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 super_block *sb, 290 struct ext4_super_block *es) 291 { 292 struct ext4_sb_info *sbi = EXT4_SB(sb); 293 int offset = offsetof(struct ext4_super_block, s_checksum); 294 __u32 csum; 295 296 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 297 298 return cpu_to_le32(csum); 299 } 300 301 static int ext4_superblock_csum_verify(struct super_block *sb, 302 struct ext4_super_block *es) 303 { 304 if (!ext4_has_feature_metadata_csum(sb)) 305 return 1; 306 307 return es->s_checksum == ext4_superblock_csum(sb, es); 308 } 309 310 void ext4_superblock_csum_set(struct super_block *sb) 311 { 312 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 313 314 if (!ext4_has_feature_metadata_csum(sb)) 315 return; 316 317 es->s_checksum = ext4_superblock_csum(sb, es); 318 } 319 320 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 321 struct ext4_group_desc *bg) 322 { 323 return le32_to_cpu(bg->bg_block_bitmap_lo) | 324 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 325 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 326 } 327 328 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 329 struct ext4_group_desc *bg) 330 { 331 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 332 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 333 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 334 } 335 336 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 337 struct ext4_group_desc *bg) 338 { 339 return le32_to_cpu(bg->bg_inode_table_lo) | 340 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 341 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 342 } 343 344 __u32 ext4_free_group_clusters(struct super_block *sb, 345 struct ext4_group_desc *bg) 346 { 347 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 348 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 349 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 350 } 351 352 __u32 ext4_free_inodes_count(struct super_block *sb, 353 struct ext4_group_desc *bg) 354 { 355 return le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_lo)) | 356 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 357 (__u32)le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_hi)) << 16 : 0); 358 } 359 360 __u32 ext4_used_dirs_count(struct super_block *sb, 361 struct ext4_group_desc *bg) 362 { 363 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 364 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 365 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 366 } 367 368 __u32 ext4_itable_unused_count(struct super_block *sb, 369 struct ext4_group_desc *bg) 370 { 371 return le16_to_cpu(bg->bg_itable_unused_lo) | 372 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 373 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 374 } 375 376 void ext4_block_bitmap_set(struct super_block *sb, 377 struct ext4_group_desc *bg, ext4_fsblk_t blk) 378 { 379 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 380 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 381 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 382 } 383 384 void ext4_inode_bitmap_set(struct super_block *sb, 385 struct ext4_group_desc *bg, ext4_fsblk_t blk) 386 { 387 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 388 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 389 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 390 } 391 392 void ext4_inode_table_set(struct super_block *sb, 393 struct ext4_group_desc *bg, ext4_fsblk_t blk) 394 { 395 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 396 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 397 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 398 } 399 400 void ext4_free_group_clusters_set(struct super_block *sb, 401 struct ext4_group_desc *bg, __u32 count) 402 { 403 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 404 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 405 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 406 } 407 408 void ext4_free_inodes_set(struct super_block *sb, 409 struct ext4_group_desc *bg, __u32 count) 410 { 411 WRITE_ONCE(bg->bg_free_inodes_count_lo, cpu_to_le16((__u16)count)); 412 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 413 WRITE_ONCE(bg->bg_free_inodes_count_hi, cpu_to_le16(count >> 16)); 414 } 415 416 void ext4_used_dirs_set(struct super_block *sb, 417 struct ext4_group_desc *bg, __u32 count) 418 { 419 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 420 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 421 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 422 } 423 424 void ext4_itable_unused_set(struct super_block *sb, 425 struct ext4_group_desc *bg, __u32 count) 426 { 427 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 428 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 429 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 430 } 431 432 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 433 { 434 now = clamp_val(now, 0, (1ull << 40) - 1); 435 436 *lo = cpu_to_le32(lower_32_bits(now)); 437 *hi = upper_32_bits(now); 438 } 439 440 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 441 { 442 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 443 } 444 #define ext4_update_tstamp(es, tstamp) \ 445 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 446 ktime_get_real_seconds()) 447 #define ext4_get_tstamp(es, tstamp) \ 448 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 449 450 /* 451 * The ext4_maybe_update_superblock() function checks and updates the 452 * superblock if needed. 453 * 454 * This function is designed to update the on-disk superblock only under 455 * certain conditions to prevent excessive disk writes and unnecessary 456 * waking of the disk from sleep. The superblock will be updated if: 457 * 1. More than sbi->s_sb_update_sec (def: 1 hour) has passed since the last 458 * superblock update 459 * 2. More than sbi->s_sb_update_kb (def: 16MB) kbs have been written since the 460 * last superblock update. 461 * 462 * @sb: The superblock 463 */ 464 static void ext4_maybe_update_superblock(struct super_block *sb) 465 { 466 struct ext4_sb_info *sbi = EXT4_SB(sb); 467 struct ext4_super_block *es = sbi->s_es; 468 journal_t *journal = sbi->s_journal; 469 time64_t now; 470 __u64 last_update; 471 __u64 lifetime_write_kbytes; 472 __u64 diff_size; 473 474 if (ext4_emergency_state(sb) || sb_rdonly(sb) || 475 !(sb->s_flags & SB_ACTIVE) || !journal || 476 journal->j_flags & JBD2_UNMOUNT) 477 return; 478 479 now = ktime_get_real_seconds(); 480 last_update = ext4_get_tstamp(es, s_wtime); 481 482 if (likely(now - last_update < sbi->s_sb_update_sec)) 483 return; 484 485 lifetime_write_kbytes = sbi->s_kbytes_written + 486 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 487 sbi->s_sectors_written_start) >> 1); 488 489 /* Get the number of kilobytes not written to disk to account 490 * for statistics and compare with a multiple of 16 MB. This 491 * is used to determine when the next superblock commit should 492 * occur (i.e. not more often than once per 16MB if there was 493 * less written in an hour). 494 */ 495 diff_size = lifetime_write_kbytes - le64_to_cpu(es->s_kbytes_written); 496 497 if (diff_size > sbi->s_sb_update_kb) 498 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 499 } 500 501 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 502 { 503 struct super_block *sb = journal->j_private; 504 505 BUG_ON(txn->t_state == T_FINISHED); 506 507 ext4_process_freed_data(sb, txn->t_tid); 508 ext4_maybe_update_superblock(sb); 509 } 510 511 /* 512 * This writepage callback for write_cache_pages() 513 * takes care of a few cases after page cleaning. 514 * 515 * write_cache_pages() already checks for dirty pages 516 * and calls clear_page_dirty_for_io(), which we want, 517 * to write protect the pages. 518 * 519 * However, we may have to redirty a page (see below.) 520 */ 521 static int ext4_journalled_writepage_callback(struct folio *folio, 522 struct writeback_control *wbc, 523 void *data) 524 { 525 transaction_t *transaction = (transaction_t *) data; 526 struct buffer_head *bh, *head; 527 struct journal_head *jh; 528 529 bh = head = folio_buffers(folio); 530 do { 531 /* 532 * We have to redirty a page in these cases: 533 * 1) If buffer is dirty, it means the page was dirty because it 534 * contains a buffer that needs checkpointing. So the dirty bit 535 * needs to be preserved so that checkpointing writes the buffer 536 * properly. 537 * 2) If buffer is not part of the committing transaction 538 * (we may have just accidentally come across this buffer because 539 * inode range tracking is not exact) or if the currently running 540 * transaction already contains this buffer as well, dirty bit 541 * needs to be preserved so that the buffer gets writeprotected 542 * properly on running transaction's commit. 543 */ 544 jh = bh2jh(bh); 545 if (buffer_dirty(bh) || 546 (jh && (jh->b_transaction != transaction || 547 jh->b_next_transaction))) { 548 folio_redirty_for_writepage(wbc, folio); 549 goto out; 550 } 551 } while ((bh = bh->b_this_page) != head); 552 553 out: 554 return AOP_WRITEPAGE_ACTIVATE; 555 } 556 557 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 558 { 559 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 560 struct writeback_control wbc = { 561 .sync_mode = WB_SYNC_ALL, 562 .nr_to_write = LONG_MAX, 563 .range_start = jinode->i_dirty_start, 564 .range_end = jinode->i_dirty_end, 565 }; 566 567 return write_cache_pages(mapping, &wbc, 568 ext4_journalled_writepage_callback, 569 jinode->i_transaction); 570 } 571 572 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 573 { 574 int ret; 575 576 if (ext4_should_journal_data(jinode->i_vfs_inode)) 577 ret = ext4_journalled_submit_inode_data_buffers(jinode); 578 else 579 ret = ext4_normal_submit_inode_data_buffers(jinode); 580 return ret; 581 } 582 583 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 584 { 585 int ret = 0; 586 587 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 588 ret = jbd2_journal_finish_inode_data_buffers(jinode); 589 590 return ret; 591 } 592 593 static bool system_going_down(void) 594 { 595 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 596 || system_state == SYSTEM_RESTART; 597 } 598 599 struct ext4_err_translation { 600 int code; 601 int errno; 602 }; 603 604 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 605 606 static struct ext4_err_translation err_translation[] = { 607 EXT4_ERR_TRANSLATE(EIO), 608 EXT4_ERR_TRANSLATE(ENOMEM), 609 EXT4_ERR_TRANSLATE(EFSBADCRC), 610 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 611 EXT4_ERR_TRANSLATE(ENOSPC), 612 EXT4_ERR_TRANSLATE(ENOKEY), 613 EXT4_ERR_TRANSLATE(EROFS), 614 EXT4_ERR_TRANSLATE(EFBIG), 615 EXT4_ERR_TRANSLATE(EEXIST), 616 EXT4_ERR_TRANSLATE(ERANGE), 617 EXT4_ERR_TRANSLATE(EOVERFLOW), 618 EXT4_ERR_TRANSLATE(EBUSY), 619 EXT4_ERR_TRANSLATE(ENOTDIR), 620 EXT4_ERR_TRANSLATE(ENOTEMPTY), 621 EXT4_ERR_TRANSLATE(ESHUTDOWN), 622 EXT4_ERR_TRANSLATE(EFAULT), 623 }; 624 625 static int ext4_errno_to_code(int errno) 626 { 627 int i; 628 629 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 630 if (err_translation[i].errno == errno) 631 return err_translation[i].code; 632 return EXT4_ERR_UNKNOWN; 633 } 634 635 static void save_error_info(struct super_block *sb, int error, 636 __u32 ino, __u64 block, 637 const char *func, unsigned int line) 638 { 639 struct ext4_sb_info *sbi = EXT4_SB(sb); 640 641 /* We default to EFSCORRUPTED error... */ 642 if (error == 0) 643 error = EFSCORRUPTED; 644 645 spin_lock(&sbi->s_error_lock); 646 sbi->s_add_error_count++; 647 sbi->s_last_error_code = error; 648 sbi->s_last_error_line = line; 649 sbi->s_last_error_ino = ino; 650 sbi->s_last_error_block = block; 651 sbi->s_last_error_func = func; 652 sbi->s_last_error_time = ktime_get_real_seconds(); 653 if (!sbi->s_first_error_time) { 654 sbi->s_first_error_code = error; 655 sbi->s_first_error_line = line; 656 sbi->s_first_error_ino = ino; 657 sbi->s_first_error_block = block; 658 sbi->s_first_error_func = func; 659 sbi->s_first_error_time = sbi->s_last_error_time; 660 } 661 spin_unlock(&sbi->s_error_lock); 662 } 663 664 /* Deal with the reporting of failure conditions on a filesystem such as 665 * inconsistencies detected or read IO failures. 666 * 667 * On ext2, we can store the error state of the filesystem in the 668 * superblock. That is not possible on ext4, because we may have other 669 * write ordering constraints on the superblock which prevent us from 670 * writing it out straight away; and given that the journal is about to 671 * be aborted, we can't rely on the current, or future, transactions to 672 * write out the superblock safely. 673 * 674 * We'll just use the jbd2_journal_abort() error code to record an error in 675 * the journal instead. On recovery, the journal will complain about 676 * that error until we've noted it down and cleared it. 677 * 678 * If force_ro is set, we unconditionally force the filesystem into an 679 * ABORT|READONLY state, unless the error response on the fs has been set to 680 * panic in which case we take the easy way out and panic immediately. This is 681 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 682 * at a critical moment in log management. 683 */ 684 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 685 __u32 ino, __u64 block, 686 const char *func, unsigned int line) 687 { 688 journal_t *journal = EXT4_SB(sb)->s_journal; 689 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 690 691 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 692 if (test_opt(sb, WARN_ON_ERROR)) 693 WARN_ON_ONCE(1); 694 695 if (!continue_fs && !ext4_emergency_ro(sb) && journal) 696 jbd2_journal_abort(journal, -EIO); 697 698 if (!bdev_read_only(sb->s_bdev)) { 699 save_error_info(sb, error, ino, block, func, line); 700 /* 701 * In case the fs should keep running, we need to writeout 702 * superblock through the journal. Due to lock ordering 703 * constraints, it may not be safe to do it right here so we 704 * defer superblock flushing to a workqueue. We just need to be 705 * careful when the journal is already shutting down. If we get 706 * here in that case, just update the sb directly as the last 707 * transaction won't commit anyway. 708 */ 709 if (continue_fs && journal && 710 !ext4_test_mount_flag(sb, EXT4_MF_JOURNAL_DESTROY)) 711 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 712 else 713 ext4_commit_super(sb); 714 } 715 716 /* 717 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 718 * could panic during 'reboot -f' as the underlying device got already 719 * disabled. 720 */ 721 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 722 panic("EXT4-fs (device %s): panic forced after error\n", 723 sb->s_id); 724 } 725 726 if (ext4_emergency_ro(sb) || continue_fs) 727 return; 728 729 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 730 /* 731 * We don't set SB_RDONLY because that requires sb->s_umount 732 * semaphore and setting it without proper remount procedure is 733 * confusing code such as freeze_super() leading to deadlocks 734 * and other problems. 735 */ 736 set_bit(EXT4_FLAGS_EMERGENCY_RO, &EXT4_SB(sb)->s_ext4_flags); 737 } 738 739 static void update_super_work(struct work_struct *work) 740 { 741 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 742 s_sb_upd_work); 743 journal_t *journal = sbi->s_journal; 744 handle_t *handle; 745 746 /* 747 * If the journal is still running, we have to write out superblock 748 * through the journal to avoid collisions of other journalled sb 749 * updates. 750 * 751 * We use directly jbd2 functions here to avoid recursing back into 752 * ext4 error handling code during handling of previous errors. 753 */ 754 if (!ext4_emergency_state(sbi->s_sb) && 755 !sb_rdonly(sbi->s_sb) && journal) { 756 struct buffer_head *sbh = sbi->s_sbh; 757 bool call_notify_err = false; 758 759 handle = jbd2_journal_start(journal, 1); 760 if (IS_ERR(handle)) 761 goto write_directly; 762 if (jbd2_journal_get_write_access(handle, sbh)) { 763 jbd2_journal_stop(handle); 764 goto write_directly; 765 } 766 767 if (sbi->s_add_error_count > 0) 768 call_notify_err = true; 769 770 ext4_update_super(sbi->s_sb); 771 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 772 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 773 "superblock detected"); 774 clear_buffer_write_io_error(sbh); 775 set_buffer_uptodate(sbh); 776 } 777 778 if (jbd2_journal_dirty_metadata(handle, sbh)) { 779 jbd2_journal_stop(handle); 780 goto write_directly; 781 } 782 jbd2_journal_stop(handle); 783 784 if (call_notify_err) 785 ext4_notify_error_sysfs(sbi); 786 787 return; 788 } 789 write_directly: 790 /* 791 * Write through journal failed. Write sb directly to get error info 792 * out and hope for the best. 793 */ 794 ext4_commit_super(sbi->s_sb); 795 ext4_notify_error_sysfs(sbi); 796 } 797 798 #define ext4_error_ratelimit(sb) \ 799 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 800 "EXT4-fs error") 801 802 void __ext4_error(struct super_block *sb, const char *function, 803 unsigned int line, bool force_ro, int error, __u64 block, 804 const char *fmt, ...) 805 { 806 struct va_format vaf; 807 va_list args; 808 809 if (unlikely(ext4_emergency_state(sb))) 810 return; 811 812 trace_ext4_error(sb, function, line); 813 if (ext4_error_ratelimit(sb)) { 814 va_start(args, fmt); 815 vaf.fmt = fmt; 816 vaf.va = &args; 817 printk(KERN_CRIT 818 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 819 sb->s_id, function, line, current->comm, &vaf); 820 va_end(args); 821 } 822 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED); 823 824 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 825 } 826 827 void __ext4_error_inode(struct inode *inode, const char *function, 828 unsigned int line, ext4_fsblk_t block, int error, 829 const char *fmt, ...) 830 { 831 va_list args; 832 struct va_format vaf; 833 834 if (unlikely(ext4_emergency_state(inode->i_sb))) 835 return; 836 837 trace_ext4_error(inode->i_sb, function, line); 838 if (ext4_error_ratelimit(inode->i_sb)) { 839 va_start(args, fmt); 840 vaf.fmt = fmt; 841 vaf.va = &args; 842 if (block) 843 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 844 "inode #%lu: block %llu: comm %s: %pV\n", 845 inode->i_sb->s_id, function, line, inode->i_ino, 846 block, current->comm, &vaf); 847 else 848 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 849 "inode #%lu: comm %s: %pV\n", 850 inode->i_sb->s_id, function, line, inode->i_ino, 851 current->comm, &vaf); 852 va_end(args); 853 } 854 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED); 855 856 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 857 function, line); 858 } 859 860 void __ext4_error_file(struct file *file, const char *function, 861 unsigned int line, ext4_fsblk_t block, 862 const char *fmt, ...) 863 { 864 va_list args; 865 struct va_format vaf; 866 struct inode *inode = file_inode(file); 867 char pathname[80], *path; 868 869 if (unlikely(ext4_emergency_state(inode->i_sb))) 870 return; 871 872 trace_ext4_error(inode->i_sb, function, line); 873 if (ext4_error_ratelimit(inode->i_sb)) { 874 path = file_path(file, pathname, sizeof(pathname)); 875 if (IS_ERR(path)) 876 path = "(unknown)"; 877 va_start(args, fmt); 878 vaf.fmt = fmt; 879 vaf.va = &args; 880 if (block) 881 printk(KERN_CRIT 882 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 883 "block %llu: comm %s: path %s: %pV\n", 884 inode->i_sb->s_id, function, line, inode->i_ino, 885 block, current->comm, path, &vaf); 886 else 887 printk(KERN_CRIT 888 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 889 "comm %s: path %s: %pV\n", 890 inode->i_sb->s_id, function, line, inode->i_ino, 891 current->comm, path, &vaf); 892 va_end(args); 893 } 894 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED); 895 896 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 897 function, line); 898 } 899 900 const char *ext4_decode_error(struct super_block *sb, int errno, 901 char nbuf[16]) 902 { 903 char *errstr = NULL; 904 905 switch (errno) { 906 case -EFSCORRUPTED: 907 errstr = "Corrupt filesystem"; 908 break; 909 case -EFSBADCRC: 910 errstr = "Filesystem failed CRC"; 911 break; 912 case -EIO: 913 errstr = "IO failure"; 914 break; 915 case -ENOMEM: 916 errstr = "Out of memory"; 917 break; 918 case -EROFS: 919 if (!sb || (EXT4_SB(sb)->s_journal && 920 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 921 errstr = "Journal has aborted"; 922 else 923 errstr = "Readonly filesystem"; 924 break; 925 default: 926 /* If the caller passed in an extra buffer for unknown 927 * errors, textualise them now. Else we just return 928 * NULL. */ 929 if (nbuf) { 930 /* Check for truncated error codes... */ 931 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 932 errstr = nbuf; 933 } 934 break; 935 } 936 937 return errstr; 938 } 939 940 /* __ext4_std_error decodes expected errors from journaling functions 941 * automatically and invokes the appropriate error response. */ 942 943 void __ext4_std_error(struct super_block *sb, const char *function, 944 unsigned int line, int errno) 945 { 946 char nbuf[16]; 947 const char *errstr; 948 949 if (unlikely(ext4_emergency_state(sb))) 950 return; 951 952 /* Special case: if the error is EROFS, and we're not already 953 * inside a transaction, then there's really no point in logging 954 * an error. */ 955 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 956 return; 957 958 if (ext4_error_ratelimit(sb)) { 959 errstr = ext4_decode_error(sb, errno, nbuf); 960 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 961 sb->s_id, function, line, errstr); 962 } 963 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED); 964 965 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 966 } 967 968 void __ext4_msg(struct super_block *sb, 969 const char *prefix, const char *fmt, ...) 970 { 971 struct va_format vaf; 972 va_list args; 973 974 if (sb) { 975 atomic_inc(&EXT4_SB(sb)->s_msg_count); 976 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 977 "EXT4-fs")) 978 return; 979 } 980 981 va_start(args, fmt); 982 vaf.fmt = fmt; 983 vaf.va = &args; 984 if (sb) 985 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 986 else 987 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 988 va_end(args); 989 } 990 991 static int ext4_warning_ratelimit(struct super_block *sb) 992 { 993 atomic_inc(&EXT4_SB(sb)->s_warning_count); 994 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 995 "EXT4-fs warning"); 996 } 997 998 void __ext4_warning(struct super_block *sb, const char *function, 999 unsigned int line, const char *fmt, ...) 1000 { 1001 struct va_format vaf; 1002 va_list args; 1003 1004 if (!ext4_warning_ratelimit(sb)) 1005 return; 1006 1007 va_start(args, fmt); 1008 vaf.fmt = fmt; 1009 vaf.va = &args; 1010 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 1011 sb->s_id, function, line, &vaf); 1012 va_end(args); 1013 } 1014 1015 void __ext4_warning_inode(const struct inode *inode, const char *function, 1016 unsigned int line, const char *fmt, ...) 1017 { 1018 struct va_format vaf; 1019 va_list args; 1020 1021 if (!ext4_warning_ratelimit(inode->i_sb)) 1022 return; 1023 1024 va_start(args, fmt); 1025 vaf.fmt = fmt; 1026 vaf.va = &args; 1027 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 1028 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 1029 function, line, inode->i_ino, current->comm, &vaf); 1030 va_end(args); 1031 } 1032 1033 void __ext4_grp_locked_error(const char *function, unsigned int line, 1034 struct super_block *sb, ext4_group_t grp, 1035 unsigned long ino, ext4_fsblk_t block, 1036 const char *fmt, ...) 1037 __releases(bitlock) 1038 __acquires(bitlock) 1039 { 1040 struct va_format vaf; 1041 va_list args; 1042 1043 if (unlikely(ext4_emergency_state(sb))) 1044 return; 1045 1046 trace_ext4_error(sb, function, line); 1047 if (ext4_error_ratelimit(sb)) { 1048 va_start(args, fmt); 1049 vaf.fmt = fmt; 1050 vaf.va = &args; 1051 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1052 sb->s_id, function, line, grp); 1053 if (ino) 1054 printk(KERN_CONT "inode %lu: ", ino); 1055 if (block) 1056 printk(KERN_CONT "block %llu:", 1057 (unsigned long long) block); 1058 printk(KERN_CONT "%pV\n", &vaf); 1059 va_end(args); 1060 } 1061 1062 if (test_opt(sb, ERRORS_CONT)) { 1063 if (test_opt(sb, WARN_ON_ERROR)) 1064 WARN_ON_ONCE(1); 1065 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1066 if (!bdev_read_only(sb->s_bdev)) { 1067 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1068 line); 1069 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 1070 } 1071 return; 1072 } 1073 ext4_unlock_group(sb, grp); 1074 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1075 /* 1076 * We only get here in the ERRORS_RO case; relocking the group 1077 * may be dangerous, but nothing bad will happen since the 1078 * filesystem will have already been marked read/only and the 1079 * journal has been aborted. We return 1 as a hint to callers 1080 * who might what to use the return value from 1081 * ext4_grp_locked_error() to distinguish between the 1082 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1083 * aggressively from the ext4 function in question, with a 1084 * more appropriate error code. 1085 */ 1086 ext4_lock_group(sb, grp); 1087 return; 1088 } 1089 1090 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1091 ext4_group_t group, 1092 unsigned int flags) 1093 { 1094 struct ext4_sb_info *sbi = EXT4_SB(sb); 1095 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1096 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1097 int ret; 1098 1099 if (!grp || !gdp) 1100 return; 1101 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1102 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1103 &grp->bb_state); 1104 if (!ret) 1105 percpu_counter_sub(&sbi->s_freeclusters_counter, 1106 grp->bb_free); 1107 } 1108 1109 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1110 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1111 &grp->bb_state); 1112 if (!ret && gdp) { 1113 int count; 1114 1115 count = ext4_free_inodes_count(sb, gdp); 1116 percpu_counter_sub(&sbi->s_freeinodes_counter, 1117 count); 1118 } 1119 } 1120 } 1121 1122 void ext4_update_dynamic_rev(struct super_block *sb) 1123 { 1124 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1125 1126 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1127 return; 1128 1129 ext4_warning(sb, 1130 "updating to rev %d because of new feature flag, " 1131 "running e2fsck is recommended", 1132 EXT4_DYNAMIC_REV); 1133 1134 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1135 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1136 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1137 /* leave es->s_feature_*compat flags alone */ 1138 /* es->s_uuid will be set by e2fsck if empty */ 1139 1140 /* 1141 * The rest of the superblock fields should be zero, and if not it 1142 * means they are likely already in use, so leave them alone. We 1143 * can leave it up to e2fsck to clean up any inconsistencies there. 