 } 7224 7225 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum) 7226 { 7227 switch (type) { 7228 case USRQUOTA: 7229 return qf_inum == EXT4_USR_QUOTA_INO; 7230 case GRPQUOTA: 7231 return qf_inum == EXT4_GRP_QUOTA_INO; 7232 case PRJQUOTA: 7233 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO; 7234 default: 7235 BUG(); 7236 } 7237 } 7238 7239 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 7240 unsigned int flags) 7241 { 7242 int err; 7243 struct inode *qf_inode; 7244 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7245 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7246 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7247 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7248 }; 7249 7250 BUG_ON(!ext4_has_feature_quota(sb)); 7251 7252 if (!qf_inums[type]) 7253 return -EPERM; 7254 7255 if (!ext4_check_quota_inum(type, qf_inums[type])) { 7256 ext4_error(sb, "Bad quota inum: %lu, type: %d", 7257 qf_inums[type], type); 7258 return -EUCLEAN; 7259 } 7260 7261 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 7262 if (IS_ERR(qf_inode)) { 7263 ext4_error(sb, "Bad quota inode: %lu, type: %d", 7264 qf_inums[type], type); 7265 return PTR_ERR(qf_inode); 7266 } 7267 7268 /* Don't account quota for quota files to avoid recursion */ 7269 qf_inode->i_flags |= S_NOQUOTA; 7270 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 7271 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 7272 if (err) 7273 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 7274 iput(qf_inode); 7275 7276 return err; 7277 } 7278 7279 /* Enable usage tracking for all quota types. */ 7280 int ext4_enable_quotas(struct super_block *sb) 7281 { 7282 int type, err = 0; 7283 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7284 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7285 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7286 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7287 }; 7288 bool quota_mopt[EXT4_MAXQUOTAS] = { 7289 test_opt(sb, USRQUOTA), 7290 test_opt(sb, GRPQUOTA), 7291 test_opt(sb, PRJQUOTA), 7292 }; 7293 7294 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 7295 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 7296 if (qf_inums[type]) { 7297 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 7298 DQUOT_USAGE_ENABLED | 7299 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 7300 if (err) { 7301 ext4_warning(sb, 7302 "Failed to enable quota tracking " 7303 "(type=%d, err=%d, ino=%lu). " 7304 "Please run e2fsck to fix.", type, 7305 err, qf_inums[type]); 7306 7307 ext4_quotas_off(sb, type); 7308 return err; 7309 } 7310 } 7311 } 7312 return 0; 7313 } 7314 7315 static int ext4_quota_off(struct super_block *sb, int type) 7316 { 7317 struct inode *inode = sb_dqopt(sb)->files[type]; 7318 handle_t *handle; 7319 int err; 7320 7321 /* Force all delayed allocation blocks to be allocated. 7322 * Caller already holds s_umount sem */ 7323 if (test_opt(sb, DELALLOC)) 7324 sync_filesystem(sb); 7325 7326 if (!inode || !igrab(inode)) 7327 goto out; 7328 7329 err = dquot_quota_off(sb, type); 7330 if (err || ext4_has_feature_quota(sb)) 7331 goto out_put; 7332 /* 7333 * When the filesystem was remounted read-only first, we cannot cleanup 7334 * inode flags here. Bad luck but people should be using QUOTA feature 7335 * these days anyway. 7336 */ 7337 if (sb_rdonly(sb)) 7338 goto out_put; 7339 7340 inode_lock(inode); 7341 /* 7342 * Update modification times of quota files when userspace can 7343 * start looking at them. If we fail, we return success anyway since 7344 * this is not a hard failure and quotas are already disabled. 