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_destroy_lazyinit_thread(void); 83 static void ext4_unregister_li_request(struct super_block *sb); 84 static void ext4_clear_request_list(void); 85 static struct inode *ext4_get_journal_inode(struct super_block *sb, 86 unsigned int journal_inum); 87 static int ext4_validate_options(struct fs_context *fc); 88 static int ext4_check_opt_consistency(struct fs_context *fc, 89 struct super_block *sb); 90 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb); 91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param); 92 static int ext4_get_tree(struct fs_context *fc); 93 static int ext4_reconfigure(struct fs_context *fc); 94 static void ext4_fc_free(struct fs_context *fc); 95 static int ext4_init_fs_context(struct fs_context *fc); 96 static void ext4_kill_sb(struct super_block *sb); 97 static const struct fs_parameter_spec ext4_param_specs[]; 98 99 /* 100 * Lock ordering 101 * 102 * page fault path: 103 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start 104 * -> page lock -> i_data_sem (rw) 105 * 106 * buffered write path: 107 * sb_start_write -> i_mutex -> mmap_lock 108 * sb_start_write -> i_mutex -> transaction start -> page lock -> 109 * i_data_sem (rw) 110 * 111 * truncate: 112 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) -> 113 * page lock 114 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start -> 115 * i_data_sem (rw) 116 * 117 * direct IO: 118 * sb_start_write -> i_mutex -> mmap_lock 119 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 120 * 121 * writepages: 122 * transaction start -> page lock(s) -> i_data_sem (rw) 123 */ 124 125 static const struct fs_context_operations ext4_context_ops = { 126 .parse_param = ext4_parse_param, 127 .get_tree = ext4_get_tree, 128 .reconfigure = ext4_reconfigure, 129 .free = ext4_fc_free, 130 }; 131 132 133 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 134 static struct file_system_type ext2_fs_type = { 135 .owner = THIS_MODULE, 136 .name = "ext2", 137 .init_fs_context = ext4_init_fs_context, 138 .parameters = ext4_param_specs, 139 .kill_sb = ext4_kill_sb, 140 .fs_flags = FS_REQUIRES_DEV, 141 }; 142 MODULE_ALIAS_FS("ext2"); 143 MODULE_ALIAS("ext2"); 144 #define IS_EXT2_SB(sb) ((sb)->s_type == &ext2_fs_type) 145 #else 146 #define IS_EXT2_SB(sb) (0) 147 #endif 148 149 150 static struct file_system_type ext3_fs_type = { 151 .owner = THIS_MODULE, 152 .name = "ext3", 153 .init_fs_context = ext4_init_fs_context, 154 .parameters = ext4_param_specs, 155 .kill_sb = ext4_kill_sb, 156 .fs_flags = FS_REQUIRES_DEV, 157 }; 158 MODULE_ALIAS_FS("ext3"); 159 MODULE_ALIAS("ext3"); 160 #define IS_EXT3_SB(sb) ((sb)->s_type == &ext3_fs_type) 161 162 163 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 164 bh_end_io_t *end_io) 165 { 166 /* 167 * buffer's verified bit is no longer valid after reading from 168 * disk again due to write out error, clear it to make sure we 169 * recheck the buffer contents. 170 */ 171 clear_buffer_verified(bh); 172 173 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 174 get_bh(bh); 175 submit_bh(REQ_OP_READ | op_flags, bh); 176 } 177 178 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags, 179 bh_end_io_t *end_io) 180 { 181 BUG_ON(!buffer_locked(bh)); 182 183 if (ext4_buffer_uptodate(bh)) { 184 unlock_buffer(bh); 185 return; 186 } 187 __ext4_read_bh(bh, op_flags, end_io); 188 } 189 190 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io) 191 { 192 BUG_ON(!buffer_locked(bh)); 193 194 if (ext4_buffer_uptodate(bh)) { 195 unlock_buffer(bh); 196 return 0; 197 } 198 199 __ext4_read_bh(bh, op_flags, end_io); 200 201 wait_on_buffer(bh); 202 if (buffer_uptodate(bh)) 203 return 0; 204 return -EIO; 205 } 206 207 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 208 { 209 lock_buffer(bh); 210 if (!wait) { 211 ext4_read_bh_nowait(bh, op_flags, NULL); 212 return 0; 213 } 214 return ext4_read_bh(bh, op_flags, NULL); 215 } 216 217 /* 218 * This works like __bread_gfp() except it uses ERR_PTR for error 219 * returns. Currently with sb_bread it's impossible to distinguish 220 * between ENOMEM and EIO situations (since both result in a NULL 221 * return. 222 */ 223 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 224 sector_t block, 225 blk_opf_t op_flags, gfp_t gfp) 226 { 227 struct buffer_head *bh; 228 int ret; 229 230 bh = sb_getblk_gfp(sb, block, gfp); 231 if (bh == NULL) 232 return ERR_PTR(-ENOMEM); 233 if (ext4_buffer_uptodate(bh)) 234 return bh; 235 236 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 237 if (ret) { 238 put_bh(bh); 239 return ERR_PTR(ret); 240 } 241 return bh; 242 } 243 244 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 245 blk_opf_t op_flags) 246 { 247 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 248 } 249 250 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 251 sector_t block) 252 { 253 return __ext4_sb_bread_gfp(sb, block, 0, 0); 254 } 255 256 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 257 { 258 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 259 260 if (likely(bh)) { 261 if (trylock_buffer(bh)) 262 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL); 263 brelse(bh); 264 } 265 } 266 267 static int ext4_verify_csum_type(struct super_block *sb, 268 struct ext4_super_block *es) 269 { 270 if (!ext4_has_feature_metadata_csum(sb)) 271 return 1; 272 273 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 274 } 275 276 __le32 ext4_superblock_csum(struct super_block *sb, 277 struct ext4_super_block *es) 278 { 279 struct ext4_sb_info *sbi = EXT4_SB(sb); 280 int offset = offsetof(struct ext4_super_block, s_checksum); 281 __u32 csum; 282 283 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 284 285 return cpu_to_le32(csum); 286 } 287 288 static int ext4_superblock_csum_verify(struct super_block *sb, 289 struct ext4_super_block *es) 290 { 291 if (!ext4_has_metadata_csum(sb)) 292 return 1; 293 294 return es->s_checksum == ext4_superblock_csum(sb, es); 295 } 296 297 void ext4_superblock_csum_set(struct super_block *sb) 298 { 299 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 300 301 if (!ext4_has_metadata_csum(sb)) 302 return; 303 304 es->s_checksum = ext4_superblock_csum(sb, es); 305 } 306 307 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 308 struct ext4_group_desc *bg) 309 { 310 return le32_to_cpu(bg->bg_block_bitmap_lo) | 311 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 312 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 313 } 314 315 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 316 struct ext4_group_desc *bg) 317 { 318 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 319 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 320 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 321 } 322 323 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 324 struct ext4_group_desc *bg) 325 { 326 return le32_to_cpu(bg->bg_inode_table_lo) | 327 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 328 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 329 } 330 331 __u32 ext4_free_group_clusters(struct super_block *sb, 332 struct ext4_group_desc *bg) 333 { 334 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 335 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 336 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 337 } 338 339 __u32 ext4_free_inodes_count(struct super_block *sb, 340 struct ext4_group_desc *bg) 341 { 342 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 343 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 344 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 345 } 346 347 __u32 ext4_used_dirs_count(struct super_block *sb, 348 struct ext4_group_desc *bg) 349 { 350 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 351 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 352 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 353 } 354 355 __u32 ext4_itable_unused_count(struct super_block *sb, 356 struct ext4_group_desc *bg) 357 { 358 return le16_to_cpu(bg->bg_itable_unused_lo) | 359 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 360 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 361 } 362 363 void ext4_block_bitmap_set(struct super_block *sb, 364 struct ext4_group_desc *bg, ext4_fsblk_t blk) 365 { 366 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 367 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 368 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 369 } 370 371 void ext4_inode_bitmap_set(struct super_block *sb, 372 struct ext4_group_desc *bg, ext4_fsblk_t blk) 373 { 374 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 375 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 376 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 377 } 378 379 void ext4_inode_table_set(struct super_block *sb, 380 struct ext4_group_desc *bg, ext4_fsblk_t blk) 381 { 382 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 383 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 384 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 385 } 386 387 void ext4_free_group_clusters_set(struct super_block *sb, 388 struct ext4_group_desc *bg, __u32 count) 389 { 390 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 391 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 392 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 393 } 394 395 void ext4_free_inodes_set(struct super_block *sb, 396 struct ext4_group_desc *bg, __u32 count) 397 { 398 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 399 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 400 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 401 } 402 403 void ext4_used_dirs_set(struct super_block *sb, 404 struct ext4_group_desc *bg, __u32 count) 405 { 406 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 407 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 408 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 409 } 410 411 void ext4_itable_unused_set(struct super_block *sb, 412 struct ext4_group_desc *bg, __u32 count) 413 { 414 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 415 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 416 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 417 } 418 419 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 420 { 421 now = clamp_val(now, 0, (1ull << 40) - 1); 422 423 *lo = cpu_to_le32(lower_32_bits(now)); 424 *hi = upper_32_bits(now); 425 } 426 427 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 428 { 429 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 430 } 431 #define ext4_update_tstamp(es, tstamp) \ 432 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 433 ktime_get_real_seconds()) 434 #define ext4_get_tstamp(es, tstamp) \ 435 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 436 437 #define EXT4_SB_REFRESH_INTERVAL_SEC (3600) /* seconds (1 hour) */ 438 #define EXT4_SB_REFRESH_INTERVAL_KB (16384) /* kilobytes (16MB) */ 439 440 /* 441 * The ext4_maybe_update_superblock() function checks and updates the 442 * superblock if needed. 443 * 444 * This function is designed to update the on-disk superblock only under 445 * certain conditions to prevent excessive disk writes and unnecessary 446 * waking of the disk from sleep. The superblock will be updated if: 447 * 1. More than an hour has passed since the last superblock update, and 448 * 2. More than 16MB have been written since the last superblock update. 449 * 450 * @sb: The superblock 451 */ 452 static void ext4_maybe_update_superblock(struct super_block *sb) 453 { 454 struct ext4_sb_info *sbi = EXT4_SB(sb); 455 struct ext4_super_block *es = sbi->s_es; 456 journal_t *journal = sbi->s_journal; 457 time64_t now; 458 __u64 last_update; 459 __u64 lifetime_write_kbytes; 460 __u64 diff_size; 461 462 if (sb_rdonly(sb) || !(sb->s_flags & SB_ACTIVE) || 463 !journal || (journal->j_flags & JBD2_UNMOUNT)) 464 return; 465 466 now = ktime_get_real_seconds(); 467 last_update = ext4_get_tstamp(es, s_wtime); 468 469 if (likely(now - last_update < EXT4_SB_REFRESH_INTERVAL_SEC)) 470 return; 471 472 lifetime_write_kbytes = sbi->s_kbytes_written + 473 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 474 sbi->s_sectors_written_start) >> 1); 475 476 /* Get the number of kilobytes not written to disk to account 477 * for statistics and compare with a multiple of 16 MB. This 478 * is used to determine when the next superblock commit should 479 * occur (i.e. not more often than once per 16MB if there was 480 * less written in an hour). 481 */ 482 diff_size = lifetime_write_kbytes - le64_to_cpu(es->s_kbytes_written); 483 484 if (diff_size > EXT4_SB_REFRESH_INTERVAL_KB) 485 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 486 } 487 488 /* 489 * The del_gendisk() function uninitializes the disk-specific data 490 * structures, including the bdi structure, without telling anyone 491 * else. Once this happens, any attempt to call mark_buffer_dirty() 492 * (for example, by ext4_commit_super), will cause a kernel OOPS. 493 * This is a kludge to prevent these oops until we can put in a proper 494 * hook in del_gendisk() to inform the VFS and file system layers. 495 */ 496 static int block_device_ejected(struct super_block *sb) 497 { 498 struct inode *bd_inode = sb->s_bdev->bd_inode; 499 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 500 501 return bdi->dev == NULL; 502 } 503 504 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 505 { 506 struct super_block *sb = journal->j_private; 507 struct ext4_sb_info *sbi = EXT4_SB(sb); 508 int error = is_journal_aborted(journal); 509 struct ext4_journal_cb_entry *jce; 510 511 BUG_ON(txn->t_state == T_FINISHED); 512 513 ext4_process_freed_data(sb, txn->t_tid); 514 ext4_maybe_update_superblock(sb); 515 516 spin_lock(&sbi->s_md_lock); 517 while (!list_empty(&txn->t_private_list)) { 518 jce = list_entry(txn->t_private_list.next, 519 struct ext4_journal_cb_entry, jce_list); 520 list_del_init(&jce->jce_list); 521 spin_unlock(&sbi->s_md_lock); 522 jce->jce_func(sb, jce, error); 523 spin_lock(&sbi->s_md_lock); 524 } 525 spin_unlock(&sbi->s_md_lock); 526 } 527 528 /* 529 * This writepage callback for write_cache_pages() 530 * takes care of a few cases after page cleaning. 531 * 532 * write_cache_pages() already checks for dirty pages 533 * and calls clear_page_dirty_for_io(), which we want, 534 * to write protect the pages. 535 * 536 * However, we may have to redirty a page (see below.) 537 */ 538 static int ext4_journalled_writepage_callback(struct folio *folio, 539 struct writeback_control *wbc, 540 void *data) 541 { 542 transaction_t *transaction = (transaction_t *) data; 543 struct buffer_head *bh, *head; 544 struct journal_head *jh; 545 546 bh = head = folio_buffers(folio); 547 do { 548 /* 549 * We have to redirty a page in these cases: 550 * 1) If buffer is dirty, it means the page was dirty because it 551 * contains a buffer that needs checkpointing. So the dirty bit 552 * needs to be preserved so that checkpointing writes the buffer 553 * properly. 554 * 2) If buffer is not part of the committing transaction 555 * (we may have just accidentally come across this buffer because 556 * inode range tracking is not exact) or if the currently running 557 * transaction already contains this buffer as well, dirty bit 558 * needs to be preserved so that the buffer gets writeprotected 559 * properly on running transaction's commit. 560 */ 561 jh = bh2jh(bh); 562 if (buffer_dirty(bh) || 563 (jh && (jh->b_transaction != transaction || 564 jh->b_next_transaction))) { 565 folio_redirty_for_writepage(wbc, folio); 566 goto out; 567 } 568 } while ((bh = bh->b_this_page) != head); 569 570 out: 571 return AOP_WRITEPAGE_ACTIVATE; 572 } 573 574 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 575 { 576 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 577 struct writeback_control wbc = { 578 .sync_mode = WB_SYNC_ALL, 579 .nr_to_write = LONG_MAX, 580 .range_start = jinode->i_dirty_start, 581 .range_end = jinode->i_dirty_end, 582 }; 583 584 return write_cache_pages(mapping, &wbc, 585 ext4_journalled_writepage_callback, 586 jinode->i_transaction); 587 } 588 589 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 590 { 591 int ret; 592 593 if (ext4_should_journal_data(jinode->i_vfs_inode)) 594 ret = ext4_journalled_submit_inode_data_buffers(jinode); 595 else 596 ret = ext4_normal_submit_inode_data_buffers(jinode); 597 return ret; 598 } 599 600 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 601 { 602 int ret = 0; 603 604 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 605 ret = jbd2_journal_finish_inode_data_buffers(jinode); 606 607 return ret; 608 } 609 610 static bool system_going_down(void) 611 { 612 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 613 || system_state == SYSTEM_RESTART; 614 } 615 616 struct ext4_err_translation { 617 int code; 618 int errno; 619 }; 620 621 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 622 623 static struct ext4_err_translation err_translation[] = { 624 EXT4_ERR_TRANSLATE(EIO), 625 EXT4_ERR_TRANSLATE(ENOMEM), 626 EXT4_ERR_TRANSLATE(EFSBADCRC), 627 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 628 EXT4_ERR_TRANSLATE(ENOSPC), 629 EXT4_ERR_TRANSLATE(ENOKEY), 630 EXT4_ERR_TRANSLATE(EROFS), 631 EXT4_ERR_TRANSLATE(EFBIG), 632 EXT4_ERR_TRANSLATE(EEXIST), 633 EXT4_ERR_TRANSLATE(ERANGE), 634 EXT4_ERR_TRANSLATE(EOVERFLOW), 635 EXT4_ERR_TRANSLATE(EBUSY), 636 EXT4_ERR_TRANSLATE(ENOTDIR), 637 EXT4_ERR_TRANSLATE(ENOTEMPTY), 638 EXT4_ERR_TRANSLATE(ESHUTDOWN), 639 EXT4_ERR_TRANSLATE(EFAULT), 640 }; 641 642 static int ext4_errno_to_code(int errno) 643 { 644 int i; 645 646 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 647 if (err_translation[i].errno == errno) 648 return err_translation[i].code; 649 return EXT4_ERR_UNKNOWN; 650 } 651 652 static void save_error_info(struct super_block *sb, int error, 653 __u32 ino, __u64 block, 654 const char *func, unsigned int line) 655 { 656 struct ext4_sb_info *sbi = EXT4_SB(sb); 657 658 /* We default to EFSCORRUPTED error... */ 659 if (error == 0) 660 error = EFSCORRUPTED; 661 662 spin_lock(&sbi->s_error_lock); 663 sbi->s_add_error_count++; 664 sbi->s_last_error_code = error; 665 sbi->s_last_error_line = line; 666 sbi->s_last_error_ino = ino; 667 sbi->s_last_error_block = block; 668 sbi->s_last_error_func = func; 669 sbi->s_last_error_time = ktime_get_real_seconds(); 670 if (!sbi->s_first_error_time) { 671 sbi->s_first_error_code = error; 672 sbi->s_first_error_line = line; 673 sbi->s_first_error_ino = ino; 674 sbi->s_first_error_block = block; 675 sbi->s_first_error_func = func; 676 sbi->s_first_error_time = sbi->s_last_error_time; 677 } 678 spin_unlock(&sbi->s_error_lock); 679 } 680 681 /* Deal with the reporting of failure conditions on a filesystem such as 682 * inconsistencies detected or read IO failures. 683 * 684 * On ext2, we can store the error state of the filesystem in the 685 * superblock. That is not possible on ext4, because we may have other 686 * write ordering constraints on the superblock which prevent us from 687 * writing it out straight away; and given that the journal is about to 688 * be aborted, we can't rely on the current, or future, transactions to 689 * write out the superblock safely. 690 * 691 * We'll just use the jbd2_journal_abort() error code to record an error in 692 * the journal instead. On recovery, the journal will complain about 693 * that error until we've noted it down and cleared it. 694 * 695 * If force_ro is set, we unconditionally force the filesystem into an 696 * ABORT|READONLY state, unless the error response on the fs has been set to 697 * panic in which case we take the easy way out and panic immediately. This is 698 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 699 * at a critical moment in log management. 700 */ 701 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 702 __u32 ino, __u64 block, 703 const char *func, unsigned int line) 704 { 705 journal_t *journal = EXT4_SB(sb)->s_journal; 706 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 707 708 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 709 if (test_opt(sb, WARN_ON_ERROR)) 710 WARN_ON_ONCE(1); 711 712 if (!continue_fs && !sb_rdonly(sb)) { 713 set_bit(EXT4_FLAGS_SHUTDOWN, &EXT4_SB(sb)->s_ext4_flags); 714 if (journal) 715 jbd2_journal_abort(journal, -EIO); 716 } 717 718 if (!bdev_read_only(sb->s_bdev)) { 719 save_error_info(sb, error, ino, block, func, line); 720 /* 721 * In case the fs should keep running, we need to writeout 722 * superblock through the journal. Due to lock ordering 723 * constraints, it may not be safe to do it right here so we 724 * defer superblock flushing to a workqueue. 725 */ 726 if (continue_fs && journal) 727 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 728 else 729 ext4_commit_super(sb); 730 } 731 732 /* 733 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 734 * could panic during 'reboot -f' as the underlying device got already 735 * disabled. 736 */ 737 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 738 panic("EXT4-fs (device %s): panic forced after error\n", 739 sb->s_id); 740 } 741 742 if (sb_rdonly(sb) || continue_fs) 743 return; 744 745 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 746 /* 747 * Make sure updated value of ->s_mount_flags will be visible before 748 * ->s_flags update 749 */ 750 smp_wmb(); 751 sb->s_flags |= SB_RDONLY; 752 } 753 754 static void update_super_work(struct work_struct *work) 755 { 756 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 757 s_sb_upd_work); 758 journal_t *journal = sbi->s_journal; 759 handle_t *handle; 760 761 /* 762 * If the journal is still running, we have to write out superblock 763 * through the journal to avoid collisions of other journalled sb 764 * updates. 765 * 766 * We use directly jbd2 functions here to avoid recursing back into 767 * ext4 error handling code during handling of previous errors. 768 */ 769 if (!sb_rdonly(sbi->s_sb) && journal) { 770 struct buffer_head *sbh = sbi->s_sbh; 771 bool call_notify_err = false; 772 773 handle = jbd2_journal_start(journal, 1); 774 if (IS_ERR(handle)) 775 goto write_directly; 776 if (jbd2_journal_get_write_access(handle, sbh)) { 777 jbd2_journal_stop(handle); 778 goto write_directly; 779 } 780 781 if (sbi->s_add_error_count > 0) 782 call_notify_err = true; 783 784 ext4_update_super(sbi->s_sb); 785 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 786 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 787 "superblock detected"); 788 clear_buffer_write_io_error(sbh); 789 set_buffer_uptodate(sbh); 790 } 791 792 if (jbd2_journal_dirty_metadata(handle, sbh)) { 793 jbd2_journal_stop(handle); 794 goto write_directly; 795 } 796 jbd2_journal_stop(handle); 797 798 if (call_notify_err) 799 ext4_notify_error_sysfs(sbi); 800 801 return; 802 } 803 write_directly: 804 /* 805 * Write through journal failed. Write sb directly to get error info 806 * out and hope for the best. 807 */ 808 ext4_commit_super(sbi->s_sb); 809 ext4_notify_error_sysfs(sbi); 810 } 811 812 #define ext4_error_ratelimit(sb) \ 813 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 814 "EXT4-fs error") 815 816 void __ext4_error(struct super_block *sb, const char *function, 817 unsigned int line, bool force_ro, int error, __u64 block, 818 const char *fmt, ...) 819 { 820 struct va_format vaf; 821 va_list args; 822 823 if (unlikely(ext4_forced_shutdown(sb))) 824 return; 825 826 trace_ext4_error(sb, function, line); 827 if (ext4_error_ratelimit(sb)) { 828 va_start(args, fmt); 829 vaf.fmt = fmt; 830 vaf.va = &args; 831 printk(KERN_CRIT 832 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 833 sb->s_id, function, line, current->comm, &vaf); 834 va_end(args); 835 } 836 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED); 837 838 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 839 } 840 841 void __ext4_error_inode(struct inode *inode, const char *function, 842 unsigned int line, ext4_fsblk_t block, int error, 843 const char *fmt, ...) 844 { 845 va_list args; 846 struct va_format vaf; 847 848 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 849 return; 850 851 trace_ext4_error(inode->i_sb, function, line); 852 if (ext4_error_ratelimit(inode->i_sb)) { 853 va_start(args, fmt); 854 vaf.fmt = fmt; 855 vaf.va = &args; 856 if (block) 857 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 858 "inode #%lu: block %llu: comm %s: %pV\n", 859 inode->i_sb->s_id, function, line, inode->i_ino, 860 block, current->comm, &vaf); 861 else 862 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 863 "inode #%lu: comm %s: %pV\n", 864 inode->i_sb->s_id, function, line, inode->i_ino, 865 current->comm, &vaf); 866 va_end(args); 867 } 868 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED); 869 870 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 871 function, line); 872 } 873 874 void __ext4_error_file(struct file *file, const char *function, 875 unsigned int line, ext4_fsblk_t block, 876 const char *fmt, ...) 877 { 878 va_list args; 879 struct va_format vaf; 880 struct inode *inode = file_inode(file); 881 char pathname[80], *path; 882 883 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 884 return; 885 886 trace_ext4_error(inode->i_sb, function, line); 887 if (ext4_error_ratelimit(inode->i_sb)) { 888 path = file_path(file, pathname, sizeof(pathname)); 889 if (IS_ERR(path)) 890 path = "(unknown)"; 891 va_start(args, fmt); 892 vaf.fmt = fmt; 893 vaf.va = &args; 894 if (block) 895 printk(KERN_CRIT 896 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 897 "block %llu: comm %s: path %s: %pV\n", 898 inode->i_sb->s_id, function, line, inode->i_ino, 899 block, current->comm, path, &vaf); 900 else 901 printk(KERN_CRIT 902 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 903 "comm %s: path %s: %pV\n", 904 inode->i_sb->s_id, function, line, inode->i_ino, 905 current->comm, path, &vaf); 906 va_end(args); 907 } 908 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED); 909 910 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 911 function, line); 912 } 913 914 const char *ext4_decode_error(struct super_block *sb, int errno, 915 char nbuf[16]) 916 { 917 char *errstr = NULL; 918 919 switch (errno) { 920 case -EFSCORRUPTED: 921 errstr = "Corrupt filesystem"; 922 break; 923 case -EFSBADCRC: 924 errstr = "Filesystem failed CRC"; 925 break; 926 case -EIO: 927 errstr = "IO failure"; 928 break; 929 case -ENOMEM: 930 errstr = "Out of memory"; 931 break; 932 case -EROFS: 933 if (!sb || (EXT4_SB(sb)->s_journal && 934 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 935 errstr = "Journal has aborted"; 936 else 937 errstr = "Readonly filesystem"; 938 break; 939 default: 940 /* If the caller passed in an extra buffer for unknown 941 * errors, textualise them now. Else we just return 942 * NULL. */ 943 if (nbuf) { 944 /* Check for truncated error codes... */ 945 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 946 errstr = nbuf; 947 } 948 break; 949 } 950 951 return errstr; 952 } 953 954 /* __ext4_std_error decodes expected errors from journaling functions 955 * automatically and invokes the appropriate error response. */ 956 957 void __ext4_std_error(struct super_block *sb, const char *function, 958 unsigned int line, int errno) 959 { 960 char nbuf[16]; 961 const char *errstr; 962 963 if (unlikely(ext4_forced_shutdown(sb))) 964 return; 965 966 /* Special case: if the error is EROFS, and we're not already 967 * inside a transaction, then there's really no point in logging 968 * an error. */ 969 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 970 return; 971 972 if (ext4_error_ratelimit(sb)) { 973 errstr = ext4_decode_error(sb, errno, nbuf); 974 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 975 sb->s_id, function, line, errstr); 976 } 977 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED); 978 979 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 980 } 981 982 void __ext4_msg(struct super_block *sb, 983 const char *prefix, const char *fmt, ...) 