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 | __GFP_NOWARN); 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 = generic_drop_inode(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 return 0; 2008 } 2009 2010 #ifdef CONFIG_QUOTA 2011 /* 2012 * Note the name of the specified quota file. 2013 */ 2014 static int note_qf_name(struct fs_context *fc, int qtype, 2015 struct fs_parameter *param) 2016 { 2017 struct ext4_fs_context *ctx = fc->fs_private; 2018 char *qname; 2019 2020 if (param->size < 1) { 2021 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 2022 return -EINVAL; 2023 } 2024 if (strchr(param->string, '/')) { 2025 ext4_msg(NULL, KERN_ERR, 2026 "quotafile must be on filesystem root"); 2027 return -EINVAL; 2028 } 2029 if (ctx->s_qf_names[qtype]) { 2030 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 2031 ext4_msg(NULL, KERN_ERR, 2032 "%s quota file already specified", 2033 QTYPE2NAME(qtype)); 2034 return -EINVAL; 2035 } 2036 return 0; 2037 } 2038 2039 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 2040 if (!qname) { 2041 ext4_msg(NULL, KERN_ERR, 2042 "Not enough memory for storing quotafile name"); 2043 return -ENOMEM; 2044 } 2045 ctx->s_qf_names[qtype] = qname; 2046 ctx->qname_spec |= 1 << qtype; 2047 ctx->spec |= EXT4_SPEC_JQUOTA; 2048 return 0; 2049 } 2050 2051 /* 2052 * Clear the name of the specified quota file. 2053 */ 2054 static int unnote_qf_name(struct fs_context *fc, int qtype) 2055 { 2056 struct ext4_fs_context *ctx = fc->fs_private; 2057 2058 kfree(ctx->s_qf_names[qtype]); 2059 2060 ctx->s_qf_names[qtype] = NULL; 2061 ctx->qname_spec |= 1 << qtype; 2062 ctx->spec |= EXT4_SPEC_JQUOTA; 2063 return 0; 2064 } 2065 #endif 2066 2067 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2068 struct ext4_fs_context *ctx) 2069 { 2070 int err; 2071 2072 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2073 ext4_msg(NULL, KERN_WARNING, 2074 "test_dummy_encryption option not supported"); 2075 return -EINVAL; 2076 } 2077 err = fscrypt_parse_test_dummy_encryption(param, 2078 &ctx->dummy_enc_policy); 2079 if (err == -EINVAL) { 2080 ext4_msg(NULL, KERN_WARNING, 2081 "Value of option \"%s\" is unrecognized", param->key); 2082 } else if (err == -EEXIST) { 2083 ext4_msg(NULL, KERN_WARNING, 2084 "Conflicting test_dummy_encryption options"); 2085 return -EINVAL; 2086 } 2087 return err; 2088 } 2089 2090 #define EXT4_SET_CTX(name) \ 2091 static inline __maybe_unused \ 2092 void ctx_set_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2093 { \ 2094 ctx->mask_s_##name |= flag; \ 2095 ctx->vals_s_##name |= flag; \ 2096 } 2097 2098 #define EXT4_CLEAR_CTX(name) \ 2099 static inline __maybe_unused \ 2100 void ctx_clear_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2101 { \ 2102 ctx->mask_s_##name |= flag; \ 2103 ctx->vals_s_##name &= ~flag; \ 2104 } 2105 2106 #define EXT4_TEST_CTX(name) \ 2107 static inline unsigned long \ 2108 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2109 { \ 2110 return (ctx->vals_s_##name & flag); \ 2111 } 2112 2113 EXT4_SET_CTX(flags); /* set only */ 2114 EXT4_SET_CTX(mount_opt); 2115 EXT4_CLEAR_CTX(mount_opt); 2116 EXT4_TEST_CTX(mount_opt); 2117 EXT4_SET_CTX(mount_opt2); 2118 EXT4_CLEAR_CTX(mount_opt2); 2119 EXT4_TEST_CTX(mount_opt2); 2120 2121 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2122 { 2123 struct ext4_fs_context *ctx = fc->fs_private; 2124 struct fs_parse_result result; 2125 const struct mount_opts *m; 2126 int is_remount; 2127 int token; 2128 2129 token = fs_parse(fc, ext4_param_specs, param, &result); 2130 if (token < 0) 2131 return token; 2132 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2133 2134 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2135 if (token == m->token) 2136 break; 2137 2138 ctx->opt_flags |= m->flags; 2139 2140 if (m->flags & MOPT_EXPLICIT) { 2141 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2142 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2143 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2144 ctx_set_mount_opt2(ctx, 2145 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2146 } else 2147 return -EINVAL; 2148 } 2149 2150 if (m->flags & MOPT_NOSUPPORT) { 2151 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2152 param->key); 2153 return 0; 2154 } 2155 2156 switch (token) { 2157 #ifdef CONFIG_QUOTA 2158 case Opt_usrjquota: 2159 if (!*param->string) 2160 return unnote_qf_name(fc, USRQUOTA); 2161 else 2162 return note_qf_name(fc, USRQUOTA, param); 2163 case Opt_grpjquota: 2164 if (!*param->string) 2165 return unnote_qf_name(fc, GRPQUOTA); 2166 else 2167 return note_qf_name(fc, GRPQUOTA, param); 2168 #endif 2169 case Opt_sb: 2170 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2171 ext4_msg(NULL, KERN_WARNING, 2172 "Ignoring %s option on remount", param->key); 2173 } else { 2174 ctx->s_sb_block = result.uint_32; 2175 ctx->spec |= EXT4_SPEC_s_sb_block; 2176 } 2177 return 0; 2178 case Opt_removed: 2179 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2180 param->key); 2181 return 0; 2182 case Opt_inlinecrypt: 2183 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2184 ctx_set_flags(ctx, SB_INLINECRYPT); 2185 #else 2186 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2187 #endif 2188 return 0; 2189 case Opt_errors: 2190 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2191 ctx_set_mount_opt(ctx, result.uint_32); 2192 return 0; 2193 #ifdef CONFIG_QUOTA 2194 case Opt_jqfmt: 2195 ctx->s_jquota_fmt = result.uint_32; 2196 ctx->spec |= EXT4_SPEC_JQFMT; 2197 return 0; 2198 #endif 2199 case Opt_data: 2200 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2201 ctx_set_mount_opt(ctx, result.uint_32); 2202 ctx->spec |= EXT4_SPEC_DATAJ; 2203 return 0; 2204 case Opt_commit: 2205 if (result.uint_32 == 0) 2206 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE; 2207 else if (result.uint_32 > INT_MAX / HZ) { 2208 ext4_msg(NULL, KERN_ERR, 2209 "Invalid commit interval %d, " 2210 "must be smaller than %d", 2211 result.uint_32, INT_MAX / HZ); 2212 return -EINVAL; 2213 } 2214 ctx->s_commit_interval = HZ * result.uint_32; 2215 ctx->spec |= EXT4_SPEC_s_commit_interval; 2216 return 0; 2217 case Opt_debug_want_extra_isize: 2218 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2219 ext4_msg(NULL, KERN_ERR, 2220 "Invalid want_extra_isize %d", result.uint_32); 2221 return -EINVAL; 2222 } 2223 ctx->s_want_extra_isize = result.uint_32; 2224 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2225 return 0; 2226 case Opt_max_batch_time: 2227 ctx->s_max_batch_time = result.uint_32; 2228 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2229 return 0; 2230 case Opt_min_batch_time: 2231 ctx->s_min_batch_time = result.uint_32; 2232 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2233 return 0; 2234 case Opt_inode_readahead_blks: 2235 if (result.uint_32 && 2236 (result.uint_32 > (1 << 30) || 2237 !is_power_of_2(result.uint_32))) { 2238 ext4_msg(NULL, KERN_ERR, 2239 "EXT4-fs: inode_readahead_blks must be " 2240 "0 or a power of 2 smaller than 2^31"); 2241 return -EINVAL; 2242 } 2243 ctx->s_inode_readahead_blks = result.uint_32; 2244 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2245 return 0; 2246 case Opt_init_itable: 2247 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2248 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2249 if (param->type == fs_value_is_string) 2250 ctx->s_li_wait_mult = result.uint_32; 2251 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2252 return 0; 2253 case Opt_max_dir_size_kb: 2254 ctx->s_max_dir_size_kb = result.uint_32; 2255 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2256 return 0; 2257 #ifdef CONFIG_EXT4_DEBUG 2258 case Opt_fc_debug_max_replay: 2259 ctx->s_fc_debug_max_replay = result.uint_32; 2260 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2261 return 0; 2262 #endif 2263 case Opt_stripe: 2264 ctx->s_stripe = result.uint_32; 2265 ctx->spec |= EXT4_SPEC_s_stripe; 2266 return 0; 2267 case Opt_resuid: 2268 ctx->s_resuid = result.uid; 2269 ctx->spec |= EXT4_SPEC_s_resuid; 2270 return 0; 2271 case Opt_resgid: 2272 ctx->s_resgid = result.gid; 2273 ctx->spec |= EXT4_SPEC_s_resgid; 2274 return 0; 2275 case Opt_journal_dev: 2276 if (is_remount) { 2277 ext4_msg(NULL, KERN_ERR, 2278 "Cannot specify journal on remount"); 2279 return -EINVAL; 2280 } 2281 ctx->journal_devnum = result.uint_32; 2282 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2283 return 0; 2284 case Opt_journal_path: 2285 { 2286 struct inode *journal_inode; 2287 struct path path; 2288 int error; 2289 2290 if (is_remount) { 2291 ext4_msg(NULL, KERN_ERR, 2292 "Cannot specify journal on remount"); 2293 return -EINVAL; 2294 } 2295 2296 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path); 2297 if (error) { 2298 ext4_msg(NULL, KERN_ERR, "error: could not find " 2299 "journal device path"); 2300 return -EINVAL; 2301 } 2302 2303 journal_inode = d_inode(path.dentry); 2304 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2305 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2306 path_put(&path); 2307 return 0; 2308 } 2309 case Opt_journal_ioprio: 2310 if (result.uint_32 > 7) { 2311 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2312 " (must be 0-7)"); 2313 return -EINVAL; 2314 } 2315 ctx->journal_ioprio = 2316 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2317 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2318 return 0; 2319 case Opt_test_dummy_encryption: 2320 return ext4_parse_test_dummy_encryption(param, ctx); 2321 case Opt_dax: 2322 case Opt_dax_type: 2323 #ifdef CONFIG_FS_DAX 2324 { 2325 int type = (token == Opt_dax) ? 2326 Opt_dax : result.uint_32; 2327 2328 switch (type) { 2329 case Opt_dax: 2330 case Opt_dax_always: 2331 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2332 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2333 break; 2334 case Opt_dax_never: 2335 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2336 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2337 break; 2338 case Opt_dax_inode: 2339 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2340 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2341 /* Strictly for printing options */ 2342 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2343 break; 2344 } 2345 return 0; 2346 } 2347 #else 2348 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2349 return -EINVAL; 2350 #endif 2351 case Opt_data_err: 2352 if (result.uint_32 == Opt_data_err_abort) 2353 ctx_set_mount_opt(ctx, m->mount_opt); 2354 else if (result.uint_32 == Opt_data_err_ignore) 2355 ctx_clear_mount_opt(ctx, m->mount_opt); 2356 return 0; 2357 case Opt_mb_optimize_scan: 2358 if (result.int_32 == 1) { 2359 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2360 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2361 } else if (result.int_32 == 0) { 2362 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2363 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2364 } else { 2365 ext4_msg(NULL, KERN_WARNING, 2366 "mb_optimize_scan should be set to 0 or 1."); 2367 return -EINVAL; 2368 } 2369 return 0; 2370 } 2371 2372 /* 2373 * At this point we should only be getting options requiring MOPT_SET, 2374 * or MOPT_CLEAR. Anything else is a bug 2375 */ 2376 if (m->token == Opt_err) { 2377 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2378 param->key); 2379 WARN_ON(1); 2380 return -EINVAL; 2381 } 2382 2383 else { 2384 unsigned int set = 0; 2385 2386 if ((param->type == fs_value_is_flag) || 2387 result.uint_32 > 0) 2388 set = 1; 2389 2390 if (m->flags & MOPT_CLEAR) 2391 set = !set; 2392 else if (unlikely(!(m->flags & MOPT_SET))) { 2393 ext4_msg(NULL, KERN_WARNING, 2394 "buggy handling of option %s", 2395 param->key); 2396 WARN_ON(1); 2397 return -EINVAL; 2398 } 2399 if (m->flags & MOPT_2) { 2400 if (set != 0) 2401 ctx_set_mount_opt2(ctx, m->mount_opt); 2402 else 2403 ctx_clear_mount_opt2(ctx, m->mount_opt); 2404 } else { 2405 if (set != 0) 2406 ctx_set_mount_opt(ctx, m->mount_opt); 2407 else 2408 ctx_clear_mount_opt(ctx, m->mount_opt); 2409 } 2410 } 2411 2412 return 0; 2413 } 2414 2415 static int parse_options(struct fs_context *fc, char *options) 2416 { 2417 struct fs_parameter param; 2418 int ret; 2419 char *key; 2420 2421 if (!options) 2422 return 0; 2423 2424 while ((key = strsep(&options, ",")) != NULL) { 2425 if (*key) { 2426 size_t v_len = 0; 2427 char *value = strchr(key, '='); 2428 2429 param.type = fs_value_is_flag; 2430 param.string = NULL; 2431 2432 if (value) { 2433 if (value == key) 2434 continue; 2435 2436 *value++ = 0; 2437 v_len = strlen(value); 2438 param.string = kmemdup_nul(value, v_len, 2439 GFP_KERNEL); 2440 if (!param.string) 2441 return -ENOMEM; 2442 param.type = fs_value_is_string; 2443 } 2444 2445 param.key = key; 2446 param.size = v_len; 2447 2448 ret = ext4_parse_param(fc, ¶m); 2449 kfree(param.string); 2450 if (ret < 0) 2451 return ret; 2452 } 2453 } 2454 2455 ret = ext4_validate_options(fc); 2456 if (ret < 0) 2457 return ret; 2458 2459 return 0; 2460 } 2461 2462 static int parse_apply_sb_mount_options(struct super_block *sb, 2463 struct ext4_fs_context *m_ctx) 2464 { 2465 struct ext4_sb_info *sbi = EXT4_SB(sb); 2466 char *s_mount_opts = NULL; 2467 struct ext4_fs_context *s_ctx = NULL; 2468 struct fs_context *fc = NULL; 2469 int ret = -ENOMEM; 2470 2471 if (!sbi->s_es->s_mount_opts[0]) 2472 return 0; 2473 2474 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2475 sizeof(sbi->s_es->s_mount_opts), 2476 GFP_KERNEL); 2477 if (!s_mount_opts) 2478 return ret; 2479 2480 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2481 if (!fc) 2482 goto out_free; 2483 2484 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2485 if (!s_ctx) 2486 goto out_free; 2487 2488 fc->fs_private = s_ctx; 2489 fc->s_fs_info = sbi; 2490 2491 ret = parse_options(fc, s_mount_opts); 2492 if (ret < 0) 2493 goto parse_failed; 2494 2495 ret = ext4_check_opt_consistency(fc, sb); 2496 if (ret < 0) { 2497 parse_failed: 2498 ext4_msg(sb, KERN_WARNING, 2499 "failed to parse options in superblock: %s", 2500 s_mount_opts); 2501 ret = 0; 2502 goto out_free; 2503 } 2504 2505 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2506 m_ctx->journal_devnum = s_ctx->journal_devnum; 2507 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2508 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2509 2510 ext4_apply_options(fc, sb); 2511 ret = 0; 2512 2513 out_free: 2514 if (fc) { 2515 ext4_fc_free(fc); 2516 kfree(fc); 2517 } 2518 kfree(s_mount_opts); 2519 return ret; 2520 } 2521 2522 static void ext4_apply_quota_options(struct fs_context *fc, 2523 struct super_block *sb) 2524 { 2525 #ifdef CONFIG_QUOTA 2526 bool quota_feature = ext4_has_feature_quota(sb); 2527 struct ext4_fs_context *ctx = fc->fs_private; 2528 struct ext4_sb_info *sbi = EXT4_SB(sb); 2529 char *qname; 2530 int i; 2531 2532 if (quota_feature) 2533 return; 2534 2535 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2536 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2537 if (!(ctx->qname_spec & (1 << i))) 2538 continue; 2539 2540 qname = ctx->s_qf_names[i]; /* May be NULL */ 2541 if (qname) 2542 set_opt(sb, QUOTA); 2543 ctx->s_qf_names[i] = NULL; 2544 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2545 lockdep_is_held(&sb->s_umount)); 2546 if (qname) 2547 kfree_rcu_mightsleep(qname); 2548 } 2549 } 2550 2551 if (ctx->spec & EXT4_SPEC_JQFMT) 2552 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2553 #endif 2554 } 2555 2556 /* 2557 * Check quota settings consistency. 2558 */ 2559 static int ext4_check_quota_consistency(struct fs_context *fc, 2560 struct super_block *sb) 2561 { 2562 #ifdef CONFIG_QUOTA 2563 struct ext4_fs_context *ctx = fc->fs_private; 2564 struct ext4_sb_info *sbi = EXT4_SB(sb); 2565 bool quota_feature = ext4_has_feature_quota(sb); 2566 bool quota_loaded = sb_any_quota_loaded(sb); 2567 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2568 int quota_flags, i; 2569 2570 /* 2571 * We do the test below only for project quotas. 'usrquota' and 2572 * 'grpquota' mount options are allowed even without quota feature 2573 * to support legacy quotas in quota files. 2574 */ 2575 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2576 !ext4_has_feature_project(sb)) { 2577 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2578 "Cannot enable project quota enforcement."); 2579 return -EINVAL; 2580 } 2581 2582 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2583 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2584 if (quota_loaded && 2585 ctx->mask_s_mount_opt & quota_flags && 2586 !ctx_test_mount_opt(ctx, quota_flags)) 2587 goto err_quota_change; 2588 2589 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2590 2591 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2592 if (!(ctx->qname_spec & (1 << i))) 2593 continue; 2594 2595 if (quota_loaded && 2596 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2597 goto err_jquota_change; 2598 2599 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2600 strcmp(get_qf_name(sb, sbi, i), 2601 ctx->s_qf_names[i]) != 0) 2602 goto err_jquota_specified; 2603 } 2604 2605 if (quota_feature) { 2606 ext4_msg(NULL, KERN_INFO, 2607 "Journaled quota options ignored when " 2608 "QUOTA feature is enabled"); 2609 return 0; 2610 } 2611 } 2612 2613 if (ctx->spec & EXT4_SPEC_JQFMT) { 2614 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2615 goto err_jquota_change; 2616 if (quota_feature) { 2617 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2618 "ignored when QUOTA feature is enabled"); 2619 return 0; 2620 } 2621 } 2622 2623 /* Make sure we don't mix old and new quota format */ 2624 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2625 ctx->s_qf_names[USRQUOTA]); 2626 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2627 ctx->s_qf_names[GRPQUOTA]); 2628 2629 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2630 test_opt(sb, USRQUOTA)); 2631 2632 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2633 test_opt(sb, GRPQUOTA)); 2634 2635 if (usr_qf_name) { 2636 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2637 usrquota = false; 2638 } 2639 if (grp_qf_name) { 2640 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2641 grpquota = false; 2642 } 2643 2644 if (usr_qf_name || grp_qf_name) { 2645 if (usrquota || grpquota) { 2646 ext4_msg(NULL, KERN_ERR, "old and new quota " 2647 "format mixing"); 2648 return -EINVAL; 2649 } 2650 2651 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2652 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2653 "not specified"); 2654 return -EINVAL; 2655 } 2656 } 2657 2658 return 0; 2659 2660 err_quota_change: 2661 ext4_msg(NULL, KERN_ERR, 2662 "Cannot change quota options when quota turned on"); 2663 return -EINVAL; 2664 err_jquota_change: 2665 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2666 "options when quota turned on"); 2667 return -EINVAL; 2668 err_jquota_specified: 2669 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2670 QTYPE2NAME(i)); 2671 return -EINVAL; 2672 #else 2673 return 0; 2674 #endif 2675 } 2676 2677 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2678 struct super_block *sb) 2679 { 2680 const struct ext4_fs_context *ctx = fc->fs_private; 2681 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2682 2683 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2684 return 0; 2685 2686 if (!ext4_has_feature_encrypt(sb)) { 2687 ext4_msg(NULL, KERN_WARNING, 2688 "test_dummy_encryption requires encrypt feature"); 2689 return -EINVAL; 2690 } 2691 /* 2692 * This mount option is just for testing, and it's not worthwhile to 2693 * implement the extra complexity (e.g. RCU protection) that would be 2694 * needed to allow it to be set or changed during remount. We do allow 2695 * it to be specified during remount, but only if there is no change. 