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 52 #include "ext4.h" 53 #include "ext4_extents.h" /* Needed for trace points definition */ 54 #include "ext4_jbd2.h" 55 #include "xattr.h" 56 #include "acl.h" 57 #include "mballoc.h" 58 #include "fsmap.h" 59 60 #define CREATE_TRACE_POINTS 61 #include <trace/events/ext4.h> 62 63 static struct ext4_lazy_init *ext4_li_info; 64 static DEFINE_MUTEX(ext4_li_mtx); 65 static struct ratelimit_state ext4_mount_msg_ratelimit; 66 67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 68 unsigned long journal_devnum); 69 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 70 static void ext4_update_super(struct super_block *sb); 71 static int ext4_commit_super(struct super_block *sb); 72 static int ext4_mark_recovery_complete(struct super_block *sb, 73 struct ext4_super_block *es); 74 static int ext4_clear_journal_err(struct super_block *sb, 75 struct ext4_super_block *es); 76 static int ext4_sync_fs(struct super_block *sb, int wait); 77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 78 static int ext4_unfreeze(struct super_block *sb); 79 static int ext4_freeze(struct super_block *sb); 80 static inline int ext2_feature_set_ok(struct super_block *sb); 81 static inline int ext3_feature_set_ok(struct super_block *sb); 82 static void ext4_unregister_li_request(struct super_block *sb); 83 static void ext4_clear_request_list(void); 84 static struct inode *ext4_get_journal_inode(struct super_block *sb, 85 unsigned int journal_inum); 86 static int ext4_validate_options(struct fs_context *fc); 87 static int ext4_check_opt_consistency(struct fs_context *fc, 88 struct super_block *sb); 89 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb); 90 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param); 91 static int ext4_get_tree(struct fs_context *fc); 92 static int ext4_reconfigure(struct fs_context *fc); 93 static void ext4_fc_free(struct fs_context *fc); 94 static int ext4_init_fs_context(struct fs_context *fc); 95 static void ext4_kill_sb(struct super_block *sb); 96 static const struct fs_parameter_spec ext4_param_specs[]; 97 98 /* 99 * Lock ordering 100 * 101 * page fault path: 102 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start 103 * -> page lock -> i_data_sem (rw) 104 * 105 * buffered write path: 106 * sb_start_write -> i_mutex -> mmap_lock 107 * sb_start_write -> i_mutex -> transaction start -> page lock -> 108 * i_data_sem (rw) 109 * 110 * truncate: 111 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) -> 112 * page lock 113 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start -> 114 * i_data_sem (rw) 115 * 116 * direct IO: 117 * sb_start_write -> i_mutex -> mmap_lock 118 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 119 * 120 * writepages: 121 * transaction start -> page lock(s) -> i_data_sem (rw) 122 */ 123 124 static const struct fs_context_operations ext4_context_ops = { 125 .parse_param = ext4_parse_param, 126 .get_tree = ext4_get_tree, 127 .reconfigure = ext4_reconfigure, 128 .free = ext4_fc_free, 129 }; 130 131 132 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 133 static struct file_system_type ext2_fs_type = { 134 .owner = THIS_MODULE, 135 .name = "ext2", 136 .init_fs_context = ext4_init_fs_context, 137 .parameters = ext4_param_specs, 138 .kill_sb = ext4_kill_sb, 139 .fs_flags = FS_REQUIRES_DEV, 140 }; 141 MODULE_ALIAS_FS("ext2"); 142 MODULE_ALIAS("ext2"); 143 #define IS_EXT2_SB(sb) ((sb)->s_type == &ext2_fs_type) 144 #else 145 #define IS_EXT2_SB(sb) (0) 146 #endif 147 148 149 static struct file_system_type ext3_fs_type = { 150 .owner = THIS_MODULE, 151 .name = "ext3", 152 .init_fs_context = ext4_init_fs_context, 153 .parameters = ext4_param_specs, 154 .kill_sb = ext4_kill_sb, 155 .fs_flags = FS_REQUIRES_DEV, 156 }; 157 MODULE_ALIAS_FS("ext3"); 158 MODULE_ALIAS("ext3"); 159 #define IS_EXT3_SB(sb) ((sb)->s_type == &ext3_fs_type) 160 161 162 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 163 bh_end_io_t *end_io, bool simu_fail) 164 { 165 if (simu_fail) { 166 clear_buffer_uptodate(bh); 167 unlock_buffer(bh); 168 return; 169 } 170 171 /* 172 * buffer's verified bit is no longer valid after reading from 173 * disk again due to write out error, clear it to make sure we 174 * recheck the buffer contents. 175 */ 176 clear_buffer_verified(bh); 177 178 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 179 get_bh(bh); 180 submit_bh(REQ_OP_READ | op_flags, bh); 181 } 182 183 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags, 184 bh_end_io_t *end_io, bool simu_fail) 185 { 186 BUG_ON(!buffer_locked(bh)); 187 188 if (ext4_buffer_uptodate(bh)) { 189 unlock_buffer(bh); 190 return; 191 } 192 __ext4_read_bh(bh, op_flags, end_io, simu_fail); 193 } 194 195 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 196 bh_end_io_t *end_io, bool simu_fail) 197 { 198 BUG_ON(!buffer_locked(bh)); 199 200 if (ext4_buffer_uptodate(bh)) { 201 unlock_buffer(bh); 202 return 0; 203 } 204 205 __ext4_read_bh(bh, op_flags, end_io, simu_fail); 206 207 wait_on_buffer(bh); 208 if (buffer_uptodate(bh)) 209 return 0; 210 return -EIO; 211 } 212 213 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 214 { 215 lock_buffer(bh); 216 if (!wait) { 217 ext4_read_bh_nowait(bh, op_flags, NULL, false); 218 return 0; 219 } 220 return ext4_read_bh(bh, op_flags, NULL, false); 221 } 222 223 /* 224 * This works like __bread_gfp() except it uses ERR_PTR for error 225 * returns. Currently with sb_bread it's impossible to distinguish 226 * between ENOMEM and EIO situations (since both result in a NULL 227 * return. 228 */ 229 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 230 sector_t block, 231 blk_opf_t op_flags, gfp_t gfp) 232 { 233 struct buffer_head *bh; 234 int ret; 235 236 bh = sb_getblk_gfp(sb, block, gfp); 237 if (bh == NULL) 238 return ERR_PTR(-ENOMEM); 239 if (ext4_buffer_uptodate(bh)) 240 return bh; 241 242 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 243 if (ret) { 244 put_bh(bh); 245 return ERR_PTR(ret); 246 } 247 return bh; 248 } 249 250 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 251 blk_opf_t op_flags) 252 { 253 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping, 254 ~__GFP_FS) | __GFP_MOVABLE; 255 256 return __ext4_sb_bread_gfp(sb, block, op_flags, gfp); 257 } 258 259 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 260 sector_t block) 261 { 262 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping, 263 ~__GFP_FS); 264 265 return __ext4_sb_bread_gfp(sb, block, 0, gfp); 266 } 267 268 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 269 { 270 struct buffer_head *bh = bdev_getblk(sb->s_bdev, block, 271 sb->s_blocksize, GFP_NOWAIT); 272 273 if (likely(bh)) { 274 if (trylock_buffer(bh)) 275 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL, false); 276 brelse(bh); 277 } 278 } 279 280 static int ext4_verify_csum_type(struct super_block *sb, 281 struct ext4_super_block *es) 282 { 283 if (!ext4_has_feature_metadata_csum(sb)) 284 return 1; 285 286 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 287 } 288 289 __le32 ext4_superblock_csum(struct ext4_super_block *es) 290 { 291 int offset = offsetof(struct ext4_super_block, s_checksum); 292 __u32 csum; 293 294 csum = ext4_chksum(~0, (char *)es, offset); 295 296 return cpu_to_le32(csum); 297 } 298 299 static int ext4_superblock_csum_verify(struct super_block *sb, 300 struct ext4_super_block *es) 301 { 302 if (!ext4_has_feature_metadata_csum(sb)) 303 return 1; 304 305 return es->s_checksum == ext4_superblock_csum(es); 306 } 307 308 void ext4_superblock_csum_set(struct super_block *sb) 309 { 310 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 311 312 if (!ext4_has_feature_metadata_csum(sb)) 313 return; 314 315 es->s_checksum = ext4_superblock_csum(es); 316 } 317 318 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 319 struct ext4_group_desc *bg) 320 { 321 return le32_to_cpu(bg->bg_block_bitmap_lo) | 322 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 323 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 324 } 325 326 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 327 struct ext4_group_desc *bg) 328 { 329 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 330 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 331 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 332 } 333 334 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 335 struct ext4_group_desc *bg) 336 { 337 return le32_to_cpu(bg->bg_inode_table_lo) | 338 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 339 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 340 } 341 342 __u32 ext4_free_group_clusters(struct super_block *sb, 343 struct ext4_group_desc *bg) 344 { 345 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 346 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 347 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 348 } 349 350 __u32 ext4_free_inodes_count(struct super_block *sb, 351 struct ext4_group_desc *bg) 352 { 353 return le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_lo)) | 354 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 355 (__u32)le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_hi)) << 16 : 0); 356 } 357 358 __u32 ext4_used_dirs_count(struct super_block *sb, 359 struct ext4_group_desc *bg) 360 { 361 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 362 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 363 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 364 } 365 366 __u32 ext4_itable_unused_count(struct super_block *sb, 367 struct ext4_group_desc *bg) 368 { 369 return le16_to_cpu(bg->bg_itable_unused_lo) | 370 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 371 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 372 } 373 374 void ext4_block_bitmap_set(struct super_block *sb, 375 struct ext4_group_desc *bg, ext4_fsblk_t blk) 376 { 377 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 378 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 379 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 380 } 381 382 void ext4_inode_bitmap_set(struct super_block *sb, 383 struct ext4_group_desc *bg, ext4_fsblk_t blk) 384 { 385 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 386 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 387 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 388 } 389 390 void ext4_inode_table_set(struct super_block *sb, 391 struct ext4_group_desc *bg, ext4_fsblk_t blk) 392 { 393 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 394 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 395 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 396 } 397 398 void ext4_free_group_clusters_set(struct super_block *sb, 399 struct ext4_group_desc *bg, __u32 count) 400 { 401 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 402 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 403 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 404 } 405 406 void ext4_free_inodes_set(struct super_block *sb, 407 struct ext4_group_desc *bg, __u32 count) 408 { 409 WRITE_ONCE(bg->bg_free_inodes_count_lo, cpu_to_le16((__u16)count)); 410 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 411 WRITE_ONCE(bg->bg_free_inodes_count_hi, cpu_to_le16(count >> 16)); 412 } 413 414 void ext4_used_dirs_set(struct super_block *sb, 415 struct ext4_group_desc *bg, __u32 count) 416 { 417 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 418 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 419 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 420 } 421 422 void ext4_itable_unused_set(struct super_block *sb, 423 struct ext4_group_desc *bg, __u32 count) 424 { 425 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 426 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 427 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 428 } 429 430 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 431 { 432 now = clamp_val(now, 0, (1ull << 40) - 1); 433 434 *lo = cpu_to_le32(lower_32_bits(now)); 435 *hi = upper_32_bits(now); 436 } 437 438 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 439 { 440 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 441 } 442 #define ext4_update_tstamp(es, tstamp) \ 443 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 444 ktime_get_real_seconds()) 445 #define ext4_get_tstamp(es, tstamp) \ 446 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 447 448 /* 449 * The ext4_maybe_update_superblock() function checks and updates the 450 * superblock if needed. 451 * 452 * This function is designed to update the on-disk superblock only under 453 * certain conditions to prevent excessive disk writes and unnecessary 454 * waking of the disk from sleep. The superblock will be updated if: 455 * 1. More than sbi->s_sb_update_sec (def: 1 hour) has passed since the last 456 * superblock update 457 * 2. More than sbi->s_sb_update_kb (def: 16MB) kbs have been written since the 458 * last superblock update. 459 * 460 * @sb: The superblock 461 */ 462 static void ext4_maybe_update_superblock(struct super_block *sb) 463 { 464 struct ext4_sb_info *sbi = EXT4_SB(sb); 465 struct ext4_super_block *es = sbi->s_es; 466 journal_t *journal = sbi->s_journal; 467 time64_t now; 468 __u64 last_update; 469 __u64 lifetime_write_kbytes; 470 __u64 diff_size; 471 472 if (ext4_emergency_state(sb) || sb_rdonly(sb) || 473 !(sb->s_flags & SB_ACTIVE) || !journal || 474 journal->j_flags & JBD2_UNMOUNT) 475 return; 476 477 now = ktime_get_real_seconds(); 478 last_update = ext4_get_tstamp(es, s_wtime); 479 480 if (likely(now - last_update < sbi->s_sb_update_sec)) 481 return; 482 483 lifetime_write_kbytes = sbi->s_kbytes_written + 484 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 485 sbi->s_sectors_written_start) >> 1); 486 487 /* Get the number of kilobytes not written to disk to account 488 * for statistics and compare with a multiple of 16 MB. This 489 * is used to determine when the next superblock commit should 490 * occur (i.e. not more often than once per 16MB if there was 491 * less written in an hour). 492 */ 493 diff_size = lifetime_write_kbytes - le64_to_cpu(es->s_kbytes_written); 494 495 if (diff_size > sbi->s_sb_update_kb) 496 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 497 } 498 499 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 500 { 501 struct super_block *sb = journal->j_private; 502 503 BUG_ON(txn->t_state == T_FINISHED); 504 505 ext4_process_freed_data(sb, txn->t_tid); 506 ext4_maybe_update_superblock(sb); 507 } 508 509 static bool ext4_journalled_writepage_needs_redirty(struct jbd2_inode *jinode, 510 struct folio *folio) 511 { 512 struct buffer_head *bh, *head; 513 struct journal_head *jh; 514 515 bh = head = folio_buffers(folio); 516 do { 517 /* 518 * We have to redirty a page in these cases: 519 * 1) If buffer is dirty, it means the page was dirty because it 520 * contains a buffer that needs checkpointing. So the dirty bit 521 * needs to be preserved so that checkpointing writes the buffer 522 * properly. 523 * 2) If buffer is not part of the committing transaction 524 * (we may have just accidentally come across this buffer because 525 * inode range tracking is not exact) or if the currently running 526 * transaction already contains this buffer as well, dirty bit 527 * needs to be preserved so that the buffer gets writeprotected 528 * properly on running transaction's commit. 529 */ 530 jh = bh2jh(bh); 531 if (buffer_dirty(bh) || 532 (jh && (jh->b_transaction != jinode->i_transaction || 533 jh->b_next_transaction))) 534 return true; 535 } while ((bh = bh->b_this_page) != head); 536 537 return false; 538 } 539 540 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 541 { 542 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 543 struct writeback_control wbc = { 544 .sync_mode = WB_SYNC_ALL, 545 .nr_to_write = LONG_MAX, 546 .range_start = jinode->i_dirty_start, 547 .range_end = jinode->i_dirty_end, 548 }; 549 struct folio *folio = NULL; 550 int error; 551 552 /* 553 * writeback_iter() already checks for dirty pages and calls 554 * folio_clear_dirty_for_io(), which we want to write protect the 555 * folios. 556 * 557 * However, we may have to redirty a folio sometimes. 558 */ 559 while ((folio = writeback_iter(mapping, &wbc, folio, &error))) { 560 if (ext4_journalled_writepage_needs_redirty(jinode, folio)) 561 folio_redirty_for_writepage(&wbc, folio); 562 folio_unlock(folio); 563 } 564 565 return error; 566 } 567 568 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 569 { 570 int ret; 571 572 if (ext4_should_journal_data(jinode->i_vfs_inode)) 573 ret = ext4_journalled_submit_inode_data_buffers(jinode); 574 else 575 ret = ext4_normal_submit_inode_data_buffers(jinode); 576 return ret; 577 } 578 579 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 580 { 581 int ret = 0; 582 583 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 584 ret = jbd2_journal_finish_inode_data_buffers(jinode); 585 586 return ret; 587 } 588 589 static bool system_going_down(void) 590 { 591 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 592 || system_state == SYSTEM_RESTART; 593 } 594 595 struct ext4_err_translation { 596 int code; 597 int errno; 598 }; 599 600 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 601 602 static struct ext4_err_translation err_translation[] = { 603 EXT4_ERR_TRANSLATE(EIO), 604 EXT4_ERR_TRANSLATE(ENOMEM), 605 EXT4_ERR_TRANSLATE(EFSBADCRC), 606 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 607 EXT4_ERR_TRANSLATE(ENOSPC), 608 EXT4_ERR_TRANSLATE(ENOKEY), 609 EXT4_ERR_TRANSLATE(EROFS), 610 EXT4_ERR_TRANSLATE(EFBIG), 611 EXT4_ERR_TRANSLATE(EEXIST), 612 EXT4_ERR_TRANSLATE(ERANGE), 613 EXT4_ERR_TRANSLATE(EOVERFLOW), 614 EXT4_ERR_TRANSLATE(EBUSY), 615 EXT4_ERR_TRANSLATE(ENOTDIR), 616 EXT4_ERR_TRANSLATE(ENOTEMPTY), 617 EXT4_ERR_TRANSLATE(ESHUTDOWN), 618 EXT4_ERR_TRANSLATE(EFAULT), 619 }; 620 621 static int ext4_errno_to_code(int errno) 622 { 623 int i; 624 625 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 626 if (err_translation[i].errno == errno) 627 return err_translation[i].code; 628 return EXT4_ERR_UNKNOWN; 629 } 630 631 static void save_error_info(struct super_block *sb, int error, 632 __u32 ino, __u64 block, 633 const char *func, unsigned int line) 634 { 635 struct ext4_sb_info *sbi = EXT4_SB(sb); 636 637 /* We default to EFSCORRUPTED error... */ 638 if (error == 0) 639 error = EFSCORRUPTED; 640 641 spin_lock(&sbi->s_error_lock); 642 sbi->s_add_error_count++; 643 sbi->s_last_error_code = error; 644 sbi->s_last_error_line = line; 645 sbi->s_last_error_ino = ino; 646 sbi->s_last_error_block = block; 647 sbi->s_last_error_func = func; 648 sbi->s_last_error_time = ktime_get_real_seconds(); 649 if (!sbi->s_first_error_time) { 650 sbi->s_first_error_code = error; 651 sbi->s_first_error_line = line; 652 sbi->s_first_error_ino = ino; 653 sbi->s_first_error_block = block; 654 sbi->s_first_error_func = func; 655 sbi->s_first_error_time = sbi->s_last_error_time; 656 } 657 spin_unlock(&sbi->s_error_lock); 658 } 659 660 /* Deal with the reporting of failure conditions on a filesystem such as 661 * inconsistencies detected or read IO failures. 662 * 663 * On ext2, we can store the error state of the filesystem in the 664 * superblock. That is not possible on ext4, because we may have other 665 * write ordering constraints on the superblock which prevent us from 666 * writing it out straight away; and given that the journal is about to 667 * be aborted, we can't rely on the current, or future, transactions to 668 * write out the superblock safely. 669 * 670 * We'll just use the jbd2_journal_abort() error code to record an error in 671 * the journal instead. On recovery, the journal will complain about 672 * that error until we've noted it down and cleared it. 673 * 674 * If force_ro is set, we unconditionally force the filesystem into an 675 * ABORT|READONLY state, unless the error response on the fs has been set to 676 * panic in which case we take the easy way out and panic immediately. This is 677 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 678 * at a critical moment in log management. 679 */ 680 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 681 __u32 ino, __u64 block, 682 const char *func, unsigned int line) 683 { 684 journal_t *journal = EXT4_SB(sb)->s_journal; 685 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 686 687 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 688 if (test_opt(sb, WARN_ON_ERROR)) 689 WARN_ON_ONCE(1); 690 691 if (!continue_fs && !ext4_emergency_ro(sb) && journal) 692 jbd2_journal_abort(journal, -EIO); 693 694 if (!bdev_read_only(sb->s_bdev)) { 695 save_error_info(sb, error, ino, block, func, line); 696 /* 697 * In case the fs should keep running, we need to writeout 698 * superblock through the journal. Due to lock ordering 699 * constraints, it may not be safe to do it right here so we 700 * defer superblock flushing to a workqueue. We just need to be 701 * careful when the journal is already shutting down. If we get 702 * here in that case, just update the sb directly as the last 703 * transaction won't commit anyway. 704 */ 705 if (continue_fs && journal && 706 !ext4_test_mount_flag(sb, EXT4_MF_JOURNAL_DESTROY)) 707 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 708 else 709 ext4_commit_super(sb); 710 } 711 712 /* 713 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 714 * could panic during 'reboot -f' as the underlying device got already 715 * disabled. 716 */ 717 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 718 panic("EXT4-fs (device %s): panic forced after error\n", 719 sb->s_id); 720 } 721 722 if (ext4_emergency_ro(sb) || continue_fs) 723 return; 724 725 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 726 /* 727 * We don't set SB_RDONLY because that requires sb->s_umount 728 * semaphore and setting it without proper remount procedure is 729 * confusing code such as freeze_super() leading to deadlocks 730 * and other problems. 731 */ 732 set_bit(EXT4_FLAGS_EMERGENCY_RO, &EXT4_SB(sb)->s_ext4_flags); 733 } 734 735 static void update_super_work(struct work_struct *work) 736 { 737 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 738 s_sb_upd_work); 739 journal_t *journal = sbi->s_journal; 740 handle_t *handle; 741 742 /* 743 * If the journal is still running, we have to write out superblock 744 * through the journal to avoid collisions of other journalled sb 745 * updates. 746 * 747 * We use directly jbd2 functions here to avoid recursing back into 748 * ext4 error handling code during handling of previous errors. 749 */ 750 if (!ext4_emergency_state(sbi->s_sb) && 751 !sb_rdonly(sbi->s_sb) && journal) { 752 struct buffer_head *sbh = sbi->s_sbh; 753 bool call_notify_err = false; 754 755 handle = jbd2_journal_start(journal, 1); 756 if (IS_ERR(handle)) 757 goto write_directly; 758 if (jbd2_journal_get_write_access(handle, sbh)) { 759 jbd2_journal_stop(handle); 760 goto write_directly; 761 } 762 763 if (sbi->s_add_error_count > 0) 764 call_notify_err = true; 765 766 ext4_update_super(sbi->s_sb); 767 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 768 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 769 "superblock detected"); 770 clear_buffer_write_io_error(sbh); 771 set_buffer_uptodate(sbh); 772 } 773 774 if (jbd2_journal_dirty_metadata(handle, sbh)) { 775 jbd2_journal_stop(handle); 776 goto write_directly; 777 } 778 jbd2_journal_stop(handle); 779 780 if (call_notify_err) 781 ext4_notify_error_sysfs(sbi); 782 783 return; 784 } 785 write_directly: 786 /* 787 * Write through journal failed. Write sb directly to get error info 788 * out and hope for the best. 789 */ 790 ext4_commit_super(sbi->s_sb); 791 ext4_notify_error_sysfs(sbi); 792 } 793 794 #define ext4_error_ratelimit(sb) \ 795 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 796 "EXT4-fs error") 797 798 void __ext4_error(struct super_block *sb, const char *function, 799 unsigned int line, bool force_ro, int error, __u64 block, 800 const char *fmt, ...) 801 { 802 struct va_format vaf; 803 va_list args; 804 805 if (unlikely(ext4_emergency_state(sb))) 806 return; 807 808 trace_ext4_error(sb, function, line); 809 if (ext4_error_ratelimit(sb)) { 810 va_start(args, fmt); 811 vaf.fmt = fmt; 812 vaf.va = &args; 813 printk(KERN_CRIT 814 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 815 sb->s_id, function, line, current->comm, &vaf); 816 va_end(args); 817 } 818 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED); 819 820 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 821 } 822 823 void __ext4_error_inode(struct inode *inode, const char *function, 824 unsigned int line, ext4_fsblk_t block, int error, 825 const char *fmt, ...) 826 { 827 va_list args; 828 struct va_format vaf; 829 830 if (unlikely(ext4_emergency_state(inode->i_sb))) 831 return; 832 833 trace_ext4_error(inode->i_sb, function, line); 834 if (ext4_error_ratelimit(inode->i_sb)) { 835 va_start(args, fmt); 836 vaf.fmt = fmt; 837 vaf.va = &args; 838 if (block) 839 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 840 "inode #%lu: block %llu: comm %s: %pV\n", 841 inode->i_sb->s_id, function, line, inode->i_ino, 842 block, current->comm, &vaf); 843 else 844 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 845 "inode #%lu: comm %s: %pV\n", 846 inode->i_sb->s_id, function, line, inode->i_ino, 847 current->comm, &vaf); 848 va_end(args); 849 } 850 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED); 851 852 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 853 function, line); 854 } 855 856 void __ext4_error_file(struct file *file, const char *function, 857 unsigned int line, ext4_fsblk_t block, 858 const char *fmt, ...) 859 { 860 va_list args; 861 struct va_format vaf; 862 struct inode *inode = file_inode(file); 863 char pathname[80], *path; 864 865 if (unlikely(ext4_emergency_state(inode->i_sb))) 866 return; 867 868 trace_ext4_error(inode->i_sb, function, line); 869 if (ext4_error_ratelimit(inode->i_sb)) { 870 path = file_path(file, pathname, sizeof(pathname)); 871 if (IS_ERR(path)) 872 path = "(unknown)"; 873 va_start(args, fmt); 874 vaf.fmt = fmt; 875 vaf.va = &args; 876 if (block) 877 printk(KERN_CRIT 878 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 879 "block %llu: comm %s: path %s: %pV\n", 880 inode->i_sb->s_id, function, line, inode->i_ino, 881 block, current->comm, path, &vaf); 882 else 883 printk(KERN_CRIT 884 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 885 "comm %s: path %s: %pV\n", 886 inode->i_sb->s_id, function, line, inode->i_ino, 887 current->comm, path, &vaf); 888 va_end(args); 889 } 890 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED); 891 892 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 893 function, line); 894 } 895 896 const char *ext4_decode_error(struct super_block *sb, int errno, 897 char nbuf[16]) 898 { 899 char *errstr = NULL; 900 901 switch (errno) { 902 case -EFSCORRUPTED: 903 errstr = "Corrupt filesystem"; 904 break; 905 case -EFSBADCRC: 906 errstr = "Filesystem failed CRC"; 907 break; 908 case -EIO: 909 errstr = "IO failure"; 910 break; 911 case -ENOMEM: 912 errstr = "Out of memory"; 913 break; 914 case -EROFS: 915 if (!sb || (EXT4_SB(sb)->s_journal && 916 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 917 errstr = "Journal has aborted"; 918 else 919 errstr = "Readonly filesystem"; 920 break; 921 default: 922 /* If the caller passed in an extra buffer for unknown 923 * errors, textualise them now. Else we just return 924 * NULL. */ 925 if (nbuf) { 926 /* Check for truncated error codes... */ 927 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 928 errstr = nbuf; 929 } 930 break; 931 } 932 933 return errstr; 934 } 935 936 /* __ext4_std_error decodes expected errors from journaling functions 937 * automatically and invokes the appropriate error response. */ 938 939 void __ext4_std_error(struct super_block *sb, const char *function, 940 unsigned int line, int errno) 941 { 942 char nbuf[16]; 943 const char *errstr; 944 945 if (unlikely(ext4_emergency_state(sb))) 946 return; 947 948 /* Special case: if the error is EROFS, and we're not already 949 * inside a transaction, then there's really no point in logging 950 * an error. */ 951 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 952 return; 953 954 if (ext4_error_ratelimit(sb)) { 955 errstr = ext4_decode_error(sb, errno, nbuf); 956 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 957 sb->s_id, function, line, errstr); 958 } 959 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED); 960 961 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 962 } 963 964 void __ext4_msg(struct super_block *sb, 965 const char *prefix, const char *fmt, ...) 966 { 967 struct va_format vaf; 968 va_list args; 969 970 if (sb) { 971 atomic_inc(&EXT4_SB(sb)->s_msg_count); 972 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 973 "EXT4-fs")) 974 return; 975 } 976 977 va_start(args, fmt); 978 vaf.fmt = fmt; 979 vaf.va = &args; 980 if (sb) 981 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 982 else 983 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 984 va_end(args); 985 } 986 987 static int ext4_warning_ratelimit(struct super_block *sb) 988 { 989 atomic_inc(&EXT4_SB(sb)->s_warning_count); 990 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 991 "EXT4-fs warning"); 992 } 993 994 void __ext4_warning(struct super_block *sb, const char *function, 995 unsigned int line, const char *fmt, ...) 