984 { 985 struct va_format vaf; 986 va_list args; 987 988 if (sb) { 989 atomic_inc(&EXT4_SB(sb)->s_msg_count); 990 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 991 "EXT4-fs")) 992 return; 993 } 994 995 va_start(args, fmt); 996 vaf.fmt = fmt; 997 vaf.va = &args; 998 if (sb) 999 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 1000 else 1001 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 1002 va_end(args); 1003 } 1004 1005 static int ext4_warning_ratelimit(struct super_block *sb) 1006 { 1007 atomic_inc(&EXT4_SB(sb)->s_warning_count); 1008 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 1009 "EXT4-fs warning"); 1010 } 1011 1012 void __ext4_warning(struct super_block *sb, const char *function, 1013 unsigned int line, const char *fmt, ...) 1014 { 1015 struct va_format vaf; 1016 va_list args; 1017 1018 if (!ext4_warning_ratelimit(sb)) 1019 return; 1020 1021 va_start(args, fmt); 1022 vaf.fmt = fmt; 1023 vaf.va = &args; 1024 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 1025 sb->s_id, function, line, &vaf); 1026 va_end(args); 1027 } 1028 1029 void __ext4_warning_inode(const struct inode *inode, const char *function, 1030 unsigned int line, const char *fmt, ...) 1031 { 1032 struct va_format vaf; 1033 va_list args; 1034 1035 if (!ext4_warning_ratelimit(inode->i_sb)) 1036 return; 1037 1038 va_start(args, fmt); 1039 vaf.fmt = fmt; 1040 vaf.va = &args; 1041 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 1042 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 1043 function, line, inode->i_ino, current->comm, &vaf); 1044 va_end(args); 1045 } 1046 1047 void __ext4_grp_locked_error(const char *function, unsigned int line, 1048 struct super_block *sb, ext4_group_t grp, 1049 unsigned long ino, ext4_fsblk_t block, 1050 const char *fmt, ...) 1051 __releases(bitlock) 1052 __acquires(bitlock) 1053 { 1054 struct va_format vaf; 1055 va_list args; 1056 1057 if (unlikely(ext4_forced_shutdown(sb))) 1058 return; 1059 1060 trace_ext4_error(sb, function, line); 1061 if (ext4_error_ratelimit(sb)) { 1062 va_start(args, fmt); 1063 vaf.fmt = fmt; 1064 vaf.va = &args; 1065 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1066 sb->s_id, function, line, grp); 1067 if (ino) 1068 printk(KERN_CONT "inode %lu: ", ino); 1069 if (block) 1070 printk(KERN_CONT "block %llu:", 1071 (unsigned long long) block); 1072 printk(KERN_CONT "%pV\n", &vaf); 1073 va_end(args); 1074 } 1075 1076 if (test_opt(sb, ERRORS_CONT)) { 1077 if (test_opt(sb, WARN_ON_ERROR)) 1078 WARN_ON_ONCE(1); 1079 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1080 if (!bdev_read_only(sb->s_bdev)) { 1081 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1082 line); 1083 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 1084 } 1085 return; 1086 } 1087 ext4_unlock_group(sb, grp); 1088 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1089 /* 1090 * We only get here in the ERRORS_RO case; relocking the group 1091 * may be dangerous, but nothing bad will happen since the 1092 * filesystem will have already been marked read/only and the 1093 * journal has been aborted. We return 1 as a hint to callers 1094 * who might what to use the return value from 1095 * ext4_grp_locked_error() to distinguish between the 1096 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1097 * aggressively from the ext4 function in question, with a 1098 * more appropriate error code. 1099 */ 1100 ext4_lock_group(sb, grp); 1101 return; 1102 } 1103 1104 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1105 ext4_group_t group, 1106 unsigned int flags) 1107 { 1108 struct ext4_sb_info *sbi = EXT4_SB(sb); 1109 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1110 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1111 int ret; 1112 1113 if (!grp || !gdp) 1114 return; 1115 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1116 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1117 &grp->bb_state); 1118 if (!ret) 1119 percpu_counter_sub(&sbi->s_freeclusters_counter, 1120 grp->bb_free); 1121 } 1122 1123 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1124 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1125 &grp->bb_state); 1126 if (!ret && gdp) { 1127 int count; 1128 1129 count = ext4_free_inodes_count(sb, gdp); 1130 percpu_counter_sub(&sbi->s_freeinodes_counter, 1131 count); 1132 } 1133 } 1134 } 1135 1136 void ext4_update_dynamic_rev(struct super_block *sb) 1137 { 1138 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1139 1140 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1141 return; 1142 1143 ext4_warning(sb, 1144 "updating to rev %d because of new feature flag, " 1145 "running e2fsck is recommended", 1146 EXT4_DYNAMIC_REV); 1147 1148 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1149 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1150 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1151 /* leave es->s_feature_*compat flags alone */ 1152 /* es->s_uuid will be set by e2fsck if empty */ 1153 1154 /* 1155 * The rest of the superblock fields should be zero, and if not it 1156 * means they are likely already in use, so leave them alone. We 1157 * can leave it up to e2fsck to clean up any inconsistencies there. 1158 */ 1159 } 1160 1161 static inline struct inode *orphan_list_entry(struct list_head *l) 1162 { 1163 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1164 } 1165 1166 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1167 { 1168 struct list_head *l; 1169 1170 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1171 le32_to_cpu(sbi->s_es->s_last_orphan)); 1172 1173 printk(KERN_ERR "sb_info orphan list:\n"); 1174 list_for_each(l, &sbi->s_orphan) { 1175 struct inode *inode = orphan_list_entry(l); 1176 printk(KERN_ERR " " 1177 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1178 inode->i_sb->s_id, inode->i_ino, inode, 1179 inode->i_mode, inode->i_nlink, 1180 NEXT_ORPHAN(inode)); 1181 } 1182 } 1183 1184 #ifdef CONFIG_QUOTA 1185 static int ext4_quota_off(struct super_block *sb, int type); 1186 1187 static inline void ext4_quotas_off(struct super_block *sb, int type) 1188 { 1189 BUG_ON(type > EXT4_MAXQUOTAS); 1190 1191 /* Use our quota_off function to clear inode flags etc. */ 1192 for (type--; type >= 0; type--) 1193 ext4_quota_off(sb, type); 1194 } 1195 1196 /* 1197 * This is a helper function which is used in the mount/remount 1198 * codepaths (which holds s_umount) to fetch the quota file name. 1199 */ 1200 static inline char *get_qf_name(struct super_block *sb, 1201 struct ext4_sb_info *sbi, 1202 int type) 1203 { 1204 return rcu_dereference_protected(sbi->s_qf_names[type], 1205 lockdep_is_held(&sb->s_umount)); 1206 } 1207 #else 1208 static inline void ext4_quotas_off(struct super_block *sb, int type) 1209 { 1210 } 1211 #endif 1212 1213 static int ext4_percpu_param_init(struct ext4_sb_info *sbi) 1214 { 1215 ext4_fsblk_t block; 1216 int err; 1217 1218 block = ext4_count_free_clusters(sbi->s_sb); 1219 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block)); 1220 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 1221 GFP_KERNEL); 1222 if (!err) { 1223 unsigned long freei = ext4_count_free_inodes(sbi->s_sb); 1224 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 1225 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 1226 GFP_KERNEL); 1227 } 1228 if (!err) 1229 err = percpu_counter_init(&sbi->s_dirs_counter, 1230 ext4_count_dirs(sbi->s_sb), GFP_KERNEL); 1231 if (!err) 1232 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 1233 GFP_KERNEL); 1234 if (!err) 1235 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 1236 GFP_KERNEL); 1237 if (!err) 1238 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 1239 1240 if (err) 1241 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory"); 1242 1243 return err; 1244 } 1245 1246 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi) 1247 { 1248 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1249 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1250 percpu_counter_destroy(&sbi->s_dirs_counter); 1251 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1252 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1253 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1254 } 1255 1256 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 1257 { 1258 struct buffer_head **group_desc; 1259 int i; 1260 1261 rcu_read_lock(); 1262 group_desc = rcu_dereference(sbi->s_group_desc); 1263 for (i = 0; i < sbi->s_gdb_count; i++) 1264 brelse(group_desc[i]); 1265 kvfree(group_desc); 1266 rcu_read_unlock(); 1267 } 1268 1269 static void ext4_flex_groups_free(struct ext4_sb_info *sbi) 1270 { 1271 struct flex_groups **flex_groups; 1272 int i; 1273 1274 rcu_read_lock(); 1275 flex_groups = rcu_dereference(sbi->s_flex_groups); 1276 if (flex_groups) { 1277 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1278 kvfree(flex_groups[i]); 1279 kvfree(flex_groups); 1280 } 1281 rcu_read_unlock(); 1282 } 1283 1284 static void ext4_put_super(struct super_block *sb) 1285 { 1286 struct ext4_sb_info *sbi = EXT4_SB(sb); 1287 struct ext4_super_block *es = sbi->s_es; 1288 int aborted = 0; 1289 int err; 1290 1291 /* 1292 * Unregister sysfs before destroying jbd2 journal. 1293 * Since we could still access attr_journal_task attribute via sysfs 1294 * path which could have sbi->s_journal->j_task as NULL 1295 * Unregister sysfs before flush sbi->s_sb_upd_work. 1296 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1297 * read metadata verify failed then will queue error work. 1298 * update_super_work will call start_this_handle may trigger 1299 * BUG_ON. 1300 */ 1301 ext4_unregister_sysfs(sb); 1302 1303 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1304 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.", 1305 &sb->s_uuid); 1306 1307 ext4_unregister_li_request(sb); 1308 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 1309 1310 flush_work(&sbi->s_sb_upd_work); 1311 destroy_workqueue(sbi->rsv_conversion_wq); 1312 ext4_release_orphan_info(sb); 1313 1314 if (sbi->s_journal) { 1315 aborted = is_journal_aborted(sbi->s_journal); 1316 err = jbd2_journal_destroy(sbi->s_journal); 1317 sbi->s_journal = NULL; 1318 if ((err < 0) && !aborted) { 1319 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1320 } 1321 } 1322 1323 ext4_es_unregister_shrinker(sbi); 1324 timer_shutdown_sync(&sbi->s_err_report); 1325 ext4_release_system_zone(sb); 1326 ext4_mb_release(sb); 1327 ext4_ext_release(sb); 1328 1329 if (!sb_rdonly(sb) && !aborted) { 1330 ext4_clear_feature_journal_needs_recovery(sb); 1331 ext4_clear_feature_orphan_present(sb); 1332 es->s_state = cpu_to_le16(sbi->s_mount_state); 1333 } 1334 if (!sb_rdonly(sb)) 1335 ext4_commit_super(sb); 1336 1337 ext4_group_desc_free(sbi); 1338 ext4_flex_groups_free(sbi); 1339 ext4_percpu_param_destroy(sbi); 1340 #ifdef CONFIG_QUOTA 1341 for (int i = 0; i < EXT4_MAXQUOTAS; i++) 1342 kfree(get_qf_name(sb, sbi, i)); 1343 #endif 1344 1345 /* Debugging code just in case the in-memory inode orphan list 1346 * isn't empty. The on-disk one can be non-empty if we've 1347 * detected an error and taken the fs readonly, but the 1348 * in-memory list had better be clean by this point. */ 1349 if (!list_empty(&sbi->s_orphan)) 1350 dump_orphan_list(sb, sbi); 1351 ASSERT(list_empty(&sbi->s_orphan)); 1352 1353 sync_blockdev(sb->s_bdev); 1354 invalidate_bdev(sb->s_bdev); 1355 if (sbi->s_journal_bdev) { 1356 /* 1357 * Invalidate the journal device's buffers. We don't want them 1358 * floating about in memory - the physical journal device may 1359 * hotswapped, and it breaks the `ro-after' testing code. 1360 */ 1361 sync_blockdev(sbi->s_journal_bdev); 1362 invalidate_bdev(sbi->s_journal_bdev); 1363 } 1364 1365 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1366 sbi->s_ea_inode_cache = NULL; 1367 1368 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1369 sbi->s_ea_block_cache = NULL; 1370 1371 ext4_stop_mmpd(sbi); 1372 1373 brelse(sbi->s_sbh); 1374 sb->s_fs_info = NULL; 1375 /* 1376 * Now that we are completely done shutting down the 1377 * superblock, we need to actually destroy the kobject. 1378 */ 1379 kobject_put(&sbi->s_kobj); 1380 wait_for_completion(&sbi->s_kobj_unregister); 1381 if (sbi->s_chksum_driver) 1382 crypto_free_shash(sbi->s_chksum_driver); 1383 kfree(sbi->s_blockgroup_lock); 1384 fs_put_dax(sbi->s_daxdev, NULL); 1385 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1386 #if IS_ENABLED(CONFIG_UNICODE) 1387 utf8_unload(sb->s_encoding); 1388 #endif 1389 kfree(sbi); 1390 } 1391 1392 static struct kmem_cache *ext4_inode_cachep; 1393 1394 /* 1395 * Called inside transaction, so use GFP_NOFS 1396 */ 1397 static struct inode *ext4_alloc_inode(struct super_block *sb) 1398 { 1399 struct ext4_inode_info *ei; 1400 1401 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1402 if (!ei) 1403 return NULL; 1404 1405 inode_set_iversion(&ei->vfs_inode, 1); 1406 ei->i_flags = 0; 1407 spin_lock_init(&ei->i_raw_lock); 1408 ei->i_prealloc_node = RB_ROOT; 1409 atomic_set(&ei->i_prealloc_active, 0); 1410 rwlock_init(&ei->i_prealloc_lock); 1411 ext4_es_init_tree(&ei->i_es_tree); 1412 rwlock_init(&ei->i_es_lock); 1413 INIT_LIST_HEAD(&ei->i_es_list); 1414 ei->i_es_all_nr = 0; 1415 ei->i_es_shk_nr = 0; 1416 ei->i_es_shrink_lblk = 0; 1417 ei->i_reserved_data_blocks = 0; 1418 spin_lock_init(&(ei->i_block_reservation_lock)); 1419 ext4_init_pending_tree(&ei->i_pending_tree); 1420 #ifdef CONFIG_QUOTA 1421 ei->i_reserved_quota = 0; 1422 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1423 #endif 1424 ei->jinode = NULL; 1425 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1426 spin_lock_init(&ei->i_completed_io_lock); 1427 ei->i_sync_tid = 0; 1428 ei->i_datasync_tid = 0; 1429 atomic_set(&ei->i_unwritten, 0); 1430 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1431 ext4_fc_init_inode(&ei->vfs_inode); 1432 mutex_init(&ei->i_fc_lock); 1433 return &ei->vfs_inode; 1434 } 1435 1436 static int ext4_drop_inode(struct inode *inode) 1437 { 1438 int drop = generic_drop_inode(inode); 1439 1440 if (!drop) 1441 drop = fscrypt_drop_inode(inode); 1442 1443 trace_ext4_drop_inode(inode, drop); 1444 return drop; 1445 } 1446 1447 static void ext4_free_in_core_inode(struct inode *inode) 1448 { 1449 fscrypt_free_inode(inode); 1450 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1451 pr_warn("%s: inode %ld still in fc list", 1452 __func__, inode->i_ino); 1453 } 1454 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1455 } 1456 1457 static void ext4_destroy_inode(struct inode *inode) 1458 { 1459 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1460 ext4_msg(inode->i_sb, KERN_ERR, 1461 "Inode %lu (%p): orphan list check failed!", 1462 inode->i_ino, EXT4_I(inode)); 1463 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1464 EXT4_I(inode), sizeof(struct ext4_inode_info), 1465 true); 1466 dump_stack(); 1467 } 1468 1469 if (EXT4_I(inode)->i_reserved_data_blocks) 1470 ext4_msg(inode->i_sb, KERN_ERR, 1471 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!", 1472 inode->i_ino, EXT4_I(inode), 1473 EXT4_I(inode)->i_reserved_data_blocks); 1474 } 1475 1476 static void ext4_shutdown(struct super_block *sb) 1477 { 1478 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH); 1479 } 1480 1481 static void init_once(void *foo) 1482 { 1483 struct ext4_inode_info *ei = foo; 1484 1485 INIT_LIST_HEAD(&ei->i_orphan); 1486 init_rwsem(&ei->xattr_sem); 1487 init_rwsem(&ei->i_data_sem); 1488 inode_init_once(&ei->vfs_inode); 1489 ext4_fc_init_inode(&ei->vfs_inode); 1490 } 1491 1492 static int __init init_inodecache(void) 1493 { 1494 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1495 sizeof(struct ext4_inode_info), 0, 1496 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1497 SLAB_ACCOUNT), 1498 offsetof(struct ext4_inode_info, i_data), 1499 sizeof_field(struct ext4_inode_info, i_data), 1500 init_once); 1501 if (ext4_inode_cachep == NULL) 1502 return -ENOMEM; 1503 return 0; 1504 } 1505 1506 static void destroy_inodecache(void) 1507 { 1508 /* 1509 * Make sure all delayed rcu free inodes are flushed before we 1510 * destroy cache. 1511 */ 1512 rcu_barrier(); 1513 kmem_cache_destroy(ext4_inode_cachep); 1514 } 1515 1516 void ext4_clear_inode(struct inode *inode) 1517 { 1518 ext4_fc_del(inode); 1519 invalidate_inode_buffers(inode); 1520 clear_inode(inode); 1521 ext4_discard_preallocations(inode, 0); 1522 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1523 dquot_drop(inode); 1524 if (EXT4_I(inode)->jinode) { 1525 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1526 EXT4_I(inode)->jinode); 1527 jbd2_free_inode(EXT4_I(inode)->jinode); 1528 EXT4_I(inode)->jinode = NULL; 1529 } 1530 fscrypt_put_encryption_info(inode); 1531 fsverity_cleanup_inode(inode); 1532 } 1533 1534 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1535 u64 ino, u32 generation) 1536 { 1537 struct inode *inode; 1538 1539 /* 1540 * Currently we don't know the generation for parent directory, so 1541 * a generation of 0 means "accept any" 1542 */ 1543 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1544 if (IS_ERR(inode)) 1545 return ERR_CAST(inode); 1546 if (generation && inode->i_generation != generation) { 1547 iput(inode); 1548 return ERR_PTR(-ESTALE); 1549 } 1550 1551 return inode; 1552 } 1553 1554 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1555 int fh_len, int fh_type) 1556 { 1557 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1558 ext4_nfs_get_inode); 1559 } 1560 1561 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1562 int fh_len, int fh_type) 1563 { 1564 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1565 ext4_nfs_get_inode); 1566 } 1567 1568 static int ext4_nfs_commit_metadata(struct inode *inode) 1569 { 1570 struct writeback_control wbc = { 1571 .sync_mode = WB_SYNC_ALL 1572 }; 1573 1574 trace_ext4_nfs_commit_metadata(inode); 1575 return ext4_write_inode(inode, &wbc); 1576 } 1577 1578 #ifdef CONFIG_QUOTA 1579 static const char * const quotatypes[] = INITQFNAMES; 1580 #define QTYPE2NAME(t) (quotatypes[t]) 1581 1582 static int ext4_write_dquot(struct dquot *dquot); 1583 static int ext4_acquire_dquot(struct dquot *dquot); 1584 static int ext4_release_dquot(struct dquot *dquot); 1585 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1586 static int ext4_write_info(struct super_block *sb, int type); 1587 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1588 const struct path *path); 1589 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1590 size_t len, loff_t off); 1591 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1592 const char *data, size_t len, loff_t off); 1593 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1594 unsigned int flags); 1595 1596 static struct dquot __rcu **ext4_get_dquots(struct inode *inode) 1597 { 1598 return EXT4_I(inode)->i_dquot; 1599 } 1600 1601 static const struct dquot_operations ext4_quota_operations = { 1602 .get_reserved_space = ext4_get_reserved_space, 1603 .write_dquot = ext4_write_dquot, 1604 .acquire_dquot = ext4_acquire_dquot, 1605 .release_dquot = ext4_release_dquot, 1606 .mark_dirty = ext4_mark_dquot_dirty, 1607 .write_info = ext4_write_info, 1608 .alloc_dquot = dquot_alloc, 1609 .destroy_dquot = dquot_destroy, 1610 .get_projid = ext4_get_projid, 1611 .get_inode_usage = ext4_get_inode_usage, 1612 .get_next_id = dquot_get_next_id, 1613 }; 1614 1615 static const struct quotactl_ops ext4_qctl_operations = { 1616 .quota_on = ext4_quota_on, 1617 .quota_off = ext4_quota_off, 1618 .quota_sync = dquot_quota_sync, 1619 .get_state = dquot_get_state, 1620 .set_info = dquot_set_dqinfo, 1621 .get_dqblk = dquot_get_dqblk, 1622 .set_dqblk = dquot_set_dqblk, 1623 .get_nextdqblk = dquot_get_next_dqblk, 1624 }; 1625 #endif 1626 1627 static const struct super_operations ext4_sops = { 1628 .alloc_inode = ext4_alloc_inode, 1629 .free_inode = ext4_free_in_core_inode, 1630 .destroy_inode = ext4_destroy_inode, 1631 .write_inode = ext4_write_inode, 1632 .dirty_inode = ext4_dirty_inode, 1633 .drop_inode = ext4_drop_inode, 1634 .evict_inode = ext4_evict_inode, 1635 .put_super = ext4_put_super, 1636 .sync_fs = ext4_sync_fs, 1637 .freeze_fs = ext4_freeze, 1638 .unfreeze_fs = ext4_unfreeze, 1639 .statfs = ext4_statfs, 1640 .show_options = ext4_show_options, 1641 .shutdown = ext4_shutdown, 1642 #ifdef CONFIG_QUOTA 1643 .quota_read = ext4_quota_read, 1644 .quota_write = ext4_quota_write, 1645 .get_dquots = ext4_get_dquots, 1646 #endif 1647 }; 1648 1649 static const struct export_operations ext4_export_ops = { 1650 .fh_to_dentry = ext4_fh_to_dentry, 1651 .fh_to_parent = ext4_fh_to_parent, 1652 .get_parent = ext4_get_parent, 1653 .commit_metadata = ext4_nfs_commit_metadata, 1654 }; 1655 1656 enum { 1657 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1658 Opt_resgid, Opt_resuid, Opt_sb, 1659 Opt_nouid32, Opt_debug, Opt_removed, 1660 Opt_user_xattr, Opt_acl, 1661 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1662 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1663 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1664 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1665 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1666 Opt_inlinecrypt, 1667 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1668 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1669 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1670 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1671 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1672 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1673 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1674 Opt_inode_readahead_blks, Opt_journal_ioprio, 1675 Opt_dioread_nolock, Opt_dioread_lock, 1676 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1677 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1678 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1679 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1680 #ifdef CONFIG_EXT4_DEBUG 1681 Opt_fc_debug_max_replay, Opt_fc_debug_force 1682 #endif 1683 }; 1684 1685 static const struct constant_table ext4_param_errors[] = { 1686 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1687 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1688 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1689 {} 1690 }; 1691 1692 static const struct constant_table ext4_param_data[] = { 1693 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1694 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1695 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1696 {} 1697 }; 1698 1699 static const struct constant_table ext4_param_data_err[] = { 1700 {"abort", Opt_data_err_abort}, 1701 {"ignore", Opt_data_err_ignore}, 1702 {} 1703 }; 1704 1705 static const struct constant_table ext4_param_jqfmt[] = { 1706 {"vfsold", QFMT_VFS_OLD}, 1707 {"vfsv0", QFMT_VFS_V0}, 1708 {"vfsv1", QFMT_VFS_V1}, 1709 {} 1710 }; 1711 1712 static const struct constant_table ext4_param_dax[] = { 1713 {"always", Opt_dax_always}, 1714 {"inode", Opt_dax_inode}, 1715 {"never", Opt_dax_never}, 1716 {} 1717 }; 1718 1719 /* String parameter that allows empty argument */ 1720 #define fsparam_string_empty(NAME, OPT) \ 1721 __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL) 1722 1723 /* 1724 * Mount option specification 1725 * We don't use fsparam_flag_no because of the way we set the 1726 * options and the way we show them in _ext4_show_options(). To 1727 * keep the changes to a minimum, let's keep the negative options 1728 * separate for now. 1729 */ 1730 static const struct fs_parameter_spec ext4_param_specs[] = { 1731 fsparam_flag ("bsddf", Opt_bsd_df), 1732 fsparam_flag ("minixdf", Opt_minix_df), 1733 fsparam_flag ("grpid", Opt_grpid), 1734 fsparam_flag ("bsdgroups", Opt_grpid), 1735 fsparam_flag ("nogrpid", Opt_nogrpid), 1736 fsparam_flag ("sysvgroups", Opt_nogrpid), 1737 fsparam_u32 ("resgid", Opt_resgid), 1738 fsparam_u32 ("resuid", Opt_resuid), 1739 fsparam_u32 ("sb", Opt_sb), 1740 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1741 fsparam_flag ("nouid32", Opt_nouid32), 1742 fsparam_flag ("debug", Opt_debug), 1743 fsparam_flag ("oldalloc", Opt_removed), 1744 fsparam_flag ("orlov", Opt_removed), 1745 fsparam_flag ("user_xattr", Opt_user_xattr), 1746 fsparam_flag ("acl", Opt_acl), 1747 fsparam_flag ("norecovery", Opt_noload), 1748 fsparam_flag ("noload", Opt_noload), 1749 fsparam_flag ("bh", Opt_removed), 1750 fsparam_flag ("nobh", Opt_removed), 1751 fsparam_u32 ("commit", Opt_commit), 1752 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1753 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1754 fsparam_u32 ("journal_dev", Opt_journal_dev), 1755 fsparam_bdev ("journal_path", Opt_journal_path), 1756 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1757 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1758 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1759 fsparam_flag ("abort", Opt_abort), 1760 fsparam_enum ("data", Opt_data, ext4_param_data), 1761 fsparam_enum ("data_err", Opt_data_err, 1762 ext4_param_data_err), 1763 fsparam_string_empty 1764 ("usrjquota", Opt_usrjquota), 1765 fsparam_string_empty 1766 ("grpjquota", Opt_grpjquota), 1767 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1768 fsparam_flag ("grpquota", Opt_grpquota), 1769 fsparam_flag ("quota", Opt_quota), 1770 fsparam_flag ("noquota", Opt_noquota), 1771 fsparam_flag ("usrquota", Opt_usrquota), 1772 fsparam_flag ("prjquota", Opt_prjquota), 1773 fsparam_flag ("barrier", Opt_barrier), 1774 fsparam_u32 ("barrier", Opt_barrier), 1775 fsparam_flag ("nobarrier", Opt_nobarrier), 1776 fsparam_flag ("i_version", Opt_removed), 1777 fsparam_flag ("dax", Opt_dax), 1778 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1779 fsparam_u32 ("stripe", Opt_stripe), 1780 fsparam_flag ("delalloc", Opt_delalloc), 1781 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1782 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1783 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1784 fsparam_u32 ("debug_want_extra_isize", 1785 Opt_debug_want_extra_isize), 1786 fsparam_flag ("mblk_io_submit", Opt_removed), 1787 fsparam_flag ("nomblk_io_submit", Opt_removed), 1788 fsparam_flag ("block_validity", Opt_block_validity), 1789 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1790 fsparam_u32 ("inode_readahead_blks", 1791 Opt_inode_readahead_blks), 1792 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1793 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1794 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1795 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1796 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1797 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1798 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1799 fsparam_flag ("discard", Opt_discard), 1800 fsparam_flag ("nodiscard", Opt_nodiscard), 1801 fsparam_u32 ("init_itable", Opt_init_itable), 1802 fsparam_flag ("init_itable", Opt_init_itable), 1803 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1804 #ifdef CONFIG_EXT4_DEBUG 1805 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1806 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1807 #endif 1808 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1809 fsparam_flag ("test_dummy_encryption", 1810 Opt_test_dummy_encryption), 1811 fsparam_string ("test_dummy_encryption", 1812 Opt_test_dummy_encryption), 1813 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1814 fsparam_flag ("nombcache", Opt_nombcache), 1815 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1816 fsparam_flag ("prefetch_block_bitmaps", 1817 Opt_removed), 1818 fsparam_flag ("no_prefetch_block_bitmaps", 1819 Opt_no_prefetch_block_bitmaps), 1820 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1821 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1822 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1823 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1824 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1825 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1826 {} 1827 }; 1828 1829 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1830 1831 #define MOPT_SET 0x0001 1832 #define MOPT_CLEAR 