2696 */ 2697 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2698 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2699 &ctx->dummy_enc_policy)) 2700 return 0; 2701 ext4_msg(NULL, KERN_WARNING, 2702 "Can't set or change test_dummy_encryption on remount"); 2703 return -EINVAL; 2704 } 2705 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2706 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2707 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2708 &ctx->dummy_enc_policy)) 2709 return 0; 2710 ext4_msg(NULL, KERN_WARNING, 2711 "Conflicting test_dummy_encryption options"); 2712 return -EINVAL; 2713 } 2714 return 0; 2715 } 2716 2717 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2718 struct super_block *sb) 2719 { 2720 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2721 /* if already set, it was already verified to be the same */ 2722 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2723 return; 2724 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2725 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2726 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2727 } 2728 2729 static int ext4_check_opt_consistency(struct fs_context *fc, 2730 struct super_block *sb) 2731 { 2732 struct ext4_fs_context *ctx = fc->fs_private; 2733 struct ext4_sb_info *sbi = fc->s_fs_info; 2734 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2735 int err; 2736 2737 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2738 ext4_msg(NULL, KERN_ERR, 2739 "Mount option(s) incompatible with ext2"); 2740 return -EINVAL; 2741 } 2742 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2743 ext4_msg(NULL, KERN_ERR, 2744 "Mount option(s) incompatible with ext3"); 2745 return -EINVAL; 2746 } 2747 2748 if (ctx->s_want_extra_isize > 2749 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2750 ext4_msg(NULL, KERN_ERR, 2751 "Invalid want_extra_isize %d", 2752 ctx->s_want_extra_isize); 2753 return -EINVAL; 2754 } 2755 2756 err = ext4_check_test_dummy_encryption(fc, sb); 2757 if (err) 2758 return err; 2759 2760 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2761 if (!sbi->s_journal) { 2762 ext4_msg(NULL, KERN_WARNING, 2763 "Remounting file system with no journal " 2764 "so ignoring journalled data option"); 2765 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2766 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2767 test_opt(sb, DATA_FLAGS)) { 2768 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2769 "on remount"); 2770 return -EINVAL; 2771 } 2772 } 2773 2774 if (is_remount) { 2775 if (!sbi->s_journal && 2776 ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT)) { 2777 ext4_msg(NULL, KERN_WARNING, 2778 "Remounting fs w/o journal so ignoring data_err option"); 2779 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT); 2780 } 2781 2782 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2783 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2784 ext4_msg(NULL, KERN_ERR, "can't mount with " 2785 "both data=journal and dax"); 2786 return -EINVAL; 2787 } 2788 2789 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2790 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2791 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2792 fail_dax_change_remount: 2793 ext4_msg(NULL, KERN_ERR, "can't change " 2794 "dax mount option while remounting"); 2795 return -EINVAL; 2796 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2797 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2798 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2799 goto fail_dax_change_remount; 2800 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2801 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2802 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2803 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2804 goto fail_dax_change_remount; 2805 } 2806 } 2807 2808 return ext4_check_quota_consistency(fc, sb); 2809 } 2810 2811 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2812 { 2813 struct ext4_fs_context *ctx = fc->fs_private; 2814 struct ext4_sb_info *sbi = fc->s_fs_info; 2815 2816 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2817 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2818 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2819 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2820 sb->s_flags &= ~ctx->mask_s_flags; 2821 sb->s_flags |= ctx->vals_s_flags; 2822 2823 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2824 APPLY(s_commit_interval); 2825 APPLY(s_stripe); 2826 APPLY(s_max_batch_time); 2827 APPLY(s_min_batch_time); 2828 APPLY(s_want_extra_isize); 2829 APPLY(s_inode_readahead_blks); 2830 APPLY(s_max_dir_size_kb); 2831 APPLY(s_li_wait_mult); 2832 APPLY(s_resgid); 2833 APPLY(s_resuid); 2834 2835 #ifdef CONFIG_EXT4_DEBUG 2836 APPLY(s_fc_debug_max_replay); 2837 #endif 2838 2839 ext4_apply_quota_options(fc, sb); 2840 ext4_apply_test_dummy_encryption(ctx, sb); 2841 } 2842 2843 2844 static int ext4_validate_options(struct fs_context *fc) 2845 { 2846 #ifdef CONFIG_QUOTA 2847 struct ext4_fs_context *ctx = fc->fs_private; 2848 char *usr_qf_name, *grp_qf_name; 2849 2850 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2851 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2852 2853 if (usr_qf_name || grp_qf_name) { 2854 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2855 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2856 2857 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2858 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2859 2860 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2861 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2862 ext4_msg(NULL, KERN_ERR, "old and new quota " 2863 "format mixing"); 2864 return -EINVAL; 2865 } 2866 } 2867 #endif 2868 return 1; 2869 } 2870 2871 static inline void ext4_show_quota_options(struct seq_file *seq, 2872 struct super_block *sb) 2873 { 2874 #if defined(CONFIG_QUOTA) 2875 struct ext4_sb_info *sbi = EXT4_SB(sb); 2876 char *usr_qf_name, *grp_qf_name; 2877 2878 if (sbi->s_jquota_fmt) { 2879 char *fmtname = ""; 2880 2881 switch (sbi->s_jquota_fmt) { 2882 case QFMT_VFS_OLD: 2883 fmtname = "vfsold"; 2884 break; 2885 case QFMT_VFS_V0: 2886 fmtname = "vfsv0"; 2887 break; 2888 case QFMT_VFS_V1: 2889 fmtname = "vfsv1"; 2890 break; 2891 } 2892 seq_printf(seq, ",jqfmt=%s", fmtname); 2893 } 2894 2895 rcu_read_lock(); 2896 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2897 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2898 if (usr_qf_name) 2899 seq_show_option(seq, "usrjquota", usr_qf_name); 2900 if (grp_qf_name) 2901 seq_show_option(seq, "grpjquota", grp_qf_name); 2902 rcu_read_unlock(); 2903 #endif 2904 } 2905 2906 static const char *token2str(int token) 2907 { 2908 const struct fs_parameter_spec *spec; 2909 2910 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2911 if (spec->opt == token && !spec->type) 2912 break; 2913 return spec->name; 2914 } 2915 2916 /* 2917 * Show an option if 2918 * - it's set to a non-default value OR 2919 * - if the per-sb default is different from the global default 2920 */ 2921 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2922 int nodefs) 2923 { 2924 struct ext4_sb_info *sbi = EXT4_SB(sb); 2925 struct ext4_super_block *es = sbi->s_es; 2926 int def_errors; 2927 const struct mount_opts *m; 2928 char sep = nodefs ? '\n' : ','; 2929 2930 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2931 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2932 2933 if (sbi->s_sb_block != 1) 2934 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2935 2936 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2937 int want_set = m->flags & MOPT_SET; 2938 int opt_2 = m->flags & MOPT_2; 2939 unsigned int mount_opt, def_mount_opt; 2940 2941 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2942 m->flags & MOPT_SKIP) 2943 continue; 2944 2945 if (opt_2) { 2946 mount_opt = sbi->s_mount_opt2; 2947 def_mount_opt = sbi->s_def_mount_opt2; 2948 } else { 2949 mount_opt = sbi->s_mount_opt; 2950 def_mount_opt = sbi->s_def_mount_opt; 2951 } 2952 /* skip if same as the default */ 2953 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt))) 2954 continue; 2955 /* select Opt_noFoo vs Opt_Foo */ 2956 if ((want_set && 2957 (mount_opt & m->mount_opt) != m->mount_opt) || 2958 (!want_set && (mount_opt & m->mount_opt))) 2959 continue; 2960 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2961 } 2962 2963 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2964 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2965 SEQ_OPTS_PRINT("resuid=%u", 2966 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2967 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2968 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2969 SEQ_OPTS_PRINT("resgid=%u", 2970 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2971 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2972 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2973 SEQ_OPTS_PUTS("errors=remount-ro"); 2974 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2975 SEQ_OPTS_PUTS("errors=continue"); 2976 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2977 SEQ_OPTS_PUTS("errors=panic"); 2978 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2979 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2980 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2981 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2982 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2983 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2984 if (nodefs || sbi->s_stripe) 2985 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2986 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2987 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 2988 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2989 SEQ_OPTS_PUTS("data=journal"); 2990 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2991 SEQ_OPTS_PUTS("data=ordered"); 2992 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2993 SEQ_OPTS_PUTS("data=writeback"); 2994 } 2995 if (nodefs || 2996 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2997 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2998 sbi->s_inode_readahead_blks); 2999 3000 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 3001 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 3002 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 3003 if (nodefs || sbi->s_max_dir_size_kb) 3004 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 3005 if (test_opt(sb, DATA_ERR_ABORT)) 3006 SEQ_OPTS_PUTS("data_err=abort"); 3007 3008 fscrypt_show_test_dummy_encryption(seq, sep, sb); 3009 3010 if (sb->s_flags & SB_INLINECRYPT) 3011 SEQ_OPTS_PUTS("inlinecrypt"); 3012 3013 if (test_opt(sb, DAX_ALWAYS)) { 3014 if (IS_EXT2_SB(sb)) 3015 SEQ_OPTS_PUTS("dax"); 3016 else 3017 SEQ_OPTS_PUTS("dax=always"); 3018 } else if (test_opt2(sb, DAX_NEVER)) { 3019 SEQ_OPTS_PUTS("dax=never"); 3020 } else if (test_opt2(sb, DAX_INODE)) { 3021 SEQ_OPTS_PUTS("dax=inode"); 3022 } 3023 3024 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3025 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3026 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 3027 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3028 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3029 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 3030 } 3031 3032 if (nodefs && !test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) 3033 SEQ_OPTS_PUTS("prefetch_block_bitmaps"); 3034 3035 if (ext4_emergency_ro(sb)) 3036 SEQ_OPTS_PUTS("emergency_ro"); 3037 3038 if (ext4_forced_shutdown(sb)) 3039 SEQ_OPTS_PUTS("shutdown"); 3040 3041 ext4_show_quota_options(seq, sb); 3042 return 0; 3043 } 3044 3045 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 3046 { 3047 return _ext4_show_options(seq, root->d_sb, 0); 3048 } 3049 3050 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3051 { 3052 struct super_block *sb = seq->private; 3053 int rc; 3054 3055 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3056 rc = _ext4_show_options(seq, sb, 1); 3057 seq_putc(seq, '\n'); 3058 return rc; 3059 } 3060 3061 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3062 int read_only) 3063 { 3064 struct ext4_sb_info *sbi = EXT4_SB(sb); 3065 int err = 0; 3066 3067 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3068 ext4_msg(sb, KERN_ERR, "revision level too high, " 3069 "forcing read-only mode"); 3070 err = -EROFS; 3071 goto done; 3072 } 3073 if (read_only) 3074 goto done; 3075 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3076 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3077 "running e2fsck is recommended"); 3078 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3079 ext4_msg(sb, KERN_WARNING, 3080 "warning: mounting fs with errors, " 3081 "running e2fsck is recommended"); 3082 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3083 le16_to_cpu(es->s_mnt_count) >= 3084 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3085 ext4_msg(sb, KERN_WARNING, 3086 "warning: maximal mount count reached, " 3087 "running e2fsck is recommended"); 3088 else if (le32_to_cpu(es->s_checkinterval) && 3089 (ext4_get_tstamp(es, s_lastcheck) + 3090 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3091 ext4_msg(sb, KERN_WARNING, 3092 "warning: checktime reached, " 3093 "running e2fsck is recommended"); 3094 if (!sbi->s_journal) 3095 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3096 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3097 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3098 le16_add_cpu(&es->s_mnt_count, 1); 3099 ext4_update_tstamp(es, s_mtime); 3100 if (sbi->s_journal) { 3101 ext4_set_feature_journal_needs_recovery(sb); 3102 if (ext4_has_feature_orphan_file(sb)) 3103 ext4_set_feature_orphan_present(sb); 3104 } 3105 3106 err = ext4_commit_super(sb); 3107 done: 3108 if (test_opt(sb, DEBUG)) 3109 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3110 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3111 sb->s_blocksize, 3112 sbi->s_groups_count, 3113 EXT4_BLOCKS_PER_GROUP(sb), 3114 EXT4_INODES_PER_GROUP(sb), 3115 sbi->s_mount_opt, sbi->s_mount_opt2); 3116 return err; 3117 } 3118 3119 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3120 { 3121 struct ext4_sb_info *sbi = EXT4_SB(sb); 3122 struct flex_groups **old_groups, **new_groups; 3123 int size, i, j; 3124 3125 if (!sbi->s_log_groups_per_flex) 3126 return 0; 3127 3128 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3129 if (size <= sbi->s_flex_groups_allocated) 3130 return 0; 3131 3132 new_groups = kvzalloc(roundup_pow_of_two(size * 3133 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3134 if (!new_groups) { 3135 ext4_msg(sb, KERN_ERR, 3136 "not enough memory for %d flex group pointers", size); 3137 return -ENOMEM; 3138 } 3139 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3140 new_groups[i] = kvzalloc(roundup_pow_of_two( 3141 sizeof(struct flex_groups)), 3142 GFP_KERNEL); 3143 if (!new_groups[i]) { 3144 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3145 kvfree(new_groups[j]); 3146 kvfree(new_groups); 3147 ext4_msg(sb, KERN_ERR, 3148 "not enough memory for %d flex groups", size); 3149 return -ENOMEM; 3150 } 3151 } 3152 rcu_read_lock(); 3153 old_groups = rcu_dereference(sbi->s_flex_groups); 3154 if (old_groups) 3155 memcpy(new_groups, old_groups, 3156 (sbi->s_flex_groups_allocated * 3157 sizeof(struct flex_groups *))); 3158 rcu_read_unlock(); 3159 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3160 sbi->s_flex_groups_allocated = size; 3161 if (old_groups) 3162 ext4_kvfree_array_rcu(old_groups); 3163 return 0; 3164 } 3165 3166 static int ext4_fill_flex_info(struct super_block *sb) 3167 { 3168 struct ext4_sb_info *sbi = EXT4_SB(sb); 3169 struct ext4_group_desc *gdp = NULL; 3170 struct flex_groups *fg; 3171 ext4_group_t flex_group; 3172 int i, err; 3173 3174 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3175 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3176 sbi->s_log_groups_per_flex = 0; 3177 return 1; 3178 } 3179 3180 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3181 if (err) 3182 goto failed; 3183 3184 for (i = 0; i < sbi->s_groups_count; i++) { 3185 gdp = ext4_get_group_desc(sb, i, NULL); 3186 3187 flex_group = ext4_flex_group(sbi, i); 3188 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3189 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3190 atomic64_add(ext4_free_group_clusters(sb, gdp), 3191 &fg->free_clusters); 3192 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3193 } 3194 3195 return 1; 3196 failed: 3197 return 0; 3198 } 3199 3200 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3201 struct ext4_group_desc *gdp) 3202 { 3203 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3204 __u16 crc = 0; 3205 __le32 le_group = cpu_to_le32(block_group); 3206 struct ext4_sb_info *sbi = EXT4_SB(sb); 3207 3208 if (ext4_has_feature_metadata_csum(sbi->s_sb)) { 3209 /* Use new metadata_csum algorithm */ 3210 __u32 csum32; 3211 __u16 dummy_csum = 0; 3212 3213 csum32 = ext4_chksum(sbi->s_csum_seed, (__u8 *)&le_group, 3214 sizeof(le_group)); 3215 csum32 = ext4_chksum(csum32, (__u8 *)gdp, offset); 3216 csum32 = ext4_chksum(csum32, (__u8 *)&dummy_csum, 3217 sizeof(dummy_csum)); 3218 offset += sizeof(dummy_csum); 3219 if (offset < sbi->s_desc_size) 3220 csum32 = ext4_chksum(csum32, (__u8 *)gdp + offset, 3221 sbi->s_desc_size - offset); 3222 3223 crc = csum32 & 0xFFFF; 3224 goto out; 3225 } 3226 3227 /* old crc16 code */ 3228 if (!ext4_has_feature_gdt_csum(sb)) 3229 return 0; 3230 3231 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3232 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3233 crc = crc16(crc, (__u8 *)gdp, offset); 3234 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3235 /* for checksum of struct ext4_group_desc do the rest...*/ 3236 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size) 3237 crc = crc16(crc, (__u8 *)gdp + offset, 3238 sbi->s_desc_size - offset); 3239 3240 out: 3241 return cpu_to_le16(crc); 3242 } 3243 3244 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3245 struct ext4_group_desc *gdp) 3246 { 3247 if (ext4_has_group_desc_csum(sb) && 3248 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3249 return 0; 3250 3251 return 1; 3252 } 3253 3254 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3255 struct ext4_group_desc *gdp) 3256 { 3257 if (!ext4_has_group_desc_csum(sb)) 3258 return; 3259 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3260 } 3261 3262 /* Called at mount-time, super-block is locked */ 3263 static int ext4_check_descriptors(struct super_block *sb, 3264 ext4_fsblk_t sb_block, 3265 ext4_group_t *first_not_zeroed) 3266 { 3267 struct ext4_sb_info *sbi = EXT4_SB(sb); 3268 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3269 ext4_fsblk_t last_block; 3270 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3271 ext4_fsblk_t block_bitmap; 3272 ext4_fsblk_t inode_bitmap; 3273 ext4_fsblk_t inode_table; 3274 int flexbg_flag = 0; 3275 ext4_group_t i, grp = sbi->s_groups_count; 3276 3277 if (ext4_has_feature_flex_bg(sb)) 3278 flexbg_flag = 1; 3279 3280 ext4_debug("Checking group descriptors"); 3281 3282 for (i = 0; i < sbi->s_groups_count; i++) { 3283 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3284 3285 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3286 last_block = ext4_blocks_count(sbi->s_es) - 1; 3287 else 3288 last_block = first_block + 3289 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3290 3291 if ((grp == sbi->s_groups_count) && 3292 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3293 grp = i; 3294 3295 block_bitmap = ext4_block_bitmap(sb, gdp); 3296 if (block_bitmap == sb_block) { 3297 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3298 "Block bitmap for group %u overlaps " 3299 "superblock", i); 3300 if (!sb_rdonly(sb)) 3301 return 0; 3302 } 3303 if (block_bitmap >= sb_block + 1 && 3304 block_bitmap <= last_bg_block) { 3305 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3306 "Block bitmap for group %u overlaps " 3307 "block group descriptors", i); 3308 if (!sb_rdonly(sb)) 3309 return 0; 3310 } 3311 if (block_bitmap < first_block || block_bitmap > last_block) { 3312 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3313 "Block bitmap for group %u not in group " 3314 "(block %llu)!", i, block_bitmap); 3315 return 0; 3316 } 3317 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3318 if (inode_bitmap == sb_block) { 3319 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3320 "Inode bitmap for group %u overlaps " 3321 "superblock", i); 3322 if (!sb_rdonly(sb)) 3323 return 0; 3324 } 3325 if (inode_bitmap >= sb_block + 1 && 3326 inode_bitmap <= last_bg_block) { 3327 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3328 "Inode bitmap for group %u overlaps " 3329 "block group descriptors", i); 3330 if (!sb_rdonly(sb)) 3331 return 0; 3332 } 3333 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3334 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3335 "Inode bitmap for group %u not in group " 3336 "(block %llu)!", i, inode_bitmap); 3337 return 0; 3338 } 3339 inode_table = ext4_inode_table(sb, gdp); 3340 if (inode_table == sb_block) { 3341 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3342 "Inode table for group %u overlaps " 3343 "superblock", i); 3344 if (!sb_rdonly(sb)) 3345 return 0; 3346 } 3347 if (inode_table >= sb_block + 1 && 3348 inode_table <= last_bg_block) { 3349 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3350 "Inode table for group %u overlaps " 3351 "block group descriptors", i); 3352 if (!sb_rdonly(sb)) 3353 return 0; 3354 } 3355 if (inode_table < first_block || 3356 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3357 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3358 "Inode table for group %u not in group " 3359 "(block %llu)!", i, inode_table); 3360 return 0; 3361 } 3362 ext4_lock_group(sb, i); 3363 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3364 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3365 "Checksum for group %u failed (%u!