1144 */ 1145 } 1146 1147 static inline struct inode *orphan_list_entry(struct list_head *l) 1148 { 1149 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1150 } 1151 1152 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1153 { 1154 struct list_head *l; 1155 1156 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1157 le32_to_cpu(sbi->s_es->s_last_orphan)); 1158 1159 printk(KERN_ERR "sb_info orphan list:\n"); 1160 list_for_each(l, &sbi->s_orphan) { 1161 struct inode *inode = orphan_list_entry(l); 1162 printk(KERN_ERR " " 1163 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1164 inode->i_sb->s_id, inode->i_ino, inode, 1165 inode->i_mode, inode->i_nlink, 1166 NEXT_ORPHAN(inode)); 1167 } 1168 } 1169 1170 #ifdef CONFIG_QUOTA 1171 static int ext4_quota_off(struct super_block *sb, int type); 1172 1173 static inline void ext4_quotas_off(struct super_block *sb, int type) 1174 { 1175 BUG_ON(type > EXT4_MAXQUOTAS); 1176 1177 /* Use our quota_off function to clear inode flags etc. */ 1178 for (type--; type >= 0; type--) 1179 ext4_quota_off(sb, type); 1180 } 1181 1182 /* 1183 * This is a helper function which is used in the mount/remount 1184 * codepaths (which holds s_umount) to fetch the quota file name. 1185 */ 1186 static inline char *get_qf_name(struct super_block *sb, 1187 struct ext4_sb_info *sbi, 1188 int type) 1189 { 1190 return rcu_dereference_protected(sbi->s_qf_names[type], 1191 lockdep_is_held(&sb->s_umount)); 1192 } 1193 #else 1194 static inline void ext4_quotas_off(struct super_block *sb, int type) 1195 { 1196 } 1197 #endif 1198 1199 static int ext4_percpu_param_init(struct ext4_sb_info *sbi) 1200 { 1201 ext4_fsblk_t block; 1202 int err; 1203 1204 block = ext4_count_free_clusters(sbi->s_sb); 1205 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block)); 1206 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 1207 GFP_KERNEL); 1208 if (!err) { 1209 unsigned long freei = ext4_count_free_inodes(sbi->s_sb); 1210 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 1211 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 1212 GFP_KERNEL); 1213 } 1214 if (!err) 1215 err = percpu_counter_init(&sbi->s_dirs_counter, 1216 ext4_count_dirs(sbi->s_sb), GFP_KERNEL); 1217 if (!err) 1218 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 1219 GFP_KERNEL); 1220 if (!err) 1221 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 1222 GFP_KERNEL); 1223 if (!err) 1224 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 1225 1226 if (err) 1227 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory"); 1228 1229 return err; 1230 } 1231 1232 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi) 1233 { 1234 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1235 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1236 percpu_counter_destroy(&sbi->s_dirs_counter); 1237 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1238 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1239 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1240 } 1241 1242 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 1243 { 1244 struct buffer_head **group_desc; 1245 int i; 1246 1247 rcu_read_lock(); 1248 group_desc = rcu_dereference(sbi->s_group_desc); 1249 for (i = 0; i < sbi->s_gdb_count; i++) 1250 brelse(group_desc[i]); 1251 kvfree(group_desc); 1252 rcu_read_unlock(); 1253 } 1254 1255 static void ext4_flex_groups_free(struct ext4_sb_info *sbi) 1256 { 1257 struct flex_groups **flex_groups; 1258 int i; 1259 1260 rcu_read_lock(); 1261 flex_groups = rcu_dereference(sbi->s_flex_groups); 1262 if (flex_groups) { 1263 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1264 kvfree(flex_groups[i]); 1265 kvfree(flex_groups); 1266 } 1267 rcu_read_unlock(); 1268 } 1269 1270 static void ext4_put_super(struct super_block *sb) 1271 { 1272 struct ext4_sb_info *sbi = EXT4_SB(sb); 1273 struct ext4_super_block *es = sbi->s_es; 1274 int aborted = 0; 1275 int err; 1276 1277 /* 1278 * Unregister sysfs before destroying jbd2 journal. 1279 * Since we could still access attr_journal_task attribute via sysfs 1280 * path which could have sbi->s_journal->j_task as NULL 1281 * Unregister sysfs before flush sbi->s_sb_upd_work. 1282 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1283 * read metadata verify failed then will queue error work. 1284 * update_super_work will call start_this_handle may trigger 1285 * BUG_ON. 1286 */ 1287 ext4_unregister_sysfs(sb); 1288 1289 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1290 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.", 1291 &sb->s_uuid); 1292 1293 ext4_unregister_li_request(sb); 1294 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 1295 1296 destroy_workqueue(sbi->rsv_conversion_wq); 1297 ext4_release_orphan_info(sb); 1298 1299 if (sbi->s_journal) { 1300 aborted = is_journal_aborted(sbi->s_journal); 1301 err = ext4_journal_destroy(sbi, sbi->s_journal); 1302 if ((err < 0) && !aborted) { 1303 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1304 } 1305 } else 1306 flush_work(&sbi->s_sb_upd_work); 1307 1308 ext4_es_unregister_shrinker(sbi); 1309 timer_shutdown_sync(&sbi->s_err_report); 1310 ext4_release_system_zone(sb); 1311 ext4_mb_release(sb); 1312 ext4_ext_release(sb); 1313 1314 if (!ext4_emergency_state(sb) && !sb_rdonly(sb)) { 1315 if (!aborted) { 1316 ext4_clear_feature_journal_needs_recovery(sb); 1317 ext4_clear_feature_orphan_present(sb); 1318 es->s_state = cpu_to_le16(sbi->s_mount_state); 1319 } 1320 ext4_commit_super(sb); 1321 } 1322 1323 ext4_group_desc_free(sbi); 1324 ext4_flex_groups_free(sbi); 1325 1326 WARN_ON_ONCE(!(sbi->s_mount_state & EXT4_ERROR_FS) && 1327 percpu_counter_sum(&sbi->s_dirtyclusters_counter)); 1328 ext4_percpu_param_destroy(sbi); 1329 #ifdef CONFIG_QUOTA 1330 for (int i = 0; i < EXT4_MAXQUOTAS; i++) 1331 kfree(get_qf_name(sb, sbi, i)); 1332 #endif 1333 1334 /* Debugging code just in case the in-memory inode orphan list 1335 * isn't empty. The on-disk one can be non-empty if we've 1336 * detected an error and taken the fs readonly, but the 1337 * in-memory list had better be clean by this point. */ 1338 if (!list_empty(&sbi->s_orphan)) 1339 dump_orphan_list(sb, sbi); 1340 ASSERT(list_empty(&sbi->s_orphan)); 1341 1342 sync_blockdev(sb->s_bdev); 1343 invalidate_bdev(sb->s_bdev); 1344 if (sbi->s_journal_bdev_file) { 1345 /* 1346 * Invalidate the journal device's buffers. We don't want them 1347 * floating about in memory - the physical journal device may 1348 * hotswapped, and it breaks the `ro-after' testing code. 1349 */ 1350 sync_blockdev(file_bdev(sbi->s_journal_bdev_file)); 1351 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 1352 } 1353 1354 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1355 sbi->s_ea_inode_cache = NULL; 1356 1357 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1358 sbi->s_ea_block_cache = NULL; 1359 1360 ext4_stop_mmpd(sbi); 1361 1362 brelse(sbi->s_sbh); 1363 sb->s_fs_info = NULL; 1364 /* 1365 * Now that we are completely done shutting down the 1366 * superblock, we need to actually destroy the kobject. 1367 */ 1368 kobject_put(&sbi->s_kobj); 1369 wait_for_completion(&sbi->s_kobj_unregister); 1370 kfree(sbi->s_blockgroup_lock); 1371 fs_put_dax(sbi->s_daxdev, NULL); 1372 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1373 #if IS_ENABLED(CONFIG_UNICODE) 1374 utf8_unload(sb->s_encoding); 1375 #endif 1376 kfree(sbi); 1377 } 1378 1379 static struct kmem_cache *ext4_inode_cachep; 1380 1381 /* 1382 * Called inside transaction, so use GFP_NOFS 1383 */ 1384 static struct inode *ext4_alloc_inode(struct super_block *sb) 1385 { 1386 struct ext4_inode_info *ei; 1387 1388 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1389 if (!ei) 1390 return NULL; 1391 1392 inode_set_iversion(&ei->vfs_inode, 1); 1393 ei->i_flags = 0; 1394 spin_lock_init(&ei->i_raw_lock); 1395 ei->i_prealloc_node = RB_ROOT; 1396 atomic_set(&ei->i_prealloc_active, 0); 1397 rwlock_init(&ei->i_prealloc_lock); 1398 ext4_es_init_tree(&ei->i_es_tree); 1399 rwlock_init(&ei->i_es_lock); 1400 INIT_LIST_HEAD(&ei->i_es_list); 1401 ei->i_es_all_nr = 0; 1402 ei->i_es_shk_nr = 0; 1403 ei->i_es_shrink_lblk = 0; 1404 ei->i_reserved_data_blocks = 0; 1405 spin_lock_init(&(ei->i_block_reservation_lock)); 1406 ext4_init_pending_tree(&ei->i_pending_tree); 1407 #ifdef CONFIG_QUOTA 1408 ei->i_reserved_quota = 0; 1409 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1410 #endif 1411 ei->jinode = NULL; 1412 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1413 spin_lock_init(&ei->i_completed_io_lock); 1414 ei->i_sync_tid = 0; 1415 ei->i_datasync_tid = 0; 1416 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1417 ext4_fc_init_inode(&ei->vfs_inode); 1418 mutex_init(&ei->i_fc_lock); 1419 return &ei->vfs_inode; 1420 } 1421 1422 static int ext4_drop_inode(struct inode *inode) 1423 { 1424 int drop = generic_drop_inode(inode); 1425 1426 if (!drop) 1427 drop = fscrypt_drop_inode(inode); 1428 1429 trace_ext4_drop_inode(inode, drop); 1430 return drop; 1431 } 1432 1433 static void ext4_free_in_core_inode(struct inode *inode) 1434 { 1435 fscrypt_free_inode(inode); 1436 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1437 pr_warn("%s: inode %ld still in fc list", 1438 __func__, inode->i_ino); 1439 } 1440 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1441 } 1442 1443 static void ext4_destroy_inode(struct inode *inode) 1444 { 1445 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1446 ext4_msg(inode->i_sb, KERN_ERR, 1447 "Inode %lu (%p): orphan list check failed!", 1448 inode->i_ino, EXT4_I(inode)); 1449 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1450 EXT4_I(inode), sizeof(struct ext4_inode_info), 1451 true); 1452 dump_stack(); 1453 } 1454 1455 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ERROR_FS) && 1456 WARN_ON_ONCE(EXT4_I(inode)->i_reserved_data_blocks)) 1457 ext4_msg(inode->i_sb, KERN_ERR, 1458 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!", 1459 inode->i_ino, EXT4_I(inode), 1460 EXT4_I(inode)->i_reserved_data_blocks); 1461 } 1462 1463 static void ext4_shutdown(struct super_block *sb) 1464 { 1465 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH); 1466 } 1467 1468 static void init_once(void *foo) 1469 { 1470 struct ext4_inode_info *ei = foo; 1471 1472 INIT_LIST_HEAD(&ei->i_orphan); 1473 init_rwsem(&ei->xattr_sem); 1474 init_rwsem(&ei->i_data_sem); 1475 inode_init_once(&ei->vfs_inode); 1476 ext4_fc_init_inode(&ei->vfs_inode); 1477 } 1478 1479 static int __init init_inodecache(void) 1480 { 1481 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1482 sizeof(struct ext4_inode_info), 0, 1483 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT, 1484 offsetof(struct ext4_inode_info, i_data), 1485 sizeof_field(struct ext4_inode_info, i_data), 1486 init_once); 1487 if (ext4_inode_cachep == NULL) 1488 return -ENOMEM; 1489 return 0; 1490 } 1491 1492 static void destroy_inodecache(void) 1493 { 1494 /* 1495 * Make sure all delayed rcu free inodes are flushed before we 1496 * destroy cache. 1497 */ 1498 rcu_barrier(); 1499 kmem_cache_destroy(ext4_inode_cachep); 1500 } 1501 1502 void ext4_clear_inode(struct inode *inode) 1503 { 1504 ext4_fc_del(inode); 1505 invalidate_inode_buffers(inode); 1506 clear_inode(inode); 1507 ext4_discard_preallocations(inode); 1508 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1509 dquot_drop(inode); 1510 if (EXT4_I(inode)->jinode) { 1511 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1512 EXT4_I(inode)->jinode); 1513 jbd2_free_inode(EXT4_I(inode)->jinode); 1514 EXT4_I(inode)->jinode = NULL; 1515 } 1516 fscrypt_put_encryption_info(inode); 1517 fsverity_cleanup_inode(inode); 1518 } 1519 1520 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1521 u64 ino, u32 generation) 1522 { 1523 struct inode *inode; 1524 1525 /* 1526 * Currently we don't know the generation for parent directory, so 1527 * a generation of 0 means "accept any" 1528 */ 1529 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1530 if (IS_ERR(inode)) 1531 return ERR_CAST(inode); 1532 if (generation && inode->i_generation != generation) { 1533 iput(inode); 1534 return ERR_PTR(-ESTALE); 1535 } 1536 1537 return inode; 1538 } 1539 1540 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1541 int fh_len, int fh_type) 1542 { 1543 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1544 ext4_nfs_get_inode); 1545 } 1546 1547 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1548 int fh_len, int fh_type) 1549 { 1550 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1551 ext4_nfs_get_inode); 1552 } 1553 1554 static int ext4_nfs_commit_metadata(struct inode *inode) 1555 { 1556 struct writeback_control wbc = { 1557 .sync_mode = WB_SYNC_ALL 1558 }; 1559 1560 trace_ext4_nfs_commit_metadata(inode); 1561 return ext4_write_inode(inode, &wbc); 1562 } 1563 1564 #ifdef CONFIG_QUOTA 1565 static const char * const quotatypes[] = INITQFNAMES; 1566 #define QTYPE2NAME(t) (quotatypes[t]) 1567 1568 static int ext4_write_dquot(struct dquot *dquot); 1569 static int ext4_acquire_dquot(struct dquot *dquot); 1570 static int ext4_release_dquot(struct dquot *dquot); 1571 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1572 static int ext4_write_info(struct super_block *sb, int type); 1573 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1574 const struct path *path); 1575 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1576 size_t len, loff_t off); 1577 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1578 const char *data, size_t len, loff_t off); 1579 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1580 unsigned int flags); 1581 1582 static struct dquot __rcu **ext4_get_dquots(struct inode *inode) 1583 { 1584 return EXT4_I(inode)->i_dquot; 1585 } 1586 1587 static const struct dquot_operations ext4_quota_operations = { 1588 .get_reserved_space = ext4_get_reserved_space, 1589 .write_dquot = ext4_write_dquot, 1590 .acquire_dquot = ext4_acquire_dquot, 1591 .release_dquot = ext4_release_dquot, 1592 .mark_dirty = ext4_mark_dquot_dirty, 1593 .write_info = ext4_write_info, 1594 .alloc_dquot = dquot_alloc, 1595 .destroy_dquot = dquot_destroy, 1596 .get_projid = ext4_get_projid, 1597 .get_inode_usage = ext4_get_inode_usage, 1598 .get_next_id = dquot_get_next_id, 1599 }; 1600 1601 static const struct quotactl_ops ext4_qctl_operations = { 1602 .quota_on = ext4_quota_on, 1603 .quota_off = ext4_quota_off, 1604 .quota_sync = dquot_quota_sync, 1605 .get_state = dquot_get_state, 1606 .set_info = dquot_set_dqinfo, 1607 .get_dqblk = dquot_get_dqblk, 1608 .set_dqblk = dquot_set_dqblk, 1609 .get_nextdqblk = dquot_get_next_dqblk, 1610 }; 1611 #endif 1612 1613 static const struct super_operations ext4_sops = { 1614 .alloc_inode = ext4_alloc_inode, 1615 .free_inode = ext4_free_in_core_inode, 1616 .destroy_inode = ext4_destroy_inode, 1617 .write_inode = ext4_write_inode, 1618 .dirty_inode = ext4_dirty_inode, 1619 .drop_inode = ext4_drop_inode, 1620 .evict_inode = ext4_evict_inode, 1621 .put_super = ext4_put_super, 1622 .sync_fs = ext4_sync_fs, 1623 .freeze_fs = ext4_freeze, 1624 .unfreeze_fs = ext4_unfreeze, 1625 .statfs = ext4_statfs, 1626 .show_options = ext4_show_options, 1627 .shutdown = ext4_shutdown, 1628 #ifdef CONFIG_QUOTA 1629 .quota_read = ext4_quota_read, 1630 .quota_write = ext4_quota_write, 1631 .get_dquots = ext4_get_dquots, 1632 #endif 1633 }; 1634 1635 static const struct export_operations ext4_export_ops = { 1636 .encode_fh = generic_encode_ino32_fh, 1637 .fh_to_dentry = ext4_fh_to_dentry, 1638 .fh_to_parent = ext4_fh_to_parent, 1639 .get_parent = ext4_get_parent, 1640 .commit_metadata = ext4_nfs_commit_metadata, 1641 }; 1642 1643 enum { 1644 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1645 Opt_resgid, Opt_resuid, Opt_sb, 1646 Opt_nouid32, Opt_debug, Opt_removed, 1647 Opt_user_xattr, Opt_acl, 1648 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1649 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1650 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1651 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1652 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1653 Opt_inlinecrypt, 1654 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1655 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1656 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1657 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1658 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1659 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1660 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1661 Opt_inode_readahead_blks, Opt_journal_ioprio, 1662 Opt_dioread_nolock, Opt_dioread_lock, 1663 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1664 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1665 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1666 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1667 #ifdef CONFIG_EXT4_DEBUG 1668 Opt_fc_debug_max_replay, Opt_fc_debug_force 1669 #endif 1670 }; 1671 1672 static const struct constant_table ext4_param_errors[] = { 1673 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1674 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1675 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1676 {} 1677 }; 1678 1679 static const struct constant_table ext4_param_data[] = { 1680 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1681 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1682 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1683 {} 1684 }; 1685 1686 static const struct constant_table ext4_param_data_err[] = { 1687 {"abort", Opt_data_err_abort}, 1688 {"ignore", Opt_data_err_ignore}, 1689 {} 1690 }; 1691 1692 static const struct constant_table ext4_param_jqfmt[] = { 1693 {"vfsold", QFMT_VFS_OLD}, 1694 {"vfsv0", QFMT_VFS_V0}, 1695 {"vfsv1", QFMT_VFS_V1}, 1696 {} 1697 }; 1698 1699 static const struct constant_table ext4_param_dax[] = { 1700 {"always", Opt_dax_always}, 1701 {"inode", Opt_dax_inode}, 1702 {"never", Opt_dax_never}, 1703 {} 1704 }; 1705 1706 /* 1707 * Mount option specification 1708 * We don't use fsparam_flag_no because of the way we set the 1709 * options and the way we show them in _ext4_show_options(). To 1710 * keep the changes to a minimum, let's keep the negative options 1711 * separate for now. 1712 */ 1713 static const struct fs_parameter_spec ext4_param_specs[] = { 1714 fsparam_flag ("bsddf", Opt_bsd_df), 1715 fsparam_flag ("minixdf", Opt_minix_df), 1716 fsparam_flag ("grpid", Opt_grpid), 1717 fsparam_flag ("bsdgroups", Opt_grpid), 1718 fsparam_flag ("nogrpid", Opt_nogrpid), 1719 fsparam_flag ("sysvgroups", Opt_nogrpid), 1720 fsparam_gid ("resgid", Opt_resgid), 1721 fsparam_uid ("resuid", Opt_resuid), 1722 fsparam_u32 ("sb", Opt_sb), 1723 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1724 fsparam_flag ("nouid32", Opt_nouid32), 1725 fsparam_flag ("debug", Opt_debug), 1726 fsparam_flag ("oldalloc", Opt_removed), 1727 fsparam_flag ("orlov", Opt_removed), 1728 fsparam_flag ("user_xattr", Opt_user_xattr), 1729 fsparam_flag ("acl", Opt_acl), 1730 fsparam_flag ("norecovery", Opt_noload), 1731 fsparam_flag ("noload", Opt_noload), 1732 fsparam_flag ("bh", Opt_removed), 1733 fsparam_flag ("nobh", Opt_removed), 1734 fsparam_u32 ("commit", Opt_commit), 1735 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1736 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1737 fsparam_u32 ("journal_dev", Opt_journal_dev), 1738 fsparam_bdev ("journal_path", Opt_journal_path), 1739 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1740 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1741 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1742 fsparam_flag ("abort", Opt_abort), 1743 fsparam_enum ("data", Opt_data, ext4_param_data), 1744 fsparam_enum ("data_err", Opt_data_err, 1745 ext4_param_data_err), 1746 fsparam_string_empty 1747 ("usrjquota", Opt_usrjquota), 1748 fsparam_string_empty 1749 ("grpjquota", Opt_grpjquota), 1750 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1751 fsparam_flag ("grpquota", Opt_grpquota), 1752 fsparam_flag ("quota", Opt_quota), 1753 fsparam_flag ("noquota", Opt_noquota), 1754 fsparam_flag ("usrquota", Opt_usrquota), 1755 fsparam_flag ("prjquota", Opt_prjquota), 1756 fsparam_flag ("barrier", Opt_barrier), 1757 fsparam_u32 ("barrier", Opt_barrier), 1758 fsparam_flag ("nobarrier", Opt_nobarrier), 1759 fsparam_flag ("i_version", Opt_removed), 1760 fsparam_flag ("dax", Opt_dax), 1761 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1762 fsparam_u32 ("stripe", Opt_stripe), 1763 fsparam_flag ("delalloc", Opt_delalloc), 1764 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1765 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1766 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1767 fsparam_u32 ("debug_want_extra_isize", 1768 Opt_debug_want_extra_isize), 1769 fsparam_flag ("mblk_io_submit", Opt_removed), 1770 fsparam_flag ("nomblk_io_submit", Opt_removed), 1771 fsparam_flag ("block_validity", Opt_block_validity), 1772 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1773 fsparam_u32 ("inode_readahead_blks", 1774 Opt_inode_readahead_blks), 1775 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1776 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1777 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1778 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1779 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1780 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1781 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1782 fsparam_flag ("discard", Opt_discard), 1783 fsparam_flag ("nodiscard", Opt_nodiscard), 1784 fsparam_u32 ("init_itable", Opt_init_itable), 1785 fsparam_flag ("init_itable", Opt_init_itable), 1786 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1787 #ifdef CONFIG_EXT4_DEBUG 1788 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1789 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1790 #endif 1791 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1792 fsparam_flag ("test_dummy_encryption", 1793 Opt_test_dummy_encryption), 1794 fsparam_string ("test_dummy_encryption", 1795 Opt_test_dummy_encryption), 1796 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1797 fsparam_flag ("nombcache", Opt_nombcache), 1798 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1799 fsparam_flag ("prefetch_block_bitmaps", 1800 Opt_removed), 1801 fsparam_flag ("no_prefetch_block_bitmaps", 1802 Opt_no_prefetch_block_bitmaps), 1803 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1804 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1805 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1806 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1807 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1808 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1809 {} 1810 }; 1811 1812 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1813 1814 #define MOPT_SET 0x0001 1815 #define MOPT_CLEAR 0x0002 1816 #define MOPT_NOSUPPORT 0x0004 1817 #define MOPT_EXPLICIT 0x0008 1818 #ifdef CONFIG_QUOTA 1819 #define MOPT_Q 0 1820 #define MOPT_QFMT 0x0010 1821 #else 1822 #define MOPT_Q MOPT_NOSUPPORT 1823 #define MOPT_QFMT MOPT_NOSUPPORT 1824 #endif 1825 #define MOPT_NO_EXT2 0x0020 1826 #define MOPT_NO_EXT3 0x0040 1827 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1828 #define MOPT_SKIP 0x0080 1829 #define MOPT_2 0x0100 1830 1831 static const struct mount_opts { 1832 int token; 1833 int mount_opt; 1834 int flags; 1835 } ext4_mount_opts[] = { 1836 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1837 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1838 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1839 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1840 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1841 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1842 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1843 MOPT_EXT4_ONLY | MOPT_SET}, 1844 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1845 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1846 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1847 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1848 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1849 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1850 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1851 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1852 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1853 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1854 {Opt_commit, 0, MOPT_NO_EXT2}, 1855 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1856 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1857 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1858 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1859 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1860 EXT4_MOUNT_JOURNAL_CHECKSUM), 1861 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1862 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1863 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1864 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1865 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1866 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1867 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1868 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1869 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1870 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1871 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1872 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1873 {Opt_data, 0, MOPT_NO_EXT2}, 1874 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1875 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1876 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1877 #else 1878 {Opt_acl, 0, MOPT_NOSUPPORT}, 1879 #endif 1880 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1881 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1882 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1883 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1884 MOPT_SET | MOPT_Q}, 1885 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1886 MOPT_SET | MOPT_Q}, 1887 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1888 MOPT_SET | MOPT_Q}, 1889 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1890 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1891 MOPT_CLEAR | MOPT_Q}, 1892 {Opt_usrjquota, 0, MOPT_Q}, 1893 {Opt_grpjquota, 0, MOPT_Q}, 1894 {Opt_jqfmt, 0, MOPT_QFMT}, 1895 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1896 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1897 MOPT_SET}, 1898 #ifdef CONFIG_EXT4_DEBUG 1899 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1900 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1901 #endif 1902 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2}, 1903 {Opt_err, 0, 0} 1904 }; 1905 1906 #if IS_ENABLED(CONFIG_UNICODE) 1907 static const struct ext4_sb_encodings { 1908 __u16 magic; 1909 char *name; 1910 unsigned int version; 1911 } ext4_sb_encoding_map[] = { 1912 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1913 }; 1914 1915 static const struct ext4_sb_encodings * 1916 ext4_sb_read_encoding(const struct ext4_super_block *es) 1917 { 1918 __u16 magic = le16_to_cpu(es->s_encoding); 1919 int i; 1920 1921 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1922 if (magic == ext4_sb_encoding_map[i].magic) 1923 return &ext4_sb_encoding_map[i]; 1924 1925 return NULL; 1926 } 1927 #endif 1928 1929 #define EXT4_SPEC_JQUOTA (1 << 0) 1930 #define EXT4_SPEC_JQFMT (1 << 1) 1931 #define EXT4_SPEC_DATAJ (1 << 2) 1932 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1933 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1934 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1935 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 1936 #define EXT4_SPEC_s_max_batch_time (1 << 8) 1937 #define EXT4_SPEC_s_min_batch_time (1 << 9) 1938 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 1939 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 1940 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 1941 #define EXT4_SPEC_s_stripe (1 << 13) 1942 #define EXT4_SPEC_s_resuid (1 << 14) 1943 #define EXT4_SPEC_s_resgid (1 << 15) 1944 #define EXT4_SPEC_s_commit_interval (1 << 16) 1945 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 1946 #define EXT4_SPEC_s_sb_block (1 << 18) 1947 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 1948 1949 struct ext4_fs_context { 1950 char *s_qf_names[EXT4_MAXQUOTAS]; 1951 struct fscrypt_dummy_policy dummy_enc_policy; 1952 int s_jquota_fmt; /* Format of quota to use */ 1953 #ifdef CONFIG_EXT4_DEBUG 1954 int s_fc_debug_max_replay; 1955 #endif 1956 unsigned short qname_spec; 1957 unsigned long vals_s_flags; /* Bits to set in s_flags */ 1958 unsigned long mask_s_flags; /* Bits changed in s_flags */ 1959 unsigned long journal_devnum; 1960 unsigned long s_commit_interval; 1961 unsigned long s_stripe; 1962 unsigned int s_inode_readahead_blks; 1963 unsigned int s_want_extra_isize; 1964 unsigned int s_li_wait_mult; 1965 unsigned int s_max_dir_size_kb; 1966 unsigned int journal_ioprio; 1967 unsigned int vals_s_mount_opt; 1968 unsigned int mask_s_mount_opt; 1969 unsigned int vals_s_mount_opt2; 1970 unsigned int mask_s_mount_opt2; 1971 unsigned int opt_flags; /* MOPT flags */ 1972 unsigned int spec; 1973 u32 s_max_batch_time; 1974 u32 