0x0002 1833 #define MOPT_NOSUPPORT 0x0004 1834 #define MOPT_EXPLICIT 0x0008 1835 #ifdef CONFIG_QUOTA 1836 #define MOPT_Q 0 1837 #define MOPT_QFMT 0x0010 1838 #else 1839 #define MOPT_Q MOPT_NOSUPPORT 1840 #define MOPT_QFMT MOPT_NOSUPPORT 1841 #endif 1842 #define MOPT_NO_EXT2 0x0020 1843 #define MOPT_NO_EXT3 0x0040 1844 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1845 #define MOPT_SKIP 0x0080 1846 #define MOPT_2 0x0100 1847 1848 static const struct mount_opts { 1849 int token; 1850 int mount_opt; 1851 int flags; 1852 } ext4_mount_opts[] = { 1853 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1854 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1855 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1856 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1857 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1858 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1859 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1860 MOPT_EXT4_ONLY | MOPT_SET}, 1861 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1862 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1863 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1864 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1865 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1866 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1867 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1868 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1869 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1870 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1871 {Opt_commit, 0, MOPT_NO_EXT2}, 1872 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1873 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1874 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1875 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1876 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1877 EXT4_MOUNT_JOURNAL_CHECKSUM), 1878 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1879 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1880 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1881 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1882 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1883 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1884 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1885 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1886 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1887 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1888 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1889 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1890 {Opt_data, 0, MOPT_NO_EXT2}, 1891 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1892 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1893 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1894 #else 1895 {Opt_acl, 0, MOPT_NOSUPPORT}, 1896 #endif 1897 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1898 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1899 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1900 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1901 MOPT_SET | MOPT_Q}, 1902 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1903 MOPT_SET | MOPT_Q}, 1904 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1905 MOPT_SET | MOPT_Q}, 1906 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1907 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1908 MOPT_CLEAR | MOPT_Q}, 1909 {Opt_usrjquota, 0, MOPT_Q}, 1910 {Opt_grpjquota, 0, MOPT_Q}, 1911 {Opt_jqfmt, 0, MOPT_QFMT}, 1912 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1913 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1914 MOPT_SET}, 1915 #ifdef CONFIG_EXT4_DEBUG 1916 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1917 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1918 #endif 1919 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2}, 1920 {Opt_err, 0, 0} 1921 }; 1922 1923 #if IS_ENABLED(CONFIG_UNICODE) 1924 static const struct ext4_sb_encodings { 1925 __u16 magic; 1926 char *name; 1927 unsigned int version; 1928 } ext4_sb_encoding_map[] = { 1929 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1930 }; 1931 1932 static const struct ext4_sb_encodings * 1933 ext4_sb_read_encoding(const struct ext4_super_block *es) 1934 { 1935 __u16 magic = le16_to_cpu(es->s_encoding); 1936 int i; 1937 1938 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1939 if (magic == ext4_sb_encoding_map[i].magic) 1940 return &ext4_sb_encoding_map[i]; 1941 1942 return NULL; 1943 } 1944 #endif 1945 1946 #define EXT4_SPEC_JQUOTA (1 << 0) 1947 #define EXT4_SPEC_JQFMT (1 << 1) 1948 #define EXT4_SPEC_DATAJ (1 << 2) 1949 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1950 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1951 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1952 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 1953 #define EXT4_SPEC_s_max_batch_time (1 << 8) 1954 #define EXT4_SPEC_s_min_batch_time (1 << 9) 1955 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 1956 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 1957 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 1958 #define EXT4_SPEC_s_stripe (1 << 13) 1959 #define EXT4_SPEC_s_resuid (1 << 14) 1960 #define EXT4_SPEC_s_resgid (1 << 15) 1961 #define EXT4_SPEC_s_commit_interval (1 << 16) 1962 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 1963 #define EXT4_SPEC_s_sb_block (1 << 18) 1964 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 1965 1966 struct ext4_fs_context { 1967 char *s_qf_names[EXT4_MAXQUOTAS]; 1968 struct fscrypt_dummy_policy dummy_enc_policy; 1969 int s_jquota_fmt; /* Format of quota to use */ 1970 #ifdef CONFIG_EXT4_DEBUG 1971 int s_fc_debug_max_replay; 1972 #endif 1973 unsigned short qname_spec; 1974 unsigned long vals_s_flags; /* Bits to set in s_flags */ 1975 unsigned long mask_s_flags; /* Bits changed in s_flags */ 1976 unsigned long journal_devnum; 1977 unsigned long s_commit_interval; 1978 unsigned long s_stripe; 1979 unsigned int s_inode_readahead_blks; 1980 unsigned int s_want_extra_isize; 1981 unsigned int s_li_wait_mult; 1982 unsigned int s_max_dir_size_kb; 1983 unsigned int journal_ioprio; 1984 unsigned int vals_s_mount_opt; 1985 unsigned int mask_s_mount_opt; 1986 unsigned int vals_s_mount_opt2; 1987 unsigned int mask_s_mount_opt2; 1988 unsigned int opt_flags; /* MOPT flags */ 1989 unsigned int spec; 1990 u32 s_max_batch_time; 1991 u32 s_min_batch_time; 1992 kuid_t s_resuid; 1993 kgid_t s_resgid; 1994 ext4_fsblk_t s_sb_block; 1995 }; 1996 1997 static void ext4_fc_free(struct fs_context *fc) 1998 { 1999 struct ext4_fs_context *ctx = fc->fs_private; 2000 int i; 2001 2002 if (!ctx) 2003 return; 2004 2005 for (i = 0; i < EXT4_MAXQUOTAS; i++) 2006 kfree(ctx->s_qf_names[i]); 2007 2008 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 2009 kfree(ctx); 2010 } 2011 2012 int ext4_init_fs_context(struct fs_context *fc) 2013 { 2014 struct ext4_fs_context *ctx; 2015 2016 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2017 if (!ctx) 2018 return -ENOMEM; 2019 2020 fc->fs_private = ctx; 2021 fc->ops = &ext4_context_ops; 2022 2023 return 0; 2024 } 2025 2026 #ifdef CONFIG_QUOTA 2027 /* 2028 * Note the name of the specified quota file. 2029 */ 2030 static int note_qf_name(struct fs_context *fc, int qtype, 2031 struct fs_parameter *param) 2032 { 2033 struct ext4_fs_context *ctx = fc->fs_private; 2034 char *qname; 2035 2036 if (param->size < 1) { 2037 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 2038 return -EINVAL; 2039 } 2040 if (strchr(param->string, '/')) { 2041 ext4_msg(NULL, KERN_ERR, 2042 "quotafile must be on filesystem root"); 2043 return -EINVAL; 2044 } 2045 if (ctx->s_qf_names[qtype]) { 2046 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 2047 ext4_msg(NULL, KERN_ERR, 2048 "%s quota file already specified", 2049 QTYPE2NAME(qtype)); 2050 return -EINVAL; 2051 } 2052 return 0; 2053 } 2054 2055 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 2056 if (!qname) { 2057 ext4_msg(NULL, KERN_ERR, 2058 "Not enough memory for storing quotafile name"); 2059 return -ENOMEM; 2060 } 2061 ctx->s_qf_names[qtype] = qname; 2062 ctx->qname_spec |= 1 << qtype; 2063 ctx->spec |= EXT4_SPEC_JQUOTA; 2064 return 0; 2065 } 2066 2067 /* 2068 * Clear the name of the specified quota file. 2069 */ 2070 static int unnote_qf_name(struct fs_context *fc, int qtype) 2071 { 2072 struct ext4_fs_context *ctx = fc->fs_private; 2073 2074 if (ctx->s_qf_names[qtype]) 2075 kfree(ctx->s_qf_names[qtype]); 2076 2077 ctx->s_qf_names[qtype] = NULL; 2078 ctx->qname_spec |= 1 << qtype; 2079 ctx->spec |= EXT4_SPEC_JQUOTA; 2080 return 0; 2081 } 2082 #endif 2083 2084 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2085 struct ext4_fs_context *ctx) 2086 { 2087 int err; 2088 2089 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2090 ext4_msg(NULL, KERN_WARNING, 2091 "test_dummy_encryption option not supported"); 2092 return -EINVAL; 2093 } 2094 err = fscrypt_parse_test_dummy_encryption(param, 2095 &ctx->dummy_enc_policy); 2096 if (err == -EINVAL) { 2097 ext4_msg(NULL, KERN_WARNING, 2098 "Value of option \"%s\" is unrecognized", param->key); 2099 } else if (err == -EEXIST) { 2100 ext4_msg(NULL, KERN_WARNING, 2101 "Conflicting test_dummy_encryption options"); 2102 return -EINVAL; 2103 } 2104 return err; 2105 } 2106 2107 #define EXT4_SET_CTX(name) \ 2108 static inline void ctx_set_##name(struct ext4_fs_context *ctx, \ 2109 unsigned long flag) \ 2110 { \ 2111 ctx->mask_s_##name |= flag; \ 2112 ctx->vals_s_##name |= flag; \ 2113 } 2114 2115 #define EXT4_CLEAR_CTX(name) \ 2116 static inline void ctx_clear_##name(struct ext4_fs_context *ctx, \ 2117 unsigned long flag) \ 2118 { \ 2119 ctx->mask_s_##name |= flag; \ 2120 ctx->vals_s_##name &= ~flag; \ 2121 } 2122 2123 #define EXT4_TEST_CTX(name) \ 2124 static inline unsigned long \ 2125 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2126 { \ 2127 return (ctx->vals_s_##name & flag); \ 2128 } 2129 2130 EXT4_SET_CTX(flags); /* set only */ 2131 EXT4_SET_CTX(mount_opt); 2132 EXT4_CLEAR_CTX(mount_opt); 2133 EXT4_TEST_CTX(mount_opt); 2134 EXT4_SET_CTX(mount_opt2); 2135 EXT4_CLEAR_CTX(mount_opt2); 2136 EXT4_TEST_CTX(mount_opt2); 2137 2138 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2139 { 2140 struct ext4_fs_context *ctx = fc->fs_private; 2141 struct fs_parse_result result; 2142 const struct mount_opts *m; 2143 int is_remount; 2144 kuid_t uid; 2145 kgid_t gid; 2146 int token; 2147 2148 token = fs_parse(fc, ext4_param_specs, param, &result); 2149 if (token < 0) 2150 return token; 2151 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2152 2153 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2154 if (token == m->token) 2155 break; 2156 2157 ctx->opt_flags |= m->flags; 2158 2159 if (m->flags & MOPT_EXPLICIT) { 2160 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2161 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2162 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2163 ctx_set_mount_opt2(ctx, 2164 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2165 } else 2166 return -EINVAL; 2167 } 2168 2169 if (m->flags & MOPT_NOSUPPORT) { 2170 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2171 param->key); 2172 return 0; 2173 } 2174 2175 switch (token) { 2176 #ifdef CONFIG_QUOTA 2177 case Opt_usrjquota: 2178 if (!*param->string) 2179 return unnote_qf_name(fc, USRQUOTA); 2180 else 2181 return note_qf_name(fc, USRQUOTA, param); 2182 case Opt_grpjquota: 2183 if (!*param->string) 2184 return unnote_qf_name(fc, GRPQUOTA); 2185 else 2186 return note_qf_name(fc, GRPQUOTA, param); 2187 #endif 2188 case Opt_sb: 2189 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2190 ext4_msg(NULL, KERN_WARNING, 2191 "Ignoring %s option on remount", param->key); 2192 } else { 2193 ctx->s_sb_block = result.uint_32; 2194 ctx->spec |= EXT4_SPEC_s_sb_block; 2195 } 2196 return 0; 2197 case Opt_removed: 2198 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2199 param->key); 2200 return 0; 2201 case Opt_inlinecrypt: 2202 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2203 ctx_set_flags(ctx, SB_INLINECRYPT); 2204 #else 2205 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2206 #endif 2207 return 0; 2208 case Opt_errors: 2209 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2210 ctx_set_mount_opt(ctx, result.uint_32); 2211 return 0; 2212 #ifdef CONFIG_QUOTA 2213 case Opt_jqfmt: 2214 ctx->s_jquota_fmt = result.uint_32; 2215 ctx->spec |= EXT4_SPEC_JQFMT; 2216 return 0; 2217 #endif 2218 case Opt_data: 2219 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2220 ctx_set_mount_opt(ctx, result.uint_32); 2221 ctx->spec |= EXT4_SPEC_DATAJ; 2222 return 0; 2223 case Opt_commit: 2224 if (result.uint_32 == 0) 2225 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE; 2226 else if (result.uint_32 > INT_MAX / HZ) { 2227 ext4_msg(NULL, KERN_ERR, 2228 "Invalid commit interval %d, " 2229 "must be smaller than %d", 2230 result.uint_32, INT_MAX / HZ); 2231 return -EINVAL; 2232 } 2233 ctx->s_commit_interval = HZ * result.uint_32; 2234 ctx->spec |= EXT4_SPEC_s_commit_interval; 2235 return 0; 2236 case Opt_debug_want_extra_isize: 2237 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2238 ext4_msg(NULL, KERN_ERR, 2239 "Invalid want_extra_isize %d", result.uint_32); 2240 return -EINVAL; 2241 } 2242 ctx->s_want_extra_isize = result.uint_32; 2243 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2244 return 0; 2245 case Opt_max_batch_time: 2246 ctx->s_max_batch_time = result.uint_32; 2247 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2248 return 0; 2249 case Opt_min_batch_time: 2250 ctx->s_min_batch_time = result.uint_32; 2251 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2252 return 0; 2253 case Opt_inode_readahead_blks: 2254 if (result.uint_32 && 2255 (result.uint_32 > (1 << 30) || 2256 !is_power_of_2(result.uint_32))) { 2257 ext4_msg(NULL, KERN_ERR, 2258 "EXT4-fs: inode_readahead_blks must be " 2259 "0 or a power of 2 smaller than 2^31"); 2260 return -EINVAL; 2261 } 2262 ctx->s_inode_readahead_blks = result.uint_32; 2263 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2264 return 0; 2265 case Opt_init_itable: 2266 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2267 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2268 if (param->type == fs_value_is_string) 2269 ctx->s_li_wait_mult = result.uint_32; 2270 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2271 return 0; 2272 case Opt_max_dir_size_kb: 2273 ctx->s_max_dir_size_kb = result.uint_32; 2274 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2275 return 0; 2276 #ifdef CONFIG_EXT4_DEBUG 2277 case Opt_fc_debug_max_replay: 2278 ctx->s_fc_debug_max_replay = result.uint_32; 2279 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2280 return 0; 2281 #endif 2282 case Opt_stripe: 2283 ctx->s_stripe = result.uint_32; 2284 ctx->spec |= EXT4_SPEC_s_stripe; 2285 return 0; 2286 case Opt_resuid: 2287 uid = make_kuid(current_user_ns(), result.uint_32); 2288 if (!uid_valid(uid)) { 2289 ext4_msg(NULL, KERN_ERR, "Invalid uid value %d", 2290 result.uint_32); 2291 return -EINVAL; 2292 } 2293 ctx->s_resuid = uid; 2294 ctx->spec |= EXT4_SPEC_s_resuid; 2295 return 0; 2296 case Opt_resgid: 2297 gid = make_kgid(current_user_ns(), result.uint_32); 2298 if (!gid_valid(gid)) { 2299 ext4_msg(NULL, KERN_ERR, "Invalid gid value %d", 2300 result.uint_32); 2301 return -EINVAL; 2302 } 2303 ctx->s_resgid = gid; 2304 ctx->spec |= EXT4_SPEC_s_resgid; 2305 return 0; 2306 case Opt_journal_dev: 2307 if (is_remount) { 2308 ext4_msg(NULL, KERN_ERR, 2309 "Cannot specify journal on remount"); 2310 return -EINVAL; 2311 } 2312 ctx->journal_devnum = result.uint_32; 2313 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2314 return 0; 2315 case Opt_journal_path: 2316 { 2317 struct inode *journal_inode; 2318 struct path path; 2319 int error; 2320 2321 if (is_remount) { 2322 ext4_msg(NULL, KERN_ERR, 2323 "Cannot specify journal on remount"); 2324 return -EINVAL; 2325 } 2326 2327 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path); 2328 if (error) { 2329 ext4_msg(NULL, KERN_ERR, "error: could not find " 2330 "journal device path"); 2331 return -EINVAL; 2332 } 2333 2334 journal_inode = d_inode(path.dentry); 2335 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2336 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2337 path_put(&path); 2338 return 0; 2339 } 2340 case Opt_journal_ioprio: 2341 if (result.uint_32 > 7) { 2342 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2343 " (must be 0-7)"); 2344 return -EINVAL; 2345 } 2346 ctx->journal_ioprio = 2347 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2348 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2349 return 0; 2350 case Opt_test_dummy_encryption: 2351 return ext4_parse_test_dummy_encryption(param, ctx); 2352 case Opt_dax: 2353 case Opt_dax_type: 2354 #ifdef CONFIG_FS_DAX 2355 { 2356 int type = (token == Opt_dax) ? 2357 Opt_dax : result.uint_32; 2358 2359 switch (type) { 2360 case Opt_dax: 2361 case Opt_dax_always: 2362 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2363 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2364 break; 2365 case Opt_dax_never: 2366 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2367 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2368 break; 2369 case Opt_dax_inode: 2370 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2371 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2372 /* Strictly for printing options */ 2373 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2374 break; 2375 } 2376 return 0; 2377 } 2378 #else 2379 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2380 return -EINVAL; 2381 #endif 2382 case Opt_data_err: 2383 if (result.uint_32 == Opt_data_err_abort) 2384 ctx_set_mount_opt(ctx, m->mount_opt); 2385 else if (result.uint_32 == Opt_data_err_ignore) 2386 ctx_clear_mount_opt(ctx, m->mount_opt); 2387 return 0; 2388 case Opt_mb_optimize_scan: 2389 if (result.int_32 == 1) { 2390 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2391 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2392 } else if (result.int_32 == 0) { 2393 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2394 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2395 } else { 2396 ext4_msg(NULL, KERN_WARNING, 2397 "mb_optimize_scan should be set to 0 or 1."); 2398 return -EINVAL; 2399 } 2400 return 0; 2401 } 2402 2403 /* 2404 * At this point we should only be getting options requiring MOPT_SET, 2405 * or MOPT_CLEAR. Anything else is a bug 2406 */ 2407 if (m->token == Opt_err) { 2408 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2409 param->key); 2410 WARN_ON(1); 2411 return -EINVAL; 2412 } 2413 2414 else { 2415 unsigned int set = 0; 2416 2417 if ((param->type == fs_value_is_flag) || 2418 result.uint_32 > 0) 2419 set = 1; 2420 2421 if (m->flags & MOPT_CLEAR) 2422 set = !set; 2423 else if (unlikely(!(m->flags & MOPT_SET))) { 2424 ext4_msg(NULL, KERN_WARNING, 2425 "buggy handling of option %s", 2426 param->key); 2427 WARN_ON(1); 2428 return -EINVAL; 2429 } 2430 if (m->flags & MOPT_2) { 2431 if (set != 0) 2432 ctx_set_mount_opt2(ctx, m->mount_opt); 2433 else 2434 ctx_clear_mount_opt2(ctx, m->mount_opt); 2435 } else { 2436 if (set != 0) 2437 ctx_set_mount_opt(ctx, m->mount_opt); 2438 else 2439 ctx_clear_mount_opt(ctx, m->mount_opt); 2440 } 2441 } 2442 2443 return 0; 2444 } 2445 2446 static int parse_options(struct fs_context *fc, char *options) 2447 { 2448 struct fs_parameter param; 2449 int ret; 2450 char *key; 2451 2452 if (!options) 2453 return 0; 2454 2455 while ((key = strsep(&options, ",")) != NULL) { 2456 if (*key) { 2457 size_t v_len = 0; 2458 char *value = strchr(key, '='); 2459 2460 param.type = fs_value_is_flag; 2461 param.string = NULL; 2462 2463 if (value) { 2464 if (value == key) 2465 continue; 2466 2467 *value++ = 0; 2468 v_len = strlen(value); 2469 param.string = kmemdup_nul(value, v_len, 2470 GFP_KERNEL); 2471 if (!param.string) 2472 return -ENOMEM; 2473 param.type = fs_value_is_string; 2474 } 2475 2476 param.key = key; 2477 param.size = v_len; 2478 2479 ret = ext4_parse_param(fc, ¶m); 2480 if (param.string) 2481 kfree(param.string); 2482 if (ret < 0) 2483 return ret; 2484 } 2485 } 2486 2487 ret = ext4_validate_options(fc); 2488 if (ret < 0) 2489 return ret; 2490 2491 return 0; 2492 } 2493 2494 static int parse_apply_sb_mount_options(struct super_block *sb, 2495 struct ext4_fs_context *m_ctx) 2496 { 2497 struct ext4_sb_info *sbi = EXT4_SB(sb); 2498 char *s_mount_opts = NULL; 2499 struct ext4_fs_context *s_ctx = NULL; 2500 struct fs_context *fc = NULL; 2501 int ret = -ENOMEM; 2502 2503 if (!sbi->s_es->s_mount_opts[0]) 2504 return 0; 2505 2506 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2507 sizeof(sbi->s_es->s_mount_opts), 2508 GFP_KERNEL); 2509 if (!s_mount_opts) 2510 return ret; 2511 2512 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2513 if (!fc) 2514 goto out_free; 2515 2516 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2517 if (!s_ctx) 2518 goto out_free; 2519 2520 fc->fs_private = s_ctx; 2521 fc->s_fs_info = sbi; 2522 2523 ret = parse_options(fc, s_mount_opts); 2524 if (ret < 0) 2525 goto parse_failed; 2526 2527 ret = ext4_check_opt_consistency(fc, sb); 2528 if (ret < 0) { 2529 parse_failed: 2530 ext4_msg(sb, KERN_WARNING, 2531 "failed to parse options in superblock: %s", 2532 s_mount_opts); 2533 ret = 0; 2534 goto out_free; 2535 } 2536 2537 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2538 m_ctx->journal_devnum = s_ctx->journal_devnum; 2539 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2540 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2541 2542 ext4_apply_options(fc, sb); 2543 ret = 0; 2544 2545 out_free: 2546 if (fc) { 2547 ext4_fc_free(fc); 2548 kfree(fc); 2549 } 2550 kfree(s_mount_opts); 2551 return ret; 2552 } 2553 2554 static void ext4_apply_quota_options(struct fs_context *fc, 2555 struct super_block *sb) 2556 { 2557 #ifdef CONFIG_QUOTA 2558 bool quota_feature = ext4_has_feature_quota(sb); 2559 struct ext4_fs_context *ctx = fc->fs_private; 2560 struct ext4_sb_info *sbi = EXT4_SB(sb); 2561 char *qname; 2562 int i; 2563 2564 if (quota_feature) 2565 return; 2566 2567 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2568 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2569 if (!(ctx->qname_spec & (1 << i))) 2570 continue; 2571 2572 qname = ctx->s_qf_names[i]; /* May be NULL */ 2573 if (qname) 2574 set_opt(sb, QUOTA); 2575 ctx->s_qf_names[i] = NULL; 2576 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2577 lockdep_is_held(&sb->s_umount)); 2578 if (qname) 2579 kfree_rcu_mightsleep(qname); 2580 } 2581 } 2582 2583 if (ctx->spec & EXT4_SPEC_JQFMT) 2584 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2585 #endif 2586 } 2587 2588 /* 2589 * Check quota settings consistency. 2590 */ 2591 static int ext4_check_quota_consistency(struct fs_context *fc, 2592 struct super_block *sb) 2593 { 2594 #ifdef CONFIG_QUOTA 2595 struct ext4_fs_context *ctx = fc->fs_private; 2596 struct ext4_sb_info *sbi = EXT4_SB(sb); 2597 bool quota_feature = ext4_has_feature_quota(sb); 2598 bool quota_loaded = sb_any_quota_loaded(sb); 2599 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2600 int quota_flags, i; 2601 2602 /* 2603 * We do the test below only for project quotas. 'usrquota' and 2604 * 'grpquota' mount options are allowed even without quota feature 2605 * to support legacy quotas in quota files. 2606 */ 2607 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2608 !ext4_has_feature_project(sb)) { 2609 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2610 "Cannot enable project quota enforcement."); 2611 return -EINVAL; 2612 } 2613 2614 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2615 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2616 if (quota_loaded && 2617 ctx->mask_s_mount_opt & quota_flags && 2618 !ctx_test_mount_opt(ctx, quota_flags)) 2619 goto err_quota_change; 2620 2621 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2622 2623 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2624 if (!(ctx->qname_spec & (1 << i))) 2625 continue; 2626 2627 if (quota_loaded && 2628 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2629 goto err_jquota_change; 2630 2631 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2632 strcmp(get_qf_name(sb, sbi, i), 2633 ctx->s_qf_names[i]) != 0) 2634 goto err_jquota_specified; 2635 } 2636 2637 if (quota_feature) { 2638 ext4_msg(NULL, KERN_INFO, 2639 "Journaled quota options ignored when " 2640 "QUOTA feature is enabled"); 2641 return 0; 2642 } 2643 } 2644 2645 if (ctx->spec & EXT4_SPEC_JQFMT) { 2646 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2647 goto err_jquota_change; 2648 if (quota_feature) { 2649 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2650 "ignored when QUOTA feature is enabled"); 2651 return 0; 2652 } 2653 } 2654 2655 /* Make sure we don't mix old and new quota format */ 2656 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2657 ctx->s_qf_names[USRQUOTA]); 2658 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2659 ctx->s_qf_names[GRPQUOTA]); 2660 2661 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2662 test_opt(sb, USRQUOTA)); 2663 2664 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2665 test_opt(sb, GRPQUOTA)); 2666 2667 if (usr_qf_name) { 2668 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2669 usrquota = false; 2670 } 2671 if (grp_qf_name) { 2672 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2673 grpquota = false; 2674 } 2675 2676 if (usr_qf_name || grp_qf_name) { 2677 if (usrquota || grpquota) { 2678 ext4_msg(NULL, KERN_ERR, "old and new quota " 2679 "format mixing"); 2680 return -EINVAL; 2681 } 2682 2683 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2684 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2685 "not specified"); 2686 return -EINVAL; 2687 } 2688 } 2689 2690 return 0; 2691 2692 err_quota_change: 2693 ext4_msg(NULL, KERN_ERR, 2694 "Cannot change quota options when quota turned on"); 2695 return -EINVAL; 2696 err_jquota_change: 2697 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2698 "options when quota turned on"); 2699 return -EINVAL; 2700 err_jquota_specified: 2701 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2702 QTYPE2NAME(i)); 2703 return -EINVAL; 2704 #else 2705 return 0; 2706 #endif 2707 } 2708 2709 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2710 struct super_block *sb) 2711 { 2712 const struct ext4_fs_context *ctx = fc->fs_private; 2713 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2714 2715 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2716 return 0; 2717 2718 if (!ext4_has_feature_encrypt(sb)) { 2719 ext4_msg(NULL, KERN_WARNING, 2720 "test_dummy_encryption requires encrypt feature"); 2721 return -EINVAL; 2722 } 2723 /* 2724 * This mount option is just for testing, and it's not worthwhile to 2725 * implement the extra complexity (e.g. RCU protection) that would be 2726 * needed to allow it to be set or changed during remount. We do allow 2727 * it to be specified during remount, but only if there is no change. 