=%u)", 3366 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3367 gdp)), le16_to_cpu(gdp->bg_checksum)); 3368 if (!sb_rdonly(sb)) { 3369 ext4_unlock_group(sb, i); 3370 return 0; 3371 } 3372 } 3373 ext4_unlock_group(sb, i); 3374 if (!flexbg_flag) 3375 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3376 } 3377 if (NULL != first_not_zeroed) 3378 *first_not_zeroed = grp; 3379 return 1; 3380 } 3381 3382 /* 3383 * Maximal extent format file size. 3384 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3385 * extent format containers, within a sector_t, and within i_blocks 3386 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3387 * so that won't be a limiting factor. 3388 * 3389 * However there is other limiting factor. We do store extents in the form 3390 * of starting block and length, hence the resulting length of the extent 3391 * covering maximum file size must fit into on-disk format containers as 3392 * well. Given that length is always by 1 unit bigger than max unit (because 3393 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3394 * 3395 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3396 */ 3397 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3398 { 3399 loff_t res; 3400 loff_t upper_limit = MAX_LFS_FILESIZE; 3401 3402 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3403 3404 if (!has_huge_files) { 3405 upper_limit = (1LL << 32) - 1; 3406 3407 /* total blocks in file system block size */ 3408 upper_limit >>= (blkbits - 9); 3409 upper_limit <<= blkbits; 3410 } 3411 3412 /* 3413 * 32-bit extent-start container, ee_block. We lower the maxbytes 3414 * by one fs block, so ee_len can cover the extent of maximum file 3415 * size 3416 */ 3417 res = (1LL << 32) - 1; 3418 res <<= blkbits; 3419 3420 /* Sanity check against vm- & vfs- imposed limits */ 3421 if (res > upper_limit) 3422 res = upper_limit; 3423 3424 return res; 3425 } 3426 3427 /* 3428 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3429 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3430 * We need to be 1 filesystem block less than the 2^48 sector limit. 3431 */ 3432 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3433 { 3434 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3435 int meta_blocks; 3436 unsigned int ppb = 1 << (bits - 2); 3437 3438 /* 3439 * This is calculated to be the largest file size for a dense, block 3440 * mapped file such that the file's total number of 512-byte sectors, 3441 * including data and all indirect blocks, does not exceed (2^48 - 1). 3442 * 3443 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3444 * number of 512-byte sectors of the file. 3445 */ 3446 if (!has_huge_files) { 3447 /* 3448 * !has_huge_files or implies that the inode i_block field 3449 * represents total file blocks in 2^32 512-byte sectors == 3450 * size of vfs inode i_blocks * 8 3451 */ 3452 upper_limit = (1LL << 32) - 1; 3453 3454 /* total blocks in file system block size */ 3455 upper_limit >>= (bits - 9); 3456 3457 } else { 3458 /* 3459 * We use 48 bit ext4_inode i_blocks 3460 * With EXT4_HUGE_FILE_FL set the i_blocks 3461 * represent total number of blocks in 3462 * file system block size 3463 */ 3464 upper_limit = (1LL << 48) - 1; 3465 3466 } 3467 3468 /* Compute how many blocks we can address by block tree */ 3469 res += ppb; 3470 res += ppb * ppb; 3471 res += ((loff_t)ppb) * ppb * ppb; 3472 /* Compute how many metadata blocks are needed */ 3473 meta_blocks = 1; 3474 meta_blocks += 1 + ppb; 3475 meta_blocks += 1 + ppb + ppb * ppb; 3476 /* Does block tree limit file size? */ 3477 if (res + meta_blocks <= upper_limit) 3478 goto check_lfs; 3479 3480 res = upper_limit; 3481 /* How many metadata blocks are needed for addressing upper_limit? */ 3482 upper_limit -= EXT4_NDIR_BLOCKS; 3483 /* indirect blocks */ 3484 meta_blocks = 1; 3485 upper_limit -= ppb; 3486 /* double indirect blocks */ 3487 if (upper_limit < ppb * ppb) { 3488 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3489 res -= meta_blocks; 3490 goto check_lfs; 3491 } 3492 meta_blocks += 1 + ppb; 3493 upper_limit -= ppb * ppb; 3494 /* tripple indirect blocks for the rest */ 3495 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3496 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3497 res -= meta_blocks; 3498 check_lfs: 3499 res <<= bits; 3500 if (res > MAX_LFS_FILESIZE) 3501 res = MAX_LFS_FILESIZE; 3502 3503 return res; 3504 } 3505 3506 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3507 ext4_fsblk_t logical_sb_block, int nr) 3508 { 3509 struct ext4_sb_info *sbi = EXT4_SB(sb); 3510 ext4_group_t bg, first_meta_bg; 3511 int has_super = 0; 3512 3513 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3514 3515 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3516 return logical_sb_block + nr + 1; 3517 bg = sbi->s_desc_per_block * nr; 3518 if (ext4_bg_has_super(sb, bg)) 3519 has_super = 1; 3520 3521 /* 3522 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3523 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3524 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3525 * compensate. 3526 */ 3527 if (sb->s_blocksize == 1024 && nr == 0 && 3528 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3529 has_super++; 3530 3531 return (has_super + ext4_group_first_block_no(sb, bg)); 3532 } 3533 3534 /** 3535 * ext4_get_stripe_size: Get the stripe size. 3536 * @sbi: In memory super block info 3537 * 3538 * If we have specified it via mount option, then 3539 * use the mount option value. If the value specified at mount time is 3540 * greater than the blocks per group use the super block value. 3541 * If the super block value is greater than blocks per group return 0. 3542 * Allocator needs it be less than blocks per group. 3543 * 3544 */ 3545 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3546 { 3547 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3548 unsigned long stripe_width = 3549 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3550 int ret; 3551 3552 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3553 ret = sbi->s_stripe; 3554 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3555 ret = stripe_width; 3556 else if (stride && stride <= sbi->s_blocks_per_group) 3557 ret = stride; 3558 else 3559 ret = 0; 3560 3561 /* 3562 * If the stripe width is 1, this makes no sense and 3563 * we set it to 0 to turn off stripe handling code. 3564 */ 3565 if (ret <= 1) 3566 ret = 0; 3567 3568 return ret; 3569 } 3570 3571 /* 3572 * Check whether this filesystem can be mounted based on 3573 * the features present and the RDONLY/RDWR mount requested. 3574 * Returns 1 if this filesystem can be mounted as requested, 3575 * 0 if it cannot be. 3576 */ 3577 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3578 { 3579 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3580 ext4_msg(sb, KERN_ERR, 3581 "Couldn't mount because of " 3582 "unsupported optional features (%x)", 3583 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3584 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3585 return 0; 3586 } 3587 3588 if (!IS_ENABLED(CONFIG_UNICODE) && ext4_has_feature_casefold(sb)) { 3589 ext4_msg(sb, KERN_ERR, 3590 "Filesystem with casefold feature cannot be " 3591 "mounted without CONFIG_UNICODE"); 3592 return 0; 3593 } 3594 3595 if (readonly) 3596 return 1; 3597 3598 if (ext4_has_feature_readonly(sb)) { 3599 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3600 sb->s_flags |= SB_RDONLY; 3601 return 1; 3602 } 3603 3604 /* Check that feature set is OK for a read-write mount */ 3605 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3606 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3607 "unsupported optional features (%x)", 3608 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3609 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3610 return 0; 3611 } 3612 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3613 ext4_msg(sb, KERN_ERR, 3614 "Can't support bigalloc feature without " 3615 "extents feature\n"); 3616 return 0; 3617 } 3618 3619 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3620 if (!readonly && (ext4_has_feature_quota(sb) || 3621 ext4_has_feature_project(sb))) { 3622 ext4_msg(sb, KERN_ERR, 3623 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3624 return 0; 3625 } 3626 #endif /* CONFIG_QUOTA */ 3627 return 1; 3628 } 3629 3630 /* 3631 * This function is called once a day if we have errors logged 3632 * on the file system 3633 */ 3634 static void print_daily_error_info(struct timer_list *t) 3635 { 3636 struct ext4_sb_info *sbi = timer_container_of(sbi, t, s_err_report); 3637 struct super_block *sb = sbi->s_sb; 3638 struct ext4_super_block *es = sbi->s_es; 3639 3640 if (es->s_error_count) 3641 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3642 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3643 le32_to_cpu(es->s_error_count)); 3644 if (es->s_first_error_time) { 3645 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3646 sb->s_id, 3647 ext4_get_tstamp(es, s_first_error_time), 3648 (int) sizeof(es->s_first_error_func), 3649 es->s_first_error_func, 3650 le32_to_cpu(es->s_first_error_line)); 3651 if (es->s_first_error_ino) 3652 printk(KERN_CONT ": inode %u", 3653 le32_to_cpu(es->s_first_error_ino)); 3654 if (es->s_first_error_block) 3655 printk(KERN_CONT ": block %llu", (unsigned long long) 3656 le64_to_cpu(es->s_first_error_block)); 3657 printk(KERN_CONT "\n"); 3658 } 3659 if (es->s_last_error_time) { 3660 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3661 sb->s_id, 3662 ext4_get_tstamp(es, s_last_error_time), 3663 (int) sizeof(es->s_last_error_func), 3664 es->s_last_error_func, 3665 le32_to_cpu(es->s_last_error_line)); 3666 if (es->s_last_error_ino) 3667 printk(KERN_CONT ": inode %u", 3668 le32_to_cpu(es->s_last_error_ino)); 3669 if (es->s_last_error_block) 3670 printk(KERN_CONT ": block %llu", (unsigned long long) 3671 le64_to_cpu(es->s_last_error_block)); 3672 printk(KERN_CONT "\n"); 3673 } 3674 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3675 } 3676 3677 /* Find next suitable group and run ext4_init_inode_table */ 3678 static int ext4_run_li_request(struct ext4_li_request *elr) 3679 { 3680 struct ext4_group_desc *gdp = NULL; 3681 struct super_block *sb = elr->lr_super; 3682 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3683 ext4_group_t group = elr->lr_next_group; 3684 unsigned int prefetch_ios = 0; 3685 int ret = 0; 3686 int nr = EXT4_SB(sb)->s_mb_prefetch; 3687 u64 start_time; 3688 3689 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3690 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios); 3691 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr); 3692 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr); 3693 if (group >= elr->lr_next_group) { 3694 ret = 1; 3695 if (elr->lr_first_not_zeroed != ngroups && 3696 !ext4_emergency_state(sb) && !sb_rdonly(sb) && 3697 test_opt(sb, INIT_INODE_TABLE)) { 3698 elr->lr_next_group = elr->lr_first_not_zeroed; 3699 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3700 ret = 0; 3701 } 3702 } 3703 return ret; 3704 } 3705 3706 for (; group < ngroups; group++) { 3707 gdp = ext4_get_group_desc(sb, group, NULL); 3708 if (!gdp) { 3709 ret = 1; 3710 break; 3711 } 3712 3713 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3714 break; 3715 } 3716 3717 if (group >= ngroups) 3718 ret = 1; 3719 3720 if (!ret) { 3721 start_time = ktime_get_ns(); 3722 ret = ext4_init_inode_table(sb, group, 3723 elr->lr_timeout ? 0 : 1); 3724 trace_ext4_lazy_itable_init(sb, group); 3725 if (elr->lr_timeout == 0) { 3726 elr->lr_timeout = nsecs_to_jiffies((ktime_get_ns() - start_time) * 3727 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3728 } 3729 elr->lr_next_sched = jiffies + elr->lr_timeout; 3730 elr->lr_next_group = group + 1; 3731 } 3732 return ret; 3733 } 3734 3735 /* 3736 * Remove lr_request from the list_request and free the 3737 * request structure. Should be called with li_list_mtx held 3738 */ 3739 static void ext4_remove_li_request(struct ext4_li_request *elr) 3740 { 3741 if (!elr) 3742 return; 3743 3744 list_del(&elr->lr_request); 3745 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3746 kfree(elr); 3747 } 3748 3749 static void ext4_unregister_li_request(struct super_block *sb) 3750 { 3751 mutex_lock(&ext4_li_mtx); 3752 if (!ext4_li_info) { 3753 mutex_unlock(&ext4_li_mtx); 3754 return; 3755 } 3756 3757 mutex_lock(&ext4_li_info->li_list_mtx); 3758 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3759 mutex_unlock(&ext4_li_info->li_list_mtx); 3760 mutex_unlock(&ext4_li_mtx); 3761 } 3762 3763 static struct task_struct *ext4_lazyinit_task; 3764 3765 /* 3766 * This is the function where ext4lazyinit thread lives. It walks 3767 * through the request list searching for next scheduled filesystem. 3768 * When such a fs is found, run the lazy initialization request 3769 * (ext4_rn_li_request) and keep track of the time spend in this 3770 * function. Based on that time we compute next schedule time of 3771 * the request. When walking through the list is complete, compute 3772 * next waking time and put itself into sleep. 3773 */ 3774 static int ext4_lazyinit_thread(void *arg) 3775 { 3776 struct ext4_lazy_init *eli = arg; 3777 struct list_head *pos, *n; 3778 struct ext4_li_request *elr; 3779 unsigned long next_wakeup, cur; 3780 3781 BUG_ON(NULL == eli); 3782 set_freezable(); 3783 3784 cont_thread: 3785 while (true) { 3786 bool next_wakeup_initialized = false; 3787 3788 next_wakeup = 0; 3789 mutex_lock(&eli->li_list_mtx); 3790 if (list_empty(&eli->li_request_list)) { 3791 mutex_unlock(&eli->li_list_mtx); 3792 goto exit_thread; 3793 } 3794 list_for_each_safe(pos, n, &eli->li_request_list) { 3795 int err = 0; 3796 int progress = 0; 3797 elr = list_entry(pos, struct ext4_li_request, 3798 lr_request); 3799 3800 if (time_before(jiffies, elr->lr_next_sched)) { 3801 if (!next_wakeup_initialized || 3802 time_before(elr->lr_next_sched, next_wakeup)) { 3803 next_wakeup = elr->lr_next_sched; 3804 next_wakeup_initialized = true; 3805 } 3806 continue; 3807 } 3808 if (down_read_trylock(&elr->lr_super->s_umount)) { 3809 if (sb_start_write_trylock(elr->lr_super)) { 3810 progress = 1; 3811 /* 3812 * We hold sb->s_umount, sb can not 3813 * be removed from the list, it is 3814 * now safe to drop li_list_mtx 3815 */ 3816 mutex_unlock(&eli->li_list_mtx); 3817 err = ext4_run_li_request(elr); 3818 sb_end_write(elr->lr_super); 3819 mutex_lock(&eli->li_list_mtx); 3820 n = pos->next; 3821 } 3822 up_read((&elr->lr_super->s_umount)); 3823 } 3824 /* error, remove the lazy_init job */ 3825 if (err) { 3826 ext4_remove_li_request(elr); 3827 continue; 3828 } 3829 if (!progress) { 3830 elr->lr_next_sched = jiffies + 3831 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3832 } 3833 if (!next_wakeup_initialized || 3834 time_before(elr->lr_next_sched, next_wakeup)) { 3835 next_wakeup = elr->lr_next_sched; 3836 next_wakeup_initialized = true; 3837 } 3838 } 3839 mutex_unlock(&eli->li_list_mtx); 3840 3841 try_to_freeze(); 3842 3843 cur = jiffies; 3844 if (!next_wakeup_initialized || time_after_eq(cur, next_wakeup)) { 3845 cond_resched(); 3846 continue; 3847 } 3848 3849 schedule_timeout_interruptible(next_wakeup - cur); 3850 3851 if (kthread_should_stop()) { 3852 ext4_clear_request_list(); 3853 goto exit_thread; 3854 } 3855 } 3856 3857 exit_thread: 3858 /* 3859 * It looks like the request list is empty, but we need 3860 * to check it under the li_list_mtx lock, to prevent any 3861 * additions into it, and of course we should lock ext4_li_mtx 3862 * to atomically free the list and ext4_li_info, because at 3863 * this point another ext4 filesystem could be registering 3864 * new one. 3865 */ 3866 mutex_lock(&ext4_li_mtx); 3867 mutex_lock(&eli->li_list_mtx); 3868 if (!list_empty(&eli->li_request_list)) { 3869 mutex_unlock(&eli->li_list_mtx); 3870 mutex_unlock(&ext4_li_mtx); 3871 goto cont_thread; 3872 } 3873 mutex_unlock(&eli->li_list_mtx); 3874 kfree(ext4_li_info); 3875 ext4_li_info = NULL; 3876 mutex_unlock(&ext4_li_mtx); 3877 3878 return 0; 3879 } 3880 3881 static void ext4_clear_request_list(void) 3882 { 3883 struct list_head *pos, *n; 3884 struct ext4_li_request *elr; 3885 3886 mutex_lock(&ext4_li_info->li_list_mtx); 3887 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3888 elr = list_entry(pos, struct ext4_li_request, 3889 lr_request); 3890 ext4_remove_li_request(elr); 3891 } 3892 mutex_unlock(&ext4_li_info->li_list_mtx); 3893 } 3894 3895 static int ext4_run_lazyinit_thread(void) 3896 { 3897 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3898 ext4_li_info, "ext4lazyinit"); 3899 if (IS_ERR(ext4_lazyinit_task)) { 3900 int err = PTR_ERR(ext4_lazyinit_task); 3901 ext4_clear_request_list(); 3902 kfree(ext4_li_info); 3903 ext4_li_info = NULL; 3904 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3905 "initialization thread\n", 3906 err); 3907 return err; 3908 } 3909 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3910 return 0; 3911 } 3912 3913 /* 3914 * Check whether it make sense to run itable init. thread or not. 3915 * If there is at least one uninitialized inode table, return 3916 * corresponding group number, else the loop goes through all 3917 * groups and return total number of groups. 3918 */ 3919 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3920 { 3921 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3922 struct ext4_group_desc *gdp = NULL; 3923 3924 if (!ext4_has_group_desc_csum(sb)) 3925 return ngroups; 3926 3927 for (group = 0; group < ngroups; group++) { 3928 gdp = ext4_get_group_desc(sb, group, NULL); 3929 if (!gdp) 3930 continue; 3931 3932 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3933 break; 3934 } 3935 3936 return group; 3937 } 3938 3939 static int ext4_li_info_new(void) 3940 { 3941 struct ext4_lazy_init *eli = NULL; 3942 3943 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3944 if (!eli) 3945 return -ENOMEM; 3946 3947 INIT_LIST_HEAD(&eli->li_request_list); 3948 mutex_init(&eli->li_list_mtx); 3949 3950 eli->li_state |= EXT4_LAZYINIT_QUIT; 3951 3952 ext4_li_info = eli; 3953 3954 return 0; 3955 } 3956 3957 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3958 ext4_group_t start) 3959 { 3960 struct ext4_li_request *elr; 3961 3962 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3963 if (!elr) 3964 return NULL; 3965 3966 elr->lr_super = sb; 3967 elr->lr_first_not_zeroed = start; 3968 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3969 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3970 elr->lr_next_group = start; 3971 } else { 3972 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3973 } 3974 3975 /* 3976 * Randomize first schedule time of the request to 3977 * spread the inode table initialization requests 3978 * better. 3979 */ 3980 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3981 return elr; 3982 } 3983 3984 int ext4_register_li_request(struct super_block *sb, 3985 ext4_group_t first_not_zeroed) 3986 { 3987 struct ext4_sb_info *sbi = EXT4_SB(sb); 3988 struct ext4_li_request *elr = NULL; 3989 ext4_group_t ngroups = sbi->s_groups_count; 3990 int ret = 0; 3991 3992 mutex_lock(&ext4_li_mtx); 3993 if (sbi->s_li_request != NULL) { 3994 /* 3995 * Reset timeout so it can be computed again, because 3996 * s_li_wait_mult might have changed. 3997 */ 3998 sbi->s_li_request->lr_timeout = 0; 3999 goto out; 4000 } 4001 4002 if (ext4_emergency_state(sb) || sb_rdonly(sb) || 4003 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 4004 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 4005 goto out; 4006 4007 elr = ext4_li_request_new(sb, first_not_zeroed); 4008 if (!elr) { 4009 ret = -ENOMEM; 4010 goto out; 4011 } 4012 4013 if (NULL == ext4_li_info) { 4014 ret = ext4_li_info_new(); 4015 if (ret) 4016 goto out; 4017 } 4018 4019 mutex_lock(&ext4_li_info->li_list_mtx); 4020 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 4021 mutex_unlock(&ext4_li_info->li_list_mtx); 4022 4023 sbi->s_li_request = elr; 4024 /* 4025 * set elr to NULL here since it has been inserted to 4026 * the request_list and the removal and free of it is 4027 * handled by ext4_clear_request_list from now on. 4028 */ 4029 elr = NULL; 4030 4031 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 4032 ret = ext4_run_lazyinit_thread(); 4033 if (ret) 4034 goto out; 4035 } 4036 out: 4037 mutex_unlock(&ext4_li_mtx); 4038 if (ret) 4039 kfree(elr); 4040 return ret; 4041 } 4042 4043 /* 4044 * We do not need to lock anything since this is called on 4045 * module unload. 