996 { 997 struct va_format vaf; 998 va_list args; 999 1000 if (!ext4_warning_ratelimit(sb)) 1001 return; 1002 1003 va_start(args, fmt); 1004 vaf.fmt = fmt; 1005 vaf.va = &args; 1006 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 1007 sb->s_id, function, line, &vaf); 1008 va_end(args); 1009 } 1010 1011 void __ext4_warning_inode(const struct inode *inode, const char *function, 1012 unsigned int line, const char *fmt, ...) 1013 { 1014 struct va_format vaf; 1015 va_list args; 1016 1017 if (!ext4_warning_ratelimit(inode->i_sb)) 1018 return; 1019 1020 va_start(args, fmt); 1021 vaf.fmt = fmt; 1022 vaf.va = &args; 1023 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 1024 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 1025 function, line, inode->i_ino, current->comm, &vaf); 1026 va_end(args); 1027 } 1028 1029 void __ext4_grp_locked_error(const char *function, unsigned int line, 1030 struct super_block *sb, ext4_group_t grp, 1031 unsigned long ino, ext4_fsblk_t block, 1032 const char *fmt, ...) 1033 __releases(bitlock) 1034 __acquires(bitlock) 1035 { 1036 struct va_format vaf; 1037 va_list args; 1038 1039 if (unlikely(ext4_emergency_state(sb))) 1040 return; 1041 1042 trace_ext4_error(sb, function, line); 1043 if (ext4_error_ratelimit(sb)) { 1044 va_start(args, fmt); 1045 vaf.fmt = fmt; 1046 vaf.va = &args; 1047 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1048 sb->s_id, function, line, grp); 1049 if (ino) 1050 printk(KERN_CONT "inode %lu: ", ino); 1051 if (block) 1052 printk(KERN_CONT "block %llu:", 1053 (unsigned long long) block); 1054 printk(KERN_CONT "%pV\n", &vaf); 1055 va_end(args); 1056 } 1057 1058 if (test_opt(sb, ERRORS_CONT)) { 1059 if (test_opt(sb, WARN_ON_ERROR)) 1060 WARN_ON_ONCE(1); 1061 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1062 if (!bdev_read_only(sb->s_bdev)) { 1063 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1064 line); 1065 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 1066 } 1067 return; 1068 } 1069 ext4_unlock_group(sb, grp); 1070 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1071 /* 1072 * We only get here in the ERRORS_RO case; relocking the group 1073 * may be dangerous, but nothing bad will happen since the 1074 * filesystem will have already been marked read/only and the 1075 * journal has been aborted. We return 1 as a hint to callers 1076 * who might what to use the return value from 1077 * ext4_grp_locked_error() to distinguish between the 1078 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1079 * aggressively from the ext4 function in question, with a 1080 * more appropriate error code. 1081 */ 1082 ext4_lock_group(sb, grp); 1083 return; 1084 } 1085 1086 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1087 ext4_group_t group, 1088 unsigned int flags) 1089 { 1090 struct ext4_sb_info *sbi = EXT4_SB(sb); 1091 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1092 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1093 int ret; 1094 1095 if (!grp || !gdp) 1096 return; 1097 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1098 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1099 &grp->bb_state); 1100 if (!ret) 1101 percpu_counter_sub(&sbi->s_freeclusters_counter, 1102 grp->bb_free); 1103 } 1104 1105 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1106 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1107 &grp->bb_state); 1108 if (!ret && gdp) { 1109 int count; 1110 1111 count = ext4_free_inodes_count(sb, gdp); 1112 percpu_counter_sub(&sbi->s_freeinodes_counter, 1113 count); 1114 } 1115 } 1116 } 1117 1118 void ext4_update_dynamic_rev(struct super_block *sb) 1119 { 1120 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1121 1122 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1123 return; 1124 1125 ext4_warning(sb, 1126 "updating to rev %d because of new feature flag, " 1127 "running e2fsck is recommended", 1128 EXT4_DYNAMIC_REV); 1129 1130 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1131 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1132 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1133 /* leave es->s_feature_*compat flags alone */ 1134 /* es->s_uuid will be set by e2fsck if empty */ 1135 1136 /* 1137 * The rest of the superblock fields should be zero, and if not it 1138 * means they are likely already in use, so leave them alone. We 1139 * can leave it up to e2fsck to clean up any inconsistencies there. 1140 */ 1141 } 1142 1143 static inline struct inode *orphan_list_entry(struct list_head *l) 1144 { 1145 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1146 } 1147 1148 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1149 { 1150 struct list_head *l; 1151 1152 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1153 le32_to_cpu(sbi->s_es->s_last_orphan)); 1154 1155 printk(KERN_ERR "sb_info orphan list:\n"); 1156 list_for_each(l, &sbi->s_orphan) { 1157 struct inode *inode = orphan_list_entry(l); 1158 printk(KERN_ERR " " 1159 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1160 inode->i_sb->s_id, inode->i_ino, inode, 1161 inode->i_mode, inode->i_nlink, 1162 NEXT_ORPHAN(inode)); 1163 } 1164 } 1165 1166 #ifdef CONFIG_QUOTA 1167 static int ext4_quota_off(struct super_block *sb, int type); 1168 1169 static inline void ext4_quotas_off(struct super_block *sb, int type) 1170 { 1171 BUG_ON(type > EXT4_MAXQUOTAS); 1172 1173 /* Use our quota_off function to clear inode flags etc. */ 1174 for (type--; type >= 0; type--) 1175 ext4_quota_off(sb, type); 1176 } 1177 1178 /* 1179 * This is a helper function which is used in the mount/remount 1180 * codepaths (which holds s_umount) to fetch the quota file name. 1181 */ 1182 static inline char *get_qf_name(struct super_block *sb, 1183 struct ext4_sb_info *sbi, 1184 int type) 1185 { 1186 return rcu_dereference_protected(sbi->s_qf_names[type], 1187 lockdep_is_held(&sb->s_umount)); 1188 } 1189 #else 1190 static inline void ext4_quotas_off(struct super_block *sb, int type) 1191 { 1192 } 1193 #endif 1194 1195 static int ext4_percpu_param_init(struct ext4_sb_info *sbi) 1196 { 1197 ext4_fsblk_t block; 1198 int err; 1199 1200 block = ext4_count_free_clusters(sbi->s_sb); 1201 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block)); 1202 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 1203 GFP_KERNEL); 1204 if (!err) { 1205 unsigned long freei = ext4_count_free_inodes(sbi->s_sb); 1206 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 1207 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 1208 GFP_KERNEL); 1209 } 1210 if (!err) 1211 err = percpu_counter_init(&sbi->s_dirs_counter, 1212 ext4_count_dirs(sbi->s_sb), GFP_KERNEL); 1213 if (!err) 1214 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 1215 GFP_KERNEL); 1216 if (!err) 1217 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 1218 GFP_KERNEL); 1219 if (!err) 1220 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 1221 1222 if (err) 1223 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory"); 1224 1225 return err; 1226 } 1227 1228 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi) 1229 { 1230 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1231 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1232 percpu_counter_destroy(&sbi->s_dirs_counter); 1233 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1234 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1235 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1236 } 1237 1238 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 1239 { 1240 struct buffer_head **group_desc; 1241 int i; 1242 1243 rcu_read_lock(); 1244 group_desc = rcu_dereference(sbi->s_group_desc); 1245 for (i = 0; i < sbi->s_gdb_count; i++) 1246 brelse(group_desc[i]); 1247 kvfree(group_desc); 1248 rcu_read_unlock(); 1249 } 1250 1251 static void ext4_flex_groups_free(struct ext4_sb_info *sbi) 1252 { 1253 struct flex_groups **flex_groups; 1254 int i; 1255 1256 rcu_read_lock(); 1257 flex_groups = rcu_dereference(sbi->s_flex_groups); 1258 if (flex_groups) { 1259 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1260 kvfree(flex_groups[i]); 1261 kvfree(flex_groups); 1262 } 1263 rcu_read_unlock(); 1264 } 1265 1266 static void ext4_put_super(struct super_block *sb) 1267 { 1268 struct ext4_sb_info *sbi = EXT4_SB(sb); 1269 struct ext4_super_block *es = sbi->s_es; 1270 int aborted = 0; 1271 int err; 1272 1273 /* 1274 * Unregister sysfs before destroying jbd2 journal. 1275 * Since we could still access attr_journal_task attribute via sysfs 1276 * path which could have sbi->s_journal->j_task as NULL 1277 * Unregister sysfs before flush sbi->s_sb_upd_work. 1278 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1279 * read metadata verify failed then will queue error work. 1280 * update_super_work will call start_this_handle may trigger 1281 * BUG_ON. 1282 */ 1283 ext4_unregister_sysfs(sb); 1284 1285 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1286 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.", 1287 &sb->s_uuid); 1288 1289 ext4_unregister_li_request(sb); 1290 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 1291 1292 destroy_workqueue(sbi->rsv_conversion_wq); 1293 ext4_release_orphan_info(sb); 1294 1295 if (sbi->s_journal) { 1296 aborted = is_journal_aborted(sbi->s_journal); 1297 err = ext4_journal_destroy(sbi, sbi->s_journal); 1298 if ((err < 0) && !aborted) { 1299 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1300 } 1301 } else 1302 flush_work(&sbi->s_sb_upd_work); 1303 1304 ext4_es_unregister_shrinker(sbi); 1305 timer_shutdown_sync(&sbi->s_err_report); 1306 ext4_release_system_zone(sb); 1307 ext4_mb_release(sb); 1308 ext4_ext_release(sb); 1309 1310 if (!ext4_emergency_state(sb) && !sb_rdonly(sb)) { 1311 if (!aborted) { 1312 ext4_clear_feature_journal_needs_recovery(sb); 1313 ext4_clear_feature_orphan_present(sb); 1314 es->s_state = cpu_to_le16(sbi->s_mount_state); 1315 } 1316 ext4_commit_super(sb); 1317 } 1318 1319 ext4_group_desc_free(sbi); 1320 ext4_flex_groups_free(sbi); 1321 1322 WARN_ON_ONCE(!(sbi->s_mount_state & EXT4_ERROR_FS) && 1323 percpu_counter_sum(&sbi->s_dirtyclusters_counter)); 1324 ext4_percpu_param_destroy(sbi); 1325 #ifdef CONFIG_QUOTA 1326 for (int i = 0; i < EXT4_MAXQUOTAS; i++) 1327 kfree(get_qf_name(sb, sbi, i)); 1328 #endif 1329 1330 /* Debugging code just in case the in-memory inode orphan list 1331 * isn't empty. The on-disk one can be non-empty if we've 1332 * detected an error and taken the fs readonly, but the 1333 * in-memory list had better be clean by this point. */ 1334 if (!list_empty(&sbi->s_orphan)) 1335 dump_orphan_list(sb, sbi); 1336 ASSERT(list_empty(&sbi->s_orphan)); 1337 1338 sync_blockdev(sb->s_bdev); 1339 invalidate_bdev(sb->s_bdev); 1340 if (sbi->s_journal_bdev_file) { 1341 /* 1342 * Invalidate the journal device's buffers. We don't want them 1343 * floating about in memory - the physical journal device may 1344 * hotswapped, and it breaks the `ro-after' testing code. 1345 */ 1346 sync_blockdev(file_bdev(sbi->s_journal_bdev_file)); 1347 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 1348 } 1349 1350 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1351 sbi->s_ea_inode_cache = NULL; 1352 1353 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1354 sbi->s_ea_block_cache = NULL; 1355 1356 ext4_stop_mmpd(sbi); 1357 1358 brelse(sbi->s_sbh); 1359 sb->s_fs_info = NULL; 1360 /* 1361 * Now that we are completely done shutting down the 1362 * superblock, we need to actually destroy the kobject. 1363 */ 1364 kobject_put(&sbi->s_kobj); 1365 wait_for_completion(&sbi->s_kobj_unregister); 1366 kfree(sbi->s_blockgroup_lock); 1367 fs_put_dax(sbi->s_daxdev, NULL); 1368 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1369 #if IS_ENABLED(CONFIG_UNICODE) 1370 utf8_unload(sb->s_encoding); 1371 #endif 1372 kfree(sbi); 1373 } 1374 1375 static struct kmem_cache *ext4_inode_cachep; 1376 1377 /* 1378 * Called inside transaction, so use GFP_NOFS 1379 */ 1380 static struct inode *ext4_alloc_inode(struct super_block *sb) 1381 { 1382 struct ext4_inode_info *ei; 1383 1384 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1385 if (!ei) 1386 return NULL; 1387 1388 inode_set_iversion(&ei->vfs_inode, 1); 1389 ei->i_flags = 0; 1390 spin_lock_init(&ei->i_raw_lock); 1391 ei->i_prealloc_node = RB_ROOT; 1392 atomic_set(&ei->i_prealloc_active, 0); 1393 rwlock_init(&ei->i_prealloc_lock); 1394 ext4_es_init_tree(&ei->i_es_tree); 1395 rwlock_init(&ei->i_es_lock); 1396 INIT_LIST_HEAD(&ei->i_es_list); 1397 ei->i_es_all_nr = 0; 1398 ei->i_es_shk_nr = 0; 1399 ei->i_es_shrink_lblk = 0; 1400 ei->i_reserved_data_blocks = 0; 1401 spin_lock_init(&(ei->i_block_reservation_lock)); 1402 ext4_init_pending_tree(&ei->i_pending_tree); 1403 #ifdef CONFIG_QUOTA 1404 ei->i_reserved_quota = 0; 1405 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1406 #endif 1407 ei->jinode = NULL; 1408 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1409 spin_lock_init(&ei->i_completed_io_lock); 1410 ei->i_sync_tid = 0; 1411 ei->i_datasync_tid = 0; 1412 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1413 ext4_fc_init_inode(&ei->vfs_inode); 1414 spin_lock_init(&ei->i_fc_lock); 1415 return &ei->vfs_inode; 1416 } 1417 1418 static int ext4_drop_inode(struct inode *inode) 1419 { 1420 int drop = inode_generic_drop(inode); 1421 1422 if (!drop) 1423 drop = fscrypt_drop_inode(inode); 1424 1425 trace_ext4_drop_inode(inode, drop); 1426 return drop; 1427 } 1428 1429 static void ext4_free_in_core_inode(struct inode *inode) 1430 { 1431 fscrypt_free_inode(inode); 1432 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1433 pr_warn("%s: inode %ld still in fc list", 1434 __func__, inode->i_ino); 1435 } 1436 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1437 } 1438 1439 static void ext4_destroy_inode(struct inode *inode) 1440 { 1441 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1442 ext4_msg(inode->i_sb, KERN_ERR, 1443 "Inode %lu (%p): orphan list check failed!", 1444 inode->i_ino, EXT4_I(inode)); 1445 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1446 EXT4_I(inode), sizeof(struct ext4_inode_info), 1447 true); 1448 dump_stack(); 1449 } 1450 1451 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ERROR_FS) && 1452 WARN_ON_ONCE(EXT4_I(inode)->i_reserved_data_blocks)) 1453 ext4_msg(inode->i_sb, KERN_ERR, 1454 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!", 1455 inode->i_ino, EXT4_I(inode), 1456 EXT4_I(inode)->i_reserved_data_blocks); 1457 } 1458 1459 static void ext4_shutdown(struct super_block *sb) 1460 { 1461 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH); 1462 } 1463 1464 static void init_once(void *foo) 1465 { 1466 struct ext4_inode_info *ei = foo; 1467 1468 INIT_LIST_HEAD(&ei->i_orphan); 1469 init_rwsem(&ei->xattr_sem); 1470 init_rwsem(&ei->i_data_sem); 1471 inode_init_once(&ei->vfs_inode); 1472 ext4_fc_init_inode(&ei->vfs_inode); 1473 #ifdef CONFIG_FS_ENCRYPTION 1474 ei->i_crypt_info = NULL; 1475 #endif 1476 #ifdef CONFIG_FS_VERITY 1477 ei->i_verity_info = NULL; 1478 #endif 1479 } 1480 1481 static int __init init_inodecache(void) 1482 { 1483 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1484 sizeof(struct ext4_inode_info), 0, 1485 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT, 1486 offsetof(struct ext4_inode_info, i_data), 1487 sizeof_field(struct ext4_inode_info, i_data), 1488 init_once); 1489 if (ext4_inode_cachep == NULL) 1490 return -ENOMEM; 1491 return 0; 1492 } 1493 1494 static void destroy_inodecache(void) 1495 { 1496 /* 1497 * Make sure all delayed rcu free inodes are flushed before we 1498 * destroy cache. 1499 */ 1500 rcu_barrier(); 1501 kmem_cache_destroy(ext4_inode_cachep); 1502 } 1503 1504 void ext4_clear_inode(struct inode *inode) 1505 { 1506 ext4_fc_del(inode); 1507 invalidate_inode_buffers(inode); 1508 clear_inode(inode); 1509 ext4_discard_preallocations(inode); 1510 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1511 dquot_drop(inode); 1512 if (EXT4_I(inode)->jinode) { 1513 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1514 EXT4_I(inode)->jinode); 1515 jbd2_free_inode(EXT4_I(inode)->jinode); 1516 EXT4_I(inode)->jinode = NULL; 1517 } 1518 fscrypt_put_encryption_info(inode); 1519 fsverity_cleanup_inode(inode); 1520 } 1521 1522 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1523 u64 ino, u32 generation) 1524 { 1525 struct inode *inode; 1526 1527 /* 1528 * Currently we don't know the generation for parent directory, so 1529 * a generation of 0 means "accept any" 1530 */ 1531 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1532 if (IS_ERR(inode)) 1533 return ERR_CAST(inode); 1534 if (generation && inode->i_generation != generation) { 1535 iput(inode); 1536 return ERR_PTR(-ESTALE); 1537 } 1538 1539 return inode; 1540 } 1541 1542 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1543 int fh_len, int fh_type) 1544 { 1545 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1546 ext4_nfs_get_inode); 1547 } 1548 1549 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1550 int fh_len, int fh_type) 1551 { 1552 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1553 ext4_nfs_get_inode); 1554 } 1555 1556 static int ext4_nfs_commit_metadata(struct inode *inode) 1557 { 1558 struct writeback_control wbc = { 1559 .sync_mode = WB_SYNC_ALL 1560 }; 1561 1562 trace_ext4_nfs_commit_metadata(inode); 1563 return ext4_write_inode(inode, &wbc); 1564 } 1565 1566 #ifdef CONFIG_QUOTA 1567 static const char * const quotatypes[] = INITQFNAMES; 1568 #define QTYPE2NAME(t) (quotatypes[t]) 1569 1570 static int ext4_write_dquot(struct dquot *dquot); 1571 static int ext4_acquire_dquot(struct dquot *dquot); 1572 static int ext4_release_dquot(struct dquot *dquot); 1573 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1574 static int ext4_write_info(struct super_block *sb, int type); 1575 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1576 const struct path *path); 1577 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1578 size_t len, loff_t off); 1579 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1580 const char *data, size_t len, loff_t off); 1581 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1582 unsigned int flags); 1583 1584 static struct dquot __rcu **ext4_get_dquots(struct inode *inode) 1585 { 1586 return EXT4_I(inode)->i_dquot; 1587 } 1588 1589 static const struct dquot_operations ext4_quota_operations = { 1590 .get_reserved_space = ext4_get_reserved_space, 1591 .write_dquot = ext4_write_dquot, 1592 .acquire_dquot = ext4_acquire_dquot, 1593 .release_dquot = ext4_release_dquot, 1594 .mark_dirty = ext4_mark_dquot_dirty, 1595 .write_info = ext4_write_info, 1596 .alloc_dquot = dquot_alloc, 1597 .destroy_dquot = dquot_destroy, 1598 .get_projid = ext4_get_projid, 1599 .get_inode_usage = ext4_get_inode_usage, 1600 .get_next_id = dquot_get_next_id, 1601 }; 1602 1603 static const struct quotactl_ops ext4_qctl_operations = { 1604 .quota_on = ext4_quota_on, 1605 .quota_off = ext4_quota_off, 1606 .quota_sync = dquot_quota_sync, 1607 .get_state = dquot_get_state, 1608 .set_info = dquot_set_dqinfo, 1609 .get_dqblk = dquot_get_dqblk, 1610 .set_dqblk = dquot_set_dqblk, 1611 .get_nextdqblk = dquot_get_next_dqblk, 1612 }; 1613 #endif 1614 1615 static const struct super_operations ext4_sops = { 1616 .alloc_inode = ext4_alloc_inode, 1617 .free_inode = ext4_free_in_core_inode, 1618 .destroy_inode = ext4_destroy_inode, 1619 .write_inode = ext4_write_inode, 1620 .dirty_inode = ext4_dirty_inode, 1621 .drop_inode = ext4_drop_inode, 1622 .evict_inode = ext4_evict_inode, 1623 .put_super = ext4_put_super, 1624 .sync_fs = ext4_sync_fs, 1625 .freeze_fs = ext4_freeze, 1626 .unfreeze_fs = ext4_unfreeze, 1627 .statfs = ext4_statfs, 1628 .show_options = ext4_show_options, 1629 .shutdown = ext4_shutdown, 1630 #ifdef CONFIG_QUOTA 1631 .quota_read = ext4_quota_read, 1632 .quota_write = ext4_quota_write, 1633 .get_dquots = ext4_get_dquots, 1634 #endif 1635 }; 1636 1637 static const struct export_operations ext4_export_ops = { 1638 .encode_fh = generic_encode_ino32_fh, 1639 .fh_to_dentry = ext4_fh_to_dentry, 1640 .fh_to_parent = ext4_fh_to_parent, 1641 .get_parent = ext4_get_parent, 1642 .commit_metadata = ext4_nfs_commit_metadata, 1643 }; 1644 1645 enum { 1646 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1647 Opt_resgid, Opt_resuid, Opt_sb, 1648 Opt_nouid32, Opt_debug, Opt_removed, 1649 Opt_user_xattr, Opt_acl, 1650 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1651 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1652 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1653 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1654 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1655 Opt_inlinecrypt, 1656 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1657 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1658 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1659 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1660 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1661 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1662 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1663 Opt_inode_readahead_blks, Opt_journal_ioprio, 1664 Opt_dioread_nolock, Opt_dioread_lock, 1665 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1666 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1667 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1668 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1669 #ifdef CONFIG_EXT4_DEBUG 1670 Opt_fc_debug_max_replay, Opt_fc_debug_force 1671 #endif 1672 }; 1673 1674 static const struct constant_table ext4_param_errors[] = { 1675 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1676 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1677 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1678 {} 1679 }; 1680 1681 static const struct constant_table ext4_param_data[] = { 1682 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1683 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1684 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1685 {} 1686 }; 1687 1688 static const struct constant_table ext4_param_data_err[] = { 1689 {"abort", Opt_data_err_abort}, 1690 {"ignore", Opt_data_err_ignore}, 1691 {} 1692 }; 1693 1694 static const struct constant_table ext4_param_jqfmt[] = { 1695 {"vfsold", QFMT_VFS_OLD}, 1696 {"vfsv0", QFMT_VFS_V0}, 1697 {"vfsv1", QFMT_VFS_V1}, 1698 {} 1699 }; 1700 1701 static const struct constant_table ext4_param_dax[] = { 1702 {"always", Opt_dax_always}, 1703 {"inode", Opt_dax_inode}, 1704 {"never", Opt_dax_never}, 1705 {} 1706 }; 1707 1708 /* 1709 * Mount option specification 1710 * We don't use fsparam_flag_no because of the way we set the 1711 * options and the way we show them in _ext4_show_options(). To 1712 * keep the changes to a minimum, let's keep the negative options 1713 * separate for now. 1714 */ 1715 static const struct fs_parameter_spec ext4_param_specs[] = { 1716 fsparam_flag ("bsddf", Opt_bsd_df), 1717 fsparam_flag ("minixdf", Opt_minix_df), 1718 fsparam_flag ("grpid", Opt_grpid), 1719 fsparam_flag ("bsdgroups", Opt_grpid), 1720 fsparam_flag ("nogrpid", Opt_nogrpid), 1721 fsparam_flag ("sysvgroups", Opt_nogrpid), 1722 fsparam_gid ("resgid", Opt_resgid), 1723 fsparam_uid ("resuid", Opt_resuid), 1724 fsparam_u32 ("sb", Opt_sb), 1725 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1726 fsparam_flag ("nouid32", Opt_nouid32), 1727 fsparam_flag ("debug", Opt_debug), 1728 fsparam_flag ("oldalloc", Opt_removed), 1729 fsparam_flag ("orlov", Opt_removed), 1730 fsparam_flag ("user_xattr", Opt_user_xattr), 1731 fsparam_flag ("acl", Opt_acl), 1732 fsparam_flag ("norecovery", Opt_noload), 1733 fsparam_flag ("noload", Opt_noload), 1734 fsparam_flag ("bh", Opt_removed), 1735 fsparam_flag ("nobh", Opt_removed), 1736 fsparam_u32 ("commit", Opt_commit), 1737 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1738 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1739 fsparam_u32 ("journal_dev", Opt_journal_dev), 1740 fsparam_bdev ("journal_path", Opt_journal_path), 1741 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1742 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1743 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1744 fsparam_flag ("abort", Opt_abort), 1745 fsparam_enum ("data", Opt_data, ext4_param_data), 1746 fsparam_enum ("data_err", Opt_data_err, 1747 ext4_param_data_err), 1748 fsparam_string_empty 1749 ("usrjquota", Opt_usrjquota), 1750 fsparam_string_empty 1751 ("grpjquota", Opt_grpjquota), 1752 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1753 fsparam_flag ("grpquota", Opt_grpquota), 1754 fsparam_flag ("quota", Opt_quota), 1755 fsparam_flag ("noquota", Opt_noquota), 1756 fsparam_flag ("usrquota", Opt_usrquota), 1757 fsparam_flag ("prjquota", Opt_prjquota), 1758 fsparam_flag ("barrier", Opt_barrier), 1759 fsparam_u32 ("barrier", Opt_barrier), 1760 fsparam_flag ("nobarrier", Opt_nobarrier), 1761 fsparam_flag ("i_version", Opt_removed), 1762 fsparam_flag ("dax", Opt_dax), 1763 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1764 fsparam_u32 ("stripe", Opt_stripe), 1765 fsparam_flag ("delalloc", Opt_delalloc), 1766 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1767 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1768 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1769 fsparam_u32 ("debug_want_extra_isize", 1770 Opt_debug_want_extra_isize), 1771 fsparam_flag ("mblk_io_submit", Opt_removed), 1772 fsparam_flag ("nomblk_io_submit", Opt_removed), 1773 fsparam_flag ("block_validity", Opt_block_validity), 1774 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1775 fsparam_u32 ("inode_readahead_blks", 1776 Opt_inode_readahead_blks), 1777 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1778 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1779 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1780 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1781 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1782 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1783 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1784 fsparam_flag ("discard", Opt_discard), 1785 fsparam_flag ("nodiscard", Opt_nodiscard), 1786 fsparam_u32 ("init_itable", Opt_init_itable), 1787 fsparam_flag ("init_itable", Opt_init_itable), 1788 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1789 #ifdef CONFIG_EXT4_DEBUG 1790 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1791 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1792 #endif 1793 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1794 fsparam_flag ("test_dummy_encryption", 1795 Opt_test_dummy_encryption), 1796 fsparam_string ("test_dummy_encryption", 1797 Opt_test_dummy_encryption), 1798 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1799 fsparam_flag ("nombcache", Opt_nombcache), 1800 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1801 fsparam_flag ("prefetch_block_bitmaps", 1802 Opt_removed), 1803 fsparam_flag ("no_prefetch_block_bitmaps", 1804 Opt_no_prefetch_block_bitmaps), 1805 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1806 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1807 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1808 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1809 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1810 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1811 {} 1812 }; 1813 1814 1815 #define MOPT_SET 0x0001 1816 #define MOPT_CLEAR 0x0002 1817 #define MOPT_NOSUPPORT 0x0004 1818 #define MOPT_EXPLICIT 0x0008 1819 #ifdef CONFIG_QUOTA 1820 #define MOPT_Q 0 1821 #define MOPT_QFMT 0x0010 1822 #else 1823 #define MOPT_Q MOPT_NOSUPPORT 1824 #define MOPT_QFMT MOPT_NOSUPPORT 1825 #endif 1826 #define MOPT_NO_EXT2 0x0020 1827 #define MOPT_NO_EXT3 0x0040 1828 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1829 #define MOPT_SKIP 0x0080 1830 #define MOPT_2 0x0100 1831 1832 static const struct mount_opts { 1833 int token; 1834 int mount_opt; 1835 int flags; 1836 } ext4_mount_opts[] = { 1837 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1838 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1839 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1840 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1841 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1842 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1843 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1844 MOPT_EXT4_ONLY | MOPT_SET}, 1845 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1846 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1847 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1848 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1849 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1850 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1851 