2728 */ 2729 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2730 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2731 &ctx->dummy_enc_policy)) 2732 return 0; 2733 ext4_msg(NULL, KERN_WARNING, 2734 "Can't set or change test_dummy_encryption on remount"); 2735 return -EINVAL; 2736 } 2737 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2738 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2739 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2740 &ctx->dummy_enc_policy)) 2741 return 0; 2742 ext4_msg(NULL, KERN_WARNING, 2743 "Conflicting test_dummy_encryption options"); 2744 return -EINVAL; 2745 } 2746 return 0; 2747 } 2748 2749 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2750 struct super_block *sb) 2751 { 2752 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2753 /* if already set, it was already verified to be the same */ 2754 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2755 return; 2756 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2757 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2758 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2759 } 2760 2761 static int ext4_check_opt_consistency(struct fs_context *fc, 2762 struct super_block *sb) 2763 { 2764 struct ext4_fs_context *ctx = fc->fs_private; 2765 struct ext4_sb_info *sbi = fc->s_fs_info; 2766 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2767 int err; 2768 2769 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2770 ext4_msg(NULL, KERN_ERR, 2771 "Mount option(s) incompatible with ext2"); 2772 return -EINVAL; 2773 } 2774 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2775 ext4_msg(NULL, KERN_ERR, 2776 "Mount option(s) incompatible with ext3"); 2777 return -EINVAL; 2778 } 2779 2780 if (ctx->s_want_extra_isize > 2781 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2782 ext4_msg(NULL, KERN_ERR, 2783 "Invalid want_extra_isize %d", 2784 ctx->s_want_extra_isize); 2785 return -EINVAL; 2786 } 2787 2788 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) { 2789 int blocksize = 2790 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2791 if (blocksize < PAGE_SIZE) 2792 ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an " 2793 "experimental mount option 'dioread_nolock' " 2794 "for blocksize < PAGE_SIZE"); 2795 } 2796 2797 err = ext4_check_test_dummy_encryption(fc, sb); 2798 if (err) 2799 return err; 2800 2801 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2802 if (!sbi->s_journal) { 2803 ext4_msg(NULL, KERN_WARNING, 2804 "Remounting file system with no journal " 2805 "so ignoring journalled data option"); 2806 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2807 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2808 test_opt(sb, DATA_FLAGS)) { 2809 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2810 "on remount"); 2811 return -EINVAL; 2812 } 2813 } 2814 2815 if (is_remount) { 2816 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2817 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2818 ext4_msg(NULL, KERN_ERR, "can't mount with " 2819 "both data=journal and dax"); 2820 return -EINVAL; 2821 } 2822 2823 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2824 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2825 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2826 fail_dax_change_remount: 2827 ext4_msg(NULL, KERN_ERR, "can't change " 2828 "dax mount option while remounting"); 2829 return -EINVAL; 2830 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2831 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2832 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2833 goto fail_dax_change_remount; 2834 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2835 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2836 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2837 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2838 goto fail_dax_change_remount; 2839 } 2840 } 2841 2842 return ext4_check_quota_consistency(fc, sb); 2843 } 2844 2845 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2846 { 2847 struct ext4_fs_context *ctx = fc->fs_private; 2848 struct ext4_sb_info *sbi = fc->s_fs_info; 2849 2850 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2851 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2852 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2853 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2854 sb->s_flags &= ~ctx->mask_s_flags; 2855 sb->s_flags |= ctx->vals_s_flags; 2856 2857 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2858 APPLY(s_commit_interval); 2859 APPLY(s_stripe); 2860 APPLY(s_max_batch_time); 2861 APPLY(s_min_batch_time); 2862 APPLY(s_want_extra_isize); 2863 APPLY(s_inode_readahead_blks); 2864 APPLY(s_max_dir_size_kb); 2865 APPLY(s_li_wait_mult); 2866 APPLY(s_resgid); 2867 APPLY(s_resuid); 2868 2869 #ifdef CONFIG_EXT4_DEBUG 2870 APPLY(s_fc_debug_max_replay); 2871 #endif 2872 2873 ext4_apply_quota_options(fc, sb); 2874 ext4_apply_test_dummy_encryption(ctx, sb); 2875 } 2876 2877 2878 static int ext4_validate_options(struct fs_context *fc) 2879 { 2880 #ifdef CONFIG_QUOTA 2881 struct ext4_fs_context *ctx = fc->fs_private; 2882 char *usr_qf_name, *grp_qf_name; 2883 2884 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2885 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2886 2887 if (usr_qf_name || grp_qf_name) { 2888 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2889 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2890 2891 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2892 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2893 2894 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2895 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2896 ext4_msg(NULL, KERN_ERR, "old and new quota " 2897 "format mixing"); 2898 return -EINVAL; 2899 } 2900 } 2901 #endif 2902 return 1; 2903 } 2904 2905 static inline void ext4_show_quota_options(struct seq_file *seq, 2906 struct super_block *sb) 2907 { 2908 #if defined(CONFIG_QUOTA) 2909 struct ext4_sb_info *sbi = EXT4_SB(sb); 2910 char *usr_qf_name, *grp_qf_name; 2911 2912 if (sbi->s_jquota_fmt) { 2913 char *fmtname = ""; 2914 2915 switch (sbi->s_jquota_fmt) { 2916 case QFMT_VFS_OLD: 2917 fmtname = "vfsold"; 2918 break; 2919 case QFMT_VFS_V0: 2920 fmtname = "vfsv0"; 2921 break; 2922 case QFMT_VFS_V1: 2923 fmtname = "vfsv1"; 2924 break; 2925 } 2926 seq_printf(seq, ",jqfmt=%s", fmtname); 2927 } 2928 2929 rcu_read_lock(); 2930 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2931 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2932 if (usr_qf_name) 2933 seq_show_option(seq, "usrjquota", usr_qf_name); 2934 if (grp_qf_name) 2935 seq_show_option(seq, "grpjquota", grp_qf_name); 2936 rcu_read_unlock(); 2937 #endif 2938 } 2939 2940 static const char *token2str(int token) 2941 { 2942 const struct fs_parameter_spec *spec; 2943 2944 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2945 if (spec->opt == token && !spec->type) 2946 break; 2947 return spec->name; 2948 } 2949 2950 /* 2951 * Show an option if 2952 * - it's set to a non-default value OR 2953 * - if the per-sb default is different from the global default 2954 */ 2955 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2956 int nodefs) 2957 { 2958 struct ext4_sb_info *sbi = EXT4_SB(sb); 2959 struct ext4_super_block *es = sbi->s_es; 2960 int def_errors; 2961 const struct mount_opts *m; 2962 char sep = nodefs ? '\n' : ','; 2963 2964 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2965 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2966 2967 if (sbi->s_sb_block != 1) 2968 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2969 2970 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2971 int want_set = m->flags & MOPT_SET; 2972 int opt_2 = m->flags & MOPT_2; 2973 unsigned int mount_opt, def_mount_opt; 2974 2975 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2976 m->flags & MOPT_SKIP) 2977 continue; 2978 2979 if (opt_2) { 2980 mount_opt = sbi->s_mount_opt2; 2981 def_mount_opt = sbi->s_def_mount_opt2; 2982 } else { 2983 mount_opt = sbi->s_mount_opt; 2984 def_mount_opt = sbi->s_def_mount_opt; 2985 } 2986 /* skip if same as the default */ 2987 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt))) 2988 continue; 2989 /* select Opt_noFoo vs Opt_Foo */ 2990 if ((want_set && 2991 (mount_opt & m->mount_opt) != m->mount_opt) || 2992 (!want_set && (mount_opt & m->mount_opt))) 2993 continue; 2994 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2995 } 2996 2997 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2998 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2999 SEQ_OPTS_PRINT("resuid=%u", 3000 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 3001 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 3002 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 3003 SEQ_OPTS_PRINT("resgid=%u", 3004 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 3005 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 3006 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 3007 SEQ_OPTS_PUTS("errors=remount-ro"); 3008 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 3009 SEQ_OPTS_PUTS("errors=continue"); 3010 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 3011 SEQ_OPTS_PUTS("errors=panic"); 3012 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 3013 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 3014 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 3015 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 3016 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 3017 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 3018 if (nodefs || sbi->s_stripe) 3019 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 3020 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 3021 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 3022 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 3023 SEQ_OPTS_PUTS("data=journal"); 3024 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 3025 SEQ_OPTS_PUTS("data=ordered"); 3026 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 3027 SEQ_OPTS_PUTS("data=writeback"); 3028 } 3029 if (nodefs || 3030 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 3031 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 3032 sbi->s_inode_readahead_blks); 3033 3034 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 3035 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 3036 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 3037 if (nodefs || sbi->s_max_dir_size_kb) 3038 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 3039 if (test_opt(sb, DATA_ERR_ABORT)) 3040 SEQ_OPTS_PUTS("data_err=abort"); 3041 3042 fscrypt_show_test_dummy_encryption(seq, sep, sb); 3043 3044 if (sb->s_flags & SB_INLINECRYPT) 3045 SEQ_OPTS_PUTS("inlinecrypt"); 3046 3047 if (test_opt(sb, DAX_ALWAYS)) { 3048 if (IS_EXT2_SB(sb)) 3049 SEQ_OPTS_PUTS("dax"); 3050 else 3051 SEQ_OPTS_PUTS("dax=always"); 3052 } else if (test_opt2(sb, DAX_NEVER)) { 3053 SEQ_OPTS_PUTS("dax=never"); 3054 } else if (test_opt2(sb, DAX_INODE)) { 3055 SEQ_OPTS_PUTS("dax=inode"); 3056 } 3057 3058 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3059 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3060 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 3061 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3062 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3063 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 3064 } 3065 3066 ext4_show_quota_options(seq, sb); 3067 return 0; 3068 } 3069 3070 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 3071 { 3072 return _ext4_show_options(seq, root->d_sb, 0); 3073 } 3074 3075 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3076 { 3077 struct super_block *sb = seq->private; 3078 int rc; 3079 3080 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3081 rc = _ext4_show_options(seq, sb, 1); 3082 seq_puts(seq, "\n"); 3083 return rc; 3084 } 3085 3086 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3087 int read_only) 3088 { 3089 struct ext4_sb_info *sbi = EXT4_SB(sb); 3090 int err = 0; 3091 3092 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3093 ext4_msg(sb, KERN_ERR, "revision level too high, " 3094 "forcing read-only mode"); 3095 err = -EROFS; 3096 goto done; 3097 } 3098 if (read_only) 3099 goto done; 3100 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3101 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3102 "running e2fsck is recommended"); 3103 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3104 ext4_msg(sb, KERN_WARNING, 3105 "warning: mounting fs with errors, " 3106 "running e2fsck is recommended"); 3107 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3108 le16_to_cpu(es->s_mnt_count) >= 3109 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3110 ext4_msg(sb, KERN_WARNING, 3111 "warning: maximal mount count reached, " 3112 "running e2fsck is recommended"); 3113 else if (le32_to_cpu(es->s_checkinterval) && 3114 (ext4_get_tstamp(es, s_lastcheck) + 3115 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3116 ext4_msg(sb, KERN_WARNING, 3117 "warning: checktime reached, " 3118 "running e2fsck is recommended"); 3119 if (!sbi->s_journal) 3120 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3121 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3122 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3123 le16_add_cpu(&es->s_mnt_count, 1); 3124 ext4_update_tstamp(es, s_mtime); 3125 if (sbi->s_journal) { 3126 ext4_set_feature_journal_needs_recovery(sb); 3127 if (ext4_has_feature_orphan_file(sb)) 3128 ext4_set_feature_orphan_present(sb); 3129 } 3130 3131 err = ext4_commit_super(sb); 3132 done: 3133 if (test_opt(sb, DEBUG)) 3134 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3135 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3136 sb->s_blocksize, 3137 sbi->s_groups_count, 3138 EXT4_BLOCKS_PER_GROUP(sb), 3139 EXT4_INODES_PER_GROUP(sb), 3140 sbi->s_mount_opt, sbi->s_mount_opt2); 3141 return err; 3142 } 3143 3144 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3145 { 3146 struct ext4_sb_info *sbi = EXT4_SB(sb); 3147 struct flex_groups **old_groups, **new_groups; 3148 int size, i, j; 3149 3150 if (!sbi->s_log_groups_per_flex) 3151 return 0; 3152 3153 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3154 if (size <= sbi->s_flex_groups_allocated) 3155 return 0; 3156 3157 new_groups = kvzalloc(roundup_pow_of_two(size * 3158 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3159 if (!new_groups) { 3160 ext4_msg(sb, KERN_ERR, 3161 "not enough memory for %d flex group pointers", size); 3162 return -ENOMEM; 3163 } 3164 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3165 new_groups[i] = kvzalloc(roundup_pow_of_two( 3166 sizeof(struct flex_groups)), 3167 GFP_KERNEL); 3168 if (!new_groups[i]) { 3169 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3170 kvfree(new_groups[j]); 3171 kvfree(new_groups); 3172 ext4_msg(sb, KERN_ERR, 3173 "not enough memory for %d flex groups", size); 3174 return -ENOMEM; 3175 } 3176 } 3177 rcu_read_lock(); 3178 old_groups = rcu_dereference(sbi->s_flex_groups); 3179 if (old_groups) 3180 memcpy(new_groups, old_groups, 3181 (sbi->s_flex_groups_allocated * 3182 sizeof(struct flex_groups *))); 3183 rcu_read_unlock(); 3184 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3185 sbi->s_flex_groups_allocated = size; 3186 if (old_groups) 3187 ext4_kvfree_array_rcu(old_groups); 3188 return 0; 3189 } 3190 3191 static int ext4_fill_flex_info(struct super_block *sb) 3192 { 3193 struct ext4_sb_info *sbi = EXT4_SB(sb); 3194 struct ext4_group_desc *gdp = NULL; 3195 struct flex_groups *fg; 3196 ext4_group_t flex_group; 3197 int i, err; 3198 3199 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3200 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3201 sbi->s_log_groups_per_flex = 0; 3202 return 1; 3203 } 3204 3205 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3206 if (err) 3207 goto failed; 3208 3209 for (i = 0; i < sbi->s_groups_count; i++) { 3210 gdp = ext4_get_group_desc(sb, i, NULL); 3211 3212 flex_group = ext4_flex_group(sbi, i); 3213 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3214 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3215 atomic64_add(ext4_free_group_clusters(sb, gdp), 3216 &fg->free_clusters); 3217 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3218 } 3219 3220 return 1; 3221 failed: 3222 return 0; 3223 } 3224 3225 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3226 struct ext4_group_desc *gdp) 3227 { 3228 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3229 __u16 crc = 0; 3230 __le32 le_group = cpu_to_le32(block_group); 3231 struct ext4_sb_info *sbi = EXT4_SB(sb); 3232 3233 if (ext4_has_metadata_csum(sbi->s_sb)) { 3234 /* Use new metadata_csum algorithm */ 3235 __u32 csum32; 3236 __u16 dummy_csum = 0; 3237 3238 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 3239 sizeof(le_group)); 3240 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 3241 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 3242 sizeof(dummy_csum)); 3243 offset += sizeof(dummy_csum); 3244 if (offset < sbi->s_desc_size) 3245 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 3246 sbi->s_desc_size - offset); 3247 3248 crc = csum32 & 0xFFFF; 3249 goto out; 3250 } 3251 3252 /* old crc16 code */ 3253 if (!ext4_has_feature_gdt_csum(sb)) 3254 return 0; 3255 3256 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3257 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3258 crc = crc16(crc, (__u8 *)gdp, offset); 3259 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3260 /* for checksum of struct ext4_group_desc do the rest...*/ 3261 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size) 3262 crc = crc16(crc, (__u8 *)gdp + offset, 3263 sbi->s_desc_size - offset); 3264 3265 out: 3266 return cpu_to_le16(crc); 3267 } 3268 3269 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3270 struct ext4_group_desc *gdp) 3271 { 3272 if (ext4_has_group_desc_csum(sb) && 3273 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3274 return 0; 3275 3276 return 1; 3277 } 3278 3279 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3280 struct ext4_group_desc *gdp) 3281 { 3282 if (!ext4_has_group_desc_csum(sb)) 3283 return; 3284 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3285 } 3286 3287 /* Called at mount-time, super-block is locked */ 3288 static int ext4_check_descriptors(struct super_block *sb, 3289 ext4_fsblk_t sb_block, 3290 ext4_group_t *first_not_zeroed) 3291 { 3292 struct ext4_sb_info *sbi = EXT4_SB(sb); 3293 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3294 ext4_fsblk_t last_block; 3295 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3296 ext4_fsblk_t block_bitmap; 3297 ext4_fsblk_t inode_bitmap; 3298 ext4_fsblk_t inode_table; 3299 int flexbg_flag = 0; 3300 ext4_group_t i, grp = sbi->s_groups_count; 3301 3302 if (ext4_has_feature_flex_bg(sb)) 3303 flexbg_flag = 1; 3304 3305 ext4_debug("Checking group descriptors"); 3306 3307 for (i = 0; i < sbi->s_groups_count; i++) { 3308 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3309 3310 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3311 last_block = ext4_blocks_count(sbi->s_es) - 1; 3312 else 3313 last_block = first_block + 3314 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3315 3316 if ((grp == sbi->s_groups_count) && 3317 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3318 grp = i; 3319 3320 block_bitmap = ext4_block_bitmap(sb, gdp); 3321 if (block_bitmap == sb_block) { 3322 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3323 "Block bitmap for group %u overlaps " 3324 "superblock", i); 3325 if (!sb_rdonly(sb)) 3326 return 0; 3327 } 3328 if (block_bitmap >= sb_block + 1 && 3329 block_bitmap <= last_bg_block) { 3330 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3331 "Block bitmap for group %u overlaps " 3332 "block group descriptors", i); 3333 if (!sb_rdonly(sb)) 3334 return 0; 3335 } 3336 if (block_bitmap < first_block || block_bitmap > last_block) { 3337 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3338 "Block bitmap for group %u not in group " 3339 "(block %llu)!", i, block_bitmap); 3340 return 0; 3341 } 3342 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3343 if (inode_bitmap == sb_block) { 3344 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3345 "Inode bitmap for group %u overlaps " 3346 "superblock", i); 3347 if (!sb_rdonly(sb)) 3348 return 0; 3349 } 3350 if (inode_bitmap >= sb_block + 1 && 3351 inode_bitmap <= last_bg_block) { 3352 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3353 "Inode bitmap for group %u overlaps " 3354 "block group descriptors", i); 3355 if (!sb_rdonly(sb)) 3356 return 0; 3357 } 3358 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3359 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3360 "Inode bitmap for group %u not in group " 3361 "(block %llu)!", i, inode_bitmap); 3362 return 0; 3363 } 3364 inode_table = ext4_inode_table(sb, gdp); 3365 if (inode_table == sb_block) { 3366 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3367 "Inode table for group %u overlaps " 3368 "superblock", i); 3369 if (!sb_rdonly(sb)) 3370 return 0; 3371 } 3372 if (inode_table >= sb_block + 1 && 3373 inode_table <= last_bg_block) { 3374 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3375 "Inode table for group %u overlaps " 3376 "block group descriptors", i); 3377 if (!sb_rdonly(sb)) 3378 return 0; 3379 } 3380 if (inode_table < first_block || 3381 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3382 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3383 "Inode table for group %u not in group " 3384 "(block %llu)!", i, inode_table); 3385 return 0; 3386 } 3387 ext4_lock_group(sb, i); 3388 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3389 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3390 "Checksum for group %u failed (%u!=%u)", 3391 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3392 gdp)), le16_to_cpu(gdp->bg_checksum)); 3393 if (!sb_rdonly(sb)) { 3394 ext4_unlock_group(sb, i); 3395 return 0; 3396 } 3397 } 3398 ext4_unlock_group(sb, i); 3399 if (!flexbg_flag) 3400 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3401 } 3402 if (NULL != first_not_zeroed) 3403 *first_not_zeroed = grp; 3404 return 1; 3405 } 3406 3407 /* 3408 * Maximal extent format file size. 3409 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3410 * extent format containers, within a sector_t, and within i_blocks 3411 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3412 * so that won't be a limiting factor. 3413 * 3414 * However there is other limiting factor. We do store extents in the form 3415 * of starting block and length, hence the resulting length of the extent 3416 * covering maximum file size must fit into on-disk format containers as 3417 * well. Given that length is always by 1 unit bigger than max unit (because 3418 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3419 * 3420 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3421 */ 3422 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3423 { 3424 loff_t res; 3425 loff_t upper_limit = MAX_LFS_FILESIZE; 3426 3427 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3428 3429 if (!has_huge_files) { 3430 upper_limit = (1LL << 32) - 1; 3431 3432 /* total blocks in file system block size */ 3433 upper_limit >>= (blkbits - 9); 3434 upper_limit <<= blkbits; 3435 } 3436 3437 /* 3438 * 32-bit extent-start container, ee_block. We lower the maxbytes 3439 * by one fs block, so ee_len can cover the extent of maximum file 3440 * size 3441 */ 3442 res = (1LL << 32) - 1; 3443 res <<= blkbits; 3444 3445 /* Sanity check against vm- & vfs- imposed limits */ 3446 if (res > upper_limit) 3447 res = upper_limit; 3448 3449 return res; 3450 } 3451 3452 /* 3453 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3454 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3455 * We need to be 1 filesystem block less than the 2^48 sector limit. 3456 */ 3457 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3458 { 3459 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3460 int meta_blocks; 3461 unsigned int ppb = 1 << (bits - 2); 3462 3463 /* 3464 * This is calculated to be the largest file size for a dense, block 3465 * mapped file such that the file's total number of 512-byte sectors, 3466 * including data and all indirect blocks, does not exceed (2^48 - 1). 3467 * 3468 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3469 * number of 512-byte sectors of the file. 3470 */ 3471 if (!has_huge_files) { 3472 /* 3473 * !has_huge_files or implies that the inode i_block field 3474 * represents total file blocks in 2^32 512-byte sectors == 3475 * size of vfs inode i_blocks * 8 3476 */ 3477 upper_limit = (1LL << 32) - 1; 3478 3479 /* total blocks in file system block size */ 3480 upper_limit >>= (bits - 9); 3481 3482 } else { 3483 /* 3484 * We use 48 bit ext4_inode i_blocks 3485 * With EXT4_HUGE_FILE_FL set the i_blocks 3486 * represent total number of blocks in 3487 * file system block size 3488 */ 3489 upper_limit = (1LL << 48) - 1; 3490 3491 } 3492 3493 /* Compute how many blocks we can address by block tree */ 3494 res += ppb; 3495 res += ppb * ppb; 3496 res += ((loff_t)ppb) * ppb * ppb; 3497 /* Compute how many metadata blocks are needed */ 3498 meta_blocks = 1; 3499 meta_blocks += 1 + ppb; 3500 meta_blocks += 1 + ppb + ppb * ppb; 3501 /* Does block tree limit file size? */ 3502 if (res + meta_blocks <= upper_limit) 3503 goto check_lfs; 3504 3505 res = upper_limit; 3506 /* How many metadata blocks are needed for addressing upper_limit? */ 3507 upper_limit -= EXT4_NDIR_BLOCKS; 3508 /* indirect blocks */ 3509 meta_blocks = 1; 3510 upper_limit -= ppb; 3511 /* double indirect blocks */ 3512 if (upper_limit < ppb * ppb) { 3513 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3514 res -= meta_blocks; 3515 goto check_lfs; 3516 } 3517 meta_blocks += 1 + ppb; 3518 upper_limit -= ppb * ppb; 3519 /* tripple indirect blocks for the rest */ 3520 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3521 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3522 res -= meta_blocks; 3523 check_lfs: 3524 res <<= bits; 3525 if (res > MAX_LFS_FILESIZE) 3526 res = MAX_LFS_FILESIZE; 3527 3528 return res; 3529 } 3530 3531 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3532 ext4_fsblk_t logical_sb_block, int nr) 3533 { 3534 struct ext4_sb_info *sbi = EXT4_SB(sb); 3535 ext4_group_t bg, first_meta_bg; 3536 int has_super = 0; 3537 3538 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3539 3540 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3541 return logical_sb_block + nr + 1; 3542 bg = sbi->s_desc_per_block * nr; 3543 if (ext4_bg_has_super(sb, bg)) 3544 has_super = 1; 3545 3546 /* 3547 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3548 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3549 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3550 * compensate. 3551 */ 3552 if (sb->s_blocksize == 1024 && nr == 0 && 3553 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3554 has_super++; 3555 3556 return (has_super + ext4_group_first_block_no(sb, bg)); 3557 } 3558 3559 /** 3560 * ext4_get_stripe_size: Get the stripe size. 3561 * @sbi: In memory super block info 3562 * 3563 * If we have specified it via mount option, then 3564 * use the mount option value. If the value specified at mount time is 3565 * greater than the blocks per group use the super block value. 3566 * If the super block value is greater than blocks per group return 0. 3567 * Allocator needs it be less than blocks per group. 3568 * 3569 */ 3570 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3571 { 3572 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3573 unsigned long stripe_width = 3574 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3575 int ret; 3576 3577 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3578 ret = sbi->s_stripe; 3579 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3580 ret = stripe_width; 3581 else if (stride && stride <= sbi->s_blocks_per_group) 3582 ret = stride; 3583 else 3584 ret = 0; 3585 3586 /* 3587 * If the stripe width is 1, this makes no sense and 3588 * we set it to 0 to turn off stripe handling code. 3589 */ 3590 if (ret <= 1) 3591 ret = 0; 3592 3593 return ret; 3594 } 3595 3596 /* 3597 * Check whether this filesystem can be mounted based on 3598 * the features present and the RDONLY/RDWR mount requested. 