4046 */ 4047 static void ext4_destroy_lazyinit_thread(void) 4048 { 4049 /* 4050 * If thread exited earlier 4051 * there's nothing to be done. 4052 */ 4053 if (!ext4_li_info || !ext4_lazyinit_task) 4054 return; 4055 4056 kthread_stop(ext4_lazyinit_task); 4057 } 4058 4059 static int set_journal_csum_feature_set(struct super_block *sb) 4060 { 4061 int ret = 1; 4062 int compat, incompat; 4063 struct ext4_sb_info *sbi = EXT4_SB(sb); 4064 4065 if (ext4_has_feature_metadata_csum(sb)) { 4066 /* journal checksum v3 */ 4067 compat = 0; 4068 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4069 } else { 4070 /* journal checksum v1 */ 4071 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4072 incompat = 0; 4073 } 4074 4075 jbd2_journal_clear_features(sbi->s_journal, 4076 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4077 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4078 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4079 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4080 ret = jbd2_journal_set_features(sbi->s_journal, 4081 compat, 0, 4082 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4083 incompat); 4084 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4085 ret = jbd2_journal_set_features(sbi->s_journal, 4086 compat, 0, 4087 incompat); 4088 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4089 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4090 } else { 4091 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4092 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4093 } 4094 4095 return ret; 4096 } 4097 4098 /* 4099 * Note: calculating the overhead so we can be compatible with 4100 * historical BSD practice is quite difficult in the face of 4101 * clusters/bigalloc. This is because multiple metadata blocks from 4102 * different block group can end up in the same allocation cluster. 4103 * Calculating the exact overhead in the face of clustered allocation 4104 * requires either O(all block bitmaps) in memory or O(number of block 4105 * groups**2) in time. We will still calculate the superblock for 4106 * older file systems --- and if we come across with a bigalloc file 4107 * system with zero in s_overhead_clusters the estimate will be close to 4108 * correct especially for very large cluster sizes --- but for newer 4109 * file systems, it's better to calculate this figure once at mkfs 4110 * time, and store it in the superblock. If the superblock value is 4111 * present (even for non-bigalloc file systems), we will use it. 4112 */ 4113 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4114 char *buf) 4115 { 4116 struct ext4_sb_info *sbi = EXT4_SB(sb); 4117 struct ext4_group_desc *gdp; 4118 ext4_fsblk_t first_block, last_block, b; 4119 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4120 int s, j, count = 0; 4121 int has_super = ext4_bg_has_super(sb, grp); 4122 4123 if (!ext4_has_feature_bigalloc(sb)) 4124 return (has_super + ext4_bg_num_gdb(sb, grp) + 4125 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4126 sbi->s_itb_per_group + 2); 4127 4128 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4129 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4130 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4131 for (i = 0; i < ngroups; i++) { 4132 gdp = ext4_get_group_desc(sb, i, NULL); 4133 b = ext4_block_bitmap(sb, gdp); 4134 if (b >= first_block && b <= last_block) { 4135 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4136 count++; 4137 } 4138 b = ext4_inode_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_table(sb, gdp); 4144 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4145 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4146 int c = EXT4_B2C(sbi, b - first_block); 4147 ext4_set_bit(c, buf); 4148 count++; 4149 } 4150 if (i != grp) 4151 continue; 4152 s = 0; 4153 if (ext4_bg_has_super(sb, grp)) { 4154 ext4_set_bit(s++, buf); 4155 count++; 4156 } 4157 j = ext4_bg_num_gdb(sb, grp); 4158 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4159 ext4_error(sb, "Invalid number of block group " 4160 "descriptor blocks: %d", j); 4161 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4162 } 4163 count += j; 4164 for (; j > 0; j--) 4165 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4166 } 4167 if (!count) 4168 return 0; 4169 return EXT4_CLUSTERS_PER_GROUP(sb) - 4170 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4171 } 4172 4173 /* 4174 * Compute the overhead and stash it in sbi->s_overhead 4175 */ 4176 int ext4_calculate_overhead(struct super_block *sb) 4177 { 4178 struct ext4_sb_info *sbi = EXT4_SB(sb); 4179 struct ext4_super_block *es = sbi->s_es; 4180 struct inode *j_inode; 4181 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4182 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4183 ext4_fsblk_t overhead = 0; 4184 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4185 4186 if (!buf) 4187 return -ENOMEM; 4188 4189 /* 4190 * Compute the overhead (FS structures). This is constant 4191 * for a given filesystem unless the number of block groups 4192 * changes so we cache the previous value until it does. 4193 */ 4194 4195 /* 4196 * All of the blocks before first_data_block are overhead 4197 */ 4198 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4199 4200 /* 4201 * Add the overhead found in each block group 4202 */ 4203 for (i = 0; i < ngroups; i++) { 4204 int blks; 4205 4206 blks = count_overhead(sb, i, buf); 4207 overhead += blks; 4208 if (blks) 4209 memset(buf, 0, PAGE_SIZE); 4210 cond_resched(); 4211 } 4212 4213 /* 4214 * Add the internal journal blocks whether the journal has been 4215 * loaded or not 4216 */ 4217 if (sbi->s_journal && !sbi->s_journal_bdev_file) 4218 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4219 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4220 /* j_inum for internal journal is non-zero */ 4221 j_inode = ext4_get_journal_inode(sb, j_inum); 4222 if (!IS_ERR(j_inode)) { 4223 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4224 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4225 iput(j_inode); 4226 } else { 4227 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4228 } 4229 } 4230 sbi->s_overhead = overhead; 4231 smp_wmb(); 4232 free_page((unsigned long) buf); 4233 return 0; 4234 } 4235 4236 static void ext4_set_resv_clusters(struct super_block *sb) 4237 { 4238 ext4_fsblk_t resv_clusters; 4239 struct ext4_sb_info *sbi = EXT4_SB(sb); 4240 4241 /* 4242 * There's no need to reserve anything when we aren't using extents. 4243 * The space estimates are exact, there are no unwritten extents, 4244 * hole punching doesn't need new metadata... This is needed especially 4245 * to keep ext2/3 backward compatibility. 4246 */ 4247 if (!ext4_has_feature_extents(sb)) 4248 return; 4249 /* 4250 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4251 * This should cover the situations where we can not afford to run 4252 * out of space like for example punch hole, or converting 4253 * unwritten extents in delalloc path. In most cases such 4254 * allocation would require 1, or 2 blocks, higher numbers are 4255 * very rare. 4256 */ 4257 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4258 sbi->s_cluster_bits); 4259 4260 do_div(resv_clusters, 50); 4261 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4262 4263 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4264 } 4265 4266 static const char *ext4_quota_mode(struct super_block *sb) 4267 { 4268 #ifdef CONFIG_QUOTA 4269 if (!ext4_quota_capable(sb)) 4270 return "none"; 4271 4272 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4273 return "journalled"; 4274 else 4275 return "writeback"; 4276 #else 4277 return "disabled"; 4278 #endif 4279 } 4280 4281 static void ext4_setup_csum_trigger(struct super_block *sb, 4282 enum ext4_journal_trigger_type type, 4283 void (*trigger)( 4284 struct jbd2_buffer_trigger_type *type, 4285 struct buffer_head *bh, 4286 void *mapped_data, 4287 size_t size)) 4288 { 4289 struct ext4_sb_info *sbi = EXT4_SB(sb); 4290 4291 sbi->s_journal_triggers[type].sb = sb; 4292 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4293 } 4294 4295 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4296 { 4297 if (!sbi) 4298 return; 4299 4300 kfree(sbi->s_blockgroup_lock); 4301 fs_put_dax(sbi->s_daxdev, NULL); 4302 kfree(sbi); 4303 } 4304 4305 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4306 { 4307 struct ext4_sb_info *sbi; 4308 4309 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4310 if (!sbi) 4311 return NULL; 4312 4313 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4314 NULL, NULL); 4315 4316 sbi->s_blockgroup_lock = 4317 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4318 4319 if (!sbi->s_blockgroup_lock) 4320 goto err_out; 4321 4322 sb->s_fs_info = sbi; 4323 sbi->s_sb = sb; 4324 return sbi; 4325 err_out: 4326 fs_put_dax(sbi->s_daxdev, NULL); 4327 kfree(sbi); 4328 return NULL; 4329 } 4330 4331 static void ext4_set_def_opts(struct super_block *sb, 4332 struct ext4_super_block *es) 4333 { 4334 unsigned long def_mount_opts; 4335 4336 /* Set defaults before we parse the mount options */ 4337 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4338 set_opt(sb, INIT_INODE_TABLE); 4339 if (def_mount_opts & EXT4_DEFM_DEBUG) 4340 set_opt(sb, DEBUG); 4341 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4342 set_opt(sb, GRPID); 4343 if (def_mount_opts & EXT4_DEFM_UID16) 4344 set_opt(sb, NO_UID32); 4345 /* xattr user namespace & acls are now defaulted on */ 4346 set_opt(sb, XATTR_USER); 4347 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4348 set_opt(sb, POSIX_ACL); 4349 #endif 4350 if (ext4_has_feature_fast_commit(sb)) 4351 set_opt2(sb, JOURNAL_FAST_COMMIT); 4352 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4353 if (ext4_has_feature_metadata_csum(sb)) 4354 set_opt(sb, JOURNAL_CHECKSUM); 4355 4356 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4357 set_opt(sb, JOURNAL_DATA); 4358 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4359 set_opt(sb, ORDERED_DATA); 4360 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4361 set_opt(sb, WRITEBACK_DATA); 4362 4363 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4364 set_opt(sb, ERRORS_PANIC); 4365 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4366 set_opt(sb, ERRORS_CONT); 4367 else 4368 set_opt(sb, ERRORS_RO); 4369 /* block_validity enabled by default; disable with noblock_validity */ 4370 set_opt(sb, BLOCK_VALIDITY); 4371 if (def_mount_opts & EXT4_DEFM_DISCARD) 4372 set_opt(sb, DISCARD); 4373 4374 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4375 set_opt(sb, BARRIER); 4376 4377 /* 4378 * enable delayed allocation by default 4379 * Use -o nodelalloc to turn it off 4380 */ 4381 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4382 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4383 set_opt(sb, DELALLOC); 4384 4385 if (sb->s_blocksize <= PAGE_SIZE) 4386 set_opt(sb, DIOREAD_NOLOCK); 4387 } 4388 4389 static int ext4_handle_clustersize(struct super_block *sb) 4390 { 4391 struct ext4_sb_info *sbi = EXT4_SB(sb); 4392 struct ext4_super_block *es = sbi->s_es; 4393 int clustersize; 4394 4395 /* Handle clustersize */ 4396 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4397 if (ext4_has_feature_bigalloc(sb)) { 4398 if (clustersize < sb->s_blocksize) { 4399 ext4_msg(sb, KERN_ERR, 4400 "cluster size (%d) smaller than " 4401 "block size (%lu)", clustersize, sb->s_blocksize); 4402 return -EINVAL; 4403 } 4404 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4405 le32_to_cpu(es->s_log_block_size); 4406 } else { 4407 if (clustersize != sb->s_blocksize) { 4408 ext4_msg(sb, KERN_ERR, 4409 "fragment/cluster size (%d) != " 4410 "block size (%lu)", clustersize, sb->s_blocksize); 4411 return -EINVAL; 4412 } 4413 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4414 ext4_msg(sb, KERN_ERR, 4415 "#blocks per group too big: %lu", 4416 sbi->s_blocks_per_group); 4417 return -EINVAL; 4418 } 4419 sbi->s_cluster_bits = 0; 4420 } 4421 sbi->s_clusters_per_group = le32_to_cpu(es->s_clusters_per_group); 4422 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4423 ext4_msg(sb, KERN_ERR, "#clusters per group too big: %lu", 4424 sbi->s_clusters_per_group); 4425 return -EINVAL; 4426 } 4427 if (sbi->s_blocks_per_group != 4428 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4429 ext4_msg(sb, KERN_ERR, 4430 "blocks per group (%lu) and clusters per group (%lu) inconsistent", 4431 sbi->s_blocks_per_group, sbi->s_clusters_per_group); 4432 return -EINVAL; 4433 } 4434 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4435 4436 /* Do we have standard group size of clustersize * 8 blocks ? */ 4437 if (sbi->s_blocks_per_group == clustersize << 3) 4438 set_opt2(sb, STD_GROUP_SIZE); 4439 4440 return 0; 4441 } 4442 4443 /* 4444 * ext4_atomic_write_init: Initializes filesystem min & max atomic write units. 4445 * With non-bigalloc filesystem awu will be based upon filesystem blocksize 4446 * & bdev awu units. 4447 * With bigalloc it will be based upon bigalloc cluster size & bdev awu units. 4448 * @sb: super block 4449 */ 4450 static void ext4_atomic_write_init(struct super_block *sb) 4451 { 4452 struct ext4_sb_info *sbi = EXT4_SB(sb); 4453 struct block_device *bdev = sb->s_bdev; 4454 unsigned int clustersize = EXT4_CLUSTER_SIZE(sb); 4455 4456 if (!bdev_can_atomic_write(bdev)) 4457 return; 4458 4459 if (!ext4_has_feature_extents(sb)) 4460 return; 4461 4462 sbi->s_awu_min = max(sb->s_blocksize, 4463 bdev_atomic_write_unit_min_bytes(bdev)); 4464 sbi->s_awu_max = min(clustersize, 4465 bdev_atomic_write_unit_max_bytes(bdev)); 4466 if (sbi->s_awu_min && sbi->s_awu_max && 4467 sbi->s_awu_min <= sbi->s_awu_max) { 4468 ext4_msg(sb, KERN_NOTICE, "Supports (experimental) DIO atomic writes awu_min: %u, awu_max: %u", 4469 sbi->s_awu_min, sbi->s_awu_max); 4470 } else { 4471 sbi->s_awu_min = 0; 4472 sbi->s_awu_max = 0; 4473 } 4474 } 4475 4476 static void ext4_fast_commit_init(struct super_block *sb) 4477 { 4478 struct ext4_sb_info *sbi = EXT4_SB(sb); 4479 4480 /* Initialize fast commit stuff */ 4481 atomic_set(&sbi->s_fc_subtid, 0); 4482 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4483 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4484 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4485 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4486 sbi->s_fc_bytes = 0; 4487 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4488 sbi->s_fc_ineligible_tid = 0; 4489 mutex_init(&sbi->s_fc_lock); 4490 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4491 sbi->s_fc_replay_state.fc_regions = NULL; 4492 sbi->s_fc_replay_state.fc_regions_size = 0; 4493 sbi->s_fc_replay_state.fc_regions_used = 0; 4494 sbi->s_fc_replay_state.fc_regions_valid = 0; 4495 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4496 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4497 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4498 } 4499 4500 static int ext4_inode_info_init(struct super_block *sb, 4501 struct ext4_super_block *es) 4502 { 4503 struct ext4_sb_info *sbi = EXT4_SB(sb); 4504 4505 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4506 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4507 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4508 } else { 4509 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4510 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4511 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4512 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4513 sbi->s_first_ino); 4514 return -EINVAL; 4515 } 4516 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4517 (!is_power_of_2(sbi->s_inode_size)) || 4518 (sbi->s_inode_size > sb->s_blocksize)) { 4519 ext4_msg(sb, KERN_ERR, 4520 "unsupported inode size: %d", 4521 sbi->s_inode_size); 4522 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4523 return -EINVAL; 4524 } 4525 /* 4526 * i_atime_extra is the last extra field available for 4527 * [acm]times in struct ext4_inode. Checking for that 4528 * field should suffice to ensure we have extra space 4529 * for all three. 4530 */ 4531 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4532 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4533 sb->s_time_gran = 1; 4534 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4535 } else { 4536 sb->s_time_gran = NSEC_PER_SEC; 4537 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4538 } 4539 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4540 } 4541 4542 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4543 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4544 EXT4_GOOD_OLD_INODE_SIZE; 4545 if (ext4_has_feature_extra_isize(sb)) { 4546 unsigned v, max = (sbi->s_inode_size - 4547 EXT4_GOOD_OLD_INODE_SIZE); 4548 4549 v = le16_to_cpu(es->s_want_extra_isize); 4550 if (v > max) { 4551 ext4_msg(sb, KERN_ERR, 4552 "bad s_want_extra_isize: %d", v); 4553 return -EINVAL; 4554 } 4555 if (sbi->s_want_extra_isize < v) 4556 sbi->s_want_extra_isize = v; 4557 4558 v = le16_to_cpu(es->s_min_extra_isize); 4559 if (v > max) { 4560 ext4_msg(sb, KERN_ERR, 4561 "bad s_min_extra_isize: %d", v); 4562 return -EINVAL; 4563 } 4564 if (sbi->s_want_extra_isize < v) 4565 sbi->s_want_extra_isize = v; 4566 } 4567 } 4568 4569 return 0; 4570 } 4571 4572 #if IS_ENABLED(CONFIG_UNICODE) 4573 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4574 { 4575 const struct ext4_sb_encodings *encoding_info; 4576 struct unicode_map *encoding; 4577 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4578 4579 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4580 return 0; 4581 4582 encoding_info = ext4_sb_read_encoding(es); 4583 if (!encoding_info) { 4584 ext4_msg(sb, KERN_ERR, 4585 "Encoding requested by superblock is unknown"); 4586 return -EINVAL; 4587 } 4588 4589 encoding = utf8_load(encoding_info->version); 4590 if (IS_ERR(encoding)) { 4591 ext4_msg(sb, KERN_ERR, 4592 "can't mount with superblock charset: %s-%u.%u.%u " 4593 "not supported by the kernel. flags: 0x%x.", 4594 encoding_info->name, 4595 unicode_major(encoding_info->version), 4596 unicode_minor(encoding_info->version), 4597 unicode_rev(encoding_info->version), 4598 encoding_flags); 4599 return -EINVAL; 4600 } 4601 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4602 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4603 unicode_major(encoding_info->version), 4604 unicode_minor(encoding_info->version), 4605 unicode_rev(encoding_info->version), 4606 encoding_flags); 4607 4608 sb->s_encoding = encoding; 4609 sb->s_encoding_flags = encoding_flags; 4610 4611 return 0; 4612 } 4613 #else 4614 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4615 { 4616 return 0; 4617 } 4618 #endif 4619 4620 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4621 { 4622 struct ext4_sb_info *sbi = EXT4_SB(sb); 4623 4624 /* Warn if metadata_csum and gdt_csum are both set. */ 4625 if (ext4_has_feature_metadata_csum(sb) && 4626 ext4_has_feature_gdt_csum(sb)) 4627 ext4_warning(sb, "metadata_csum and uninit_bg are " 4628 "redundant flags; please run fsck."); 4629 4630 /* Check for a known checksum algorithm */ 4631 if (!ext4_verify_csum_type(sb, es)) { 4632 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4633 "unknown checksum algorithm."); 4634 return -EINVAL; 4635 } 4636 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4637 ext4_orphan_file_block_trigger); 4638 4639 /* Check superblock checksum */ 4640 if (!ext4_superblock_csum_verify(sb, es)) { 4641 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4642 "invalid superblock checksum. Run e2fsck?"); 4643 return -EFSBADCRC; 4644 } 4645 4646 /* Precompute checksum seed for all metadata */ 4647 if (ext4_has_feature_csum_seed(sb)) 4648 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4649 else if (ext4_has_feature_metadata_csum(sb) || 4650 ext4_has_feature_ea_inode(sb)) 4651 sbi->s_csum_seed = ext4_chksum(~0, es->s_uuid, 4652 sizeof(es->s_uuid)); 4653 return 0; 4654 } 4655 4656 static int ext4_check_feature_compatibility(struct super_block *sb, 4657 struct ext4_super_block *es, 4658 int silent) 4659 { 4660 struct ext4_sb_info *sbi = EXT4_SB(sb); 4661 4662 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4663 (ext4_has_compat_features(sb) || 4664 ext4_has_ro_compat_features(sb) || 4665 ext4_has_incompat_features(sb))) 4666 ext4_msg(sb, KERN_WARNING, 4667 "feature flags set on rev 0 fs, " 4668 "running e2fsck is recommended"); 4669 4670 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4671 set_opt2(sb, HURD_COMPAT); 4672 if (ext4_has_feature_64bit(sb)) { 4673 ext4_msg(sb, KERN_ERR, 4674 "The Hurd can't support 64-bit file systems"); 4675 return -EINVAL; 4676 } 4677 4678 /* 4679 * ea_inode feature uses l_i_version field which is not 4680 * available in HURD_COMPAT mode. 4681 */ 4682 if (ext4_has_feature_ea_inode(sb)) { 4683 ext4_msg(sb, KERN_ERR, 4684 "ea_inode feature is not supported for Hurd"); 4685 return -EINVAL; 4686 } 4687 } 4688 4689 if (IS_EXT2_SB(sb)) { 4690 if (ext2_feature_set_ok(sb)) 4691 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4692 "using the ext4 subsystem"); 4693 else { 4694 /* 4695 * If we're probing be silent, if this looks like 4696 * it's actually an ext[34] filesystem. 4697 */ 4698 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4699 return -EINVAL; 4700 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4701 "to feature incompatibilities"); 4702 return -EINVAL; 4703 } 4704 } 4705 4706 if (IS_EXT3_SB(sb)) { 4707 if (ext3_feature_set_ok(sb)) 4708 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4709 "using the ext4 subsystem"); 4710 else { 4711 /* 4712 * If we're probing be silent, if this looks like 4713 * it's actually an ext4 filesystem. 4714 */ 4715 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4716 return -EINVAL; 4717 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4718 "to feature incompatibilities"); 4719 return -EINVAL; 4720 } 4721 } 4722 4723 /* 4724 * Check feature flags regardless of the revision level, since we 4725 * previously didn't change the revision level when setting the flags, 4726 * so there is a chance incompat flags are set on a rev 0 filesystem. 