s_min_batch_time; 1975 kuid_t s_resuid; 1976 kgid_t s_resgid; 1977 ext4_fsblk_t s_sb_block; 1978 }; 1979 1980 static void ext4_fc_free(struct fs_context *fc) 1981 { 1982 struct ext4_fs_context *ctx = fc->fs_private; 1983 int i; 1984 1985 if (!ctx) 1986 return; 1987 1988 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1989 kfree(ctx->s_qf_names[i]); 1990 1991 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 1992 kfree(ctx); 1993 } 1994 1995 int ext4_init_fs_context(struct fs_context *fc) 1996 { 1997 struct ext4_fs_context *ctx; 1998 1999 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2000 if (!ctx) 2001 return -ENOMEM; 2002 2003 fc->fs_private = ctx; 2004 fc->ops = &ext4_context_ops; 2005 2006 return 0; 2007 } 2008 2009 #ifdef CONFIG_QUOTA 2010 /* 2011 * Note the name of the specified quota file. 2012 */ 2013 static int note_qf_name(struct fs_context *fc, int qtype, 2014 struct fs_parameter *param) 2015 { 2016 struct ext4_fs_context *ctx = fc->fs_private; 2017 char *qname; 2018 2019 if (param->size < 1) { 2020 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 2021 return -EINVAL; 2022 } 2023 if (strchr(param->string, '/')) { 2024 ext4_msg(NULL, KERN_ERR, 2025 "quotafile must be on filesystem root"); 2026 return -EINVAL; 2027 } 2028 if (ctx->s_qf_names[qtype]) { 2029 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 2030 ext4_msg(NULL, KERN_ERR, 2031 "%s quota file already specified", 2032 QTYPE2NAME(qtype)); 2033 return -EINVAL; 2034 } 2035 return 0; 2036 } 2037 2038 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 2039 if (!qname) { 2040 ext4_msg(NULL, KERN_ERR, 2041 "Not enough memory for storing quotafile name"); 2042 return -ENOMEM; 2043 } 2044 ctx->s_qf_names[qtype] = qname; 2045 ctx->qname_spec |= 1 << qtype; 2046 ctx->spec |= EXT4_SPEC_JQUOTA; 2047 return 0; 2048 } 2049 2050 /* 2051 * Clear the name of the specified quota file. 2052 */ 2053 static int unnote_qf_name(struct fs_context *fc, int qtype) 2054 { 2055 struct ext4_fs_context *ctx = fc->fs_private; 2056 2057 kfree(ctx->s_qf_names[qtype]); 2058 2059 ctx->s_qf_names[qtype] = NULL; 2060 ctx->qname_spec |= 1 << qtype; 2061 ctx->spec |= EXT4_SPEC_JQUOTA; 2062 return 0; 2063 } 2064 #endif 2065 2066 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2067 struct ext4_fs_context *ctx) 2068 { 2069 int err; 2070 2071 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2072 ext4_msg(NULL, KERN_WARNING, 2073 "test_dummy_encryption option not supported"); 2074 return -EINVAL; 2075 } 2076 err = fscrypt_parse_test_dummy_encryption(param, 2077 &ctx->dummy_enc_policy); 2078 if (err == -EINVAL) { 2079 ext4_msg(NULL, KERN_WARNING, 2080 "Value of option \"%s\" is unrecognized", param->key); 2081 } else if (err == -EEXIST) { 2082 ext4_msg(NULL, KERN_WARNING, 2083 "Conflicting test_dummy_encryption options"); 2084 return -EINVAL; 2085 } 2086 return err; 2087 } 2088 2089 #define EXT4_SET_CTX(name) \ 2090 static inline __maybe_unused \ 2091 void ctx_set_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2092 { \ 2093 ctx->mask_s_##name |= flag; \ 2094 ctx->vals_s_##name |= flag; \ 2095 } 2096 2097 #define EXT4_CLEAR_CTX(name) \ 2098 static inline __maybe_unused \ 2099 void ctx_clear_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2100 { \ 2101 ctx->mask_s_##name |= flag; \ 2102 ctx->vals_s_##name &= ~flag; \ 2103 } 2104 2105 #define EXT4_TEST_CTX(name) \ 2106 static inline unsigned long \ 2107 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2108 { \ 2109 return (ctx->vals_s_##name & flag); \ 2110 } 2111 2112 EXT4_SET_CTX(flags); /* set only */ 2113 EXT4_SET_CTX(mount_opt); 2114 EXT4_CLEAR_CTX(mount_opt); 2115 EXT4_TEST_CTX(mount_opt); 2116 EXT4_SET_CTX(mount_opt2); 2117 EXT4_CLEAR_CTX(mount_opt2); 2118 EXT4_TEST_CTX(mount_opt2); 2119 2120 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2121 { 2122 struct ext4_fs_context *ctx = fc->fs_private; 2123 struct fs_parse_result result; 2124 const struct mount_opts *m; 2125 int is_remount; 2126 int token; 2127 2128 token = fs_parse(fc, ext4_param_specs, param, &result); 2129 if (token < 0) 2130 return token; 2131 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2132 2133 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2134 if (token == m->token) 2135 break; 2136 2137 ctx->opt_flags |= m->flags; 2138 2139 if (m->flags & MOPT_EXPLICIT) { 2140 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2141 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2142 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2143 ctx_set_mount_opt2(ctx, 2144 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2145 } else 2146 return -EINVAL; 2147 } 2148 2149 if (m->flags & MOPT_NOSUPPORT) { 2150 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2151 param->key); 2152 return 0; 2153 } 2154 2155 switch (token) { 2156 #ifdef CONFIG_QUOTA 2157 case Opt_usrjquota: 2158 if (!*param->string) 2159 return unnote_qf_name(fc, USRQUOTA); 2160 else 2161 return note_qf_name(fc, USRQUOTA, param); 2162 case Opt_grpjquota: 2163 if (!*param->string) 2164 return unnote_qf_name(fc, GRPQUOTA); 2165 else 2166 return note_qf_name(fc, GRPQUOTA, param); 2167 #endif 2168 case Opt_sb: 2169 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2170 ext4_msg(NULL, KERN_WARNING, 2171 "Ignoring %s option on remount", param->key); 2172 } else { 2173 ctx->s_sb_block = result.uint_32; 2174 ctx->spec |= EXT4_SPEC_s_sb_block; 2175 } 2176 return 0; 2177 case Opt_removed: 2178 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2179 param->key); 2180 return 0; 2181 case Opt_inlinecrypt: 2182 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2183 ctx_set_flags(ctx, SB_INLINECRYPT); 2184 #else 2185 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2186 #endif 2187 return 0; 2188 case Opt_errors: 2189 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2190 ctx_set_mount_opt(ctx, result.uint_32); 2191 return 0; 2192 #ifdef CONFIG_QUOTA 2193 case Opt_jqfmt: 2194 ctx->s_jquota_fmt = result.uint_32; 2195 ctx->spec |= EXT4_SPEC_JQFMT; 2196 return 0; 2197 #endif 2198 case Opt_data: 2199 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2200 ctx_set_mount_opt(ctx, result.uint_32); 2201 ctx->spec |= EXT4_SPEC_DATAJ; 2202 return 0; 2203 case Opt_commit: 2204 if (result.uint_32 == 0) 2205 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE; 2206 else if (result.uint_32 > INT_MAX / HZ) { 2207 ext4_msg(NULL, KERN_ERR, 2208 "Invalid commit interval %d, " 2209 "must be smaller than %d", 2210 result.uint_32, INT_MAX / HZ); 2211 return -EINVAL; 2212 } 2213 ctx->s_commit_interval = HZ * result.uint_32; 2214 ctx->spec |= EXT4_SPEC_s_commit_interval; 2215 return 0; 2216 case Opt_debug_want_extra_isize: 2217 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2218 ext4_msg(NULL, KERN_ERR, 2219 "Invalid want_extra_isize %d", result.uint_32); 2220 return -EINVAL; 2221 } 2222 ctx->s_want_extra_isize = result.uint_32; 2223 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2224 return 0; 2225 case Opt_max_batch_time: 2226 ctx->s_max_batch_time = result.uint_32; 2227 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2228 return 0; 2229 case Opt_min_batch_time: 2230 ctx->s_min_batch_time = result.uint_32; 2231 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2232 return 0; 2233 case Opt_inode_readahead_blks: 2234 if (result.uint_32 && 2235 (result.uint_32 > (1 << 30) || 2236 !is_power_of_2(result.uint_32))) { 2237 ext4_msg(NULL, KERN_ERR, 2238 "EXT4-fs: inode_readahead_blks must be " 2239 "0 or a power of 2 smaller than 2^31"); 2240 return -EINVAL; 2241 } 2242 ctx->s_inode_readahead_blks = result.uint_32; 2243 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2244 return 0; 2245 case Opt_init_itable: 2246 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2247 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2248 if (param->type == fs_value_is_string) 2249 ctx->s_li_wait_mult = result.uint_32; 2250 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2251 return 0; 2252 case Opt_max_dir_size_kb: 2253 ctx->s_max_dir_size_kb = result.uint_32; 2254 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2255 return 0; 2256 #ifdef CONFIG_EXT4_DEBUG 2257 case Opt_fc_debug_max_replay: 2258 ctx->s_fc_debug_max_replay = result.uint_32; 2259 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2260 return 0; 2261 #endif 2262 case Opt_stripe: 2263 ctx->s_stripe = result.uint_32; 2264 ctx->spec |= EXT4_SPEC_s_stripe; 2265 return 0; 2266 case Opt_resuid: 2267 ctx->s_resuid = result.uid; 2268 ctx->spec |= EXT4_SPEC_s_resuid; 2269 return 0; 2270 case Opt_resgid: 2271 ctx->s_resgid = result.gid; 2272 ctx->spec |= EXT4_SPEC_s_resgid; 2273 return 0; 2274 case Opt_journal_dev: 2275 if (is_remount) { 2276 ext4_msg(NULL, KERN_ERR, 2277 "Cannot specify journal on remount"); 2278 return -EINVAL; 2279 } 2280 ctx->journal_devnum = result.uint_32; 2281 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2282 return 0; 2283 case Opt_journal_path: 2284 { 2285 struct inode *journal_inode; 2286 struct path path; 2287 int error; 2288 2289 if (is_remount) { 2290 ext4_msg(NULL, KERN_ERR, 2291 "Cannot specify journal on remount"); 2292 return -EINVAL; 2293 } 2294 2295 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path); 2296 if (error) { 2297 ext4_msg(NULL, KERN_ERR, "error: could not find " 2298 "journal device path"); 2299 return -EINVAL; 2300 } 2301 2302 journal_inode = d_inode(path.dentry); 2303 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2304 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2305 path_put(&path); 2306 return 0; 2307 } 2308 case Opt_journal_ioprio: 2309 if (result.uint_32 > 7) { 2310 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2311 " (must be 0-7)"); 2312 return -EINVAL; 2313 } 2314 ctx->journal_ioprio = 2315 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2316 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2317 return 0; 2318 case Opt_test_dummy_encryption: 2319 return ext4_parse_test_dummy_encryption(param, ctx); 2320 case Opt_dax: 2321 case Opt_dax_type: 2322 #ifdef CONFIG_FS_DAX 2323 { 2324 int type = (token == Opt_dax) ? 2325 Opt_dax : result.uint_32; 2326 2327 switch (type) { 2328 case Opt_dax: 2329 case Opt_dax_always: 2330 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2331 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2332 break; 2333 case Opt_dax_never: 2334 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2335 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2336 break; 2337 case Opt_dax_inode: 2338 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2339 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2340 /* Strictly for printing options */ 2341 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2342 break; 2343 } 2344 return 0; 2345 } 2346 #else 2347 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2348 return -EINVAL; 2349 #endif 2350 case Opt_data_err: 2351 if (result.uint_32 == Opt_data_err_abort) 2352 ctx_set_mount_opt(ctx, m->mount_opt); 2353 else if (result.uint_32 == Opt_data_err_ignore) 2354 ctx_clear_mount_opt(ctx, m->mount_opt); 2355 return 0; 2356 case Opt_mb_optimize_scan: 2357 if (result.int_32 == 1) { 2358 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2359 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2360 } else if (result.int_32 == 0) { 2361 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2362 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2363 } else { 2364 ext4_msg(NULL, KERN_WARNING, 2365 "mb_optimize_scan should be set to 0 or 1."); 2366 return -EINVAL; 2367 } 2368 return 0; 2369 } 2370 2371 /* 2372 * At this point we should only be getting options requiring MOPT_SET, 2373 * or MOPT_CLEAR. Anything else is a bug 2374 */ 2375 if (m->token == Opt_err) { 2376 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2377 param->key); 2378 WARN_ON(1); 2379 return -EINVAL; 2380 } 2381 2382 else { 2383 unsigned int set = 0; 2384 2385 if ((param->type == fs_value_is_flag) || 2386 result.uint_32 > 0) 2387 set = 1; 2388 2389 if (m->flags & MOPT_CLEAR) 2390 set = !set; 2391 else if (unlikely(!(m->flags & MOPT_SET))) { 2392 ext4_msg(NULL, KERN_WARNING, 2393 "buggy handling of option %s", 2394 param->key); 2395 WARN_ON(1); 2396 return -EINVAL; 2397 } 2398 if (m->flags & MOPT_2) { 2399 if (set != 0) 2400 ctx_set_mount_opt2(ctx, m->mount_opt); 2401 else 2402 ctx_clear_mount_opt2(ctx, m->mount_opt); 2403 } else { 2404 if (set != 0) 2405 ctx_set_mount_opt(ctx, m->mount_opt); 2406 else 2407 ctx_clear_mount_opt(ctx, m->mount_opt); 2408 } 2409 } 2410 2411 return 0; 2412 } 2413 2414 static int parse_options(struct fs_context *fc, char *options) 2415 { 2416 struct fs_parameter param; 2417 int ret; 2418 char *key; 2419 2420 if (!options) 2421 return 0; 2422 2423 while ((key = strsep(&options, ",")) != NULL) { 2424 if (*key) { 2425 size_t v_len = 0; 2426 char *value = strchr(key, '='); 2427 2428 param.type = fs_value_is_flag; 2429 param.string = NULL; 2430 2431 if (value) { 2432 if (value == key) 2433 continue; 2434 2435 *value++ = 0; 2436 v_len = strlen(value); 2437 param.string = kmemdup_nul(value, v_len, 2438 GFP_KERNEL); 2439 if (!param.string) 2440 return -ENOMEM; 2441 param.type = fs_value_is_string; 2442 } 2443 2444 param.key = key; 2445 param.size = v_len; 2446 2447 ret = ext4_parse_param(fc, ¶m); 2448 kfree(param.string); 2449 if (ret < 0) 2450 return ret; 2451 } 2452 } 2453 2454 ret = ext4_validate_options(fc); 2455 if (ret < 0) 2456 return ret; 2457 2458 return 0; 2459 } 2460 2461 static int parse_apply_sb_mount_options(struct super_block *sb, 2462 struct ext4_fs_context *m_ctx) 2463 { 2464 struct ext4_sb_info *sbi = EXT4_SB(sb); 2465 char *s_mount_opts = NULL; 2466 struct ext4_fs_context *s_ctx = NULL; 2467 struct fs_context *fc = NULL; 2468 int ret = -ENOMEM; 2469 2470 if (!sbi->s_es->s_mount_opts[0]) 2471 return 0; 2472 2473 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2474 sizeof(sbi->s_es->s_mount_opts), 2475 GFP_KERNEL); 2476 if (!s_mount_opts) 2477 return ret; 2478 2479 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2480 if (!fc) 2481 goto out_free; 2482 2483 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2484 if (!s_ctx) 2485 goto out_free; 2486 2487 fc->fs_private = s_ctx; 2488 fc->s_fs_info = sbi; 2489 2490 ret = parse_options(fc, s_mount_opts); 2491 if (ret < 0) 2492 goto parse_failed; 2493 2494 ret = ext4_check_opt_consistency(fc, sb); 2495 if (ret < 0) { 2496 parse_failed: 2497 ext4_msg(sb, KERN_WARNING, 2498 "failed to parse options in superblock: %s", 2499 s_mount_opts); 2500 ret = 0; 2501 goto out_free; 2502 } 2503 2504 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2505 m_ctx->journal_devnum = s_ctx->journal_devnum; 2506 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2507 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2508 2509 ext4_apply_options(fc, sb); 2510 ret = 0; 2511 2512 out_free: 2513 if (fc) { 2514 ext4_fc_free(fc); 2515 kfree(fc); 2516 } 2517 kfree(s_mount_opts); 2518 return ret; 2519 } 2520 2521 static void ext4_apply_quota_options(struct fs_context *fc, 2522 struct super_block *sb) 2523 { 2524 #ifdef CONFIG_QUOTA 2525 bool quota_feature = ext4_has_feature_quota(sb); 2526 struct ext4_fs_context *ctx = fc->fs_private; 2527 struct ext4_sb_info *sbi = EXT4_SB(sb); 2528 char *qname; 2529 int i; 2530 2531 if (quota_feature) 2532 return; 2533 2534 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2535 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2536 if (!(ctx->qname_spec & (1 << i))) 2537 continue; 2538 2539 qname = ctx->s_qf_names[i]; /* May be NULL */ 2540 if (qname) 2541 set_opt(sb, QUOTA); 2542 ctx->s_qf_names[i] = NULL; 2543 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2544 lockdep_is_held(&sb->s_umount)); 2545 if (qname) 2546 kfree_rcu_mightsleep(qname); 2547 } 2548 } 2549 2550 if (ctx->spec & EXT4_SPEC_JQFMT) 2551 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2552 #endif 2553 } 2554 2555 /* 2556 * Check quota settings consistency. 2557 */ 2558 static int ext4_check_quota_consistency(struct fs_context *fc, 2559 struct super_block *sb) 2560 { 2561 #ifdef CONFIG_QUOTA 2562 struct ext4_fs_context *ctx = fc->fs_private; 2563 struct ext4_sb_info *sbi = EXT4_SB(sb); 2564 bool quota_feature = ext4_has_feature_quota(sb); 2565 bool quota_loaded = sb_any_quota_loaded(sb); 2566 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2567 int quota_flags, i; 2568 2569 /* 2570 * We do the test below only for project quotas. 'usrquota' and 2571 * 'grpquota' mount options are allowed even without quota feature 2572 * to support legacy quotas in quota files. 2573 */ 2574 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2575 !ext4_has_feature_project(sb)) { 2576 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2577 "Cannot enable project quota enforcement."); 2578 return -EINVAL; 2579 } 2580 2581 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2582 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2583 if (quota_loaded && 2584 ctx->mask_s_mount_opt & quota_flags && 2585 !ctx_test_mount_opt(ctx, quota_flags)) 2586 goto err_quota_change; 2587 2588 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2589 2590 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2591 if (!(ctx->qname_spec & (1 << i))) 2592 continue; 2593 2594 if (quota_loaded && 2595 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2596 goto err_jquota_change; 2597 2598 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2599 strcmp(get_qf_name(sb, sbi, i), 2600 ctx->s_qf_names[i]) != 0) 2601 goto err_jquota_specified; 2602 } 2603 2604 if (quota_feature) { 2605 ext4_msg(NULL, KERN_INFO, 2606 "Journaled quota options ignored when " 2607 "QUOTA feature is enabled"); 2608 return 0; 2609 } 2610 } 2611 2612 if (ctx->spec & EXT4_SPEC_JQFMT) { 2613 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2614 goto err_jquota_change; 2615 if (quota_feature) { 2616 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2617 "ignored when QUOTA feature is enabled"); 2618 return 0; 2619 } 2620 } 2621 2622 /* Make sure we don't mix old and new quota format */ 2623 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2624 ctx->s_qf_names[USRQUOTA]); 2625 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2626 ctx->s_qf_names[GRPQUOTA]); 2627 2628 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2629 test_opt(sb, USRQUOTA)); 2630 2631 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2632 test_opt(sb, GRPQUOTA)); 2633 2634 if (usr_qf_name) { 2635 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2636 usrquota = false; 2637 } 2638 if (grp_qf_name) { 2639 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2640 grpquota = false; 2641 } 2642 2643 if (usr_qf_name || grp_qf_name) { 2644 if (usrquota || grpquota) { 2645 ext4_msg(NULL, KERN_ERR, "old and new quota " 2646 "format mixing"); 2647 return -EINVAL; 2648 } 2649 2650 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2651 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2652 "not specified"); 2653 return -EINVAL; 2654 } 2655 } 2656 2657 return 0; 2658 2659 err_quota_change: 2660 ext4_msg(NULL, KERN_ERR, 2661 "Cannot change quota options when quota turned on"); 2662 return -EINVAL; 2663 err_jquota_change: 2664 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2665 "options when quota turned on"); 2666 return -EINVAL; 2667 err_jquota_specified: 2668 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2669 QTYPE2NAME(i)); 2670 return -EINVAL; 2671 #else 2672 return 0; 2673 #endif 2674 } 2675 2676 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2677 struct super_block *sb) 2678 { 2679 const struct ext4_fs_context *ctx = fc->fs_private; 2680 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2681 2682 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2683 return 0; 2684 2685 if (!ext4_has_feature_encrypt(sb)) { 2686 ext4_msg(NULL, KERN_WARNING, 2687 "test_dummy_encryption requires encrypt feature"); 2688 return -EINVAL; 2689 } 2690 /* 2691 * This mount option is just for testing, and it's not worthwhile to 2692 * implement the extra complexity (e.g. RCU protection) that would be 2693 * needed to allow it to be set or changed during remount. We do allow 2694 * it to be specified during remount, but only if there is no change. 2695 */ 2696 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2697 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2698 &ctx->dummy_enc_policy)) 2699 return 0; 2700 ext4_msg(NULL, KERN_WARNING, 2701 "Can't set or change test_dummy_encryption on remount"); 2702 return -EINVAL; 2703 } 2704 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2705 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2706 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2707 &ctx->dummy_enc_policy)) 2708 return 0; 2709 ext4_msg(NULL, KERN_WARNING, 2710 "Conflicting test_dummy_encryption options"); 2711 return -EINVAL; 2712 } 2713 return 0; 2714 } 2715 2716 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2717 struct super_block *sb) 2718 { 2719 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2720 /* if already set, it was already verified to be the same */ 2721 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2722 return; 2723 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2724 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2725 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2726 } 2727 2728 static int ext4_check_opt_consistency(struct fs_context *fc, 2729 struct super_block *sb) 2730 { 2731 struct ext4_fs_context *ctx = fc->fs_private; 2732 struct ext4_sb_info *sbi = fc->s_fs_info; 2733 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2734 int err; 2735 2736 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2737 ext4_msg(NULL, KERN_ERR, 2738 "Mount option(s) incompatible with ext2"); 2739 return -EINVAL; 2740 } 2741 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2742 ext4_msg(NULL, KERN_ERR, 2743 "Mount option(s) incompatible with ext3"); 2744 return -EINVAL; 2745 } 2746 2747 if (ctx->s_want_extra_isize > 2748 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2749 ext4_msg(NULL, KERN_ERR, 2750 "Invalid want_extra_isize %d", 2751 ctx->s_want_extra_isize); 2752 return -EINVAL; 2753 } 2754 2755 err = ext4_check_test_dummy_encryption(fc, sb); 2756 if (err) 2757 return err; 2758 2759 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2760 if (!sbi->s_journal) { 2761 ext4_msg(NULL, KERN_WARNING, 2762 "Remounting file system with no journal " 2763 "so ignoring journalled data option"); 2764 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2765 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2766 test_opt(sb, DATA_FLAGS)) { 2767 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2768 "on remount"); 2769 return -EINVAL; 2770 } 2771 } 2772 2773 if (is_remount) { 2774 if (!sbi->s_journal && 2775 ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT)) { 2776 ext4_msg(NULL, KERN_WARNING, 2777 "Remounting fs w/o journal so ignoring data_err option"); 2778 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_ERR_ABORT); 2779 } 2780 2781 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2782 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2783 ext4_msg(NULL, KERN_ERR, "can't mount with " 2784 "both data=journal and dax"); 2785 return -EINVAL; 2786 } 2787 2788 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2789 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2790 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2791 fail_dax_change_remount: 2792 ext4_msg(NULL, KERN_ERR, "can't change " 2793 "dax mount option while remounting"); 2794 return -EINVAL; 2795 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2796 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2797 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2798 goto fail_dax_change_remount; 2799 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2800 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2801 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2802 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2803 goto fail_dax_change_remount; 2804 } 2805 } 2806 2807 return ext4_check_quota_consistency(fc, sb); 2808 } 2809 2810 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2811 { 2812 struct ext4_fs_context *ctx = fc->fs_private; 2813 struct ext4_sb_info *sbi = fc->s_fs_info; 2814 2815 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2816 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2817 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2818 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2819 sb->s_flags &= ~ctx->mask_s_flags; 2820 sb->s_flags |= ctx->vals_s_flags; 2821 2822 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2823 APPLY(s_commit_interval); 2824 APPLY(s_stripe); 2825 APPLY(s_max_batch_time); 2826 APPLY(s_min_batch_time); 2827 APPLY(s_want_extra_isize); 2828 APPLY(s_inode_readahead_blks); 2829 APPLY(s_max_dir_size_kb); 2830 APPLY(s_li_wait_mult); 2831 APPLY(s_resgid); 2832 APPLY(s_resuid); 2833 2834 #ifdef CONFIG_EXT4_DEBUG 2835 APPLY(s_fc_debug_max_replay); 2836 #endif 2837 2838 ext4_apply_quota_options(fc, sb); 2839 ext4_apply_test_dummy_encryption(ctx, sb); 2840 } 2841 2842 2843 static int ext4_validate_options(struct fs_context *fc) 2844 { 2845 #ifdef CONFIG_QUOTA 2846 struct ext4_fs_context *ctx = fc->fs_private; 2847 char *usr_qf_name, *grp_qf_name; 2848 2849 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2850 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2851 2852 if (usr_qf_name || grp_qf_name) { 2853 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2854 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2855 2856 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2857 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2858 2859 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2860 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2861 ext4_msg(NULL, KERN_ERR, "old and new quota " 2862 "format mixing"); 2863 return -EINVAL; 2864 } 2865 } 2866 #endif 2867 return 1; 2868 } 2869 2870 static inline void ext4_show_quota_options(struct seq_file *seq, 2871 struct super_block *sb) 2872 { 2873 #if defined(CONFIG_QUOTA) 2874 struct ext4_sb_info *sbi = EXT4_SB(sb); 2875 char *usr_qf_name, *grp_qf_name; 2876 2877 if (sbi->s_jquota_fmt) { 2878 char *fmtname = ""; 2879 2880 switch (sbi->s_jquota_fmt) { 2881 case QFMT_VFS_OLD: 2882 fmtname = "vfsold"; 2883 break; 2884 case QFMT_VFS_V0: 2885 fmtname = "vfsv0"; 2886 break; 2887 case QFMT_VFS_V1: 2888 fmtname = "vfsv1"; 2889 break; 2890 } 2891 seq_printf(seq, ",jqfmt=%s", fmtname); 2892 } 2893 2894 rcu_read_lock(); 2895 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2896 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2897 if (usr_qf_name) 2898 seq_show_option(seq, "usrjquota", usr_qf_name); 2899 if (grp_qf_name) 2900 seq_show_option(seq, "grpjquota", grp_qf_name); 2901 rcu_read_unlock(); 2902 #endif 2903 } 2904 2905 static const char *token2str(int token) 2906 { 2907 const struct fs_parameter_spec *spec; 2908 2909 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2910 if (spec->opt == token && !spec->type) 2911 break; 2912 return spec->name; 2913 } 2914 2915 /* 2916 * Show an option if 2917 * - it's set to a non-default value OR 2918 * - if the per-sb default is different from the global default 2919 */ 2920 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2921 int nodefs) 2922 { 2923 struct ext4_sb_info *sbi = EXT4_SB(sb); 2924 struct ext4_super_block *es = sbi->s_es; 2925 int def_errors; 2926 const struct mount_opts *m; 2927 char sep = nodefs ? '\n' : ','; 2928 2929 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2930 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2931 2932 if (sbi->s_sb_block != 1) 2933 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2934 2935 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2936 int want_set = m->flags & MOPT_SET; 2937 int opt_2 = m->flags & MOPT_2; 2938 unsigned int mount_opt, def_mount_opt; 2939 2940 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2941 m->flags & MOPT_SKIP) 2942 continue; 2943 2944 if (opt_2) { 2945 