3599 * Returns 1 if this filesystem can be mounted as requested, 3600 * 0 if it cannot be. 3601 */ 3602 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3603 { 3604 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3605 ext4_msg(sb, KERN_ERR, 3606 "Couldn't mount because of " 3607 "unsupported optional features (%x)", 3608 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3609 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3610 return 0; 3611 } 3612 3613 #if !IS_ENABLED(CONFIG_UNICODE) 3614 if (ext4_has_feature_casefold(sb)) { 3615 ext4_msg(sb, KERN_ERR, 3616 "Filesystem with casefold feature cannot be " 3617 "mounted without CONFIG_UNICODE"); 3618 return 0; 3619 } 3620 #endif 3621 3622 if (readonly) 3623 return 1; 3624 3625 if (ext4_has_feature_readonly(sb)) { 3626 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3627 sb->s_flags |= SB_RDONLY; 3628 return 1; 3629 } 3630 3631 /* Check that feature set is OK for a read-write mount */ 3632 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3633 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3634 "unsupported optional features (%x)", 3635 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3636 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3637 return 0; 3638 } 3639 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3640 ext4_msg(sb, KERN_ERR, 3641 "Can't support bigalloc feature without " 3642 "extents feature\n"); 3643 return 0; 3644 } 3645 3646 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3647 if (!readonly && (ext4_has_feature_quota(sb) || 3648 ext4_has_feature_project(sb))) { 3649 ext4_msg(sb, KERN_ERR, 3650 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3651 return 0; 3652 } 3653 #endif /* CONFIG_QUOTA */ 3654 return 1; 3655 } 3656 3657 /* 3658 * This function is called once a day if we have errors logged 3659 * on the file system 3660 */ 3661 static void print_daily_error_info(struct timer_list *t) 3662 { 3663 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3664 struct super_block *sb = sbi->s_sb; 3665 struct ext4_super_block *es = sbi->s_es; 3666 3667 if (es->s_error_count) 3668 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3669 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3670 le32_to_cpu(es->s_error_count)); 3671 if (es->s_first_error_time) { 3672 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3673 sb->s_id, 3674 ext4_get_tstamp(es, s_first_error_time), 3675 (int) sizeof(es->s_first_error_func), 3676 es->s_first_error_func, 3677 le32_to_cpu(es->s_first_error_line)); 3678 if (es->s_first_error_ino) 3679 printk(KERN_CONT ": inode %u", 3680 le32_to_cpu(es->s_first_error_ino)); 3681 if (es->s_first_error_block) 3682 printk(KERN_CONT ": block %llu", (unsigned long long) 3683 le64_to_cpu(es->s_first_error_block)); 3684 printk(KERN_CONT "\n"); 3685 } 3686 if (es->s_last_error_time) { 3687 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3688 sb->s_id, 3689 ext4_get_tstamp(es, s_last_error_time), 3690 (int) sizeof(es->s_last_error_func), 3691 es->s_last_error_func, 3692 le32_to_cpu(es->s_last_error_line)); 3693 if (es->s_last_error_ino) 3694 printk(KERN_CONT ": inode %u", 3695 le32_to_cpu(es->s_last_error_ino)); 3696 if (es->s_last_error_block) 3697 printk(KERN_CONT ": block %llu", (unsigned long long) 3698 le64_to_cpu(es->s_last_error_block)); 3699 printk(KERN_CONT "\n"); 3700 } 3701 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3702 } 3703 3704 /* Find next suitable group and run ext4_init_inode_table */ 3705 static int ext4_run_li_request(struct ext4_li_request *elr) 3706 { 3707 struct ext4_group_desc *gdp = NULL; 3708 struct super_block *sb = elr->lr_super; 3709 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3710 ext4_group_t group = elr->lr_next_group; 3711 unsigned int prefetch_ios = 0; 3712 int ret = 0; 3713 int nr = EXT4_SB(sb)->s_mb_prefetch; 3714 u64 start_time; 3715 3716 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3717 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios); 3718 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr); 3719 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr); 3720 if (group >= elr->lr_next_group) { 3721 ret = 1; 3722 if (elr->lr_first_not_zeroed != ngroups && 3723 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3724 elr->lr_next_group = elr->lr_first_not_zeroed; 3725 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3726 ret = 0; 3727 } 3728 } 3729 return ret; 3730 } 3731 3732 for (; group < ngroups; group++) { 3733 gdp = ext4_get_group_desc(sb, group, NULL); 3734 if (!gdp) { 3735 ret = 1; 3736 break; 3737 } 3738 3739 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3740 break; 3741 } 3742 3743 if (group >= ngroups) 3744 ret = 1; 3745 3746 if (!ret) { 3747 start_time = ktime_get_real_ns(); 3748 ret = ext4_init_inode_table(sb, group, 3749 elr->lr_timeout ? 0 : 1); 3750 trace_ext4_lazy_itable_init(sb, group); 3751 if (elr->lr_timeout == 0) { 3752 elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) * 3753 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3754 } 3755 elr->lr_next_sched = jiffies + elr->lr_timeout; 3756 elr->lr_next_group = group + 1; 3757 } 3758 return ret; 3759 } 3760 3761 /* 3762 * Remove lr_request from the list_request and free the 3763 * request structure. Should be called with li_list_mtx held 3764 */ 3765 static void ext4_remove_li_request(struct ext4_li_request *elr) 3766 { 3767 if (!elr) 3768 return; 3769 3770 list_del(&elr->lr_request); 3771 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3772 kfree(elr); 3773 } 3774 3775 static void ext4_unregister_li_request(struct super_block *sb) 3776 { 3777 mutex_lock(&ext4_li_mtx); 3778 if (!ext4_li_info) { 3779 mutex_unlock(&ext4_li_mtx); 3780 return; 3781 } 3782 3783 mutex_lock(&ext4_li_info->li_list_mtx); 3784 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3785 mutex_unlock(&ext4_li_info->li_list_mtx); 3786 mutex_unlock(&ext4_li_mtx); 3787 } 3788 3789 static struct task_struct *ext4_lazyinit_task; 3790 3791 /* 3792 * This is the function where ext4lazyinit thread lives. It walks 3793 * through the request list searching for next scheduled filesystem. 3794 * When such a fs is found, run the lazy initialization request 3795 * (ext4_rn_li_request) and keep track of the time spend in this 3796 * function. Based on that time we compute next schedule time of 3797 * the request. When walking through the list is complete, compute 3798 * next waking time and put itself into sleep. 3799 */ 3800 static int ext4_lazyinit_thread(void *arg) 3801 { 3802 struct ext4_lazy_init *eli = arg; 3803 struct list_head *pos, *n; 3804 struct ext4_li_request *elr; 3805 unsigned long next_wakeup, cur; 3806 3807 BUG_ON(NULL == eli); 3808 set_freezable(); 3809 3810 cont_thread: 3811 while (true) { 3812 next_wakeup = MAX_JIFFY_OFFSET; 3813 3814 mutex_lock(&eli->li_list_mtx); 3815 if (list_empty(&eli->li_request_list)) { 3816 mutex_unlock(&eli->li_list_mtx); 3817 goto exit_thread; 3818 } 3819 list_for_each_safe(pos, n, &eli->li_request_list) { 3820 int err = 0; 3821 int progress = 0; 3822 elr = list_entry(pos, struct ext4_li_request, 3823 lr_request); 3824 3825 if (time_before(jiffies, elr->lr_next_sched)) { 3826 if (time_before(elr->lr_next_sched, next_wakeup)) 3827 next_wakeup = elr->lr_next_sched; 3828 continue; 3829 } 3830 if (down_read_trylock(&elr->lr_super->s_umount)) { 3831 if (sb_start_write_trylock(elr->lr_super)) { 3832 progress = 1; 3833 /* 3834 * We hold sb->s_umount, sb can not 3835 * be removed from the list, it is 3836 * now safe to drop li_list_mtx 3837 */ 3838 mutex_unlock(&eli->li_list_mtx); 3839 err = ext4_run_li_request(elr); 3840 sb_end_write(elr->lr_super); 3841 mutex_lock(&eli->li_list_mtx); 3842 n = pos->next; 3843 } 3844 up_read((&elr->lr_super->s_umount)); 3845 } 3846 /* error, remove the lazy_init job */ 3847 if (err) { 3848 ext4_remove_li_request(elr); 3849 continue; 3850 } 3851 if (!progress) { 3852 elr->lr_next_sched = jiffies + 3853 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3854 } 3855 if (time_before(elr->lr_next_sched, next_wakeup)) 3856 next_wakeup = elr->lr_next_sched; 3857 } 3858 mutex_unlock(&eli->li_list_mtx); 3859 3860 try_to_freeze(); 3861 3862 cur = jiffies; 3863 if ((time_after_eq(cur, next_wakeup)) || 3864 (MAX_JIFFY_OFFSET == next_wakeup)) { 3865 cond_resched(); 3866 continue; 3867 } 3868 3869 schedule_timeout_interruptible(next_wakeup - cur); 3870 3871 if (kthread_should_stop()) { 3872 ext4_clear_request_list(); 3873 goto exit_thread; 3874 } 3875 } 3876 3877 exit_thread: 3878 /* 3879 * It looks like the request list is empty, but we need 3880 * to check it under the li_list_mtx lock, to prevent any 3881 * additions into it, and of course we should lock ext4_li_mtx 3882 * to atomically free the list and ext4_li_info, because at 3883 * this point another ext4 filesystem could be registering 3884 * new one. 3885 */ 3886 mutex_lock(&ext4_li_mtx); 3887 mutex_lock(&eli->li_list_mtx); 3888 if (!list_empty(&eli->li_request_list)) { 3889 mutex_unlock(&eli->li_list_mtx); 3890 mutex_unlock(&ext4_li_mtx); 3891 goto cont_thread; 3892 } 3893 mutex_unlock(&eli->li_list_mtx); 3894 kfree(ext4_li_info); 3895 ext4_li_info = NULL; 3896 mutex_unlock(&ext4_li_mtx); 3897 3898 return 0; 3899 } 3900 3901 static void ext4_clear_request_list(void) 3902 { 3903 struct list_head *pos, *n; 3904 struct ext4_li_request *elr; 3905 3906 mutex_lock(&ext4_li_info->li_list_mtx); 3907 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3908 elr = list_entry(pos, struct ext4_li_request, 3909 lr_request); 3910 ext4_remove_li_request(elr); 3911 } 3912 mutex_unlock(&ext4_li_info->li_list_mtx); 3913 } 3914 3915 static int ext4_run_lazyinit_thread(void) 3916 { 3917 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3918 ext4_li_info, "ext4lazyinit"); 3919 if (IS_ERR(ext4_lazyinit_task)) { 3920 int err = PTR_ERR(ext4_lazyinit_task); 3921 ext4_clear_request_list(); 3922 kfree(ext4_li_info); 3923 ext4_li_info = NULL; 3924 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3925 "initialization thread\n", 3926 err); 3927 return err; 3928 } 3929 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3930 return 0; 3931 } 3932 3933 /* 3934 * Check whether it make sense to run itable init. thread or not. 3935 * If there is at least one uninitialized inode table, return 3936 * corresponding group number, else the loop goes through all 3937 * groups and return total number of groups. 3938 */ 3939 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3940 { 3941 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3942 struct ext4_group_desc *gdp = NULL; 3943 3944 if (!ext4_has_group_desc_csum(sb)) 3945 return ngroups; 3946 3947 for (group = 0; group < ngroups; group++) { 3948 gdp = ext4_get_group_desc(sb, group, NULL); 3949 if (!gdp) 3950 continue; 3951 3952 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3953 break; 3954 } 3955 3956 return group; 3957 } 3958 3959 static int ext4_li_info_new(void) 3960 { 3961 struct ext4_lazy_init *eli = NULL; 3962 3963 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3964 if (!eli) 3965 return -ENOMEM; 3966 3967 INIT_LIST_HEAD(&eli->li_request_list); 3968 mutex_init(&eli->li_list_mtx); 3969 3970 eli->li_state |= EXT4_LAZYINIT_QUIT; 3971 3972 ext4_li_info = eli; 3973 3974 return 0; 3975 } 3976 3977 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3978 ext4_group_t start) 3979 { 3980 struct ext4_li_request *elr; 3981 3982 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3983 if (!elr) 3984 return NULL; 3985 3986 elr->lr_super = sb; 3987 elr->lr_first_not_zeroed = start; 3988 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3989 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3990 elr->lr_next_group = start; 3991 } else { 3992 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3993 } 3994 3995 /* 3996 * Randomize first schedule time of the request to 3997 * spread the inode table initialization requests 3998 * better. 3999 */ 4000 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 4001 return elr; 4002 } 4003 4004 int ext4_register_li_request(struct super_block *sb, 4005 ext4_group_t first_not_zeroed) 4006 { 4007 struct ext4_sb_info *sbi = EXT4_SB(sb); 4008 struct ext4_li_request *elr = NULL; 4009 ext4_group_t ngroups = sbi->s_groups_count; 4010 int ret = 0; 4011 4012 mutex_lock(&ext4_li_mtx); 4013 if (sbi->s_li_request != NULL) { 4014 /* 4015 * Reset timeout so it can be computed again, because 4016 * s_li_wait_mult might have changed. 4017 */ 4018 sbi->s_li_request->lr_timeout = 0; 4019 goto out; 4020 } 4021 4022 if (sb_rdonly(sb) || 4023 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 4024 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 4025 goto out; 4026 4027 elr = ext4_li_request_new(sb, first_not_zeroed); 4028 if (!elr) { 4029 ret = -ENOMEM; 4030 goto out; 4031 } 4032 4033 if (NULL == ext4_li_info) { 4034 ret = ext4_li_info_new(); 4035 if (ret) 4036 goto out; 4037 } 4038 4039 mutex_lock(&ext4_li_info->li_list_mtx); 4040 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 4041 mutex_unlock(&ext4_li_info->li_list_mtx); 4042 4043 sbi->s_li_request = elr; 4044 /* 4045 * set elr to NULL here since it has been inserted to 4046 * the request_list and the removal and free of it is 4047 * handled by ext4_clear_request_list from now on. 4048 */ 4049 elr = NULL; 4050 4051 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 4052 ret = ext4_run_lazyinit_thread(); 4053 if (ret) 4054 goto out; 4055 } 4056 out: 4057 mutex_unlock(&ext4_li_mtx); 4058 if (ret) 4059 kfree(elr); 4060 return ret; 4061 } 4062 4063 /* 4064 * We do not need to lock anything since this is called on 4065 * module unload. 4066 */ 4067 static void ext4_destroy_lazyinit_thread(void) 4068 { 4069 /* 4070 * If thread exited earlier 4071 * there's nothing to be done. 4072 */ 4073 if (!ext4_li_info || !ext4_lazyinit_task) 4074 return; 4075 4076 kthread_stop(ext4_lazyinit_task); 4077 } 4078 4079 static int set_journal_csum_feature_set(struct super_block *sb) 4080 { 4081 int ret = 1; 4082 int compat, incompat; 4083 struct ext4_sb_info *sbi = EXT4_SB(sb); 4084 4085 if (ext4_has_metadata_csum(sb)) { 4086 /* journal checksum v3 */ 4087 compat = 0; 4088 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4089 } else { 4090 /* journal checksum v1 */ 4091 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4092 incompat = 0; 4093 } 4094 4095 jbd2_journal_clear_features(sbi->s_journal, 4096 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4097 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4098 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4099 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4100 ret = jbd2_journal_set_features(sbi->s_journal, 4101 compat, 0, 4102 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4103 incompat); 4104 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4105 ret = jbd2_journal_set_features(sbi->s_journal, 4106 compat, 0, 4107 incompat); 4108 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4109 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4110 } else { 4111 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4112 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4113 } 4114 4115 return ret; 4116 } 4117 4118 /* 4119 * Note: calculating the overhead so we can be compatible with 4120 * historical BSD practice is quite difficult in the face of 4121 * clusters/bigalloc. This is because multiple metadata blocks from 4122 * different block group can end up in the same allocation cluster. 4123 * Calculating the exact overhead in the face of clustered allocation 4124 * requires either O(all block bitmaps) in memory or O(number of block 4125 * groups**2) in time. We will still calculate the superblock for 4126 * older file systems --- and if we come across with a bigalloc file 4127 * system with zero in s_overhead_clusters the estimate will be close to 4128 * correct especially for very large cluster sizes --- but for newer 4129 * file systems, it's better to calculate this figure once at mkfs 4130 * time, and store it in the superblock. If the superblock value is 4131 * present (even for non-bigalloc file systems), we will use it. 4132 */ 4133 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4134 char *buf) 4135 { 4136 struct ext4_sb_info *sbi = EXT4_SB(sb); 4137 struct ext4_group_desc *gdp; 4138 ext4_fsblk_t first_block, last_block, b; 4139 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4140 int s, j, count = 0; 4141 int has_super = ext4_bg_has_super(sb, grp); 4142 4143 if (!ext4_has_feature_bigalloc(sb)) 4144 return (has_super + ext4_bg_num_gdb(sb, grp) + 4145 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4146 sbi->s_itb_per_group + 2); 4147 4148 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4149 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4150 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4151 for (i = 0; i < ngroups; i++) { 4152 gdp = ext4_get_group_desc(sb, i, NULL); 4153 b = ext4_block_bitmap(sb, gdp); 4154 if (b >= first_block && b <= last_block) { 4155 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4156 count++; 4157 } 4158 b = ext4_inode_bitmap(sb, gdp); 4159 if (b >= first_block && b <= last_block) { 4160 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4161 count++; 4162 } 4163 b = ext4_inode_table(sb, gdp); 4164 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4165 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4166 int c = EXT4_B2C(sbi, b - first_block); 4167 ext4_set_bit(c, buf); 4168 count++; 4169 } 4170 if (i != grp) 4171 continue; 4172 s = 0; 4173 if (ext4_bg_has_super(sb, grp)) { 4174 ext4_set_bit(s++, buf); 4175 count++; 4176 } 4177 j = ext4_bg_num_gdb(sb, grp); 4178 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4179 ext4_error(sb, "Invalid number of block group " 4180 "descriptor blocks: %d", j); 4181 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4182 } 4183 count += j; 4184 for (; j > 0; j--) 4185 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4186 } 4187 if (!count) 4188 return 0; 4189 return EXT4_CLUSTERS_PER_GROUP(sb) - 4190 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4191 } 4192 4193 /* 4194 * Compute the overhead and stash it in sbi->s_overhead 4195 */ 4196 int ext4_calculate_overhead(struct super_block *sb) 4197 { 4198 struct ext4_sb_info *sbi = EXT4_SB(sb); 4199 struct ext4_super_block *es = sbi->s_es; 4200 struct inode *j_inode; 4201 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4202 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4203 ext4_fsblk_t overhead = 0; 4204 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4205 4206 if (!buf) 4207 return -ENOMEM; 4208 4209 /* 4210 * Compute the overhead (FS structures). This is constant 4211 * for a given filesystem unless the number of block groups 4212 * changes so we cache the previous value until it does. 4213 */ 4214 4215 /* 4216 * All of the blocks before first_data_block are overhead 4217 */ 4218 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4219 4220 /* 4221 * Add the overhead found in each block group 4222 */ 4223 for (i = 0; i < ngroups; i++) { 4224 int blks; 4225 4226 blks = count_overhead(sb, i, buf); 4227 overhead += blks; 4228 if (blks) 4229 memset(buf, 0, PAGE_SIZE); 4230 cond_resched(); 4231 } 4232 4233 /* 4234 * Add the internal journal blocks whether the journal has been 4235 * loaded or not 4236 */ 4237 if (sbi->s_journal && !sbi->s_journal_bdev) 4238 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4239 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4240 /* j_inum for internal journal is non-zero */ 4241 j_inode = ext4_get_journal_inode(sb, j_inum); 4242 if (!IS_ERR(j_inode)) { 4243 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4244 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4245 iput(j_inode); 4246 } else { 4247 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4248 } 4249 } 4250 sbi->s_overhead = overhead; 4251 smp_wmb(); 4252 free_page((unsigned long) buf); 4253 return 0; 4254 } 4255 4256 static void ext4_set_resv_clusters(struct super_block *sb) 4257 { 4258 ext4_fsblk_t resv_clusters; 4259 struct ext4_sb_info *sbi = EXT4_SB(sb); 4260 4261 /* 4262 * There's no need to reserve anything when we aren't using extents. 4263 * The space estimates are exact, there are no unwritten extents, 4264 * hole punching doesn't need new metadata... This is needed especially 4265 * to keep ext2/3 backward compatibility. 4266 */ 4267 if (!ext4_has_feature_extents(sb)) 4268 return; 4269 /* 4270 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4271 * This should cover the situations where we can not afford to run 4272 * out of space like for example punch hole, or converting 4273 * unwritten extents in delalloc path. In most cases such 4274 * allocation would require 1, or 2 blocks, higher numbers are 4275 * very rare. 4276 */ 4277 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4278 sbi->s_cluster_bits); 4279 4280 do_div(resv_clusters, 50); 4281 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4282 4283 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4284 } 4285 4286 static const char *ext4_quota_mode(struct super_block *sb) 4287 { 4288 #ifdef CONFIG_QUOTA 4289 if (!ext4_quota_capable(sb)) 4290 return "none"; 4291 4292 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4293 return "journalled"; 4294 else 4295 return "writeback"; 4296 #else 4297 return "disabled"; 4298 #endif 4299 } 4300 4301 static void ext4_setup_csum_trigger(struct super_block *sb, 4302 enum ext4_journal_trigger_type type, 4303 void (*trigger)( 4304 struct jbd2_buffer_trigger_type *type, 4305 struct buffer_head *bh, 4306 void *mapped_data, 4307 size_t size)) 4308 { 4309 struct ext4_sb_info *sbi = EXT4_SB(sb); 4310 4311 sbi->s_journal_triggers[type].sb = sb; 4312 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4313 } 4314 4315 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4316 { 4317 if (!sbi) 4318 return; 4319 4320 kfree(sbi->s_blockgroup_lock); 4321 fs_put_dax(sbi->s_daxdev, NULL); 4322 kfree(sbi); 4323 } 4324 4325 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4326 { 4327 struct ext4_sb_info *sbi; 4328 4329 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4330 if (!sbi) 4331 return NULL; 4332 4333 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4334 NULL, NULL); 4335 4336 sbi->s_blockgroup_lock = 4337 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4338 4339 if (!sbi->s_blockgroup_lock) 4340 goto err_out; 4341 4342 sb->s_fs_info = sbi; 4343 sbi->s_sb = sb; 4344 return sbi; 4345 err_out: 4346 fs_put_dax(sbi->s_daxdev, NULL); 4347 kfree(sbi); 4348 return NULL; 4349 } 4350 4351 static void ext4_set_def_opts(struct super_block *sb, 4352 struct ext4_super_block *es) 4353 { 4354 unsigned long def_mount_opts; 4355 4356 /* Set defaults before we parse the mount options */ 4357 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4358 set_opt(sb, INIT_INODE_TABLE); 4359 if (def_mount_opts & EXT4_DEFM_DEBUG) 4360 set_opt(sb, DEBUG); 4361 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4362 set_opt(sb, GRPID); 4363 if (def_mount_opts & EXT4_DEFM_UID16) 4364 set_opt(sb, NO_UID32); 4365 /* xattr user namespace & acls are now defaulted on */ 4366 set_opt(sb, XATTR_USER); 4367 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4368 set_opt(sb, POSIX_ACL); 4369 #endif 4370 if (ext4_has_feature_fast_commit(sb)) 4371 set_opt2(sb, JOURNAL_FAST_COMMIT); 4372 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4373 if (ext4_has_metadata_csum(sb)) 4374 set_opt(sb, JOURNAL_CHECKSUM); 4375 4376 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4377 set_opt(sb, JOURNAL_DATA); 4378 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4379 set_opt(sb, ORDERED_DATA); 4380 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4381 set_opt(sb, WRITEBACK_DATA); 4382 4383 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4384 set_opt(sb, ERRORS_PANIC); 4385 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4386 set_opt(sb, ERRORS_CONT); 4387 else 4388 set_opt(sb, ERRORS_RO); 4389 /* block_validity enabled by default; disable with noblock_validity */ 4390 set_opt(sb, BLOCK_VALIDITY); 4391 if (def_mount_opts & EXT4_DEFM_DISCARD) 4392 set_opt(sb, DISCARD); 4393 4394 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4395 set_opt(sb, BARRIER); 4396 4397 /* 4398 * enable delayed allocation by default 4399 * Use -o nodelalloc to turn it off 4400 */ 4401 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4402 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4403 set_opt(sb, DELALLOC); 4404 4405 if (sb->s_blocksize == PAGE_SIZE) 4406 set_opt(sb, DIOREAD_NOLOCK); 4407 } 4408 4409 static int ext4_handle_clustersize(struct super_block *sb) 4410 { 4411 struct ext4_sb_info *sbi = EXT4_SB(sb); 4412 struct ext4_super_block *es = sbi->s_es; 4413 int clustersize; 4414 4415 /* Handle clustersize */ 4416 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4417 if (ext4_has_feature_bigalloc(sb)) { 4418 if (clustersize < sb->s_blocksize) { 4419 ext4_msg(sb, KERN_ERR, 4420 "cluster size (%d) smaller than " 4421 "block size (%lu)", clustersize, sb->s_blocksize); 4422 return -EINVAL; 4423 } 4424 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4425 le32_to_cpu(es->s_log_block_size); 4426 sbi->s_clusters_per_group = 4427 le32_to_cpu(es->s_clusters_per_group); 4428 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4429 ext4_msg(sb, KERN_ERR, 4430 "#clusters per group too big: %lu", 4431 sbi->s_clusters_per_group); 4432 return -EINVAL; 4433 } 4434 if (sbi->s_blocks_per_group != 4435 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4436 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4437 "clusters per group (%lu) inconsistent", 4438 sbi->s_blocks_per_group, 4439 sbi->s_clusters_per_group); 4440 return -EINVAL; 4441 } 4442 } else { 4443 if (clustersize != sb->s_blocksize) { 4444 ext4_msg(sb, KERN_ERR, 4445 "fragment/cluster size (%d) != " 4446 "block size (%lu)", clustersize, sb->s_blocksize); 4447 return -EINVAL; 4448 } 4449 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4450 ext4_msg(sb, KERN_ERR, 4451 "#blocks per group too big: %lu", 4452 sbi->s_blocks_per_group); 4453 return -EINVAL; 4454 } 4455 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4456 sbi->s_cluster_bits = 0; 4457 } 4458 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4459 4460 /* Do we have standard group size of clustersize * 8 blocks ? */ 4461 if (sbi->s_blocks_per_group == clustersize << 3) 4462 set_opt2(sb, STD_GROUP_SIZE); 4463 4464 return 0; 4465 } 4466 4467 static void ext4_fast_commit_init(struct super_block *sb) 4468 { 4469 struct ext4_sb_info *sbi = EXT4_SB(sb); 4470 4471 /* Initialize fast commit stuff */ 4472 atomic_set(&sbi->s_fc_subtid, 0); 4473 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4474 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4475 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4476 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4477 sbi->s_fc_bytes = 0; 4478 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4479 sbi->s_fc_ineligible_tid = 0; 4480 spin_lock_init(&sbi->s_fc_lock); 4481 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4482 sbi->s_fc_replay_state.fc_regions = NULL; 4483 sbi->s_fc_replay_state.fc_regions_size = 0; 4484 sbi->s_fc_replay_state.fc_regions_used = 0; 4485 sbi->s_fc_replay_state.fc_regions_valid = 0; 4486 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4487 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4488 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4489 } 4490 4491 static int ext4_inode_info_init(struct super_block *sb, 4492 struct ext4_super_block *es) 4493 { 4494 struct ext4_sb_info *sbi = EXT4_SB(sb); 4495 4496 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4497 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4498 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4499 } else { 4500 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4501 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4502 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4503 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4504 sbi->s_first_ino); 4505 return -EINVAL; 4506 } 4507 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4508 (!is_power_of_2(sbi->s_inode_size)) || 4509 (sbi->s_inode_size > sb->s_blocksize)) { 4510 ext4_msg(sb, KERN_ERR, 4511 "unsupported inode size: %d", 4512 sbi->s_inode_size); 4513 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4514 return -EINVAL; 4515 } 4516 /* 4517 * i_atime_extra is the last extra field available for 4518 * [acm]times in struct ext4_inode. Checking for that 4519 * field should suffice to ensure we have extra space 4520 * for all three. 