4727 */ 4728 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4729 return -EINVAL; 4730 4731 if (sbi->s_daxdev) { 4732 if (sb->s_blocksize == PAGE_SIZE) 4733 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4734 else 4735 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 4736 } 4737 4738 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4739 if (ext4_has_feature_inline_data(sb)) { 4740 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4741 " that may contain inline data"); 4742 return -EINVAL; 4743 } 4744 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4745 ext4_msg(sb, KERN_ERR, 4746 "DAX unsupported by block device."); 4747 return -EINVAL; 4748 } 4749 } 4750 4751 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4752 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4753 es->s_encryption_level); 4754 return -EINVAL; 4755 } 4756 4757 return 0; 4758 } 4759 4760 static int ext4_check_geometry(struct super_block *sb, 4761 struct ext4_super_block *es) 4762 { 4763 struct ext4_sb_info *sbi = EXT4_SB(sb); 4764 __u64 blocks_count; 4765 int err; 4766 4767 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 4768 ext4_msg(sb, KERN_ERR, 4769 "Number of reserved GDT blocks insanely large: %d", 4770 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4771 return -EINVAL; 4772 } 4773 /* 4774 * Test whether we have more sectors than will fit in sector_t, 4775 * and whether the max offset is addressable by the page cache. 4776 */ 4777 err = generic_check_addressable(sb->s_blocksize_bits, 4778 ext4_blocks_count(es)); 4779 if (err) { 4780 ext4_msg(sb, KERN_ERR, "filesystem" 4781 " too large to mount safely on this system"); 4782 return err; 4783 } 4784 4785 /* check blocks count against device size */ 4786 blocks_count = sb_bdev_nr_blocks(sb); 4787 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4788 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4789 "exceeds size of device (%llu blocks)", 4790 ext4_blocks_count(es), blocks_count); 4791 return -EINVAL; 4792 } 4793 4794 /* 4795 * It makes no sense for the first data block to be beyond the end 4796 * of the filesystem. 4797 */ 4798 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4799 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4800 "block %u is beyond end of filesystem (%llu)", 4801 le32_to_cpu(es->s_first_data_block), 4802 ext4_blocks_count(es)); 4803 return -EINVAL; 4804 } 4805 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4806 (sbi->s_cluster_ratio == 1)) { 4807 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4808 "block is 0 with a 1k block and cluster size"); 4809 return -EINVAL; 4810 } 4811 4812 blocks_count = (ext4_blocks_count(es) - 4813 le32_to_cpu(es->s_first_data_block) + 4814 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4815 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4816 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4817 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4818 "(block count %llu, first data block %u, " 4819 "blocks per group %lu)", blocks_count, 4820 ext4_blocks_count(es), 4821 le32_to_cpu(es->s_first_data_block), 4822 EXT4_BLOCKS_PER_GROUP(sb)); 4823 return -EINVAL; 4824 } 4825 sbi->s_groups_count = blocks_count; 4826 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4827 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4828 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4829 le32_to_cpu(es->s_inodes_count)) { 4830 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4831 le32_to_cpu(es->s_inodes_count), 4832 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4833 return -EINVAL; 4834 } 4835 4836 return 0; 4837 } 4838 4839 static int ext4_group_desc_init(struct super_block *sb, 4840 struct ext4_super_block *es, 4841 ext4_fsblk_t logical_sb_block, 4842 ext4_group_t *first_not_zeroed) 4843 { 4844 struct ext4_sb_info *sbi = EXT4_SB(sb); 4845 unsigned int db_count; 4846 ext4_fsblk_t block; 4847 int i; 4848 4849 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4850 EXT4_DESC_PER_BLOCK(sb); 4851 if (ext4_has_feature_meta_bg(sb)) { 4852 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4853 ext4_msg(sb, KERN_WARNING, 4854 "first meta block group too large: %u " 4855 "(group descriptor block count %u)", 4856 le32_to_cpu(es->s_first_meta_bg), db_count); 4857 return -EINVAL; 4858 } 4859 } 4860 rcu_assign_pointer(sbi->s_group_desc, 4861 kvmalloc_array(db_count, 4862 sizeof(struct buffer_head *), 4863 GFP_KERNEL)); 4864 if (sbi->s_group_desc == NULL) { 4865 ext4_msg(sb, KERN_ERR, "not enough memory"); 4866 return -ENOMEM; 4867 } 4868 4869 bgl_lock_init(sbi->s_blockgroup_lock); 4870 4871 /* Pre-read the descriptors into the buffer cache */ 4872 for (i = 0; i < db_count; i++) { 4873 block = descriptor_loc(sb, logical_sb_block, i); 4874 ext4_sb_breadahead_unmovable(sb, block); 4875 } 4876 4877 for (i = 0; i < db_count; i++) { 4878 struct buffer_head *bh; 4879 4880 block = descriptor_loc(sb, logical_sb_block, i); 4881 bh = ext4_sb_bread_unmovable(sb, block); 4882 if (IS_ERR(bh)) { 4883 ext4_msg(sb, KERN_ERR, 4884 "can't read group descriptor %d", i); 4885 sbi->s_gdb_count = i; 4886 return PTR_ERR(bh); 4887 } 4888 rcu_read_lock(); 4889 rcu_dereference(sbi->s_group_desc)[i] = bh; 4890 rcu_read_unlock(); 4891 } 4892 sbi->s_gdb_count = db_count; 4893 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4894 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4895 return -EFSCORRUPTED; 4896 } 4897 4898 return 0; 4899 } 4900 4901 static int ext4_load_and_init_journal(struct super_block *sb, 4902 struct ext4_super_block *es, 4903 struct ext4_fs_context *ctx) 4904 { 4905 struct ext4_sb_info *sbi = EXT4_SB(sb); 4906 int err; 4907 4908 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4909 if (err) 4910 return err; 4911 4912 if (ext4_has_feature_64bit(sb) && 4913 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4914 JBD2_FEATURE_INCOMPAT_64BIT)) { 4915 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4916 goto out; 4917 } 4918 4919 if (!set_journal_csum_feature_set(sb)) { 4920 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4921 "feature set"); 4922 goto out; 4923 } 4924 4925 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4926 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4927 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4928 ext4_msg(sb, KERN_ERR, 4929 "Failed to set fast commit journal feature"); 4930 goto out; 4931 } 4932 4933 /* We have now updated the journal if required, so we can 4934 * validate the data journaling mode. */ 4935 switch (test_opt(sb, DATA_FLAGS)) { 4936 case 0: 4937 /* No mode set, assume a default based on the journal 4938 * capabilities: ORDERED_DATA if the journal can 4939 * cope, else JOURNAL_DATA 4940 */ 4941 if (jbd2_journal_check_available_features 4942 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4943 set_opt(sb, ORDERED_DATA); 4944 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4945 } else { 4946 set_opt(sb, JOURNAL_DATA); 4947 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4948 } 4949 break; 4950 4951 case EXT4_MOUNT_ORDERED_DATA: 4952 case EXT4_MOUNT_WRITEBACK_DATA: 4953 if (!jbd2_journal_check_available_features 4954 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4955 ext4_msg(sb, KERN_ERR, "Journal does not support " 4956 "requested data journaling mode"); 4957 goto out; 4958 } 4959 break; 4960 default: 4961 break; 4962 } 4963 4964 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4965 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4966 ext4_msg(sb, KERN_ERR, "can't mount with " 4967 "journal_async_commit in data=ordered mode"); 4968 goto out; 4969 } 4970 4971 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4972 4973 sbi->s_journal->j_submit_inode_data_buffers = 4974 ext4_journal_submit_inode_data_buffers; 4975 sbi->s_journal->j_finish_inode_data_buffers = 4976 ext4_journal_finish_inode_data_buffers; 4977 4978 return 0; 4979 4980 out: 4981 ext4_journal_destroy(sbi, sbi->s_journal); 4982 return -EINVAL; 4983 } 4984 4985 static int ext4_check_journal_data_mode(struct super_block *sb) 4986 { 4987 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4988 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4989 "data=journal disables delayed allocation, " 4990 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 4991 /* can't mount with both data=journal and dioread_nolock. */ 4992 clear_opt(sb, DIOREAD_NOLOCK); 4993 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4994 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4995 ext4_msg(sb, KERN_ERR, "can't mount with " 4996 "both data=journal and delalloc"); 4997 return -EINVAL; 4998 } 4999 if (test_opt(sb, DAX_ALWAYS)) { 5000 ext4_msg(sb, KERN_ERR, "can't mount with " 5001 "both data=journal and dax"); 5002 return -EINVAL; 5003 } 5004 if (ext4_has_feature_encrypt(sb)) { 5005 ext4_msg(sb, KERN_WARNING, 5006 "encrypted files will use data=ordered " 5007 "instead of data journaling mode"); 5008 } 5009 if (test_opt(sb, DELALLOC)) 5010 clear_opt(sb, DELALLOC); 5011 } else { 5012 sb->s_iflags |= SB_I_CGROUPWB; 5013 } 5014 5015 return 0; 5016 } 5017 5018 static const char *ext4_has_journal_option(struct super_block *sb) 5019 { 5020 struct ext4_sb_info *sbi = EXT4_SB(sb); 5021 5022 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) 5023 return "journal_async_commit"; 5024 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) 5025 return "journal_checksum"; 5026 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 5027 return "commit="; 5028 if (EXT4_MOUNT_DATA_FLAGS & 5029 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) 5030 return "data="; 5031 if (test_opt(sb, DATA_ERR_ABORT)) 5032 return "data_err=abort"; 5033 return NULL; 5034 } 5035 5036 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 5037 int silent) 5038 { 5039 struct ext4_sb_info *sbi = EXT4_SB(sb); 5040 struct ext4_super_block *es; 5041 ext4_fsblk_t logical_sb_block; 5042 unsigned long offset = 0; 5043 struct buffer_head *bh; 5044 int ret = -EINVAL; 5045 int blocksize; 5046 5047 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 5048 if (!blocksize) { 5049 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 5050 return -EINVAL; 5051 } 5052 5053 /* 5054 * The ext4 superblock will not be buffer aligned for other than 1kB 5055 * block sizes. We need to calculate the offset from buffer start. 5056 */ 5057 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 5058 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5059 offset = do_div(logical_sb_block, blocksize); 5060 } else { 5061 logical_sb_block = sbi->s_sb_block; 5062 } 5063 5064 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5065 if (IS_ERR(bh)) { 5066 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 5067 return PTR_ERR(bh); 5068 } 5069 /* 5070 * Note: s_es must be initialized as soon as possible because 5071 * some ext4 macro-instructions depend on its value 5072 */ 5073 es = (struct ext4_super_block *) (bh->b_data + offset); 5074 sbi->s_es = es; 5075 sb->s_magic = le16_to_cpu(es->s_magic); 5076 if (sb->s_magic != EXT4_SUPER_MAGIC) { 5077 if (!silent) 5078 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5079 goto out; 5080 } 5081 5082 if (le32_to_cpu(es->s_log_block_size) > 5083 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5084 ext4_msg(sb, KERN_ERR, 5085 "Invalid log block size: %u", 5086 le32_to_cpu(es->s_log_block_size)); 5087 goto out; 5088 } 5089 if (le32_to_cpu(es->s_log_cluster_size) > 5090 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5091 ext4_msg(sb, KERN_ERR, 5092 "Invalid log cluster size: %u", 5093 le32_to_cpu(es->s_log_cluster_size)); 5094 goto out; 5095 } 5096 5097 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 5098 5099 /* 5100 * If the default block size is not the same as the real block size, 5101 * we need to reload it. 5102 */ 5103 if (sb->s_blocksize == blocksize) { 5104 *lsb = logical_sb_block; 5105 sbi->s_sbh = bh; 5106 return 0; 5107 } 5108 5109 /* 5110 * bh must be released before kill_bdev(), otherwise 5111 * it won't be freed and its page also. kill_bdev() 5112 * is called by sb_set_blocksize(). 5113 */ 5114 brelse(bh); 5115 /* Validate the filesystem blocksize */ 5116 if (!sb_set_blocksize(sb, blocksize)) { 5117 ext4_msg(sb, KERN_ERR, "bad block size %d", 5118 blocksize); 5119 bh = NULL; 5120 goto out; 5121 } 5122 5123 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5124 offset = do_div(logical_sb_block, blocksize); 5125 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5126 if (IS_ERR(bh)) { 5127 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5128 ret = PTR_ERR(bh); 5129 bh = NULL; 5130 goto out; 5131 } 5132 es = (struct ext4_super_block *)(bh->b_data + offset); 5133 sbi->s_es = es; 5134 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5135 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5136 goto out; 5137 } 5138 *lsb = logical_sb_block; 5139 sbi->s_sbh = bh; 5140 return 0; 5141 out: 5142 brelse(bh); 5143 return ret; 5144 } 5145 5146 static int ext4_hash_info_init(struct super_block *sb) 5147 { 5148 struct ext4_sb_info *sbi = EXT4_SB(sb); 5149 struct ext4_super_block *es = sbi->s_es; 5150 unsigned int i; 5151 5152 sbi->s_def_hash_version = es->s_def_hash_version; 5153 5154 if (sbi->s_def_hash_version > DX_HASH_LAST) { 5155 ext4_msg(sb, KERN_ERR, 5156 "Invalid default hash set in the superblock"); 5157 return -EINVAL; 5158 } else if (sbi->s_def_hash_version == DX_HASH_SIPHASH) { 5159 ext4_msg(sb, KERN_ERR, 5160 "SIPHASH is not a valid default hash value"); 5161 return -EINVAL; 5162 } 5163 5164 for (i = 0; i < 4; i++) 5165 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5166 5167 if (ext4_has_feature_dir_index(sb)) { 5168 i = le32_to_cpu(es->s_flags); 5169 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5170 sbi->s_hash_unsigned = 3; 5171 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5172 #ifdef __CHAR_UNSIGNED__ 5173 if (!sb_rdonly(sb)) 5174 es->s_flags |= 5175 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5176 sbi->s_hash_unsigned = 3; 5177 #else 5178 if (!sb_rdonly(sb)) 5179 es->s_flags |= 5180 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5181 #endif 5182 } 5183 } 5184 return 0; 5185 } 5186 5187 static int ext4_block_group_meta_init(struct super_block *sb, int silent) 5188 { 5189 struct ext4_sb_info *sbi = EXT4_SB(sb); 5190 struct ext4_super_block *es = sbi->s_es; 5191 int has_huge_files; 5192 5193 has_huge_files = ext4_has_feature_huge_file(sb); 5194 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5195 has_huge_files); 5196 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5197 5198 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5199 if (ext4_has_feature_64bit(sb)) { 5200 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5201 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5202 !is_power_of_2(sbi->s_desc_size)) { 5203 ext4_msg(sb, KERN_ERR, 5204 "unsupported descriptor size %lu", 5205 sbi->s_desc_size); 5206 return -EINVAL; 5207 } 5208 } else 5209 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5210 5211 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5212 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5213 5214 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5215 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5216 if (!silent) 5217 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5218 return -EINVAL; 5219 } 5220 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5221 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5222 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5223 sbi->s_inodes_per_group); 5224 return -EINVAL; 5225 } 5226 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5227 sbi->s_inodes_per_block; 5228 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5229 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5230 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5231 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5232 5233 return 0; 5234 } 5235 5236 /* 5237 * It's hard to get stripe aligned blocks if stripe is not aligned with 5238 * cluster, just disable stripe and alert user to simplify code and avoid 5239 * stripe aligned allocation which will rarely succeed. 5240 */ 5241 static bool ext4_is_stripe_incompatible(struct super_block *sb, unsigned long stripe) 5242 { 5243 struct ext4_sb_info *sbi = EXT4_SB(sb); 5244 return (stripe > 0 && sbi->s_cluster_ratio > 1 && 5245 stripe % sbi->s_cluster_ratio != 0); 5246 } 5247 5248 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5249 { 5250 struct ext4_super_block *es = NULL; 5251 struct ext4_sb_info *sbi = EXT4_SB(sb); 5252 ext4_fsblk_t logical_sb_block; 5253 struct inode *root; 5254 int needs_recovery; 5255 int err; 5256 ext4_group_t first_not_zeroed; 5257 struct ext4_fs_context *ctx = fc->fs_private; 5258 int silent = fc->sb_flags & SB_SILENT; 5259 5260 /* Set defaults for the variables that will be set during parsing */ 5261 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5262 ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO; 5263 5264 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5265 sbi->s_sectors_written_start = 5266 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5267 5268 err = ext4_load_super(sb, &logical_sb_block, silent); 5269 if (err) 5270 goto out_fail; 5271 5272 es = sbi->s_es; 5273 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5274 5275 err = ext4_init_metadata_csum(sb, es); 5276 if (err) 5277 goto failed_mount; 5278 5279 ext4_set_def_opts(sb, es); 5280 5281 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5282 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5283 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5284 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5285 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5286 sbi->s_sb_update_kb = EXT4_DEF_SB_UPDATE_INTERVAL_KB; 5287 sbi->s_sb_update_sec = EXT4_DEF_SB_UPDATE_INTERVAL_SEC; 5288 5289 /* 5290 * set default s_li_wait_mult for lazyinit, for the case there is 5291 * no mount option specified. 5292 */ 5293 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5294 5295 err = ext4_inode_info_init(sb, es); 5296 if (err) 5297 goto failed_mount; 5298 5299 err = parse_apply_sb_mount_options(sb, ctx); 5300 if (err < 0) 5301 goto failed_mount; 5302 5303 sbi->s_def_mount_opt = sbi->s_mount_opt; 5304 sbi->s_def_mount_opt2 = sbi->s_mount_opt2; 5305 5306 err = ext4_check_opt_consistency(fc, sb); 5307 if (err < 0) 5308 goto failed_mount; 5309 5310 ext4_apply_options(fc, sb); 5311 5312 err = ext4_encoding_init(sb, es); 5313 if (err) 5314 goto failed_mount; 5315 5316 err = ext4_check_journal_data_mode(sb); 5317 if (err) 5318 goto failed_mount; 5319 5320 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5321 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5322 5323 /* i_version is always enabled now */ 5324 sb->s_flags |= SB_I_VERSION; 5325 5326 /* HSM events are allowed by default. */ 5327 sb->s_iflags |= SB_I_ALLOW_HSM; 5328 5329 err = ext4_check_feature_compatibility(sb, es, silent); 5330 if (err) 5331 goto failed_mount; 5332 5333 err = ext4_block_group_meta_init(sb, silent); 5334 if (err) 5335 goto failed_mount; 5336 5337 err = ext4_hash_info_init(sb); 5338 if (err) 5339 goto failed_mount; 5340 5341 err = ext4_handle_clustersize(sb); 5342 if (err) 5343 goto failed_mount; 5344 5345 err = ext4_check_geometry(sb, es); 5346 if (err) 5347 goto failed_mount; 5348 5349 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5350 spin_lock_init(&sbi->s_error_lock); 5351 INIT_WORK(&sbi->s_sb_upd_work, update_super_work); 5352 5353 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5354 if (err) 5355 goto failed_mount3; 5356 5357 err = ext4_es_register_shrinker(sbi); 5358 if (err) 5359 goto failed_mount3; 5360 5361 sbi->s_stripe = ext4_get_stripe_size(sbi); 5362 if (ext4_is_stripe_incompatible(sb, sbi->s_stripe)) { 5363 ext4_msg(sb, KERN_WARNING, 5364 "stripe (%lu) is not aligned with cluster size (%u), " 5365 "stripe is disabled", 5366 sbi->s_stripe, sbi->s_cluster_ratio); 5367 sbi->s_stripe = 0; 5368 } 5369 sbi->s_extent_max_zeroout_kb = 32; 5370 5371 /* 5372 * set up enough so that it can read an inode 5373 */ 5374 sb->s_op = &ext4_sops; 5375 sb->s_export_op = &ext4_export_ops; 5376 sb->s_xattr = ext4_xattr_handlers; 5377 #ifdef CONFIG_FS_ENCRYPTION 5378 sb->s_cop = &ext4_cryptops; 5379 #endif 5380 #ifdef CONFIG_FS_VERITY 5381 sb->s_vop = &ext4_verityops; 5382 #endif 5383 #ifdef CONFIG_QUOTA 5384 sb->dq_op = &ext4_quota_operations; 5385 if (ext4_has_feature_quota(sb)) 5386 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5387 else 5388 sb->s_qcop = &ext4_qctl_operations; 5389 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5390 #endif 5391 super_set_uuid(sb, es->s_uuid, sizeof(es->s_uuid)); 5392 super_set_sysfs_name_bdev(sb); 5393 5394 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5395 mutex_init(&sbi->s_orphan_lock); 5396 5397 spin_lock_init(&sbi->s_bdev_wb_lock); 5398 5399 ext4_atomic_write_init(sb); 5400 ext4_fast_commit_init(sb); 5401 5402 sb->s_root = NULL; 5403 5404 needs_recovery = (es->s_last_orphan != 0 || 5405 ext4_has_feature_orphan_present(sb) || 5406 ext4_has_feature_journal_needs_recovery(sb)); 5407 5408 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) { 5409 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)); 5410 if (err) 5411 goto failed_mount3a; 5412 } 5413 5414 err = -EINVAL; 5415 /* 5416 * The first inode we look at is the journal inode. Don't try 5417 * root first: it may be modified in the journal! 