7345 */ 7346 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7347 if (IS_ERR(handle)) { 7348 err = PTR_ERR(handle); 7349 goto out_unlock; 7350 } 7351 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7352 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7353 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 7354 err = ext4_mark_inode_dirty(handle, inode); 7355 ext4_journal_stop(handle); 7356 out_unlock: 7357 inode_unlock(inode); 7358 out_put: 7359 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7360 iput(inode); 7361 return err; 7362 out: 7363 return dquot_quota_off(sb, type); 7364 } 7365 7366 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7367 * acquiring the locks... As quota files are never truncated and quota code 7368 * itself serializes the operations (and no one else should touch the files) 7369 * we don't have to be afraid of races */ 7370 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7371 size_t len, loff_t off) 7372 { 7373 struct inode *inode = sb_dqopt(sb)->files[type]; 7374 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7375 int offset = off & (sb->s_blocksize - 1); 7376 int tocopy; 7377 size_t toread; 7378 struct buffer_head *bh; 7379 loff_t i_size = i_size_read(inode); 7380 7381 if (off > i_size) 7382 return 0; 7383 if (off+len > i_size) 7384 len = i_size-off; 7385 toread = len; 7386 while (toread > 0) { 7387 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread); 7388 bh = ext4_bread(NULL, inode, blk, 0); 7389 if (IS_ERR(bh)) 7390 return PTR_ERR(bh); 7391 if (!bh) /* A hole? */ 7392 memset(data, 0, tocopy); 7393 else 7394 memcpy(data, bh->b_data+offset, tocopy); 7395 brelse(bh); 7396 offset = 0; 7397 toread -= tocopy; 7398 data += tocopy; 7399 blk++; 7400 } 7401 return len; 7402 } 7403 7404 /* Write to quotafile (we know the transaction is already started and has 7405 * enough credits) */ 7406 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7407 const char *data, size_t len, loff_t off) 7408 { 7409 struct inode *inode = sb_dqopt(sb)->files[type]; 7410 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7411 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7412 int retries = 0; 7413 struct buffer_head *bh; 7414 handle_t *handle = journal_current_handle(); 7415 7416 if (!handle) { 7417 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7418 " cancelled because transaction is not started", 7419 (unsigned long long)off, (unsigned long long)len); 7420 return -EIO; 7421 } 7422 /* 7423 * Since we account only one data block in transaction credits, 7424 * then it is impossible to cross a block boundary. 7425 */ 7426 if (sb->s_blocksize - offset < len) { 7427 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7428 " cancelled because not block aligned", 7429 (unsigned long long)off, (unsigned long long)len); 7430 return -EIO; 7431 } 7432 7433 do { 7434 bh = ext4_bread(handle, inode, blk, 7435 EXT4_GET_BLOCKS_CREATE | 7436 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7437 } while (PTR_ERR(bh) == -ENOSPC && 7438 ext4_should_retry_alloc(inode->i_sb, &retries)); 7439 if (IS_ERR(bh)) 7440 return PTR_ERR(bh); 7441 if (!bh) 7442 goto out; 7443 BUFFER_TRACE(bh, "get write access"); 7444 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7445 if (err) { 7446 brelse(bh); 7447 return err; 7448 } 7449 lock_buffer(bh); 7450 memcpy(bh->b_data+offset, data, len); 7451 flush_dcache_folio(bh->b_folio); 7452 unlock_buffer(bh); 7453 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7454 brelse(bh); 7455 out: 7456 if (inode->i_size < off + len) { 7457 i_size_write(inode, off + len); 7458 EXT4_I(inode)->i_disksize = inode->i_size; 7459 err2 = ext4_mark_inode_dirty(handle, inode); 7460 if (unlikely(err2 && !err)) 7461 err = err2; 7462 } 7463 return err ? err : len; 7464 } 7465 #endif 7466 7467 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7468 static inline void register_as_ext2(void) 7469 { 7470 int err = register_filesystem(&ext2_fs_type); 7471 if (err) 7472 printk(KERN_WARNING 7473 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7474 } 7475 7476 static inline void unregister_as_ext2(void) 7477 { 7478 