4521 */ 4522 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4523 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4524 sb->s_time_gran = 1; 4525 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4526 } else { 4527 sb->s_time_gran = NSEC_PER_SEC; 4528 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4529 } 4530 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4531 } 4532 4533 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4534 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4535 EXT4_GOOD_OLD_INODE_SIZE; 4536 if (ext4_has_feature_extra_isize(sb)) { 4537 unsigned v, max = (sbi->s_inode_size - 4538 EXT4_GOOD_OLD_INODE_SIZE); 4539 4540 v = le16_to_cpu(es->s_want_extra_isize); 4541 if (v > max) { 4542 ext4_msg(sb, KERN_ERR, 4543 "bad s_want_extra_isize: %d", v); 4544 return -EINVAL; 4545 } 4546 if (sbi->s_want_extra_isize < v) 4547 sbi->s_want_extra_isize = v; 4548 4549 v = le16_to_cpu(es->s_min_extra_isize); 4550 if (v > max) { 4551 ext4_msg(sb, KERN_ERR, 4552 "bad s_min_extra_isize: %d", v); 4553 return -EINVAL; 4554 } 4555 if (sbi->s_want_extra_isize < v) 4556 sbi->s_want_extra_isize = v; 4557 } 4558 } 4559 4560 return 0; 4561 } 4562 4563 #if IS_ENABLED(CONFIG_UNICODE) 4564 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4565 { 4566 const struct ext4_sb_encodings *encoding_info; 4567 struct unicode_map *encoding; 4568 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4569 4570 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4571 return 0; 4572 4573 encoding_info = ext4_sb_read_encoding(es); 4574 if (!encoding_info) { 4575 ext4_msg(sb, KERN_ERR, 4576 "Encoding requested by superblock is unknown"); 4577 return -EINVAL; 4578 } 4579 4580 encoding = utf8_load(encoding_info->version); 4581 if (IS_ERR(encoding)) { 4582 ext4_msg(sb, KERN_ERR, 4583 "can't mount with superblock charset: %s-%u.%u.%u " 4584 "not supported by the kernel. flags: 0x%x.", 4585 encoding_info->name, 4586 unicode_major(encoding_info->version), 4587 unicode_minor(encoding_info->version), 4588 unicode_rev(encoding_info->version), 4589 encoding_flags); 4590 return -EINVAL; 4591 } 4592 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4593 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4594 unicode_major(encoding_info->version), 4595 unicode_minor(encoding_info->version), 4596 unicode_rev(encoding_info->version), 4597 encoding_flags); 4598 4599 sb->s_encoding = encoding; 4600 sb->s_encoding_flags = encoding_flags; 4601 4602 return 0; 4603 } 4604 #else 4605 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4606 { 4607 return 0; 4608 } 4609 #endif 4610 4611 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4612 { 4613 struct ext4_sb_info *sbi = EXT4_SB(sb); 4614 4615 /* Warn if metadata_csum and gdt_csum are both set. */ 4616 if (ext4_has_feature_metadata_csum(sb) && 4617 ext4_has_feature_gdt_csum(sb)) 4618 ext4_warning(sb, "metadata_csum and uninit_bg are " 4619 "redundant flags; please run fsck."); 4620 4621 /* Check for a known checksum algorithm */ 4622 if (!ext4_verify_csum_type(sb, es)) { 4623 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4624 "unknown checksum algorithm."); 4625 return -EINVAL; 4626 } 4627 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4628 ext4_orphan_file_block_trigger); 4629 4630 /* Load the checksum driver */ 4631 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4632 if (IS_ERR(sbi->s_chksum_driver)) { 4633 int ret = PTR_ERR(sbi->s_chksum_driver); 4634 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4635 sbi->s_chksum_driver = NULL; 4636 return ret; 4637 } 4638 4639 /* Check superblock checksum */ 4640 if (!ext4_superblock_csum_verify(sb, es)) { 4641 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4642 "invalid superblock checksum. Run e2fsck?"); 4643 return -EFSBADCRC; 4644 } 4645 4646 /* Precompute checksum seed for all metadata */ 4647 if (ext4_has_feature_csum_seed(sb)) 4648 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4649 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4650 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4651 sizeof(es->s_uuid)); 4652 return 0; 4653 } 4654 4655 static int ext4_check_feature_compatibility(struct super_block *sb, 4656 struct ext4_super_block *es, 4657 int silent) 4658 { 4659 struct ext4_sb_info *sbi = EXT4_SB(sb); 4660 4661 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4662 (ext4_has_compat_features(sb) || 4663 ext4_has_ro_compat_features(sb) || 4664 ext4_has_incompat_features(sb))) 4665 ext4_msg(sb, KERN_WARNING, 4666 "feature flags set on rev 0 fs, " 4667 "running e2fsck is recommended"); 4668 4669 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4670 set_opt2(sb, HURD_COMPAT); 4671 if (ext4_has_feature_64bit(sb)) { 4672 ext4_msg(sb, KERN_ERR, 4673 "The Hurd can't support 64-bit file systems"); 4674 return -EINVAL; 4675 } 4676 4677 /* 4678 * ea_inode feature uses l_i_version field which is not 4679 * available in HURD_COMPAT mode. 4680 */ 4681 if (ext4_has_feature_ea_inode(sb)) { 4682 ext4_msg(sb, KERN_ERR, 4683 "ea_inode feature is not supported for Hurd"); 4684 return -EINVAL; 4685 } 4686 } 4687 4688 if (IS_EXT2_SB(sb)) { 4689 if (ext2_feature_set_ok(sb)) 4690 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4691 "using the ext4 subsystem"); 4692 else { 4693 /* 4694 * If we're probing be silent, if this looks like 4695 * it's actually an ext[34] filesystem. 4696 */ 4697 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4698 return -EINVAL; 4699 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4700 "to feature incompatibilities"); 4701 return -EINVAL; 4702 } 4703 } 4704 4705 if (IS_EXT3_SB(sb)) { 4706 if (ext3_feature_set_ok(sb)) 4707 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4708 "using the ext4 subsystem"); 4709 else { 4710 /* 4711 * If we're probing be silent, if this looks like 4712 * it's actually an ext4 filesystem. 4713 */ 4714 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4715 return -EINVAL; 4716 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4717 "to feature incompatibilities"); 4718 return -EINVAL; 4719 } 4720 } 4721 4722 /* 4723 * Check feature flags regardless of the revision level, since we 4724 * previously didn't change the revision level when setting the flags, 4725 * so there is a chance incompat flags are set on a rev 0 filesystem. 4726 */ 4727 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4728 return -EINVAL; 4729 4730 if (sbi->s_daxdev) { 4731 if (sb->s_blocksize == PAGE_SIZE) 4732 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4733 else 4734 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 4735 } 4736 4737 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4738 if (ext4_has_feature_inline_data(sb)) { 4739 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4740 " that may contain inline data"); 4741 return -EINVAL; 4742 } 4743 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4744 ext4_msg(sb, KERN_ERR, 4745 "DAX unsupported by block device."); 4746 return -EINVAL; 4747 } 4748 } 4749 4750 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4751 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4752 es->s_encryption_level); 4753 return -EINVAL; 4754 } 4755 4756 return 0; 4757 } 4758 4759 static int ext4_check_geometry(struct super_block *sb, 4760 struct ext4_super_block *es) 4761 { 4762 struct ext4_sb_info *sbi = EXT4_SB(sb); 4763 __u64 blocks_count; 4764 int err; 4765 4766 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 4767 ext4_msg(sb, KERN_ERR, 4768 "Number of reserved GDT blocks insanely large: %d", 4769 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4770 return -EINVAL; 4771 } 4772 /* 4773 * Test whether we have more sectors than will fit in sector_t, 4774 * and whether the max offset is addressable by the page cache. 4775 */ 4776 err = generic_check_addressable(sb->s_blocksize_bits, 4777 ext4_blocks_count(es)); 4778 if (err) { 4779 ext4_msg(sb, KERN_ERR, "filesystem" 4780 " too large to mount safely on this system"); 4781 return err; 4782 } 4783 4784 /* check blocks count against device size */ 4785 blocks_count = sb_bdev_nr_blocks(sb); 4786 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4787 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4788 "exceeds size of device (%llu blocks)", 4789 ext4_blocks_count(es), blocks_count); 4790 return -EINVAL; 4791 } 4792 4793 /* 4794 * It makes no sense for the first data block to be beyond the end 4795 * of the filesystem. 4796 */ 4797 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4798 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4799 "block %u is beyond end of filesystem (%llu)", 4800 le32_to_cpu(es->s_first_data_block), 4801 ext4_blocks_count(es)); 4802 return -EINVAL; 4803 } 4804 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4805 (sbi->s_cluster_ratio == 1)) { 4806 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4807 "block is 0 with a 1k block and cluster size"); 4808 return -EINVAL; 4809 } 4810 4811 blocks_count = (ext4_blocks_count(es) - 4812 le32_to_cpu(es->s_first_data_block) + 4813 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4814 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4815 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4816 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4817 "(block count %llu, first data block %u, " 4818 "blocks per group %lu)", blocks_count, 4819 ext4_blocks_count(es), 4820 le32_to_cpu(es->s_first_data_block), 4821 EXT4_BLOCKS_PER_GROUP(sb)); 4822 return -EINVAL; 4823 } 4824 sbi->s_groups_count = blocks_count; 4825 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4826 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4827 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4828 le32_to_cpu(es->s_inodes_count)) { 4829 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4830 le32_to_cpu(es->s_inodes_count), 4831 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4832 return -EINVAL; 4833 } 4834 4835 return 0; 4836 } 4837 4838 static int ext4_group_desc_init(struct super_block *sb, 4839 struct ext4_super_block *es, 4840 ext4_fsblk_t logical_sb_block, 4841 ext4_group_t *first_not_zeroed) 4842 { 4843 struct ext4_sb_info *sbi = EXT4_SB(sb); 4844 unsigned int db_count; 4845 ext4_fsblk_t block; 4846 int i; 4847 4848 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4849 EXT4_DESC_PER_BLOCK(sb); 4850 if (ext4_has_feature_meta_bg(sb)) { 4851 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4852 ext4_msg(sb, KERN_WARNING, 4853 "first meta block group too large: %u " 4854 "(group descriptor block count %u)", 4855 le32_to_cpu(es->s_first_meta_bg), db_count); 4856 return -EINVAL; 4857 } 4858 } 4859 rcu_assign_pointer(sbi->s_group_desc, 4860 kvmalloc_array(db_count, 4861 sizeof(struct buffer_head *), 4862 GFP_KERNEL)); 4863 if (sbi->s_group_desc == NULL) { 4864 ext4_msg(sb, KERN_ERR, "not enough memory"); 4865 return -ENOMEM; 4866 } 4867 4868 bgl_lock_init(sbi->s_blockgroup_lock); 4869 4870 /* Pre-read the descriptors into the buffer cache */ 4871 for (i = 0; i < db_count; i++) { 4872 block = descriptor_loc(sb, logical_sb_block, i); 4873 ext4_sb_breadahead_unmovable(sb, block); 4874 } 4875 4876 for (i = 0; i < db_count; i++) { 4877 struct buffer_head *bh; 4878 4879 block = descriptor_loc(sb, logical_sb_block, i); 4880 bh = ext4_sb_bread_unmovable(sb, block); 4881 if (IS_ERR(bh)) { 4882 ext4_msg(sb, KERN_ERR, 4883 "can't read group descriptor %d", i); 4884 sbi->s_gdb_count = i; 4885 return PTR_ERR(bh); 4886 } 4887 rcu_read_lock(); 4888 rcu_dereference(sbi->s_group_desc)[i] = bh; 4889 rcu_read_unlock(); 4890 } 4891 sbi->s_gdb_count = db_count; 4892 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4893 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4894 return -EFSCORRUPTED; 4895 } 4896 4897 return 0; 4898 } 4899 4900 static int ext4_load_and_init_journal(struct super_block *sb, 4901 struct ext4_super_block *es, 4902 struct ext4_fs_context *ctx) 4903 { 4904 struct ext4_sb_info *sbi = EXT4_SB(sb); 4905 int err; 4906 4907 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4908 if (err) 4909 return err; 4910 4911 if (ext4_has_feature_64bit(sb) && 4912 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4913 JBD2_FEATURE_INCOMPAT_64BIT)) { 4914 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4915 goto out; 4916 } 4917 4918 if (!set_journal_csum_feature_set(sb)) { 4919 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4920 "feature set"); 4921 goto out; 4922 } 4923 4924 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4925 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4926 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4927 ext4_msg(sb, KERN_ERR, 4928 "Failed to set fast commit journal feature"); 4929 goto out; 4930 } 4931 4932 /* We have now updated the journal if required, so we can 4933 * validate the data journaling mode. */ 4934 switch (test_opt(sb, DATA_FLAGS)) { 4935 case 0: 4936 /* No mode set, assume a default based on the journal 4937 * capabilities: ORDERED_DATA if the journal can 4938 * cope, else JOURNAL_DATA 4939 */ 4940 if (jbd2_journal_check_available_features 4941 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4942 set_opt(sb, ORDERED_DATA); 4943 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4944 } else { 4945 set_opt(sb, JOURNAL_DATA); 4946 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4947 } 4948 break; 4949 4950 case EXT4_MOUNT_ORDERED_DATA: 4951 case EXT4_MOUNT_WRITEBACK_DATA: 4952 if (!jbd2_journal_check_available_features 4953 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4954 ext4_msg(sb, KERN_ERR, "Journal does not support " 4955 "requested data journaling mode"); 4956 goto out; 4957 } 4958 break; 4959 default: 4960 break; 4961 } 4962 4963 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4964 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4965 ext4_msg(sb, KERN_ERR, "can't mount with " 4966 "journal_async_commit in data=ordered mode"); 4967 goto out; 4968 } 4969 4970 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4971 4972 sbi->s_journal->j_submit_inode_data_buffers = 4973 ext4_journal_submit_inode_data_buffers; 4974 sbi->s_journal->j_finish_inode_data_buffers = 4975 ext4_journal_finish_inode_data_buffers; 4976 4977 return 0; 4978 4979 out: 4980 /* flush s_sb_upd_work before destroying the journal. */ 4981 flush_work(&sbi->s_sb_upd_work); 4982 jbd2_journal_destroy(sbi->s_journal); 4983 sbi->s_journal = NULL; 4984 return -EINVAL; 4985 } 4986 4987 static int ext4_check_journal_data_mode(struct super_block *sb) 4988 { 4989 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4990 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4991 "data=journal disables delayed allocation, " 4992 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 4993 /* can't mount with both data=journal and dioread_nolock. */ 4994 clear_opt(sb, DIOREAD_NOLOCK); 4995 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4996 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4997 ext4_msg(sb, KERN_ERR, "can't mount with " 4998 "both data=journal and delalloc"); 4999 return -EINVAL; 5000 } 5001 if (test_opt(sb, DAX_ALWAYS)) { 5002 ext4_msg(sb, KERN_ERR, "can't mount with " 5003 "both data=journal and dax"); 5004 return -EINVAL; 5005 } 5006 if (ext4_has_feature_encrypt(sb)) { 5007 ext4_msg(sb, KERN_WARNING, 5008 "encrypted files will use data=ordered " 5009 "instead of data journaling mode"); 5010 } 5011 if (test_opt(sb, DELALLOC)) 5012 clear_opt(sb, DELALLOC); 5013 } else { 5014 sb->s_iflags |= SB_I_CGROUPWB; 5015 } 5016 5017 return 0; 5018 } 5019 5020 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 5021 int silent) 5022 { 5023 struct ext4_sb_info *sbi = EXT4_SB(sb); 5024 struct ext4_super_block *es; 5025 ext4_fsblk_t logical_sb_block; 5026 unsigned long offset = 0; 5027 struct buffer_head *bh; 5028 int ret = -EINVAL; 5029 int blocksize; 5030 5031 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 5032 if (!blocksize) { 5033 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 5034 return -EINVAL; 5035 } 5036 5037 /* 5038 * The ext4 superblock will not be buffer aligned for other than 1kB 5039 * block sizes. We need to calculate the offset from buffer start. 5040 */ 5041 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 5042 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5043 offset = do_div(logical_sb_block, blocksize); 5044 } else { 5045 logical_sb_block = sbi->s_sb_block; 5046 } 5047 5048 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5049 if (IS_ERR(bh)) { 5050 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 5051 return PTR_ERR(bh); 5052 } 5053 /* 5054 * Note: s_es must be initialized as soon as possible because 5055 * some ext4 macro-instructions depend on its value 5056 */ 5057 es = (struct ext4_super_block *) (bh->b_data + offset); 5058 sbi->s_es = es; 5059 sb->s_magic = le16_to_cpu(es->s_magic); 5060 if (sb->s_magic != EXT4_SUPER_MAGIC) { 5061 if (!silent) 5062 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5063 goto out; 5064 } 5065 5066 if (le32_to_cpu(es->s_log_block_size) > 5067 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5068 ext4_msg(sb, KERN_ERR, 5069 "Invalid log block size: %u", 5070 le32_to_cpu(es->s_log_block_size)); 5071 goto out; 5072 } 5073 if (le32_to_cpu(es->s_log_cluster_size) > 5074 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5075 ext4_msg(sb, KERN_ERR, 5076 "Invalid log cluster size: %u", 5077 le32_to_cpu(es->s_log_cluster_size)); 5078 goto out; 5079 } 5080 5081 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 5082 5083 /* 5084 * If the default block size is not the same as the real block size, 5085 * we need to reload it. 5086 */ 5087 if (sb->s_blocksize == blocksize) { 5088 *lsb = logical_sb_block; 5089 sbi->s_sbh = bh; 5090 return 0; 5091 } 5092 5093 /* 5094 * bh must be released before kill_bdev(), otherwise 5095 * it won't be freed and its page also. kill_bdev() 5096 * is called by sb_set_blocksize(). 5097 */ 5098 brelse(bh); 5099 /* Validate the filesystem blocksize */ 5100 if (!sb_set_blocksize(sb, blocksize)) { 5101 ext4_msg(sb, KERN_ERR, "bad block size %d", 5102 blocksize); 5103 bh = NULL; 5104 goto out; 5105 } 5106 5107 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5108 offset = do_div(logical_sb_block, blocksize); 5109 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5110 if (IS_ERR(bh)) { 5111 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5112 ret = PTR_ERR(bh); 5113 bh = NULL; 5114 goto out; 5115 } 5116 es = (struct ext4_super_block *)(bh->b_data + offset); 5117 sbi->s_es = es; 5118 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5119 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5120 goto out; 5121 } 5122 *lsb = logical_sb_block; 5123 sbi->s_sbh = bh; 5124 return 0; 5125 out: 5126 brelse(bh); 5127 return ret; 5128 } 5129 5130 static void ext4_hash_info_init(struct super_block *sb) 5131 { 5132 struct ext4_sb_info *sbi = EXT4_SB(sb); 5133 struct ext4_super_block *es = sbi->s_es; 5134 unsigned int i; 5135 5136 for (i = 0; i < 4; i++) 5137 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5138 5139 sbi->s_def_hash_version = es->s_def_hash_version; 5140 if (ext4_has_feature_dir_index(sb)) { 5141 i = le32_to_cpu(es->s_flags); 5142 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5143 sbi->s_hash_unsigned = 3; 5144 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5145 #ifdef __CHAR_UNSIGNED__ 5146 if (!sb_rdonly(sb)) 5147 es->s_flags |= 5148 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5149 sbi->s_hash_unsigned = 3; 5150 #else 5151 if (!sb_rdonly(sb)) 5152 es->s_flags |= 5153 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5154 #endif 5155 } 5156 } 5157 } 5158 5159 static int ext4_block_group_meta_init(struct super_block *sb, int silent) 5160 { 5161 struct ext4_sb_info *sbi = EXT4_SB(sb); 5162 struct ext4_super_block *es = sbi->s_es; 5163 int has_huge_files; 5164 5165 has_huge_files = ext4_has_feature_huge_file(sb); 5166 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5167 has_huge_files); 5168 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5169 5170 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5171 if (ext4_has_feature_64bit(sb)) { 5172 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5173 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5174 !is_power_of_2(sbi->s_desc_size)) { 5175 ext4_msg(sb, KERN_ERR, 5176 "unsupported descriptor size %lu", 5177 sbi->s_desc_size); 5178 return -EINVAL; 5179 } 5180 } else 5181 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5182 5183 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5184 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5185 5186 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5187 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5188 if (!silent) 5189 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5190 return -EINVAL; 5191 } 5192 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5193 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5194 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5195 sbi->s_inodes_per_group); 5196 return -EINVAL; 5197 } 5198 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5199 sbi->s_inodes_per_block; 5200 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5201 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5202 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5203 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5204 5205 return 0; 5206 } 5207 5208 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5209 { 5210 struct ext4_super_block *es = NULL; 5211 struct ext4_sb_info *sbi = EXT4_SB(sb); 5212 ext4_fsblk_t logical_sb_block; 5213 struct inode *root; 5214 int needs_recovery; 5215 int err; 5216 ext4_group_t first_not_zeroed; 5217 struct ext4_fs_context *ctx = fc->fs_private; 5218 int silent = fc->sb_flags & SB_SILENT; 5219 5220 /* Set defaults for the variables that will be set during parsing */ 5221 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5222 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5223 5224 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5225 sbi->s_sectors_written_start = 5226 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5227 5228 err = ext4_load_super(sb, &logical_sb_block, silent); 5229 if (err) 5230 goto out_fail; 5231 5232 es = sbi->s_es; 5233 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5234 5235 err = ext4_init_metadata_csum(sb, es); 5236 if (err) 5237 goto failed_mount; 5238 5239 ext4_set_def_opts(sb, es); 5240 5241 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5242 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5243 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5244 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5245 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5246 5247 /* 5248 * set default s_li_wait_mult for lazyinit, for the case there is 5249 * no mount option specified. 5250 */ 5251 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5252 5253 err = ext4_inode_info_init(sb, es); 5254 if (err) 5255 goto failed_mount; 5256 5257 err = parse_apply_sb_mount_options(sb, ctx); 5258 if (err < 0) 5259 goto failed_mount; 5260 5261 sbi->s_def_mount_opt = sbi->s_mount_opt; 5262 sbi->s_def_mount_opt2 = sbi->s_mount_opt2; 5263 5264 err = ext4_check_opt_consistency(fc, sb); 5265 if (err < 0) 5266 goto failed_mount; 5267 5268 ext4_apply_options(fc, sb); 5269 5270 err = ext4_encoding_init(sb, es); 5271 if (err) 5272 goto failed_mount; 5273 5274 err = ext4_check_journal_data_mode(sb); 5275 if (err) 5276 goto failed_mount; 5277 5278 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5279 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5280 5281 /* i_version is always enabled now */ 5282 sb->s_flags |= SB_I_VERSION; 5283 5284 err = ext4_check_feature_compatibility(sb, es, silent); 5285 if (err) 5286 goto failed_mount; 5287 5288 err = ext4_block_group_meta_init(sb, silent); 5289 if (err) 5290 goto failed_mount; 5291 5292 ext4_hash_info_init(sb); 5293 5294 err = ext4_handle_clustersize(sb); 5295 if (err) 5296 goto failed_mount; 5297 5298 err = ext4_check_geometry(sb, es); 5299 if (err) 5300 goto failed_mount; 5301 5302 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5303 spin_lock_init(&sbi->s_error_lock); 5304 INIT_WORK(&sbi->s_sb_upd_work, update_super_work); 5305 5306 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5307 if (err) 5308 goto failed_mount3; 5309 5310 err = ext4_es_register_shrinker(sbi); 5311 if (err) 5312 goto failed_mount3; 5313 5314 sbi->s_stripe = ext4_get_stripe_size(sbi); 5315 /* 5316 * It's hard to get stripe aligned blocks if stripe is not aligned with 5317 * cluster, just disable stripe and alert user to simpfy code and avoid 5318 * stripe aligned allocation which will rarely successes. 5319 */ 5320 if (sbi->s_stripe > 0 && sbi->s_cluster_ratio > 1 && 5321 sbi->s_stripe % sbi->s_cluster_ratio != 0) { 5322 ext4_msg(sb, KERN_WARNING, 5323 "stripe (%lu) is not aligned with cluster size (%u), " 5324 "stripe is disabled", 5325 sbi->s_stripe, sbi->s_cluster_ratio); 5326 sbi->s_stripe = 0; 5327 } 5328 sbi->s_extent_max_zeroout_kb = 32; 5329 5330 /* 5331 * set up enough so that it can read an inode 5332 */ 5333 sb->s_op = &ext4_sops; 5334 sb->s_export_op = &ext4_export_ops; 5335 sb->s_xattr = ext4_xattr_handlers; 5336 #ifdef CONFIG_FS_ENCRYPTION 5337 sb->s_cop = &ext4_cryptops; 5338 #endif 5339 #ifdef CONFIG_FS_VERITY 5340 sb->s_vop = &ext4_verityops; 5341 #endif 5342 #ifdef CONFIG_QUOTA 5343 sb->dq_op = &ext4_quota_operations; 5344 if (ext4_has_feature_quota(sb)) 5345 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5346 else 5347 sb->s_qcop = &ext4_qctl_operations; 5348 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5349 #endif 5350 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 5351 5352 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5353 mutex_init(&sbi->s_orphan_lock); 5354 5355 ext4_fast_commit_init(sb); 5356 5357 sb->s_root = NULL; 5358 5359 needs_recovery = (es->s_last_orphan != 0 || 5360 ext4_has_feature_orphan_present(sb) || 5361 ext4_has_feature_journal_needs_recovery(sb)); 5362 5363 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) { 5364 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)); 5365 if (err) 5366 goto failed_mount3a; 5367 } 5368 5369 err = -EINVAL; 5370 /* 5371 * The first inode we look at is the journal inode. Don't try 5372 * root first: it may be modified in the journal! 5373 */ 5374 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5375 err = ext4_load_and_init_journal(sb, es, ctx); 5376 if (err) 5377 goto failed_mount3a; 5378 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5379 ext4_has_feature_journal_needs_recovery(sb)) { 5380 ext4_msg(sb, KERN_ERR, "required journal recovery " 5381 "suppressed and not mounted read-only"); 5382 goto failed_mount3a; 5383 } else { 5384 /* Nojournal mode, all journal mount options are illegal */ 5385 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5386 ext4_msg(sb, KERN_ERR, "can't mount with " 5387 "journal_async_commit, fs mounted w/o journal"); 5388 goto failed_mount3a; 5389 } 5390 5391 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 5392 ext4_msg(sb, KERN_ERR, "can't mount with " 5393 "journal_checksum, fs mounted w/o journal"); 5394 goto failed_mount3a; 5395 } 5396 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 5397 ext4_msg(sb, KERN_ERR, "can't mount with " 5398 "commit=%lu, fs mounted w/o journal", 5399 sbi->s_commit_interval / HZ); 5400 goto failed_mount3a; 5401 } 5402 if (EXT4_MOUNT_DATA_FLAGS & 5403 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 5404 ext4_msg(sb, KERN_ERR, "can't mount with " 5405 "data=, fs mounted w/o journal"); 5406 goto failed_mount3a; 5407 } 5408 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5409 clear_opt(sb, JOURNAL_CHECKSUM); 5410 clear_opt(sb, DATA_FLAGS); 5411 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5412 sbi->s_journal = NULL; 5413 needs_recovery = 0; 5414 } 5415 5416 if (!test_opt(sb, NO_MBCACHE)) { 5417 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5418 if (!sbi->s_ea_block_cache) { 5419 ext4_msg(sb, KERN_ERR, 5420 "Failed to create ea_block_cache"); 5421 err = -EINVAL; 5422 goto failed_mount_wq; 5423 } 5424 5425 if (ext4_has_feature_ea_inode(sb)) { 5426 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5427 if (!sbi->s_ea_inode_cache) { 5428 ext4_msg(sb, KERN_ERR, 5429 "Failed to create ea_inode_cache"); 5430 err = -EINVAL; 5431 goto failed_mount_wq; 5432 } 5433 } 5434 } 5435 5436 /* 5437 * Get the # of file system overhead blocks from the 5438 * superblock if present. 