5418 */ 5419 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5420 err = ext4_load_and_init_journal(sb, es, ctx); 5421 if (err) 5422 goto failed_mount3a; 5423 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5424 ext4_has_feature_journal_needs_recovery(sb)) { 5425 ext4_msg(sb, KERN_ERR, "required journal recovery " 5426 "suppressed and not mounted read-only"); 5427 goto failed_mount3a; 5428 } else { 5429 const char *journal_option; 5430 5431 /* Nojournal mode, all journal mount options are illegal */ 5432 journal_option = ext4_has_journal_option(sb); 5433 if (journal_option != NULL) { 5434 ext4_msg(sb, KERN_ERR, 5435 "can't mount with %s, fs mounted w/o journal", 5436 journal_option); 5437 goto failed_mount3a; 5438 } 5439 5440 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5441 clear_opt(sb, JOURNAL_CHECKSUM); 5442 clear_opt(sb, DATA_FLAGS); 5443 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5444 sbi->s_journal = NULL; 5445 needs_recovery = 0; 5446 } 5447 5448 if (!test_opt(sb, NO_MBCACHE)) { 5449 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5450 if (!sbi->s_ea_block_cache) { 5451 ext4_msg(sb, KERN_ERR, 5452 "Failed to create ea_block_cache"); 5453 err = -EINVAL; 5454 goto failed_mount_wq; 5455 } 5456 5457 if (ext4_has_feature_ea_inode(sb)) { 5458 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5459 if (!sbi->s_ea_inode_cache) { 5460 ext4_msg(sb, KERN_ERR, 5461 "Failed to create ea_inode_cache"); 5462 err = -EINVAL; 5463 goto failed_mount_wq; 5464 } 5465 } 5466 } 5467 5468 /* 5469 * Get the # of file system overhead blocks from the 5470 * superblock if present. 5471 */ 5472 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5473 /* ignore the precalculated value if it is ridiculous */ 5474 if (sbi->s_overhead > ext4_blocks_count(es)) 5475 sbi->s_overhead = 0; 5476 /* 5477 * If the bigalloc feature is not enabled recalculating the 5478 * overhead doesn't take long, so we might as well just redo 5479 * it to make sure we are using the correct value. 5480 */ 5481 if (!ext4_has_feature_bigalloc(sb)) 5482 sbi->s_overhead = 0; 5483 if (sbi->s_overhead == 0) { 5484 err = ext4_calculate_overhead(sb); 5485 if (err) 5486 goto failed_mount_wq; 5487 } 5488 5489 /* 5490 * The maximum number of concurrent works can be high and 5491 * concurrency isn't really necessary. Limit it to 1. 5492 */ 5493 EXT4_SB(sb)->rsv_conversion_wq = 5494 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5495 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5496 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5497 err = -ENOMEM; 5498 goto failed_mount4; 5499 } 5500 5501 /* 5502 * The jbd2_journal_load will have done any necessary log recovery, 5503 * so we can safely mount the rest of the filesystem now. 5504 */ 5505 5506 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5507 if (IS_ERR(root)) { 5508 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5509 err = PTR_ERR(root); 5510 root = NULL; 5511 goto failed_mount4; 5512 } 5513 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5514 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5515 iput(root); 5516 err = -EFSCORRUPTED; 5517 goto failed_mount4; 5518 } 5519 5520 generic_set_sb_d_ops(sb); 5521 sb->s_root = d_make_root(root); 5522 if (!sb->s_root) { 5523 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5524 err = -ENOMEM; 5525 goto failed_mount4; 5526 } 5527 5528 err = ext4_setup_super(sb, es, sb_rdonly(sb)); 5529 if (err == -EROFS) { 5530 sb->s_flags |= SB_RDONLY; 5531 } else if (err) 5532 goto failed_mount4a; 5533 5534 ext4_set_resv_clusters(sb); 5535 5536 if (test_opt(sb, BLOCK_VALIDITY)) { 5537 err = ext4_setup_system_zone(sb); 5538 if (err) { 5539 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5540 "zone (%d)", err); 5541 goto failed_mount4a; 5542 } 5543 } 5544 ext4_fc_replay_cleanup(sb); 5545 5546 ext4_ext_init(sb); 5547 5548 /* 5549 * Enable optimize_scan if number of groups is > threshold. This can be 5550 * turned off by passing "mb_optimize_scan=0". This can also be 5551 * turned on forcefully by passing "mb_optimize_scan=1". 5552 */ 5553 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5554 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5555 set_opt2(sb, MB_OPTIMIZE_SCAN); 5556 else 5557 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5558 } 5559 5560 err = ext4_mb_init(sb); 5561 if (err) { 5562 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5563 err); 5564 goto failed_mount5; 5565 } 5566 5567 /* 5568 * We can only set up the journal commit callback once 5569 * mballoc is initialized 5570 */ 5571 if (sbi->s_journal) 5572 sbi->s_journal->j_commit_callback = 5573 ext4_journal_commit_callback; 5574 5575 err = ext4_percpu_param_init(sbi); 5576 if (err) 5577 goto failed_mount6; 5578 5579 if (ext4_has_feature_flex_bg(sb)) 5580 if (!ext4_fill_flex_info(sb)) { 5581 ext4_msg(sb, KERN_ERR, 5582 "unable to initialize " 5583 "flex_bg meta info!"); 5584 err = -ENOMEM; 5585 goto failed_mount6; 5586 } 5587 5588 err = ext4_register_li_request(sb, first_not_zeroed); 5589 if (err) 5590 goto failed_mount6; 5591 5592 err = ext4_init_orphan_info(sb); 5593 if (err) 5594 goto failed_mount7; 5595 #ifdef CONFIG_QUOTA 5596 /* Enable quota usage during mount. */ 5597 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5598 err = ext4_enable_quotas(sb); 5599 if (err) 5600 goto failed_mount8; 5601 } 5602 #endif /* CONFIG_QUOTA */ 5603 5604 /* 5605 * Save the original bdev mapping's wb_err value which could be 5606 * used to detect the metadata async write error. 5607 */ 5608 errseq_check_and_advance(&sb->s_bdev->bd_mapping->wb_err, 5609 &sbi->s_bdev_wb_err); 5610 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5611 ext4_orphan_cleanup(sb, es); 5612 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5613 /* 5614 * Update the checksum after updating free space/inode counters and 5615 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5616 * checksum in the buffer cache until it is written out and 5617 * e2fsprogs programs trying to open a file system immediately 5618 * after it is mounted can fail. 5619 */ 5620 ext4_superblock_csum_set(sb); 5621 if (needs_recovery) { 5622 ext4_msg(sb, KERN_INFO, "recovery complete"); 5623 err = ext4_mark_recovery_complete(sb, es); 5624 if (err) 5625 goto failed_mount9; 5626 } 5627 5628 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) { 5629 ext4_msg(sb, KERN_WARNING, 5630 "mounting with \"discard\" option, but the device does not support discard"); 5631 clear_opt(sb, DISCARD); 5632 } 5633 5634 if (es->s_error_count) 5635 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5636 5637 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5638 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5639 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5640 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5641 atomic_set(&sbi->s_warning_count, 0); 5642 atomic_set(&sbi->s_msg_count, 0); 5643 5644 /* Register sysfs after all initializations are complete. */ 5645 err = ext4_register_sysfs(sb); 5646 if (err) 5647 goto failed_mount9; 5648 5649 return 0; 5650 5651 failed_mount9: 5652 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 5653 failed_mount8: __maybe_unused 5654 ext4_release_orphan_info(sb); 5655 failed_mount7: 5656 ext4_unregister_li_request(sb); 5657 failed_mount6: 5658 ext4_mb_release(sb); 5659 ext4_flex_groups_free(sbi); 5660 ext4_percpu_param_destroy(sbi); 5661 failed_mount5: 5662 ext4_ext_release(sb); 5663 ext4_release_system_zone(sb); 5664 failed_mount4a: 5665 dput(sb->s_root); 5666 sb->s_root = NULL; 5667 failed_mount4: 5668 ext4_msg(sb, KERN_ERR, "mount failed"); 5669 if (EXT4_SB(sb)->rsv_conversion_wq) 5670 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5671 failed_mount_wq: 5672 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5673 sbi->s_ea_inode_cache = NULL; 5674 5675 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5676 sbi->s_ea_block_cache = NULL; 5677 5678 if (sbi->s_journal) { 5679 ext4_journal_destroy(sbi, sbi->s_journal); 5680 } 5681 failed_mount3a: 5682 ext4_es_unregister_shrinker(sbi); 5683 failed_mount3: 5684 /* flush s_sb_upd_work before sbi destroy */ 5685 flush_work(&sbi->s_sb_upd_work); 5686 ext4_stop_mmpd(sbi); 5687 timer_delete_sync(&sbi->s_err_report); 5688 ext4_group_desc_free(sbi); 5689 failed_mount: 5690 #if IS_ENABLED(CONFIG_UNICODE) 5691 utf8_unload(sb->s_encoding); 5692 #endif 5693 5694 #ifdef CONFIG_QUOTA 5695 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++) 5696 kfree(get_qf_name(sb, sbi, i)); 5697 #endif 5698 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5699 brelse(sbi->s_sbh); 5700 if (sbi->s_journal_bdev_file) { 5701 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 5702 bdev_fput(sbi->s_journal_bdev_file); 5703 } 5704 out_fail: 5705 invalidate_bdev(sb->s_bdev); 5706 sb->s_fs_info = NULL; 5707 return err; 5708 } 5709 5710 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5711 { 5712 struct ext4_fs_context *ctx = fc->fs_private; 5713 struct ext4_sb_info *sbi; 5714 const char *descr; 5715 int ret; 5716 5717 sbi = ext4_alloc_sbi(sb); 5718 if (!sbi) 5719 return -ENOMEM; 5720 5721 fc->s_fs_info = sbi; 5722 5723 /* Cleanup superblock name */ 5724 strreplace(sb->s_id, '/', '!'); 5725 5726 sbi->s_sb_block = 1; /* Default super block location */ 5727 if (ctx->spec & EXT4_SPEC_s_sb_block) 5728 sbi->s_sb_block = ctx->s_sb_block; 5729 5730 ret = __ext4_fill_super(fc, sb); 5731 if (ret < 0) 5732 goto free_sbi; 5733 5734 if (sbi->s_journal) { 5735 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5736 descr = " journalled data mode"; 5737 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5738 descr = " ordered data mode"; 5739 else 5740 descr = " writeback data mode"; 5741 } else 5742 descr = "out journal"; 5743 5744 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5745 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. " 5746 "Quota mode: %s.", &sb->s_uuid, 5747 sb_rdonly(sb) ? "ro" : "r/w", descr, 5748 ext4_quota_mode(sb)); 5749 5750 /* Update the s_overhead_clusters if necessary */ 5751 ext4_update_overhead(sb, false); 5752 return 0; 5753 5754 free_sbi: 5755 ext4_free_sbi(sbi); 5756 fc->s_fs_info = NULL; 5757 return ret; 5758 } 5759 5760 static int ext4_get_tree(struct fs_context *fc) 5761 { 5762 return get_tree_bdev(fc, ext4_fill_super); 5763 } 5764 5765 /* 5766 * Setup any per-fs journal parameters now. We'll do this both on 5767 * initial mount, once the journal has been initialised but before we've 5768 * done any recovery; and again on any subsequent remount. 5769 */ 5770 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5771 { 5772 struct ext4_sb_info *sbi = EXT4_SB(sb); 5773 5774 journal->j_commit_interval = sbi->s_commit_interval; 5775 journal->j_min_batch_time = sbi->s_min_batch_time; 5776 journal->j_max_batch_time = sbi->s_max_batch_time; 5777 ext4_fc_init(sb, journal); 5778 5779 write_lock(&journal->j_state_lock); 5780 if (test_opt(sb, BARRIER)) 5781 journal->j_flags |= JBD2_BARRIER; 5782 else 5783 journal->j_flags &= ~JBD2_BARRIER; 5784 /* 5785 * Always enable journal cycle record option, letting the journal 5786 * records log transactions continuously between each mount. 5787 */ 5788 journal->j_flags |= JBD2_CYCLE_RECORD; 5789 write_unlock(&journal->j_state_lock); 5790 } 5791 5792 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5793 unsigned int journal_inum) 5794 { 5795 struct inode *journal_inode; 5796 5797 /* 5798 * Test for the existence of a valid inode on disk. Bad things 5799 * happen if we iget() an unused inode, as the subsequent iput() 5800 * will try to delete it. 5801 */ 5802 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5803 if (IS_ERR(journal_inode)) { 5804 ext4_msg(sb, KERN_ERR, "no journal found"); 5805 return ERR_CAST(journal_inode); 5806 } 5807 if (!journal_inode->i_nlink) { 5808 make_bad_inode(journal_inode); 5809 iput(journal_inode); 5810 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5811 return ERR_PTR(-EFSCORRUPTED); 5812 } 5813 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) { 5814 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5815 iput(journal_inode); 5816 return ERR_PTR(-EFSCORRUPTED); 5817 } 5818 5819 ext4_debug("Journal inode found at %p: %lld bytes\n", 5820 journal_inode, journal_inode->i_size); 5821 return journal_inode; 5822 } 5823 5824 static int ext4_journal_bmap(journal_t *journal, sector_t *block) 5825 { 5826 struct ext4_map_blocks map; 5827 int ret; 5828 5829 if (journal->j_inode == NULL) 5830 return 0; 5831 5832 map.m_lblk = *block; 5833 map.m_len = 1; 5834 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0); 5835 if (ret <= 0) { 5836 ext4_msg(journal->j_inode->i_sb, KERN_CRIT, 5837 "journal bmap failed: block %llu ret %d\n", 5838 *block, ret); 5839 jbd2_journal_abort(journal, ret ? ret : -EIO); 5840 return ret; 5841 } 5842 *block = map.m_pblk; 5843 return 0; 5844 } 5845 5846 static journal_t *ext4_open_inode_journal(struct super_block *sb, 5847 unsigned int journal_inum) 5848 { 5849 struct inode *journal_inode; 5850 journal_t *journal; 5851 5852 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5853 if (IS_ERR(journal_inode)) 5854 return ERR_CAST(journal_inode); 5855 5856 journal = jbd2_journal_init_inode(journal_inode); 5857 if (IS_ERR(journal)) { 5858 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5859 iput(journal_inode); 5860 return ERR_CAST(journal); 5861 } 5862 journal->j_private = sb; 5863 journal->j_bmap = ext4_journal_bmap; 5864 ext4_init_journal_params(sb, journal); 5865 return journal; 5866 } 5867 5868 static struct file *ext4_get_journal_blkdev(struct super_block *sb, 5869 dev_t j_dev, ext4_fsblk_t *j_start, 5870 ext4_fsblk_t *j_len) 5871 { 5872 struct buffer_head *bh; 5873 struct block_device *bdev; 5874 struct file *bdev_file; 5875 int hblock, blocksize; 5876 ext4_fsblk_t sb_block; 5877 unsigned long offset; 5878 struct ext4_super_block *es; 5879 int errno; 5880 5881 bdev_file = bdev_file_open_by_dev(j_dev, 5882 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES, 5883 sb, &fs_holder_ops); 5884 if (IS_ERR(bdev_file)) { 5885 ext4_msg(sb, KERN_ERR, 5886 "failed to open journal device unknown-block(%u,%u) %ld", 5887 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev_file)); 5888 return bdev_file; 5889 } 5890 5891 bdev = file_bdev(bdev_file); 5892 blocksize = sb->s_blocksize; 5893 hblock = bdev_logical_block_size(bdev); 5894 if (blocksize < hblock) { 5895 ext4_msg(sb, KERN_ERR, 5896 "blocksize too small for journal device"); 5897 errno = -EINVAL; 5898 goto out_bdev; 5899 } 5900 5901 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5902 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5903 set_blocksize(bdev_file, blocksize); 5904 bh = __bread(bdev, sb_block, blocksize); 5905 if (!bh) { 5906 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5907 "external journal"); 5908 errno = -EINVAL; 5909 goto out_bdev; 5910 } 5911 5912 es = (struct ext4_super_block *) (bh->b_data + offset); 5913 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5914 !(le32_to_cpu(es->s_feature_incompat) & 5915 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5916 ext4_msg(sb, KERN_ERR, "external journal has bad superblock"); 5917 errno = -EFSCORRUPTED; 5918 goto out_bh; 5919 } 5920 5921 if ((le32_to_cpu(es->s_feature_ro_compat) & 5922 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5923 es->s_checksum != ext4_superblock_csum(es)) { 5924 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock"); 5925 errno = -EFSCORRUPTED; 5926 goto out_bh; 5927 } 5928 5929 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5930 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5931 errno = -EFSCORRUPTED; 5932 goto out_bh; 5933 } 5934 5935 *j_start = sb_block + 1; 5936 *j_len = ext4_blocks_count(es); 5937 brelse(bh); 5938 return bdev_file; 5939 5940 out_bh: 5941 brelse(bh); 5942 out_bdev: 5943 bdev_fput(bdev_file); 5944 return ERR_PTR(errno); 5945 } 5946 5947 static journal_t *ext4_open_dev_journal(struct super_block *sb, 5948 dev_t j_dev) 5949 { 5950 journal_t *journal; 5951 ext4_fsblk_t j_start; 5952 ext4_fsblk_t j_len; 5953 struct file *bdev_file; 5954 int errno = 0; 5955 5956 bdev_file = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len); 5957 if (IS_ERR(bdev_file)) 5958 return ERR_CAST(bdev_file); 5959 5960 journal = jbd2_journal_init_dev(file_bdev(bdev_file), sb->s_bdev, j_start, 5961 j_len, sb->s_blocksize); 5962 if (IS_ERR(journal)) { 5963 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5964 errno = PTR_ERR(journal); 5965 goto out_bdev; 5966 } 5967 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5968 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5969 "user (unsupported) - %d", 5970 be32_to_cpu(journal->j_superblock->s_nr_users)); 5971 errno = -EINVAL; 5972 goto out_journal; 5973 } 5974 journal->j_private = sb; 5975 EXT4_SB(sb)->s_journal_bdev_file = bdev_file; 5976 ext4_init_journal_params(sb, journal); 5977 return journal; 5978 5979 out_journal: 5980 ext4_journal_destroy(EXT4_SB(sb), journal); 5981 out_bdev: 5982 bdev_fput(bdev_file); 5983 return ERR_PTR(errno); 5984 } 5985 5986 static int ext4_load_journal(struct super_block *sb, 5987 struct ext4_super_block *es, 5988 unsigned long journal_devnum) 5989 { 5990 journal_t *journal; 5991 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5992 dev_t journal_dev; 5993 int err = 0; 5994 int really_read_only; 5995 int journal_dev_ro; 5996 5997 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5998 return -EFSCORRUPTED; 5999 6000 if (journal_devnum && 6001 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6002 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 6003 "numbers have changed"); 6004 journal_dev = new_decode_dev(journal_devnum); 6005 } else 6006 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 6007 6008 if (journal_inum && journal_dev) { 6009 ext4_msg(sb, KERN_ERR, 6010 "filesystem has both journal inode and journal device!"); 6011 return -EINVAL; 6012 } 6013 6014 if (journal_inum) { 6015 journal = ext4_open_inode_journal(sb, journal_inum); 6016 if (IS_ERR(journal)) 6017 return PTR_ERR(journal); 6018 } else { 6019 journal = ext4_open_dev_journal(sb, journal_dev); 6020 if (IS_ERR(journal)) 6021 return PTR_ERR(journal); 6022 } 6023 6024 journal_dev_ro = bdev_read_only(journal->j_dev); 6025 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 6026 6027 if (journal_dev_ro && !sb_rdonly(sb)) { 6028 ext4_msg(sb, KERN_ERR, 6029 "journal device read-only, try mounting with '-o ro'"); 6030 err = -EROFS; 6031 goto err_out; 6032 } 6033 6034 /* 6035 * Are we loading a blank journal or performing recovery after a 6036 * crash? For recovery, we need to check in advance whether we 6037 * can get read-write access to the device. 6038 */ 6039 if (ext4_has_feature_journal_needs_recovery(sb)) { 6040 if (sb_rdonly(sb)) { 6041 ext4_msg(sb, KERN_INFO, "INFO: recovery " 6042 "required on readonly filesystem"); 6043 if (really_read_only) { 6044 ext4_msg(sb, KERN_ERR, "write access " 6045 "unavailable, cannot proceed " 6046 "(try mounting with noload)"); 6047 err = -EROFS; 6048 goto err_out; 6049 } 6050 ext4_msg(sb, KERN_INFO, "write access will " 6051 "be enabled during recovery"); 6052 } 6053 } 6054 6055 if (!