mount_opt = sbi->s_mount_opt2; 2946 def_mount_opt = sbi->s_def_mount_opt2; 2947 } else { 2948 mount_opt = sbi->s_mount_opt; 2949 def_mount_opt = sbi->s_def_mount_opt; 2950 } 2951 /* skip if same as the default */ 2952 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt))) 2953 continue; 2954 /* select Opt_noFoo vs Opt_Foo */ 2955 if ((want_set && 2956 (mount_opt & m->mount_opt) != m->mount_opt) || 2957 (!want_set && (mount_opt & m->mount_opt))) 2958 continue; 2959 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2960 } 2961 2962 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2963 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2964 SEQ_OPTS_PRINT("resuid=%u", 2965 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2966 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2967 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2968 SEQ_OPTS_PRINT("resgid=%u", 2969 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2970 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2971 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2972 SEQ_OPTS_PUTS("errors=remount-ro"); 2973 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2974 SEQ_OPTS_PUTS("errors=continue"); 2975 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2976 SEQ_OPTS_PUTS("errors=panic"); 2977 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2978 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2979 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2980 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2981 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2982 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2983 if (nodefs || sbi->s_stripe) 2984 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2985 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2986 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 2987 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2988 SEQ_OPTS_PUTS("data=journal"); 2989 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2990 SEQ_OPTS_PUTS("data=ordered"); 2991 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2992 SEQ_OPTS_PUTS("data=writeback"); 2993 } 2994 if (nodefs || 2995 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2996 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2997 sbi->s_inode_readahead_blks); 2998 2999 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 3000 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 3001 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 3002 if (nodefs || sbi->s_max_dir_size_kb) 3003 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 3004 if (test_opt(sb, DATA_ERR_ABORT)) 3005 SEQ_OPTS_PUTS("data_err=abort"); 3006 3007 fscrypt_show_test_dummy_encryption(seq, sep, sb); 3008 3009 if (sb->s_flags & SB_INLINECRYPT) 3010 SEQ_OPTS_PUTS("inlinecrypt"); 3011 3012 if (test_opt(sb, DAX_ALWAYS)) { 3013 if (IS_EXT2_SB(sb)) 3014 SEQ_OPTS_PUTS("dax"); 3015 else 3016 SEQ_OPTS_PUTS("dax=always"); 3017 } else if (test_opt2(sb, DAX_NEVER)) { 3018 SEQ_OPTS_PUTS("dax=never"); 3019 } else if (test_opt2(sb, DAX_INODE)) { 3020 SEQ_OPTS_PUTS("dax=inode"); 3021 } 3022 3023 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3024 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3025 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 3026 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3027 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3028 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 3029 } 3030 3031 if (nodefs && !test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) 3032 SEQ_OPTS_PUTS("prefetch_block_bitmaps"); 3033 3034 if (ext4_emergency_ro(sb)) 3035 SEQ_OPTS_PUTS("emergency_ro"); 3036 3037 if (ext4_forced_shutdown(sb)) 3038 SEQ_OPTS_PUTS("shutdown"); 3039 3040 ext4_show_quota_options(seq, sb); 3041 return 0; 3042 } 3043 3044 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 3045 { 3046 return _ext4_show_options(seq, root->d_sb, 0); 3047 } 3048 3049 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3050 { 3051 struct super_block *sb = seq->private; 3052 int rc; 3053 3054 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3055 rc = _ext4_show_options(seq, sb, 1); 3056 seq_putc(seq, '\n'); 3057 return rc; 3058 } 3059 3060 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3061 int read_only) 3062 { 3063 struct ext4_sb_info *sbi = EXT4_SB(sb); 3064 int err = 0; 3065 3066 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3067 ext4_msg(sb, KERN_ERR, "revision level too high, " 3068 "forcing read-only mode"); 3069 err = -EROFS; 3070 goto done; 3071 } 3072 if (read_only) 3073 goto done; 3074 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3075 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3076 "running e2fsck is recommended"); 3077 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3078 ext4_msg(sb, KERN_WARNING, 3079 "warning: mounting fs with errors, " 3080 "running e2fsck is recommended"); 3081 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3082 le16_to_cpu(es->s_mnt_count) >= 3083 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3084 ext4_msg(sb, KERN_WARNING, 3085 "warning: maximal mount count reached, " 3086 "running e2fsck is recommended"); 3087 else if (le32_to_cpu(es->s_checkinterval) && 3088 (ext4_get_tstamp(es, s_lastcheck) + 3089 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3090 ext4_msg(sb, KERN_WARNING, 3091 "warning: checktime reached, " 3092 "running e2fsck is recommended"); 3093 if (!sbi->s_journal) 3094 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3095 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3096 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3097 le16_add_cpu(&es->s_mnt_count, 1); 3098 ext4_update_tstamp(es, s_mtime); 3099 if (sbi->s_journal) { 3100 ext4_set_feature_journal_needs_recovery(sb); 3101 if (ext4_has_feature_orphan_file(sb)) 3102 ext4_set_feature_orphan_present(sb); 3103 } 3104 3105 err = ext4_commit_super(sb); 3106 done: 3107 if (test_opt(sb, DEBUG)) 3108 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3109 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3110 sb->s_blocksize, 3111 sbi->s_groups_count, 3112 EXT4_BLOCKS_PER_GROUP(sb), 3113 EXT4_INODES_PER_GROUP(sb), 3114 sbi->s_mount_opt, sbi->s_mount_opt2); 3115 return err; 3116 } 3117 3118 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3119 { 3120 struct ext4_sb_info *sbi = EXT4_SB(sb); 3121 struct flex_groups **old_groups, **new_groups; 3122 int size, i, j; 3123 3124 if (!sbi->s_log_groups_per_flex) 3125 return 0; 3126 3127 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3128 if (size <= sbi->s_flex_groups_allocated) 3129 return 0; 3130 3131 new_groups = kvzalloc(roundup_pow_of_two(size * 3132 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3133 if (!new_groups) { 3134 ext4_msg(sb, KERN_ERR, 3135 "not enough memory for %d flex group pointers", size); 3136 return -ENOMEM; 3137 } 3138 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3139 new_groups[i] = kvzalloc(roundup_pow_of_two( 3140 sizeof(struct flex_groups)), 3141 GFP_KERNEL); 3142 if (!new_groups[i]) { 3143 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3144 kvfree(new_groups[j]); 3145 kvfree(new_groups); 3146 ext4_msg(sb, KERN_ERR, 3147 "not enough memory for %d flex groups", size); 3148 return -ENOMEM; 3149 } 3150 } 3151 rcu_read_lock(); 3152 old_groups = rcu_dereference(sbi->s_flex_groups); 3153 if (old_groups) 3154 memcpy(new_groups, old_groups, 3155 (sbi->s_flex_groups_allocated * 3156 sizeof(struct flex_groups *))); 3157 rcu_read_unlock(); 3158 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3159 sbi->s_flex_groups_allocated = size; 3160 if (old_groups) 3161 ext4_kvfree_array_rcu(old_groups); 3162 return 0; 3163 } 3164 3165 static int ext4_fill_flex_info(struct super_block *sb) 3166 { 3167 struct ext4_sb_info *sbi = EXT4_SB(sb); 3168 struct ext4_group_desc *gdp = NULL; 3169 struct flex_groups *fg; 3170 ext4_group_t flex_group; 3171 int i, err; 3172 3173 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3174 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3175 sbi->s_log_groups_per_flex = 0; 3176 return 1; 3177 } 3178 3179 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3180 if (err) 3181 goto failed; 3182 3183 for (i = 0; i < sbi->s_groups_count; i++) { 3184 gdp = ext4_get_group_desc(sb, i, NULL); 3185 3186 flex_group = ext4_flex_group(sbi, i); 3187 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3188 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3189 atomic64_add(ext4_free_group_clusters(sb, gdp), 3190 &fg->free_clusters); 3191 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3192 } 3193 3194 return 1; 3195 failed: 3196 return 0; 3197 } 3198 3199 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3200 struct ext4_group_desc *gdp) 3201 { 3202 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3203 __u16 crc = 0; 3204 __le32 le_group = cpu_to_le32(block_group); 3205 struct ext4_sb_info *sbi = EXT4_SB(sb); 3206 3207 if (ext4_has_feature_metadata_csum(sbi->s_sb)) { 3208 /* Use new metadata_csum algorithm */ 3209 __u32 csum32; 3210 __u16 dummy_csum = 0; 3211 3212 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 3213 sizeof(le_group)); 3214 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 3215 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 3216 sizeof(dummy_csum)); 3217 offset += sizeof(dummy_csum); 3218 if (offset < sbi->s_desc_size) 3219 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 3220 sbi->s_desc_size - offset); 3221 3222 crc = csum32 & 0xFFFF; 3223 goto out; 3224 } 3225 3226 /* old crc16 code */ 3227 if (!ext4_has_feature_gdt_csum(sb)) 3228 return 0; 3229 3230 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3231 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3232 crc = crc16(crc, (__u8 *)gdp, offset); 3233 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3234 /* for checksum of struct ext4_group_desc do the rest...*/ 3235 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size) 3236 crc = crc16(crc, (__u8 *)gdp + offset, 3237 sbi->s_desc_size - offset); 3238 3239 out: 3240 return cpu_to_le16(crc); 3241 } 3242 3243 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3244 struct ext4_group_desc *gdp) 3245 { 3246 if (ext4_has_group_desc_csum(sb) && 3247 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3248 return 0; 3249 3250 return 1; 3251 } 3252 3253 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3254 struct ext4_group_desc *gdp) 3255 { 3256 if (!ext4_has_group_desc_csum(sb)) 3257 return; 3258 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3259 } 3260 3261 /* Called at mount-time, super-block is locked */ 3262 static int ext4_check_descriptors(struct super_block *sb, 3263 ext4_fsblk_t sb_block, 3264 ext4_group_t *first_not_zeroed) 3265 { 3266 struct ext4_sb_info *sbi = EXT4_SB(sb); 3267 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3268 ext4_fsblk_t last_block; 3269 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3270 ext4_fsblk_t block_bitmap; 3271 ext4_fsblk_t inode_bitmap; 3272 ext4_fsblk_t inode_table; 3273 int flexbg_flag = 0; 3274 ext4_group_t i, grp = sbi->s_groups_count; 3275 3276 if (ext4_has_feature_flex_bg(sb)) 3277 flexbg_flag = 1; 3278 3279 ext4_debug("Checking group descriptors"); 3280 3281 for (i = 0; i < sbi->s_groups_count; i++) { 3282 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3283 3284 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3285 last_block = ext4_blocks_count(sbi->s_es) - 1; 3286 else 3287 last_block = first_block + 3288 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3289 3290 if ((grp == sbi->s_groups_count) && 3291 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3292 grp = i; 3293 3294 block_bitmap = ext4_block_bitmap(sb, gdp); 3295 if (block_bitmap == sb_block) { 3296 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3297 "Block bitmap for group %u overlaps " 3298 "superblock", i); 3299 if (!sb_rdonly(sb)) 3300 return 0; 3301 } 3302 if (block_bitmap >= sb_block + 1 && 3303 block_bitmap <= last_bg_block) { 3304 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3305 "Block bitmap for group %u overlaps " 3306 "block group descriptors", i); 3307 if (!sb_rdonly(sb)) 3308 return 0; 3309 } 3310 if (block_bitmap < first_block || block_bitmap > last_block) { 3311 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3312 "Block bitmap for group %u not in group " 3313 "(block %llu)!", i, block_bitmap); 3314 return 0; 3315 } 3316 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3317 if (inode_bitmap == sb_block) { 3318 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3319 "Inode bitmap for group %u overlaps " 3320 "superblock", i); 3321 if (!sb_rdonly(sb)) 3322 return 0; 3323 } 3324 if (inode_bitmap >= sb_block + 1 && 3325 inode_bitmap <= last_bg_block) { 3326 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3327 "Inode bitmap for group %u overlaps " 3328 "block group descriptors", i); 3329 if (!sb_rdonly(sb)) 3330 return 0; 3331 } 3332 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3333 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3334 "Inode bitmap for group %u not in group " 3335 "(block %llu)!", i, inode_bitmap); 3336 return 0; 3337 } 3338 inode_table = ext4_inode_table(sb, gdp); 3339 if (inode_table == sb_block) { 3340 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3341 "Inode table for group %u overlaps " 3342 "superblock", i); 3343 if (!sb_rdonly(sb)) 3344 return 0; 3345 } 3346 if (inode_table >= sb_block + 1 && 3347 inode_table <= last_bg_block) { 3348 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3349 "Inode table for group %u overlaps " 3350 "block group descriptors", i); 3351 if (!sb_rdonly(sb)) 3352 return 0; 3353 } 3354 if (inode_table < first_block || 3355 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3356 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3357 "Inode table for group %u not in group " 3358 "(block %llu)!", i, inode_table); 3359 return 0; 3360 } 3361 ext4_lock_group(sb, i); 3362 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3363 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3364 "Checksum for group %u failed (%u!=%u)", 3365 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3366 gdp)), le16_to_cpu(gdp->bg_checksum)); 3367 if (!sb_rdonly(sb)) { 3368 ext4_unlock_group(sb, i); 3369 return 0; 3370 } 3371 } 3372 ext4_unlock_group(sb, i); 3373 if (!flexbg_flag) 3374 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3375 } 3376 if (NULL != first_not_zeroed) 3377 *first_not_zeroed = grp; 3378 return 1; 3379 } 3380 3381 /* 3382 * Maximal extent format file size. 3383 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3384 * extent format containers, within a sector_t, and within i_blocks 3385 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3386 * so that won't be a limiting factor. 3387 * 3388 * However there is other limiting factor. We do store extents in the form 3389 * of starting block and length, hence the resulting length of the extent 3390 * covering maximum file size must fit into on-disk format containers as 3391 * well. Given that length is always by 1 unit bigger than max unit (because 3392 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3393 * 3394 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3395 */ 3396 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3397 { 3398 loff_t res; 3399 loff_t upper_limit = MAX_LFS_FILESIZE; 3400 3401 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3402 3403 if (!has_huge_files) { 3404 upper_limit = (1LL << 32) - 1; 3405 3406 /* total blocks in file system block size */ 3407 upper_limit >>= (blkbits - 9); 3408 upper_limit <<= blkbits; 3409 } 3410 3411 /* 3412 * 32-bit extent-start container, ee_block. We lower the maxbytes 3413 * by one fs block, so ee_len can cover the extent of maximum file 3414 * size 3415 */ 3416 res = (1LL << 32) - 1; 3417 res <<= blkbits; 3418 3419 /* Sanity check against vm- & vfs- imposed limits */ 3420 if (res > upper_limit) 3421 res = upper_limit; 3422 3423 return res; 3424 } 3425 3426 /* 3427 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3428 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3429 * We need to be 1 filesystem block less than the 2^48 sector limit. 3430 */ 3431 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3432 { 3433 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3434 int meta_blocks; 3435 unsigned int ppb = 1 << (bits - 2); 3436 3437 /* 3438 * This is calculated to be the largest file size for a dense, block 3439 * mapped file such that the file's total number of 512-byte sectors, 3440 * including data and all indirect blocks, does not exceed (2^48 - 1). 3441 * 3442 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3443 * number of 512-byte sectors of the file. 3444 */ 3445 if (!has_huge_files) { 3446 /* 3447 * !has_huge_files or implies that the inode i_block field 3448 * represents total file blocks in 2^32 512-byte sectors == 3449 * size of vfs inode i_blocks * 8 3450 */ 3451 upper_limit = (1LL << 32) - 1; 3452 3453 /* total blocks in file system block size */ 3454 upper_limit >>= (bits - 9); 3455 3456 } else { 3457 /* 3458 * We use 48 bit ext4_inode i_blocks 3459 * With EXT4_HUGE_FILE_FL set the i_blocks 3460 * represent total number of blocks in 3461 * file system block size 3462 */ 3463 upper_limit = (1LL << 48) - 1; 3464 3465 } 3466 3467 /* Compute how many blocks we can address by block tree */ 3468 res += ppb; 3469 res += ppb * ppb; 3470 res += ((loff_t)ppb) * ppb * ppb; 3471 /* Compute how many metadata blocks are needed */ 3472 meta_blocks = 1; 3473 meta_blocks += 1 + ppb; 3474 meta_blocks += 1 + ppb + ppb * ppb; 3475 /* Does block tree limit file size? */ 3476 if (res + meta_blocks <= upper_limit) 3477 goto check_lfs; 3478 3479 res = upper_limit; 3480 /* How many metadata blocks are needed for addressing upper_limit? */ 3481 upper_limit -= EXT4_NDIR_BLOCKS; 3482 /* indirect blocks */ 3483 meta_blocks = 1; 3484 upper_limit -= ppb; 3485 /* double indirect blocks */ 3486 if (upper_limit < ppb * ppb) { 3487 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3488 res -= meta_blocks; 3489 goto check_lfs; 3490 } 3491 meta_blocks += 1 + ppb; 3492 upper_limit -= ppb * ppb; 3493 /* tripple indirect blocks for the rest */ 3494 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3495 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3496 res -= meta_blocks; 3497 check_lfs: 3498 res <<= bits; 3499 if (res > MAX_LFS_FILESIZE) 3500 res = MAX_LFS_FILESIZE; 3501 3502 return res; 3503 } 3504 3505 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3506 ext4_fsblk_t logical_sb_block, int nr) 3507 { 3508 struct ext4_sb_info *sbi = EXT4_SB(sb); 3509 ext4_group_t bg, first_meta_bg; 3510 int has_super = 0; 3511 3512 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3513 3514 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3515 return logical_sb_block + nr + 1; 3516 bg = sbi->s_desc_per_block * nr; 3517 if (ext4_bg_has_super(sb, bg)) 3518 has_super = 1; 3519 3520 /* 3521 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3522 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3523 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3524 * compensate. 3525 */ 3526 if (sb->s_blocksize == 1024 && nr == 0 && 3527 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3528 has_super++; 3529 3530 return (has_super + ext4_group_first_block_no(sb, bg)); 3531 } 3532 3533 /** 3534 * ext4_get_stripe_size: Get the stripe size. 3535 * @sbi: In memory super block info 3536 * 3537 * If we have specified it via mount option, then 3538 * use the mount option value. If the value specified at mount time is 3539 * greater than the blocks per group use the super block value. 3540 * If the super block value is greater than blocks per group return 0. 3541 * Allocator needs it be less than blocks per group. 3542 * 3543 */ 3544 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3545 { 3546 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3547 unsigned long stripe_width = 3548 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3549 int ret; 3550 3551 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3552 ret = sbi->s_stripe; 3553 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3554 ret = stripe_width; 3555 else if (stride && stride <= sbi->s_blocks_per_group) 3556 ret = stride; 3557 else 3558 ret = 0; 3559 3560 /* 3561 * If the stripe width is 1, this makes no sense and 3562 * we set it to 0 to turn off stripe handling code. 3563 */ 3564 if (ret <= 1) 3565 ret = 0; 3566 3567 return ret; 3568 } 3569 3570 /* 3571 * Check whether this filesystem can be mounted based on 3572 * the features present and the RDONLY/RDWR mount requested. 3573 * Returns 1 if this filesystem can be mounted as requested, 3574 * 0 if it cannot be. 3575 */ 3576 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3577 { 3578 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3579 ext4_msg(sb, KERN_ERR, 3580 "Couldn't mount because of " 3581 "unsupported optional features (%x)", 3582 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3583 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3584 return 0; 3585 } 3586 3587 if (!IS_ENABLED(CONFIG_UNICODE) && ext4_has_feature_casefold(sb)) { 3588 ext4_msg(sb, KERN_ERR, 3589 "Filesystem with casefold feature cannot be " 3590 "mounted without CONFIG_UNICODE"); 3591 return 0; 3592 } 3593 3594 if (readonly) 3595 return 1; 3596 3597 if (ext4_has_feature_readonly(sb)) { 3598 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3599 sb->s_flags |= SB_RDONLY; 3600 return 1; 3601 } 3602 3603 /* Check that feature set is OK for a read-write mount */ 3604 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3605 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3606 "unsupported optional features (%x)", 3607 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3608 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3609 return 0; 3610 } 3611 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3612 ext4_msg(sb, KERN_ERR, 3613 "Can't support bigalloc feature without " 3614 "extents feature\n"); 3615 return 0; 3616 } 3617 3618 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3619 if (!readonly && (ext4_has_feature_quota(sb) || 3620 ext4_has_feature_project(sb))) { 3621 ext4_msg(sb, KERN_ERR, 3622 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3623 return 0; 3624 } 3625 #endif /* CONFIG_QUOTA */ 3626 return 1; 3627 } 3628 3629 /* 3630 * This function is called once a day if we have errors logged 3631 * on the file system 3632 */ 3633 static void print_daily_error_info(struct timer_list *t) 3634 { 3635 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3636 struct super_block *sb = sbi->s_sb; 3637 struct ext4_super_block *es = sbi->s_es; 3638 3639 if (es->s_error_count) 3640 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3641 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3642 le32_to_cpu(es->s_error_count)); 3643 if (es->s_first_error_time) { 3644 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3645 sb->s_id, 3646 ext4_get_tstamp(es, s_first_error_time), 3647 (int) sizeof(es->s_first_error_func), 3648 es->s_first_error_func, 3649 le32_to_cpu(es->s_first_error_line)); 3650 if (es->s_first_error_ino) 3651 printk(KERN_CONT ": inode %u", 3652 le32_to_cpu(es->s_first_error_ino)); 3653 if (es->s_first_error_block) 3654 printk(KERN_CONT ": block %llu", (unsigned long long) 3655 le64_to_cpu(es->s_first_error_block)); 3656 printk(KERN_CONT "\n"); 3657 } 3658 if (es->s_last_error_time) { 3659 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3660 sb->s_id, 3661 ext4_get_tstamp(es, s_last_error_time), 3662 (int) sizeof(es->s_last_error_func), 3663 es->s_last_error_func, 3664 le32_to_cpu(es->s_last_error_line)); 3665 if (es->s_last_error_ino) 3666 printk(KERN_CONT ": inode %u", 3667 le32_to_cpu(es->s_last_error_ino)); 3668 if (es->s_last_error_block) 3669 printk(KERN_CONT ": block %llu", (unsigned long long) 3670 le64_to_cpu(es->s_last_error_block)); 3671 printk(KERN_CONT "\n"); 3672 } 3673 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3674 } 3675 3676 /* Find next suitable group and run ext4_init_inode_table */ 3677 static int ext4_run_li_request(struct ext4_li_request *elr) 3678 { 3679 struct ext4_group_desc *gdp = NULL; 3680 struct super_block *sb = elr->lr_super; 3681 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3682 ext4_group_t group = elr->lr_next_group; 3683 unsigned int prefetch_ios = 0; 3684 int ret = 0; 3685 int nr = EXT4_SB(sb)->s_mb_prefetch; 3686 u64 start_time; 3687 3688 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3689 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios); 3690 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr); 3691 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr); 3692 if (group >= elr->lr_next_group) { 3693 ret = 1; 3694 if (elr->lr_first_not_zeroed != ngroups && 3695 !ext4_emergency_state(sb) && !sb_rdonly(sb) && 3696 test_opt(sb, INIT_INODE_TABLE)) { 3697 elr->lr_next_group = elr->lr_first_not_zeroed; 3698 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3699 ret = 0; 3700 } 3701 } 3702 return ret; 3703 } 3704 3705 for (; group < ngroups; group++) { 3706 gdp = ext4_get_group_desc(sb, group, NULL); 3707 if (!gdp) { 3708 ret = 1; 3709 break; 3710 } 3711 3712 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3713 break; 3714 } 3715 3716 if (group >= ngroups) 3717 ret = 1; 3718 3719 if (!ret) { 3720 start_time = ktime_get_ns(); 3721 ret = ext4_init_inode_table(sb, group, 3722 elr->lr_timeout ? 0 : 1); 3723 trace_ext4_lazy_itable_init(sb, group); 3724 if (elr->lr_timeout == 0) { 3725 elr->lr_timeout = nsecs_to_jiffies((ktime_get_ns() - start_time) * 3726 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3727 } 3728 elr->lr_next_sched = jiffies + elr->lr_timeout; 3729 elr->lr_next_group = group + 1; 3730 } 3731 return ret; 3732 } 3733 3734 /* 3735 * Remove lr_request from the list_request and free the 3736 * request structure. Should be called with li_list_mtx held 3737 */ 3738 static void ext4_remove_li_request(struct ext4_li_request *elr) 3739 { 3740 if (!elr) 3741 return; 3742 3743 list_del(&elr->lr_request); 3744 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3745 kfree(elr); 3746 } 3747 3748 static void ext4_unregister_li_request(struct super_block *sb) 3749 { 3750 mutex_lock(&ext4_li_mtx); 3751 if (!ext4_li_info) { 3752 mutex_unlock(&ext4_li_mtx); 3753 return; 3754 } 3755 3756 mutex_lock(&ext4_li_info->li_list_mtx); 3757 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3758 mutex_unlock(&ext4_li_info->li_list_mtx); 3759 mutex_unlock(&ext4_li_mtx); 3760 } 3761 3762 static struct task_struct *ext4_lazyinit_task; 3763 3764 /* 3765 * This is the function where ext4lazyinit thread lives. It walks 3766 * through the request list searching for next scheduled filesystem. 3767 * When such a fs is found, run the lazy initialization request 3768 * (ext4_rn_li_request) and keep track of the time spend in this 3769 * function. Based on that time we compute next schedule time of 3770 * the request. When walking through the list is complete, compute 3771 * next waking time and put itself into sleep. 