5439 */ 5440 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5441 /* ignore the precalculated value if it is ridiculous */ 5442 if (sbi->s_overhead > ext4_blocks_count(es)) 5443 sbi->s_overhead = 0; 5444 /* 5445 * If the bigalloc feature is not enabled recalculating the 5446 * overhead doesn't take long, so we might as well just redo 5447 * it to make sure we are using the correct value. 5448 */ 5449 if (!ext4_has_feature_bigalloc(sb)) 5450 sbi->s_overhead = 0; 5451 if (sbi->s_overhead == 0) { 5452 err = ext4_calculate_overhead(sb); 5453 if (err) 5454 goto failed_mount_wq; 5455 } 5456 5457 /* 5458 * The maximum number of concurrent works can be high and 5459 * concurrency isn't really necessary. Limit it to 1. 5460 */ 5461 EXT4_SB(sb)->rsv_conversion_wq = 5462 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5463 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5464 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5465 err = -ENOMEM; 5466 goto failed_mount4; 5467 } 5468 5469 /* 5470 * The jbd2_journal_load will have done any necessary log recovery, 5471 * so we can safely mount the rest of the filesystem now. 5472 */ 5473 5474 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5475 if (IS_ERR(root)) { 5476 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5477 err = PTR_ERR(root); 5478 root = NULL; 5479 goto failed_mount4; 5480 } 5481 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5482 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5483 iput(root); 5484 err = -EFSCORRUPTED; 5485 goto failed_mount4; 5486 } 5487 5488 sb->s_root = d_make_root(root); 5489 if (!sb->s_root) { 5490 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5491 err = -ENOMEM; 5492 goto failed_mount4; 5493 } 5494 5495 err = ext4_setup_super(sb, es, sb_rdonly(sb)); 5496 if (err == -EROFS) { 5497 sb->s_flags |= SB_RDONLY; 5498 } else if (err) 5499 goto failed_mount4a; 5500 5501 ext4_set_resv_clusters(sb); 5502 5503 if (test_opt(sb, BLOCK_VALIDITY)) { 5504 err = ext4_setup_system_zone(sb); 5505 if (err) { 5506 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5507 "zone (%d)", err); 5508 goto failed_mount4a; 5509 } 5510 } 5511 ext4_fc_replay_cleanup(sb); 5512 5513 ext4_ext_init(sb); 5514 5515 /* 5516 * Enable optimize_scan if number of groups is > threshold. This can be 5517 * turned off by passing "mb_optimize_scan=0". This can also be 5518 * turned on forcefully by passing "mb_optimize_scan=1". 5519 */ 5520 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5521 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5522 set_opt2(sb, MB_OPTIMIZE_SCAN); 5523 else 5524 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5525 } 5526 5527 err = ext4_mb_init(sb); 5528 if (err) { 5529 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5530 err); 5531 goto failed_mount5; 5532 } 5533 5534 /* 5535 * We can only set up the journal commit callback once 5536 * mballoc is initialized 5537 */ 5538 if (sbi->s_journal) 5539 sbi->s_journal->j_commit_callback = 5540 ext4_journal_commit_callback; 5541 5542 err = ext4_percpu_param_init(sbi); 5543 if (err) 5544 goto failed_mount6; 5545 5546 if (ext4_has_feature_flex_bg(sb)) 5547 if (!ext4_fill_flex_info(sb)) { 5548 ext4_msg(sb, KERN_ERR, 5549 "unable to initialize " 5550 "flex_bg meta info!"); 5551 err = -ENOMEM; 5552 goto failed_mount6; 5553 } 5554 5555 err = ext4_register_li_request(sb, first_not_zeroed); 5556 if (err) 5557 goto failed_mount6; 5558 5559 err = ext4_register_sysfs(sb); 5560 if (err) 5561 goto failed_mount7; 5562 5563 err = ext4_init_orphan_info(sb); 5564 if (err) 5565 goto failed_mount8; 5566 #ifdef CONFIG_QUOTA 5567 /* Enable quota usage during mount. */ 5568 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5569 err = ext4_enable_quotas(sb); 5570 if (err) 5571 goto failed_mount9; 5572 } 5573 #endif /* CONFIG_QUOTA */ 5574 5575 /* 5576 * Save the original bdev mapping's wb_err value which could be 5577 * used to detect the metadata async write error. 5578 */ 5579 spin_lock_init(&sbi->s_bdev_wb_lock); 5580 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5581 &sbi->s_bdev_wb_err); 5582 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5583 ext4_orphan_cleanup(sb, es); 5584 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5585 /* 5586 * Update the checksum after updating free space/inode counters and 5587 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5588 * checksum in the buffer cache until it is written out and 5589 * e2fsprogs programs trying to open a file system immediately 5590 * after it is mounted can fail. 5591 */ 5592 ext4_superblock_csum_set(sb); 5593 if (needs_recovery) { 5594 ext4_msg(sb, KERN_INFO, "recovery complete"); 5595 err = ext4_mark_recovery_complete(sb, es); 5596 if (err) 5597 goto failed_mount10; 5598 } 5599 5600 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) 5601 ext4_msg(sb, KERN_WARNING, 5602 "mounting with \"discard\" option, but the device does not support discard"); 5603 5604 if (es->s_error_count) 5605 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5606 5607 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5608 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5609 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5610 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5611 atomic_set(&sbi->s_warning_count, 0); 5612 atomic_set(&sbi->s_msg_count, 0); 5613 5614 return 0; 5615 5616 failed_mount10: 5617 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 5618 failed_mount9: __maybe_unused 5619 ext4_release_orphan_info(sb); 5620 failed_mount8: 5621 ext4_unregister_sysfs(sb); 5622 kobject_put(&sbi->s_kobj); 5623 failed_mount7: 5624 ext4_unregister_li_request(sb); 5625 failed_mount6: 5626 ext4_mb_release(sb); 5627 ext4_flex_groups_free(sbi); 5628 ext4_percpu_param_destroy(sbi); 5629 failed_mount5: 5630 ext4_ext_release(sb); 5631 ext4_release_system_zone(sb); 5632 failed_mount4a: 5633 dput(sb->s_root); 5634 sb->s_root = NULL; 5635 failed_mount4: 5636 ext4_msg(sb, KERN_ERR, "mount failed"); 5637 if (EXT4_SB(sb)->rsv_conversion_wq) 5638 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5639 failed_mount_wq: 5640 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5641 sbi->s_ea_inode_cache = NULL; 5642 5643 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5644 sbi->s_ea_block_cache = NULL; 5645 5646 if (sbi->s_journal) { 5647 /* flush s_sb_upd_work before journal destroy. */ 5648 flush_work(&sbi->s_sb_upd_work); 5649 jbd2_journal_destroy(sbi->s_journal); 5650 sbi->s_journal = NULL; 5651 } 5652 failed_mount3a: 5653 ext4_es_unregister_shrinker(sbi); 5654 failed_mount3: 5655 /* flush s_sb_upd_work before sbi destroy */ 5656 flush_work(&sbi->s_sb_upd_work); 5657 del_timer_sync(&sbi->s_err_report); 5658 ext4_stop_mmpd(sbi); 5659 ext4_group_desc_free(sbi); 5660 failed_mount: 5661 if (sbi->s_chksum_driver) 5662 crypto_free_shash(sbi->s_chksum_driver); 5663 5664 #if IS_ENABLED(CONFIG_UNICODE) 5665 utf8_unload(sb->s_encoding); 5666 #endif 5667 5668 #ifdef CONFIG_QUOTA 5669 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++) 5670 kfree(get_qf_name(sb, sbi, i)); 5671 #endif 5672 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5673 brelse(sbi->s_sbh); 5674 if (sbi->s_journal_bdev) { 5675 invalidate_bdev(sbi->s_journal_bdev); 5676 blkdev_put(sbi->s_journal_bdev, sb); 5677 } 5678 out_fail: 5679 invalidate_bdev(sb->s_bdev); 5680 sb->s_fs_info = NULL; 5681 return err; 5682 } 5683 5684 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5685 { 5686 struct ext4_fs_context *ctx = fc->fs_private; 5687 struct ext4_sb_info *sbi; 5688 const char *descr; 5689 int ret; 5690 5691 sbi = ext4_alloc_sbi(sb); 5692 if (!sbi) 5693 return -ENOMEM; 5694 5695 fc->s_fs_info = sbi; 5696 5697 /* Cleanup superblock name */ 5698 strreplace(sb->s_id, '/', '!'); 5699 5700 sbi->s_sb_block = 1; /* Default super block location */ 5701 if (ctx->spec & EXT4_SPEC_s_sb_block) 5702 sbi->s_sb_block = ctx->s_sb_block; 5703 5704 ret = __ext4_fill_super(fc, sb); 5705 if (ret < 0) 5706 goto free_sbi; 5707 5708 if (sbi->s_journal) { 5709 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5710 descr = " journalled data mode"; 5711 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5712 descr = " ordered data mode"; 5713 else 5714 descr = " writeback data mode"; 5715 } else 5716 descr = "out journal"; 5717 5718 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5719 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. " 5720 "Quota mode: %s.", &sb->s_uuid, 5721 sb_rdonly(sb) ? "ro" : "r/w", descr, 5722 ext4_quota_mode(sb)); 5723 5724 /* Update the s_overhead_clusters if necessary */ 5725 ext4_update_overhead(sb, false); 5726 return 0; 5727 5728 free_sbi: 5729 ext4_free_sbi(sbi); 5730 fc->s_fs_info = NULL; 5731 return ret; 5732 } 5733 5734 static int ext4_get_tree(struct fs_context *fc) 5735 { 5736 return get_tree_bdev(fc, ext4_fill_super); 5737 } 5738 5739 /* 5740 * Setup any per-fs journal parameters now. We'll do this both on 5741 * initial mount, once the journal has been initialised but before we've 5742 * done any recovery; and again on any subsequent remount. 5743 */ 5744 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5745 { 5746 struct ext4_sb_info *sbi = EXT4_SB(sb); 5747 5748 journal->j_commit_interval = sbi->s_commit_interval; 5749 journal->j_min_batch_time = sbi->s_min_batch_time; 5750 journal->j_max_batch_time = sbi->s_max_batch_time; 5751 ext4_fc_init(sb, journal); 5752 5753 write_lock(&journal->j_state_lock); 5754 if (test_opt(sb, BARRIER)) 5755 journal->j_flags |= JBD2_BARRIER; 5756 else 5757 journal->j_flags &= ~JBD2_BARRIER; 5758 if (test_opt(sb, DATA_ERR_ABORT)) 5759 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5760 else 5761 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5762 /* 5763 * Always enable journal cycle record option, letting the journal 5764 * records log transactions continuously between each mount. 5765 */ 5766 journal->j_flags |= JBD2_CYCLE_RECORD; 5767 write_unlock(&journal->j_state_lock); 5768 } 5769 5770 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5771 unsigned int journal_inum) 5772 { 5773 struct inode *journal_inode; 5774 5775 /* 5776 * Test for the existence of a valid inode on disk. Bad things 5777 * happen if we iget() an unused inode, as the subsequent iput() 5778 * will try to delete it. 5779 */ 5780 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5781 if (IS_ERR(journal_inode)) { 5782 ext4_msg(sb, KERN_ERR, "no journal found"); 5783 return ERR_CAST(journal_inode); 5784 } 5785 if (!journal_inode->i_nlink) { 5786 make_bad_inode(journal_inode); 5787 iput(journal_inode); 5788 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5789 return ERR_PTR(-EFSCORRUPTED); 5790 } 5791 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) { 5792 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5793 iput(journal_inode); 5794 return ERR_PTR(-EFSCORRUPTED); 5795 } 5796 5797 ext4_debug("Journal inode found at %p: %lld bytes\n", 5798 journal_inode, journal_inode->i_size); 5799 return journal_inode; 5800 } 5801 5802 static int ext4_journal_bmap(journal_t *journal, sector_t *block) 5803 { 5804 struct ext4_map_blocks map; 5805 int ret; 5806 5807 if (journal->j_inode == NULL) 5808 return 0; 5809 5810 map.m_lblk = *block; 5811 map.m_len = 1; 5812 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0); 5813 if (ret <= 0) { 5814 ext4_msg(journal->j_inode->i_sb, KERN_CRIT, 5815 "journal bmap failed: block %llu ret %d\n", 5816 *block, ret); 5817 jbd2_journal_abort(journal, ret ? ret : -EIO); 5818 return ret; 5819 } 5820 *block = map.m_pblk; 5821 return 0; 5822 } 5823 5824 static journal_t *ext4_open_inode_journal(struct super_block *sb, 5825 unsigned int journal_inum) 5826 { 5827 struct inode *journal_inode; 5828 journal_t *journal; 5829 5830 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5831 if (IS_ERR(journal_inode)) 5832 return ERR_CAST(journal_inode); 5833 5834 journal = jbd2_journal_init_inode(journal_inode); 5835 if (IS_ERR(journal)) { 5836 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5837 iput(journal_inode); 5838 return ERR_CAST(journal); 5839 } 5840 journal->j_private = sb; 5841 journal->j_bmap = ext4_journal_bmap; 5842 ext4_init_journal_params(sb, journal); 5843 return journal; 5844 } 5845 5846 static struct block_device *ext4_get_journal_blkdev(struct super_block *sb, 5847 dev_t j_dev, ext4_fsblk_t *j_start, 5848 ext4_fsblk_t *j_len) 5849 { 5850 struct buffer_head *bh; 5851 struct block_device *bdev; 5852 int hblock, blocksize; 5853 ext4_fsblk_t sb_block; 5854 unsigned long offset; 5855 struct ext4_super_block *es; 5856 int errno; 5857 5858 /* see get_tree_bdev why this is needed and safe */ 5859 up_write(&sb->s_umount); 5860 bdev = blkdev_get_by_dev(j_dev, BLK_OPEN_READ | BLK_OPEN_WRITE, sb, 5861 &fs_holder_ops); 5862 down_write(&sb->s_umount); 5863 if (IS_ERR(bdev)) { 5864 ext4_msg(sb, KERN_ERR, 5865 "failed to open journal device unknown-block(%u,%u) %ld", 5866 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev)); 5867 return ERR_CAST(bdev); 5868 } 5869 5870 blocksize = sb->s_blocksize; 5871 hblock = bdev_logical_block_size(bdev); 5872 if (blocksize < hblock) { 5873 ext4_msg(sb, KERN_ERR, 5874 "blocksize too small for journal device"); 5875 errno = -EINVAL; 5876 goto out_bdev; 5877 } 5878 5879 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5880 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5881 set_blocksize(bdev, blocksize); 5882 bh = __bread(bdev, sb_block, blocksize); 5883 if (!bh) { 5884 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5885 "external journal"); 5886 errno = -EINVAL; 5887 goto out_bdev; 5888 } 5889 5890 es = (struct ext4_super_block *) (bh->b_data + offset); 5891 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5892 !(le32_to_cpu(es->s_feature_incompat) & 5893 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5894 ext4_msg(sb, KERN_ERR, "external journal has bad superblock"); 5895 errno = -EFSCORRUPTED; 5896 goto out_bh; 5897 } 5898 5899 if ((le32_to_cpu(es->s_feature_ro_compat) & 5900 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5901 es->s_checksum != ext4_superblock_csum(sb, es)) { 5902 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock"); 5903 errno = -EFSCORRUPTED; 5904 goto out_bh; 5905 } 5906 5907 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5908 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5909 errno = -EFSCORRUPTED; 5910 goto out_bh; 5911 } 5912 5913 *j_start = sb_block + 1; 5914 *j_len = ext4_blocks_count(es); 5915 brelse(bh); 5916 return bdev; 5917 5918 out_bh: 5919 brelse(bh); 5920 out_bdev: 5921 blkdev_put(bdev, sb); 5922 return ERR_PTR(errno); 5923 } 5924 5925 static journal_t *ext4_open_dev_journal(struct super_block *sb, 5926 dev_t j_dev) 5927 { 5928 journal_t *journal; 5929 ext4_fsblk_t j_start; 5930 ext4_fsblk_t j_len; 5931 struct block_device *journal_bdev; 5932 int errno = 0; 5933 5934 journal_bdev = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len); 5935 if (IS_ERR(journal_bdev)) 5936 return ERR_CAST(journal_bdev); 5937 5938 journal = jbd2_journal_init_dev(journal_bdev, sb->s_bdev, j_start, 5939 j_len, sb->s_blocksize); 5940 if (IS_ERR(journal)) { 5941 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5942 errno = PTR_ERR(journal); 5943 goto out_bdev; 5944 } 5945 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5946 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5947 "user (unsupported) - %d", 5948 be32_to_cpu(journal->j_superblock->s_nr_users)); 5949 errno = -EINVAL; 5950 goto out_journal; 5951 } 5952 journal->j_private = sb; 5953 EXT4_SB(sb)->s_journal_bdev = journal_bdev; 5954 ext4_init_journal_params(sb, journal); 5955 return journal; 5956 5957 out_journal: 5958 jbd2_journal_destroy(journal); 5959 out_bdev: 5960 blkdev_put(journal_bdev, sb); 5961 return ERR_PTR(errno); 5962 } 5963 5964 static int ext4_load_journal(struct super_block *sb, 5965 struct ext4_super_block *es, 5966 unsigned long journal_devnum) 5967 { 5968 journal_t *journal; 5969 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5970 dev_t journal_dev; 5971 int err = 0; 5972 int really_read_only; 5973 int journal_dev_ro; 5974 5975 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5976 return -EFSCORRUPTED; 5977 5978 if (journal_devnum && 5979 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5980 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5981 "numbers have changed"); 5982 journal_dev = new_decode_dev(journal_devnum); 5983 } else 5984 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5985 5986 if (journal_inum && journal_dev) { 5987 ext4_msg(sb, KERN_ERR, 5988 "filesystem has both journal inode and journal device!"); 5989 return -EINVAL; 5990 } 5991 5992 if (journal_inum) { 5993 journal = ext4_open_inode_journal(sb, journal_inum); 5994 if (IS_ERR(journal)) 5995 return PTR_ERR(journal); 5996 } else { 5997 journal = ext4_open_dev_journal(sb, journal_dev); 5998 if (IS_ERR(journal)) 5999 return PTR_ERR(journal); 6000 } 6001 6002 journal_dev_ro = bdev_read_only(journal->j_dev); 6003 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 6004 6005 if (journal_dev_ro && !sb_rdonly(sb)) { 6006 ext4_msg(sb, KERN_ERR, 6007 "journal device read-only, try mounting with '-o ro'"); 6008 err = -EROFS; 6009 goto err_out; 6010 } 6011 6012 /* 6013 * Are we loading a blank journal or performing recovery after a 6014 * crash? For recovery, we need to check in advance whether we 6015 * can get read-write access to the device. 6016 */ 6017 if (ext4_has_feature_journal_needs_recovery(sb)) { 6018 if (sb_rdonly(sb)) { 6019 ext4_msg(sb, KERN_INFO, "INFO: recovery " 6020 "required on readonly filesystem"); 6021 if (really_read_only) { 6022 ext4_msg(sb, KERN_ERR, "write access " 6023 "unavailable, cannot proceed " 6024 "(try mounting with noload)"); 6025 err = -EROFS; 6026 goto err_out; 6027 } 6028 ext4_msg(sb, KERN_INFO, "write access will " 6029 "be enabled during recovery"); 6030 } 6031 } 6032 6033 if (!(journal->j_flags & JBD2_BARRIER)) 6034 ext4_msg(sb, KERN_INFO, "barriers disabled"); 6035 6036 if (!ext4_has_feature_journal_needs_recovery(sb)) 6037 err = jbd2_journal_wipe(journal, !really_read_only); 6038 if (!err) { 6039 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 6040 __le16 orig_state; 6041 bool changed = false; 6042 6043 if (save) 6044 memcpy(save, ((char *) es) + 6045 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 6046 err = jbd2_journal_load(journal); 6047 if (save && memcmp(((char *) es) + EXT4_S_ERR_START, 6048 save, EXT4_S_ERR_LEN)) { 6049 memcpy(((char *) es) + EXT4_S_ERR_START, 6050 save, EXT4_S_ERR_LEN); 6051 changed = true; 6052 } 6053 kfree(save); 6054 orig_state = es->s_state; 6055 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state & 6056 EXT4_ERROR_FS); 6057 if (orig_state != es->s_state) 6058 changed = true; 6059 /* Write out restored error information to the superblock */ 6060 if (changed && !really_read_only) { 6061 int err2; 6062 err2 = ext4_commit_super(sb); 6063 err = err ? : err2; 6064 } 6065 } 6066 6067 if (err) { 6068 ext4_msg(sb, KERN_ERR, "error loading journal"); 6069 goto err_out; 6070 } 6071 6072 EXT4_SB(sb)->s_journal = journal; 6073 err = ext4_clear_journal_err(sb, es); 6074 if (err) { 6075 EXT4_SB(sb)->s_journal = NULL; 6076 jbd2_journal_destroy(journal); 6077 return err; 6078 } 6079 6080 if (!really_read_only && journal_devnum && 6081 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6082 es->s_journal_dev = cpu_to_le32(journal_devnum); 6083 ext4_commit_super(sb); 6084 } 6085 if (!really_read_only && journal_inum && 6086 journal_inum != le32_to_cpu(es->s_journal_inum)) { 6087 es->s_journal_inum = cpu_to_le32(journal_inum); 6088 ext4_commit_super(sb); 6089 } 6090 6091 return 0; 6092 6093 err_out: 6094 jbd2_journal_destroy(journal); 6095 return err; 6096 } 6097 6098 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 6099 static void ext4_update_super(struct super_block *sb) 6100 { 6101 struct ext4_sb_info *sbi = EXT4_SB(sb); 6102 struct ext4_super_block *es = sbi->s_es; 6103 struct buffer_head *sbh = sbi->s_sbh; 6104 6105 lock_buffer(sbh); 6106 /* 6107 * If the file system is mounted read-only, don't update the 6108 * superblock write time. This avoids updating the superblock 6109 * write time when we are mounting the root file system 6110 * read/only but we need to replay the journal; at that point, 6111 * for people who are east of GMT and who make their clock 6112 * tick in localtime for Windows bug-for-bug compatibility, 6113 * the clock is set in the future, and this will cause e2fsck 6114 * to complain and force a full file system check. 6115 */ 6116 if (!sb_rdonly(sb)) 6117 ext4_update_tstamp(es, s_wtime); 6118 es->s_kbytes_written = 6119 cpu_to_le64(sbi->s_kbytes_written + 6120 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 6121 sbi->s_sectors_written_start) >> 1)); 6122 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 6123 ext4_free_blocks_count_set(es, 6124 EXT4_C2B(sbi, percpu_counter_sum_positive( 6125 &sbi->s_freeclusters_counter))); 6126 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 6127 es->s_free_inodes_count = 6128 cpu_to_le32(percpu_counter_sum_positive( 6129 &sbi->s_freeinodes_counter)); 6130 /* Copy error information to the on-disk superblock */ 6131 spin_lock(&sbi->s_error_lock); 6132 if (sbi->s_add_error_count > 0) { 6133 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6134 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6135 __ext4_update_tstamp(&es->s_first_error_time, 6136 &es->s_first_error_time_hi, 6137 sbi->s_first_error_time); 6138 strncpy(es->s_first_error_func, sbi->s_first_error_func, 6139 sizeof(es->s_first_error_func)); 6140 es->s_first_error_line = 6141 cpu_to_le32(sbi->s_first_error_line); 6142 es->s_first_error_ino = 6143 cpu_to_le32(sbi->s_first_error_ino); 6144 es->s_first_error_block = 6145 cpu_to_le64(sbi->s_first_error_block); 6146 es->s_first_error_errcode = 6147 ext4_errno_to_code(sbi->s_first_error_code); 6148 } 6149 __ext4_update_tstamp(&es->s_last_error_time, 6150 &es->s_last_error_time_hi, 6151 sbi->s_last_error_time); 6152 strncpy(es->s_last_error_func, sbi->s_last_error_func, 6153 sizeof(es->s_last_error_func)); 6154 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6155 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6156 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6157 es->s_last_error_errcode = 6158 ext4_errno_to_code(sbi->s_last_error_code); 6159 /* 6160 * Start the daily error reporting function if it hasn't been 6161 * started already 6162 */ 6163 if (!es->s_error_count) 6164 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6165 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6166 sbi->s_add_error_count = 0; 6167 } 6168 spin_unlock(&sbi->s_error_lock); 6169 6170 ext4_superblock_csum_set(sb); 6171 unlock_buffer(sbh); 6172 } 6173 6174 static int ext4_commit_super(struct super_block *sb) 6175 { 6176 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6177 6178 if (!sbh) 6179 return -EINVAL; 6180 if (block_device_ejected(sb)) 6181 return -ENODEV; 6182 6183 ext4_update_super(sb); 6184 6185 lock_buffer(sbh); 6186 /* Buffer got discarded which means block device got invalidated */ 6187 if (!buffer_mapped(sbh)) { 6188 unlock_buffer(sbh); 6189 return -EIO; 6190 } 6191 6192 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6193 /* 6194 * Oh, dear. A previous attempt to write the 6195 * superblock failed. This could happen because the 6196 * USB device was yanked out. Or it could happen to 6197 * be a transient write error and maybe the block will 6198 * be remapped. Nothing we can do but to retry the 6199 * write and hope for the best. 6200 */ 6201 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6202 "superblock detected"); 6203 clear_buffer_write_io_error(sbh); 6204 set_buffer_uptodate(sbh); 6205 } 6206 get_bh(sbh); 6207 /* Clear potential dirty bit if it was journalled update */ 6208 clear_buffer_dirty(sbh); 6209 sbh->b_end_io = end_buffer_write_sync; 6210 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6211 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6212 wait_on_buffer(sbh); 6213 if (buffer_write_io_error(sbh)) { 6214 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6215 "superblock"); 6216 clear_buffer_write_io_error(sbh); 6217 set_buffer_uptodate(sbh); 6218 return -EIO; 6219 } 6220 return 0; 6221 } 6222 6223 /* 6224 * Have we just finished recovery? If so, and if we are mounting (or 6225 * remounting) the filesystem readonly, then we will end up with a 6226 * consistent fs on disk. Record that fact. 6227 */ 6228 static int ext4_mark_recovery_complete(struct super_block *sb, 6229 struct ext4_super_block *es) 6230 { 6231 int err; 6232 journal_t *journal = EXT4_SB(sb)->s_journal; 6233 6234 if (!ext4_has_feature_journal(sb)) { 6235 if (journal != NULL) { 6236 ext4_error(sb, "Journal got removed while the fs was " 6237 "mounted!"); 6238 return -EFSCORRUPTED; 6239 } 6240 return 0; 6241 } 6242 jbd2_journal_lock_updates(journal); 6243 err = jbd2_journal_flush(journal, 0); 6244 if (err < 0) 6245 goto out; 6246 6247 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6248 ext4_has_feature_orphan_present(sb))) { 6249 if (!ext4_orphan_file_empty(sb)) { 6250 ext4_error(sb, "Orphan file not empty on read-only fs."); 6251 err = -EFSCORRUPTED; 6252 goto out; 6253 } 6254 ext4_clear_feature_journal_needs_recovery(sb); 6255 ext4_clear_feature_orphan_present(sb); 6256 ext4_commit_super(sb); 6257 } 6258 out: 6259 jbd2_journal_unlock_updates(journal); 6260 return err; 6261 } 6262 6263 /* 6264 * If we are mounting (or read-write remounting) a filesystem whose journal 6265 * has recorded an error from a previous lifetime, move that error to the 6266 * main filesystem now. 6267 */ 6268 static int ext4_clear_journal_err(struct super_block *sb, 6269 struct ext4_super_block *es) 6270 { 6271 journal_t *journal; 6272 int j_errno; 6273 const char *errstr; 6274 6275 if (!ext4_has_feature_journal(sb)) { 6276 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6277 return -EFSCORRUPTED; 6278 } 6279 6280 journal = EXT4_SB(sb)->s_journal; 6281 6282 /* 6283 * Now check for any error status which may have been recorded in the 6284 * journal by a prior ext4_error() or ext4_abort() 6285 */ 6286 6287 j_errno = jbd2_journal_errno(journal); 6288 if (j_errno) { 6289 char nbuf[16]; 6290 6291 errstr = ext4_decode_error(sb, j_errno, nbuf); 6292 ext4_warning(sb, "Filesystem error recorded " 6293 "from previous mount: %s", errstr); 6294 6295 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6296 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6297 j_errno = ext4_commit_super(sb); 6298 if (j_errno) 6299 return j_errno; 6300 ext4_warning(sb, "Marked fs in need of filesystem check."); 6301 6302 jbd2_journal_clear_err(journal); 6303 jbd2_journal_update_sb_errno(journal); 6304 } 6305 return 0; 6306 } 6307 6308 /* 6309 * Force the running and committing transactions to commit, 6310 * and wait on the commit. 