(journal->j_flags & JBD2_BARRIER)) 6056 ext4_msg(sb, KERN_INFO, "barriers disabled"); 6057 6058 if (!ext4_has_feature_journal_needs_recovery(sb)) 6059 err = jbd2_journal_wipe(journal, !really_read_only); 6060 if (!err) { 6061 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 6062 __le16 orig_state; 6063 bool changed = false; 6064 6065 if (save) 6066 memcpy(save, ((char *) es) + 6067 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 6068 err = jbd2_journal_load(journal); 6069 if (save && memcmp(((char *) es) + EXT4_S_ERR_START, 6070 save, EXT4_S_ERR_LEN)) { 6071 memcpy(((char *) es) + EXT4_S_ERR_START, 6072 save, EXT4_S_ERR_LEN); 6073 changed = true; 6074 } 6075 kfree(save); 6076 orig_state = es->s_state; 6077 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state & 6078 EXT4_ERROR_FS); 6079 if (orig_state != es->s_state) 6080 changed = true; 6081 /* Write out restored error information to the superblock */ 6082 if (changed && !really_read_only) { 6083 int err2; 6084 err2 = ext4_commit_super(sb); 6085 err = err ? : err2; 6086 } 6087 } 6088 6089 if (err) { 6090 ext4_msg(sb, KERN_ERR, "error loading journal"); 6091 goto err_out; 6092 } 6093 6094 EXT4_SB(sb)->s_journal = journal; 6095 err = ext4_clear_journal_err(sb, es); 6096 if (err) { 6097 ext4_journal_destroy(EXT4_SB(sb), journal); 6098 return err; 6099 } 6100 6101 if (!really_read_only && journal_devnum && 6102 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6103 es->s_journal_dev = cpu_to_le32(journal_devnum); 6104 ext4_commit_super(sb); 6105 } 6106 if (!really_read_only && journal_inum && 6107 journal_inum != le32_to_cpu(es->s_journal_inum)) { 6108 es->s_journal_inum = cpu_to_le32(journal_inum); 6109 ext4_commit_super(sb); 6110 } 6111 6112 return 0; 6113 6114 err_out: 6115 ext4_journal_destroy(EXT4_SB(sb), journal); 6116 return err; 6117 } 6118 6119 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 6120 static void ext4_update_super(struct super_block *sb) 6121 { 6122 struct ext4_sb_info *sbi = EXT4_SB(sb); 6123 struct ext4_super_block *es = sbi->s_es; 6124 struct buffer_head *sbh = sbi->s_sbh; 6125 6126 lock_buffer(sbh); 6127 /* 6128 * If the file system is mounted read-only, don't update the 6129 * superblock write time. This avoids updating the superblock 6130 * write time when we are mounting the root file system 6131 * read/only but we need to replay the journal; at that point, 6132 * for people who are east of GMT and who make their clock 6133 * tick in localtime for Windows bug-for-bug compatibility, 6134 * the clock is set in the future, and this will cause e2fsck 6135 * to complain and force a full file system check. 6136 */ 6137 if (!sb_rdonly(sb)) 6138 ext4_update_tstamp(es, s_wtime); 6139 es->s_kbytes_written = 6140 cpu_to_le64(sbi->s_kbytes_written + 6141 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 6142 sbi->s_sectors_written_start) >> 1)); 6143 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 6144 ext4_free_blocks_count_set(es, 6145 EXT4_C2B(sbi, percpu_counter_sum_positive( 6146 &sbi->s_freeclusters_counter))); 6147 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 6148 es->s_free_inodes_count = 6149 cpu_to_le32(percpu_counter_sum_positive( 6150 &sbi->s_freeinodes_counter)); 6151 /* Copy error information to the on-disk superblock */ 6152 spin_lock(&sbi->s_error_lock); 6153 if (sbi->s_add_error_count > 0) { 6154 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6155 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6156 __ext4_update_tstamp(&es->s_first_error_time, 6157 &es->s_first_error_time_hi, 6158 sbi->s_first_error_time); 6159 strtomem_pad(es->s_first_error_func, 6160 sbi->s_first_error_func, 0); 6161 es->s_first_error_line = 6162 cpu_to_le32(sbi->s_first_error_line); 6163 es->s_first_error_ino = 6164 cpu_to_le32(sbi->s_first_error_ino); 6165 es->s_first_error_block = 6166 cpu_to_le64(sbi->s_first_error_block); 6167 es->s_first_error_errcode = 6168 ext4_errno_to_code(sbi->s_first_error_code); 6169 } 6170 __ext4_update_tstamp(&es->s_last_error_time, 6171 &es->s_last_error_time_hi, 6172 sbi->s_last_error_time); 6173 strtomem_pad(es->s_last_error_func, sbi->s_last_error_func, 0); 6174 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6175 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6176 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6177 es->s_last_error_errcode = 6178 ext4_errno_to_code(sbi->s_last_error_code); 6179 /* 6180 * Start the daily error reporting function if it hasn't been 6181 * started already 6182 */ 6183 if (!es->s_error_count) 6184 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6185 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6186 sbi->s_add_error_count = 0; 6187 } 6188 spin_unlock(&sbi->s_error_lock); 6189 6190 ext4_superblock_csum_set(sb); 6191 unlock_buffer(sbh); 6192 } 6193 6194 static int ext4_commit_super(struct super_block *sb) 6195 { 6196 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6197 6198 if (!sbh) 6199 return -EINVAL; 6200 6201 ext4_update_super(sb); 6202 6203 lock_buffer(sbh); 6204 /* Buffer got discarded which means block device got invalidated */ 6205 if (!buffer_mapped(sbh)) { 6206 unlock_buffer(sbh); 6207 return -EIO; 6208 } 6209 6210 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6211 /* 6212 * Oh, dear. A previous attempt to write the 6213 * superblock failed. This could happen because the 6214 * USB device was yanked out. Or it could happen to 6215 * be a transient write error and maybe the block will 6216 * be remapped. Nothing we can do but to retry the 6217 * write and hope for the best. 6218 */ 6219 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6220 "superblock detected"); 6221 clear_buffer_write_io_error(sbh); 6222 set_buffer_uptodate(sbh); 6223 } 6224 get_bh(sbh); 6225 /* Clear potential dirty bit if it was journalled update */ 6226 clear_buffer_dirty(sbh); 6227 sbh->b_end_io = end_buffer_write_sync; 6228 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6229 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6230 wait_on_buffer(sbh); 6231 if (buffer_write_io_error(sbh)) { 6232 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6233 "superblock"); 6234 clear_buffer_write_io_error(sbh); 6235 set_buffer_uptodate(sbh); 6236 return -EIO; 6237 } 6238 return 0; 6239 } 6240 6241 /* 6242 * Have we just finished recovery? If so, and if we are mounting (or 6243 * remounting) the filesystem readonly, then we will end up with a 6244 * consistent fs on disk. Record that fact. 6245 */ 6246 static int ext4_mark_recovery_complete(struct super_block *sb, 6247 struct ext4_super_block *es) 6248 { 6249 int err; 6250 journal_t *journal = EXT4_SB(sb)->s_journal; 6251 6252 if (!ext4_has_feature_journal(sb)) { 6253 if (journal != NULL) { 6254 ext4_error(sb, "Journal got removed while the fs was " 6255 "mounted!"); 6256 return -EFSCORRUPTED; 6257 } 6258 return 0; 6259 } 6260 jbd2_journal_lock_updates(journal); 6261 err = jbd2_journal_flush(journal, 0); 6262 if (err < 0) 6263 goto out; 6264 6265 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6266 ext4_has_feature_orphan_present(sb))) { 6267 if (!ext4_orphan_file_empty(sb)) { 6268 ext4_error(sb, "Orphan file not empty on read-only fs."); 6269 err = -EFSCORRUPTED; 6270 goto out; 6271 } 6272 ext4_clear_feature_journal_needs_recovery(sb); 6273 ext4_clear_feature_orphan_present(sb); 6274 ext4_commit_super(sb); 6275 } 6276 out: 6277 jbd2_journal_unlock_updates(journal); 6278 return err; 6279 } 6280 6281 /* 6282 * If we are mounting (or read-write remounting) a filesystem whose journal 6283 * has recorded an error from a previous lifetime, move that error to the 6284 * main filesystem now. 6285 */ 6286 static int ext4_clear_journal_err(struct super_block *sb, 6287 struct ext4_super_block *es) 6288 { 6289 journal_t *journal; 6290 int j_errno; 6291 const char *errstr; 6292 6293 if (!ext4_has_feature_journal(sb)) { 6294 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6295 return -EFSCORRUPTED; 6296 } 6297 6298 journal = EXT4_SB(sb)->s_journal; 6299 6300 /* 6301 * Now check for any error status which may have been recorded in the 6302 * journal by a prior ext4_error() or ext4_abort() 6303 */ 6304 6305 j_errno = jbd2_journal_errno(journal); 6306 if (j_errno) { 6307 char nbuf[16]; 6308 6309 errstr = ext4_decode_error(sb, j_errno, nbuf); 6310 ext4_warning(sb, "Filesystem error recorded " 6311 "from previous mount: %s", errstr); 6312 6313 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6314 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6315 j_errno = ext4_commit_super(sb); 6316 if (j_errno) 6317 return j_errno; 6318 ext4_warning(sb, "Marked fs in need of filesystem check."); 6319 6320 jbd2_journal_clear_err(journal); 6321 jbd2_journal_update_sb_errno(journal); 6322 } 6323 return 0; 6324 } 6325 6326 /* 6327 * Force the running and committing transactions to commit, 6328 * and wait on the commit. 6329 */ 6330 int ext4_force_commit(struct super_block *sb) 6331 { 6332 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal); 6333 } 6334 6335 static int ext4_sync_fs(struct super_block *sb, int wait) 6336 { 6337 int ret = 0; 6338 tid_t target; 6339 bool needs_barrier = false; 6340 struct ext4_sb_info *sbi = EXT4_SB(sb); 6341 6342 ret = ext4_emergency_state(sb); 6343 if (unlikely(ret)) 6344 return ret; 6345 6346 trace_ext4_sync_fs(sb, wait); 6347 flush_workqueue(sbi->rsv_conversion_wq); 6348 /* 6349 * Writeback quota in non-journalled quota case - journalled quota has 6350 * no dirty dquots 6351 */ 6352 dquot_writeback_dquots(sb, -1); 6353 /* 6354 * Data writeback is possible w/o journal transaction, so barrier must 6355 * being sent at the end of the function. But we can skip it if 6356 * transaction_commit will do it for us. 6357 */ 6358 if (sbi->s_journal) { 6359 target = jbd2_get_latest_transaction(sbi->s_journal); 6360 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6361 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6362 needs_barrier = true; 6363 6364 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6365 if (wait) 6366 ret = jbd2_log_wait_commit(sbi->s_journal, 6367 target); 6368 } 6369 } else if (wait && test_opt(sb, BARRIER)) 6370 needs_barrier = true; 6371 if (needs_barrier) { 6372 int err; 6373 err = blkdev_issue_flush(sb->s_bdev); 6374 if (!ret) 6375 ret = err; 6376 } 6377 6378 return ret; 6379 } 6380 6381 /* 6382 * LVM calls this function before a (read-only) snapshot is created. This 6383 * gives us a chance to flush the journal completely and mark the fs clean. 6384 * 6385 * Note that only this function cannot bring a filesystem to be in a clean 6386 * state independently. It relies on upper layer to stop all data & metadata 6387 * modifications. 6388 */ 6389 static int ext4_freeze(struct super_block *sb) 6390 { 6391 int error = 0; 6392 journal_t *journal = EXT4_SB(sb)->s_journal; 6393 6394 if (journal) { 6395 /* Now we set up the journal barrier. */ 6396 jbd2_journal_lock_updates(journal); 6397 6398 /* 6399 * Don't clear the needs_recovery flag if we failed to 6400 * flush the journal. 6401 */ 6402 error = jbd2_journal_flush(journal, 0); 6403 if (error < 0) 6404 goto out; 6405 6406 /* Journal blocked and flushed, clear needs_recovery flag. */ 6407 ext4_clear_feature_journal_needs_recovery(sb); 6408 if (ext4_orphan_file_empty(sb)) 6409 ext4_clear_feature_orphan_present(sb); 6410 } 6411 6412 error = ext4_commit_super(sb); 6413 out: 6414 if (journal) 6415 /* we rely on upper layer to stop further updates */ 6416 jbd2_journal_unlock_updates(journal); 6417 return error; 6418 } 6419 6420 /* 6421 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6422 * flag here, even though the filesystem is not technically dirty yet. 6423 */ 6424 static int ext4_unfreeze(struct super_block *sb) 6425 { 6426 if (ext4_emergency_state(sb)) 6427 return 0; 6428 6429 if (EXT4_SB(sb)->s_journal) { 6430 /* Reset the needs_recovery flag before the fs is unlocked. */ 6431 ext4_set_feature_journal_needs_recovery(sb); 6432 if (ext4_has_feature_orphan_file(sb)) 6433 ext4_set_feature_orphan_present(sb); 6434 } 6435 6436 ext4_commit_super(sb); 6437 return 0; 6438 } 6439 6440 /* 6441 * Structure to save mount options for ext4_remount's benefit 6442 */ 6443 struct ext4_mount_options { 6444 unsigned long s_mount_opt; 6445 unsigned long s_mount_opt2; 6446 kuid_t s_resuid; 6447 kgid_t s_resgid; 6448 unsigned long s_commit_interval; 6449 u32 s_min_batch_time, s_max_batch_time; 6450 #ifdef CONFIG_QUOTA 6451 int s_jquota_fmt; 6452 char *s_qf_names[EXT4_MAXQUOTAS]; 6453 #endif 6454 }; 6455 6456 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6457 { 6458 struct ext4_fs_context *ctx = fc->fs_private; 6459 struct ext4_super_block *es; 6460 struct ext4_sb_info *sbi = EXT4_SB(sb); 6461 unsigned long old_sb_flags; 6462 struct ext4_mount_options old_opts; 6463 ext4_group_t g; 6464 int err = 0; 6465 int alloc_ctx; 6466 #ifdef CONFIG_QUOTA 6467 int enable_quota = 0; 6468 int i, j; 6469 char *to_free[EXT4_MAXQUOTAS]; 6470 #endif 6471 6472 6473 /* Store the original options */ 6474 old_sb_flags = sb->s_flags; 6475 old_opts.s_mount_opt = sbi->s_mount_opt; 6476 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6477 old_opts.s_resuid = sbi->s_resuid; 6478 old_opts.s_resgid = sbi->s_resgid; 6479 old_opts.s_commit_interval = sbi->s_commit_interval; 6480 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6481 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6482 #ifdef CONFIG_QUOTA 6483 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6484 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6485 if (sbi->s_qf_names[i]) { 6486 char *qf_name = get_qf_name(sb, sbi, i); 6487 6488 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6489 if (!old_opts.s_qf_names[i]) { 6490 for (j = 0; j < i; j++) 6491 kfree(old_opts.s_qf_names[j]); 6492 return -ENOMEM; 6493 } 6494 } else 6495 old_opts.s_qf_names[i] = NULL; 6496 #endif 6497 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6498 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6499 ctx->journal_ioprio = 6500 sbi->s_journal->j_task->io_context->ioprio; 6501 else 6502 ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO; 6503 6504 } 6505 6506 if ((ctx->spec & EXT4_SPEC_s_stripe) && 6507 ext4_is_stripe_incompatible(sb, ctx->s_stripe)) { 6508 ext4_msg(sb, KERN_WARNING, 6509 "stripe (%lu) is not aligned with cluster size (%u), " 6510 "stripe is disabled", 6511 ctx->s_stripe, sbi->s_cluster_ratio); 6512 ctx->s_stripe = 0; 6513 } 6514 6515 /* 6516 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause 6517 * two calls to ext4_should_dioread_nolock() to return inconsistent 6518 * values, triggering WARN_ON in ext4_add_complete_io(). we grab 6519 * here s_writepages_rwsem to avoid race between writepages ops and 6520 * remount. 6521 */ 6522 alloc_ctx = ext4_writepages_down_write(sb); 6523 ext4_apply_options(fc, sb); 6524 ext4_writepages_up_write(sb, alloc_ctx); 6525 6526 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6527 test_opt(sb, JOURNAL_CHECKSUM)) { 6528 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6529 "during remount not supported; ignoring"); 6530 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6531 } 6532 6533 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6534 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6535 ext4_msg(sb, KERN_ERR, "can't mount with " 6536 "both data=journal and delalloc"); 6537 err = -EINVAL; 6538 goto restore_opts; 6539 } 6540 if (test_opt(sb, DIOREAD_NOLOCK)) { 6541 ext4_msg(sb, KERN_ERR, "can't mount with " 6542 "both data=journal and dioread_nolock"); 6543 err = -EINVAL; 6544 goto restore_opts; 6545 } 6546 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6547 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6548 ext4_msg(sb, KERN_ERR, "can't mount with " 6549 "journal_async_commit in data=ordered mode"); 6550 err = -EINVAL; 6551 goto restore_opts; 6552 } 6553 } 6554 6555 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6556 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6557 err = -EINVAL; 6558 goto restore_opts; 6559 } 6560 6561 if ((old_opts.s_mount_opt & EXT4_MOUNT_DELALLOC) && 6562 !test_opt(sb, DELALLOC)) { 6563 ext4_msg(sb, KERN_ERR, "can't disable delalloc during remount"); 6564 err = -EINVAL; 6565 goto restore_opts; 6566 } 6567 6568 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6569 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6570 6571 es = sbi->s_es; 6572 6573 if (sbi->s_journal) { 6574 ext4_init_journal_params(sb, sbi->s_journal); 6575 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6576 } 6577 6578 /* Flush outstanding errors before changing fs state */ 6579 flush_work(&sbi->s_sb_upd_work); 6580 6581 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6582 if (ext4_emergency_state(sb)) { 6583 err = -EROFS; 6584 goto restore_opts; 6585 } 6586 6587 if (fc->sb_flags & SB_RDONLY) { 6588 err = sync_filesystem(sb); 6589 if (err < 0) 6590 goto restore_opts; 6591 err = dquot_suspend(sb, -1); 6592 if (err < 0) 6593 goto restore_opts; 6594 6595 /* 6596 * First of all, the unconditional stuff we have to do 6597 * to disable replay of the journal when we next remount 6598 */ 6599 sb->s_flags |= SB_RDONLY; 6600 6601 /* 6602 * OK, test if we are remounting a valid rw partition 6603 * readonly, and if so set the rdonly flag and then 6604 * mark the partition as valid again. 6605 */ 6606 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6607 (sbi->s_mount_state & EXT4_VALID_FS)) 6608 es->s_state = cpu_to_le16(sbi->s_mount_state); 6609 6610 if (sbi->s_journal) { 6611 /* 6612 * We let remount-ro finish even if marking fs 6613 * as clean failed... 6614 */ 6615 ext4_mark_recovery_complete(sb, es); 6616 } 6617 } else { 6618 /* Make sure we can mount this feature set readwrite */ 6619 if (ext4_has_feature_readonly(sb) || 6620 !ext4_feature_set_ok(sb, 0)) { 6621 err = -EROFS; 6622 goto restore_opts; 6623 } 6624 /* 6625 * Make sure the group descriptor checksums 6626 * are sane. If they aren't, refuse to remount r/w. 6627 */ 6628 for (g = 0; g < sbi->s_groups_count; g++) { 6629 struct ext4_group_desc *gdp = 6630 ext4_get_group_desc(sb, g, NULL); 6631 6632 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6633 ext4_msg(sb, KERN_ERR, 6634 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6635 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6636 le16_to_cpu(gdp->bg_checksum)); 6637 err = -EFSBADCRC; 6638 goto restore_opts; 6639 } 6640 } 6641 6642 /* 6643 * If we have an unprocessed orphan list hanging 6644 * around from a previously readonly bdev mount, 6645 * require a full umount/remount for now. 6646 */ 6647 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6648 ext4_msg(sb, KERN_WARNING, "Couldn't " 6649 "remount RDWR because of unprocessed " 6650 "orphan inode list. Please " 6651 "umount/remount instead"); 6652 err = -EINVAL; 6653 goto restore_opts; 6654 } 6655 6656 /* 6657 * Mounting a RDONLY partition read-write, so reread 6658 * and store the current valid flag. (It may have 6659 * been changed by e2fsck since we originally mounted 6660 * the partition.) 6661 */ 6662 if (sbi->s_journal) { 6663 err = ext4_clear_journal_err(sb, es); 6664 if (err) 6665 goto restore_opts; 6666 } 6667 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6668 ~EXT4_FC_REPLAY); 6669 6670 err = ext4_setup_super(sb, es, 0); 6671 if (err) 6672 goto restore_opts; 6673 6674 sb->s_flags &= ~SB_RDONLY; 6675 if (ext4_has_feature_mmp(sb)) { 6676 err = ext4_multi_mount_protect(sb, 6677 le64_to_cpu(es->s_mmp_block)); 6678 if (err) 6679 goto restore_opts; 6680 } 6681 #ifdef CONFIG_QUOTA 6682 enable_quota = 1; 6683 #endif 6684 } 6685 } 6686 6687 /* 6688 * Handle creation of system zone data early because it can fail. 6689 * Releasing of existing data is done when we are sure remount will 6690 * succeed. 6691 */ 6692 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6693 err = ext4_setup_system_zone(sb); 6694 if (err) 6695 goto restore_opts; 6696 } 6697 6698 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6699 err = ext4_commit_super(sb); 6700 if (err) 6701 goto restore_opts; 6702 } 6703 6704 #ifdef CONFIG_QUOTA 6705 if (enable_quota) { 6706 if (sb_any_quota_suspended(sb)) 6707 dquot_resume(sb, -1); 6708 else if (ext4_has_feature_quota(sb)) { 6709 err = ext4_enable_quotas(sb); 6710 if (err) 6711 goto restore_opts; 6712 } 6713 } 6714 /* Release old quota file names */ 6715 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6716 kfree(old_opts.s_qf_names[i]); 6717 #endif 6718 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6719 ext4_release_system_zone(sb); 6720 6721 /* 6722 * Reinitialize lazy itable initialization thread based on 6723 * current settings 6724 */ 6725 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6726 ext4_unregister_li_request(sb); 6727 else { 6728 ext4_group_t first_not_zeroed; 6729 first_not_zeroed = ext4_has_uninit_itable(sb); 6730 ext4_register_li_request(sb, first_not_zeroed); 6731 } 6732 6733 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6734 ext4_stop_mmpd(sbi); 6735 6736 /* 6737 * Handle aborting the filesystem as the last thing during remount to 6738 * avoid obsure errors during remount when some option changes fail to 6739 * apply due to shutdown filesystem. 