3772 */ 3773 static int ext4_lazyinit_thread(void *arg) 3774 { 3775 struct ext4_lazy_init *eli = arg; 3776 struct list_head *pos, *n; 3777 struct ext4_li_request *elr; 3778 unsigned long next_wakeup, cur; 3779 3780 BUG_ON(NULL == eli); 3781 set_freezable(); 3782 3783 cont_thread: 3784 while (true) { 3785 bool next_wakeup_initialized = false; 3786 3787 next_wakeup = 0; 3788 mutex_lock(&eli->li_list_mtx); 3789 if (list_empty(&eli->li_request_list)) { 3790 mutex_unlock(&eli->li_list_mtx); 3791 goto exit_thread; 3792 } 3793 list_for_each_safe(pos, n, &eli->li_request_list) { 3794 int err = 0; 3795 int progress = 0; 3796 elr = list_entry(pos, struct ext4_li_request, 3797 lr_request); 3798 3799 if (time_before(jiffies, elr->lr_next_sched)) { 3800 if (!next_wakeup_initialized || 3801 time_before(elr->lr_next_sched, next_wakeup)) { 3802 next_wakeup = elr->lr_next_sched; 3803 next_wakeup_initialized = true; 3804 } 3805 continue; 3806 } 3807 if (down_read_trylock(&elr->lr_super->s_umount)) { 3808 if (sb_start_write_trylock(elr->lr_super)) { 3809 progress = 1; 3810 /* 3811 * We hold sb->s_umount, sb can not 3812 * be removed from the list, it is 3813 * now safe to drop li_list_mtx 3814 */ 3815 mutex_unlock(&eli->li_list_mtx); 3816 err = ext4_run_li_request(elr); 3817 sb_end_write(elr->lr_super); 3818 mutex_lock(&eli->li_list_mtx); 3819 n = pos->next; 3820 } 3821 up_read((&elr->lr_super->s_umount)); 3822 } 3823 /* error, remove the lazy_init job */ 3824 if (err) { 3825 ext4_remove_li_request(elr); 3826 continue; 3827 } 3828 if (!progress) { 3829 elr->lr_next_sched = jiffies + 3830 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3831 } 3832 if (!next_wakeup_initialized || 3833 time_before(elr->lr_next_sched, next_wakeup)) { 3834 next_wakeup = elr->lr_next_sched; 3835 next_wakeup_initialized = true; 3836 } 3837 } 3838 mutex_unlock(&eli->li_list_mtx); 3839 3840 try_to_freeze(); 3841 3842 cur = jiffies; 3843 if (!next_wakeup_initialized || time_after_eq(cur, next_wakeup)) { 3844 cond_resched(); 3845 continue; 3846 } 3847 3848 schedule_timeout_interruptible(next_wakeup - cur); 3849 3850 if (kthread_should_stop()) { 3851 ext4_clear_request_list(); 3852 goto exit_thread; 3853 } 3854 } 3855 3856 exit_thread: 3857 /* 3858 * It looks like the request list is empty, but we need 3859 * to check it under the li_list_mtx lock, to prevent any 3860 * additions into it, and of course we should lock ext4_li_mtx 3861 * to atomically free the list and ext4_li_info, because at 3862 * this point another ext4 filesystem could be registering 3863 * new one. 3864 */ 3865 mutex_lock(&ext4_li_mtx); 3866 mutex_lock(&eli->li_list_mtx); 3867 if (!list_empty(&eli->li_request_list)) { 3868 mutex_unlock(&eli->li_list_mtx); 3869 mutex_unlock(&ext4_li_mtx); 3870 goto cont_thread; 3871 } 3872 mutex_unlock(&eli->li_list_mtx); 3873 kfree(ext4_li_info); 3874 ext4_li_info = NULL; 3875 mutex_unlock(&ext4_li_mtx); 3876 3877 return 0; 3878 } 3879 3880 static void ext4_clear_request_list(void) 3881 { 3882 struct list_head *pos, *n; 3883 struct ext4_li_request *elr; 3884 3885 mutex_lock(&ext4_li_info->li_list_mtx); 3886 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3887 elr = list_entry(pos, struct ext4_li_request, 3888 lr_request); 3889 ext4_remove_li_request(elr); 3890 } 3891 mutex_unlock(&ext4_li_info->li_list_mtx); 3892 } 3893 3894 static int ext4_run_lazyinit_thread(void) 3895 { 3896 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3897 ext4_li_info, "ext4lazyinit"); 3898 if (IS_ERR(ext4_lazyinit_task)) { 3899 int err = PTR_ERR(ext4_lazyinit_task); 3900 ext4_clear_request_list(); 3901 kfree(ext4_li_info); 3902 ext4_li_info = NULL; 3903 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3904 "initialization thread\n", 3905 err); 3906 return err; 3907 } 3908 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3909 return 0; 3910 } 3911 3912 /* 3913 * Check whether it make sense to run itable init. thread or not. 3914 * If there is at least one uninitialized inode table, return 3915 * corresponding group number, else the loop goes through all 3916 * groups and return total number of groups. 3917 */ 3918 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3919 { 3920 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3921 struct ext4_group_desc *gdp = NULL; 3922 3923 if (!ext4_has_group_desc_csum(sb)) 3924 return ngroups; 3925 3926 for (group = 0; group < ngroups; group++) { 3927 gdp = ext4_get_group_desc(sb, group, NULL); 3928 if (!gdp) 3929 continue; 3930 3931 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3932 break; 3933 } 3934 3935 return group; 3936 } 3937 3938 static int ext4_li_info_new(void) 3939 { 3940 struct ext4_lazy_init *eli = NULL; 3941 3942 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3943 if (!eli) 3944 return -ENOMEM; 3945 3946 INIT_LIST_HEAD(&eli->li_request_list); 3947 mutex_init(&eli->li_list_mtx); 3948 3949 eli->li_state |= EXT4_LAZYINIT_QUIT; 3950 3951 ext4_li_info = eli; 3952 3953 return 0; 3954 } 3955 3956 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3957 ext4_group_t start) 3958 { 3959 struct ext4_li_request *elr; 3960 3961 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3962 if (!elr) 3963 return NULL; 3964 3965 elr->lr_super = sb; 3966 elr->lr_first_not_zeroed = start; 3967 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3968 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3969 elr->lr_next_group = start; 3970 } else { 3971 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3972 } 3973 3974 /* 3975 * Randomize first schedule time of the request to 3976 * spread the inode table initialization requests 3977 * better. 3978 */ 3979 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3980 return elr; 3981 } 3982 3983 int ext4_register_li_request(struct super_block *sb, 3984 ext4_group_t first_not_zeroed) 3985 { 3986 struct ext4_sb_info *sbi = EXT4_SB(sb); 3987 struct ext4_li_request *elr = NULL; 3988 ext4_group_t ngroups = sbi->s_groups_count; 3989 int ret = 0; 3990 3991 mutex_lock(&ext4_li_mtx); 3992 if (sbi->s_li_request != NULL) { 3993 /* 3994 * Reset timeout so it can be computed again, because 3995 * s_li_wait_mult might have changed. 3996 */ 3997 sbi->s_li_request->lr_timeout = 0; 3998 goto out; 3999 } 4000 4001 if (ext4_emergency_state(sb) || sb_rdonly(sb) || 4002 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 4003 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 4004 goto out; 4005 4006 elr = ext4_li_request_new(sb, first_not_zeroed); 4007 if (!elr) { 4008 ret = -ENOMEM; 4009 goto out; 4010 } 4011 4012 if (NULL == ext4_li_info) { 4013 ret = ext4_li_info_new(); 4014 if (ret) 4015 goto out; 4016 } 4017 4018 mutex_lock(&ext4_li_info->li_list_mtx); 4019 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 4020 mutex_unlock(&ext4_li_info->li_list_mtx); 4021 4022 sbi->s_li_request = elr; 4023 /* 4024 * set elr to NULL here since it has been inserted to 4025 * the request_list and the removal and free of it is 4026 * handled by ext4_clear_request_list from now on. 4027 */ 4028 elr = NULL; 4029 4030 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 4031 ret = ext4_run_lazyinit_thread(); 4032 if (ret) 4033 goto out; 4034 } 4035 out: 4036 mutex_unlock(&ext4_li_mtx); 4037 if (ret) 4038 kfree(elr); 4039 return ret; 4040 } 4041 4042 /* 4043 * We do not need to lock anything since this is called on 4044 * module unload. 4045 */ 4046 static void ext4_destroy_lazyinit_thread(void) 4047 { 4048 /* 4049 * If thread exited earlier 4050 * there's nothing to be done. 4051 */ 4052 if (!ext4_li_info || !ext4_lazyinit_task) 4053 return; 4054 4055 kthread_stop(ext4_lazyinit_task); 4056 } 4057 4058 static int set_journal_csum_feature_set(struct super_block *sb) 4059 { 4060 int ret = 1; 4061 int compat, incompat; 4062 struct ext4_sb_info *sbi = EXT4_SB(sb); 4063 4064 if (ext4_has_feature_metadata_csum(sb)) { 4065 /* journal checksum v3 */ 4066 compat = 0; 4067 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4068 } else { 4069 /* journal checksum v1 */ 4070 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4071 incompat = 0; 4072 } 4073 4074 jbd2_journal_clear_features(sbi->s_journal, 4075 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4076 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4077 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4078 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4079 ret = jbd2_journal_set_features(sbi->s_journal, 4080 compat, 0, 4081 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4082 incompat); 4083 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4084 ret = jbd2_journal_set_features(sbi->s_journal, 4085 compat, 0, 4086 incompat); 4087 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4088 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4089 } else { 4090 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4091 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4092 } 4093 4094 return ret; 4095 } 4096 4097 /* 4098 * Note: calculating the overhead so we can be compatible with 4099 * historical BSD practice is quite difficult in the face of 4100 * clusters/bigalloc. This is because multiple metadata blocks from 4101 * different block group can end up in the same allocation cluster. 4102 * Calculating the exact overhead in the face of clustered allocation 4103 * requires either O(all block bitmaps) in memory or O(number of block 4104 * groups**2) in time. We will still calculate the superblock for 4105 * older file systems --- and if we come across with a bigalloc file 4106 * system with zero in s_overhead_clusters the estimate will be close to 4107 * correct especially for very large cluster sizes --- but for newer 4108 * file systems, it's better to calculate this figure once at mkfs 4109 * time, and store it in the superblock. If the superblock value is 4110 * present (even for non-bigalloc file systems), we will use it. 4111 */ 4112 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4113 char *buf) 4114 { 4115 struct ext4_sb_info *sbi = EXT4_SB(sb); 4116 struct ext4_group_desc *gdp; 4117 ext4_fsblk_t first_block, last_block, b; 4118 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4119 int s, j, count = 0; 4120 int has_super = ext4_bg_has_super(sb, grp); 4121 4122 if (!ext4_has_feature_bigalloc(sb)) 4123 return (has_super + ext4_bg_num_gdb(sb, grp) + 4124 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4125 sbi->s_itb_per_group + 2); 4126 4127 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4128 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4129 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4130 for (i = 0; i < ngroups; i++) { 4131 gdp = ext4_get_group_desc(sb, i, NULL); 4132 b = ext4_block_bitmap(sb, gdp); 4133 if (b >= first_block && b <= last_block) { 4134 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4135 count++; 4136 } 4137 b = ext4_inode_bitmap(sb, gdp); 4138 if (b >= first_block && b <= last_block) { 4139 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4140 count++; 4141 } 4142 b = ext4_inode_table(sb, gdp); 4143 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4144 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4145 int c = EXT4_B2C(sbi, b - first_block); 4146 ext4_set_bit(c, buf); 4147 count++; 4148 } 4149 if (i != grp) 4150 continue; 4151 s = 0; 4152 if (ext4_bg_has_super(sb, grp)) { 4153 ext4_set_bit(s++, buf); 4154 count++; 4155 } 4156 j = ext4_bg_num_gdb(sb, grp); 4157 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4158 ext4_error(sb, "Invalid number of block group " 4159 "descriptor blocks: %d", j); 4160 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4161 } 4162 count += j; 4163 for (; j > 0; j--) 4164 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4165 } 4166 if (!count) 4167 return 0; 4168 return EXT4_CLUSTERS_PER_GROUP(sb) - 4169 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4170 } 4171 4172 /* 4173 * Compute the overhead and stash it in sbi->s_overhead 4174 */ 4175 int ext4_calculate_overhead(struct super_block *sb) 4176 { 4177 struct ext4_sb_info *sbi = EXT4_SB(sb); 4178 struct ext4_super_block *es = sbi->s_es; 4179 struct inode *j_inode; 4180 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4181 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4182 ext4_fsblk_t overhead = 0; 4183 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4184 4185 if (!buf) 4186 return -ENOMEM; 4187 4188 /* 4189 * Compute the overhead (FS structures). This is constant 4190 * for a given filesystem unless the number of block groups 4191 * changes so we cache the previous value until it does. 4192 */ 4193 4194 /* 4195 * All of the blocks before first_data_block are overhead 4196 */ 4197 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4198 4199 /* 4200 * Add the overhead found in each block group 4201 */ 4202 for (i = 0; i < ngroups; i++) { 4203 int blks; 4204 4205 blks = count_overhead(sb, i, buf); 4206 overhead += blks; 4207 if (blks) 4208 memset(buf, 0, PAGE_SIZE); 4209 cond_resched(); 4210 } 4211 4212 /* 4213 * Add the internal journal blocks whether the journal has been 4214 * loaded or not 4215 */ 4216 if (sbi->s_journal && !sbi->s_journal_bdev_file) 4217 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4218 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4219 /* j_inum for internal journal is non-zero */ 4220 j_inode = ext4_get_journal_inode(sb, j_inum); 4221 if (!IS_ERR(j_inode)) { 4222 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4223 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4224 iput(j_inode); 4225 } else { 4226 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4227 } 4228 } 4229 sbi->s_overhead = overhead; 4230 smp_wmb(); 4231 free_page((unsigned long) buf); 4232 return 0; 4233 } 4234 4235 static void ext4_set_resv_clusters(struct super_block *sb) 4236 { 4237 ext4_fsblk_t resv_clusters; 4238 struct ext4_sb_info *sbi = EXT4_SB(sb); 4239 4240 /* 4241 * There's no need to reserve anything when we aren't using extents. 4242 * The space estimates are exact, there are no unwritten extents, 4243 * hole punching doesn't need new metadata... This is needed especially 4244 * to keep ext2/3 backward compatibility. 4245 */ 4246 if (!ext4_has_feature_extents(sb)) 4247 return; 4248 /* 4249 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4250 * This should cover the situations where we can not afford to run 4251 * out of space like for example punch hole, or converting 4252 * unwritten extents in delalloc path. In most cases such 4253 * allocation would require 1, or 2 blocks, higher numbers are 4254 * very rare. 4255 */ 4256 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4257 sbi->s_cluster_bits); 4258 4259 do_div(resv_clusters, 50); 4260 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4261 4262 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4263 } 4264 4265 static const char *ext4_quota_mode(struct super_block *sb) 4266 { 4267 #ifdef CONFIG_QUOTA 4268 if (!ext4_quota_capable(sb)) 4269 return "none"; 4270 4271 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4272 return "journalled"; 4273 else 4274 return "writeback"; 4275 #else 4276 return "disabled"; 4277 #endif 4278 } 4279 4280 static void ext4_setup_csum_trigger(struct super_block *sb, 4281 enum ext4_journal_trigger_type type, 4282 void (*trigger)( 4283 struct jbd2_buffer_trigger_type *type, 4284 struct buffer_head *bh, 4285 void *mapped_data, 4286 size_t size)) 4287 { 4288 struct ext4_sb_info *sbi = EXT4_SB(sb); 4289 4290 sbi->s_journal_triggers[type].sb = sb; 4291 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4292 } 4293 4294 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4295 { 4296 if (!sbi) 4297 return; 4298 4299 kfree(sbi->s_blockgroup_lock); 4300 fs_put_dax(sbi->s_daxdev, NULL); 4301 kfree(sbi); 4302 } 4303 4304 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4305 { 4306 struct ext4_sb_info *sbi; 4307 4308 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4309 if (!sbi) 4310 return NULL; 4311 4312 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4313 NULL, NULL); 4314 4315 sbi->s_blockgroup_lock = 4316 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4317 4318 if (!sbi->s_blockgroup_lock) 4319 goto err_out; 4320 4321 sb->s_fs_info = sbi; 4322 sbi->s_sb = sb; 4323 return sbi; 4324 err_out: 4325 fs_put_dax(sbi->s_daxdev, NULL); 4326 kfree(sbi); 4327 return NULL; 4328 } 4329 4330 static void ext4_set_def_opts(struct super_block *sb, 4331 struct ext4_super_block *es) 4332 { 4333 unsigned long def_mount_opts; 4334 4335 /* Set defaults before we parse the mount options */ 4336 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4337 set_opt(sb, INIT_INODE_TABLE); 4338 if (def_mount_opts & EXT4_DEFM_DEBUG) 4339 set_opt(sb, DEBUG); 4340 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4341 set_opt(sb, GRPID); 4342 if (def_mount_opts & EXT4_DEFM_UID16) 4343 set_opt(sb, NO_UID32); 4344 /* xattr user namespace & acls are now defaulted on */ 4345 set_opt(sb, XATTR_USER); 4346 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4347 set_opt(sb, POSIX_ACL); 4348 #endif 4349 if (ext4_has_feature_fast_commit(sb)) 4350 set_opt2(sb, JOURNAL_FAST_COMMIT); 4351 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4352 if (ext4_has_feature_metadata_csum(sb)) 4353 set_opt(sb, JOURNAL_CHECKSUM); 4354 4355 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4356 set_opt(sb, JOURNAL_DATA); 4357 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4358 set_opt(sb, ORDERED_DATA); 4359 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4360 set_opt(sb, WRITEBACK_DATA); 4361 4362 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4363 set_opt(sb, ERRORS_PANIC); 4364 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4365 set_opt(sb, ERRORS_CONT); 4366 else 4367 set_opt(sb, ERRORS_RO); 4368 /* block_validity enabled by default; disable with noblock_validity */ 4369 set_opt(sb, BLOCK_VALIDITY); 4370 if (def_mount_opts & EXT4_DEFM_DISCARD) 4371 set_opt(sb, DISCARD); 4372 4373 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4374 set_opt(sb, BARRIER); 4375 4376 /* 4377 * enable delayed allocation by default 4378 * Use -o nodelalloc to turn it off 4379 */ 4380 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4381 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4382 set_opt(sb, DELALLOC); 4383 4384 if (sb->s_blocksize <= PAGE_SIZE) 4385 set_opt(sb, DIOREAD_NOLOCK); 4386 } 4387 4388 static int ext4_handle_clustersize(struct super_block *sb) 4389 { 4390 struct ext4_sb_info *sbi = EXT4_SB(sb); 4391 struct ext4_super_block *es = sbi->s_es; 4392 int clustersize; 4393 4394 /* Handle clustersize */ 4395 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4396 if (ext4_has_feature_bigalloc(sb)) { 4397 if (clustersize < sb->s_blocksize) { 4398 ext4_msg(sb, KERN_ERR, 4399 "cluster size (%d) smaller than " 4400 "block size (%lu)", clustersize, sb->s_blocksize); 4401 return -EINVAL; 4402 } 4403 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4404 le32_to_cpu(es->s_log_block_size); 4405 } else { 4406 if (clustersize != sb->s_blocksize) { 4407 ext4_msg(sb, KERN_ERR, 4408 "fragment/cluster size (%d) != " 4409 "block size (%lu)", clustersize, sb->s_blocksize); 4410 return -EINVAL; 4411 } 4412 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4413 ext4_msg(sb, KERN_ERR, 4414 "#blocks per group too big: %lu", 4415 sbi->s_blocks_per_group); 4416 return -EINVAL; 4417 } 4418 sbi->s_cluster_bits = 0; 4419 } 4420 sbi->s_clusters_per_group = le32_to_cpu(es->s_clusters_per_group); 4421 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4422 ext4_msg(sb, KERN_ERR, "#clusters per group too big: %lu", 4423 sbi->s_clusters_per_group); 4424 return -EINVAL; 4425 } 4426 if (sbi->s_blocks_per_group != 4427 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4428 ext4_msg(sb, KERN_ERR, 4429 "blocks per group (%lu) and clusters per group (%lu) inconsistent", 4430 sbi->s_blocks_per_group, sbi->s_clusters_per_group); 4431 return -EINVAL; 4432 } 4433 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4434 4435 /* Do we have standard group size of clustersize * 8 blocks ? */ 4436 if (sbi->s_blocks_per_group == clustersize << 3) 4437 set_opt2(sb, STD_GROUP_SIZE); 4438 4439 return 0; 4440 } 4441 4442 /* 4443 * ext4_atomic_write_init: Initializes filesystem min & max atomic write units. 4444 * @sb: super block 4445 * TODO: Later add support for bigalloc 4446 */ 4447 static void ext4_atomic_write_init(struct super_block *sb) 4448 { 4449 struct ext4_sb_info *sbi = EXT4_SB(sb); 4450 struct block_device *bdev = sb->s_bdev; 4451 4452 if (!bdev_can_atomic_write(bdev)) 4453 return; 4454 4455 if (!ext4_has_feature_extents(sb)) 4456 return; 4457 4458 sbi->s_awu_min = max(sb->s_blocksize, 4459 bdev_atomic_write_unit_min_bytes(bdev)); 4460 sbi->s_awu_max = min(sb->s_blocksize, 4461 bdev_atomic_write_unit_max_bytes(bdev)); 4462 if (sbi->s_awu_min && sbi->s_awu_max && 4463 sbi->s_awu_min <= sbi->s_awu_max) { 4464 ext4_msg(sb, KERN_NOTICE, "Supports (experimental) DIO atomic writes awu_min: %u, awu_max: %u", 4465 sbi->s_awu_min, sbi->s_awu_max); 4466 } else { 4467 sbi->s_awu_min = 0; 4468 sbi->s_awu_max = 0; 4469 } 4470 } 4471 4472 static void ext4_fast_commit_init(struct super_block *sb) 4473 { 4474 struct ext4_sb_info *sbi = EXT4_SB(sb); 4475 4476 /* Initialize fast commit stuff */ 4477 atomic_set(&sbi->s_fc_subtid, 0); 4478 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4479 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4480 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4481 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4482 sbi->s_fc_bytes = 0; 4483 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4484 sbi->s_fc_ineligible_tid = 0; 4485 spin_lock_init(&sbi->s_fc_lock); 4486 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4487 sbi->s_fc_replay_state.fc_regions = NULL; 4488 sbi->s_fc_replay_state.fc_regions_size = 0; 4489 sbi->s_fc_replay_state.fc_regions_used = 0; 4490 sbi->s_fc_replay_state.fc_regions_valid = 0; 4491 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4492 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4493 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4494 } 4495 4496 static int ext4_inode_info_init(struct super_block *sb, 4497 struct ext4_super_block *es) 4498 { 4499 struct ext4_sb_info *sbi = EXT4_SB(sb); 4500 4501 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4502 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4503 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4504 } else { 4505 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4506 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4507 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4508 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4509 sbi->s_first_ino); 4510 return -EINVAL; 4511 } 4512 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4513 (!is_power_of_2(sbi->s_inode_size)) || 4514 (sbi->s_inode_size > sb->s_blocksize)) { 4515 ext4_msg(sb, KERN_ERR, 4516 "unsupported inode size: %d", 4517 sbi->s_inode_size); 4518 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4519 return -EINVAL; 4520 } 4521 /* 4522 * i_atime_extra is the last extra field available for 4523 * [acm]times in struct ext4_inode. Checking for that 4524 * field should suffice to ensure we have extra space 4525 * for all three. 4526 */ 4527 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4528 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4529 sb->s_time_gran = 1; 4530 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4531 } else { 4532 sb->s_time_gran = NSEC_PER_SEC; 4533 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4534 } 4535 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4536 } 4537 4538 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4539 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4540 EXT4_GOOD_OLD_INODE_SIZE; 4541 if (ext4_has_feature_extra_isize(sb)) { 4542 unsigned v, max = (sbi->s_inode_size - 4543 EXT4_GOOD_OLD_INODE_SIZE); 4544 4545 v = le16_to_cpu(es->s_want_extra_isize); 4546 if (v > max) { 4547 ext4_msg(sb, KERN_ERR, 4548 "bad s_want_extra_isize: %d", v); 4549 return -EINVAL; 4550 } 4551 if (sbi->s_want_extra_isize < v) 4552 sbi->s_want_extra_isize = v; 4553 4554 v = le16_to_cpu(es->s_min_extra_isize); 4555 if (v > max) { 4556 ext4_msg(sb, KERN_ERR, 4557 "bad s_min_extra_isize: %d", v); 4558 return -EINVAL; 4559 } 4560 if (sbi->s_want_extra_isize < v) 4561 sbi->s_want_extra_isize = v; 4562 } 4563 } 4564 4565 return 0; 4566 } 4567 4568 #if IS_ENABLED(CONFIG_UNICODE) 4569 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4570 { 4571 const struct ext4_sb_encodings *encoding_info; 4572 struct unicode_map *encoding; 4573 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4574 4575 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4576 return 0; 4577 4578 encoding_info = ext4_sb_read_encoding(es); 4579 if (!encoding_info) { 4580 ext4_msg(sb, KERN_ERR, 4581 "Encoding requested by superblock is unknown"); 4582 return -EINVAL; 4583 } 4584 4585 encoding = utf8_load(encoding_info->version); 4586 if (IS_ERR(encoding)) { 4587 ext4_msg(sb, KERN_ERR, 4588 "can't mount with superblock charset: %s-%u.%u.%u " 4589 "not supported by the kernel. flags: 0x%x.", 4590 encoding_info->name, 4591 unicode_major(encoding_info->version), 4592 unicode_minor(encoding_info->version), 4593 unicode_rev(encoding_info->version), 4594 encoding_flags); 4595 return -EINVAL; 4596 } 4597 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4598 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4599 unicode_major(encoding_info->version), 4600 unicode_minor(encoding_info->version), 4601 unicode_rev(encoding_info->version), 4602 encoding_flags); 4603 4604 sb->s_encoding = encoding; 4605 sb->s_encoding_flags = encoding_flags; 4606 4607 return 0; 4608 } 4609 #else 4610 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4611 { 4612 return 0; 4613 } 4614 #endif 4615 4616 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4617 { 4618 struct ext4_sb_info *sbi = EXT4_SB(sb); 4619 4620 /* Warn if metadata_csum and gdt_csum are both set. */ 4621 if (ext4_has_feature_metadata_csum(sb) && 4622 ext4_has_feature_gdt_csum(sb)) 4623 ext4_warning(sb, "metadata_csum and uninit_bg are " 4624 "redundant flags; please run fsck."); 4625 4626 /* Check for a known checksum algorithm */ 4627 if (!ext4_verify_csum_type(sb, es)) { 4628 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4629 "unknown checksum algorithm."); 4630 return -EINVAL; 4631 } 4632 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4633 ext4_orphan_file_block_trigger); 4634 4635 /* Check superblock checksum */ 4636 if (!ext4_superblock_csum_verify(sb, es)) { 4637 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4638 "invalid superblock checksum. Run e2fsck?"); 4639 return -EFSBADCRC; 4640 } 4641 4642 /* Precompute checksum seed for all metadata */ 4643 if (ext4_has_feature_csum_seed(sb)) 4644 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4645 else if (ext4_has_feature_metadata_csum(sb) || 4646 ext4_has_feature_ea_inode(sb)) 4647 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4648 sizeof(es->s_uuid)); 4649 return 0; 4650 } 4651 4652 static int ext4_check_feature_compatibility(struct super_block *sb, 4653 struct ext4_super_block *es, 4654 int silent) 4655 { 4656 struct ext4_sb_info *sbi = EXT4_SB(sb); 4657 4658 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4659 (ext4_has_compat_features(sb) || 4660 ext4_has_ro_compat_features(sb) || 4661 ext4_has_incompat_features(sb))) 4662 ext4_msg(sb, KERN_WARNING, 4663 "feature flags set on rev 0 fs, " 4664 "running e2fsck is recommended"); 4665 4666 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4667 set_opt2(sb, HURD_COMPAT); 4668 if (ext4_has_feature_64bit(sb)) { 4669 ext4_msg(sb, KERN_ERR, 4670 "The Hurd can't support 64-bit file systems"); 4671 return -EINVAL; 4672 } 4673 4674 /* 4675 * ea_inode feature uses l_i_version field which is not 4676 * available in HURD_COMPAT mode. 4677 */ 4678 if (ext4_has_feature_ea_inode(sb)) { 4679 ext4_msg(sb, KERN_ERR, 4680 "ea_inode feature is not supported for Hurd"); 4681 return -EINVAL; 4682 } 4683 } 4684 4685 if (IS_EXT2_SB(sb)) { 4686 if (ext2_feature_set_ok(sb)) 4687 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4688 "using the ext4 subsystem"); 4689 else { 4690 /* 4691 * If we're probing be silent, if this looks like 4692 * it's actually an ext[34] filesystem. 4693 */ 4694 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4695 return -EINVAL; 4696 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4697 "to feature incompatibilities"); 4698 return -EINVAL; 4699 } 4700 } 4701 4702 if (IS_EXT3_SB(sb)) { 4703 if (ext3_feature_set_ok(sb)) 4704 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4705 "using the ext4 subsystem"); 4706 else { 4707 /* 4708 * If we're probing be silent, if this looks like 4709 * it's actually an ext4 filesystem. 