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1852 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1853 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1854 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1855 {Opt_commit, 0, MOPT_NO_EXT2}, 1856 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1857 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1858 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1859 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1860 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1861 EXT4_MOUNT_JOURNAL_CHECKSUM), 1862 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1863 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1864 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1865 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1866 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1867 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1868 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1869 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1870 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1871 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1872 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1873 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1874 {Opt_data, 0, MOPT_NO_EXT2}, 1875 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1876 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1877 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1878 #else 1879 {Opt_acl, 0, MOPT_NOSUPPORT}, 1880 #endif 1881 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1882 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1883 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1884 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1885 MOPT_SET | MOPT_Q}, 1886 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1887 MOPT_SET | MOPT_Q}, 1888 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1889 MOPT_SET | MOPT_Q}, 1890 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1891 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1892 MOPT_CLEAR | MOPT_Q}, 1893 {Opt_usrjquota, 0, MOPT_Q}, 1894 {Opt_grpjquota, 0, MOPT_Q}, 1895 {Opt_jqfmt, 0, MOPT_QFMT}, 1896 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1897 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1898 MOPT_SET}, 1899 #ifdef CONFIG_EXT4_DEBUG 1900 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1901 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1902 #endif 1903 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2}, 1904 {Opt_err, 0, 0} 1905 }; 1906 1907 #if IS_ENABLED(CONFIG_UNICODE) 1908 static const struct ext4_sb_encodings { 1909 __u16 magic; 1910 char *name; 1911 unsigned int version; 1912 } ext4_sb_encoding_map[] = { 1913 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1914 }; 1915 1916 static const struct ext4_sb_encodings * 1917 ext4_sb_read_encoding(const struct ext4_super_block *es) 1918 { 1919 __u16 magic = le16_to_cpu(es->s_encoding); 1920 int i; 1921 1922 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1923 if (magic == ext4_sb_encoding_map[i].magic) 1924 return &ext4_sb_encoding_map[i]; 1925 1926 return NULL; 1927 } 1928 #endif 1929 1930 #define EXT4_SPEC_JQUOTA (1 << 0) 1931 #define EXT4_SPEC_JQFMT (1 << 1) 1932 #define EXT4_SPEC_DATAJ (1 << 2) 1933 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1934 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1935 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1936 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 1937 #define EXT4_SPEC_s_max_batch_time (1 << 8) 1938 #define EXT4_SPEC_s_min_batch_time (1 << 9) 1939 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 1940 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 1941 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 1942 #define EXT4_SPEC_s_stripe (1 << 13) 1943 #define EXT4_SPEC_s_resuid (1 << 14) 1944 #define EXT4_SPEC_s_resgid (1 << 15) 1945 #define EXT4_SPEC_s_commit_interval (1 << 16) 1946 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 1947 #define EXT4_SPEC_s_sb_block (1 << 18) 1948 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 1949 1950 struct ext4_fs_context { 1951 char *s_qf_names[EXT4_MAXQUOTAS]; 1952 struct fscrypt_dummy_policy dummy_enc_policy; 1953 int s_jquota_fmt; /* Format of quota to use */ 1954 #ifdef CONFIG_EXT4_DEBUG 1955 int s_fc_debug_max_replay; 1956 #endif 1957 unsigned short qname_spec; 1958 unsigned long vals_s_flags; /* Bits to set in s_flags */ 1959 unsigned long mask_s_flags; /* Bits changed in s_flags */ 1960 unsigned long journal_devnum; 1961 unsigned long s_commit_interval; 1962 unsigned long s_stripe; 1963 unsigned int s_inode_readahead_blks; 1964 unsigned int s_want_extra_isize; 1965 unsigned int s_li_wait_mult; 1966 unsigned int s_max_dir_size_kb; 1967 unsigned int journal_ioprio; 1968 unsigned int vals_s_mount_opt; 1969 unsigned int mask_s_mount_opt; 1970 unsigned int vals_s_mount_opt2; 1971 unsigned int mask_s_mount_opt2; 1972 unsigned int opt_flags; /* MOPT flags */ 1973 unsigned int spec; 1974 u32 s_max_batch_time; 1975 u32 s_min_batch_time; 1976 kuid_t s_resuid; 1977 kgid_t s_resgid; 1978 ext4_fsblk_t s_sb_block; 1979 }; 1980 1981 static void ext4_fc_free(struct fs_context *fc) 1982 { 1983 struct ext4_fs_context *ctx = fc->fs_private; 1984 int i; 1985 1986 if (!ctx) 1987 return; 1988 1989 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1990 kfree(ctx->s_qf_names[i]); 1991 1992 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 1993 kfree(ctx); 1994 } 1995 1996 int ext4_init_fs_context(struct fs_context *fc) 1997 { 1998 struct ext4_fs_context *ctx; 1999 2000 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2001 if (!ctx) 2002 return -ENOMEM; 2003 2004 fc->fs_private = ctx; 2005 fc->ops = &ext4_context_ops; 2006 2007 /* i_version is always enabled now */ 2008 fc->sb_flags |= SB_I_VERSION; 2009 2010 return 0; 2011 } 2012 2013 #ifdef CONFIG_QUOTA 2014 /* 2015 * Note the name of the specified quota file. 2016 */ 2017 static int note_qf_name(struct fs_context *fc, int qtype, 2018 struct fs_parameter *param) 2019 { 2020 struct ext4_fs_context *ctx = fc->fs_private; 2021 char *qname; 2022 2023 if (param->size < 1) { 2024 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 2025 return -EINVAL; 2026 } 2027 if (strchr(param->string, '/')) { 2028 ext4_msg(NULL, KERN_ERR, 2029 "quotafile must be on filesystem root"); 2030 return -EINVAL; 2031 } 2032 if (ctx->s_qf_names[qtype]) { 2033 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 2034 ext4_msg(NULL, KERN_ERR, 2035 "%s quota file already specified", 2036 QTYPE2NAME(qtype)); 2037 return -EINVAL; 2038 } 2039 return 0; 2040 } 2041 2042 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 2043 if (!qname) { 2044 ext4_msg(NULL, KERN_ERR, 2045 "Not enough memory for storing quotafile name"); 2046 return -ENOMEM; 2047 } 2048 ctx->s_qf_names[qtype] = qname; 2049 ctx->qname_spec |= 1 << qtype; 2050 ctx->spec |= EXT4_SPEC_JQUOTA; 2051 return 0; 2052 } 2053 2054 /* 2055 * Clear the name of the specified quota file. 2056 */ 2057 static int unnote_qf_name(struct fs_context *fc, int qtype) 2058 { 2059 struct ext4_fs_context *ctx = fc->fs_private; 2060 2061 kfree(ctx->s_qf_names[qtype]); 2062 2063 ctx->s_qf_names[qtype] = NULL; 2064 ctx->qname_spec |= 1 << qtype; 2065 ctx->spec |= EXT4_SPEC_JQUOTA; 2066 return 0; 2067 } 2068 #endif 2069 2070 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2071 struct ext4_fs_context *ctx) 2072 { 2073 int err; 2074 2075 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2076 ext4_msg(NULL, KERN_WARNING, 2077 "test_dummy_encryption option not supported"); 2078 return -EINVAL; 2079 } 2080 err = fscrypt_parse_test_dummy_encryption(param, 2081 &ctx->dummy_enc_policy); 2082 if (err == -EINVAL) { 2083 ext4_msg(NULL, KERN_WARNING, 2084 "Value of option \"%s\" is unrecognized", param->key); 2085 } else if (err == -EEXIST) { 2086 ext4_msg(NULL, KERN_WARNING, 2087 "Conflicting test_dummy_encryption options"); 2088 return -EINVAL; 2089 } 2090 return err; 2091 } 2092 2093 #define EXT4_SET_CTX(name) \ 2094 static inline __maybe_unused \ 2095 void ctx_set_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2096 { \ 2097 ctx->mask_s_##name |= flag; \ 2098 ctx->vals_s_##name |= flag; \ 2099 } 2100 2101 #define EXT4_CLEAR_CTX(name) \ 2102 static inline __maybe_unused \ 2103 void ctx_clear_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2104 { \ 2105 ctx->mask_s_##name |= flag; \ 2106 ctx->vals_s_##name &= ~flag; \ 2107 } 2108 2109 #define EXT4_TEST_CTX(name) \ 2110 static inline unsigned long \ 2111 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2112 { \ 2113 return (ctx->vals_s_##name & flag); \ 2114 } 2115 2116 EXT4_SET_CTX(flags); /* set only */ 2117 EXT4_SET_CTX(mount_opt); 2118 EXT4_CLEAR_CTX(mount_opt); 2119 EXT4_TEST_CTX(mount_opt); 2120 EXT4_SET_CTX(mount_opt2); 2121 EXT4_CLEAR_CTX(mount_opt2); 2122 EXT4_TEST_CTX(mount_opt2); 2123 2124 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2125 { 2126 struct ext4_fs_context *ctx = fc->fs_private; 2127 struct fs_parse_result result; 2128 const struct mount_opts *m; 2129 int is_remount; 2130 int token; 2131 2132 token = fs_parse(fc, ext4_param_specs, param, &result); 2133 if (token < 0) 2134 return token; 2135 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2136 2137 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2138 if (token == m->token) 2139 break; 2140 2141 ctx->opt_flags |= m->flags; 2142 2143 if (m->flags & MOPT_EXPLICIT) { 2144 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2145 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2146 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2147 ctx_set_mount_opt2(ctx, 2148 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2149 } else 2150 return -EINVAL; 2151 } 2152 2153 if (m->flags & MOPT_NOSUPPORT) { 2154 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2155 param->key); 2156 return 0; 2157 } 2158 2159 switch (token) { 2160 #ifdef CONFIG_QUOTA 2161 case Opt_usrjquota: 2162 if (!*param->string) 2163 return unnote_qf_name(fc, USRQUOTA); 2164 else 2165 return note_qf_name(fc, USRQUOTA, param); 2166 case Opt_grpjquota: 2167 if (!*param->string) 2168 return unnote_qf_name(fc, GRPQUOTA); 2169 else 2170 return note_qf_name(fc, GRPQUOTA, param); 2171 #endif 2172 case Opt_sb: 2173 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2174 ext4_msg(NULL, KERN_WARNING, 2175 "Ignoring %s option on remount", param->key); 2176 } else { 2177 ctx->s_sb_block = result.uint_32; 2178 ctx->spec |= EXT4_SPEC_s_sb_block; 2179 } 2180 return 0; 2181 case Opt_removed: 2182 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2183 param->key); 2184 return 0; 2185 case Opt_inlinecrypt: 2186 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2187 ctx_set_flags(ctx, SB_INLINECRYPT); 2188 #else 2189 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2190 #endif 2191 return 0; 2192 case Opt_errors: 2193 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2194 ctx_set_mount_opt(ctx, result.uint_32); 2195 return 0; 2196 #ifdef CONFIG_QUOTA 2197 case Opt_jqfmt: 2198 ctx->s_jquota_fmt = result.uint_32; 2199 ctx->spec |= EXT4_SPEC_JQFMT; 2200 return 0; 2201 #endif 2202 case Opt_data: 2203 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2204 ctx_set_mount_opt(ctx, result.uint_32); 2205 ctx->spec |= EXT4_SPEC_DATAJ; 2206 return 0; 2207 case Opt_commit: 2208 if (result.uint_32 == 0) 2209 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE; 2210 else if (result.uint_32 > INT_MAX / HZ) { 2211 ext4_msg(NULL, KERN_ERR, 2212 "Invalid commit interval %d, " 2213 "must be smaller than %d", 2214 result.uint_32, INT_MAX / HZ); 2215 return -EINVAL; 2216 } 2217 ctx->s_commit_interval = HZ * result.uint_32; 2218 ctx->spec |= EXT4_SPEC_s_commit_interval; 2219 return 0; 2220 case Opt_debug_want_extra_isize: 2221 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2222 ext4_msg(NULL, KERN_ERR, 2223 "Invalid want_extra_isize %d", result.uint_32); 2224 return -EINVAL; 2225 } 2226 ctx->s_want_extra_isize = result.uint_32; 2227 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2228 return 0; 2229 case Opt_max_batch_time: 2230 ctx->s_max_batch_time = result.uint_32; 2231 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2232 return 0; 2233 case Opt_min_batch_time: 2234 ctx->s_min_batch_time = result.uint_32; 2235 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2236 return 0; 2237 case Opt_inode_readahead_blks: 2238 if (result.uint_32 && 2239 (result.uint_32 > (1 << 30) || 2240 !is_power_of_2(result.uint_32))) { 2241 ext4_msg(NULL, KERN_ERR, 2242 "EXT4-fs: inode_readahead_blks must be " 2243 "0 or a power of 2 smaller than 2^31"); 2244 return -EINVAL; 2245 } 2246 ctx->s_inode_readahead_blks = result.uint_32; 2247 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2248 return 0; 2249 case Opt_init_itable: 2250 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2251 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2252 if (param->type == fs_value_is_string) 2253 ctx->s_li_wait_mult = result.uint_32; 2254 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2255 return 0; 2256 case Opt_max_dir_size_kb: 2257 ctx->s_max_dir_size_kb = result.uint_32; 2258 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2259 return 0; 2260 #ifdef CONFIG_EXT4_DEBUG 2261 case Opt_fc_debug_max_replay: 2262 ctx->s_fc_debug_max_replay = result.uint_32; 2263 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2264 return 0; 2265 #endif 2266 case Opt_stripe: 2267 ctx->s_stripe = result.uint_32; 2268 ctx->spec |= EXT4_SPEC_s_stripe; 2269 return 0; 2270 case Opt_resuid: 2271 ctx->s_resuid = result.uid; 2272 ctx->spec |= EXT4_SPEC_s_resuid; 2273 return 0; 2274 case Opt_resgid: 2275 ctx->s_resgid = result.gid; 2276 ctx->spec |= EXT4_SPEC_s_resgid; 2277 return 0; 2278 case Opt_journal_dev: 2279 if (is_remount) { 2280 ext4_msg(NULL, KERN_ERR, 2281 "Cannot specify journal on remount"); 2282 return -EINVAL; 2283 } 2284 ctx->journal_devnum = result.uint_32; 2285 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2286 return 0; 2287 case Opt_journal_path: 2288 { 2289 struct inode *journal_inode; 2290 struct path path; 2291 int error; 2292 2293 if (is_remount) { 2294 ext4_msg(NULL, KERN_ERR, 2295 "Cannot specify journal on remount"); 2296 return -EINVAL; 2297 } 2298 2299 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path); 2300 if (error) { 2301 ext4_msg(NULL, KERN_ERR, "error: could not find " 2302 "journal device path"); 2303 return -EINVAL; 2304 } 2305 2306 journal_inode = d_inode(path.dentry); 2307 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2308 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2309 path_put(&path); 2310 return 0; 2311 } 2312 case Opt_journal_ioprio: 2313 if (result.uint_32 > 7) { 2314 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2315 " (must be 0-7)"); 2316 return -EINVAL; 2317 } 2318 ctx->journal_ioprio = 2319 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2320 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2321 return 0; 2322 case Opt_test_dummy_encryption: 2323 return ext4_parse_test_dummy_encryption(param, ctx); 2324 case Opt_dax: 2325 case Opt_dax_type: 2326 #ifdef CONFIG_FS_DAX 2327 { 2328 int type = (token == Opt_dax) ? 2329 Opt_dax : result.uint_32; 2330 2331 switch (type) { 2332 case Opt_dax: 2333 case Opt_dax_always: 2334 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2335 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2336 break; 2337 case Opt_dax_never: 2338 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2339 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2340 break; 2341 case Opt_dax_inode: 2342 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2343 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2344 /* Strictly for printing options */ 2345 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2346 break; 2347 } 2348 return 0; 2349 } 2350 #else 2351 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2352 return -EINVAL; 2353 #endif 2354 case Opt_data_err: 2355 if (result.uint_32 == Opt_data_err_abort) 2356 ctx_set_mount_opt(ctx, m->mount_opt); 2357 else if (result.uint_32 == Opt_data_err_ignore) 2358 ctx_clear_mount_opt(ctx, m->mount_opt); 2359 return 0; 2360 case Opt_mb_optimize_scan: 2361 if (result.int_32 == 1) { 2362 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2363 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2364 } else if (result.int_32 == 0) { 2365 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2366 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2367 } else { 2368 ext4_msg(NULL, KERN_WARNING, 2369 "mb_optimize_scan should be set to 0 or 1."); 2370 return -EINVAL; 2371 } 2372 return 0; 2373 } 2374 2375 /* 2376 * At this point we should only be getting options requiring MOPT_SET, 2377 * or MOPT_CLEAR. Anything else is a bug 2378 */ 2379 if (m->token == Opt_err) { 2380 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2381 param->key); 2382 WARN_ON(1); 2383 return -EINVAL; 2384 } 2385 2386 else { 2387 unsigned int set = 0; 2388 2389 if ((param->type == fs_value_is_flag) || 2390 result.uint_32 > 0) 2391 set = 1; 2392 2393 if (m->flags & MOPT_CLEAR) 2394 set = !set; 2395 else if (unlikely(!(m->flags & MOPT_SET))) { 2396 ext4_msg(NULL, KERN_WARNING, 2397 "buggy handling of option %s", 2398 param->key); 2399 WARN_ON(1); 2400 return -EINVAL; 2401 } 2402 if (m->flags & MOPT_2) { 2403 if (set != 0) 2404 ctx_set_mount_opt2(ctx, m->mount_opt); 2405 else 2406 ctx_clear_mount_opt2(ctx, m->mount_opt); 2407 } else { 2408 if (set != 0) 2409 ctx_set_mount_opt(ctx, m->mount_opt); 2410 else 2411 ctx_clear_mount_opt(ctx, m->mount_opt); 2412 } 2413 } 2414 2415 return 0; 2416 } 2417 2418 static int parse_options(struct fs_context *fc, char *options) 2419 { 2420 struct fs_parameter param; 2421 int ret; 2422 char *key; 2423 2424 if (!options) 2425 return 0; 2426 2427 while ((key = strsep(&options, ",")) != NULL) { 2428 if (*key) { 2429 size_t v_len = 0; 2430 char *value = strchr(key, '='); 2431 2432 param.type = fs_value_is_flag; 2433 param.string = NULL; 2434 2435 if (value) { 2436 if (value == key) 2437 continue; 2438 2439 *value++ = 0; 2440 v_len = strlen(value); 2441 param.string = kmemdup_nul(value, v_len, 2442 GFP_KERNEL); 2443 if (!param.string) 2444 return -ENOMEM; 2445 param.type = fs_value_is_string; 2446 } 2447 2448 param.key = key; 2449 param.size = v_len; 2450 2451 ret = ext4_parse_param(fc, ¶m); 2452 kfree(param.string); 2453 if (ret < 0) 2454 return ret; 2455 } 2456 } 2457 2458 ret = ext4_validate_options(fc); 2459 if (ret < 0) 2460 return ret; 2461 2462 return 0; 2463 } 2464 2465 static int parse_apply_sb_mount_options(struct super_block *sb, 2466 struct ext4_fs_context *m_ctx) 2467 { 2468 struct ext4_sb_info *sbi = EXT4_SB(sb); 2469 char *s_mount_opts = NULL; 2470 struct ext4_fs_context *s_ctx = NULL; 2471 struct fs_context *fc = NULL; 2472 int ret = -ENOMEM; 2473 2474 if (!sbi->s_es->s_mount_opts[0]) 2475 return 0; 2476 2477 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2478 sizeof(sbi->s_es->s_mount_opts), 2479 GFP_KERNEL); 2480 if (!s_mount_opts) 2481 return ret; 2482 2483 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2484 if (!fc) 2485 goto out_free; 2486 2487 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2488 if (!s_ctx) 2489 goto out_free; 2490 2491 fc->fs_private = s_ctx; 2492 fc->s_fs_info = sbi; 2493 2494 ret = parse_options(fc, s_mount_opts); 2495 if (ret < 0) 2496 goto parse_failed; 2497 2498 ret = ext4_check_opt_consistency(fc, sb); 2499 if (ret < 0) { 2500 parse_failed: 2501 ext4_msg(sb, KERN_WARNING, 2502 "failed to parse options in superblock: %s", 2503 s_mount_opts); 2504 ret = 0; 2505 goto out_free; 2506 } 2507 2508 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2509 m_ctx->journal_devnum = s_ctx->journal_devnum; 2510 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2511 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2512 2513 ext4_apply_options(fc, sb); 2514 ret = 0; 2515 2516 out_free: 2517 if (fc) { 2518 ext4_fc_free(fc); 2519 kfree(fc); 2520 } 2521 kfree(s_mount_opts); 2522 return ret; 2523 } 2524 2525 static void ext4_apply_quota_options(struct fs_context *fc, 2526 struct super_block *sb) 2527 { 2528 #ifdef CONFIG_QUOTA 2529 bool quota_feature = ext4_has_feature_quota(sb); 2530 struct ext4_fs_context *ctx = fc->fs_private; 2531 struct ext4_sb_info *sbi = EXT4_SB(sb); 2532 char *qname; 2533 int i; 2534 2535 if (quota_feature) 2536 return; 2537 2538 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2539 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2540 if (!(ctx->qname_spec & (1 << i))) 2541 continue; 2542 2543 qname = ctx->s_qf_names[i]; /* May be NULL */ 2544 if (qname) 2545 set_opt(sb, QUOTA); 2546 ctx->s_qf_names[i] = NULL; 2547 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2548 lockdep_is_held(&sb->s_umount)); 2549 if (qname) 2550 kfree_rcu_mightsleep(qname); 2551 } 2552 } 2553 2554 if (ctx->spec & EXT4_SPEC_JQFMT) 2555 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2556 #endif 2557 } 2558 2559 /* 2560 * Check quota settings consistency. 2561 */ 2562 static int ext4_check_quota_consistency(struct fs_context *fc, 2563 struct super_block *sb) 2564 { 2565 #ifdef CONFIG_QUOTA 2566 struct ext4_fs_context *ctx = fc->fs_private; 2567 struct ext4_sb_info *sbi = EXT4_SB(sb); 2568 bool quota_feature = ext4_has_feature_quota(sb); 2569 bool quota_loaded = sb_any_quota_loaded(sb); 2570 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2571 int quota_flags, i; 2572 2573 /* 2574 * We do the test below only for project quotas. 'usrquota' and 2575 * 'grpquota' mount options are allowed even without quota feature 2576 * to support legacy quotas in quota files. 2577 */ 2578 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2579 !ext4_has_feature_project(sb)) { 2580 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2581 "Cannot enable project quota enforcement."); 2582 return -EINVAL; 2583 } 2584 2585 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2586 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2587 if (quota_loaded && 2588 ctx->mask_s_mount_opt & quota_flags && 2589 !ctx_test_mount_opt(ctx, quota_flags)) 2590 goto err_quota_change; 2591 2592 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2593 2594 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2595 if (!(ctx->qname_spec & (1 << i))) 2596 continue; 2597 2598 if (quota_loaded && 2599 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2600 goto err_jquota_change; 2601 2602 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2603 strcmp(get_qf_name(sb, sbi, i), 2604 ctx->s_qf_names[i]) != 0) 2605 goto err_jquota_specified; 2606 } 2607 2608 if (quota_feature) { 2609 ext4_msg(NULL, KERN_INFO, 2610 "Journaled quota options ignored when " 2611 "QUOTA feature is enabled"); 2612 return 0; 2613 } 2614 } 2615 2616 if (ctx->spec & EXT4_SPEC_JQFMT) { 2617 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2618 goto err_jquota_change; 2619 if (quota_feature) { 2620 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2621 "ignored when QUOTA feature is enabled"); 2622 return 0; 2623 } 2624 } 2625 2626 /* Make sure we don't mix old and new quota format */ 2627 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2628 ctx->s_qf_names[USRQUOTA]); 2629 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2630 ctx->s_qf_names[GRPQUOTA]); 2631 2632 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2633 test_opt(sb, USRQUOTA)); 2634 2635 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2636 test_opt(sb, GRPQUOTA)); 2637 2638 if (usr_qf_name) { 2639 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2640 usrquota = false; 2641 } 2642 if (grp_qf_name) { 2643 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2644 grpquota = false; 2645 } 2646 2647 if (usr_qf_name || grp_qf_name) { 2648 if (usrquota || grpquota) { 2649 ext4_msg(NULL, KERN_ERR, "old and new quota " 2650 "format mixing"); 2651 return -EINVAL; 2652 } 2653 2654 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2655 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2656 "not specified"); 2657 return -EINVAL; 2658 } 2659 } 2660 2661 return 0; 2662 2663 err_quota_change: 2664 ext4_msg(NULL, KERN_ERR, 2665 "Cannot change quota options when quota turned on"); 2666 return -EINVAL; 2667 err_jquota_change: 2668 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2669 "options when quota turned on"); 2670 return -EINVAL; 2671 err_jquota_specified: 2672 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2673 QTYPE2NAME(i)); 2674 return -EINVAL; 2675 #else 2676 return 0; 2677 #endif 2678 } 2679 2680 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2681 struct super_block *sb) 2682 { 2683 const struct ext4_fs_context *ctx = fc->fs_private; 2684 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2685 2686 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2687 return 0; 2688 2689 if (!ext4_has_feature_encrypt(sb)) { 2690 ext4_msg(NULL, KERN_WARNING, 2691 "test_dummy_encryption requires encrypt feature"); 2692 return -EINVAL; 2693 } 2694 /* 2695 * This mount option is just for testing, and it's not worthwhile to 2696 * implement the extra complexity (e.g. RCU protection) that would be 2697 * needed to allow it to be set or changed during remount. We do allow 2698 * it to be specified during remount, but only if there is no change. 