6740 */ 6741 if (test_opt2(sb, ABORT)) 6742 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6743 6744 return 0; 6745 6746 restore_opts: 6747 /* 6748 * If there was a failing r/w to ro transition, we may need to 6749 * re-enable quota 6750 */ 6751 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) && 6752 sb_any_quota_suspended(sb)) 6753 dquot_resume(sb, -1); 6754 6755 alloc_ctx = ext4_writepages_down_write(sb); 6756 sb->s_flags = old_sb_flags; 6757 sbi->s_mount_opt = old_opts.s_mount_opt; 6758 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6759 sbi->s_resuid = old_opts.s_resuid; 6760 sbi->s_resgid = old_opts.s_resgid; 6761 sbi->s_commit_interval = old_opts.s_commit_interval; 6762 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6763 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6764 ext4_writepages_up_write(sb, alloc_ctx); 6765 6766 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6767 ext4_release_system_zone(sb); 6768 #ifdef CONFIG_QUOTA 6769 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6770 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6771 to_free[i] = get_qf_name(sb, sbi, i); 6772 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6773 } 6774 synchronize_rcu(); 6775 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6776 kfree(to_free[i]); 6777 #endif 6778 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6779 ext4_stop_mmpd(sbi); 6780 return err; 6781 } 6782 6783 static int ext4_reconfigure(struct fs_context *fc) 6784 { 6785 struct super_block *sb = fc->root->d_sb; 6786 int ret; 6787 bool old_ro = sb_rdonly(sb); 6788 6789 fc->s_fs_info = EXT4_SB(sb); 6790 6791 ret = ext4_check_opt_consistency(fc, sb); 6792 if (ret < 0) 6793 return ret; 6794 6795 ret = __ext4_remount(fc, sb); 6796 if (ret < 0) 6797 return ret; 6798 6799 ext4_msg(sb, KERN_INFO, "re-mounted %pU%s.", 6800 &sb->s_uuid, 6801 (old_ro != sb_rdonly(sb)) ? (sb_rdonly(sb) ? " ro" : " r/w") : ""); 6802 6803 return 0; 6804 } 6805 6806 #ifdef CONFIG_QUOTA 6807 static int ext4_statfs_project(struct super_block *sb, 6808 kprojid_t projid, struct kstatfs *buf) 6809 { 6810 struct kqid qid; 6811 struct dquot *dquot; 6812 u64 limit; 6813 u64 curblock; 6814 6815 qid = make_kqid_projid(projid); 6816 dquot = dqget(sb, qid); 6817 if (IS_ERR(dquot)) 6818 return PTR_ERR(dquot); 6819 spin_lock(&dquot->dq_dqb_lock); 6820 6821 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6822 dquot->dq_dqb.dqb_bhardlimit); 6823 limit >>= sb->s_blocksize_bits; 6824 6825 if (limit) { 6826 uint64_t remaining = 0; 6827 6828 curblock = (dquot->dq_dqb.dqb_curspace + 6829 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6830 if (limit > curblock) 6831 remaining = limit - curblock; 6832 6833 buf->f_blocks = min(buf->f_blocks, limit); 6834 buf->f_bfree = min(buf->f_bfree, remaining); 6835 buf->f_bavail = min(buf->f_bavail, remaining); 6836 } 6837 6838 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6839 dquot->dq_dqb.dqb_ihardlimit); 6840 if (limit) { 6841 uint64_t remaining = 0; 6842 6843 if (limit > dquot->dq_dqb.dqb_curinodes) 6844 remaining = limit - dquot->dq_dqb.dqb_curinodes; 6845 6846 buf->f_files = min(buf->f_files, limit); 6847 buf->f_ffree = min(buf->f_ffree, remaining); 6848 } 6849 6850 spin_unlock(&dquot->dq_dqb_lock); 6851 dqput(dquot); 6852 return 0; 6853 } 6854 #endif 6855 6856 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6857 { 6858 struct super_block *sb = dentry->d_sb; 6859 struct ext4_sb_info *sbi = EXT4_SB(sb); 6860 struct ext4_super_block *es = sbi->s_es; 6861 ext4_fsblk_t overhead = 0, resv_blocks; 6862 s64 bfree; 6863 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6864 6865 if (!test_opt(sb, MINIX_DF)) 6866 overhead = sbi->s_overhead; 6867 6868 buf->f_type = EXT4_SUPER_MAGIC; 6869 buf->f_bsize = sb->s_blocksize; 6870 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6871 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6872 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6873 /* prevent underflow in case that few free space is available */ 6874 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6875 buf->f_bavail = buf->f_bfree - 6876 (ext4_r_blocks_count(es) + resv_blocks); 6877 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6878 buf->f_bavail = 0; 6879 buf->f_files = le32_to_cpu(es->s_inodes_count); 6880 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6881 buf->f_namelen = EXT4_NAME_LEN; 6882 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6883 6884 #ifdef CONFIG_QUOTA 6885 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6886 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6887 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6888 #endif 6889 return 0; 6890 } 6891 6892 6893 #ifdef CONFIG_QUOTA 6894 6895 /* 6896 * Helper functions so that transaction is started before we acquire dqio_sem 6897 * to keep correct lock ordering of transaction > dqio_sem 6898 */ 6899 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6900 { 6901 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6902 } 6903 6904 static int ext4_write_dquot(struct dquot *dquot) 6905 { 6906 int ret, err; 6907 handle_t *handle; 6908 struct inode *inode; 6909 6910 inode = dquot_to_inode(dquot); 6911 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6912 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6913 if (IS_ERR(handle)) 6914 return PTR_ERR(handle); 6915 ret = dquot_commit(dquot); 6916 if (ret < 0) 6917 ext4_error_err(dquot->dq_sb, -ret, 6918 "Failed to commit dquot type %d", 6919 dquot->dq_id.type); 6920 err = ext4_journal_stop(handle); 6921 if (!ret) 6922 ret = err; 6923 return ret; 6924 } 6925 6926 static int ext4_acquire_dquot(struct dquot *dquot) 6927 { 6928 int ret, err; 6929 handle_t *handle; 6930 6931 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6932 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6933 if (IS_ERR(handle)) 6934 return PTR_ERR(handle); 6935 ret = dquot_acquire(dquot); 6936 if (ret < 0) 6937 ext4_error_err(dquot->dq_sb, -ret, 6938 "Failed to acquire dquot type %d", 6939 dquot->dq_id.type); 6940 err = ext4_journal_stop(handle); 6941 if (!ret) 6942 ret = err; 6943 return ret; 6944 } 6945 6946 static int ext4_release_dquot(struct dquot *dquot) 6947 { 6948 int ret, err; 6949 handle_t *handle; 6950 bool freeze_protected = false; 6951 6952 /* 6953 * Trying to sb_start_intwrite() in a running transaction 6954 * can result in a deadlock. Further, running transactions 6955 * are already protected from freezing. 6956 */ 6957 if (!ext4_journal_current_handle()) { 6958 sb_start_intwrite(dquot->dq_sb); 6959 freeze_protected = true; 6960 } 6961 6962 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6963 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6964 if (IS_ERR(handle)) { 6965 /* Release dquot anyway to avoid endless cycle in dqput() */ 6966 dquot_release(dquot); 6967 if (freeze_protected) 6968 sb_end_intwrite(dquot->dq_sb); 6969 return PTR_ERR(handle); 6970 } 6971 ret = dquot_release(dquot); 6972 if (ret < 0) 6973 ext4_error_err(dquot->dq_sb, -ret, 6974 "Failed to release dquot type %d", 6975 dquot->dq_id.type); 6976 err = ext4_journal_stop(handle); 6977 if (!ret) 6978 ret = err; 6979 6980 if (freeze_protected) 6981 sb_end_intwrite(dquot->dq_sb); 6982 6983 return ret; 6984 } 6985 6986 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6987 { 6988 struct super_block *sb = dquot->dq_sb; 6989 6990 if (ext4_is_quota_journalled(sb)) { 6991 dquot_mark_dquot_dirty(dquot); 6992 return ext4_write_dquot(dquot); 6993 } else { 6994 return dquot_mark_dquot_dirty(dquot); 6995 } 6996 } 6997 6998 static int ext4_write_info(struct super_block *sb, int type) 6999 { 7000 int ret, err; 7001 handle_t *handle; 7002 7003 /* Data block + inode block */ 7004 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 7005 if (IS_ERR(handle)) 7006 return PTR_ERR(handle); 7007 ret = dquot_commit_info(sb, type); 7008 err = ext4_journal_stop(handle); 7009 if (!ret) 7010 ret = err; 7011 return ret; 7012 } 7013 7014 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 7015 { 7016 struct ext4_inode_info *ei = EXT4_I(inode); 7017 7018 /* The first argument of lockdep_set_subclass has to be 7019 * *exactly* the same as the argument to init_rwsem() --- in 7020 * this case, in init_once() --- or lockdep gets unhappy 7021 * because the name of the lock is set using the 7022 * stringification of the argument to init_rwsem(). 7023 */ 7024 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 7025 lockdep_set_subclass(&ei->i_data_sem, subclass); 7026 } 7027 7028 /* 7029 * Standard function to be called on quota_on 7030 */ 7031 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 7032 const struct path *path) 7033 { 7034 int err; 7035 7036 if (!test_opt(sb, QUOTA)) 7037 return -EINVAL; 7038 7039 /* Quotafile not on the same filesystem? */ 7040 if (path->dentry->d_sb != sb) 7041 return -EXDEV; 7042 7043 /* Quota already enabled for this file? */ 7044 if (IS_NOQUOTA(d_inode(path->dentry))) 7045 return -EBUSY; 7046 7047 /* Journaling quota? */ 7048 if (EXT4_SB(sb)->s_qf_names[type]) { 7049 /* Quotafile not in fs root? */ 7050 if (path->dentry->d_parent != sb->s_root) 7051 ext4_msg(sb, KERN_WARNING, 7052 "Quota file not on filesystem root. " 7053 "Journaled quota will not work"); 7054 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 7055 } else { 7056 /* 7057 * Clear the flag just in case mount options changed since 7058 * last time. 7059 */ 7060 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 7061 } 7062 7063 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 7064 err = dquot_quota_on(sb, type, format_id, path); 7065 if (!err) { 7066 struct inode *inode = d_inode(path->dentry); 7067 handle_t *handle; 7068 7069 /* 7070 * Set inode flags to prevent userspace from messing with quota 7071 * files. If this fails, we return success anyway since quotas 7072 * are already enabled and this is not a hard failure. 7073 */ 7074 inode_lock(inode); 7075 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7076 if (IS_ERR(handle)) 7077 goto unlock_inode; 7078 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 7079 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 7080 S_NOATIME | S_IMMUTABLE); 7081 err = ext4_mark_inode_dirty(handle, inode); 7082 ext4_journal_stop(handle); 7083 unlock_inode: 7084 inode_unlock(inode); 7085 if (err) 7086 dquot_quota_off(sb, type); 7087 } 7088 if (err) 7089 lockdep_set_quota_inode(path->dentry->d_inode, 7090 I_DATA_SEM_NORMAL); 7091 return err; 7092 } 7093 7094 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum) 7095 { 7096 switch (type) { 7097 case USRQUOTA: 7098 return qf_inum == EXT4_USR_QUOTA_INO; 7099 case GRPQUOTA: 7100 return qf_inum == EXT4_GRP_QUOTA_INO; 7101 case PRJQUOTA: 7102 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO; 7103 default: 7104 BUG(); 7105 } 7106 } 7107 7108 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 7109 unsigned int flags) 7110 { 7111 int err; 7112 struct inode *qf_inode; 7113 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7114 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7115 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7116 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7117 }; 7118 7119 BUG_ON(!ext4_has_feature_quota(sb)); 7120 7121 if (!qf_inums[type]) 7122 return -EPERM; 7123 7124 if (!ext4_check_quota_inum(type, qf_inums[type])) { 7125 ext4_error(sb, "Bad quota inum: %lu, type: %d", 7126 qf_inums[type], type); 7127 return -EUCLEAN; 7128 } 7129 7130 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 7131 if (IS_ERR(qf_inode)) { 7132 ext4_error(sb, "Bad quota inode: %lu, type: %d", 7133 qf_inums[type], type); 7134 return PTR_ERR(qf_inode); 7135 } 7136 7137 /* Don't account quota for quota files to avoid recursion */ 7138 qf_inode->i_flags |= S_NOQUOTA; 7139 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 7140 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 7141 if (err) 7142 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 7143 iput(qf_inode); 7144 7145 return err; 7146 } 7147 7148 /* Enable usage tracking for all quota types. */ 7149 int ext4_enable_quotas(struct super_block *sb) 7150 { 7151 int type, err = 0; 7152 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7153 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7154 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7155 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7156 }; 7157 bool quota_mopt[EXT4_MAXQUOTAS] = { 7158 test_opt(sb, USRQUOTA), 7159 test_opt(sb, GRPQUOTA), 7160 test_opt(sb, PRJQUOTA), 7161 }; 7162 7163 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 7164 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 7165 if (qf_inums[type]) { 7166 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 7167 DQUOT_USAGE_ENABLED | 7168 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 7169 if (err) { 7170 ext4_warning(sb, 7171 "Failed to enable quota tracking " 7172 "(type=%d, err=%d, ino=%lu). " 7173 "Please run e2fsck to fix.", type, 7174 err, qf_inums[type]); 7175 7176 ext4_quotas_off(sb, type); 7177 return err; 7178 } 7179 } 7180 } 7181 return 0; 7182 } 7183 7184 static int ext4_quota_off(struct super_block *sb, int type) 7185 { 7186 struct inode *inode = sb_dqopt(sb)->files[type]; 7187 handle_t *handle; 7188 int err; 7189 7190 /* Force all delayed allocation blocks to be allocated. 7191 * Caller already holds s_umount sem */ 7192 if (test_opt(sb, DELALLOC)) 7193 sync_filesystem(sb); 7194 7195 if (!inode || !igrab(inode)) 7196 goto out; 7197 7198 err = dquot_quota_off(sb, type); 7199 if (err || ext4_has_feature_quota(sb)) 7200 goto out_put; 7201 /* 7202 * When the filesystem was remounted read-only first, we cannot cleanup 7203 * inode flags here. Bad luck but people should be using QUOTA feature 7204 * these days anyway. 7205 */ 7206 if (sb_rdonly(sb)) 7207 goto out_put; 7208 7209 inode_lock(inode); 7210 /* 7211 * Update modification times of quota files when userspace can 7212 * start looking at them. If we fail, we return success anyway since 7213 * this is not a hard failure and quotas are already disabled. 7214 */ 7215 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7216 if (IS_ERR(handle)) { 7217 err = PTR_ERR(handle); 7218 goto out_unlock; 7219 } 7220 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7221 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7222 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 7223 err = ext4_mark_inode_dirty(handle, inode); 7224 ext4_journal_stop(handle); 7225 out_unlock: 7226 inode_unlock(inode); 7227 out_put: 7228 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7229 iput(inode); 7230 return err; 7231 out: 7232 return dquot_quota_off(sb, type); 7233 } 7234 7235 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7236 * acquiring the locks... As quota files are never truncated and quota code 7237 * itself serializes the operations (and no one else should touch the files) 7238 * we don't have to be afraid of races */ 7239 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7240 size_t len, loff_t off) 7241 { 7242 struct inode *inode = sb_dqopt(sb)->files[type]; 7243 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7244 int offset = off & (sb->s_blocksize - 1); 7245 int tocopy; 7246 size_t toread; 7247 struct buffer_head *bh; 7248 loff_t i_size = i_size_read(inode); 7249 7250 if (off > i_size) 7251 return 0; 7252 if (off+len > i_size) 7253 len = i_size-off; 7254 toread = len; 7255 while (toread > 0) { 7256 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread); 7257 bh = ext4_bread(NULL, inode, blk, 0); 7258 if (IS_ERR(bh)) 7259 return PTR_ERR(bh); 7260 if (!bh) /* A hole? */ 7261 memset(data, 0, tocopy); 7262 else 7263 memcpy(data, bh->b_data+offset, tocopy); 7264 brelse(bh); 7265 offset = 0; 7266 toread -= tocopy; 7267 data += tocopy; 7268 blk++; 7269 } 7270 return len; 7271 } 7272 7273 /* Write to quotafile (we know the transaction is already started and has 7274 * enough credits) */ 7275 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7276 const char *data, size_t len, loff_t off) 7277 { 7278 struct inode *inode = sb_dqopt(sb)->files[type]; 7279 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7280 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7281 int retries = 0; 7282 struct buffer_head *bh; 7283 handle_t *handle = journal_current_handle(); 7284 7285 if (!handle) { 7286 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7287 " cancelled because transaction is not started", 7288 (unsigned long long)off, (unsigned long long)len); 7289 return -EIO; 7290 } 7291 /* 7292 * Since we account only one data block in transaction credits, 7293 * then it is impossible to cross a block boundary. 7294 */ 7295 if (sb->s_blocksize - offset < len) { 7296 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7297 " cancelled because not block aligned", 7298 (unsigned long long)off, (unsigned long long)len); 7299 return -EIO; 7300 } 7301 7302 do { 7303 bh = ext4_bread(handle, inode, blk, 7304 EXT4_GET_BLOCKS_CREATE | 7305 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7306 } while (PTR_ERR(bh) == -ENOSPC && 7307 ext4_should_retry_alloc(inode->i_sb, &retries)); 7308 if (IS_ERR(bh)) 7309 return PTR_ERR(bh); 7310 if (!bh) 7311 goto out; 7312 BUFFER_TRACE(bh, "get write access"); 7313 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7314 if (err) { 7315 brelse(bh); 7316 return err; 7317 } 7318 lock_buffer(bh); 7319 memcpy(bh->b_data+offset, data, len); 7320 flush_dcache_folio(bh->b_folio); 7321 unlock_buffer(bh); 7322 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7323 brelse(bh); 7324 out: 7325 if (inode->i_size < off + len) { 7326 i_size_write(inode, off + len); 7327 EXT4_I(inode)->i_disksize = inode->i_size; 7328 err2 = ext4_mark_inode_dirty(handle, inode); 7329 if (unlikely(err2 && !err)) 7330 err = err2; 7331 } 7332 return err ? err : len; 7333 } 7334 #endif 7335 7336 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7337 static inline void register_as_ext2(void) 7338 { 7339 int err = register_filesystem(&ext2_fs_type); 7340 if (err) 7341 printk(KERN_WARNING 7342 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7343 } 7344 7345 static inline void unregister_as_ext2(void) 7346 { 7347 unregister_filesystem(&ext2_fs_type); 7348 } 7349 7350 static inline int ext2_feature_set_ok(struct super_block *sb) 7351 { 7352 if (ext4_has_unknown_ext2_incompat_features(sb)) 7353 return 0; 7354 if (sb_rdonly(sb)) 7355 return 1; 7356 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7357 return 0; 7358 return 1; 7359 } 7360 #else 7361 static inline void register_as_ext2(void) { } 7362 static inline void unregister_as_ext2(void) { } 7363 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7364 #endif 7365 7366 static inline void register_as_ext3(void) 7367 { 7368 int err = register_filesystem(&ext3_fs_type); 7369 if (err) 7370 printk(KERN_WARNING 7371 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7372 } 7373 7374 static inline void unregister_as_ext3(void) 7375 { 7376 unregister_filesystem(&ext3_fs_type); 7377 } 7378 7379 static inline int ext3_feature_set_ok(struct super_block *sb) 7380 { 7381 if (ext4_has_unknown_ext3_incompat_features(sb)) 7382 return 0; 7383 if (!ext4_has_feature_journal(sb)) 7384 return 0; 7385 if (sb_rdonly(sb)) 7386 return 1; 7387 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7388 return 0; 7389 return 1; 7390 } 7391 7392 static void ext4_kill_sb(struct super_block *sb) 7393 { 7394 struct ext4_sb_info *sbi = EXT4_SB(sb); 7395 struct file *bdev_file = sbi ? sbi->s_journal_bdev_file : NULL; 7396 7397 kill_block_super(sb); 7398 7399 if (bdev_file) 7400 bdev_fput(bdev_file); 7401 } 7402 7403 static struct file_system_type ext4_fs_type = { 7404 .owner = THIS_MODULE, 7405 .name = "ext4", 7406 .init_fs_context = ext4_init_fs_context, 7407 .parameters = ext4_param_specs, 7408 .kill_sb = ext4_kill_sb, 7409 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP | FS_MGTIME, 7410 }; 7411 MODULE_ALIAS_FS("ext4"); 7412 7413 static int __init ext4_init_fs(void) 7414 { 7415 int err; 7416 7417 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7418 ext4_li_info = NULL; 7419 7420 /* Build-time check for flags consistency */ 7421 ext4_check_flag_values(); 7422 7423 err = ext4_init_es(); 7424 if (err) 7425 return err; 7426 7427 err = ext4_init_pending(); 7428 if (err) 7429 goto out7; 7430 7431 err = ext4_init_post_read_processing(); 7432 if (err) 7433 goto out6; 7434 7435 err = ext4_init_pageio(); 7436 if (err) 7437 goto out5; 7438 7439 err = ext4_init_system_zone(); 7440 if (err) 7441 goto out4; 7442 7443 err = ext4_init_sysfs(); 7444 if (err) 7445 goto out3; 7446 7447 err = ext4_init_mballoc(); 7448 if (err) 7449 goto out2; 7450 err = init_inodecache(); 7451 if (err) 7452 goto out1; 7453 7454 err = ext4_fc_init_dentry_cache(); 7455 if (err) 7456 goto out05; 7457 7458 register_as_ext3(); 7459 register_as_ext2(); 7460 err = register_filesystem(&ext4_fs_type); 7461 if (err) 7462 goto out; 7463 7464 return 0; 7465 out: 7466 unregister_as_ext2(); 7467 unregister_as_ext3(); 7468 ext4_fc_destroy_dentry_cache(); 7469 out05: 7470 destroy_inodecache(); 7471 out1: 7472 ext4_exit_mballoc(); 7473 out2: 7474 ext4_exit_sysfs(); 7475 out3: 7476 ext4_exit_system_zone(); 7477 out4: 7478 ext4_exit_pageio(); 7479 out5: 7480 ext4_exit_post_read_processing(); 7481 out6: 7482 ext4_exit_pending(); 7483 out7: 7484 ext4_exit_es(); 7485 7486 return err; 7487 } 7488 7489 static void __exit ext4_exit_fs(void) 7490 { 7491 ext4_destroy_lazyinit_thread(); 7492 unregister_as_ext2(); 7493 unregister_as_ext3(); 7494 unregister_filesystem(&ext4_fs_type); 7495 ext4_fc_destroy_dentry_cache(); 7496 destroy_inodecache(); 7497 ext4_exit_mballoc(); 7498 ext4_exit_sysfs(); 7499 ext4_exit_system_zone(); 7500 ext4_exit_pageio(); 7501 ext4_exit_post_read_processing(); 7502 ext4_exit_es(); 7503 ext4_exit_pending(); 7504 } 7505 7506 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7507 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7508 MODULE_LICENSE("GPL"); 7509 module_init(ext4_init_fs) 7510 module_exit(ext4_exit_fs) 7511