4710 */ 4711 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4712 return -EINVAL; 4713 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4714 "to feature incompatibilities"); 4715 return -EINVAL; 4716 } 4717 } 4718 4719 /* 4720 * Check feature flags regardless of the revision level, since we 4721 * previously didn't change the revision level when setting the flags, 4722 * so there is a chance incompat flags are set on a rev 0 filesystem. 4723 */ 4724 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4725 return -EINVAL; 4726 4727 if (sbi->s_daxdev) { 4728 if (sb->s_blocksize == PAGE_SIZE) 4729 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4730 else 4731 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 4732 } 4733 4734 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4735 if (ext4_has_feature_inline_data(sb)) { 4736 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4737 " that may contain inline data"); 4738 return -EINVAL; 4739 } 4740 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4741 ext4_msg(sb, KERN_ERR, 4742 "DAX unsupported by block device."); 4743 return -EINVAL; 4744 } 4745 } 4746 4747 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4748 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4749 es->s_encryption_level); 4750 return -EINVAL; 4751 } 4752 4753 return 0; 4754 } 4755 4756 static int ext4_check_geometry(struct super_block *sb, 4757 struct ext4_super_block *es) 4758 { 4759 struct ext4_sb_info *sbi = EXT4_SB(sb); 4760 __u64 blocks_count; 4761 int err; 4762 4763 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 4764 ext4_msg(sb, KERN_ERR, 4765 "Number of reserved GDT blocks insanely large: %d", 4766 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4767 return -EINVAL; 4768 } 4769 /* 4770 * Test whether we have more sectors than will fit in sector_t, 4771 * and whether the max offset is addressable by the page cache. 4772 */ 4773 err = generic_check_addressable(sb->s_blocksize_bits, 4774 ext4_blocks_count(es)); 4775 if (err) { 4776 ext4_msg(sb, KERN_ERR, "filesystem" 4777 " too large to mount safely on this system"); 4778 return err; 4779 } 4780 4781 /* check blocks count against device size */ 4782 blocks_count = sb_bdev_nr_blocks(sb); 4783 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4784 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4785 "exceeds size of device (%llu blocks)", 4786 ext4_blocks_count(es), blocks_count); 4787 return -EINVAL; 4788 } 4789 4790 /* 4791 * It makes no sense for the first data block to be beyond the end 4792 * of the filesystem. 4793 */ 4794 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4795 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4796 "block %u is beyond end of filesystem (%llu)", 4797 le32_to_cpu(es->s_first_data_block), 4798 ext4_blocks_count(es)); 4799 return -EINVAL; 4800 } 4801 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4802 (sbi->s_cluster_ratio == 1)) { 4803 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4804 "block is 0 with a 1k block and cluster size"); 4805 return -EINVAL; 4806 } 4807 4808 blocks_count = (ext4_blocks_count(es) - 4809 le32_to_cpu(es->s_first_data_block) + 4810 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4811 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4812 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4813 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4814 "(block count %llu, first data block %u, " 4815 "blocks per group %lu)", blocks_count, 4816 ext4_blocks_count(es), 4817 le32_to_cpu(es->s_first_data_block), 4818 EXT4_BLOCKS_PER_GROUP(sb)); 4819 return -EINVAL; 4820 } 4821 sbi->s_groups_count = blocks_count; 4822 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4823 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4824 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4825 le32_to_cpu(es->s_inodes_count)) { 4826 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4827 le32_to_cpu(es->s_inodes_count), 4828 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4829 return -EINVAL; 4830 } 4831 4832 return 0; 4833 } 4834 4835 static int ext4_group_desc_init(struct super_block *sb, 4836 struct ext4_super_block *es, 4837 ext4_fsblk_t logical_sb_block, 4838 ext4_group_t *first_not_zeroed) 4839 { 4840 struct ext4_sb_info *sbi = EXT4_SB(sb); 4841 unsigned int db_count; 4842 ext4_fsblk_t block; 4843 int i; 4844 4845 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4846 EXT4_DESC_PER_BLOCK(sb); 4847 if (ext4_has_feature_meta_bg(sb)) { 4848 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4849 ext4_msg(sb, KERN_WARNING, 4850 "first meta block group too large: %u " 4851 "(group descriptor block count %u)", 4852 le32_to_cpu(es->s_first_meta_bg), db_count); 4853 return -EINVAL; 4854 } 4855 } 4856 rcu_assign_pointer(sbi->s_group_desc, 4857 kvmalloc_array(db_count, 4858 sizeof(struct buffer_head *), 4859 GFP_KERNEL)); 4860 if (sbi->s_group_desc == NULL) { 4861 ext4_msg(sb, KERN_ERR, "not enough memory"); 4862 return -ENOMEM; 4863 } 4864 4865 bgl_lock_init(sbi->s_blockgroup_lock); 4866 4867 /* Pre-read the descriptors into the buffer cache */ 4868 for (i = 0; i < db_count; i++) { 4869 block = descriptor_loc(sb, logical_sb_block, i); 4870 ext4_sb_breadahead_unmovable(sb, block); 4871 } 4872 4873 for (i = 0; i < db_count; i++) { 4874 struct buffer_head *bh; 4875 4876 block = descriptor_loc(sb, logical_sb_block, i); 4877 bh = ext4_sb_bread_unmovable(sb, block); 4878 if (IS_ERR(bh)) { 4879 ext4_msg(sb, KERN_ERR, 4880 "can't read group descriptor %d", i); 4881 sbi->s_gdb_count = i; 4882 return PTR_ERR(bh); 4883 } 4884 rcu_read_lock(); 4885 rcu_dereference(sbi->s_group_desc)[i] = bh; 4886 rcu_read_unlock(); 4887 } 4888 sbi->s_gdb_count = db_count; 4889 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4890 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4891 return -EFSCORRUPTED; 4892 } 4893 4894 return 0; 4895 } 4896 4897 static int ext4_load_and_init_journal(struct super_block *sb, 4898 struct ext4_super_block *es, 4899 struct ext4_fs_context *ctx) 4900 { 4901 struct ext4_sb_info *sbi = EXT4_SB(sb); 4902 int err; 4903 4904 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4905 if (err) 4906 return err; 4907 4908 if (ext4_has_feature_64bit(sb) && 4909 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4910 JBD2_FEATURE_INCOMPAT_64BIT)) { 4911 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4912 goto out; 4913 } 4914 4915 if (!set_journal_csum_feature_set(sb)) { 4916 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4917 "feature set"); 4918 goto out; 4919 } 4920 4921 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4922 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4923 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4924 ext4_msg(sb, KERN_ERR, 4925 "Failed to set fast commit journal feature"); 4926 goto out; 4927 } 4928 4929 /* We have now updated the journal if required, so we can 4930 * validate the data journaling mode. */ 4931 switch (test_opt(sb, DATA_FLAGS)) { 4932 case 0: 4933 /* No mode set, assume a default based on the journal 4934 * capabilities: ORDERED_DATA if the journal can 4935 * cope, else JOURNAL_DATA 4936 */ 4937 if (jbd2_journal_check_available_features 4938 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4939 set_opt(sb, ORDERED_DATA); 4940 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4941 } else { 4942 set_opt(sb, JOURNAL_DATA); 4943 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4944 } 4945 break; 4946 4947 case EXT4_MOUNT_ORDERED_DATA: 4948 case EXT4_MOUNT_WRITEBACK_DATA: 4949 if (!jbd2_journal_check_available_features 4950 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4951 ext4_msg(sb, KERN_ERR, "Journal does not support " 4952 "requested data journaling mode"); 4953 goto out; 4954 } 4955 break; 4956 default: 4957 break; 4958 } 4959 4960 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4961 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4962 ext4_msg(sb, KERN_ERR, "can't mount with " 4963 "journal_async_commit in data=ordered mode"); 4964 goto out; 4965 } 4966 4967 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4968 4969 sbi->s_journal->j_submit_inode_data_buffers = 4970 ext4_journal_submit_inode_data_buffers; 4971 sbi->s_journal->j_finish_inode_data_buffers = 4972 ext4_journal_finish_inode_data_buffers; 4973 4974 return 0; 4975 4976 out: 4977 ext4_journal_destroy(sbi, sbi->s_journal); 4978 return -EINVAL; 4979 } 4980 4981 static int ext4_check_journal_data_mode(struct super_block *sb) 4982 { 4983 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4984 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4985 "data=journal disables delayed allocation, " 4986 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 4987 /* can't mount with both data=journal and dioread_nolock. */ 4988 clear_opt(sb, DIOREAD_NOLOCK); 4989 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4990 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4991 ext4_msg(sb, KERN_ERR, "can't mount with " 4992 "both data=journal and delalloc"); 4993 return -EINVAL; 4994 } 4995 if (test_opt(sb, DAX_ALWAYS)) { 4996 ext4_msg(sb, KERN_ERR, "can't mount with " 4997 "both data=journal and dax"); 4998 return -EINVAL; 4999 } 5000 if (ext4_has_feature_encrypt(sb)) { 5001 ext4_msg(sb, KERN_WARNING, 5002 "encrypted files will use data=ordered " 5003 "instead of data journaling mode"); 5004 } 5005 if (test_opt(sb, DELALLOC)) 5006 clear_opt(sb, DELALLOC); 5007 } else { 5008 sb->s_iflags |= SB_I_CGROUPWB; 5009 } 5010 5011 return 0; 5012 } 5013 5014 static const char *ext4_has_journal_option(struct super_block *sb) 5015 { 5016 struct ext4_sb_info *sbi = EXT4_SB(sb); 5017 5018 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) 5019 return "journal_async_commit"; 5020 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) 5021 return "journal_checksum"; 5022 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 5023 return "commit="; 5024 if (EXT4_MOUNT_DATA_FLAGS & 5025 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) 5026 return "data="; 5027 if (test_opt(sb, DATA_ERR_ABORT)) 5028 return "data_err=abort"; 5029 return NULL; 5030 } 5031 5032 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 5033 int silent) 5034 { 5035 struct ext4_sb_info *sbi = EXT4_SB(sb); 5036 struct ext4_super_block *es; 5037 ext4_fsblk_t logical_sb_block; 5038 unsigned long offset = 0; 5039 struct buffer_head *bh; 5040 int ret = -EINVAL; 5041 int blocksize; 5042 5043 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 5044 if (!blocksize) { 5045 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 5046 return -EINVAL; 5047 } 5048 5049 /* 5050 * The ext4 superblock will not be buffer aligned for other than 1kB 5051 * block sizes. We need to calculate the offset from buffer start. 5052 */ 5053 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 5054 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5055 offset = do_div(logical_sb_block, blocksize); 5056 } else { 5057 logical_sb_block = sbi->s_sb_block; 5058 } 5059 5060 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5061 if (IS_ERR(bh)) { 5062 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 5063 return PTR_ERR(bh); 5064 } 5065 /* 5066 * Note: s_es must be initialized as soon as possible because 5067 * some ext4 macro-instructions depend on its value 5068 */ 5069 es = (struct ext4_super_block *) (bh->b_data + offset); 5070 sbi->s_es = es; 5071 sb->s_magic = le16_to_cpu(es->s_magic); 5072 if (sb->s_magic != EXT4_SUPER_MAGIC) { 5073 if (!silent) 5074 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5075 goto out; 5076 } 5077 5078 if (le32_to_cpu(es->s_log_block_size) > 5079 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5080 ext4_msg(sb, KERN_ERR, 5081 "Invalid log block size: %u", 5082 le32_to_cpu(es->s_log_block_size)); 5083 goto out; 5084 } 5085 if (le32_to_cpu(es->s_log_cluster_size) > 5086 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5087 ext4_msg(sb, KERN_ERR, 5088 "Invalid log cluster size: %u", 5089 le32_to_cpu(es->s_log_cluster_size)); 5090 goto out; 5091 } 5092 5093 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 5094 5095 /* 5096 * If the default block size is not the same as the real block size, 5097 * we need to reload it. 5098 */ 5099 if (sb->s_blocksize == blocksize) { 5100 *lsb = logical_sb_block; 5101 sbi->s_sbh = bh; 5102 return 0; 5103 } 5104 5105 /* 5106 * bh must be released before kill_bdev(), otherwise 5107 * it won't be freed and its page also. kill_bdev() 5108 * is called by sb_set_blocksize(). 5109 */ 5110 brelse(bh); 5111 /* Validate the filesystem blocksize */ 5112 if (!sb_set_blocksize(sb, blocksize)) { 5113 ext4_msg(sb, KERN_ERR, "bad block size %d", 5114 blocksize); 5115 bh = NULL; 5116 goto out; 5117 } 5118 5119 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5120 offset = do_div(logical_sb_block, blocksize); 5121 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5122 if (IS_ERR(bh)) { 5123 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5124 ret = PTR_ERR(bh); 5125 bh = NULL; 5126 goto out; 5127 } 5128 es = (struct ext4_super_block *)(bh->b_data + offset); 5129 sbi->s_es = es; 5130 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5131 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5132 goto out; 5133 } 5134 *lsb = logical_sb_block; 5135 sbi->s_sbh = bh; 5136 return 0; 5137 out: 5138 brelse(bh); 5139 return ret; 5140 } 5141 5142 static int ext4_hash_info_init(struct super_block *sb) 5143 { 5144 struct ext4_sb_info *sbi = EXT4_SB(sb); 5145 struct ext4_super_block *es = sbi->s_es; 5146 unsigned int i; 5147 5148 sbi->s_def_hash_version = es->s_def_hash_version; 5149 5150 if (sbi->s_def_hash_version > DX_HASH_LAST) { 5151 ext4_msg(sb, KERN_ERR, 5152 "Invalid default hash set in the superblock"); 5153 return -EINVAL; 5154 } else if (sbi->s_def_hash_version == DX_HASH_SIPHASH) { 5155 ext4_msg(sb, KERN_ERR, 5156 "SIPHASH is not a valid default hash value"); 5157 return -EINVAL; 5158 } 5159 5160 for (i = 0; i < 4; i++) 5161 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5162 5163 if (ext4_has_feature_dir_index(sb)) { 5164 i = le32_to_cpu(es->s_flags); 5165 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5166 sbi->s_hash_unsigned = 3; 5167 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5168 #ifdef __CHAR_UNSIGNED__ 5169 if (!sb_rdonly(sb)) 5170 es->s_flags |= 5171 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5172 sbi->s_hash_unsigned = 3; 5173 #else 5174 if (!sb_rdonly(sb)) 5175 es->s_flags |= 5176 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5177 #endif 5178 } 5179 } 5180 return 0; 5181 } 5182 5183 static int ext4_block_group_meta_init(struct super_block *sb, int silent) 5184 { 5185 struct ext4_sb_info *sbi = EXT4_SB(sb); 5186 struct ext4_super_block *es = sbi->s_es; 5187 int has_huge_files; 5188 5189 has_huge_files = ext4_has_feature_huge_file(sb); 5190 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5191 has_huge_files); 5192 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5193 5194 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5195 if (ext4_has_feature_64bit(sb)) { 5196 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5197 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5198 !is_power_of_2(sbi->s_desc_size)) { 5199 ext4_msg(sb, KERN_ERR, 5200 "unsupported descriptor size %lu", 5201 sbi->s_desc_size); 5202 return -EINVAL; 5203 } 5204 } else 5205 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5206 5207 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5208 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5209 5210 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5211 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5212 if (!silent) 5213 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5214 return -EINVAL; 5215 } 5216 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5217 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5218 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5219 sbi->s_inodes_per_group); 5220 return -EINVAL; 5221 } 5222 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5223 sbi->s_inodes_per_block; 5224 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5225 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5226 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5227 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5228 5229 return 0; 5230 } 5231 5232 /* 5233 * It's hard to get stripe aligned blocks if stripe is not aligned with 5234 * cluster, just disable stripe and alert user to simplify code and avoid 5235 * stripe aligned allocation which will rarely succeed. 5236 */ 5237 static bool ext4_is_stripe_incompatible(struct super_block *sb, unsigned long stripe) 5238 { 5239 struct ext4_sb_info *sbi = EXT4_SB(sb); 5240 return (stripe > 0 && sbi->s_cluster_ratio > 1 && 5241 stripe % sbi->s_cluster_ratio != 0); 5242 } 5243 5244 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5245 { 5246 struct ext4_super_block *es = NULL; 5247 struct ext4_sb_info *sbi = EXT4_SB(sb); 5248 ext4_fsblk_t logical_sb_block; 5249 struct inode *root; 5250 int needs_recovery; 5251 int err; 5252 ext4_group_t first_not_zeroed; 5253 struct ext4_fs_context *ctx = fc->fs_private; 5254 int silent = fc->sb_flags & SB_SILENT; 5255 5256 /* Set defaults for the variables that will be set during parsing */ 5257 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5258 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5259 5260 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5261 sbi->s_sectors_written_start = 5262 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5263 5264 err = ext4_load_super(sb, &logical_sb_block, silent); 5265 if (err) 5266 goto out_fail; 5267 5268 es = sbi->s_es; 5269 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5270 5271 err = ext4_init_metadata_csum(sb, es); 5272 if (err) 5273 goto failed_mount; 5274 5275 ext4_set_def_opts(sb, es); 5276 5277 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5278 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5279 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5280 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5281 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5282 sbi->s_sb_update_kb = EXT4_DEF_SB_UPDATE_INTERVAL_KB; 5283 sbi->s_sb_update_sec = EXT4_DEF_SB_UPDATE_INTERVAL_SEC; 5284 5285 /* 5286 * set default s_li_wait_mult for lazyinit, for the case there is 5287 * no mount option specified. 5288 */ 5289 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5290 5291 err = ext4_inode_info_init(sb, es); 5292 if (err) 5293 goto failed_mount; 5294 5295 err = parse_apply_sb_mount_options(sb, ctx); 5296 if (err < 0) 5297 goto failed_mount; 5298 5299 sbi->s_def_mount_opt = sbi->s_mount_opt; 5300 sbi->s_def_mount_opt2 = sbi->s_mount_opt2; 5301 5302 err = ext4_check_opt_consistency(fc, sb); 5303 if (err < 0) 5304 goto failed_mount; 5305 5306 ext4_apply_options(fc, sb); 5307 5308 err = ext4_encoding_init(sb, es); 5309 if (err) 5310 goto failed_mount; 5311 5312 err = ext4_check_journal_data_mode(sb); 5313 if (err) 5314 goto failed_mount; 5315 5316 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5317 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5318 5319 /* i_version is always enabled now */ 5320 sb->s_flags |= SB_I_VERSION; 5321 5322 /* HSM events are allowed by default. */ 5323 sb->s_iflags |= SB_I_ALLOW_HSM; 5324 5325 err = ext4_check_feature_compatibility(sb, es, silent); 5326 if (err) 5327 goto failed_mount; 5328 5329 err = ext4_block_group_meta_init(sb, silent); 5330 if (err) 5331 goto failed_mount; 5332 5333 err = ext4_hash_info_init(sb); 5334 if (err) 5335 goto failed_mount; 5336 5337 err = ext4_handle_clustersize(sb); 5338 if (err) 5339 goto failed_mount; 5340 5341 err = ext4_check_geometry(sb, es); 5342 if (err) 5343 goto failed_mount; 5344 5345 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5346 spin_lock_init(&sbi->s_error_lock); 5347 INIT_WORK(&sbi->s_sb_upd_work, update_super_work); 5348 5349 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5350 if (err) 5351 goto failed_mount3; 5352 5353 err = ext4_es_register_shrinker(sbi); 5354 if (err) 5355 goto failed_mount3; 5356 5357 sbi->s_stripe = ext4_get_stripe_size(sbi); 5358 if (ext4_is_stripe_incompatible(sb, sbi->s_stripe)) { 5359 ext4_msg(sb, KERN_WARNING, 5360 "stripe (%lu) is not aligned with cluster size (%u), " 5361 "stripe is disabled", 5362 sbi->s_stripe, sbi->s_cluster_ratio); 5363 sbi->s_stripe = 0; 5364 } 5365 sbi->s_extent_max_zeroout_kb = 32; 5366 5367 /* 5368 * set up enough so that it can read an inode 5369 */ 5370 sb->s_op = &ext4_sops; 5371 sb->s_export_op = &ext4_export_ops; 5372 sb->s_xattr = ext4_xattr_handlers; 5373 #ifdef CONFIG_FS_ENCRYPTION 5374 sb->s_cop = &ext4_cryptops; 5375 #endif 5376 #ifdef CONFIG_FS_VERITY 5377 sb->s_vop = &ext4_verityops; 5378 #endif 5379 #ifdef CONFIG_QUOTA 5380 sb->dq_op = &ext4_quota_operations; 5381 if (ext4_has_feature_quota(sb)) 5382 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5383 else 5384 sb->s_qcop = &ext4_qctl_operations; 5385 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5386 #endif 5387 super_set_uuid(sb, es->s_uuid, sizeof(es->s_uuid)); 5388 super_set_sysfs_name_bdev(sb); 5389 5390 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5391 mutex_init(&sbi->s_orphan_lock); 5392 5393 spin_lock_init(&sbi->s_bdev_wb_lock); 5394 5395 ext4_atomic_write_init(sb); 5396 ext4_fast_commit_init(sb); 5397 5398 sb->s_root = NULL; 5399 5400 needs_recovery = (es->s_last_orphan != 0 || 5401 ext4_has_feature_orphan_present(sb) || 5402 ext4_has_feature_journal_needs_recovery(sb)); 5403 5404 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) { 5405 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)); 5406 if (err) 5407 goto failed_mount3a; 5408 } 5409 5410 err = -EINVAL; 5411 /* 5412 * The first inode we look at is the journal inode. Don't try 5413 * root first: it may be modified in the journal! 5414 */ 5415 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5416 err = ext4_load_and_init_journal(sb, es, ctx); 5417 if (err) 5418 goto failed_mount3a; 5419 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5420 ext4_has_feature_journal_needs_recovery(sb)) { 5421 ext4_msg(sb, KERN_ERR, "required journal recovery " 5422 "suppressed and not mounted read-only"); 5423 goto failed_mount3a; 5424 } else { 5425 const char *journal_option; 5426 5427 /* Nojournal mode, all journal mount options are illegal */ 5428 journal_option = ext4_has_journal_option(sb); 5429 if (journal_option != NULL) { 5430 ext4_msg(sb, KERN_ERR, 5431 "can't mount with %s, fs mounted w/o journal", 5432 journal_option); 5433 goto failed_mount3a; 5434 } 5435 5436 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5437 clear_opt(sb, JOURNAL_CHECKSUM); 5438 clear_opt(sb, DATA_FLAGS); 5439 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5440 sbi->s_journal = NULL; 5441 needs_recovery = 0; 5442 } 5443 5444 if (!test_opt(sb, NO_MBCACHE)) { 5445 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5446 if (!sbi->s_ea_block_cache) { 5447 ext4_msg(sb, KERN_ERR, 5448 "Failed to create ea_block_cache"); 5449 err = -EINVAL; 5450 goto failed_mount_wq; 5451 } 5452 5453 if (ext4_has_feature_ea_inode(sb)) { 5454 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5455 if (!sbi->s_ea_inode_cache) { 5456 ext4_msg(sb, KERN_ERR, 5457 "Failed to create ea_inode_cache"); 5458 err = -EINVAL; 5459 goto failed_mount_wq; 5460 } 5461 } 5462 } 5463 5464 /* 5465 * Get the # of file system overhead blocks from the 5466 * superblock if present. 5467 */ 5468 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5469 /* ignore the precalculated value if it is ridiculous */ 5470 if (sbi->s_overhead > ext4_blocks_count(es)) 5471 sbi->s_overhead = 0; 5472 /* 5473 * If the bigalloc feature is not enabled recalculating the 5474 * overhead doesn't take long, so we might as well just redo 5475 * it to make sure we are using the correct value. 5476 */ 5477 if (!ext4_has_feature_bigalloc(sb)) 5478 sbi->s_overhead = 0; 5479 if (sbi->s_overhead == 0) { 5480 err = ext4_calculate_overhead(sb); 5481 if (err) 5482 goto failed_mount_wq; 5483 } 5484 5485 /* 5486 * The maximum number of concurrent works can be high and 5487 * concurrency isn't really necessary. Limit it to 1. 5488 */ 5489 EXT4_SB(sb)->rsv_conversion_wq = 5490 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5491 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5492 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5493 err = -ENOMEM; 5494 goto failed_mount4; 5495 } 5496 5497 /* 5498 * The jbd2_journal_load will have done any necessary log recovery, 5499 * so we can safely mount the rest of the filesystem now. 5500 */ 5501 5502 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5503 if (IS_ERR(root)) { 5504 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5505 err = PTR_ERR(root); 5506 root = NULL; 5507 goto failed_mount4; 5508 } 5509 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5510 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5511 iput(root); 5512 err = -EFSCORRUPTED; 5513 goto failed_mount4; 5514 } 5515 5516 generic_set_sb_d_ops(sb); 5517 sb->s_root = d_make_root(root); 5518 if (!sb->s_root) { 5519 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5520 err = -ENOMEM; 5521 goto failed_mount4; 5522 } 5523 5524 err = ext4_setup_super(sb, es, sb_rdonly(sb)); 5525 if (err == -EROFS) { 5526 sb->s_flags |= SB_RDONLY; 5527 } else if (err) 5528 goto failed_mount4a; 5529 5530 ext4_set_resv_clusters(sb); 5531 5532 if (test_opt(sb, BLOCK_VALIDITY)) { 5533 err = ext4_setup_system_zone(sb); 5534 if (err) { 5535 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5536 "zone (%d)", err); 5537 goto failed_mount4a; 5538 } 5539 } 5540 ext4_fc_replay_cleanup(sb); 5541 5542 ext4_ext_init(sb); 5543 5544 /* 5545 * Enable optimize_scan if number of groups is > threshold. This can be 5546 * turned off by passing "mb_optimize_scan=0". This can also be 5547 * turned on forcefully by passing "mb_optimize_scan=1". 5548 */ 5549 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5550 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5551 set_opt2(sb, MB_OPTIMIZE_SCAN); 5552 else 5553 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5554 } 5555 5556 err = ext4_mb_init(sb); 5557 if (err) { 5558 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5559 err); 5560 goto failed_mount5; 5561 } 5562 5563 /* 5564 * We can only set up the journal commit callback once 5565 * mballoc is initialized 5566 */ 5567 if (sbi->s_journal) 5568 sbi->s_journal->j_commit_callback = 5569 ext4_journal_commit_callback; 5570 5571 err = ext4_percpu_param_init(sbi); 5572 if (err) 5573 goto failed_mount6; 5574 5575 if (ext4_has_feature_flex_bg(sb)) 5576 if (!ext4_fill_flex_info(sb)) { 5577 ext4_msg(sb, KERN_ERR, 5578 "unable to initialize " 5579 "flex_bg meta info!"); 5580 err = -ENOMEM; 5581 goto failed_mount6; 5582 } 5583 5584 err = ext4_register_li_request(sb, first_not_zeroed); 5585 if (err) 5586 goto failed_mount6; 5587 5588 err = ext4_init_orphan_info(sb); 5589 if (err) 5590 goto failed_mount7; 5591 #ifdef CONFIG_QUOTA 5592 /* Enable quota usage during mount. */ 5593 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5594 err = ext4_enable_quotas(sb); 5595 if (err) 5596 goto failed_mount8; 5597 } 5598 #endif /* CONFIG_QUOTA */ 5599 5600 /* 5601 * Save the original bdev mapping's wb_err value which could be 5602 * used to detect the metadata async write error. 