2699 */ 2700 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2701 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2702 &ctx->dummy_enc_policy)) 2703 return 0; 2704 ext4_msg(NULL, KERN_WARNING, 2705 "Can't set or change test_dummy_encryption on remount"); 2706 return -EINVAL; 2707 } 2708 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2709 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2710 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2711 &ctx->dummy_enc_policy)) 2712 return 0; 2713 ext4_msg(NULL, KERN_WARNING, 2714 "Conflicting test_dummy_encryption options"); 2715 return -EINVAL; 2716 } 2717 return 0; 2718 } 2719 2720 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2721 struct super_block *sb) 2722 { 2723 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2724 /* if already set, it was already verified to be the same */ 2725 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2726 return; 2727 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2728 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2729 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2730 } 2731 2732 static int ext4_check_opt_consistency(struct fs_context *fc, 2733 struct super_block *sb) 2734 { 2735 struct ext4_fs_context *ctx = fc->fs_private; 2736 struct ext4_sb_info *sbi = fc->s_fs_info; 2737 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2738 int err; 2739 2740 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2741 ext4_msg(NULL, KERN_ERR, 2742 "Mount option(s) incompatible with ext2"); 2743 return -EINVAL; 2744 } 2745 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2746 ext4_msg(NULL, KERN_ERR, 2747 "Mount option(s) incompatible with ext3"); 2748 return -EINVAL; 2749 } 2750 2751 if (ctx->s_want_extra_isize > 2752 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2753 ext4_msg(NULL, KERN_ERR, 2754 "Invalid want_extra_isize %d", 2755 ctx->s_want_extra_isize); 2756 return -EINVAL; 2757 } 2758 2759 err = ext4_check_test_dummy_encryption(fc, sb); 2760 if (err) 2761 return err; 2762 2763 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2764 if (!sbi->s_journal) { 2765 ext4_msg(NULL, KERN_WARNING, 2766 "Remounting file system with no journal " 2767 "so ignoring journalled data option"); 2768 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2769 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2770 test_opt(sb, DATA_FLAGS)) { 2771 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2772 "on remount"); 2773 return -EINVAL; 2774 } 2775 } 2776 2777 if (is_remount) { 2778 if (!sbi->s_journal && 2779 ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT)) { 2780 ext4_msg(NULL, KERN_WARNING, 2781 "Remounting fs w/o journal so ignoring data_err option"); 2782 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT); 2783 } 2784 2785 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2786 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2787 ext4_msg(NULL, KERN_ERR, "can't mount with " 2788 "both data=journal and dax"); 2789 return -EINVAL; 2790 } 2791 2792 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2793 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2794 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2795 fail_dax_change_remount: 2796 ext4_msg(NULL, KERN_ERR, "can't change " 2797 "dax mount option while remounting"); 2798 return -EINVAL; 2799 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2800 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2801 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2802 goto fail_dax_change_remount; 2803 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2804 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2805 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2806 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2807 goto fail_dax_change_remount; 2808 } 2809 } 2810 2811 return ext4_check_quota_consistency(fc, sb); 2812 } 2813 2814 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2815 { 2816 struct ext4_fs_context *ctx = fc->fs_private; 2817 struct ext4_sb_info *sbi = fc->s_fs_info; 2818 2819 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2820 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2821 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2822 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2823 sb->s_flags &= ~ctx->mask_s_flags; 2824 sb->s_flags |= ctx->vals_s_flags; 2825 2826 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2827 APPLY(s_commit_interval); 2828 APPLY(s_stripe); 2829 APPLY(s_max_batch_time); 2830 APPLY(s_min_batch_time); 2831 APPLY(s_want_extra_isize); 2832 APPLY(s_inode_readahead_blks); 2833 APPLY(s_max_dir_size_kb); 2834 APPLY(s_li_wait_mult); 2835 APPLY(s_resgid); 2836 APPLY(s_resuid); 2837 2838 #ifdef CONFIG_EXT4_DEBUG 2839 APPLY(s_fc_debug_max_replay); 2840 #endif 2841 2842 ext4_apply_quota_options(fc, sb); 2843 ext4_apply_test_dummy_encryption(ctx, sb); 2844 } 2845 2846 2847 static int ext4_validate_options(struct fs_context *fc) 2848 { 2849 #ifdef CONFIG_QUOTA 2850 struct ext4_fs_context *ctx = fc->fs_private; 2851 char *usr_qf_name, *grp_qf_name; 2852 2853 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2854 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2855 2856 if (usr_qf_name || grp_qf_name) { 2857 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2858 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2859 2860 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2861 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2862 2863 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2864 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2865 ext4_msg(NULL, KERN_ERR, "old and new quota " 2866 "format mixing"); 2867 return -EINVAL; 2868 } 2869 } 2870 #endif 2871 return 1; 2872 } 2873 2874 static inline void ext4_show_quota_options(struct seq_file *seq, 2875 struct super_block *sb) 2876 { 2877 #if defined(CONFIG_QUOTA) 2878 struct ext4_sb_info *sbi = EXT4_SB(sb); 2879 char *usr_qf_name, *grp_qf_name; 2880 2881 if (sbi->s_jquota_fmt) { 2882 char *fmtname = ""; 2883 2884 switch (sbi->s_jquota_fmt) { 2885 case QFMT_VFS_OLD: 2886 fmtname = "vfsold"; 2887 break; 2888 case QFMT_VFS_V0: 2889 fmtname = "vfsv0"; 2890 break; 2891 case QFMT_VFS_V1: 2892 fmtname = "vfsv1"; 2893 break; 2894 } 2895 seq_printf(seq, ",jqfmt=%s", fmtname); 2896 } 2897 2898 rcu_read_lock(); 2899 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2900 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2901 if (usr_qf_name) 2902 seq_show_option(seq, "usrjquota", usr_qf_name); 2903 if (grp_qf_name) 2904 seq_show_option(seq, "grpjquota", grp_qf_name); 2905 rcu_read_unlock(); 2906 #endif 2907 } 2908 2909 static const char *token2str(int token) 2910 { 2911 const struct fs_parameter_spec *spec; 2912 2913 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2914 if (spec->opt == token && !spec->type) 2915 break; 2916 return spec->name; 2917 } 2918 2919 /* 2920 * Show an option if 2921 * - it's set to a non-default value OR 2922 * - if the per-sb default is different from the global default 2923 */ 2924 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2925 int nodefs) 2926 { 2927 struct ext4_sb_info *sbi = EXT4_SB(sb); 2928 struct ext4_super_block *es = sbi->s_es; 2929 int def_errors; 2930 const struct mount_opts *m; 2931 char sep = nodefs ? '\n' : ','; 2932 2933 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2934 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2935 2936 if (sbi->s_sb_block != 1) 2937 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2938 2939 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2940 int want_set = m->flags & MOPT_SET; 2941 int opt_2 = m->flags & MOPT_2; 2942 unsigned int mount_opt, def_mount_opt; 2943 2944 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2945 m->flags & MOPT_SKIP) 2946 continue; 2947 2948 if (opt_2) { 2949 mount_opt = sbi->s_mount_opt2; 2950 def_mount_opt = sbi->s_def_mount_opt2; 2951 } else { 2952 mount_opt = sbi->s_mount_opt; 2953 def_mount_opt = sbi->s_def_mount_opt; 2954 } 2955 /* skip if same as the default */ 2956 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt))) 2957 continue; 2958 /* select Opt_noFoo vs Opt_Foo */ 2959 if ((want_set && 2960 (mount_opt & m->mount_opt) != m->mount_opt) || 2961 (!want_set && (mount_opt & m->mount_opt))) 2962 continue; 2963 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2964 } 2965 2966 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2967 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2968 SEQ_OPTS_PRINT("resuid=%u", 2969 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2970 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2971 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2972 SEQ_OPTS_PRINT("resgid=%u", 2973 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2974 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2975 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2976 SEQ_OPTS_PUTS("errors=remount-ro"); 2977 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2978 SEQ_OPTS_PUTS("errors=continue"); 2979 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2980 SEQ_OPTS_PUTS("errors=panic"); 2981 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2982 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2983 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2984 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2985 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2986 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2987 if (nodefs && sb->s_flags & SB_I_VERSION) 2988 SEQ_OPTS_PUTS("i_version"); 2989 if (nodefs || sbi->s_stripe) 2990 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2991 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2992 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 2993 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2994 SEQ_OPTS_PUTS("data=journal"); 2995 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2996 SEQ_OPTS_PUTS("data=ordered"); 2997 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2998 SEQ_OPTS_PUTS("data=writeback"); 2999 } 3000 if (nodefs || 3001 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 3002 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 3003 sbi->s_inode_readahead_blks); 3004 3005 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 3006 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 3007 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 3008 if (nodefs || sbi->s_max_dir_size_kb) 3009 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 3010 if (test_opt(sb, DATA_ERR_ABORT)) 3011 SEQ_OPTS_PUTS("data_err=abort"); 3012 3013 fscrypt_show_test_dummy_encryption(seq, sep, sb); 3014 3015 if (sb->s_flags & SB_INLINECRYPT) 3016 SEQ_OPTS_PUTS("inlinecrypt"); 3017 3018 if (test_opt(sb, DAX_ALWAYS)) { 3019 if (IS_EXT2_SB(sb)) 3020 SEQ_OPTS_PUTS("dax"); 3021 else 3022 SEQ_OPTS_PUTS("dax=always"); 3023 } else if (test_opt2(sb, DAX_NEVER)) { 3024 SEQ_OPTS_PUTS("dax=never"); 3025 } else if (test_opt2(sb, DAX_INODE)) { 3026 SEQ_OPTS_PUTS("dax=inode"); 3027 } 3028 3029 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3030 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3031 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 3032 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3033 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3034 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 3035 } 3036 3037 if (nodefs && !test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) 3038 SEQ_OPTS_PUTS("prefetch_block_bitmaps"); 3039 3040 if (ext4_emergency_ro(sb)) 3041 SEQ_OPTS_PUTS("emergency_ro"); 3042 3043 if (ext4_forced_shutdown(sb)) 3044 SEQ_OPTS_PUTS("shutdown"); 3045 3046 ext4_show_quota_options(seq, sb); 3047 return 0; 3048 } 3049 3050 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 3051 { 3052 return _ext4_show_options(seq, root->d_sb, 0); 3053 } 3054 3055 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3056 { 3057 struct super_block *sb = seq->private; 3058 int rc; 3059 3060 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3061 rc = _ext4_show_options(seq, sb, 1); 3062 seq_putc(seq, '\n'); 3063 return rc; 3064 } 3065 3066 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3067 int read_only) 3068 { 3069 struct ext4_sb_info *sbi = EXT4_SB(sb); 3070 int err = 0; 3071 3072 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3073 ext4_msg(sb, KERN_ERR, "revision level too high, " 3074 "forcing read-only mode"); 3075 err = -EROFS; 3076 goto done; 3077 } 3078 if (read_only) 3079 goto done; 3080 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3081 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3082 "running e2fsck is recommended"); 3083 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3084 ext4_msg(sb, KERN_WARNING, 3085 "warning: mounting fs with errors, " 3086 "running e2fsck is recommended"); 3087 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3088 le16_to_cpu(es->s_mnt_count) >= 3089 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3090 ext4_msg(sb, KERN_WARNING, 3091 "warning: maximal mount count reached, " 3092 "running e2fsck is recommended"); 3093 else if (le32_to_cpu(es->s_checkinterval) && 3094 (ext4_get_tstamp(es, s_lastcheck) + 3095 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3096 ext4_msg(sb, KERN_WARNING, 3097 "warning: checktime reached, " 3098 "running e2fsck is recommended"); 3099 if (!sbi->s_journal) 3100 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3101 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3102 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3103 le16_add_cpu(&es->s_mnt_count, 1); 3104 ext4_update_tstamp(es, s_mtime); 3105 if (sbi->s_journal) { 3106 ext4_set_feature_journal_needs_recovery(sb); 3107 if (ext4_has_feature_orphan_file(sb)) 3108 ext4_set_feature_orphan_present(sb); 3109 } 3110 3111 err = ext4_commit_super(sb); 3112 done: 3113 if (test_opt(sb, DEBUG)) 3114 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3115 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3116 sb->s_blocksize, 3117 sbi->s_groups_count, 3118 EXT4_BLOCKS_PER_GROUP(sb), 3119 EXT4_INODES_PER_GROUP(sb), 3120 sbi->s_mount_opt, sbi->s_mount_opt2); 3121 return err; 3122 } 3123 3124 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3125 { 3126 struct ext4_sb_info *sbi = EXT4_SB(sb); 3127 struct flex_groups **old_groups, **new_groups; 3128 int size, i, j; 3129 3130 if (!sbi->s_log_groups_per_flex) 3131 return 0; 3132 3133 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3134 if (size <= sbi->s_flex_groups_allocated) 3135 return 0; 3136 3137 new_groups = kvzalloc(roundup_pow_of_two(size * 3138 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3139 if (!new_groups) { 3140 ext4_msg(sb, KERN_ERR, 3141 "not enough memory for %d flex group pointers", size); 3142 return -ENOMEM; 3143 } 3144 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3145 new_groups[i] = kvzalloc(roundup_pow_of_two( 3146 sizeof(struct flex_groups)), 3147 GFP_KERNEL); 3148 if (!new_groups[i]) { 3149 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3150 kvfree(new_groups[j]); 3151 kvfree(new_groups); 3152 ext4_msg(sb, KERN_ERR, 3153 "not enough memory for %d flex groups", size); 3154 return -ENOMEM; 3155 } 3156 } 3157 rcu_read_lock(); 3158 old_groups = rcu_dereference(sbi->s_flex_groups); 3159 if (old_groups) 3160 memcpy(new_groups, old_groups, 3161 (sbi->s_flex_groups_allocated * 3162 sizeof(struct flex_groups *))); 3163 rcu_read_unlock(); 3164 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3165 sbi->s_flex_groups_allocated = size; 3166 if (old_groups) 3167 ext4_kvfree_array_rcu(old_groups); 3168 return 0; 3169 } 3170 3171 static int ext4_fill_flex_info(struct super_block *sb) 3172 { 3173 struct ext4_sb_info *sbi = EXT4_SB(sb); 3174 struct ext4_group_desc *gdp = NULL; 3175 struct flex_groups *fg; 3176 ext4_group_t flex_group; 3177 int i, err; 3178 3179 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3180 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3181 sbi->s_log_groups_per_flex = 0; 3182 return 1; 3183 } 3184 3185 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3186 if (err) 3187 goto failed; 3188 3189 for (i = 0; i < sbi->s_groups_count; i++) { 3190 gdp = ext4_get_group_desc(sb, i, NULL); 3191 3192 flex_group = ext4_flex_group(sbi, i); 3193 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3194 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3195 atomic64_add(ext4_free_group_clusters(sb, gdp), 3196 &fg->free_clusters); 3197 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3198 } 3199 3200 return 1; 3201 failed: 3202 return 0; 3203 } 3204 3205 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3206 struct ext4_group_desc *gdp) 3207 { 3208 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3209 __u16 crc = 0; 3210 __le32 le_group = cpu_to_le32(block_group); 3211 struct ext4_sb_info *sbi = EXT4_SB(sb); 3212 3213 if (ext4_has_feature_metadata_csum(sbi->s_sb)) { 3214 /* Use new metadata_csum algorithm */ 3215 __u32 csum32; 3216 __u16 dummy_csum = 0; 3217 3218 csum32 = ext4_chksum(sbi->s_csum_seed, (__u8 *)&le_group, 3219 sizeof(le_group)); 3220 csum32 = ext4_chksum(csum32, (__u8 *)gdp, offset); 3221 csum32 = ext4_chksum(csum32, (__u8 *)&dummy_csum, 3222 sizeof(dummy_csum)); 3223 offset += sizeof(dummy_csum); 3224 if (offset < sbi->s_desc_size) 3225 csum32 = ext4_chksum(csum32, (__u8 *)gdp + offset, 3226 sbi->s_desc_size - offset); 3227 3228 crc = csum32 & 0xFFFF; 3229 goto out; 3230 } 3231 3232 /* old crc16 code */ 3233 if (!ext4_has_feature_gdt_csum(sb)) 3234 return 0; 3235 3236 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3237 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3238 crc = crc16(crc, (__u8 *)gdp, offset); 3239 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3240 /* for checksum of struct ext4_group_desc do the rest...*/ 3241 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size) 3242 crc = crc16(crc, (__u8 *)gdp + offset, 3243 sbi->s_desc_size - offset); 3244 3245 out: 3246 return cpu_to_le16(crc); 3247 } 3248 3249 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3250 struct ext4_group_desc *gdp) 3251 { 3252 if (ext4_has_group_desc_csum(sb) && 3253 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3254 return 0; 3255 3256 return 1; 3257 } 3258 3259 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3260 struct ext4_group_desc *gdp) 3261 { 3262 if (!ext4_has_group_desc_csum(sb)) 3263 return; 3264 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3265 } 3266 3267 /* Called at mount-time, super-block is locked */ 3268 static int ext4_check_descriptors(struct super_block *sb, 3269 ext4_fsblk_t sb_block, 3270 ext4_group_t *first_not_zeroed) 3271 { 3272 struct ext4_sb_info *sbi = EXT4_SB(sb); 3273 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3274 ext4_fsblk_t last_block; 3275 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3276 ext4_fsblk_t block_bitmap; 3277 ext4_fsblk_t inode_bitmap; 3278 ext4_fsblk_t inode_table; 3279 int flexbg_flag = 0; 3280 ext4_group_t i, grp = sbi->s_groups_count; 3281 3282 if (ext4_has_feature_flex_bg(sb)) 3283 flexbg_flag = 1; 3284 3285 ext4_debug("Checking group descriptors"); 3286 3287 for (i = 0; i < sbi->s_groups_count; i++) { 3288 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3289 3290 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3291 last_block = ext4_blocks_count(sbi->s_es) - 1; 3292 else 3293 last_block = first_block + 3294 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3295 3296 if ((grp == sbi->s_groups_count) && 3297 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3298 grp = i; 3299 3300 block_bitmap = ext4_block_bitmap(sb, gdp); 3301 if (block_bitmap == sb_block) { 3302 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3303 "Block bitmap for group %u overlaps " 3304 "superblock", i); 3305 if (!sb_rdonly(sb)) 3306 return 0; 3307 } 3308 if (block_bitmap >= sb_block + 1 && 3309 block_bitmap <= last_bg_block) { 3310 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3311 "Block bitmap for group %u overlaps " 3312 "block group descriptors", i); 3313 if (!sb_rdonly(sb)) 3314 return 0; 3315 } 3316 if (block_bitmap < first_block || block_bitmap > last_block) { 3317 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3318 "Block bitmap for group %u not in group " 3319 "(block %llu)!", i, block_bitmap); 3320 return 0; 3321 } 3322 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3323 if (inode_bitmap == sb_block) { 3324 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3325 "Inode bitmap for group %u overlaps " 3326 "superblock", i); 3327 if (!sb_rdonly(sb)) 3328 return 0; 3329 } 3330 if (inode_bitmap >= sb_block + 1 && 3331 inode_bitmap <= last_bg_block) { 3332 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3333 "Inode bitmap for group %u overlaps " 3334 "block group descriptors", i); 3335 if (!sb_rdonly(sb)) 3336 return 0; 3337 } 3338 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3339 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3340 "Inode bitmap for group %u not in group " 3341 "(block %llu)!", i, inode_bitmap); 3342 return 0; 3343 } 3344 inode_table = ext4_inode_table(sb, gdp); 3345 if (inode_table == sb_block) { 3346 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3347 "Inode table for group %u overlaps " 3348 "superblock", i); 3349 if (!sb_rdonly(sb)) 3350 return 0; 3351 } 3352 if (inode_table >= sb_block + 1 && 3353 inode_table <= last_bg_block) { 3354 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3355 "Inode table for group %u overlaps " 3356 "block group descriptors", i); 3357 if (!sb_rdonly(sb)) 3358 return 0; 3359 } 3360 if (inode_table < first_block || 3361 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3362 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3363 "Inode table for group %u not in group " 3364 "(block %llu)!", i, inode_table); 3365 return 0; 3366 } 3367 ext4_lock_group(sb, i); 3368 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3369 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3370 "Checksum for group %u failed (%u!=%u)", 3371 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3372 gdp)), le16_to_cpu(gdp->bg_checksum)); 3373 if (!sb_rdonly(sb)) { 3374 ext4_unlock_group(sb, i); 3375 return 0; 3376 } 3377 } 3378 ext4_unlock_group(sb, i); 3379 if (!flexbg_flag) 3380 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3381 } 3382 if (NULL != first_not_zeroed) 3383 *first_not_zeroed = grp; 3384 return 1; 3385 } 3386 3387 /* 3388 * Maximal extent format file size. 3389 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3390 * extent format containers, within a sector_t, and within i_blocks 3391 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3392 * so that won't be a limiting factor. 3393 * 3394 * However there is other limiting factor. We do store extents in the form 3395 * of starting block and length, hence the resulting length of the extent 3396 * covering maximum file size must fit into on-disk format containers as 3397 * well. Given that length is always by 1 unit bigger than max unit (because 3398 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3399 * 3400 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3401 */ 3402 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3403 { 3404 loff_t res; 3405 loff_t upper_limit = MAX_LFS_FILESIZE; 3406 3407 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3408 3409 if (!has_huge_files) { 3410 upper_limit = (1LL << 32) - 1; 3411 3412 /* total blocks in file system block size */ 3413 upper_limit >>= (blkbits - 9); 3414 upper_limit <<= blkbits; 3415 } 3416 3417 /* 3418 * 32-bit extent-start container, ee_block. We lower the maxbytes 3419 * by one fs block, so ee_len can cover the extent of maximum file 3420 * size 3421 */ 3422 res = (1LL << 32) - 1; 3423 res <<= blkbits; 3424 3425 /* Sanity check against vm- & vfs- imposed limits */ 3426 if (res > upper_limit) 3427 res = upper_limit; 3428 3429 return res; 3430 } 3431 3432 /* 3433 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3434 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3435 * We need to be 1 filesystem block less than the 2^48 sector limit. 3436 */ 3437 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3438 { 3439 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3440 int meta_blocks; 3441 unsigned int ppb = 1 << (bits - 2); 3442 3443 /* 3444 * This is calculated to be the largest file size for a dense, block 3445 * mapped file such that the file's total number of 512-byte sectors, 3446 * including data and all indirect blocks, does not exceed (2^48 - 1). 3447 * 3448 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3449 * number of 512-byte sectors of the file. 3450 */ 3451 if (!has_huge_files) { 3452 /* 3453 * !has_huge_files or implies that the inode i_block field 3454 * represents total file blocks in 2^32 512-byte sectors == 3455 * size of vfs inode i_blocks * 8 3456 */ 3457 upper_limit = (1LL << 32) - 1; 3458 3459 /* total blocks in file system block size */ 3460 upper_limit >>= (bits - 9); 3461 3462 } else { 3463 /* 3464 * We use 48 bit ext4_inode i_blocks 3465 * With EXT4_HUGE_FILE_FL set the i_blocks 3466 * represent total number of blocks in 3467 * file system block size 3468 */ 3469 upper_limit = (1LL << 48) - 1; 3470 3471 } 3472 3473 /* Compute how many blocks we can address by block tree */ 3474 res += ppb; 3475 res += ppb * ppb; 3476 res += ((loff_t)ppb) * ppb * ppb; 3477 /* Compute how many metadata blocks are needed */ 3478 meta_blocks = 1; 3479 meta_blocks += 1 + ppb; 3480 meta_blocks += 1 + ppb + ppb * ppb; 3481 /* Does block tree limit file size? */ 3482 if (res + meta_blocks <= upper_limit) 3483 goto check_lfs; 3484 3485 res = upper_limit; 3486 /* How many metadata blocks are needed for addressing upper_limit? */ 3487 upper_limit -= EXT4_NDIR_BLOCKS; 3488 /* indirect blocks */ 3489 meta_blocks = 1; 3490 upper_limit -= ppb; 3491 /* double indirect blocks */ 3492 if (upper_limit < ppb * ppb) { 3493 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3494 res -= meta_blocks; 3495 goto check_lfs; 3496 } 3497 meta_blocks += 1 + ppb; 3498 upper_limit -= ppb * ppb; 3499 /* tripple indirect blocks for the rest */ 3500 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3501 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3502 res -= meta_blocks; 3503 check_lfs: 3504 res <<= bits; 3505 if (res > MAX_LFS_FILESIZE) 3506 res = MAX_LFS_FILESIZE; 3507 3508 return res; 3509 } 3510 3511 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3512 ext4_fsblk_t logical_sb_block, int nr) 3513 { 3514 struct ext4_sb_info *sbi = EXT4_SB(sb); 3515 ext4_group_t bg, first_meta_bg; 3516 int has_super = 0; 3517 3518 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3519 3520 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3521 return logical_sb_block + nr + 1; 3522 bg = sbi->s_desc_per_block * nr; 3523 if (ext4_bg_has_super(sb, bg)) 3524 has_super = 1; 3525 3526 /* 3527 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3528 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3529 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3530 * compensate. 3531 */ 3532 if (sb->s_blocksize == 1024 && nr == 0 && 3533 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3534 has_super++; 3535 3536 return (has_super + ext4_group_first_block_no(sb, bg)); 3537 } 3538 3539 /** 3540 * ext4_get_stripe_size: Get the stripe size. 3541 * @sbi: In memory super block info 3542 * 3543 * If we have specified it via mount option, then 3544 * use the mount option value. If the value specified at mount time is 3545 * greater than the blocks per group use the super block value. 3546 * If the super block value is greater than blocks per group return 0. 3547 * Allocator needs it be less than blocks per group. 3548 * 3549 */ 3550 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3551 { 3552 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3553 unsigned long stripe_width = 3554 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3555 int ret; 3556 3557 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3558 ret = sbi->s_stripe; 3559 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3560 ret = stripe_width; 3561 else if (stride && stride <= sbi->s_blocks_per_group) 3562 ret = stride; 3563 else 3564 ret = 0; 3565 3566 /* 3567 * If the stripe width is 1, this makes no sense and 3568 * we set it to 0 to turn off stripe handling code. 3569 */ 3570 if (ret <= 1) 3571 ret = 0; 3572 3573 return ret; 3574 } 3575 3576 /* 3577 * Check whether this filesystem can be mounted based on 3578 * the features present and the RDONLY/RDWR mount requested. 3579 * Returns 1 if this filesystem can be mounted as requested, 3580 * 0 if it cannot be. 3581 */ 3582 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3583 { 3584 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3585 ext4_msg(sb, KERN_ERR, 3586 "Couldn't mount because of " 3587 "unsupported optional features (%x)", 3588 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3589 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3590 return 0; 3591 } 3592 3593 if (!IS_ENABLED(CONFIG_UNICODE) && ext4_has_feature_casefold(sb)) { 3594 ext4_msg(sb, KERN_ERR, 3595 "Filesystem with casefold feature cannot be " 3596 "mounted without CONFIG_UNICODE"); 3597 return 0; 3598 } 3599 3600 if (readonly) 3601 return 1; 3602 3603 if (ext4_has_feature_readonly(sb)) { 3604 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3605 sb->s_flags |= SB_RDONLY; 3606 return 1; 3607 } 3608 3609 /* Check that feature set is OK for a read-write mount */ 3610 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3611 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3612 "unsupported optional features (%x)", 3613 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3614 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3615 return 0; 3616 } 3617 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3618 ext4_msg(sb, KERN_ERR, 3619 "Can't support bigalloc feature without " 3620 "extents feature\n"); 3621 return 0; 3622 } 3623 3624 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3625 if (!readonly && (ext4_has_feature_quota(sb) || 3626 ext4_has_feature_project(sb))) { 3627 ext4_msg(sb, KERN_ERR, 3628 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3629 return 0; 3630 } 3631 #endif /* CONFIG_QUOTA */ 3632 return 1; 3633 } 3634 3635 /* 3636 * This function is called once a day if we have errors logged 3637 * on the file system 3638 */ 3639 static void print_daily_error_info(struct timer_list *t) 3640 { 3641 struct ext4_sb_info *sbi = timer_container_of(sbi, t, s_err_report); 3642 struct super_block *sb = sbi->s_sb; 3643 struct ext4_super_block *es = sbi->s_es; 3644 3645 if (es->s_error_count) 3646 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3647 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3648 le32_to_cpu(es->s_error_count)); 3649 if (es->s_first_error_time) { 3650 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3651 sb->s_id, 3652 ext4_get_tstamp(es, s_first_error_time), 3653 (int) sizeof(es->s_first_error_func), 3654 es->s_first_error_func, 3655 le32_to_cpu(es->s_first_error_line)); 3656 if (es->s_first_error_ino) 3657 printk(KERN_CONT ": inode %u", 3658 le32_to_cpu(es->s_first_error_ino)); 3659 if (es->s_first_error_block) 3660 printk(KERN_CONT ": block %llu", (unsigned long long) 3661 le64_to_cpu(es->s_first_error_block)); 3662 printk(KERN_CONT "\n"); 3663 } 3664 if (es->s_last_error_time) { 3665 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3666 sb->s_id, 3667 ext4_get_tstamp(es, s_last_error_time), 3668 (int) sizeof(es->s_last_error_func), 3669 es->s_last_error_func, 3670 le32_to_cpu(es->s_last_error_line)); 3671 if (es->s_last_error_ino) 3672 printk(KERN_CONT ": inode %u", 3673 le32_to_cpu(es->s_last_error_ino)); 3674 if (es->s_last_error_block) 3675 printk(KERN_CONT ": block %llu", (unsigned long long) 3676 le64_to_cpu(es->s_last_error_block)); 3677 printk(KERN_CONT "\n"); 3678 } 3679 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3680 } 3681 3682 /* Find next suitable group and run ext4_init_inode_table */ 3683 static int ext4_run_li_request(struct ext4_li_request *elr) 3684 { 3685 struct ext4_group_desc *gdp = NULL; 3686 struct super_block *sb = elr->lr_super; 3687 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3688 ext4_group_t group = elr->lr_next_group; 3689 unsigned int prefetch_ios = 0; 3690 int ret = 0; 3691 int nr = EXT4_SB(sb)->s_mb_prefetch; 3692 u64 start_time; 3693 3694 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3695 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios); 3696 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr); 3697 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr); 3698 if (group >= elr->lr_next_group) { 3699 ret = 1; 3700 if (elr->lr_first_not_zeroed != ngroups && 3701 !ext4_emergency_state(sb) && !sb_rdonly(sb) && 3702 test_opt(sb, INIT_INODE_TABLE)) { 3703 elr->lr_next_group = elr->lr_first_not_zeroed; 3704 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3705 ret = 0; 3706 } 3707 } 3708 return ret; 3709 } 3710 3711 for (; group < ngroups; group++) { 3712 gdp = ext4_get_group_desc(sb, group, NULL); 3713 if (!gdp) { 3714 ret = 1; 3715 break; 3716 } 3717 3718 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3719 break; 3720 } 3721 3722 if (group >= ngroups) 3723 ret = 1; 3724 3725 if (!ret) { 3726 start_time = ktime_get_ns(); 3727 ret = ext4_init_inode_table(sb, group, 3728 elr->lr_timeout ? 