unregister_filesystem(&ext2_fs_type); 7479 } 7480 7481 static inline int ext2_feature_set_ok(struct super_block *sb) 7482 { 7483 if (ext4_has_unknown_ext2_incompat_features(sb)) 7484 return 0; 7485 if (sb_rdonly(sb)) 7486 return 1; 7487 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7488 return 0; 7489 return 1; 7490 } 7491 #else 7492 static inline void register_as_ext2(void) { } 7493 static inline void unregister_as_ext2(void) { } 7494 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7495 #endif 7496 7497 static inline void register_as_ext3(void) 7498 { 7499 int err = register_filesystem(&ext3_fs_type); 7500 if (err) 7501 printk(KERN_WARNING 7502 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7503 } 7504 7505 static inline void unregister_as_ext3(void) 7506 { 7507 unregister_filesystem(&ext3_fs_type); 7508 } 7509 7510 static inline int ext3_feature_set_ok(struct super_block *sb) 7511 { 7512 if (ext4_has_unknown_ext3_incompat_features(sb)) 7513 return 0; 7514 if (!ext4_has_feature_journal(sb)) 7515 return 0; 7516 if (sb_rdonly(sb)) 7517 return 1; 7518 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7519 return 0; 7520 return 1; 7521 } 7522 7523 static void ext4_kill_sb(struct super_block *sb) 7524 { 7525 struct ext4_sb_info *sbi = EXT4_SB(sb); 7526 struct file *bdev_file = sbi ? sbi->s_journal_bdev_file : NULL; 7527 7528 kill_block_super(sb); 7529 7530 if (bdev_file) 7531 fs_bdev_file_release(bdev_file, sb); 7532 } 7533 7534 static struct file_system_type ext4_fs_type = { 7535 .owner = THIS_MODULE, 7536 .name = "ext4", 7537 .init_fs_context = ext4_init_fs_context, 7538 .parameters = ext4_param_specs, 7539 .kill_sb = ext4_kill_sb, 7540 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP | FS_MGTIME | 7541 FS_LBS, 7542 }; 7543 MODULE_ALIAS_FS("ext4"); 7544 7545 static int __init ext4_init_fs(void) 7546 { 7547 int err; 7548 7549 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7550 ext4_li_info = NULL; 7551 7552 /* Build-time check for flags consistency */ 7553 ext4_check_flag_values(); 7554 7555 err = ext4_init_es(); 7556 if (err) 7557 return err; 7558 7559 err = ext4_init_pending(); 7560 if (err) 7561 goto out7; 7562 7563 err = ext4_init_verity_caches(); 7564 if (err) 7565 goto out6; 7566 7567 err = ext4_init_pageio(); 7568 if (err) 7569 goto out5; 7570 7571 err = ext4_init_system_zone(); 7572 if (err) 7573 goto out4; 7574 7575 err = ext4_init_sysfs(); 7576 if (err) 7577 goto out3; 7578 7579 err = ext4_init_mballoc(); 7580 if (err) 7581 goto out2; 7582 err = init_inodecache(); 7583 if (err) 7584 goto out1; 7585 7586 err = ext4_fc_init_dentry_cache(); 7587 if (err) 7588 goto out05; 7589 7590 register_as_ext3(); 7591 register_as_ext2(); 7592 err = register_filesystem(&ext4_fs_type); 7593 if (err) 7594 goto out; 7595 7596 return 0; 7597 out: 7598 unregister_as_ext2(); 7599 unregister_as_ext3(); 7600 ext4_fc_destroy_dentry_cache(); 7601 out05: 7602 destroy_inodecache(); 7603 out1: 7604 ext4_exit_mballoc(); 7605 out2: 7606 ext4_exit_sysfs(); 7607 out3: 7608 ext4_exit_system_zone(); 7609 out4: 7610 ext4_exit_pageio(); 7611 out5: 7612 ext4_exit_verity_caches(); 7613 out6: 7614 ext4_exit_pending(); 7615 out7: 7616 ext4_exit_es(); 7617 7618 return err; 7619 } 7620 7621 static void __exit ext4_exit_fs(void) 7622 { 7623 ext4_destroy_lazyinit_thread(); 7624 unregister_as_ext2(); 7625 unregister_as_ext3(); 7626 unregister_filesystem(&ext4_fs_type); 7627 ext4_fc_destroy_dentry_cache(); 7628 destroy_inodecache(); 7629 ext4_exit_mballoc(); 7630 ext4_exit_sysfs(); 7631 ext4_exit_system_zone(); 7632 ext4_exit_pageio(); 7633 ext4_exit_verity_caches(); 7634 ext4_exit_es(); 7635 ext4_exit_pending(); 7636 } 7637 7638 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7639 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7640 MODULE_LICENSE("GPL"); 7641 module_init(ext4_init_fs) 7642 module_exit(ext4_exit_fs) 7643