5603 */ 5604 errseq_check_and_advance(&sb->s_bdev->bd_mapping->wb_err, 5605 &sbi->s_bdev_wb_err); 5606 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5607 ext4_orphan_cleanup(sb, es); 5608 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5609 /* 5610 * Update the checksum after updating free space/inode counters and 5611 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5612 * checksum in the buffer cache until it is written out and 5613 * e2fsprogs programs trying to open a file system immediately 5614 * after it is mounted can fail. 5615 */ 5616 ext4_superblock_csum_set(sb); 5617 if (needs_recovery) { 5618 ext4_msg(sb, KERN_INFO, "recovery complete"); 5619 err = ext4_mark_recovery_complete(sb, es); 5620 if (err) 5621 goto failed_mount9; 5622 } 5623 5624 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) { 5625 ext4_msg(sb, KERN_WARNING, 5626 "mounting with \"discard\" option, but the device does not support discard"); 5627 clear_opt(sb, DISCARD); 5628 } 5629 5630 if (es->s_error_count) 5631 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5632 5633 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5634 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5635 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5636 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5637 atomic_set(&sbi->s_warning_count, 0); 5638 atomic_set(&sbi->s_msg_count, 0); 5639 5640 /* Register sysfs after all initializations are complete. */ 5641 err = ext4_register_sysfs(sb); 5642 if (err) 5643 goto failed_mount9; 5644 5645 return 0; 5646 5647 failed_mount9: 5648 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 5649 failed_mount8: __maybe_unused 5650 ext4_release_orphan_info(sb); 5651 failed_mount7: 5652 ext4_unregister_li_request(sb); 5653 failed_mount6: 5654 ext4_mb_release(sb); 5655 ext4_flex_groups_free(sbi); 5656 ext4_percpu_param_destroy(sbi); 5657 failed_mount5: 5658 ext4_ext_release(sb); 5659 ext4_release_system_zone(sb); 5660 failed_mount4a: 5661 dput(sb->s_root); 5662 sb->s_root = NULL; 5663 failed_mount4: 5664 ext4_msg(sb, KERN_ERR, "mount failed"); 5665 if (EXT4_SB(sb)->rsv_conversion_wq) 5666 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5667 failed_mount_wq: 5668 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5669 sbi->s_ea_inode_cache = NULL; 5670 5671 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5672 sbi->s_ea_block_cache = NULL; 5673 5674 if (sbi->s_journal) { 5675 ext4_journal_destroy(sbi, sbi->s_journal); 5676 } 5677 failed_mount3a: 5678 ext4_es_unregister_shrinker(sbi); 5679 failed_mount3: 5680 /* flush s_sb_upd_work before sbi destroy */ 5681 flush_work(&sbi->s_sb_upd_work); 5682 ext4_stop_mmpd(sbi); 5683 del_timer_sync(&sbi->s_err_report); 5684 ext4_group_desc_free(sbi); 5685 failed_mount: 5686 #if IS_ENABLED(CONFIG_UNICODE) 5687 utf8_unload(sb->s_encoding); 5688 #endif 5689 5690 #ifdef CONFIG_QUOTA 5691 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++) 5692 kfree(get_qf_name(sb, sbi, i)); 5693 #endif 5694 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5695 brelse(sbi->s_sbh); 5696 if (sbi->s_journal_bdev_file) { 5697 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 5698 bdev_fput(sbi->s_journal_bdev_file); 5699 } 5700 out_fail: 5701 invalidate_bdev(sb->s_bdev); 5702 sb->s_fs_info = NULL; 5703 return err; 5704 } 5705 5706 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5707 { 5708 struct ext4_fs_context *ctx = fc->fs_private; 5709 struct ext4_sb_info *sbi; 5710 const char *descr; 5711 int ret; 5712 5713 sbi = ext4_alloc_sbi(sb); 5714 if (!sbi) 5715 return -ENOMEM; 5716 5717 fc->s_fs_info = sbi; 5718 5719 /* Cleanup superblock name */ 5720 strreplace(sb->s_id, '/', '!'); 5721 5722 sbi->s_sb_block = 1; /* Default super block location */ 5723 if (ctx->spec & EXT4_SPEC_s_sb_block) 5724 sbi->s_sb_block = ctx->s_sb_block; 5725 5726 ret = __ext4_fill_super(fc, sb); 5727 if (ret < 0) 5728 goto free_sbi; 5729 5730 if (sbi->s_journal) { 5731 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5732 descr = " journalled data mode"; 5733 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5734 descr = " ordered data mode"; 5735 else 5736 descr = " writeback data mode"; 5737 } else 5738 descr = "out journal"; 5739 5740 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5741 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. " 5742 "Quota mode: %s.", &sb->s_uuid, 5743 sb_rdonly(sb) ? "ro" : "r/w", descr, 5744 ext4_quota_mode(sb)); 5745 5746 /* Update the s_overhead_clusters if necessary */ 5747 ext4_update_overhead(sb, false); 5748 return 0; 5749 5750 free_sbi: 5751 ext4_free_sbi(sbi); 5752 fc->s_fs_info = NULL; 5753 return ret; 5754 } 5755 5756 static int ext4_get_tree(struct fs_context *fc) 5757 { 5758 return get_tree_bdev(fc, ext4_fill_super); 5759 } 5760 5761 /* 5762 * Setup any per-fs journal parameters now. We'll do this both on 5763 * initial mount, once the journal has been initialised but before we've 5764 * done any recovery; and again on any subsequent remount. 5765 */ 5766 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5767 { 5768 struct ext4_sb_info *sbi = EXT4_SB(sb); 5769 5770 journal->j_commit_interval = sbi->s_commit_interval; 5771 journal->j_min_batch_time = sbi->s_min_batch_time; 5772 journal->j_max_batch_time = sbi->s_max_batch_time; 5773 ext4_fc_init(sb, journal); 5774 5775 write_lock(&journal->j_state_lock); 5776 if (test_opt(sb, BARRIER)) 5777 journal->j_flags |= JBD2_BARRIER; 5778 else 5779 journal->j_flags &= ~JBD2_BARRIER; 5780 /* 5781 * Always enable journal cycle record option, letting the journal 5782 * records log transactions continuously between each mount. 5783 */ 5784 journal->j_flags |= JBD2_CYCLE_RECORD; 5785 write_unlock(&journal->j_state_lock); 5786 } 5787 5788 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5789 unsigned int journal_inum) 5790 { 5791 struct inode *journal_inode; 5792 5793 /* 5794 * Test for the existence of a valid inode on disk. Bad things 5795 * happen if we iget() an unused inode, as the subsequent iput() 5796 * will try to delete it. 5797 */ 5798 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5799 if (IS_ERR(journal_inode)) { 5800 ext4_msg(sb, KERN_ERR, "no journal found"); 5801 return ERR_CAST(journal_inode); 5802 } 5803 if (!journal_inode->i_nlink) { 5804 make_bad_inode(journal_inode); 5805 iput(journal_inode); 5806 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5807 return ERR_PTR(-EFSCORRUPTED); 5808 } 5809 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) { 5810 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5811 iput(journal_inode); 5812 return ERR_PTR(-EFSCORRUPTED); 5813 } 5814 5815 ext4_debug("Journal inode found at %p: %lld bytes\n", 5816 journal_inode, journal_inode->i_size); 5817 return journal_inode; 5818 } 5819 5820 static int ext4_journal_bmap(journal_t *journal, sector_t *block) 5821 { 5822 struct ext4_map_blocks map; 5823 int ret; 5824 5825 if (journal->j_inode == NULL) 5826 return 0; 5827 5828 map.m_lblk = *block; 5829 map.m_len = 1; 5830 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0); 5831 if (ret <= 0) { 5832 ext4_msg(journal->j_inode->i_sb, KERN_CRIT, 5833 "journal bmap failed: block %llu ret %d\n", 5834 *block, ret); 5835 jbd2_journal_abort(journal, ret ? ret : -EIO); 5836 return ret; 5837 } 5838 *block = map.m_pblk; 5839 return 0; 5840 } 5841 5842 static journal_t *ext4_open_inode_journal(struct super_block *sb, 5843 unsigned int journal_inum) 5844 { 5845 struct inode *journal_inode; 5846 journal_t *journal; 5847 5848 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5849 if (IS_ERR(journal_inode)) 5850 return ERR_CAST(journal_inode); 5851 5852 journal = jbd2_journal_init_inode(journal_inode); 5853 if (IS_ERR(journal)) { 5854 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5855 iput(journal_inode); 5856 return ERR_CAST(journal); 5857 } 5858 journal->j_private = sb; 5859 journal->j_bmap = ext4_journal_bmap; 5860 ext4_init_journal_params(sb, journal); 5861 return journal; 5862 } 5863 5864 static struct file *ext4_get_journal_blkdev(struct super_block *sb, 5865 dev_t j_dev, ext4_fsblk_t *j_start, 5866 ext4_fsblk_t *j_len) 5867 { 5868 struct buffer_head *bh; 5869 struct block_device *bdev; 5870 struct file *bdev_file; 5871 int hblock, blocksize; 5872 ext4_fsblk_t sb_block; 5873 unsigned long offset; 5874 struct ext4_super_block *es; 5875 int errno; 5876 5877 bdev_file = bdev_file_open_by_dev(j_dev, 5878 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES, 5879 sb, &fs_holder_ops); 5880 if (IS_ERR(bdev_file)) { 5881 ext4_msg(sb, KERN_ERR, 5882 "failed to open journal device unknown-block(%u,%u) %ld", 5883 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev_file)); 5884 return bdev_file; 5885 } 5886 5887 bdev = file_bdev(bdev_file); 5888 blocksize = sb->s_blocksize; 5889 hblock = bdev_logical_block_size(bdev); 5890 if (blocksize < hblock) { 5891 ext4_msg(sb, KERN_ERR, 5892 "blocksize too small for journal device"); 5893 errno = -EINVAL; 5894 goto out_bdev; 5895 } 5896 5897 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5898 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5899 set_blocksize(bdev_file, blocksize); 5900 bh = __bread(bdev, sb_block, blocksize); 5901 if (!bh) { 5902 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5903 "external journal"); 5904 errno = -EINVAL; 5905 goto out_bdev; 5906 } 5907 5908 es = (struct ext4_super_block *) (bh->b_data + offset); 5909 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5910 !(le32_to_cpu(es->s_feature_incompat) & 5911 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5912 ext4_msg(sb, KERN_ERR, "external journal has bad superblock"); 5913 errno = -EFSCORRUPTED; 5914 goto out_bh; 5915 } 5916 5917 if ((le32_to_cpu(es->s_feature_ro_compat) & 5918 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5919 es->s_checksum != ext4_superblock_csum(sb, es)) { 5920 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock"); 5921 errno = -EFSCORRUPTED; 5922 goto out_bh; 5923 } 5924 5925 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5926 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5927 errno = -EFSCORRUPTED; 5928 goto out_bh; 5929 } 5930 5931 *j_start = sb_block + 1; 5932 *j_len = ext4_blocks_count(es); 5933 brelse(bh); 5934 return bdev_file; 5935 5936 out_bh: 5937 brelse(bh); 5938 out_bdev: 5939 bdev_fput(bdev_file); 5940 return ERR_PTR(errno); 5941 } 5942 5943 static journal_t *ext4_open_dev_journal(struct super_block *sb, 5944 dev_t j_dev) 5945 { 5946 journal_t *journal; 5947 ext4_fsblk_t j_start; 5948 ext4_fsblk_t j_len; 5949 struct file *bdev_file; 5950 int errno = 0; 5951 5952 bdev_file = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len); 5953 if (IS_ERR(bdev_file)) 5954 return ERR_CAST(bdev_file); 5955 5956 journal = jbd2_journal_init_dev(file_bdev(bdev_file), sb->s_bdev, j_start, 5957 j_len, sb->s_blocksize); 5958 if (IS_ERR(journal)) { 5959 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5960 errno = PTR_ERR(journal); 5961 goto out_bdev; 5962 } 5963 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5964 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5965 "user (unsupported) - %d", 5966 be32_to_cpu(journal->j_superblock->s_nr_users)); 5967 errno = -EINVAL; 5968 goto out_journal; 5969 } 5970 journal->j_private = sb; 5971 EXT4_SB(sb)->s_journal_bdev_file = bdev_file; 5972 ext4_init_journal_params(sb, journal); 5973 return journal; 5974 5975 out_journal: 5976 ext4_journal_destroy(EXT4_SB(sb), journal); 5977 out_bdev: 5978 bdev_fput(bdev_file); 5979 return ERR_PTR(errno); 5980 } 5981 5982 static int ext4_load_journal(struct super_block *sb, 5983 struct ext4_super_block *es, 5984 unsigned long journal_devnum) 5985 { 5986 journal_t *journal; 5987 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5988 dev_t journal_dev; 5989 int err = 0; 5990 int really_read_only; 5991 int journal_dev_ro; 5992 5993 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5994 return -EFSCORRUPTED; 5995 5996 if (journal_devnum && 5997 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5998 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5999 "numbers have changed"); 6000 journal_dev = new_decode_dev(journal_devnum); 6001 } else 6002 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 6003 6004 if (journal_inum && journal_dev) { 6005 ext4_msg(sb, KERN_ERR, 6006 "filesystem has both journal inode and journal device!"); 6007 return -EINVAL; 6008 } 6009 6010 if (journal_inum) { 6011 journal = ext4_open_inode_journal(sb, journal_inum); 6012 if (IS_ERR(journal)) 6013 return PTR_ERR(journal); 6014 } else { 6015 journal = ext4_open_dev_journal(sb, journal_dev); 6016 if (IS_ERR(journal)) 6017 return PTR_ERR(journal); 6018 } 6019 6020 journal_dev_ro = bdev_read_only(journal->j_dev); 6021 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 6022 6023 if (journal_dev_ro && !sb_rdonly(sb)) { 6024 ext4_msg(sb, KERN_ERR, 6025 "journal device read-only, try mounting with '-o ro'"); 6026 err = -EROFS; 6027 goto err_out; 6028 } 6029 6030 /* 6031 * Are we loading a blank journal or performing recovery after a 6032 * crash? For recovery, we need to check in advance whether we 6033 * can get read-write access to the device. 6034 */ 6035 if (ext4_has_feature_journal_needs_recovery(sb)) { 6036 if (sb_rdonly(sb)) { 6037 ext4_msg(sb, KERN_INFO, "INFO: recovery " 6038 "required on readonly filesystem"); 6039 if (really_read_only) { 6040 ext4_msg(sb, KERN_ERR, "write access " 6041 "unavailable, cannot proceed " 6042 "(try mounting with noload)"); 6043 err = -EROFS; 6044 goto err_out; 6045 } 6046 ext4_msg(sb, KERN_INFO, "write access will " 6047 "be enabled during recovery"); 6048 } 6049 } 6050 6051 if (!(journal->j_flags & JBD2_BARRIER)) 6052 ext4_msg(sb, KERN_INFO, "barriers disabled"); 6053 6054 if (!ext4_has_feature_journal_needs_recovery(sb)) 6055 err = jbd2_journal_wipe(journal, !really_read_only); 6056 if (!err) { 6057 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 6058 __le16 orig_state; 6059 bool changed = false; 6060 6061 if (save) 6062 memcpy(save, ((char *) es) + 6063 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 6064 err = jbd2_journal_load(journal); 6065 if (save && memcmp(((char *) es) + EXT4_S_ERR_START, 6066 save, EXT4_S_ERR_LEN)) { 6067 memcpy(((char *) es) + EXT4_S_ERR_START, 6068 save, EXT4_S_ERR_LEN); 6069 changed = true; 6070 } 6071 kfree(save); 6072 orig_state = es->s_state; 6073 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state & 6074 EXT4_ERROR_FS); 6075 if (orig_state != es->s_state) 6076 changed = true; 6077 /* Write out restored error information to the superblock */ 6078 if (changed && !really_read_only) { 6079 int err2; 6080 err2 = ext4_commit_super(sb); 6081 err = err ? : err2; 6082 } 6083 } 6084 6085 if (err) { 6086 ext4_msg(sb, KERN_ERR, "error loading journal"); 6087 goto err_out; 6088 } 6089 6090 EXT4_SB(sb)->s_journal = journal; 6091 err = ext4_clear_journal_err(sb, es); 6092 if (err) { 6093 ext4_journal_destroy(EXT4_SB(sb), journal); 6094 return err; 6095 } 6096 6097 if (!really_read_only && journal_devnum && 6098 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6099 es->s_journal_dev = cpu_to_le32(journal_devnum); 6100 ext4_commit_super(sb); 6101 } 6102 if (!really_read_only && journal_inum && 6103 journal_inum != le32_to_cpu(es->s_journal_inum)) { 6104 es->s_journal_inum = cpu_to_le32(journal_inum); 6105 ext4_commit_super(sb); 6106 } 6107 6108 return 0; 6109 6110 err_out: 6111 ext4_journal_destroy(EXT4_SB(sb), journal); 6112 return err; 6113 } 6114 6115 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 6116 static void ext4_update_super(struct super_block *sb) 6117 { 6118 struct ext4_sb_info *sbi = EXT4_SB(sb); 6119 struct ext4_super_block *es = sbi->s_es; 6120 struct buffer_head *sbh = sbi->s_sbh; 6121 6122 lock_buffer(sbh); 6123 /* 6124 * If the file system is mounted read-only, don't update the 6125 * superblock write time. This avoids updating the superblock 6126 * write time when we are mounting the root file system 6127 * read/only but we need to replay the journal; at that point, 6128 * for people who are east of GMT and who make their clock 6129 * tick in localtime for Windows bug-for-bug compatibility, 6130 * the clock is set in the future, and this will cause e2fsck 6131 * to complain and force a full file system check. 6132 */ 6133 if (!sb_rdonly(sb)) 6134 ext4_update_tstamp(es, s_wtime); 6135 es->s_kbytes_written = 6136 cpu_to_le64(sbi->s_kbytes_written + 6137 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 6138 sbi->s_sectors_written_start) >> 1)); 6139 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 6140 ext4_free_blocks_count_set(es, 6141 EXT4_C2B(sbi, percpu_counter_sum_positive( 6142 &sbi->s_freeclusters_counter))); 6143 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 6144 es->s_free_inodes_count = 6145 cpu_to_le32(percpu_counter_sum_positive( 6146 &sbi->s_freeinodes_counter)); 6147 /* Copy error information to the on-disk superblock */ 6148 spin_lock(&sbi->s_error_lock); 6149 if (sbi->s_add_error_count > 0) { 6150 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6151 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6152 __ext4_update_tstamp(&es->s_first_error_time, 6153 &es->s_first_error_time_hi, 6154 sbi->s_first_error_time); 6155 strtomem_pad(es->s_first_error_func, 6156 sbi->s_first_error_func, 0); 6157 es->s_first_error_line = 6158 cpu_to_le32(sbi->s_first_error_line); 6159 es->s_first_error_ino = 6160 cpu_to_le32(sbi->s_first_error_ino); 6161 es->s_first_error_block = 6162 cpu_to_le64(sbi->s_first_error_block); 6163 es->s_first_error_errcode = 6164 ext4_errno_to_code(sbi->s_first_error_code); 6165 } 6166 __ext4_update_tstamp(&es->s_last_error_time, 6167 &es->s_last_error_time_hi, 6168 sbi->s_last_error_time); 6169 strtomem_pad(es->s_last_error_func, sbi->s_last_error_func, 0); 6170 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6171 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6172 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6173 es->s_last_error_errcode = 6174 ext4_errno_to_code(sbi->s_last_error_code); 6175 /* 6176 * Start the daily error reporting function if it hasn't been 6177 * started already 6178 */ 6179 if (!es->s_error_count) 6180 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6181 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6182 sbi->s_add_error_count = 0; 6183 } 6184 spin_unlock(&sbi->s_error_lock); 6185 6186 ext4_superblock_csum_set(sb); 6187 unlock_buffer(sbh); 6188 } 6189 6190 static int ext4_commit_super(struct super_block *sb) 6191 { 6192 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6193 6194 if (!sbh) 6195 return -EINVAL; 6196 6197 ext4_update_super(sb); 6198 6199 lock_buffer(sbh); 6200 /* Buffer got discarded which means block device got invalidated */ 6201 if (!buffer_mapped(sbh)) { 6202 unlock_buffer(sbh); 6203 return -EIO; 6204 } 6205 6206 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6207 /* 6208 * Oh, dear. A previous attempt to write the 6209 * superblock failed. This could happen because the 6210 * USB device was yanked out. Or it could happen to 6211 * be a transient write error and maybe the block will 6212 * be remapped. Nothing we can do but to retry the 6213 * write and hope for the best. 6214 */ 6215 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6216 "superblock detected"); 6217 clear_buffer_write_io_error(sbh); 6218 set_buffer_uptodate(sbh); 6219 } 6220 get_bh(sbh); 6221 /* Clear potential dirty bit if it was journalled update */ 6222 clear_buffer_dirty(sbh); 6223 sbh->b_end_io = end_buffer_write_sync; 6224 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6225 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6226 wait_on_buffer(sbh); 6227 if (buffer_write_io_error(sbh)) { 6228 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6229 "superblock"); 6230 clear_buffer_write_io_error(sbh); 6231 set_buffer_uptodate(sbh); 6232 return -EIO; 6233 } 6234 return 0; 6235 } 6236 6237 /* 6238 * Have we just finished recovery? If so, and if we are mounting (or 6239 * remounting) the filesystem readonly, then we will end up with a 6240 * consistent fs on disk. Record that fact. 6241 */ 6242 static int ext4_mark_recovery_complete(struct super_block *sb, 6243 struct ext4_super_block *es) 6244 { 6245 int err; 6246 journal_t *journal = EXT4_SB(sb)->s_journal; 6247 6248 if (!ext4_has_feature_journal(sb)) { 6249 if (journal != NULL) { 6250 ext4_error(sb, "Journal got removed while the fs was " 6251 "mounted!"); 6252 return -EFSCORRUPTED; 6253 } 6254 return 0; 6255 } 6256 jbd2_journal_lock_updates(journal); 6257 err = jbd2_journal_flush(journal, 0); 6258 if (err < 0) 6259 goto out; 6260 6261 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6262 ext4_has_feature_orphan_present(sb))) { 6263 if (!ext4_orphan_file_empty(sb)) { 6264 ext4_error(sb, "Orphan file not empty on read-only fs."); 6265 err = -EFSCORRUPTED; 6266 goto out; 6267 } 6268 ext4_clear_feature_journal_needs_recovery(sb); 6269 ext4_clear_feature_orphan_present(sb); 6270 ext4_commit_super(sb); 6271 } 6272 out: 6273 jbd2_journal_unlock_updates(journal); 6274 return err; 6275 } 6276 6277 /* 6278 * If we are mounting (or read-write remounting) a filesystem whose journal 6279 * has recorded an error from a previous lifetime, move that error to the 6280 * main filesystem now. 6281 */ 6282 static int ext4_clear_journal_err(struct super_block *sb, 6283 struct ext4_super_block *es) 6284 { 6285 journal_t *journal; 6286 int j_errno; 6287 const char *errstr; 6288 6289 if (!ext4_has_feature_journal(sb)) { 6290 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6291 return -EFSCORRUPTED; 6292 } 6293 6294 journal = EXT4_SB(sb)->s_journal; 6295 6296 /* 6297 * Now check for any error status which may have been recorded in the 6298 * journal by a prior ext4_error() or ext4_abort() 6299 */ 6300 6301 j_errno = jbd2_journal_errno(journal); 6302 if (j_errno) { 6303 char nbuf[16]; 6304 6305 errstr = ext4_decode_error(sb, j_errno, nbuf); 6306 ext4_warning(sb, "Filesystem error recorded " 6307 "from previous mount: %s", errstr); 6308 6309 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6310 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6311 j_errno = ext4_commit_super(sb); 6312 if (j_errno) 6313 return j_errno; 6314 ext4_warning(sb, "Marked fs in need of filesystem check."); 6315 6316 jbd2_journal_clear_err(journal); 6317 jbd2_journal_update_sb_errno(journal); 6318 } 6319 return 0; 6320 } 6321 6322 /* 6323 * Force the running and committing transactions to commit, 6324 * and wait on the commit. 6325 */ 6326 int ext4_force_commit(struct super_block *sb) 6327 { 6328 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal); 6329 } 6330 6331 static int ext4_sync_fs(struct super_block *sb, int wait) 6332 { 6333 int ret = 0; 6334 tid_t target; 6335 bool needs_barrier = false; 6336 struct ext4_sb_info *sbi = EXT4_SB(sb); 6337 6338 ret = ext4_emergency_state(sb); 6339 if (unlikely(ret)) 6340 return ret; 6341 6342 trace_ext4_sync_fs(sb, wait); 6343 flush_workqueue(sbi->rsv_conversion_wq); 6344 /* 6345 * Writeback quota in non-journalled quota case - journalled quota has 6346 * no dirty dquots 6347 */ 6348 dquot_writeback_dquots(sb, -1); 6349 /* 6350 * Data writeback is possible w/o journal transaction, so barrier must 6351 * being sent at the end of the function. But we can skip it if 6352 * transaction_commit will do it for us. 6353 */ 6354 if (sbi->s_journal) { 6355 target = jbd2_get_latest_transaction(sbi->s_journal); 6356 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6357 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6358 needs_barrier = true; 6359 6360 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6361 if (wait) 6362 ret = jbd2_log_wait_commit(sbi->s_journal, 6363 target); 6364 } 6365 } else if (wait && test_opt(sb, BARRIER)) 6366 needs_barrier = true; 6367 if (needs_barrier) { 6368 int err; 6369 err = blkdev_issue_flush(sb->s_bdev); 6370 if (!ret) 6371 ret = err; 6372 } 6373 6374 return ret; 6375 } 6376 6377 /* 6378 * LVM calls this function before a (read-only) snapshot is created. This 6379 * gives us a chance to flush the journal completely and mark the fs clean. 6380 * 6381 * Note that only this function cannot bring a filesystem to be in a clean 6382 * state independently. It relies on upper layer to stop all data & metadata 6383 * modifications. 6384 */ 6385 static int ext4_freeze(struct super_block *sb) 6386 { 6387 int error = 0; 6388 journal_t *journal = EXT4_SB(sb)->s_journal; 6389 6390 if (journal) { 6391 /* Now we set up the journal barrier. */ 6392 jbd2_journal_lock_updates(journal); 6393 6394 /* 6395 * Don't clear the needs_recovery flag if we failed to 6396 * flush the journal. 6397 */ 6398 error = jbd2_journal_flush(journal, 0); 6399 if (error < 0) 6400 goto out; 6401 6402 /* Journal blocked and flushed, clear needs_recovery flag. */ 6403 ext4_clear_feature_journal_needs_recovery(sb); 6404 if (ext4_orphan_file_empty(sb)) 6405 ext4_clear_feature_orphan_present(sb); 6406 } 6407 6408 error = ext4_commit_super(sb); 6409 out: 6410 if (journal) 6411 /* we rely on upper layer to stop further updates */ 6412 jbd2_journal_unlock_updates(journal); 6413 return error; 6414 } 6415 6416 /* 6417 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6418 * flag here, even though the filesystem is not technically dirty yet. 6419 */ 6420 static int ext4_unfreeze(struct super_block *sb) 6421 { 6422 if (ext4_emergency_state(sb)) 6423 return 0; 6424 6425 if (EXT4_SB(sb)->s_journal) { 6426 /* Reset the needs_recovery flag before the fs is unlocked. */ 6427 ext4_set_feature_journal_needs_recovery(sb); 6428 if (ext4_has_feature_orphan_file(sb)) 6429 ext4_set_feature_orphan_present(sb); 6430 } 6431 6432 ext4_commit_super(sb); 6433 return 0; 6434 } 6435 6436 /* 6437 * Structure to save mount options for ext4_remount's benefit 6438 */ 6439 struct ext4_mount_options { 6440 unsigned long s_mount_opt; 6441 unsigned long s_mount_opt2; 6442 kuid_t s_resuid; 6443 kgid_t s_resgid; 6444 unsigned long s_commit_interval; 6445 u32 s_min_batch_time, s_max_batch_time; 6446 #ifdef CONFIG_QUOTA 6447 int s_jquota_fmt; 6448 char *s_qf_names[EXT4_MAXQUOTAS]; 6449 #endif 6450 }; 6451 6452 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6453 { 6454 struct ext4_fs_context *ctx = fc->fs_private; 6455 struct ext4_super_block *es; 6456 struct ext4_sb_info *sbi = EXT4_SB(sb); 6457 unsigned long old_sb_flags; 6458 struct ext4_mount_options old_opts; 6459 ext4_group_t g; 6460 int err = 0; 6461 int alloc_ctx; 6462 #ifdef CONFIG_QUOTA 6463 int enable_quota = 0; 6464 int i, j; 6465 char *to_free[EXT4_MAXQUOTAS]; 6466 #endif 6467 6468 6469 /* Store the original options */ 6470 old_sb_flags = sb->s_flags; 6471 old_opts.s_mount_opt = sbi->s_mount_opt; 6472 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6473 old_opts.s_resuid = sbi->s_resuid; 6474 old_opts.s_resgid = sbi->s_resgid; 6475 old_opts.s_commit_interval = sbi->s_commit_interval; 6476 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6477 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6478 #ifdef CONFIG_QUOTA 6479 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6480 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6481 if (sbi->s_qf_names[i]) { 6482 char *qf_name = get_qf_name(sb, sbi, i); 6483 6484 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6485 if (!old_opts.s_qf_names[i]) { 6486 for (j = 0; j < i; j++) 6487 kfree(old_opts.s_qf_names[j]); 6488 return -ENOMEM; 6489 } 6490 } else 6491 old_opts.s_qf_names[i] = NULL; 6492 #endif 6493 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6494 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6495 ctx->journal_ioprio = 6496 sbi->s_journal->j_task->io_context->ioprio; 6497 else 6498 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 6499 6500 } 6501 6502 if ((ctx->spec & EXT4_SPEC_s_stripe) && 6503 ext4_is_stripe_incompatible(sb, ctx->s_stripe)) { 6504 ext4_msg(sb, KERN_WARNING, 6505 "stripe (%lu) is not aligned with cluster size (%u), " 6506 "stripe is disabled", 6507 ctx->s_stripe, sbi->s_cluster_ratio); 6508 ctx->s_stripe = 0; 6509 } 6510 6511 /* 6512 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause 6513 * two calls to ext4_should_dioread_nolock() to return inconsistent 6514 * values, triggering WARN_ON in ext4_add_complete_io(). we grab 6515 * here s_writepages_rwsem to avoid race between writepages ops and 6516 * remount. 