0 : 1); 3729 trace_ext4_lazy_itable_init(sb, group); 3730 if (elr->lr_timeout == 0) { 3731 elr->lr_timeout = nsecs_to_jiffies((ktime_get_ns() - start_time) * 3732 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3733 } 3734 elr->lr_next_sched = jiffies + elr->lr_timeout; 3735 elr->lr_next_group = group + 1; 3736 } 3737 return ret; 3738 } 3739 3740 /* 3741 * Remove lr_request from the list_request and free the 3742 * request structure. Should be called with li_list_mtx held 3743 */ 3744 static void ext4_remove_li_request(struct ext4_li_request *elr) 3745 { 3746 if (!elr) 3747 return; 3748 3749 list_del(&elr->lr_request); 3750 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3751 kfree(elr); 3752 } 3753 3754 static void ext4_unregister_li_request(struct super_block *sb) 3755 { 3756 mutex_lock(&ext4_li_mtx); 3757 if (!ext4_li_info) { 3758 mutex_unlock(&ext4_li_mtx); 3759 return; 3760 } 3761 3762 mutex_lock(&ext4_li_info->li_list_mtx); 3763 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3764 mutex_unlock(&ext4_li_info->li_list_mtx); 3765 mutex_unlock(&ext4_li_mtx); 3766 } 3767 3768 static struct task_struct *ext4_lazyinit_task; 3769 3770 /* 3771 * This is the function where ext4lazyinit thread lives. It walks 3772 * through the request list searching for next scheduled filesystem. 3773 * When such a fs is found, run the lazy initialization request 3774 * (ext4_rn_li_request) and keep track of the time spend in this 3775 * function. Based on that time we compute next schedule time of 3776 * the request. When walking through the list is complete, compute 3777 * next waking time and put itself into sleep. 3778 */ 3779 static int ext4_lazyinit_thread(void *arg) 3780 { 3781 struct ext4_lazy_init *eli = arg; 3782 struct list_head *pos, *n; 3783 struct ext4_li_request *elr; 3784 unsigned long next_wakeup, cur; 3785 3786 BUG_ON(NULL == eli); 3787 set_freezable(); 3788 3789 cont_thread: 3790 while (true) { 3791 bool next_wakeup_initialized = false; 3792 3793 next_wakeup = 0; 3794 mutex_lock(&eli->li_list_mtx); 3795 if (list_empty(&eli->li_request_list)) { 3796 mutex_unlock(&eli->li_list_mtx); 3797 goto exit_thread; 3798 } 3799 list_for_each_safe(pos, n, &eli->li_request_list) { 3800 int err = 0; 3801 int progress = 0; 3802 elr = list_entry(pos, struct ext4_li_request, 3803 lr_request); 3804 3805 if (time_before(jiffies, elr->lr_next_sched)) { 3806 if (!next_wakeup_initialized || 3807 time_before(elr->lr_next_sched, next_wakeup)) { 3808 next_wakeup = elr->lr_next_sched; 3809 next_wakeup_initialized = true; 3810 } 3811 continue; 3812 } 3813 if (down_read_trylock(&elr->lr_super->s_umount)) { 3814 if (sb_start_write_trylock(elr->lr_super)) { 3815 progress = 1; 3816 /* 3817 * We hold sb->s_umount, sb can not 3818 * be removed from the list, it is 3819 * now safe to drop li_list_mtx 3820 */ 3821 mutex_unlock(&eli->li_list_mtx); 3822 err = ext4_run_li_request(elr); 3823 sb_end_write(elr->lr_super); 3824 mutex_lock(&eli->li_list_mtx); 3825 n = pos->next; 3826 } 3827 up_read((&elr->lr_super->s_umount)); 3828 } 3829 /* error, remove the lazy_init job */ 3830 if (err) { 3831 ext4_remove_li_request(elr); 3832 continue; 3833 } 3834 if (!progress) { 3835 elr->lr_next_sched = jiffies + 3836 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3837 } 3838 if (!next_wakeup_initialized || 3839 time_before(elr->lr_next_sched, next_wakeup)) { 3840 next_wakeup = elr->lr_next_sched; 3841 next_wakeup_initialized = true; 3842 } 3843 } 3844 mutex_unlock(&eli->li_list_mtx); 3845 3846 try_to_freeze(); 3847 3848 cur = jiffies; 3849 if (!next_wakeup_initialized || time_after_eq(cur, next_wakeup)) { 3850 cond_resched(); 3851 continue; 3852 } 3853 3854 schedule_timeout_interruptible(next_wakeup - cur); 3855 3856 if (kthread_should_stop()) { 3857 ext4_clear_request_list(); 3858 goto exit_thread; 3859 } 3860 } 3861 3862 exit_thread: 3863 /* 3864 * It looks like the request list is empty, but we need 3865 * to check it under the li_list_mtx lock, to prevent any 3866 * additions into it, and of course we should lock ext4_li_mtx 3867 * to atomically free the list and ext4_li_info, because at 3868 * this point another ext4 filesystem could be registering 3869 * new one. 3870 */ 3871 mutex_lock(&ext4_li_mtx); 3872 mutex_lock(&eli->li_list_mtx); 3873 if (!list_empty(&eli->li_request_list)) { 3874 mutex_unlock(&eli->li_list_mtx); 3875 mutex_unlock(&ext4_li_mtx); 3876 goto cont_thread; 3877 } 3878 mutex_unlock(&eli->li_list_mtx); 3879 kfree(ext4_li_info); 3880 ext4_li_info = NULL; 3881 mutex_unlock(&ext4_li_mtx); 3882 3883 return 0; 3884 } 3885 3886 static void ext4_clear_request_list(void) 3887 { 3888 struct list_head *pos, *n; 3889 struct ext4_li_request *elr; 3890 3891 mutex_lock(&ext4_li_info->li_list_mtx); 3892 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3893 elr = list_entry(pos, struct ext4_li_request, 3894 lr_request); 3895 ext4_remove_li_request(elr); 3896 } 3897 mutex_unlock(&ext4_li_info->li_list_mtx); 3898 } 3899 3900 static int ext4_run_lazyinit_thread(void) 3901 { 3902 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3903 ext4_li_info, "ext4lazyinit"); 3904 if (IS_ERR(ext4_lazyinit_task)) { 3905 int err = PTR_ERR(ext4_lazyinit_task); 3906 ext4_clear_request_list(); 3907 kfree(ext4_li_info); 3908 ext4_li_info = NULL; 3909 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3910 "initialization thread\n", 3911 err); 3912 return err; 3913 } 3914 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3915 return 0; 3916 } 3917 3918 /* 3919 * Check whether it make sense to run itable init. thread or not. 3920 * If there is at least one uninitialized inode table, return 3921 * corresponding group number, else the loop goes through all 3922 * groups and return total number of groups. 3923 */ 3924 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3925 { 3926 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3927 struct ext4_group_desc *gdp = NULL; 3928 3929 if (!ext4_has_group_desc_csum(sb)) 3930 return ngroups; 3931 3932 for (group = 0; group < ngroups; group++) { 3933 gdp = ext4_get_group_desc(sb, group, NULL); 3934 if (!gdp) 3935 continue; 3936 3937 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3938 break; 3939 } 3940 3941 return group; 3942 } 3943 3944 static int ext4_li_info_new(void) 3945 { 3946 struct ext4_lazy_init *eli = NULL; 3947 3948 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3949 if (!eli) 3950 return -ENOMEM; 3951 3952 INIT_LIST_HEAD(&eli->li_request_list); 3953 mutex_init(&eli->li_list_mtx); 3954 3955 eli->li_state |= EXT4_LAZYINIT_QUIT; 3956 3957 ext4_li_info = eli; 3958 3959 return 0; 3960 } 3961 3962 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3963 ext4_group_t start) 3964 { 3965 struct ext4_li_request *elr; 3966 3967 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3968 if (!elr) 3969 return NULL; 3970 3971 elr->lr_super = sb; 3972 elr->lr_first_not_zeroed = start; 3973 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3974 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3975 elr->lr_next_group = start; 3976 } else { 3977 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3978 } 3979 3980 /* 3981 * Randomize first schedule time of the request to 3982 * spread the inode table initialization requests 3983 * better. 3984 */ 3985 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3986 return elr; 3987 } 3988 3989 int ext4_register_li_request(struct super_block *sb, 3990 ext4_group_t first_not_zeroed) 3991 { 3992 struct ext4_sb_info *sbi = EXT4_SB(sb); 3993 struct ext4_li_request *elr = NULL; 3994 ext4_group_t ngroups = sbi->s_groups_count; 3995 int ret = 0; 3996 3997 mutex_lock(&ext4_li_mtx); 3998 if (sbi->s_li_request != NULL) { 3999 /* 4000 * Reset timeout so it can be computed again, because 4001 * s_li_wait_mult might have changed. 4002 */ 4003 sbi->s_li_request->lr_timeout = 0; 4004 goto out; 4005 } 4006 4007 if (ext4_emergency_state(sb) || sb_rdonly(sb) || 4008 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 4009 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 4010 goto out; 4011 4012 elr = ext4_li_request_new(sb, first_not_zeroed); 4013 if (!elr) { 4014 ret = -ENOMEM; 4015 goto out; 4016 } 4017 4018 if (NULL == ext4_li_info) { 4019 ret = ext4_li_info_new(); 4020 if (ret) 4021 goto out; 4022 } 4023 4024 mutex_lock(&ext4_li_info->li_list_mtx); 4025 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 4026 mutex_unlock(&ext4_li_info->li_list_mtx); 4027 4028 sbi->s_li_request = elr; 4029 /* 4030 * set elr to NULL here since it has been inserted to 4031 * the request_list and the removal and free of it is 4032 * handled by ext4_clear_request_list from now on. 4033 */ 4034 elr = NULL; 4035 4036 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 4037 ret = ext4_run_lazyinit_thread(); 4038 if (ret) 4039 goto out; 4040 } 4041 out: 4042 mutex_unlock(&ext4_li_mtx); 4043 if (ret) 4044 kfree(elr); 4045 return ret; 4046 } 4047 4048 /* 4049 * We do not need to lock anything since this is called on 4050 * module unload. 4051 */ 4052 static void ext4_destroy_lazyinit_thread(void) 4053 { 4054 /* 4055 * If thread exited earlier 4056 * there's nothing to be done. 4057 */ 4058 if (!ext4_li_info || !ext4_lazyinit_task) 4059 return; 4060 4061 kthread_stop(ext4_lazyinit_task); 4062 } 4063 4064 static int set_journal_csum_feature_set(struct super_block *sb) 4065 { 4066 int ret = 1; 4067 int compat, incompat; 4068 struct ext4_sb_info *sbi = EXT4_SB(sb); 4069 4070 if (ext4_has_feature_metadata_csum(sb)) { 4071 /* journal checksum v3 */ 4072 compat = 0; 4073 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4074 } else { 4075 /* journal checksum v1 */ 4076 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4077 incompat = 0; 4078 } 4079 4080 jbd2_journal_clear_features(sbi->s_journal, 4081 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4082 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4083 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4084 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4085 ret = jbd2_journal_set_features(sbi->s_journal, 4086 compat, 0, 4087 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4088 incompat); 4089 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4090 ret = jbd2_journal_set_features(sbi->s_journal, 4091 compat, 0, 4092 incompat); 4093 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4094 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4095 } else { 4096 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4097 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4098 } 4099 4100 return ret; 4101 } 4102 4103 /* 4104 * Note: calculating the overhead so we can be compatible with 4105 * historical BSD practice is quite difficult in the face of 4106 * clusters/bigalloc. This is because multiple metadata blocks from 4107 * different block group can end up in the same allocation cluster. 4108 * Calculating the exact overhead in the face of clustered allocation 4109 * requires either O(all block bitmaps) in memory or O(number of block 4110 * groups**2) in time. We will still calculate the superblock for 4111 * older file systems --- and if we come across with a bigalloc file 4112 * system with zero in s_overhead_clusters the estimate will be close to 4113 * correct especially for very large cluster sizes --- but for newer 4114 * file systems, it's better to calculate this figure once at mkfs 4115 * time, and store it in the superblock. If the superblock value is 4116 * present (even for non-bigalloc file systems), we will use it. 4117 */ 4118 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4119 char *buf) 4120 { 4121 struct ext4_sb_info *sbi = EXT4_SB(sb); 4122 struct ext4_group_desc *gdp; 4123 ext4_fsblk_t first_block, last_block, b; 4124 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4125 int s, j, count = 0; 4126 int has_super = ext4_bg_has_super(sb, grp); 4127 4128 if (!ext4_has_feature_bigalloc(sb)) 4129 return (has_super + ext4_bg_num_gdb(sb, grp) + 4130 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4131 sbi->s_itb_per_group + 2); 4132 4133 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4134 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4135 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4136 for (i = 0; i < ngroups; i++) { 4137 gdp = ext4_get_group_desc(sb, i, NULL); 4138 b = ext4_block_bitmap(sb, gdp); 4139 if (b >= first_block && b <= last_block) { 4140 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4141 count++; 4142 } 4143 b = ext4_inode_bitmap(sb, gdp); 4144 if (b >= first_block && b <= last_block) { 4145 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4146 count++; 4147 } 4148 b = ext4_inode_table(sb, gdp); 4149 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4150 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4151 int c = EXT4_B2C(sbi, b - first_block); 4152 ext4_set_bit(c, buf); 4153 count++; 4154 } 4155 if (i != grp) 4156 continue; 4157 s = 0; 4158 if (ext4_bg_has_super(sb, grp)) { 4159 ext4_set_bit(s++, buf); 4160 count++; 4161 } 4162 j = ext4_bg_num_gdb(sb, grp); 4163 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4164 ext4_error(sb, "Invalid number of block group " 4165 "descriptor blocks: %d", j); 4166 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4167 } 4168 count += j; 4169 for (; j > 0; j--) 4170 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4171 } 4172 if (!count) 4173 return 0; 4174 return EXT4_CLUSTERS_PER_GROUP(sb) - 4175 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4176 } 4177 4178 /* 4179 * Compute the overhead and stash it in sbi->s_overhead 4180 */ 4181 int ext4_calculate_overhead(struct super_block *sb) 4182 { 4183 struct ext4_sb_info *sbi = EXT4_SB(sb); 4184 struct ext4_super_block *es = sbi->s_es; 4185 struct inode *j_inode; 4186 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4187 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4188 ext4_fsblk_t overhead = 0; 4189 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4190 4191 if (!buf) 4192 return -ENOMEM; 4193 4194 /* 4195 * Compute the overhead (FS structures). This is constant 4196 * for a given filesystem unless the number of block groups 4197 * changes so we cache the previous value until it does. 4198 */ 4199 4200 /* 4201 * All of the blocks before first_data_block are overhead 4202 */ 4203 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4204 4205 /* 4206 * Add the overhead found in each block group 4207 */ 4208 for (i = 0; i < ngroups; i++) { 4209 int blks; 4210 4211 blks = count_overhead(sb, i, buf); 4212 overhead += blks; 4213 if (blks) 4214 memset(buf, 0, PAGE_SIZE); 4215 cond_resched(); 4216 } 4217 4218 /* 4219 * Add the internal journal blocks whether the journal has been 4220 * loaded or not 4221 */ 4222 if (sbi->s_journal && !sbi->s_journal_bdev_file) 4223 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4224 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4225 /* j_inum for internal journal is non-zero */ 4226 j_inode = ext4_get_journal_inode(sb, j_inum); 4227 if (!IS_ERR(j_inode)) { 4228 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4229 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4230 iput(j_inode); 4231 } else { 4232 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4233 } 4234 } 4235 sbi->s_overhead = overhead; 4236 smp_wmb(); 4237 free_page((unsigned long) buf); 4238 return 0; 4239 } 4240 4241 static void ext4_set_resv_clusters(struct super_block *sb) 4242 { 4243 ext4_fsblk_t resv_clusters; 4244 struct ext4_sb_info *sbi = EXT4_SB(sb); 4245 4246 /* 4247 * There's no need to reserve anything when we aren't using extents. 4248 * The space estimates are exact, there are no unwritten extents, 4249 * hole punching doesn't need new metadata... This is needed especially 4250 * to keep ext2/3 backward compatibility. 4251 */ 4252 if (!ext4_has_feature_extents(sb)) 4253 return; 4254 /* 4255 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4256 * This should cover the situations where we can not afford to run 4257 * out of space like for example punch hole, or converting 4258 * unwritten extents in delalloc path. In most cases such 4259 * allocation would require 1, or 2 blocks, higher numbers are 4260 * very rare. 4261 */ 4262 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4263 sbi->s_cluster_bits); 4264 4265 do_div(resv_clusters, 50); 4266 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4267 4268 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4269 } 4270 4271 static const char *ext4_quota_mode(struct super_block *sb) 4272 { 4273 #ifdef CONFIG_QUOTA 4274 if (!ext4_quota_capable(sb)) 4275 return "none"; 4276 4277 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4278 return "journalled"; 4279 else 4280 return "writeback"; 4281 #else 4282 return "disabled"; 4283 #endif 4284 } 4285 4286 static void ext4_setup_csum_trigger(struct super_block *sb, 4287 enum ext4_journal_trigger_type type, 4288 void (*trigger)( 4289 struct jbd2_buffer_trigger_type *type, 4290 struct buffer_head *bh, 4291 void *mapped_data, 4292 size_t size)) 4293 { 4294 struct ext4_sb_info *sbi = EXT4_SB(sb); 4295 4296 sbi->s_journal_triggers[type].sb = sb; 4297 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4298 } 4299 4300 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4301 { 4302 if (!sbi) 4303 return; 4304 4305 kfree(sbi->s_blockgroup_lock); 4306 fs_put_dax(sbi->s_daxdev, NULL); 4307 kfree(sbi); 4308 } 4309 4310 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4311 { 4312 struct ext4_sb_info *sbi; 4313 4314 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4315 if (!sbi) 4316 return NULL; 4317 4318 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4319 NULL, NULL); 4320 4321 sbi->s_blockgroup_lock = 4322 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4323 4324 if (!sbi->s_blockgroup_lock) 4325 goto err_out; 4326 4327 sb->s_fs_info = sbi; 4328 sbi->s_sb = sb; 4329 return sbi; 4330 err_out: 4331 fs_put_dax(sbi->s_daxdev, NULL); 4332 kfree(sbi); 4333 return NULL; 4334 } 4335 4336 static void ext4_set_def_opts(struct super_block *sb, 4337 struct ext4_super_block *es) 4338 { 4339 unsigned long def_mount_opts; 4340 4341 /* Set defaults before we parse the mount options */ 4342 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4343 set_opt(sb, INIT_INODE_TABLE); 4344 if (def_mount_opts & EXT4_DEFM_DEBUG) 4345 set_opt(sb, DEBUG); 4346 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4347 set_opt(sb, GRPID); 4348 if (def_mount_opts & EXT4_DEFM_UID16) 4349 set_opt(sb, NO_UID32); 4350 /* xattr user namespace & acls are now defaulted on */ 4351 set_opt(sb, XATTR_USER); 4352 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4353 set_opt(sb, POSIX_ACL); 4354 #endif 4355 if (ext4_has_feature_fast_commit(sb)) 4356 set_opt2(sb, JOURNAL_FAST_COMMIT); 4357 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4358 if (ext4_has_feature_metadata_csum(sb)) 4359 set_opt(sb, JOURNAL_CHECKSUM); 4360 4361 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4362 set_opt(sb, JOURNAL_DATA); 4363 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4364 set_opt(sb, ORDERED_DATA); 4365 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4366 set_opt(sb, WRITEBACK_DATA); 4367 4368 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4369 set_opt(sb, ERRORS_PANIC); 4370 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4371 set_opt(sb, ERRORS_CONT); 4372 else 4373 set_opt(sb, ERRORS_RO); 4374 /* block_validity enabled by default; disable with noblock_validity */ 4375 set_opt(sb, BLOCK_VALIDITY); 4376 if (def_mount_opts & EXT4_DEFM_DISCARD) 4377 set_opt(sb, DISCARD); 4378 4379 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4380 set_opt(sb, BARRIER); 4381 4382 /* 4383 * enable delayed allocation by default 4384 * Use -o nodelalloc to turn it off 4385 */ 4386 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4387 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4388 set_opt(sb, DELALLOC); 4389 4390 if (sb->s_blocksize <= PAGE_SIZE) 4391 set_opt(sb, DIOREAD_NOLOCK); 4392 } 4393 4394 static int ext4_handle_clustersize(struct super_block *sb) 4395 { 4396 struct ext4_sb_info *sbi = EXT4_SB(sb); 4397 struct ext4_super_block *es = sbi->s_es; 4398 int clustersize; 4399 4400 /* Handle clustersize */ 4401 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4402 if (ext4_has_feature_bigalloc(sb)) { 4403 if (clustersize < sb->s_blocksize) { 4404 ext4_msg(sb, KERN_ERR, 4405 "cluster size (%d) smaller than " 4406 "block size (%lu)", clustersize, sb->s_blocksize); 4407 return -EINVAL; 4408 } 4409 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4410 le32_to_cpu(es->s_log_block_size); 4411 } else { 4412 if (clustersize != sb->s_blocksize) { 4413 ext4_msg(sb, KERN_ERR, 4414 "fragment/cluster size (%d) != " 4415 "block size (%lu)", clustersize, sb->s_blocksize); 4416 return -EINVAL; 4417 } 4418 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4419 ext4_msg(sb, KERN_ERR, 4420 "#blocks per group too big: %lu", 4421 sbi->s_blocks_per_group); 4422 return -EINVAL; 4423 } 4424 sbi->s_cluster_bits = 0; 4425 } 4426 sbi->s_clusters_per_group = le32_to_cpu(es->s_clusters_per_group); 4427 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4428 ext4_msg(sb, KERN_ERR, "#clusters per group too big: %lu", 4429 sbi->s_clusters_per_group); 4430 return -EINVAL; 4431 } 4432 if (sbi->s_blocks_per_group != 4433 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4434 ext4_msg(sb, KERN_ERR, 4435 "blocks per group (%lu) and clusters per group (%lu) inconsistent", 4436 sbi->s_blocks_per_group, sbi->s_clusters_per_group); 4437 return -EINVAL; 4438 } 4439 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4440 4441 /* Do we have standard group size of clustersize * 8 blocks ? */ 4442 if (sbi->s_blocks_per_group == clustersize << 3) 4443 set_opt2(sb, STD_GROUP_SIZE); 4444 4445 return 0; 4446 } 4447 4448 /* 4449 * ext4_atomic_write_init: Initializes filesystem min & max atomic write units. 4450 * With non-bigalloc filesystem awu will be based upon filesystem blocksize 4451 * & bdev awu units. 4452 * With bigalloc it will be based upon bigalloc cluster size & bdev awu units. 4453 * @sb: super block 4454 */ 4455 static void ext4_atomic_write_init(struct super_block *sb) 4456 { 4457 struct ext4_sb_info *sbi = EXT4_SB(sb); 4458 struct block_device *bdev = sb->s_bdev; 4459 unsigned int clustersize = EXT4_CLUSTER_SIZE(sb); 4460 4461 if (!bdev_can_atomic_write(bdev)) 4462 return; 4463 4464 if (!ext4_has_feature_extents(sb)) 4465 return; 4466 4467 sbi->s_awu_min = max(sb->s_blocksize, 4468 bdev_atomic_write_unit_min_bytes(bdev)); 4469 sbi->s_awu_max = min(clustersize, 4470 bdev_atomic_write_unit_max_bytes(bdev)); 4471 if (sbi->s_awu_min && sbi->s_awu_max && 4472 sbi->s_awu_min <= sbi->s_awu_max) { 4473 ext4_msg(sb, KERN_NOTICE, "Supports (experimental) DIO atomic writes awu_min: %u, awu_max: %u", 4474 sbi->s_awu_min, sbi->s_awu_max); 4475 } else { 4476 sbi->s_awu_min = 0; 4477 sbi->s_awu_max = 0; 4478 } 4479 } 4480 4481 static void ext4_fast_commit_init(struct super_block *sb) 4482 { 4483 struct ext4_sb_info *sbi = EXT4_SB(sb); 4484 4485 /* Initialize fast commit stuff */ 4486 atomic_set(&sbi->s_fc_subtid, 0); 4487 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4488 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4489 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4490 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4491 sbi->s_fc_bytes = 0; 4492 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4493 sbi->s_fc_ineligible_tid = 0; 4494 mutex_init(&sbi->s_fc_lock); 4495 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4496 sbi->s_fc_replay_state.fc_regions = NULL; 4497 sbi->s_fc_replay_state.fc_regions_size = 0; 4498 sbi->s_fc_replay_state.fc_regions_used = 0; 4499 sbi->s_fc_replay_state.fc_regions_valid = 0; 4500 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4501 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4502 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4503 } 4504 4505 static int ext4_inode_info_init(struct super_block *sb, 4506 struct ext4_super_block *es) 4507 { 4508 struct ext4_sb_info *sbi = EXT4_SB(sb); 4509 4510 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4511 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4512 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4513 } else { 4514 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4515 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4516 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4517 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4518 sbi->s_first_ino); 4519 return -EINVAL; 4520 } 4521 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4522 (!is_power_of_2(sbi->s_inode_size)) || 4523 (sbi->s_inode_size > sb->s_blocksize)) { 4524 ext4_msg(sb, KERN_ERR, 4525 "unsupported inode size: %d", 4526 sbi->s_inode_size); 4527 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4528 return -EINVAL; 4529 } 4530 /* 4531 * i_atime_extra is the last extra field available for 4532 * [acm]times in struct ext4_inode. Checking for that 4533 * field should suffice to ensure we have extra space 4534 * for all three. 4535 */ 4536 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4537 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4538 sb->s_time_gran = 1; 4539 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4540 } else { 4541 sb->s_time_gran = NSEC_PER_SEC; 4542 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4543 } 4544 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4545 } 4546 4547 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4548 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4549 EXT4_GOOD_OLD_INODE_SIZE; 4550 if (ext4_has_feature_extra_isize(sb)) { 4551 unsigned v, max = (sbi->s_inode_size - 4552 EXT4_GOOD_OLD_INODE_SIZE); 4553 4554 v = le16_to_cpu(es->s_want_extra_isize); 4555 if (v > max) { 4556 ext4_msg(sb, KERN_ERR, 4557 "bad s_want_extra_isize: %d", v); 4558 return -EINVAL; 4559 } 4560 if (sbi->s_want_extra_isize < v) 4561 sbi->s_want_extra_isize = v; 4562 4563 v = le16_to_cpu(es->s_min_extra_isize); 4564 if (v > max) { 4565 ext4_msg(sb, KERN_ERR, 4566 "bad s_min_extra_isize: %d", v); 4567 return -EINVAL; 4568 } 4569 if (sbi->s_want_extra_isize < v) 4570 sbi->s_want_extra_isize = v; 4571 } 4572 } 4573 4574 return 0; 4575 } 4576 4577 #if IS_ENABLED(CONFIG_UNICODE) 4578 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4579 { 4580 const struct ext4_sb_encodings *encoding_info; 4581 struct unicode_map *encoding; 4582 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4583 4584 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4585 return 0; 4586 4587 encoding_info = ext4_sb_read_encoding(es); 4588 if (!encoding_info) { 4589 ext4_msg(sb, KERN_ERR, 4590 "Encoding requested by superblock is unknown"); 4591 return -EINVAL; 4592 } 4593 4594 encoding = utf8_load(encoding_info->version); 4595 if (IS_ERR(encoding)) { 4596 ext4_msg(sb, KERN_ERR, 4597 "can't mount with superblock charset: %s-%u.%u.%u " 4598 "not supported by the kernel. flags: 0x%x.", 4599 encoding_info->name, 4600 unicode_major(encoding_info->version), 4601 unicode_minor(encoding_info->version), 4602 unicode_rev(encoding_info->version), 4603 encoding_flags); 4604 return -EINVAL; 4605 } 4606 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4607 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4608 unicode_major(encoding_info->version), 4609 unicode_minor(encoding_info->version), 4610 unicode_rev(encoding_info->version), 4611 encoding_flags); 4612 4613 sb->s_encoding = encoding; 4614 sb->s_encoding_flags = encoding_flags; 4615 4616 return 0; 4617 } 4618 #else 4619 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4620 { 4621 return 0; 4622 } 4623 #endif 4624 4625 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4626 { 4627 struct ext4_sb_info *sbi = EXT4_SB(sb); 4628 4629 /* Warn if metadata_csum and gdt_csum are both set. */ 4630 if (ext4_has_feature_metadata_csum(sb) && 4631 ext4_has_feature_gdt_csum(sb)) 4632 ext4_warning(sb, "metadata_csum and uninit_bg are " 4633 "redundant flags; please run fsck."); 4634 4635 /* Check for a known checksum algorithm */ 4636 if (!ext4_verify_csum_type(sb, es)) { 4637 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4638 "unknown checksum algorithm."); 4639 return -EINVAL; 4640 } 4641 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4642 ext4_orphan_file_block_trigger); 4643 4644 /* Check superblock checksum */ 4645 if (!ext4_superblock_csum_verify(sb, es)) { 4646 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4647 "invalid superblock checksum. Run e2fsck?"); 4648 return -EFSBADCRC; 4649 } 4650 4651 /* Precompute checksum seed for all metadata */ 4652 if (ext4_has_feature_csum_seed(sb)) 4653 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4654 else if (ext4_has_feature_metadata_csum(sb) || 4655 ext4_has_feature_ea_inode(sb)) 4656 sbi->s_csum_seed = ext4_chksum(~0, es->s_uuid, 4657 sizeof(es->s_uuid)); 4658 return 0; 4659 } 4660 4661 static int ext4_check_feature_compatibility(struct super_block *sb, 4662 struct ext4_super_block *es, 4663 int silent) 4664 { 4665 struct ext4_sb_info *sbi = EXT4_SB(sb); 4666 4667 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4668 (ext4_has_compat_features(sb) || 4669 ext4_has_ro_compat_features(sb) || 4670 ext4_has_incompat_features(sb))) 4671 ext4_msg(sb, KERN_WARNING, 4672 "feature flags set on rev 0 fs, " 4673 "running e2fsck is recommended"); 4674 4675 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4676 set_opt2(sb, HURD_COMPAT); 4677 if (ext4_has_feature_64bit(sb)) { 4678 ext4_msg(sb, KERN_ERR, 4679 "The Hurd can't support 64-bit file systems"); 4680 return -EINVAL; 4681 } 4682 4683 /* 4684 * ea_inode feature uses l_i_version field which is not 4685 * available in HURD_COMPAT mode. 4686 */ 4687 if (ext4_has_feature_ea_inode(sb)) { 4688 ext4_msg(sb, KERN_ERR, 4689 "ea_inode feature is not supported for Hurd"); 4690 return -EINVAL; 4691 } 4692 } 4693 4694 if (IS_EXT2_SB(sb)) { 4695 if (ext2_feature_set_ok(sb)) 4696 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4697 "using the ext4 subsystem"); 4698 else { 4699 /* 4700 * If we're probing be silent, if this looks like 4701 * it's actually an ext[34] filesystem. 4702 */ 4703 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4704 return -EINVAL; 4705 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4706 "to feature incompatibilities"); 4707 return -EINVAL; 4708 } 4709 } 4710 4711 if (IS_EXT3_SB(sb)) { 4712 if (ext3_feature_set_ok(sb)) 4713 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4714 "using the ext4 subsystem"); 4715 else { 4716 /* 4717 * If we're probing be silent, if this looks like 4718 * it's actually an ext4 filesystem. 4719 */ 4720 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4721 return -EINVAL; 4722 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4723 "to feature incompatibilities"); 4724 return -EINVAL; 4725 } 4726 } 4727 4728 /* 4729 * Check feature flags regardless of the revision level, since we 4730 * previously didn't change the revision level when setting the flags, 4731 * so there is a chance incompat flags are set on a rev 0 filesystem. 4732 */ 4733 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4734 return -EINVAL; 4735 4736 if (sbi->s_daxdev) { 4737 if (sb->s_blocksize == PAGE_SIZE) 4738 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4739 else 4740 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 4741 } 4742 4743 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4744 if (ext4_has_feature_inline_data(sb)) { 4745 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4746 " that may contain inline data"); 4747 return -EINVAL; 4748 } 4749 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4750 ext4_msg(sb, KERN_ERR, 4751 "DAX unsupported by block device."); 4752 return -EINVAL; 4753 } 4754 } 4755 4756 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4757 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4758 es->s_encryption_level); 4759 return -EINVAL; 4760 } 4761 4762 return 0; 4763 } 4764 4765 static int ext4_check_geometry(struct super_block *sb, 4766 struct ext4_super_block *es) 4767 { 4768 struct ext4_sb_info *sbi = EXT4_SB(sb); 4769 __u64 blocks_count; 4770 int err; 4771 4772 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 4773 ext4_msg(sb, KERN_ERR, 4774 "Number of reserved GDT blocks insanely large: %d", 4775 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4776 return -EINVAL; 4777 } 4778 /* 4779 * Test whether we have more sectors than will fit in sector_t, 4780 * and whether the max offset is addressable by the page cache. 4781 */ 4782 err = generic_check_addressable(sb->s_blocksize_bits, 4783 ext4_blocks_count(es)); 4784 if (err) { 4785 ext4_msg(sb, KERN_ERR, "filesystem" 4786 " too large to mount safely on this system"); 4787 return err; 4788 } 4789 4790 /* check blocks count against device size */ 4791 blocks_count = sb_bdev_nr_blocks(sb); 4792 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4793 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4794 "exceeds size of device (%llu blocks)", 4795 ext4_blocks_count(es), blocks_count); 4796 return -EINVAL; 4797 } 4798 4799 /* 4800 * It makes no sense for the first data block to be beyond the end 4801 * of the filesystem. 4802 */ 4803 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4804 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4805 "block %u is beyond end of filesystem (%llu)", 4806 le32_to_cpu(es->s_first_data_block), 4807 ext4_blocks_count(es)); 4808 return -EINVAL; 4809 } 4810 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4811 (sbi->s_cluster_ratio == 1)) { 4812 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4813 "block is 0 with a 1k block and cluster size"); 4814 return -EINVAL; 4815 } 4816 4817 blocks_count = (ext4_blocks_count(es) - 4818 le32_to_cpu(es->s_first_data_block) + 4819 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4820 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4821 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4822 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4823 "(block count %llu, first data block %u, " 4824 "blocks per group %lu)", blocks_count, 4825 ext4_blocks_count(es), 4826 le32_to_cpu(es->s_first_data_block), 4827 EXT4_BLOCKS_PER_GROUP(sb)); 4828 return -EINVAL; 4829 } 4830 sbi->s_groups_count = blocks_count; 4831 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4832 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4833 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4834 le32_to_cpu(es->s_inodes_count)) { 4835 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4836 le32_to_cpu(es->s_inodes_count), 4837 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4838 return -EINVAL; 4839 } 4840 4841 return 0; 4842 } 4843 4844 static int ext4_group_desc_init(struct super_block *sb, 4845 struct ext4_super_block *es, 4846 ext4_fsblk_t logical_sb_block, 4847 ext4_group_t *first_not_zeroed) 4848 { 4849 struct ext4_sb_info *sbi = EXT4_SB(sb); 4850 unsigned int db_count; 4851 ext4_fsblk_t block; 4852 int i; 4853 4854 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4855 EXT4_DESC_PER_BLOCK(sb); 4856 if (ext4_has_feature_meta_bg(sb)) { 4857 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4858 ext4_msg(sb, KERN_WARNING, 4859 "first meta block group too large: %u " 4860 "(group descriptor block count %u)", 4861 le32_to_cpu(es->s_first_meta_bg), db_count); 4862 return -EINVAL; 4863 } 4864 } 4865 rcu_assign_pointer(sbi->s_group_desc, 4866 kvmalloc_array(db_count, 4867 sizeof(struct buffer_head *), 4868 GFP_KERNEL)); 4869 if (sbi->s_group_desc == NULL) { 4870 ext4_msg(sb, KERN_ERR, "not enough memory"); 4871 return -ENOMEM; 4872 } 4873 4874 bgl_lock_init(sbi->s_blockgroup_lock); 4875 4876 /* Pre-read the descriptors into the buffer cache */ 4877 for (i = 0; i < db_count; i++) { 4878 block = descriptor_loc(sb, logical_sb_block, i); 4879 ext4_sb_breadahead_unmovable(sb, block); 4880 } 4881 4882 for (i = 0; i < db_count; i++) { 4883 struct buffer_head *bh; 4884 4885 block = descriptor_loc(sb, logical_sb_block, i); 4886 bh = ext4_sb_bread_unmovable(sb, block); 4887 if (IS_ERR(bh)) { 4888 ext4_msg(sb, KERN_ERR, 4889 "can't read group descriptor %d", i); 4890 sbi->s_gdb_count = i; 4891 return PTR_ERR(bh); 4892 } 4893 rcu_read_lock(); 4894 rcu_dereference(sbi->s_group_desc)[i] = bh; 4895 rcu_read_unlock(); 4896 } 4897 sbi->s_gdb_count = db_count; 4898 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4899 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4900 return -EFSCORRUPTED; 4901 } 4902 4903 return 0; 4904 } 4905 4906 static int ext4_load_and_init_journal(struct super_block *sb, 4907 struct ext4_super_block *es, 4908 struct ext4_fs_context *ctx) 4909 { 4910 struct ext4_sb_info *sbi = EXT4_SB(sb); 4911 int err; 4912 4913 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4914 if (err) 4915 return err; 4916 4917 if (ext4_has_feature_64bit(sb) && 4918 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4919 JBD2_FEATURE_INCOMPAT_64BIT)) { 4920 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4921 goto out; 4922 } 4923 4924 if (!set_journal_csum_feature_set(sb)) { 4925 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4926 "feature set"); 4927 goto out; 4928 } 4929 4930 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4931 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4932 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4933 ext4_msg(sb, KERN_ERR, 4934 "Failed to set fast commit journal feature"); 4935 goto out; 4936 } 4937 4938 /* We have now updated the journal if required, so we can 4939 * validate the data journaling mode. */ 4940 switch (test_opt(sb, DATA_FLAGS)) { 4941 case 0: 4942 /* No mode set, assume a default based on the journal 4943 * capabilities: ORDERED_DATA if the journal can 4944 * cope, else JOURNAL_DATA 4945 */ 4946 if (jbd2_journal_check_available_features 4947 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4948 set_opt(sb, ORDERED_DATA); 4949 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4950 } else { 4951 set_opt(sb, JOURNAL_DATA); 4952 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4953 } 4954 break; 4955 4956 case EXT4_MOUNT_ORDERED_DATA: 4957 case EXT4_MOUNT_WRITEBACK_DATA: 4958 if (!jbd2_journal_check_available_features 4959 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4960 ext4_msg(sb, KERN_ERR, "Journal does not support " 4961 "requested data journaling mode"); 4962 goto out; 4963 } 4964 break; 4965 default: 4966 break; 4967 } 4968 4969 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4970 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4971 ext4_msg(sb, KERN_ERR, "can't mount with " 4972 "journal_async_commit in data=ordered mode"); 4973 goto out; 4974 } 4975 4976 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4977 4978 sbi->s_journal->j_submit_inode_data_buffers = 4979 ext4_journal_submit_inode_data_buffers; 4980 sbi->s_journal->j_finish_inode_data_buffers = 4981 ext4_journal_finish_inode_data_buffers; 4982 4983 return 0; 4984 4985 out: 4986 ext4_journal_destroy(sbi, sbi->s_journal); 4987 return -EINVAL; 4988 } 4989 4990 static int ext4_check_journal_data_mode(struct super_block *sb) 4991 { 4992 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4993 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4994 "data=journal disables delayed allocation, " 4995 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 4996 /* can't mount with both data=journal and dioread_nolock. */ 4997 clear_opt(sb, DIOREAD_NOLOCK); 4998 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4999 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5000 ext4_msg(sb, KERN_ERR, "can't mount with " 5001 "both data=journal and delalloc"); 5002 return -EINVAL; 5003 } 5004 if (test_opt(sb, DAX_ALWAYS)) { 5005 ext4_msg(sb, KERN_ERR, "can't mount with " 5006 "both data=journal and dax"); 5007 return -EINVAL; 5008 } 5009 if (ext4_has_feature_encrypt(sb)) { 5010 ext4_msg(sb, KERN_WARNING, 5011 "encrypted files will use data=ordered " 5012 "instead of data journaling mode"); 5013 } 5014 if (test_opt(sb, DELALLOC)) 5015 clear_opt(sb, DELALLOC); 5016 } else { 5017 sb->s_iflags |= SB_I_CGROUPWB; 5018 } 5019 5020 return 0; 5021 } 5022 5023 static const char *ext4_has_journal_option(struct super_block *sb) 5024 { 5025 struct ext4_sb_info *sbi = EXT4_SB(sb); 5026 5027 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) 5028 return "journal_async_commit"; 5029 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) 5030 return "journal_checksum"; 5031 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 5032 return "commit="; 5033 if (EXT4_MOUNT_DATA_FLAGS & 5034 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) 5035 return "data="; 5036 if (test_opt(sb, DATA_ERR_ABORT)) 5037 return "data_err=abort"; 5038 return NULL; 5039 } 5040 5041 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 5042 int silent) 5043 { 5044 struct ext4_sb_info *sbi = EXT4_SB(sb); 5045 struct ext4_super_block *es; 5046 ext4_fsblk_t logical_sb_block; 5047 unsigned long offset = 0; 5048 struct buffer_head *bh; 5049 int ret = -EINVAL; 5050 int blocksize; 5051 5052 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 5053 if (!blocksize) { 5054 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 5055 return -EINVAL; 5056 } 5057 5058 /* 5059 * The ext4 superblock will not be buffer aligned for other than 1kB 5060 * block sizes. We need to calculate the offset from buffer start. 5061 */ 5062 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 5063 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5064 offset = do_div(logical_sb_block, blocksize); 5065 } else { 5066 logical_sb_block = sbi->s_sb_block; 5067 } 5068 5069 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5070 if (IS_ERR(bh)) { 5071 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 5072 return PTR_ERR(bh); 5073 } 5074 /* 5075 * Note: s_es must be initialized as soon as possible because 5076 * some ext4 macro-instructions depend on its value 5077 */ 5078 es = (struct ext4_super_block *) (bh->b_data + offset); 5079 sbi->s_es = es; 5080 sb->s_magic = le16_to_cpu(es->s_magic); 5081 if (sb->s_magic != EXT4_SUPER_MAGIC) { 5082 if (!silent) 5083 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5084 goto out; 5085 } 5086 5087 if (le32_to_cpu(es->s_log_block_size) > 5088 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5089 ext4_msg(sb, KERN_ERR, 5090 "Invalid log block size: %u", 5091 le32_to_cpu(es->s_log_block_size)); 5092 goto out; 5093 } 5094 if (le32_to_cpu(es->s_log_cluster_size) > 5095 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5096 ext4_msg(sb, KERN_ERR, 5097 "Invalid log cluster size: %u", 5098 le32_to_cpu(es->s_log_cluster_size)); 5099 goto out; 5100 } 5101 5102 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 5103 5104 /* 5105 * If the default block size is not the same as the real block size, 5106 * we need to reload it. 5107 */ 5108 if (sb->s_blocksize == blocksize) { 5109 *lsb = logical_sb_block; 5110 sbi->s_sbh = bh; 5111 return 0; 5112 } 5113 5114 /* 5115 * bh must be released before kill_bdev(), otherwise 5116 * it won't be freed and its page also. kill_bdev() 5117 * is called by sb_set_blocksize(). 5118 */ 5119 brelse(bh); 5120 /* Validate the filesystem blocksize */ 5121 if (!sb_set_blocksize(sb, blocksize)) { 5122 ext4_msg(sb, KERN_ERR, "bad block size %d", 5123 blocksize); 5124 bh = NULL; 5125 goto out; 5126 } 5127 5128 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5129 offset = do_div(logical_sb_block, blocksize); 5130 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5131 if (IS_ERR(bh)) { 5132 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5133 ret = PTR_ERR(bh); 5134 bh = NULL; 5135 goto out; 5136 } 5137 es = (struct ext4_super_block *)(bh->b_data + offset); 5138 sbi->s_es = es; 5139 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5140 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5141 goto out; 5142 } 5143 *lsb = logical_sb_block; 5144 sbi->s_sbh = bh; 5145 return 0; 5146 out: 5147 brelse(bh); 5148 return ret; 5149 } 5150 5151 static int ext4_hash_info_init(struct super_block *sb) 5152 { 5153 struct ext4_sb_info *sbi = EXT4_SB(sb); 5154 struct ext4_super_block *es = sbi->s_es; 5155 unsigned int i; 5156 5157 sbi->s_def_hash_version = es->s_def_hash_version; 5158 5159 if (sbi->s_def_hash_version > DX_HASH_LAST) { 5160 ext4_msg(sb, KERN_ERR, 5161 "Invalid default hash set in the superblock"); 5162 return -EINVAL; 5163 } else if (sbi->s_def_hash_version == DX_HASH_SIPHASH) { 5164 ext4_msg(sb, KERN_ERR, 5165 "SIPHASH is not a valid default hash value"); 5166 return -EINVAL; 5167 } 5168 5169 for (i = 0; i < 4; i++) 5170 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5171 5172 if (ext4_has_feature_dir_index(sb)) { 5173 i = le32_to_cpu(es->s_flags); 5174 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5175 sbi->s_hash_unsigned = 3; 5176 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5177 #ifdef __CHAR_UNSIGNED__ 5178 if (!sb_rdonly(sb)) 5179 es->s_flags |= 5180 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5181 sbi->s_hash_unsigned = 3; 5182 #else 5183 if (!sb_rdonly(sb)) 5184 es->s_flags |= 5185 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5186 #endif 5187 } 5188 } 5189 return 0; 5190 } 5191 5192 static int ext4_block_group_meta_init(struct super_block *sb, int silent) 5193 { 5194 struct ext4_sb_info *sbi = EXT4_SB(sb); 5195 struct ext4_super_block *es = sbi->s_es; 5196 int has_huge_files; 5197 5198 has_huge_files = ext4_has_feature_huge_file(sb); 5199 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5200 has_huge_files); 5201 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5202 5203 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5204 if (ext4_has_feature_64bit(sb)) { 5205 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5206 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5207 !is_power_of_2(sbi->s_desc_size)) { 5208 ext4_msg(sb, KERN_ERR, 5209 "unsupported descriptor size %lu", 5210 sbi->s_desc_size); 5211 return -EINVAL; 5212 } 5213 } else 5214 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5215 5216 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5217 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5218 5219 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5220 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5221 if (!silent) 5222 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5223 return -EINVAL; 5224 } 5225 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5226 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5227 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5228 sbi->s_inodes_per_group); 5229 return -EINVAL; 5230 } 5231 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5232 sbi->s_inodes_per_block; 5233 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5234 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5235 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5236 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5237 5238 return 0; 5239 } 5240 5241 /* 5242 * It's hard to get stripe aligned blocks if stripe is not aligned with 5243 * cluster, just disable stripe and alert user to simplify code and avoid 5244 * stripe aligned allocation which will rarely succeed. 5245 */ 5246 static bool ext4_is_stripe_incompatible(struct super_block *sb, unsigned long stripe) 5247 { 5248 struct ext4_sb_info *sbi = EXT4_SB(sb); 5249 return (stripe > 0 && sbi->s_cluster_ratio > 1 && 5250 stripe % sbi->s_cluster_ratio != 0); 5251 } 5252 5253 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5254 { 5255 struct ext4_super_block *es = NULL; 5256 struct ext4_sb_info *sbi = EXT4_SB(sb); 5257 ext4_fsblk_t logical_sb_block; 5258 struct inode *root; 5259 int needs_recovery; 5260 int err; 5261 ext4_group_t first_not_zeroed; 5262 struct ext4_fs_context *ctx = fc->fs_private; 5263 int silent = fc->sb_flags & SB_SILENT; 5264 5265 /* Set defaults for the variables that will be set during parsing */ 5266 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5267 ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO; 5268 5269 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5270 sbi->s_sectors_written_start = 5271 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5272 5273 err = ext4_load_super(sb, &logical_sb_block, silent); 5274 if (err) 5275 goto out_fail; 5276 5277 es = sbi->s_es; 5278 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5279 5280 err = ext4_init_metadata_csum(sb, es); 5281 if (err) 5282 goto failed_mount; 5283 5284 ext4_set_def_opts(sb, es); 5285 5286 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5287 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5288 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5289 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5290 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5291 sbi->s_sb_update_kb = EXT4_DEF_SB_UPDATE_INTERVAL_KB; 5292 sbi->s_sb_update_sec = EXT4_DEF_SB_UPDATE_INTERVAL_SEC; 5293 5294 /* 5295 * set default s_li_wait_mult for lazyinit, for the case there is 5296 * no mount option specified. 5297 */ 5298 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5299 5300 err = ext4_inode_info_init(sb, es); 5301 if (err) 5302 goto failed_mount; 5303 5304 err = parse_apply_sb_mount_options(sb, ctx); 5305 if (err < 0) 5306 goto failed_mount; 5307 5308 sbi->s_def_mount_opt = sbi->s_mount_opt; 5309 sbi->s_def_mount_opt2 = sbi->s_mount_opt2; 5310 5311 err = ext4_check_opt_consistency(fc, sb); 5312 if (err < 0) 5313 goto failed_mount; 5314 5315 ext4_apply_options(fc, sb); 5316 5317 err = ext4_encoding_init(sb, es); 5318 if (err) 5319 goto failed_mount; 5320 5321 err = ext4_check_journal_data_mode(sb); 5322 if (err) 5323 goto failed_mount; 5324 5325 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5326 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5327 5328 /* HSM events are allowed by default. */ 5329 sb->s_iflags |= SB_I_ALLOW_HSM; 5330 5331 err = ext4_check_feature_compatibility(sb, es, silent); 5332 if (err) 5333 goto failed_mount; 5334 5335 err = ext4_block_group_meta_init(sb, silent); 5336 if (err) 5337 goto failed_mount; 5338 5339 err = ext4_hash_info_init(sb); 5340 if (err) 5341 goto failed_mount; 5342 5343 err = ext4_handle_clustersize(sb); 5344 if (err) 5345 goto failed_mount; 5346 5347 err = ext4_check_geometry(sb, es); 5348 if (err) 5349 goto failed_mount; 5350 5351 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5352 spin_lock_init(&sbi->s_error_lock); 5353 INIT_WORK(&sbi->s_sb_upd_work, update_super_work); 5354 5355 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5356 if (err) 5357 goto failed_mount3; 5358 5359 err = ext4_es_register_shrinker(sbi); 5360 if (err) 5361 goto failed_mount3; 5362 5363 sbi->s_stripe = ext4_get_stripe_size(sbi); 5364 if (ext4_is_stripe_incompatible(sb, sbi->s_stripe)) { 5365 ext4_msg(sb, KERN_WARNING, 5366 "stripe (%lu) is not aligned with cluster size (%u), " 5367 "stripe is disabled", 5368 sbi->s_stripe, sbi->s_cluster_ratio); 5369 sbi->s_stripe = 0; 5370 } 5371 sbi->s_extent_max_zeroout_kb = 32; 5372 5373 /* 5374 * set up enough so that it can read an inode 5375 */ 5376 sb->s_op = &ext4_sops; 5377 sb->s_export_op = &ext4_export_ops; 5378 sb->s_xattr = ext4_xattr_handlers; 5379 #ifdef CONFIG_FS_ENCRYPTION 5380 sb->s_cop = &ext4_cryptops; 5381 #endif 5382 #ifdef CONFIG_FS_VERITY 5383 sb->s_vop = &ext4_verityops; 5384 #endif 5385 #ifdef CONFIG_QUOTA 5386 sb->dq_op = &ext4_quota_operations; 5387 if (ext4_has_feature_quota(sb)) 5388 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5389 else 5390 sb->s_qcop = &ext4_qctl_operations; 5391 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5392 #endif 5393 super_set_uuid(sb, es->s_uuid, sizeof(es->s_uuid)); 5394 super_set_sysfs_name_bdev(sb); 5395 5396 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5397 mutex_init(&sbi->s_orphan_lock); 5398 5399 spin_lock_init(&sbi->s_bdev_wb_lock); 5400 5401 ext4_atomic_write_init(sb); 5402 ext4_fast_commit_init(sb); 5403 5404 sb->s_root = NULL; 5405 5406 needs_recovery = (es->s_last_orphan != 0 || 5407 ext4_has_feature_orphan_present(sb) || 5408 ext4_has_feature_journal_needs_recovery(sb)); 5409 5410 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) { 5411 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)); 5412 if (err) 5413 goto failed_mount3a; 5414 } 5415 5416 err = -EINVAL; 5417 /* 5418 * The first inode we look at is the journal inode. Don't try 5419 * root first: it may be modified in the journal! 5420 */ 5421 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5422 err = ext4_load_and_init_journal(sb, es, ctx); 5423 if (err) 5424 goto failed_mount3a; 5425 if (bdev_read_only(sb->s_bdev)) 5426 needs_recovery = 0; 5427 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5428 ext4_has_feature_journal_needs_recovery(sb)) { 5429 ext4_msg(sb, KERN_ERR, "required journal recovery " 5430 "suppressed and not mounted read-only"); 5431 goto failed_mount3a; 5432 } else { 5433 const char *journal_option; 5434 5435 /* Nojournal mode, all journal mount options are illegal */ 5436 journal_option = ext4_has_journal_option(sb); 5437 if (journal_option != NULL) { 5438 ext4_msg(sb, KERN_ERR, 5439 "can't mount with %s, fs mounted w/o journal", 5440 journal_option); 5441 goto failed_mount3a; 5442 } 5443 5444 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5445 clear_opt(sb, JOURNAL_CHECKSUM); 5446 clear_opt(sb, DATA_FLAGS); 5447 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5448 sbi->s_journal = NULL; 5449 needs_recovery = 0; 5450 } 5451 5452 if (!test_opt(sb, NO_MBCACHE)) { 5453 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5454 if (!sbi->s_ea_block_cache) { 5455 ext4_msg(sb, KERN_ERR, 5456 "Failed to create ea_block_cache"); 5457 err = -EINVAL; 5458 goto failed_mount_wq; 5459 } 5460 5461 if (ext4_has_feature_ea_inode(sb)) { 5462 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5463 if (!sbi->s_ea_inode_cache) { 5464 ext4_msg(sb, KERN_ERR, 5465 "Failed to create ea_inode_cache"); 5466 err = -EINVAL; 5467 goto failed_mount_wq; 5468 } 5469 } 5470 } 5471 5472 /* 5473 * Get the # of file system overhead blocks from the 5474 * superblock if present. 5475 */ 5476 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5477 /* ignore the precalculated value if it is ridiculous */ 5478 if (sbi->s_overhead > ext4_blocks_count(es)) 5479 sbi->s_overhead = 0; 5480 /* 5481 * If the bigalloc feature is not enabled recalculating the 5482 * overhead doesn't take long, so we might as well just redo 5483 * it to make sure we are using the correct value. 5484 */ 5485 if (!ext4_has_feature_bigalloc(sb)) 5486 sbi->s_overhead = 0; 5487 if (sbi->s_overhead == 0) { 5488 err = ext4_calculate_overhead(sb); 5489 if (err) 5490 goto failed_mount_wq; 5491 } 5492 5493 /* 5494 * The maximum number of concurrent works can be high and 5495 * concurrency isn't really necessary. Limit it to 1. 5496 */ 5497 EXT4_SB(sb)->rsv_conversion_wq = 5498 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5499 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5500 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5501 err = -ENOMEM; 5502 goto failed_mount4; 5503 } 5504 5505 /* 5506 * The jbd2_journal_load will have done any necessary log recovery, 5507 * so we can safely mount the rest of the filesystem now. 5508 */ 5509 5510 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5511 if (IS_ERR(root)) { 5512 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5513 err = PTR_ERR(root); 5514 root = NULL; 5515 goto failed_mount4; 5516 } 5517 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5518 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5519 iput(root); 5520 err = -EFSCORRUPTED; 5521 goto failed_mount4; 5522 } 5523 5524 generic_set_sb_d_ops(sb); 5525 sb->s_root = d_make_root(root); 5526 if (!sb->s_root) { 5527 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5528 err = -ENOMEM; 5529 goto failed_mount4; 5530 } 5531 5532 err = ext4_setup_super(sb, es, sb_rdonly(sb)); 5533 if (err == -EROFS) { 5534 sb->s_flags |= SB_RDONLY; 5535 } else if (err) 5536 goto failed_mount4a; 5537 5538 ext4_set_resv_clusters(sb); 5539 5540 if (test_opt(sb, BLOCK_VALIDITY)) { 5541 err = ext4_setup_system_zone(sb); 5542 if (err) { 5543 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5544 "zone (%d)", err); 5545 goto failed_mount4a; 5546 } 5547 } 5548 ext4_fc_replay_cleanup(sb); 5549 5550 ext4_ext_init(sb); 5551 5552 /* 5553 * Enable optimize_scan if number of groups is > threshold. This can be 5554 * turned off by passing "mb_optimize_scan=0". This can also be 5555 * turned on forcefully by passing "mb_optimize_scan=1". 