6517 */ 6518 alloc_ctx = ext4_writepages_down_write(sb); 6519 ext4_apply_options(fc, sb); 6520 ext4_writepages_up_write(sb, alloc_ctx); 6521 6522 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6523 test_opt(sb, JOURNAL_CHECKSUM)) { 6524 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6525 "during remount not supported; ignoring"); 6526 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6527 } 6528 6529 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6530 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6531 ext4_msg(sb, KERN_ERR, "can't mount with " 6532 "both data=journal and delalloc"); 6533 err = -EINVAL; 6534 goto restore_opts; 6535 } 6536 if (test_opt(sb, DIOREAD_NOLOCK)) { 6537 ext4_msg(sb, KERN_ERR, "can't mount with " 6538 "both data=journal and dioread_nolock"); 6539 err = -EINVAL; 6540 goto restore_opts; 6541 } 6542 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6543 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6544 ext4_msg(sb, KERN_ERR, "can't mount with " 6545 "journal_async_commit in data=ordered mode"); 6546 err = -EINVAL; 6547 goto restore_opts; 6548 } 6549 } 6550 6551 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6552 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6553 err = -EINVAL; 6554 goto restore_opts; 6555 } 6556 6557 if ((old_opts.s_mount_opt & EXT4_MOUNT_DELALLOC) && 6558 !test_opt(sb, DELALLOC)) { 6559 ext4_msg(sb, KERN_ERR, "can't disable delalloc during remount"); 6560 err = -EINVAL; 6561 goto restore_opts; 6562 } 6563 6564 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6565 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6566 6567 es = sbi->s_es; 6568 6569 if (sbi->s_journal) { 6570 ext4_init_journal_params(sb, sbi->s_journal); 6571 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6572 } 6573 6574 /* Flush outstanding errors before changing fs state */ 6575 flush_work(&sbi->s_sb_upd_work); 6576 6577 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6578 if (ext4_emergency_state(sb)) { 6579 err = -EROFS; 6580 goto restore_opts; 6581 } 6582 6583 if (fc->sb_flags & SB_RDONLY) { 6584 err = sync_filesystem(sb); 6585 if (err < 0) 6586 goto restore_opts; 6587 err = dquot_suspend(sb, -1); 6588 if (err < 0) 6589 goto restore_opts; 6590 6591 /* 6592 * First of all, the unconditional stuff we have to do 6593 * to disable replay of the journal when we next remount 6594 */ 6595 sb->s_flags |= SB_RDONLY; 6596 6597 /* 6598 * OK, test if we are remounting a valid rw partition 6599 * readonly, and if so set the rdonly flag and then 6600 * mark the partition as valid again. 6601 */ 6602 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6603 (sbi->s_mount_state & EXT4_VALID_FS)) 6604 es->s_state = cpu_to_le16(sbi->s_mount_state); 6605 6606 if (sbi->s_journal) { 6607 /* 6608 * We let remount-ro finish even if marking fs 6609 * as clean failed... 6610 */ 6611 ext4_mark_recovery_complete(sb, es); 6612 } 6613 } else { 6614 /* Make sure we can mount this feature set readwrite */ 6615 if (ext4_has_feature_readonly(sb) || 6616 !ext4_feature_set_ok(sb, 0)) { 6617 err = -EROFS; 6618 goto restore_opts; 6619 } 6620 /* 6621 * Make sure the group descriptor checksums 6622 * are sane. If they aren't, refuse to remount r/w. 6623 */ 6624 for (g = 0; g < sbi->s_groups_count; g++) { 6625 struct ext4_group_desc *gdp = 6626 ext4_get_group_desc(sb, g, NULL); 6627 6628 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6629 ext4_msg(sb, KERN_ERR, 6630 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6631 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6632 le16_to_cpu(gdp->bg_checksum)); 6633 err = -EFSBADCRC; 6634 goto restore_opts; 6635 } 6636 } 6637 6638 /* 6639 * If we have an unprocessed orphan list hanging 6640 * around from a previously readonly bdev mount, 6641 * require a full umount/remount for now. 6642 */ 6643 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6644 ext4_msg(sb, KERN_WARNING, "Couldn't " 6645 "remount RDWR because of unprocessed " 6646 "orphan inode list. Please " 6647 "umount/remount instead"); 6648 err = -EINVAL; 6649 goto restore_opts; 6650 } 6651 6652 /* 6653 * Mounting a RDONLY partition read-write, so reread 6654 * and store the current valid flag. (It may have 6655 * been changed by e2fsck since we originally mounted 6656 * the partition.) 6657 */ 6658 if (sbi->s_journal) { 6659 err = ext4_clear_journal_err(sb, es); 6660 if (err) 6661 goto restore_opts; 6662 } 6663 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6664 ~EXT4_FC_REPLAY); 6665 6666 err = ext4_setup_super(sb, es, 0); 6667 if (err) 6668 goto restore_opts; 6669 6670 sb->s_flags &= ~SB_RDONLY; 6671 if (ext4_has_feature_mmp(sb)) { 6672 err = ext4_multi_mount_protect(sb, 6673 le64_to_cpu(es->s_mmp_block)); 6674 if (err) 6675 goto restore_opts; 6676 } 6677 #ifdef CONFIG_QUOTA 6678 enable_quota = 1; 6679 #endif 6680 } 6681 } 6682 6683 /* 6684 * Handle creation of system zone data early because it can fail. 6685 * Releasing of existing data is done when we are sure remount will 6686 * succeed. 6687 */ 6688 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6689 err = ext4_setup_system_zone(sb); 6690 if (err) 6691 goto restore_opts; 6692 } 6693 6694 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6695 err = ext4_commit_super(sb); 6696 if (err) 6697 goto restore_opts; 6698 } 6699 6700 #ifdef CONFIG_QUOTA 6701 if (enable_quota) { 6702 if (sb_any_quota_suspended(sb)) 6703 dquot_resume(sb, -1); 6704 else if (ext4_has_feature_quota(sb)) { 6705 err = ext4_enable_quotas(sb); 6706 if (err) 6707 goto restore_opts; 6708 } 6709 } 6710 /* Release old quota file names */ 6711 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6712 kfree(old_opts.s_qf_names[i]); 6713 #endif 6714 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6715 ext4_release_system_zone(sb); 6716 6717 /* 6718 * Reinitialize lazy itable initialization thread based on 6719 * current settings 6720 */ 6721 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6722 ext4_unregister_li_request(sb); 6723 else { 6724 ext4_group_t first_not_zeroed; 6725 first_not_zeroed = ext4_has_uninit_itable(sb); 6726 ext4_register_li_request(sb, first_not_zeroed); 6727 } 6728 6729 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6730 ext4_stop_mmpd(sbi); 6731 6732 /* 6733 * Handle aborting the filesystem as the last thing during remount to 6734 * avoid obsure errors during remount when some option changes fail to 6735 * apply due to shutdown filesystem. 6736 */ 6737 if (test_opt2(sb, ABORT)) 6738 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6739 6740 return 0; 6741 6742 restore_opts: 6743 /* 6744 * If there was a failing r/w to ro transition, we may need to 6745 * re-enable quota 6746 */ 6747 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) && 6748 sb_any_quota_suspended(sb)) 6749 dquot_resume(sb, -1); 6750 6751 alloc_ctx = ext4_writepages_down_write(sb); 6752 sb->s_flags = old_sb_flags; 6753 sbi->s_mount_opt = old_opts.s_mount_opt; 6754 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6755 sbi->s_resuid = old_opts.s_resuid; 6756 sbi->s_resgid = old_opts.s_resgid; 6757 sbi->s_commit_interval = old_opts.s_commit_interval; 6758 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6759 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6760 ext4_writepages_up_write(sb, alloc_ctx); 6761 6762 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6763 ext4_release_system_zone(sb); 6764 #ifdef CONFIG_QUOTA 6765 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6766 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6767 to_free[i] = get_qf_name(sb, sbi, i); 6768 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6769 } 6770 synchronize_rcu(); 6771 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6772 kfree(to_free[i]); 6773 #endif 6774 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6775 ext4_stop_mmpd(sbi); 6776 return err; 6777 } 6778 6779 static int ext4_reconfigure(struct fs_context *fc) 6780 { 6781 struct super_block *sb = fc->root->d_sb; 6782 int ret; 6783 bool old_ro = sb_rdonly(sb); 6784 6785 fc->s_fs_info = EXT4_SB(sb); 6786 6787 ret = ext4_check_opt_consistency(fc, sb); 6788 if (ret < 0) 6789 return ret; 6790 6791 ret = __ext4_remount(fc, sb); 6792 if (ret < 0) 6793 return ret; 6794 6795 ext4_msg(sb, KERN_INFO, "re-mounted %pU%s.", 6796 &sb->s_uuid, 6797 (old_ro != sb_rdonly(sb)) ? (sb_rdonly(sb) ? " ro" : " r/w") : ""); 6798 6799 return 0; 6800 } 6801 6802 #ifdef CONFIG_QUOTA 6803 static int ext4_statfs_project(struct super_block *sb, 6804 kprojid_t projid, struct kstatfs *buf) 6805 { 6806 struct kqid qid; 6807 struct dquot *dquot; 6808 u64 limit; 6809 u64 curblock; 6810 6811 qid = make_kqid_projid(projid); 6812 dquot = dqget(sb, qid); 6813 if (IS_ERR(dquot)) 6814 return PTR_ERR(dquot); 6815 spin_lock(&dquot->dq_dqb_lock); 6816 6817 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6818 dquot->dq_dqb.dqb_bhardlimit); 6819 limit >>= sb->s_blocksize_bits; 6820 6821 if (limit) { 6822 uint64_t remaining = 0; 6823 6824 curblock = (dquot->dq_dqb.dqb_curspace + 6825 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6826 if (limit > curblock) 6827 remaining = limit - curblock; 6828 6829 buf->f_blocks = min(buf->f_blocks, limit); 6830 buf->f_bfree = min(buf->f_bfree, remaining); 6831 buf->f_bavail = min(buf->f_bavail, remaining); 6832 } 6833 6834 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6835 dquot->dq_dqb.dqb_ihardlimit); 6836 if (limit) { 6837 uint64_t remaining = 0; 6838 6839 if (limit > dquot->dq_dqb.dqb_curinodes) 6840 remaining = limit - dquot->dq_dqb.dqb_curinodes; 6841 6842 buf->f_files = min(buf->f_files, limit); 6843 buf->f_ffree = min(buf->f_ffree, remaining); 6844 } 6845 6846 spin_unlock(&dquot->dq_dqb_lock); 6847 dqput(dquot); 6848 return 0; 6849 } 6850 #endif 6851 6852 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6853 { 6854 struct super_block *sb = dentry->d_sb; 6855 struct ext4_sb_info *sbi = EXT4_SB(sb); 6856 struct ext4_super_block *es = sbi->s_es; 6857 ext4_fsblk_t overhead = 0, resv_blocks; 6858 s64 bfree; 6859 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6860 6861 if (!test_opt(sb, MINIX_DF)) 6862 overhead = sbi->s_overhead; 6863 6864 buf->f_type = EXT4_SUPER_MAGIC; 6865 buf->f_bsize = sb->s_blocksize; 6866 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6867 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6868 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6869 /* prevent underflow in case that few free space is available */ 6870 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6871 buf->f_bavail = buf->f_bfree - 6872 (ext4_r_blocks_count(es) + resv_blocks); 6873 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6874 buf->f_bavail = 0; 6875 buf->f_files = le32_to_cpu(es->s_inodes_count); 6876 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6877 buf->f_namelen = EXT4_NAME_LEN; 6878 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6879 6880 #ifdef CONFIG_QUOTA 6881 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6882 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6883 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6884 #endif 6885 return 0; 6886 } 6887 6888 6889 #ifdef CONFIG_QUOTA 6890 6891 /* 6892 * Helper functions so that transaction is started before we acquire dqio_sem 6893 * to keep correct lock ordering of transaction > dqio_sem 6894 */ 6895 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6896 { 6897 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6898 } 6899 6900 static int ext4_write_dquot(struct dquot *dquot) 6901 { 6902 int ret, err; 6903 handle_t *handle; 6904 struct inode *inode; 6905 6906 inode = dquot_to_inode(dquot); 6907 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6908 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6909 if (IS_ERR(handle)) 6910 return PTR_ERR(handle); 6911 ret = dquot_commit(dquot); 6912 if (ret < 0) 6913 ext4_error_err(dquot->dq_sb, -ret, 6914 "Failed to commit dquot type %d", 6915 dquot->dq_id.type); 6916 err = ext4_journal_stop(handle); 6917 if (!ret) 6918 ret = err; 6919 return ret; 6920 } 6921 6922 static int ext4_acquire_dquot(struct dquot *dquot) 6923 { 6924 int ret, err; 6925 handle_t *handle; 6926 6927 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6928 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6929 if (IS_ERR(handle)) 6930 return PTR_ERR(handle); 6931 ret = dquot_acquire(dquot); 6932 if (ret < 0) 6933 ext4_error_err(dquot->dq_sb, -ret, 6934 "Failed to acquire dquot type %d", 6935 dquot->dq_id.type); 6936 err = ext4_journal_stop(handle); 6937 if (!ret) 6938 ret = err; 6939 return ret; 6940 } 6941 6942 static int ext4_release_dquot(struct dquot *dquot) 6943 { 6944 int ret, err; 6945 handle_t *handle; 6946 bool freeze_protected = false; 6947 6948 /* 6949 * Trying to sb_start_intwrite() in a running transaction 6950 * can result in a deadlock. Further, running transactions 6951 * are already protected from freezing. 6952 */ 6953 if (!ext4_journal_current_handle()) { 6954 sb_start_intwrite(dquot->dq_sb); 6955 freeze_protected = true; 6956 } 6957 6958 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6959 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6960 if (IS_ERR(handle)) { 6961 /* Release dquot anyway to avoid endless cycle in dqput() */ 6962 dquot_release(dquot); 6963 if (freeze_protected) 6964 sb_end_intwrite(dquot->dq_sb); 6965 return PTR_ERR(handle); 6966 } 6967 ret = dquot_release(dquot); 6968 if (ret < 0) 6969 ext4_error_err(dquot->dq_sb, -ret, 6970 "Failed to release dquot type %d", 6971 dquot->dq_id.type); 6972 err = ext4_journal_stop(handle); 6973 if (!ret) 6974 ret = err; 6975 6976 if (freeze_protected) 6977 sb_end_intwrite(dquot->dq_sb); 6978 6979 return ret; 6980 } 6981 6982 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6983 { 6984 struct super_block *sb = dquot->dq_sb; 6985 6986 if (ext4_is_quota_journalled(sb)) { 6987 dquot_mark_dquot_dirty(dquot); 6988 return ext4_write_dquot(dquot); 6989 } else { 6990 return dquot_mark_dquot_dirty(dquot); 6991 } 6992 } 6993 6994 static int ext4_write_info(struct super_block *sb, int type) 6995 { 6996 int ret, err; 6997 handle_t *handle; 6998 6999 /* Data block + inode block */ 7000 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 7001 if (IS_ERR(handle)) 7002 return PTR_ERR(handle); 7003 ret = dquot_commit_info(sb, type); 7004 err = ext4_journal_stop(handle); 7005 if (!ret) 7006 ret = err; 7007 return ret; 7008 } 7009 7010 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 7011 { 7012 struct ext4_inode_info *ei = EXT4_I(inode); 7013 7014 /* The first argument of lockdep_set_subclass has to be 7015 * *exactly* the same as the argument to init_rwsem() --- in 7016 * this case, in init_once() --- or lockdep gets unhappy 7017 * because the name of the lock is set using the 7018 * stringification of the argument to init_rwsem(). 7019 */ 7020 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 7021 lockdep_set_subclass(&ei->i_data_sem, subclass); 7022 } 7023 7024 /* 7025 * Standard function to be called on quota_on 7026 */ 7027 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 7028 const struct path *path) 7029 { 7030 int err; 7031 7032 if (!test_opt(sb, QUOTA)) 7033 return -EINVAL; 7034 7035 /* Quotafile not on the same filesystem? */ 7036 if (path->dentry->d_sb != sb) 7037 return -EXDEV; 7038 7039 /* Quota already enabled for this file? */ 7040 if (IS_NOQUOTA(d_inode(path->dentry))) 7041 return -EBUSY; 7042 7043 /* Journaling quota? */ 7044 if (EXT4_SB(sb)->s_qf_names[type]) { 7045 /* Quotafile not in fs root? */ 7046 if (path->dentry->d_parent != sb->s_root) 7047 ext4_msg(sb, KERN_WARNING, 7048 "Quota file not on filesystem root. " 7049 "Journaled quota will not work"); 7050 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 7051 } else { 7052 /* 7053 * Clear the flag just in case mount options changed since 7054 * last time. 7055 */ 7056 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 7057 } 7058 7059 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 7060 err = dquot_quota_on(sb, type, format_id, path); 7061 if (!err) { 7062 struct inode *inode = d_inode(path->dentry); 7063 handle_t *handle; 7064 7065 /* 7066 * Set inode flags to prevent userspace from messing with quota 7067 * files. If this fails, we return success anyway since quotas 7068 * are already enabled and this is not a hard failure. 7069 */ 7070 inode_lock(inode); 7071 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7072 if (IS_ERR(handle)) 7073 goto unlock_inode; 7074 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 7075 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 7076 S_NOATIME | S_IMMUTABLE); 7077 err = ext4_mark_inode_dirty(handle, inode); 7078 ext4_journal_stop(handle); 7079 unlock_inode: 7080 inode_unlock(inode); 7081 if (err) 7082 dquot_quota_off(sb, type); 7083 } 7084 if (err) 7085 lockdep_set_quota_inode(path->dentry->d_inode, 7086 I_DATA_SEM_NORMAL); 7087 return err; 7088 } 7089 7090 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum) 7091 { 7092 switch (type) { 7093 case USRQUOTA: 7094 return qf_inum == EXT4_USR_QUOTA_INO; 7095 case GRPQUOTA: 7096 return qf_inum == EXT4_GRP_QUOTA_INO; 7097 case PRJQUOTA: 7098 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO; 7099 default: 7100 BUG(); 7101 } 7102 } 7103 7104 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 7105 unsigned int flags) 7106 { 7107 int err; 7108 struct inode *qf_inode; 7109 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7110 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7111 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7112 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7113 }; 7114 7115 BUG_ON(!ext4_has_feature_quota(sb)); 7116 7117 if (!qf_inums[type]) 7118 return -EPERM; 7119 7120 if (!ext4_check_quota_inum(type, qf_inums[type])) { 7121 ext4_error(sb, "Bad quota inum: %lu, type: %d", 7122 qf_inums[type], type); 7123 return -EUCLEAN; 7124 } 7125 7126 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 7127 if (IS_ERR(qf_inode)) { 7128 ext4_error(sb, "Bad quota inode: %lu, type: %d", 7129 qf_inums[type], type); 7130 return PTR_ERR(qf_inode); 7131 } 7132 7133 /* Don't account quota for quota files to avoid recursion */ 7134 qf_inode->i_flags |= S_NOQUOTA; 7135 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 7136 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 7137 if (err) 7138 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 7139 iput(qf_inode); 7140 7141 return err; 7142 } 7143 7144 /* Enable usage tracking for all quota types. */ 7145 int ext4_enable_quotas(struct super_block *sb) 7146 { 7147 int type, err = 0; 7148 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7149 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7150 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7151 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7152 }; 7153 bool quota_mopt[EXT4_MAXQUOTAS] = { 7154 test_opt(sb, USRQUOTA), 7155 test_opt(sb, GRPQUOTA), 7156 test_opt(sb, PRJQUOTA), 7157 }; 7158 7159 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 7160 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 7161 if (qf_inums[type]) { 7162 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 7163 DQUOT_USAGE_ENABLED | 7164 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 7165 if (err) { 7166 ext4_warning(sb, 7167 "Failed to enable quota tracking " 7168 "(type=%d, err=%d, ino=%lu). " 7169 "Please run e2fsck to fix.", type, 7170 err, qf_inums[type]); 7171 7172 ext4_quotas_off(sb, type); 7173 return err; 7174 } 7175 } 7176 } 7177 return 0; 7178 } 7179 7180 static int ext4_quota_off(struct super_block *sb, int type) 7181 { 7182 struct inode *inode = sb_dqopt(sb)->files[type]; 7183 handle_t *handle; 7184 int err; 7185 7186 /* Force all delayed allocation blocks to be allocated. 7187 * Caller already holds s_umount sem */ 7188 if (test_opt(sb, DELALLOC)) 7189 sync_filesystem(sb); 7190 7191 if (!inode || !igrab(inode)) 7192 goto out; 7193 7194 err = dquot_quota_off(sb, type); 7195 if (err || ext4_has_feature_quota(sb)) 7196 goto out_put; 7197 /* 7198 * When the filesystem was remounted read-only first, we cannot cleanup 7199 * inode flags here. Bad luck but people should be using QUOTA feature 7200 * these days anyway. 7201 */ 7202 if (sb_rdonly(sb)) 7203 goto out_put; 7204 7205 inode_lock(inode); 7206 /* 7207 * Update modification times of quota files when userspace can 7208 * start looking at them. If we fail, we return success anyway since 7209 * this is not a hard failure and quotas are already disabled. 7210 */ 7211 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7212 if (IS_ERR(handle)) { 7213 err = PTR_ERR(handle); 7214 goto out_unlock; 7215 } 7216 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7217 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7218 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 7219 err = ext4_mark_inode_dirty(handle, inode); 7220 ext4_journal_stop(handle); 7221 out_unlock: 7222 inode_unlock(inode); 7223 out_put: 7224 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7225 iput(inode); 7226 return err; 7227 out: 7228 return dquot_quota_off(sb, type); 7229 } 7230 7231 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7232 * acquiring the locks... As quota files are never truncated and quota code 7233 * itself serializes the operations (and no one else should touch the files) 7234 * we don't have to be afraid of races */ 7235 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7236 size_t len, loff_t off) 7237 { 7238 struct inode *inode = sb_dqopt(sb)->files[type]; 7239 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7240 int offset = off & (sb->s_blocksize - 1); 7241 int tocopy; 7242 size_t toread; 7243 struct buffer_head *bh; 7244 loff_t i_size = i_size_read(inode); 7245 7246 if (off > i_size) 7247 return 0; 7248 if (off+len > i_size) 7249 len = i_size-off; 7250 toread = len; 7251 while (toread > 0) { 7252 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread); 7253 bh = ext4_bread(NULL, inode, blk, 0); 7254 if (IS_ERR(bh)) 7255 return PTR_ERR(bh); 7256 if (!bh) /* A hole? */ 7257 memset(data, 0, tocopy); 7258 else 7259 memcpy(data, bh->b_data+offset, tocopy); 7260 brelse(bh); 7261 offset = 0; 7262 toread -= tocopy; 7263 data += tocopy; 7264 blk++; 7265 } 7266 return len; 7267 } 7268 7269 /* Write to quotafile (we know the transaction is already started and has 7270 * enough credits) */ 7271 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7272 const char *data, size_t len, loff_t off) 7273 { 7274 struct inode *inode = sb_dqopt(sb)->files[type]; 7275 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7276 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7277 int retries = 0; 7278 struct buffer_head *bh; 7279 handle_t *handle = journal_current_handle(); 7280 7281 if (!handle) { 7282 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7283 " cancelled because transaction is not started", 7284 (unsigned long long)off, (unsigned long long)len); 7285 return -EIO; 7286 } 7287 /* 7288 * Since we account only one data block in transaction credits, 7289 * then it is impossible to cross a block boundary. 7290 */ 7291 if (sb->s_blocksize - offset < len) { 7292 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7293 " cancelled because not block aligned", 7294 (unsigned long long)off, (unsigned long long)len); 7295 return -EIO; 7296 } 7297 7298 do { 7299 bh = ext4_bread(handle, inode, blk, 7300 EXT4_GET_BLOCKS_CREATE | 7301 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7302 } while (PTR_ERR(bh) == -ENOSPC && 7303 ext4_should_retry_alloc(inode->i_sb, &retries)); 7304 if (IS_ERR(bh)) 7305 return PTR_ERR(bh); 7306 if (!bh) 7307 goto out; 7308 BUFFER_TRACE(bh, "get write access"); 7309 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7310 if (err) { 7311 brelse(bh); 7312 return err; 7313 } 7314 lock_buffer(bh); 7315 memcpy(bh->b_data+offset, data, len); 7316 flush_dcache_folio(bh->b_folio); 7317 unlock_buffer(bh); 7318 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7319 brelse(bh); 7320 out: 7321 if (inode->i_size < off + len) { 7322 i_size_write(inode, off + len); 7323 EXT4_I(inode)->i_disksize = inode->i_size; 7324 err2 = ext4_mark_inode_dirty(handle, inode); 7325 if (unlikely(err2 && !err)) 7326 err = err2; 7327 } 7328 return err ? err : len; 7329 } 7330 #endif 7331 7332 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7333 static inline void register_as_ext2(void) 7334 { 7335 int err = register_filesystem(&ext2_fs_type); 7336 if (err) 7337 printk(KERN_WARNING 7338 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7339 } 7340 7341 static inline void unregister_as_ext2(void) 7342 { 7343 unregister_filesystem(&ext2_fs_type); 7344 } 7345 7346 static inline int ext2_feature_set_ok(struct super_block *sb) 7347 { 7348 if (ext4_has_unknown_ext2_incompat_features(sb)) 7349 return 0; 7350 if (sb_rdonly(sb)) 7351 return 1; 7352 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7353 return 0; 7354 return 1; 7355 } 7356 #else 7357 static inline void register_as_ext2(void) { } 7358 static inline void unregister_as_ext2(void) { } 7359 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7360 #endif 7361 7362 static inline void register_as_ext3(void) 7363 { 7364 int err = register_filesystem(&ext3_fs_type); 7365 if (err) 7366 printk(KERN_WARNING 7367 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7368 } 7369 7370 static inline void unregister_as_ext3(void) 7371 { 7372 unregister_filesystem(&ext3_fs_type); 7373 } 7374 7375 static inline int ext3_feature_set_ok(struct super_block *sb) 7376 { 7377 if (ext4_has_unknown_ext3_incompat_features(sb)) 7378 return 0; 7379 if (!ext4_has_feature_journal(sb)) 7380 return 0; 7381 if (sb_rdonly(sb)) 7382 return 1; 7383 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7384 return 0; 7385 return 1; 7386 } 7387 7388 static void ext4_kill_sb(struct super_block *sb) 7389 { 7390 struct ext4_sb_info *sbi = EXT4_SB(sb); 7391 struct file *bdev_file = sbi ? sbi->s_journal_bdev_file : NULL; 7392 7393 kill_block_super(sb); 7394 7395 if (bdev_file) 7396 bdev_fput(bdev_file); 7397 } 7398 7399 static struct file_system_type ext4_fs_type = { 7400 .owner = THIS_MODULE, 7401 .name = "ext4", 7402 .init_fs_context = ext4_init_fs_context, 7403 .parameters = ext4_param_specs, 7404 .kill_sb = ext4_kill_sb, 7405 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP | FS_MGTIME, 7406 }; 7407 MODULE_ALIAS_FS("ext4"); 7408 7409 static int __init ext4_init_fs(void) 7410 { 7411 int err; 7412 7413 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7414 ext4_li_info = NULL; 7415 7416 /* Build-time check for flags consistency */ 7417 ext4_check_flag_values(); 7418 7419 err = ext4_init_es(); 7420 if (err) 7421 return err; 7422 7423 err = ext4_init_pending(); 7424 if (err) 7425 goto out7; 7426 7427 err = ext4_init_post_read_processing(); 7428 if (err) 7429 goto out6; 7430 7431 err = ext4_init_pageio(); 7432 if (err) 7433 goto out5; 7434 7435 err = ext4_init_system_zone(); 7436 if (err) 7437 goto out4; 7438 7439 err = ext4_init_sysfs(); 7440 if (err) 7441 goto out3; 7442 7443 err = ext4_init_mballoc(); 7444 if (err) 7445 goto out2; 7446 err = init_inodecache(); 7447 if (err) 7448 goto out1; 7449 7450 err = ext4_fc_init_dentry_cache(); 7451 if (err) 7452 goto out05; 7453 7454 register_as_ext3(); 7455 register_as_ext2(); 7456 err = register_filesystem(&ext4_fs_type); 7457 if (err) 7458 goto out; 7459 7460 return 0; 7461 out: 7462 unregister_as_ext2(); 7463 unregister_as_ext3(); 7464 ext4_fc_destroy_dentry_cache(); 7465 out05: 7466 destroy_inodecache(); 7467 out1: 7468 ext4_exit_mballoc(); 7469 out2: 7470 ext4_exit_sysfs(); 7471 out3: 7472 ext4_exit_system_zone(); 7473 out4: 7474 ext4_exit_pageio(); 7475 out5: 7476 ext4_exit_post_read_processing(); 7477 out6: 7478 ext4_exit_pending(); 7479 out7: 7480 ext4_exit_es(); 7481 7482 return err; 7483 } 7484 7485 static void __exit ext4_exit_fs(void) 7486 { 7487 ext4_destroy_lazyinit_thread(); 7488 unregister_as_ext2(); 7489 unregister_as_ext3(); 7490 unregister_filesystem(&ext4_fs_type); 7491 ext4_fc_destroy_dentry_cache(); 7492 destroy_inodecache(); 7493 ext4_exit_mballoc(); 7494 ext4_exit_sysfs(); 7495 ext4_exit_system_zone(); 7496 ext4_exit_pageio(); 7497 ext4_exit_post_read_processing(); 7498 ext4_exit_es(); 7499 ext4_exit_pending(); 7500 } 7501 7502 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7503 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7504 MODULE_LICENSE("GPL"); 7505 module_init(ext4_init_fs) 7506 module_exit(ext4_exit_fs) 7507