5556 */ 5557 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5558 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5559 set_opt2(sb, MB_OPTIMIZE_SCAN); 5560 else 5561 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5562 } 5563 5564 err = ext4_mb_init(sb); 5565 if (err) { 5566 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5567 err); 5568 goto failed_mount5; 5569 } 5570 5571 /* 5572 * We can only set up the journal commit callback once 5573 * mballoc is initialized 5574 */ 5575 if (sbi->s_journal) 5576 sbi->s_journal->j_commit_callback = 5577 ext4_journal_commit_callback; 5578 5579 err = ext4_percpu_param_init(sbi); 5580 if (err) 5581 goto failed_mount6; 5582 5583 if (ext4_has_feature_flex_bg(sb)) 5584 if (!ext4_fill_flex_info(sb)) { 5585 ext4_msg(sb, KERN_ERR, 5586 "unable to initialize " 5587 "flex_bg meta info!"); 5588 err = -ENOMEM; 5589 goto failed_mount6; 5590 } 5591 5592 err = ext4_register_li_request(sb, first_not_zeroed); 5593 if (err) 5594 goto failed_mount6; 5595 5596 err = ext4_init_orphan_info(sb); 5597 if (err) 5598 goto failed_mount7; 5599 #ifdef CONFIG_QUOTA 5600 /* Enable quota usage during mount. */ 5601 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5602 err = ext4_enable_quotas(sb); 5603 if (err) 5604 goto failed_mount8; 5605 } 5606 #endif /* CONFIG_QUOTA */ 5607 5608 /* 5609 * Save the original bdev mapping's wb_err value which could be 5610 * used to detect the metadata async write error. 5611 */ 5612 errseq_check_and_advance(&sb->s_bdev->bd_mapping->wb_err, 5613 &sbi->s_bdev_wb_err); 5614 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5615 ext4_orphan_cleanup(sb, es); 5616 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5617 /* 5618 * Update the checksum after updating free space/inode counters and 5619 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5620 * checksum in the buffer cache until it is written out and 5621 * e2fsprogs programs trying to open a file system immediately 5622 * after it is mounted can fail. 5623 */ 5624 ext4_superblock_csum_set(sb); 5625 if (needs_recovery) { 5626 ext4_msg(sb, KERN_INFO, "recovery complete"); 5627 err = ext4_mark_recovery_complete(sb, es); 5628 if (err) 5629 goto failed_mount9; 5630 } 5631 5632 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) { 5633 ext4_msg(sb, KERN_WARNING, 5634 "mounting with \"discard\" option, but the device does not support discard"); 5635 clear_opt(sb, DISCARD); 5636 } 5637 5638 if (es->s_error_count) 5639 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5640 5641 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5642 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5643 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5644 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5645 atomic_set(&sbi->s_warning_count, 0); 5646 atomic_set(&sbi->s_msg_count, 0); 5647 5648 /* Register sysfs after all initializations are complete. */ 5649 err = ext4_register_sysfs(sb); 5650 if (err) 5651 goto failed_mount9; 5652 5653 return 0; 5654 5655 failed_mount9: 5656 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 5657 failed_mount8: __maybe_unused 5658 ext4_release_orphan_info(sb); 5659 failed_mount7: 5660 ext4_unregister_li_request(sb); 5661 failed_mount6: 5662 ext4_mb_release(sb); 5663 ext4_flex_groups_free(sbi); 5664 ext4_percpu_param_destroy(sbi); 5665 failed_mount5: 5666 ext4_ext_release(sb); 5667 ext4_release_system_zone(sb); 5668 failed_mount4a: 5669 dput(sb->s_root); 5670 sb->s_root = NULL; 5671 failed_mount4: 5672 ext4_msg(sb, KERN_ERR, "mount failed"); 5673 if (EXT4_SB(sb)->rsv_conversion_wq) 5674 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5675 failed_mount_wq: 5676 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5677 sbi->s_ea_inode_cache = NULL; 5678 5679 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5680 sbi->s_ea_block_cache = NULL; 5681 5682 if (sbi->s_journal) { 5683 ext4_journal_destroy(sbi, sbi->s_journal); 5684 } 5685 failed_mount3a: 5686 ext4_es_unregister_shrinker(sbi); 5687 failed_mount3: 5688 /* flush s_sb_upd_work before sbi destroy */ 5689 flush_work(&sbi->s_sb_upd_work); 5690 ext4_stop_mmpd(sbi); 5691 timer_delete_sync(&sbi->s_err_report); 5692 ext4_group_desc_free(sbi); 5693 failed_mount: 5694 #if IS_ENABLED(CONFIG_UNICODE) 5695 utf8_unload(sb->s_encoding); 5696 #endif 5697 5698 #ifdef CONFIG_QUOTA 5699 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++) 5700 kfree(get_qf_name(sb, sbi, i)); 5701 #endif 5702 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5703 brelse(sbi->s_sbh); 5704 if (sbi->s_journal_bdev_file) { 5705 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 5706 bdev_fput(sbi->s_journal_bdev_file); 5707 } 5708 out_fail: 5709 invalidate_bdev(sb->s_bdev); 5710 sb->s_fs_info = NULL; 5711 return err; 5712 } 5713 5714 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5715 { 5716 struct ext4_fs_context *ctx = fc->fs_private; 5717 struct ext4_sb_info *sbi; 5718 const char *descr; 5719 int ret; 5720 5721 sbi = ext4_alloc_sbi(sb); 5722 if (!sbi) 5723 return -ENOMEM; 5724 5725 fc->s_fs_info = sbi; 5726 5727 /* Cleanup superblock name */ 5728 strreplace(sb->s_id, '/', '!'); 5729 5730 sbi->s_sb_block = 1; /* Default super block location */ 5731 if (ctx->spec & EXT4_SPEC_s_sb_block) 5732 sbi->s_sb_block = ctx->s_sb_block; 5733 5734 ret = __ext4_fill_super(fc, sb); 5735 if (ret < 0) 5736 goto free_sbi; 5737 5738 if (sbi->s_journal) { 5739 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5740 descr = " journalled data mode"; 5741 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5742 descr = " ordered data mode"; 5743 else 5744 descr = " writeback data mode"; 5745 } else 5746 descr = "out journal"; 5747 5748 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5749 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. " 5750 "Quota mode: %s.", &sb->s_uuid, 5751 sb_rdonly(sb) ? "ro" : "r/w", descr, 5752 ext4_quota_mode(sb)); 5753 5754 /* Update the s_overhead_clusters if necessary */ 5755 ext4_update_overhead(sb, false); 5756 return 0; 5757 5758 free_sbi: 5759 ext4_free_sbi(sbi); 5760 fc->s_fs_info = NULL; 5761 return ret; 5762 } 5763 5764 static int ext4_get_tree(struct fs_context *fc) 5765 { 5766 return get_tree_bdev(fc, ext4_fill_super); 5767 } 5768 5769 /* 5770 * Setup any per-fs journal parameters now. We'll do this both on 5771 * initial mount, once the journal has been initialised but before we've 5772 * done any recovery; and again on any subsequent remount. 5773 */ 5774 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5775 { 5776 struct ext4_sb_info *sbi = EXT4_SB(sb); 5777 5778 journal->j_commit_interval = sbi->s_commit_interval; 5779 journal->j_min_batch_time = sbi->s_min_batch_time; 5780 journal->j_max_batch_time = sbi->s_max_batch_time; 5781 ext4_fc_init(sb, journal); 5782 5783 write_lock(&journal->j_state_lock); 5784 if (test_opt(sb, BARRIER)) 5785 journal->j_flags |= JBD2_BARRIER; 5786 else 5787 journal->j_flags &= ~JBD2_BARRIER; 5788 /* 5789 * Always enable journal cycle record option, letting the journal 5790 * records log transactions continuously between each mount. 5791 */ 5792 journal->j_flags |= JBD2_CYCLE_RECORD; 5793 write_unlock(&journal->j_state_lock); 5794 } 5795 5796 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5797 unsigned int journal_inum) 5798 { 5799 struct inode *journal_inode; 5800 5801 /* 5802 * Test for the existence of a valid inode on disk. Bad things 5803 * happen if we iget() an unused inode, as the subsequent iput() 5804 * will try to delete it. 5805 */ 5806 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5807 if (IS_ERR(journal_inode)) { 5808 ext4_msg(sb, KERN_ERR, "no journal found"); 5809 return ERR_CAST(journal_inode); 5810 } 5811 if (!journal_inode->i_nlink) { 5812 make_bad_inode(journal_inode); 5813 iput(journal_inode); 5814 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5815 return ERR_PTR(-EFSCORRUPTED); 5816 } 5817 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) { 5818 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5819 iput(journal_inode); 5820 return ERR_PTR(-EFSCORRUPTED); 5821 } 5822 5823 ext4_debug("Journal inode found at %p: %lld bytes\n", 5824 journal_inode, journal_inode->i_size); 5825 return journal_inode; 5826 } 5827 5828 static int ext4_journal_bmap(journal_t *journal, sector_t *block) 5829 { 5830 struct ext4_map_blocks map; 5831 int ret; 5832 5833 if (journal->j_inode == NULL) 5834 return 0; 5835 5836 map.m_lblk = *block; 5837 map.m_len = 1; 5838 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0); 5839 if (ret <= 0) { 5840 ext4_msg(journal->j_inode->i_sb, KERN_CRIT, 5841 "journal bmap failed: block %llu ret %d\n", 5842 *block, ret); 5843 jbd2_journal_abort(journal, ret ? ret : -EIO); 5844 return ret; 5845 } 5846 *block = map.m_pblk; 5847 return 0; 5848 } 5849 5850 static journal_t *ext4_open_inode_journal(struct super_block *sb, 5851 unsigned int journal_inum) 5852 { 5853 struct inode *journal_inode; 5854 journal_t *journal; 5855 5856 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5857 if (IS_ERR(journal_inode)) 5858 return ERR_CAST(journal_inode); 5859 5860 journal = jbd2_journal_init_inode(journal_inode); 5861 if (IS_ERR(journal)) { 5862 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5863 iput(journal_inode); 5864 return ERR_CAST(journal); 5865 } 5866 journal->j_private = sb; 5867 journal->j_bmap = ext4_journal_bmap; 5868 ext4_init_journal_params(sb, journal); 5869 return journal; 5870 } 5871 5872 static struct file *ext4_get_journal_blkdev(struct super_block *sb, 5873 dev_t j_dev, ext4_fsblk_t *j_start, 5874 ext4_fsblk_t *j_len) 5875 { 5876 struct buffer_head *bh; 5877 struct block_device *bdev; 5878 struct file *bdev_file; 5879 int hblock, blocksize; 5880 ext4_fsblk_t sb_block; 5881 unsigned long offset; 5882 struct ext4_super_block *es; 5883 int errno; 5884 5885 bdev_file = bdev_file_open_by_dev(j_dev, 5886 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES, 5887 sb, &fs_holder_ops); 5888 if (IS_ERR(bdev_file)) { 5889 ext4_msg(sb, KERN_ERR, 5890 "failed to open journal device unknown-block(%u,%u) %ld", 5891 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev_file)); 5892 return bdev_file; 5893 } 5894 5895 bdev = file_bdev(bdev_file); 5896 blocksize = sb->s_blocksize; 5897 hblock = bdev_logical_block_size(bdev); 5898 if (blocksize < hblock) { 5899 ext4_msg(sb, KERN_ERR, 5900 "blocksize too small for journal device"); 5901 errno = -EINVAL; 5902 goto out_bdev; 5903 } 5904 5905 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5906 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5907 set_blocksize(bdev_file, blocksize); 5908 bh = __bread(bdev, sb_block, blocksize); 5909 if (!bh) { 5910 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5911 "external journal"); 5912 errno = -EINVAL; 5913 goto out_bdev; 5914 } 5915 5916 es = (struct ext4_super_block *) (bh->b_data + offset); 5917 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5918 !(le32_to_cpu(es->s_feature_incompat) & 5919 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5920 ext4_msg(sb, KERN_ERR, "external journal has bad superblock"); 5921 errno = -EFSCORRUPTED; 5922 goto out_bh; 5923 } 5924 5925 if ((le32_to_cpu(es->s_feature_ro_compat) & 5926 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5927 es->s_checksum != ext4_superblock_csum(es)) { 5928 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock"); 5929 errno = -EFSCORRUPTED; 5930 goto out_bh; 5931 } 5932 5933 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5934 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5935 errno = -EFSCORRUPTED; 5936 goto out_bh; 5937 } 5938 5939 *j_start = sb_block + 1; 5940 *j_len = ext4_blocks_count(es); 5941 brelse(bh); 5942 return bdev_file; 5943 5944 out_bh: 5945 brelse(bh); 5946 out_bdev: 5947 bdev_fput(bdev_file); 5948 return ERR_PTR(errno); 5949 } 5950 5951 static journal_t *ext4_open_dev_journal(struct super_block *sb, 5952 dev_t j_dev) 5953 { 5954 journal_t *journal; 5955 ext4_fsblk_t j_start; 5956 ext4_fsblk_t j_len; 5957 struct file *bdev_file; 5958 int errno = 0; 5959 5960 bdev_file = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len); 5961 if (IS_ERR(bdev_file)) 5962 return ERR_CAST(bdev_file); 5963 5964 journal = jbd2_journal_init_dev(file_bdev(bdev_file), sb->s_bdev, j_start, 5965 j_len, sb->s_blocksize); 5966 if (IS_ERR(journal)) { 5967 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5968 errno = PTR_ERR(journal); 5969 goto out_bdev; 5970 } 5971 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5972 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5973 "user (unsupported) - %d", 5974 be32_to_cpu(journal->j_superblock->s_nr_users)); 5975 errno = -EINVAL; 5976 goto out_journal; 5977 } 5978 journal->j_private = sb; 5979 EXT4_SB(sb)->s_journal_bdev_file = bdev_file; 5980 ext4_init_journal_params(sb, journal); 5981 return journal; 5982 5983 out_journal: 5984 ext4_journal_destroy(EXT4_SB(sb), journal); 5985 out_bdev: 5986 bdev_fput(bdev_file); 5987 return ERR_PTR(errno); 5988 } 5989 5990 static int ext4_load_journal(struct super_block *sb, 5991 struct ext4_super_block *es, 5992 unsigned long journal_devnum) 5993 { 5994 journal_t *journal; 5995 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5996 dev_t journal_dev; 5997 int err = 0; 5998 int really_read_only; 5999 int journal_dev_ro; 6000 6001 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 6002 return -EFSCORRUPTED; 6003 6004 if (journal_devnum && 6005 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6006 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 6007 "numbers have changed"); 6008 journal_dev = new_decode_dev(journal_devnum); 6009 } else 6010 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 6011 6012 if (journal_inum && journal_dev) { 6013 ext4_msg(sb, KERN_ERR, 6014 "filesystem has both journal inode and journal device!"); 6015 return -EINVAL; 6016 } 6017 6018 if (journal_inum) { 6019 journal = ext4_open_inode_journal(sb, journal_inum); 6020 if (IS_ERR(journal)) 6021 return PTR_ERR(journal); 6022 } else { 6023 journal = ext4_open_dev_journal(sb, journal_dev); 6024 if (IS_ERR(journal)) 6025 return PTR_ERR(journal); 6026 } 6027 6028 journal_dev_ro = bdev_read_only(journal->j_dev); 6029 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 6030 6031 if (journal_dev_ro && !sb_rdonly(sb)) { 6032 ext4_msg(sb, KERN_ERR, 6033 "journal device read-only, try mounting with '-o ro'"); 6034 err = -EROFS; 6035 goto err_out; 6036 } 6037 6038 /* 6039 * Are we loading a blank journal or performing recovery after a 6040 * crash? For recovery, we need to check in advance whether we 6041 * can get read-write access to the device. 6042 */ 6043 if (ext4_has_feature_journal_needs_recovery(sb)) { 6044 if (sb_rdonly(sb)) { 6045 ext4_msg(sb, KERN_INFO, "INFO: recovery " 6046 "required on readonly filesystem"); 6047 if (really_read_only) { 6048 ext4_msg(sb, KERN_ERR, "write access " 6049 "unavailable, cannot proceed " 6050 "(try mounting with noload)"); 6051 err = -EROFS; 6052 goto err_out; 6053 } 6054 ext4_msg(sb, KERN_INFO, "write access will " 6055 "be enabled during recovery"); 6056 } 6057 } 6058 6059 if (!(journal->j_flags & JBD2_BARRIER)) 6060 ext4_msg(sb, KERN_INFO, "barriers disabled"); 6061 6062 if (!ext4_has_feature_journal_needs_recovery(sb)) 6063 err = jbd2_journal_wipe(journal, !really_read_only); 6064 if (!err) { 6065 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 6066 __le16 orig_state; 6067 bool changed = false; 6068 6069 if (save) 6070 memcpy(save, ((char *) es) + 6071 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 6072 err = jbd2_journal_load(journal); 6073 if (save && memcmp(((char *) es) + EXT4_S_ERR_START, 6074 save, EXT4_S_ERR_LEN)) { 6075 memcpy(((char *) es) + EXT4_S_ERR_START, 6076 save, EXT4_S_ERR_LEN); 6077 changed = true; 6078 } 6079 kfree(save); 6080 orig_state = es->s_state; 6081 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state & 6082 EXT4_ERROR_FS); 6083 if (orig_state != es->s_state) 6084 changed = true; 6085 /* Write out restored error information to the superblock */ 6086 if (changed && !really_read_only) { 6087 int err2; 6088 err2 = ext4_commit_super(sb); 6089 err = err ? : err2; 6090 } 6091 } 6092 6093 if (err) { 6094 ext4_msg(sb, KERN_ERR, "error loading journal"); 6095 goto err_out; 6096 } 6097 6098 EXT4_SB(sb)->s_journal = journal; 6099 err = ext4_clear_journal_err(sb, es); 6100 if (err) { 6101 ext4_journal_destroy(EXT4_SB(sb), journal); 6102 return err; 6103 } 6104 6105 if (!really_read_only && journal_devnum && 6106 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6107 es->s_journal_dev = cpu_to_le32(journal_devnum); 6108 ext4_commit_super(sb); 6109 } 6110 if (!really_read_only && journal_inum && 6111 journal_inum != le32_to_cpu(es->s_journal_inum)) { 6112 es->s_journal_inum = cpu_to_le32(journal_inum); 6113 ext4_commit_super(sb); 6114 } 6115 6116 return 0; 6117 6118 err_out: 6119 ext4_journal_destroy(EXT4_SB(sb), journal); 6120 return err; 6121 } 6122 6123 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 6124 static void ext4_update_super(struct super_block *sb) 6125 { 6126 struct ext4_sb_info *sbi = EXT4_SB(sb); 6127 struct ext4_super_block *es = sbi->s_es; 6128 struct buffer_head *sbh = sbi->s_sbh; 6129 6130 lock_buffer(sbh); 6131 /* 6132 * If the file system is mounted read-only, don't update the 6133 * superblock write time. This avoids updating the superblock 6134 * write time when we are mounting the root file system 6135 * read/only but we need to replay the journal; at that point, 6136 * for people who are east of GMT and who make their clock 6137 * tick in localtime for Windows bug-for-bug compatibility, 6138 * the clock is set in the future, and this will cause e2fsck 6139 * to complain and force a full file system check. 6140 */ 6141 if (!sb_rdonly(sb)) 6142 ext4_update_tstamp(es, s_wtime); 6143 es->s_kbytes_written = 6144 cpu_to_le64(sbi->s_kbytes_written + 6145 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 6146 sbi->s_sectors_written_start) >> 1)); 6147 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 6148 ext4_free_blocks_count_set(es, 6149 EXT4_C2B(sbi, percpu_counter_sum_positive( 6150 &sbi->s_freeclusters_counter))); 6151 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 6152 es->s_free_inodes_count = 6153 cpu_to_le32(percpu_counter_sum_positive( 6154 &sbi->s_freeinodes_counter)); 6155 /* Copy error information to the on-disk superblock */ 6156 spin_lock(&sbi->s_error_lock); 6157 if (sbi->s_add_error_count > 0) { 6158 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6159 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6160 __ext4_update_tstamp(&es->s_first_error_time, 6161 &es->s_first_error_time_hi, 6162 sbi->s_first_error_time); 6163 strtomem_pad(es->s_first_error_func, 6164 sbi->s_first_error_func, 0); 6165 es->s_first_error_line = 6166 cpu_to_le32(sbi->s_first_error_line); 6167 es->s_first_error_ino = 6168 cpu_to_le32(sbi->s_first_error_ino); 6169 es->s_first_error_block = 6170 cpu_to_le64(sbi->s_first_error_block); 6171 es->s_first_error_errcode = 6172 ext4_errno_to_code(sbi->s_first_error_code); 6173 } 6174 __ext4_update_tstamp(&es->s_last_error_time, 6175 &es->s_last_error_time_hi, 6176 sbi->s_last_error_time); 6177 strtomem_pad(es->s_last_error_func, sbi->s_last_error_func, 0); 6178 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6179 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6180 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6181 es->s_last_error_errcode = 6182 ext4_errno_to_code(sbi->s_last_error_code); 6183 /* 6184 * Start the daily error reporting function if it hasn't been 6185 * started already 6186 */ 6187 if (!es->s_error_count) 6188 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6189 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6190 sbi->s_add_error_count = 0; 6191 } 6192 spin_unlock(&sbi->s_error_lock); 6193 6194 ext4_superblock_csum_set(sb); 6195 unlock_buffer(sbh); 6196 } 6197 6198 static int ext4_commit_super(struct super_block *sb) 6199 { 6200 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6201 6202 if (!sbh) 6203 return -EINVAL; 6204 6205 ext4_update_super(sb); 6206 6207 lock_buffer(sbh); 6208 /* Buffer got discarded which means block device got invalidated */ 6209 if (!buffer_mapped(sbh)) { 6210 unlock_buffer(sbh); 6211 return -EIO; 6212 } 6213 6214 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6215 /* 6216 * Oh, dear. A previous attempt to write the 6217 * superblock failed. This could happen because the 6218 * USB device was yanked out. Or it could happen to 6219 * be a transient write error and maybe the block will 6220 * be remapped. Nothing we can do but to retry the 6221 * write and hope for the best. 6222 */ 6223 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6224 "superblock detected"); 6225 clear_buffer_write_io_error(sbh); 6226 set_buffer_uptodate(sbh); 6227 } 6228 get_bh(sbh); 6229 /* Clear potential dirty bit if it was journalled update */ 6230 clear_buffer_dirty(sbh); 6231 sbh->b_end_io = end_buffer_write_sync; 6232 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6233 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6234 wait_on_buffer(sbh); 6235 if (buffer_write_io_error(sbh)) { 6236 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6237 "superblock"); 6238 clear_buffer_write_io_error(sbh); 6239 set_buffer_uptodate(sbh); 6240 return -EIO; 6241 } 6242 return 0; 6243 } 6244 6245 /* 6246 * Have we just finished recovery? If so, and if we are mounting (or 6247 * remounting) the filesystem readonly, then we will end up with a 6248 * consistent fs on disk. Record that fact. 6249 */ 6250 static int ext4_mark_recovery_complete(struct super_block *sb, 6251 struct ext4_super_block *es) 6252 { 6253 int err; 6254 journal_t *journal = EXT4_SB(sb)->s_journal; 6255 6256 if (!ext4_has_feature_journal(sb)) { 6257 if (journal != NULL) { 6258 ext4_error(sb, "Journal got removed while the fs was " 6259 "mounted!"); 6260 return -EFSCORRUPTED; 6261 } 6262 return 0; 6263 } 6264 jbd2_journal_lock_updates(journal); 6265 err = jbd2_journal_flush(journal, 0); 6266 if (err < 0) 6267 goto out; 6268 6269 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6270 ext4_has_feature_orphan_present(sb))) { 6271 if (!ext4_orphan_file_empty(sb)) { 6272 ext4_error(sb, "Orphan file not empty on read-only fs."); 6273 err = -EFSCORRUPTED; 6274 goto out; 6275 } 6276 ext4_clear_feature_journal_needs_recovery(sb); 6277 ext4_clear_feature_orphan_present(sb); 6278 ext4_commit_super(sb); 6279 } 6280 out: 6281 jbd2_journal_unlock_updates(journal); 6282 return err; 6283 } 6284 6285 /* 6286 * If we are mounting (or read-write remounting) a filesystem whose journal 6287 * has recorded an error from a previous lifetime, move that error to the 6288 * main filesystem now. 6289 */ 6290 static int ext4_clear_journal_err(struct super_block *sb, 6291 struct ext4_super_block *es) 6292 { 6293 journal_t *journal; 6294 int j_errno; 6295 const char *errstr; 6296 6297 if (!ext4_has_feature_journal(sb)) { 6298 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6299 return -EFSCORRUPTED; 6300 } 6301 6302 journal = EXT4_SB(sb)->s_journal; 6303 6304 /* 6305 * Now check for any error status which may have been recorded in the 6306 * journal by a prior ext4_error() or ext4_abort() 6307 */ 6308 6309 j_errno = jbd2_journal_errno(journal); 6310 if (j_errno) { 6311 char nbuf[16]; 6312 6313 errstr = ext4_decode_error(sb, j_errno, nbuf); 6314 ext4_warning(sb, "Filesystem error recorded " 6315 "from previous mount: %s", errstr); 6316 6317 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6318 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6319 j_errno = ext4_commit_super(sb); 6320 if (j_errno) 6321 return j_errno; 6322 ext4_warning(sb, "Marked fs in need of filesystem check."); 6323 6324 jbd2_journal_clear_err(journal); 6325 jbd2_journal_update_sb_errno(journal); 6326 } 6327 return 0; 6328 } 6329 6330 /* 6331 * Force the running and committing transactions to commit, 6332 * and wait on the commit. 6333 */ 6334 int ext4_force_commit(struct super_block *sb) 6335 { 6336 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal); 6337 } 6338 6339 static int ext4_sync_fs(struct super_block *sb, int wait) 6340 { 6341 int ret = 0; 6342 tid_t target; 6343 bool needs_barrier = false; 6344 struct ext4_sb_info *sbi = EXT4_SB(sb); 6345 6346 ret = ext4_emergency_state(sb); 6347 if (unlikely(ret)) 6348 return ret; 6349 6350 trace_ext4_sync_fs(sb, wait); 6351 flush_workqueue(sbi->rsv_conversion_wq); 6352 /* 6353 * Writeback quota in non-journalled quota case - journalled quota has 6354 * no dirty dquots 6355 */ 6356 dquot_writeback_dquots(sb, -1); 6357 /* 6358 * Data writeback is possible w/o journal transaction, so barrier must 6359 * being sent at the end of the function. But we can skip it if 6360 * transaction_commit will do it for us. 6361 */ 6362 if (sbi->s_journal) { 6363 target = jbd2_get_latest_transaction(sbi->s_journal); 6364 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6365 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6366 needs_barrier = true; 6367 6368 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6369 if (wait) 6370 ret = jbd2_log_wait_commit(sbi->s_journal, 6371 target); 6372 } 6373 } else if (wait && test_opt(sb, BARRIER)) 6374 needs_barrier = true; 6375 if (needs_barrier) { 6376 int err; 6377 err = blkdev_issue_flush(sb->s_bdev); 6378 if (!ret) 6379 ret = err; 6380 } 6381 6382 return ret; 6383 } 6384 6385 /* 6386 * LVM calls this function before a (read-only) snapshot is created. This 6387 * gives us a chance to flush the journal completely and mark the fs clean. 6388 * 6389 * Note that only this function cannot bring a filesystem to be in a clean 6390 * state independently. It relies on upper layer to stop all data & metadata 6391 * modifications. 6392 */ 6393 static int ext4_freeze(struct super_block *sb) 6394 { 6395 int error = 0; 6396 journal_t *journal = EXT4_SB(sb)->s_journal; 6397 6398 if (journal) { 6399 /* Now we set up the journal barrier. */ 6400 jbd2_journal_lock_updates(journal); 6401 6402 /* 6403 * Don't clear the needs_recovery flag if we failed to 6404 * flush the journal. 6405 */ 6406 error = jbd2_journal_flush(journal, 0); 6407 if (error < 0) 6408 goto out; 6409 6410 /* Journal blocked and flushed, clear needs_recovery flag. */ 6411 ext4_clear_feature_journal_needs_recovery(sb); 6412 if (ext4_orphan_file_empty(sb)) 6413 ext4_clear_feature_orphan_present(sb); 6414 } 6415 6416 error = ext4_commit_super(sb); 6417 out: 6418 if (journal) 6419 /* we rely on upper layer to stop further updates */ 6420 jbd2_journal_unlock_updates(journal); 6421 return error; 6422 } 6423 6424 /* 6425 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6426 * flag here, even though the filesystem is not technically dirty yet. 6427 */ 6428 static int ext4_unfreeze(struct super_block *sb) 6429 { 6430 if (ext4_emergency_state(sb)) 6431 return 0; 6432 6433 if (EXT4_SB(sb)->s_journal) { 6434 /* Reset the needs_recovery flag before the fs is unlocked. */ 6435 ext4_set_feature_journal_needs_recovery(sb); 6436 if (ext4_has_feature_orphan_file(sb)) 6437 ext4_set_feature_orphan_present(sb); 6438 } 6439 6440 ext4_commit_super(sb); 6441 return 0; 6442 } 6443 6444 /* 6445 * Structure to save mount options for ext4_remount's benefit 6446 */ 6447 struct ext4_mount_options { 6448 unsigned long s_mount_opt; 6449 unsigned long s_mount_opt2; 6450 kuid_t s_resuid; 6451 kgid_t s_resgid; 6452 unsigned long s_commit_interval; 6453 u32 s_min_batch_time, s_max_batch_time; 6454 #ifdef CONFIG_QUOTA 6455 int s_jquota_fmt; 6456 char *s_qf_names[EXT4_MAXQUOTAS]; 6457 #endif 6458 }; 6459 6460 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6461 { 6462 struct ext4_fs_context *ctx = fc->fs_private; 6463 struct ext4_super_block *es; 6464 struct ext4_sb_info *sbi = EXT4_SB(sb); 6465 unsigned long old_sb_flags; 6466 struct ext4_mount_options old_opts; 6467 ext4_group_t g; 6468 int err = 0; 6469 int alloc_ctx; 6470 #ifdef CONFIG_QUOTA 6471 int enable_quota = 0; 6472 int i, j; 6473 char *to_free[EXT4_MAXQUOTAS]; 6474 #endif 6475 6476 6477 /* Store the original options */ 6478 old_sb_flags = sb->s_flags; 6479 old_opts.s_mount_opt = sbi->s_mount_opt; 6480 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6481 old_opts.s_resuid = sbi->s_resuid; 6482 old_opts.s_resgid = sbi->s_resgid; 6483 old_opts.s_commit_interval = sbi->s_commit_interval; 6484 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6485 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6486 #ifdef CONFIG_QUOTA 6487 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6488 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6489 if (sbi->s_qf_names[i]) { 6490 char *qf_name = get_qf_name(sb, sbi, i); 6491 6492 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6493 if (!old_opts.s_qf_names[i]) { 6494 for (j = 0; j < i; j++) 6495 kfree(old_opts.s_qf_names[j]); 6496 return -ENOMEM; 6497 } 6498 } else 6499 old_opts.s_qf_names[i] = NULL; 6500 #endif 6501 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6502 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6503 ctx->journal_ioprio = 6504 sbi->s_journal->j_task->io_context->ioprio; 6505 else 6506 ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO; 6507 6508 } 6509 6510 if ((ctx->spec & EXT4_SPEC_s_stripe) && 6511 ext4_is_stripe_incompatible(sb, ctx->s_stripe)) { 6512 ext4_msg(sb, KERN_WARNING, 6513 "stripe (%lu) is not aligned with cluster size (%u), " 6514 "stripe is disabled", 6515 ctx->s_stripe, sbi->s_cluster_ratio); 6516 ctx->s_stripe = 0; 6517 } 6518 6519 /* 6520 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause 6521 * two calls to ext4_should_dioread_nolock() to return inconsistent 6522 * values, triggering WARN_ON in ext4_add_complete_io(). we grab 6523 * here s_writepages_rwsem to avoid race between writepages ops and 6524 * remount. 6525 */ 6526 alloc_ctx = ext4_writepages_down_write(sb); 6527 ext4_apply_options(fc, sb); 6528 ext4_writepages_up_write(sb, alloc_ctx); 6529 6530 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6531 test_opt(sb, JOURNAL_CHECKSUM)) { 6532 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6533 "during remount not supported; ignoring"); 6534 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6535 } 6536 6537 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6538 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6539 ext4_msg(sb, KERN_ERR, "can't mount with " 6540 "both data=journal and delalloc"); 6541 err = -EINVAL; 6542 goto restore_opts; 6543 } 6544 if (test_opt(sb, DIOREAD_NOLOCK)) { 6545 ext4_msg(sb, KERN_ERR, "can't mount with " 6546 "both data=journal and dioread_nolock"); 6547 err = -EINVAL; 6548 goto restore_opts; 6549 } 6550 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6551 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6552 ext4_msg(sb, KERN_ERR, "can't mount with " 6553 "journal_async_commit in data=ordered mode"); 6554 err = -EINVAL; 6555 goto restore_opts; 6556 } 6557 } 6558 6559 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6560 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6561 err = -EINVAL; 6562 goto restore_opts; 6563 } 6564 6565 if ((old_opts.s_mount_opt & EXT4_MOUNT_DELALLOC) && 6566 !test_opt(sb, DELALLOC)) { 6567 ext4_msg(sb, KERN_ERR, "can't disable delalloc during remount"); 6568 err = -EINVAL; 6569 goto restore_opts; 6570 } 6571 6572 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6573 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6574 6575 es = sbi->s_es; 6576 6577 if (sbi->s_journal) { 6578 ext4_init_journal_params(sb, sbi->s_journal); 6579 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6580 } 6581 6582 /* Flush outstanding errors before changing fs state */ 6583 flush_work(&sbi->s_sb_upd_work); 6584 6585 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6586 if (ext4_emergency_state(sb)) { 6587 err = -EROFS; 6588 goto restore_opts; 6589 } 6590 6591 if (fc->sb_flags & SB_RDONLY) { 6592 err = sync_filesystem(sb); 6593 if (err < 0) 6594 goto restore_opts; 6595 err = dquot_suspend(sb, -1); 6596 if (err < 0) 6597 goto restore_opts; 6598 6599 /* 6600 * First of all, the unconditional stuff we have to do 6601 * to disable replay of the journal when we next remount 6602 */ 6603 sb->s_flags |= SB_RDONLY; 6604 6605 /* 6606 * OK, test if we are remounting a valid rw partition 6607 * readonly, and if so set the rdonly flag and then 6608 * mark the partition as valid again. 6609 */ 6610 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6611 (sbi->s_mount_state & EXT4_VALID_FS)) 6612 es->s_state = cpu_to_le16(sbi->s_mount_state); 6613 6614 if (sbi->s_journal) { 6615 /* 6616 * We let remount-ro finish even if marking fs 6617 * as clean failed... 6618 */ 6619 ext4_mark_recovery_complete(sb, es); 6620 } 6621 } else { 6622 /* Make sure we can mount this feature set readwrite */ 6623 if (ext4_has_feature_readonly(sb) || 6624 !ext4_feature_set_ok(sb, 0)) { 6625 err = -EROFS; 6626 goto restore_opts; 6627 } 6628 /* 6629 * Make sure the group descriptor checksums 6630 * are sane. If they aren't, refuse to remount r/w. 6631 */ 6632 for (g = 0; g < sbi->s_groups_count; g++) { 6633 struct ext4_group_desc *gdp = 6634 ext4_get_group_desc(sb, g, NULL); 6635 6636 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6637 ext4_msg(sb, KERN_ERR, 6638 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6639 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6640 le16_to_cpu(gdp->bg_checksum)); 6641 err = -EFSBADCRC; 6642 goto restore_opts; 6643 } 6644 } 6645 6646 /* 6647 * If we have an unprocessed orphan list hanging 6648 * around from a previously readonly bdev mount, 6649 * require a full umount/remount for now. 6650 */ 6651 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6652 ext4_msg(sb, KERN_WARNING, "Couldn't " 6653 "remount RDWR because of unprocessed " 6654 "orphan inode list. Please " 6655 "umount/remount instead"); 6656 err = -EINVAL; 6657 goto restore_opts; 6658 } 6659 6660 /* 6661 * Mounting a RDONLY partition read-write, so reread 6662 * and store the current valid flag. (It may have 6663 * been changed by e2fsck since we originally mounted 6664 * the partition.) 6665 */ 6666 if (sbi->s_journal) { 6667 err = ext4_clear_journal_err(sb, es); 6668 if (err) 6669 goto restore_opts; 6670 } 6671 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6672 ~EXT4_FC_REPLAY); 6673 6674 err = ext4_setup_super(sb, es, 0); 6675 if (err) 6676 goto restore_opts; 6677 6678 sb->s_flags &= ~SB_RDONLY; 6679 if (ext4_has_feature_mmp(sb)) { 6680 err = ext4_multi_mount_protect(sb, 6681 le64_to_cpu(es->s_mmp_block)); 6682 if (err) 6683 goto restore_opts; 6684 } 6685 #ifdef CONFIG_QUOTA 6686 enable_quota = 1; 6687 #endif 6688 } 6689 } 6690 6691 /* 6692 * Handle creation of system zone data early because it can fail. 6693 * Releasing of existing data is done when we are sure remount will 6694 * succeed. 6695 */ 6696 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6697 err = ext4_setup_system_zone(sb); 6698 if (err) 6699 goto restore_opts; 6700 } 6701 6702 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6703 err = ext4_commit_super(sb); 6704 if (err) 6705 goto restore_opts; 6706 } 6707 6708 #ifdef CONFIG_QUOTA 6709 if (enable_quota) { 6710 if (sb_any_quota_suspended(sb)) 6711 dquot_resume(sb, -1); 6712 else if (ext4_has_feature_quota(sb)) { 6713 err = ext4_enable_quotas(sb); 6714 if (err) 6715 goto restore_opts; 6716 } 6717 } 6718 /* Release old quota file names */ 6719 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6720 kfree(old_opts.s_qf_names[i]); 6721 #endif 6722 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6723 ext4_release_system_zone(sb); 6724 6725 /* 6726 * Reinitialize lazy itable initialization thread based on 6727 * current settings 6728 */ 6729 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6730 ext4_unregister_li_request(sb); 6731 else { 6732 ext4_group_t first_not_zeroed; 6733 first_not_zeroed = ext4_has_uninit_itable(sb); 6734 ext4_register_li_request(sb, first_not_zeroed); 6735 } 6736 6737 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6738 ext4_stop_mmpd(sbi); 6739 6740 /* 6741 * Handle aborting the filesystem as the last thing during remount to 6742 * avoid obsure errors during remount when some option changes fail to 6743 * apply due to shutdown filesystem. 6744 */ 6745 if (test_opt2(sb, ABORT)) 6746 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6747 6748 return 0; 6749 6750 restore_opts: 6751 /* 6752 * If there was a failing r/w to ro transition, we may need to 6753 * re-enable quota 6754 */ 6755 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) && 6756 sb_any_quota_suspended(sb)) 6757 dquot_resume(sb, -1); 6758 6759 alloc_ctx = ext4_writepages_down_write(sb); 6760 sb->s_flags = old_sb_flags; 6761 sbi->s_mount_opt = old_opts.s_mount_opt; 6762 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6763 sbi->s_resuid = old_opts.s_resuid; 6764 sbi->s_resgid = old_opts.s_resgid; 6765 sbi->s_commit_interval = old_opts.s_commit_interval; 6766 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6767 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6768 ext4_writepages_up_write(sb, alloc_ctx); 6769 6770 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6771 ext4_release_system_zone(sb); 6772 #ifdef CONFIG_QUOTA 6773 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6774 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6775 to_free[i] = get_qf_name(sb, sbi, i); 6776 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6777 } 6778 synchronize_rcu(); 6779 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6780 kfree(to_free[i]); 6781 #endif 6782 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6783 ext4_stop_mmpd(sbi); 6784 return err; 6785 } 6786 6787 static int ext4_reconfigure(struct fs_context *fc) 6788 { 6789 struct super_block *sb = fc->root->d_sb; 6790 int ret; 6791 bool old_ro = sb_rdonly(sb); 6792 6793 fc->s_fs_info = EXT4_SB(sb); 6794 6795 ret = ext4_check_opt_consistency(fc, sb); 6796 if (ret < 0) 6797 return ret; 6798 6799 ret = __ext4_remount(fc, sb); 6800 if (ret < 0) 6801 return ret; 6802 6803 ext4_msg(sb, KERN_INFO, "re-mounted %pU%s.", 6804 &sb->s_uuid, 6805 (old_ro != sb_rdonly(sb)) ? (sb_rdonly(sb) ? " ro" : " r/w") : ""); 6806 6807 return 0; 6808 } 6809 6810 #ifdef CONFIG_QUOTA 6811 static int ext4_statfs_project(struct super_block *sb, 6812 kprojid_t projid, struct kstatfs *buf) 6813 { 6814 struct kqid qid; 6815 struct dquot *dquot; 6816 u64 limit; 6817 u64 curblock; 6818 6819 qid = make_kqid_projid(projid); 6820 dquot = dqget(sb, qid); 6821 if (IS_ERR(dquot)) 6822 return PTR_ERR(dquot); 6823 spin_lock(&dquot->dq_dqb_lock); 6824 6825 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6826 dquot->dq_dqb.dqb_bhardlimit); 6827 limit >>= sb->s_blocksize_bits; 6828 6829 if (limit) { 6830 uint64_t remaining = 0; 6831 6832 curblock = (dquot->dq_dqb.dqb_curspace + 6833 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6834 if (limit > curblock) 6835 remaining = limit - curblock; 6836 6837 buf->f_blocks = min(buf->f_blocks, limit); 6838 buf->f_bfree = min(buf->f_bfree, remaining); 6839 buf->f_bavail = min(buf->f_bavail, remaining); 6840 } 6841 6842 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6843 dquot->dq_dqb.dqb_ihardlimit); 6844 if (limit) { 6845 uint64_t remaining = 0; 6846 6847 if (limit > dquot->dq_dqb.dqb_curinodes) 6848 remaining = limit - dquot->dq_dqb.dqb_curinodes; 6849 6850 buf->f_files = min(buf->f_files, limit); 6851 buf->f_ffree = min(buf->f_ffree, remaining); 6852 } 6853 6854 spin_unlock(&dquot->dq_dqb_lock); 6855 dqput(dquot); 6856 return 0; 6857 } 6858 #endif 6859 6860 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6861 { 6862 struct super_block *sb = dentry->d_sb; 6863 struct ext4_sb_info *sbi = EXT4_SB(sb); 6864 struct ext4_super_block *es = sbi->s_es; 6865 ext4_fsblk_t overhead = 0, resv_blocks; 6866 s64 bfree; 6867 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6868 6869 if (!test_opt(sb, MINIX_DF)) 6870 overhead = sbi->s_overhead; 6871 6872 buf->f_type = EXT4_SUPER_MAGIC; 6873 buf->f_bsize = sb->s_blocksize; 6874 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6875 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6876 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6877 /* prevent underflow in case that few free space is available */ 6878 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6879 buf->f_bavail = buf->f_bfree - 6880 (ext4_r_blocks_count(es) + resv_blocks); 6881 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6882 buf->f_bavail = 0; 6883 buf->f_files = le32_to_cpu(es->s_inodes_count); 6884 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6885 buf->f_namelen = EXT4_NAME_LEN; 6886 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6887 6888 #ifdef CONFIG_QUOTA 6889 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6890 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6891 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6892 #endif 6893 return 0; 6894 } 6895 6896 6897 #ifdef CONFIG_QUOTA 6898 6899 /* 6900 * Helper functions so that transaction is started before we acquire dqio_sem 6901 * to keep correct lock ordering of transaction > dqio_sem 6902 */ 6903 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6904 { 6905 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6906 } 6907 6908 static int ext4_write_dquot(struct dquot *dquot) 6909 { 6910 int ret, err; 6911 handle_t *handle; 6912 struct inode *inode; 6913 6914 inode = dquot_to_inode(dquot); 6915 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6916 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6917 if (IS_ERR(handle)) 6918 return PTR_ERR(handle); 6919 ret = dquot_commit(dquot); 6920 if (ret < 0) 6921 ext4_error_err(dquot->dq_sb, -ret, 6922 "Failed to commit dquot type %d", 6923 dquot->dq_id.type); 6924 err = ext4_journal_stop(handle); 6925 if (!ret) 6926 ret = err; 6927 return ret; 6928 } 6929 6930 static int ext4_acquire_dquot(struct dquot *dquot) 6931 { 6932 int ret, err; 6933 handle_t *handle; 6934 6935 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6936 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6937 if (IS_ERR(handle)) 6938 return PTR_ERR(handle); 6939 ret = dquot_acquire(dquot); 6940 if (ret < 0) 6941 ext4_error_err(dquot->dq_sb, -ret, 6942 "Failed to acquire dquot type %d", 6943 dquot->dq_id.type); 6944 err = ext4_journal_stop(handle); 6945 if (!ret) 6946 ret = err; 6947 return ret; 6948 } 6949 6950 static int ext4_release_dquot(struct dquot *dquot) 6951 { 6952 int ret, err; 6953 handle_t *handle; 6954 bool freeze_protected = false; 6955 6956 /* 6957 * Trying to sb_start_intwrite() in a running transaction 6958 * can result in a deadlock. Further, running transactions 6959 * are already protected from freezing. 6960 */ 6961 if (!ext4_journal_current_handle()) { 6962 sb_start_intwrite(dquot->dq_sb); 6963 freeze_protected = true; 6964 } 6965 6966 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6967 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6968 if (IS_ERR(handle)) { 6969 /* Release dquot anyway to avoid endless cycle in dqput() */ 6970 dquot_release(dquot); 6971 if (freeze_protected) 6972 sb_end_intwrite(dquot->dq_sb); 6973 return PTR_ERR(handle); 6974 } 6975 ret = dquot_release(dquot); 6976 if (ret < 0) 6977 ext4_error_err(dquot->dq_sb, -ret, 6978 "Failed to release dquot type %d", 6979 dquot->dq_id.type); 6980 err = ext4_journal_stop(handle); 6981 if (!ret) 6982 ret = err; 6983 6984 if (freeze_protected) 6985 sb_end_intwrite(dquot->dq_sb); 6986 6987 return ret; 6988 } 6989 6990 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6991 { 6992 struct super_block *sb = dquot->dq_sb; 6993 6994 if (ext4_is_quota_journalled(sb)) { 6995 dquot_mark_dquot_dirty(dquot); 6996 return ext4_write_dquot(dquot); 6997 } else { 6998 return dquot_mark_dquot_dirty(dquot); 6999 } 7000 } 7001 7002 static int ext4_write_info(struct super_block *sb, int type) 7003 { 7004 int ret, err; 7005 handle_t *handle; 7006 7007 /* Data block + inode block */ 7008 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 7009 if (IS_ERR(handle)) 7010 return PTR_ERR(handle); 7011 ret = dquot_commit_info(sb, type); 7012 err = ext4_journal_stop(handle); 7013 if (!ret) 7014 ret = err; 7015 return ret; 7016 } 7017 7018 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 7019 { 7020 struct ext4_inode_info *ei = EXT4_I(inode); 7021 7022 /* The first argument of lockdep_set_subclass has to be 7023 * *exactly* the same as the argument to init_rwsem() --- in 7024 * this case, in init_once() --- or lockdep gets unhappy 7025 * because the name of the lock is set using the 7026 * stringification of the argument to init_rwsem(). 7027 */ 7028 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 7029 lockdep_set_subclass(&ei->i_data_sem, subclass); 7030 } 7031 7032 /* 7033 * Standard function to be called on quota_on 7034 */ 7035 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 7036 const struct path *path) 7037 { 7038 int err; 7039 7040 if (!test_opt(sb, QUOTA)) 7041 return -EINVAL; 7042 7043 /* Quotafile not on the same filesystem? */ 7044 if (path->dentry->d_sb != sb) 7045 return -EXDEV; 7046 7047 /* Quota already enabled for this file? */ 7048 if (IS_NOQUOTA(d_inode(path->dentry))) 7049 return -EBUSY; 7050 7051 /* Journaling quota? */ 7052 if (EXT4_SB(sb)->s_qf_names[type]) { 7053 /* Quotafile not in fs root? */ 7054 if (path->dentry->d_parent != sb->s_root) 7055 ext4_msg(sb, KERN_WARNING, 7056 "Quota file not on filesystem root. " 7057 "Journaled quota will not work"); 7058 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 7059 } else { 7060 /* 7061 * Clear the flag just in case mount options changed since 7062 * last time. 7063 */ 7064 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 7065 } 7066 7067 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 7068 err = dquot_quota_on(sb, type, format_id, path); 7069 if (!err) { 7070 struct inode *inode = d_inode(path->dentry); 7071 handle_t *handle; 7072 7073 /* 7074 * Set inode flags to prevent userspace from messing with quota 7075 * files. If this fails, we return success anyway since quotas 7076 * are already enabled and this is not a hard failure. 7077 */ 7078 inode_lock(inode); 7079 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7080 if (IS_ERR(handle)) 7081 goto unlock_inode; 7082 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 7083 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 7084 S_NOATIME | S_IMMUTABLE); 7085 err = ext4_mark_inode_dirty(handle, inode); 7086 ext4_journal_stop(handle); 7087 unlock_inode: 7088 inode_unlock(inode); 7089 if (err) 7090 dquot_quota_off(sb, type); 7091 } 7092 if (err) 7093 lockdep_set_quota_inode(path->dentry->d_inode, 7094 I_DATA_SEM_NORMAL); 7095 return err; 7096 } 7097 7098 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum) 7099 { 7100 switch (type) { 7101 case USRQUOTA: 7102 return qf_inum == EXT4_USR_QUOTA_INO; 7103 case GRPQUOTA: 7104 return qf_inum == EXT4_GRP_QUOTA_INO; 7105 case PRJQUOTA: 7106 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO; 7107 default: 7108 BUG(); 7109 } 7110 } 7111 7112 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 7113 unsigned int flags) 7114 { 7115 int err; 7116 struct inode *qf_inode; 7117 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7118 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7119 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7120 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7121 }; 7122 7123 BUG_ON(!ext4_has_feature_quota(sb)); 7124 7125 if (!qf_inums[type]) 7126 return -EPERM; 7127 7128 if (!ext4_check_quota_inum(type, qf_inums[type])) { 7129 ext4_error(sb, "Bad quota inum: %lu, type: %d", 7130 qf_inums[type], type); 7131 return -EUCLEAN; 7132 } 7133 7134 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 7135 if (IS_ERR(qf_inode)) { 7136 ext4_error(sb, "Bad quota inode: %lu, type: %d", 7137 qf_inums[type], type); 7138 return PTR_ERR(qf_inode); 7139 } 7140 7141 /* Don't account quota for quota files to avoid recursion */ 7142 qf_inode->i_flags |= S_NOQUOTA; 7143 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 7144 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 7145 if (err) 7146 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 7147 iput(qf_inode); 7148 7149 return err; 7150 } 7151 7152 /* Enable usage tracking for all quota types. */ 7153 int ext4_enable_quotas(struct super_block *sb) 7154 { 7155 int type, err = 0; 7156 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7157 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7158 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7159 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7160 }; 7161 bool quota_mopt[EXT4_MAXQUOTAS] = { 7162 test_opt(sb, USRQUOTA), 7163 test_opt(sb, GRPQUOTA), 7164 test_opt(sb, PRJQUOTA), 7165 }; 7166 7167 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 7168 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 7169 if (qf_inums[type]) { 7170 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 7171 DQUOT_USAGE_ENABLED | 7172 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 7173 if (err) { 7174 ext4_warning(sb, 7175 "Failed to enable quota tracking " 7176 "(type=%d, err=%d, ino=%lu). " 7177 "Please run e2fsck to fix.", type, 7178 err, qf_inums[type]); 7179 7180 ext4_quotas_off(sb, type); 7181 return err; 7182 } 7183 } 7184 } 7185 return 0; 7186 } 7187 7188 static int ext4_quota_off(struct super_block *sb, int type) 7189 { 7190 struct inode *inode = sb_dqopt(sb)->files[type]; 7191 handle_t *handle; 7192 int err; 7193 7194 /* Force all delayed allocation blocks to be allocated. 7195 * Caller already holds s_umount sem */ 7196 if (test_opt(sb, DELALLOC)) 7197 sync_filesystem(sb); 7198 7199 if (!inode || !igrab(inode)) 7200 goto out; 7201 7202 err = dquot_quota_off(sb, type); 7203 if (err || ext4_has_feature_quota(sb)) 7204 goto out_put; 7205 /* 7206 * When the filesystem was remounted read-only first, we cannot cleanup 7207 * inode flags here. Bad luck but people should be using QUOTA feature 7208 * these days anyway. 7209 */ 7210 if (sb_rdonly(sb)) 7211 goto out_put; 7212 7213 inode_lock(inode); 7214 /* 7215 * Update modification times of quota files when userspace can 7216 * start looking at them. If we fail, we return success anyway since 7217 * this is not a hard failure and quotas are already disabled. 7218 */ 7219 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7220 if (IS_ERR(handle)) { 7221 err = PTR_ERR(handle); 7222 goto out_unlock; 7223 } 7224 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7225 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7226 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 7227 err = ext4_mark_inode_dirty(handle, inode); 7228 ext4_journal_stop(handle); 7229 out_unlock: 7230 inode_unlock(inode); 7231 out_put: 7232 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7233 iput(inode); 7234 return err; 7235 out: 7236 return dquot_quota_off(sb, type); 7237 } 7238 7239 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7240 * acquiring the locks... As quota files are never truncated and quota code 7241 * itself serializes the operations (and no one else should touch the files) 7242 * we don't have to be afraid of races */ 7243 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7244 size_t len, loff_t off) 7245 { 7246 struct inode *inode = sb_dqopt(sb)->files[type]; 7247 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7248 int offset = off & (sb->s_blocksize - 1); 7249 int tocopy; 7250 size_t toread; 7251 struct buffer_head *bh; 7252 loff_t i_size = i_size_read(inode); 7253 7254 if (off > i_size) 7255 return 0; 7256 if (off+len > i_size) 7257 len = i_size-off; 7258 toread = len; 7259 while (toread > 0) { 7260 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread); 7261 bh = ext4_bread(NULL, inode, blk, 0); 7262 if (IS_ERR(bh)) 7263 return PTR_ERR(bh); 7264 if (!bh) /* A hole? */ 7265 memset(data, 0, tocopy); 7266 else 7267 memcpy(data, bh->b_data+offset, tocopy); 7268 brelse(bh); 7269 offset = 0; 7270 toread -= tocopy; 7271 data += tocopy; 7272 blk++; 7273 } 7274 return len; 7275 } 7276 7277 /* Write to quotafile (we know the transaction is already started and has 7278 * enough credits) */ 7279 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7280 const char *data, size_t len, loff_t off) 7281 { 7282 struct inode *inode = sb_dqopt(sb)->files[type]; 7283 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7284 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7285 int retries = 0; 7286 struct buffer_head *bh; 7287 handle_t *handle = journal_current_handle(); 7288 7289 if (!handle) { 7290 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7291 " cancelled because transaction is not started", 7292 (unsigned long long)off, (unsigned long long)len); 7293 return -EIO; 7294 } 7295 /* 7296 * Since we account only one data block in transaction credits, 7297 * then it is impossible to cross a block boundary. 7298 */ 7299 if (sb->s_blocksize - offset < len) { 7300 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7301 " cancelled because not block aligned", 7302 (unsigned long long)off, (unsigned long long)len); 7303 return -EIO; 7304 } 7305 7306 do { 7307 bh = ext4_bread(handle, inode, blk, 7308 EXT4_GET_BLOCKS_CREATE | 7309 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7310 } while (PTR_ERR(bh) == -ENOSPC && 7311 ext4_should_retry_alloc(inode->i_sb, &retries)); 7312 if (IS_ERR(bh)) 7313 return PTR_ERR(bh); 7314 if (!bh) 7315 goto out; 7316 BUFFER_TRACE(bh, "get write access"); 7317 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7318 if (err) { 7319 brelse(bh); 7320 return err; 7321 } 7322 lock_buffer(bh); 7323 memcpy(bh->b_data+offset, data, len); 7324 flush_dcache_folio(bh->b_folio); 7325 unlock_buffer(bh); 7326 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7327 brelse(bh); 7328 out: 7329 if (inode->i_size < off + len) { 7330 i_size_write(inode, off + len); 7331 EXT4_I(inode)->i_disksize = inode->i_size; 7332 err2 = ext4_mark_inode_dirty(handle, inode); 7333 if (unlikely(err2 && !err)) 7334 err = err2; 7335 } 7336 return err ? err : len; 7337 } 7338 #endif 7339 7340 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7341 static inline void register_as_ext2(void) 7342 { 7343 int err = register_filesystem(&ext2_fs_type); 7344 if (err) 7345 printk(KERN_WARNING 7346 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7347 } 7348 7349 static inline void unregister_as_ext2(void) 7350 { 7351 unregister_filesystem(&ext2_fs_type); 7352 } 7353 7354 static inline int ext2_feature_set_ok(struct super_block *sb) 7355 { 7356 if (ext4_has_unknown_ext2_incompat_features(sb)) 7357 return 0; 7358 if (sb_rdonly(sb)) 7359 return 1; 7360 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7361 return 0; 7362 return 1; 7363 } 7364 #else 7365 static inline void register_as_ext2(void) { } 7366 static inline void unregister_as_ext2(void) { } 7367 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7368 #endif 7369 7370 static inline void register_as_ext3(void) 7371 { 7372 int err = register_filesystem(&ext3_fs_type); 7373 if (err) 7374 printk(KERN_WARNING 7375 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7376 } 7377 7378 static inline void unregister_as_ext3(void) 7379 { 7380 unregister_filesystem(&ext3_fs_type); 7381 } 7382 7383 static inline int ext3_feature_set_ok(struct super_block *sb) 7384 { 7385 if (ext4_has_unknown_ext3_incompat_features(sb)) 7386 return 0; 7387 if (!ext4_has_feature_journal(sb)) 7388 return 0; 7389 if (sb_rdonly(sb)) 7390 return 1; 7391 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7392 return 0; 7393 return 1; 7394 } 7395 7396 static void ext4_kill_sb(struct super_block *sb) 7397 { 7398 struct ext4_sb_info *sbi = EXT4_SB(sb); 7399 struct file *bdev_file = sbi ? sbi->s_journal_bdev_file : NULL; 7400 7401 kill_block_super(sb); 7402 7403 if (bdev_file) 7404 bdev_fput(bdev_file); 7405 } 7406 7407 static struct file_system_type ext4_fs_type = { 7408 .owner = THIS_MODULE, 7409 .name = "ext4", 7410 .init_fs_context = ext4_init_fs_context, 7411 .parameters = ext4_param_specs, 7412 .kill_sb = ext4_kill_sb, 7413 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP | FS_MGTIME, 7414 }; 7415 MODULE_ALIAS_FS("ext4"); 7416 7417 static int __init ext4_init_fs(void) 7418 { 7419 int err; 7420 7421 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7422 ext4_li_info = NULL; 7423 7424 /* Build-time check for flags consistency */ 7425 ext4_check_flag_values(); 7426 7427 err = ext4_init_es(); 7428 if (err) 7429 return err; 7430 7431 err = ext4_init_pending(); 7432 if (err) 7433 goto out7; 7434 7435 err = ext4_init_post_read_processing(); 7436 if (err) 7437 goto out6; 7438 7439 err = ext4_init_pageio(); 7440 if (err) 7441 goto out5; 7442 7443 err = ext4_init_system_zone(); 7444 if (err) 7445 goto out4; 7446 7447 err = ext4_init_sysfs(); 7448 if (err) 7449 goto out3; 7450 7451 err = ext4_init_mballoc(); 7452 if (err) 7453 goto out2; 7454 err = init_inodecache(); 7455 if (err) 7456 goto out1; 7457 7458 err = ext4_fc_init_dentry_cache(); 7459 if (err) 7460 goto out05; 7461 7462 register_as_ext3(); 7463 register_as_ext2(); 7464 err = register_filesystem(&ext4_fs_type); 7465 if (err) 7466 goto out; 7467 7468 return 0; 7469 out: 7470 unregister_as_ext2(); 7471 unregister_as_ext3(); 7472 ext4_fc_destroy_dentry_cache(); 7473 out05: 7474 destroy_inodecache(); 7475 out1: 7476 ext4_exit_mballoc(); 7477 out2: 7478 ext4_exit_sysfs(); 7479 out3: 7480 ext4_exit_system_zone(); 7481 out4: 7482 ext4_exit_pageio(); 7483 out5: 7484 ext4_exit_post_read_processing(); 7485 out6: 7486 ext4_exit_pending(); 7487 out7: 7488 ext4_exit_es(); 7489 7490 return err; 7491 } 7492 7493 static void __exit ext4_exit_fs(void) 7494 { 7495 ext4_destroy_lazyinit_thread(); 7496 unregister_as_ext2(); 7497 unregister_as_ext3(); 7498 unregister_filesystem(&ext4_fs_type); 7499 ext4_fc_destroy_dentry_cache(); 7500 destroy_inodecache(); 7501 ext4_exit_mballoc(); 7502 ext4_exit_sysfs(); 7503 ext4_exit_system_zone(); 7504 ext4_exit_pageio(); 7505 ext4_exit_post_read_processing(); 7506 ext4_exit_es(); 7507 ext4_exit_pending(); 7508 } 7509 7510 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7511 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7512 MODULE_LICENSE("GPL"); 7513 module_init(ext4_init_fs) 7514 module_exit(ext4_exit_fs) 7515