6311 */ 6312 int ext4_force_commit(struct super_block *sb) 6313 { 6314 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal); 6315 } 6316 6317 static int ext4_sync_fs(struct super_block *sb, int wait) 6318 { 6319 int ret = 0; 6320 tid_t target; 6321 bool needs_barrier = false; 6322 struct ext4_sb_info *sbi = EXT4_SB(sb); 6323 6324 if (unlikely(ext4_forced_shutdown(sb))) 6325 return 0; 6326 6327 trace_ext4_sync_fs(sb, wait); 6328 flush_workqueue(sbi->rsv_conversion_wq); 6329 /* 6330 * Writeback quota in non-journalled quota case - journalled quota has 6331 * no dirty dquots 6332 */ 6333 dquot_writeback_dquots(sb, -1); 6334 /* 6335 * Data writeback is possible w/o journal transaction, so barrier must 6336 * being sent at the end of the function. But we can skip it if 6337 * transaction_commit will do it for us. 6338 */ 6339 if (sbi->s_journal) { 6340 target = jbd2_get_latest_transaction(sbi->s_journal); 6341 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6342 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6343 needs_barrier = true; 6344 6345 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6346 if (wait) 6347 ret = jbd2_log_wait_commit(sbi->s_journal, 6348 target); 6349 } 6350 } else if (wait && test_opt(sb, BARRIER)) 6351 needs_barrier = true; 6352 if (needs_barrier) { 6353 int err; 6354 err = blkdev_issue_flush(sb->s_bdev); 6355 if (!ret) 6356 ret = err; 6357 } 6358 6359 return ret; 6360 } 6361 6362 /* 6363 * LVM calls this function before a (read-only) snapshot is created. This 6364 * gives us a chance to flush the journal completely and mark the fs clean. 6365 * 6366 * Note that only this function cannot bring a filesystem to be in a clean 6367 * state independently. It relies on upper layer to stop all data & metadata 6368 * modifications. 6369 */ 6370 static int ext4_freeze(struct super_block *sb) 6371 { 6372 int error = 0; 6373 journal_t *journal = EXT4_SB(sb)->s_journal; 6374 6375 if (journal) { 6376 /* Now we set up the journal barrier. */ 6377 jbd2_journal_lock_updates(journal); 6378 6379 /* 6380 * Don't clear the needs_recovery flag if we failed to 6381 * flush the journal. 6382 */ 6383 error = jbd2_journal_flush(journal, 0); 6384 if (error < 0) 6385 goto out; 6386 6387 /* Journal blocked and flushed, clear needs_recovery flag. */ 6388 ext4_clear_feature_journal_needs_recovery(sb); 6389 if (ext4_orphan_file_empty(sb)) 6390 ext4_clear_feature_orphan_present(sb); 6391 } 6392 6393 error = ext4_commit_super(sb); 6394 out: 6395 if (journal) 6396 /* we rely on upper layer to stop further updates */ 6397 jbd2_journal_unlock_updates(journal); 6398 return error; 6399 } 6400 6401 /* 6402 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6403 * flag here, even though the filesystem is not technically dirty yet. 6404 */ 6405 static int ext4_unfreeze(struct super_block *sb) 6406 { 6407 if (ext4_forced_shutdown(sb)) 6408 return 0; 6409 6410 if (EXT4_SB(sb)->s_journal) { 6411 /* Reset the needs_recovery flag before the fs is unlocked. */ 6412 ext4_set_feature_journal_needs_recovery(sb); 6413 if (ext4_has_feature_orphan_file(sb)) 6414 ext4_set_feature_orphan_present(sb); 6415 } 6416 6417 ext4_commit_super(sb); 6418 return 0; 6419 } 6420 6421 /* 6422 * Structure to save mount options for ext4_remount's benefit 6423 */ 6424 struct ext4_mount_options { 6425 unsigned long s_mount_opt; 6426 unsigned long s_mount_opt2; 6427 kuid_t s_resuid; 6428 kgid_t s_resgid; 6429 unsigned long s_commit_interval; 6430 u32 s_min_batch_time, s_max_batch_time; 6431 #ifdef CONFIG_QUOTA 6432 int s_jquota_fmt; 6433 char *s_qf_names[EXT4_MAXQUOTAS]; 6434 #endif 6435 }; 6436 6437 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6438 { 6439 struct ext4_fs_context *ctx = fc->fs_private; 6440 struct ext4_super_block *es; 6441 struct ext4_sb_info *sbi = EXT4_SB(sb); 6442 unsigned long old_sb_flags; 6443 struct ext4_mount_options old_opts; 6444 ext4_group_t g; 6445 int err = 0; 6446 int alloc_ctx; 6447 #ifdef CONFIG_QUOTA 6448 int enable_quota = 0; 6449 int i, j; 6450 char *to_free[EXT4_MAXQUOTAS]; 6451 #endif 6452 6453 6454 /* Store the original options */ 6455 old_sb_flags = sb->s_flags; 6456 old_opts.s_mount_opt = sbi->s_mount_opt; 6457 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6458 old_opts.s_resuid = sbi->s_resuid; 6459 old_opts.s_resgid = sbi->s_resgid; 6460 old_opts.s_commit_interval = sbi->s_commit_interval; 6461 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6462 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6463 #ifdef CONFIG_QUOTA 6464 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6465 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6466 if (sbi->s_qf_names[i]) { 6467 char *qf_name = get_qf_name(sb, sbi, i); 6468 6469 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6470 if (!old_opts.s_qf_names[i]) { 6471 for (j = 0; j < i; j++) 6472 kfree(old_opts.s_qf_names[j]); 6473 return -ENOMEM; 6474 } 6475 } else 6476 old_opts.s_qf_names[i] = NULL; 6477 #endif 6478 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6479 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6480 ctx->journal_ioprio = 6481 sbi->s_journal->j_task->io_context->ioprio; 6482 else 6483 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 6484 6485 } 6486 6487 /* 6488 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause 6489 * two calls to ext4_should_dioread_nolock() to return inconsistent 6490 * values, triggering WARN_ON in ext4_add_complete_io(). we grab 6491 * here s_writepages_rwsem to avoid race between writepages ops and 6492 * remount. 6493 */ 6494 alloc_ctx = ext4_writepages_down_write(sb); 6495 ext4_apply_options(fc, sb); 6496 ext4_writepages_up_write(sb, alloc_ctx); 6497 6498 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6499 test_opt(sb, JOURNAL_CHECKSUM)) { 6500 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6501 "during remount not supported; ignoring"); 6502 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6503 } 6504 6505 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6506 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6507 ext4_msg(sb, KERN_ERR, "can't mount with " 6508 "both data=journal and delalloc"); 6509 err = -EINVAL; 6510 goto restore_opts; 6511 } 6512 if (test_opt(sb, DIOREAD_NOLOCK)) { 6513 ext4_msg(sb, KERN_ERR, "can't mount with " 6514 "both data=journal and dioread_nolock"); 6515 err = -EINVAL; 6516 goto restore_opts; 6517 } 6518 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6519 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6520 ext4_msg(sb, KERN_ERR, "can't mount with " 6521 "journal_async_commit in data=ordered mode"); 6522 err = -EINVAL; 6523 goto restore_opts; 6524 } 6525 } 6526 6527 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6528 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6529 err = -EINVAL; 6530 goto restore_opts; 6531 } 6532 6533 if (test_opt2(sb, ABORT)) 6534 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6535 6536 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6537 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6538 6539 es = sbi->s_es; 6540 6541 if (sbi->s_journal) { 6542 ext4_init_journal_params(sb, sbi->s_journal); 6543 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6544 } 6545 6546 /* Flush outstanding errors before changing fs state */ 6547 flush_work(&sbi->s_sb_upd_work); 6548 6549 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6550 if (ext4_forced_shutdown(sb)) { 6551 err = -EROFS; 6552 goto restore_opts; 6553 } 6554 6555 if (fc->sb_flags & SB_RDONLY) { 6556 err = sync_filesystem(sb); 6557 if (err < 0) 6558 goto restore_opts; 6559 err = dquot_suspend(sb, -1); 6560 if (err < 0) 6561 goto restore_opts; 6562 6563 /* 6564 * First of all, the unconditional stuff we have to do 6565 * to disable replay of the journal when we next remount 6566 */ 6567 sb->s_flags |= SB_RDONLY; 6568 6569 /* 6570 * OK, test if we are remounting a valid rw partition 6571 * readonly, and if so set the rdonly flag and then 6572 * mark the partition as valid again. 6573 */ 6574 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6575 (sbi->s_mount_state & EXT4_VALID_FS)) 6576 es->s_state = cpu_to_le16(sbi->s_mount_state); 6577 6578 if (sbi->s_journal) { 6579 /* 6580 * We let remount-ro finish even if marking fs 6581 * as clean failed... 6582 */ 6583 ext4_mark_recovery_complete(sb, es); 6584 } 6585 } else { 6586 /* Make sure we can mount this feature set readwrite */ 6587 if (ext4_has_feature_readonly(sb) || 6588 !ext4_feature_set_ok(sb, 0)) { 6589 err = -EROFS; 6590 goto restore_opts; 6591 } 6592 /* 6593 * Make sure the group descriptor checksums 6594 * are sane. If they aren't, refuse to remount r/w. 6595 */ 6596 for (g = 0; g < sbi->s_groups_count; g++) { 6597 struct ext4_group_desc *gdp = 6598 ext4_get_group_desc(sb, g, NULL); 6599 6600 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6601 ext4_msg(sb, KERN_ERR, 6602 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6603 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6604 le16_to_cpu(gdp->bg_checksum)); 6605 err = -EFSBADCRC; 6606 goto restore_opts; 6607 } 6608 } 6609 6610 /* 6611 * If we have an unprocessed orphan list hanging 6612 * around from a previously readonly bdev mount, 6613 * require a full umount/remount for now. 6614 */ 6615 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6616 ext4_msg(sb, KERN_WARNING, "Couldn't " 6617 "remount RDWR because of unprocessed " 6618 "orphan inode list. Please " 6619 "umount/remount instead"); 6620 err = -EINVAL; 6621 goto restore_opts; 6622 } 6623 6624 /* 6625 * Mounting a RDONLY partition read-write, so reread 6626 * and store the current valid flag. (It may have 6627 * been changed by e2fsck since we originally mounted 6628 * the partition.) 6629 */ 6630 if (sbi->s_journal) { 6631 err = ext4_clear_journal_err(sb, es); 6632 if (err) 6633 goto restore_opts; 6634 } 6635 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6636 ~EXT4_FC_REPLAY); 6637 6638 err = ext4_setup_super(sb, es, 0); 6639 if (err) 6640 goto restore_opts; 6641 6642 sb->s_flags &= ~SB_RDONLY; 6643 if (ext4_has_feature_mmp(sb)) { 6644 err = ext4_multi_mount_protect(sb, 6645 le64_to_cpu(es->s_mmp_block)); 6646 if (err) 6647 goto restore_opts; 6648 } 6649 #ifdef CONFIG_QUOTA 6650 enable_quota = 1; 6651 #endif 6652 } 6653 } 6654 6655 /* 6656 * Handle creation of system zone data early because it can fail. 6657 * Releasing of existing data is done when we are sure remount will 6658 * succeed. 6659 */ 6660 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6661 err = ext4_setup_system_zone(sb); 6662 if (err) 6663 goto restore_opts; 6664 } 6665 6666 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6667 err = ext4_commit_super(sb); 6668 if (err) 6669 goto restore_opts; 6670 } 6671 6672 #ifdef CONFIG_QUOTA 6673 if (enable_quota) { 6674 if (sb_any_quota_suspended(sb)) 6675 dquot_resume(sb, -1); 6676 else if (ext4_has_feature_quota(sb)) { 6677 err = ext4_enable_quotas(sb); 6678 if (err) 6679 goto restore_opts; 6680 } 6681 } 6682 /* Release old quota file names */ 6683 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6684 kfree(old_opts.s_qf_names[i]); 6685 #endif 6686 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6687 ext4_release_system_zone(sb); 6688 6689 /* 6690 * Reinitialize lazy itable initialization thread based on 6691 * current settings 6692 */ 6693 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6694 ext4_unregister_li_request(sb); 6695 else { 6696 ext4_group_t first_not_zeroed; 6697 first_not_zeroed = ext4_has_uninit_itable(sb); 6698 ext4_register_li_request(sb, first_not_zeroed); 6699 } 6700 6701 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6702 ext4_stop_mmpd(sbi); 6703 6704 return 0; 6705 6706 restore_opts: 6707 /* 6708 * If there was a failing r/w to ro transition, we may need to 6709 * re-enable quota 6710 */ 6711 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) && 6712 sb_any_quota_suspended(sb)) 6713 dquot_resume(sb, -1); 6714 6715 alloc_ctx = ext4_writepages_down_write(sb); 6716 sb->s_flags = old_sb_flags; 6717 sbi->s_mount_opt = old_opts.s_mount_opt; 6718 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6719 sbi->s_resuid = old_opts.s_resuid; 6720 sbi->s_resgid = old_opts.s_resgid; 6721 sbi->s_commit_interval = old_opts.s_commit_interval; 6722 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6723 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6724 ext4_writepages_up_write(sb, alloc_ctx); 6725 6726 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6727 ext4_release_system_zone(sb); 6728 #ifdef CONFIG_QUOTA 6729 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6730 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6731 to_free[i] = get_qf_name(sb, sbi, i); 6732 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6733 } 6734 synchronize_rcu(); 6735 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6736 kfree(to_free[i]); 6737 #endif 6738 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6739 ext4_stop_mmpd(sbi); 6740 return err; 6741 } 6742 6743 static int ext4_reconfigure(struct fs_context *fc) 6744 { 6745 struct super_block *sb = fc->root->d_sb; 6746 int ret; 6747 6748 fc->s_fs_info = EXT4_SB(sb); 6749 6750 ret = ext4_check_opt_consistency(fc, sb); 6751 if (ret < 0) 6752 return ret; 6753 6754 ret = __ext4_remount(fc, sb); 6755 if (ret < 0) 6756 return ret; 6757 6758 ext4_msg(sb, KERN_INFO, "re-mounted %pU %s. Quota mode: %s.", 6759 &sb->s_uuid, sb_rdonly(sb) ? "ro" : "r/w", 6760 ext4_quota_mode(sb)); 6761 6762 return 0; 6763 } 6764 6765 #ifdef CONFIG_QUOTA 6766 static int ext4_statfs_project(struct super_block *sb, 6767 kprojid_t projid, struct kstatfs *buf) 6768 { 6769 struct kqid qid; 6770 struct dquot *dquot; 6771 u64 limit; 6772 u64 curblock; 6773 6774 qid = make_kqid_projid(projid); 6775 dquot = dqget(sb, qid); 6776 if (IS_ERR(dquot)) 6777 return PTR_ERR(dquot); 6778 spin_lock(&dquot->dq_dqb_lock); 6779 6780 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6781 dquot->dq_dqb.dqb_bhardlimit); 6782 limit >>= sb->s_blocksize_bits; 6783 6784 if (limit && buf->f_blocks > limit) { 6785 curblock = (dquot->dq_dqb.dqb_curspace + 6786 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6787 buf->f_blocks = limit; 6788 buf->f_bfree = buf->f_bavail = 6789 (buf->f_blocks > curblock) ? 6790 (buf->f_blocks - curblock) : 0; 6791 } 6792 6793 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6794 dquot->dq_dqb.dqb_ihardlimit); 6795 if (limit && buf->f_files > limit) { 6796 buf->f_files = limit; 6797 buf->f_ffree = 6798 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6799 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6800 } 6801 6802 spin_unlock(&dquot->dq_dqb_lock); 6803 dqput(dquot); 6804 return 0; 6805 } 6806 #endif 6807 6808 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6809 { 6810 struct super_block *sb = dentry->d_sb; 6811 struct ext4_sb_info *sbi = EXT4_SB(sb); 6812 struct ext4_super_block *es = sbi->s_es; 6813 ext4_fsblk_t overhead = 0, resv_blocks; 6814 s64 bfree; 6815 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6816 6817 if (!test_opt(sb, MINIX_DF)) 6818 overhead = sbi->s_overhead; 6819 6820 buf->f_type = EXT4_SUPER_MAGIC; 6821 buf->f_bsize = sb->s_blocksize; 6822 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6823 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6824 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6825 /* prevent underflow in case that few free space is available */ 6826 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6827 buf->f_bavail = buf->f_bfree - 6828 (ext4_r_blocks_count(es) + resv_blocks); 6829 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6830 buf->f_bavail = 0; 6831 buf->f_files = le32_to_cpu(es->s_inodes_count); 6832 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6833 buf->f_namelen = EXT4_NAME_LEN; 6834 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6835 6836 #ifdef CONFIG_QUOTA 6837 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6838 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6839 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6840 #endif 6841 return 0; 6842 } 6843 6844 6845 #ifdef CONFIG_QUOTA 6846 6847 /* 6848 * Helper functions so that transaction is started before we acquire dqio_sem 6849 * to keep correct lock ordering of transaction > dqio_sem 6850 */ 6851 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6852 { 6853 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6854 } 6855 6856 static int ext4_write_dquot(struct dquot *dquot) 6857 { 6858 int ret, err; 6859 handle_t *handle; 6860 struct inode *inode; 6861 6862 inode = dquot_to_inode(dquot); 6863 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6864 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6865 if (IS_ERR(handle)) 6866 return PTR_ERR(handle); 6867 ret = dquot_commit(dquot); 6868 if (ret < 0) 6869 ext4_error_err(dquot->dq_sb, -ret, 6870 "Failed to commit dquot type %d", 6871 dquot->dq_id.type); 6872 err = ext4_journal_stop(handle); 6873 if (!ret) 6874 ret = err; 6875 return ret; 6876 } 6877 6878 static int ext4_acquire_dquot(struct dquot *dquot) 6879 { 6880 int ret, err; 6881 handle_t *handle; 6882 6883 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6884 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6885 if (IS_ERR(handle)) 6886 return PTR_ERR(handle); 6887 ret = dquot_acquire(dquot); 6888 if (ret < 0) 6889 ext4_error_err(dquot->dq_sb, -ret, 6890 "Failed to acquire dquot type %d", 6891 dquot->dq_id.type); 6892 err = ext4_journal_stop(handle); 6893 if (!ret) 6894 ret = err; 6895 return ret; 6896 } 6897 6898 static int ext4_release_dquot(struct dquot *dquot) 6899 { 6900 int ret, err; 6901 handle_t *handle; 6902 6903 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6904 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6905 if (IS_ERR(handle)) { 6906 /* Release dquot anyway to avoid endless cycle in dqput() */ 6907 dquot_release(dquot); 6908 return PTR_ERR(handle); 6909 } 6910 ret = dquot_release(dquot); 6911 if (ret < 0) 6912 ext4_error_err(dquot->dq_sb, -ret, 6913 "Failed to release dquot type %d", 6914 dquot->dq_id.type); 6915 err = ext4_journal_stop(handle); 6916 if (!ret) 6917 ret = err; 6918 return ret; 6919 } 6920 6921 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6922 { 6923 struct super_block *sb = dquot->dq_sb; 6924 6925 if (ext4_is_quota_journalled(sb)) { 6926 dquot_mark_dquot_dirty(dquot); 6927 return ext4_write_dquot(dquot); 6928 } else { 6929 return dquot_mark_dquot_dirty(dquot); 6930 } 6931 } 6932 6933 static int ext4_write_info(struct super_block *sb, int type) 6934 { 6935 int ret, err; 6936 handle_t *handle; 6937 6938 /* Data block + inode block */ 6939 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 6940 if (IS_ERR(handle)) 6941 return PTR_ERR(handle); 6942 ret = dquot_commit_info(sb, type); 6943 err = ext4_journal_stop(handle); 6944 if (!ret) 6945 ret = err; 6946 return ret; 6947 } 6948 6949 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6950 { 6951 struct ext4_inode_info *ei = EXT4_I(inode); 6952 6953 /* The first argument of lockdep_set_subclass has to be 6954 * *exactly* the same as the argument to init_rwsem() --- in 6955 * this case, in init_once() --- or lockdep gets unhappy 6956 * because the name of the lock is set using the 6957 * stringification of the argument to init_rwsem(). 6958 */ 6959 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6960 lockdep_set_subclass(&ei->i_data_sem, subclass); 6961 } 6962 6963 /* 6964 * Standard function to be called on quota_on 6965 */ 6966 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6967 const struct path *path) 6968 { 6969 int err; 6970 6971 if (!test_opt(sb, QUOTA)) 6972 return -EINVAL; 6973 6974 /* Quotafile not on the same filesystem? */ 6975 if (path->dentry->d_sb != sb) 6976 return -EXDEV; 6977 6978 /* Quota already enabled for this file? */ 6979 if (IS_NOQUOTA(d_inode(path->dentry))) 6980 return -EBUSY; 6981 6982 /* Journaling quota? */ 6983 if (EXT4_SB(sb)->s_qf_names[type]) { 6984 /* Quotafile not in fs root? */ 6985 if (path->dentry->d_parent != sb->s_root) 6986 ext4_msg(sb, KERN_WARNING, 6987 "Quota file not on filesystem root. " 6988 "Journaled quota will not work"); 6989 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6990 } else { 6991 /* 6992 * Clear the flag just in case mount options changed since 6993 * last time. 6994 */ 6995 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6996 } 6997 6998 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6999 err = dquot_quota_on(sb, type, format_id, path); 7000 if (!err) { 7001 struct inode *inode = d_inode(path->dentry); 7002 handle_t *handle; 7003 7004 /* 7005 * Set inode flags to prevent userspace from messing with quota 7006 * files. If this fails, we return success anyway since quotas 7007 * are already enabled and this is not a hard failure. 7008 */ 7009 inode_lock(inode); 7010 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7011 if (IS_ERR(handle)) 7012 goto unlock_inode; 7013 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 7014 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 7015 S_NOATIME | S_IMMUTABLE); 7016 err = ext4_mark_inode_dirty(handle, inode); 7017 ext4_journal_stop(handle); 7018 unlock_inode: 7019 inode_unlock(inode); 7020 if (err) 7021 dquot_quota_off(sb, type); 7022 } 7023 if (err) 7024 lockdep_set_quota_inode(path->dentry->d_inode, 7025 I_DATA_SEM_NORMAL); 7026 return err; 7027 } 7028 7029 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum) 7030 { 7031 switch (type) { 7032 case USRQUOTA: 7033 return qf_inum == EXT4_USR_QUOTA_INO; 7034 case GRPQUOTA: 7035 return qf_inum == EXT4_GRP_QUOTA_INO; 7036 case PRJQUOTA: 7037 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO; 7038 default: 7039 BUG(); 7040 } 7041 } 7042 7043 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 7044 unsigned int flags) 7045 { 7046 int err; 7047 struct inode *qf_inode; 7048 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7049 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7050 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7051 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7052 }; 7053 7054 BUG_ON(!ext4_has_feature_quota(sb)); 7055 7056 if (!qf_inums[type]) 7057 return -EPERM; 7058 7059 if (!ext4_check_quota_inum(type, qf_inums[type])) { 7060 ext4_error(sb, "Bad quota inum: %lu, type: %d", 7061 qf_inums[type], type); 7062 return -EUCLEAN; 7063 } 7064 7065 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 7066 if (IS_ERR(qf_inode)) { 7067 ext4_error(sb, "Bad quota inode: %lu, type: %d", 7068 qf_inums[type], type); 7069 return PTR_ERR(qf_inode); 7070 } 7071 7072 /* Don't account quota for quota files to avoid recursion */ 7073 qf_inode->i_flags |= S_NOQUOTA; 7074 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 7075 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 7076 if (err) 7077 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 7078 iput(qf_inode); 7079 7080 return err; 7081 } 7082 7083 /* Enable usage tracking for all quota types. */ 7084 int ext4_enable_quotas(struct super_block *sb) 7085 { 7086 int type, err = 0; 7087 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7088 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7089 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7090 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7091 }; 7092 bool quota_mopt[EXT4_MAXQUOTAS] = { 7093 test_opt(sb, USRQUOTA), 7094 test_opt(sb, GRPQUOTA), 7095 test_opt(sb, PRJQUOTA), 7096 }; 7097 7098 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 7099 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 7100 if (qf_inums[type]) { 7101 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 7102 DQUOT_USAGE_ENABLED | 7103 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 7104 if (err) { 7105 ext4_warning(sb, 7106 "Failed to enable quota tracking " 7107 "(type=%d, err=%d, ino=%lu). " 7108 "Please run e2fsck to fix.", type, 7109 err, qf_inums[type]); 7110 7111 ext4_quotas_off(sb, type); 7112 return err; 7113 } 7114 } 7115 } 7116 return 0; 7117 } 7118 7119 static int ext4_quota_off(struct super_block *sb, int type) 7120 { 7121 struct inode *inode = sb_dqopt(sb)->files[type]; 7122 handle_t *handle; 7123 int err; 7124 7125 /* Force all delayed allocation blocks to be allocated. 7126 * Caller already holds s_umount sem */ 7127 if (test_opt(sb, DELALLOC)) 7128 sync_filesystem(sb); 7129 7130 if (!inode || !igrab(inode)) 7131 goto out; 7132 7133 err = dquot_quota_off(sb, type); 7134 if (err || ext4_has_feature_quota(sb)) 7135 goto out_put; 7136 /* 7137 * When the filesystem was remounted read-only first, we cannot cleanup 7138 * inode flags here. Bad luck but people should be using QUOTA feature 7139 * these days anyway. 7140 */ 7141 if (sb_rdonly(sb)) 7142 goto out_put; 7143 7144 inode_lock(inode); 7145 /* 7146 * Update modification times of quota files when userspace can 7147 * start looking at them. If we fail, we return success anyway since 7148 * this is not a hard failure and quotas are already disabled. 7149 */ 7150 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7151 if (IS_ERR(handle)) { 7152 err = PTR_ERR(handle); 7153 goto out_unlock; 7154 } 7155 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7156 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7157 inode->i_mtime = inode_set_ctime_current(inode); 7158 err = ext4_mark_inode_dirty(handle, inode); 7159 ext4_journal_stop(handle); 7160 out_unlock: 7161 inode_unlock(inode); 7162 out_put: 7163 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7164 iput(inode); 7165 return err; 7166 out: 7167 return dquot_quota_off(sb, type); 7168 } 7169 7170 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7171 * acquiring the locks... As quota files are never truncated and quota code 7172 * itself serializes the operations (and no one else should touch the files) 7173 * we don't have to be afraid of races */ 7174 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7175 size_t len, loff_t off) 7176 { 7177 struct inode *inode = sb_dqopt(sb)->files[type]; 7178 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7179 int offset = off & (sb->s_blocksize - 1); 7180 int tocopy; 7181 size_t toread; 7182 struct buffer_head *bh; 7183 loff_t i_size = i_size_read(inode); 7184 7185 if (off > i_size) 7186 return 0; 7187 if (off+len > i_size) 7188 len = i_size-off; 7189 toread = len; 7190 while (toread > 0) { 7191 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread); 7192 bh = ext4_bread(NULL, inode, blk, 0); 7193 if (IS_ERR(bh)) 7194 return PTR_ERR(bh); 7195 if (!bh) /* A hole? */ 7196 memset(data, 0, tocopy); 7197 else 7198 memcpy(data, bh->b_data+offset, tocopy); 7199 brelse(bh); 7200 offset = 0; 7201 toread -= tocopy; 7202 data += tocopy; 7203 blk++; 7204 } 7205 return len; 7206 } 7207 7208 /* Write to quotafile (we know the transaction is already started and has 7209 * enough credits) */ 7210 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7211 const char *data, size_t len, loff_t off) 7212 { 7213 struct inode *inode = sb_dqopt(sb)->files[type]; 7214 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7215 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7216 int retries = 0; 7217 struct buffer_head *bh; 7218 handle_t *handle = journal_current_handle(); 7219 7220 if (!handle) { 7221 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7222 " cancelled because transaction is not started", 7223 (unsigned long long)off, (unsigned long long)len); 7224 return -EIO; 7225 } 7226 /* 7227 * Since we account only one data block in transaction credits, 7228 * then it is impossible to cross a block boundary. 7229 */ 7230 if (sb->s_blocksize - offset < len) { 7231 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7232 " cancelled because not block aligned", 7233 (unsigned long long)off, (unsigned long long)len); 7234 return -EIO; 7235 } 7236 7237 do { 7238 bh = ext4_bread(handle, inode, blk, 7239 EXT4_GET_BLOCKS_CREATE | 7240 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7241 } while (PTR_ERR(bh) == -ENOSPC && 7242 ext4_should_retry_alloc(inode->i_sb, &retries)); 7243 if (IS_ERR(bh)) 7244 return PTR_ERR(bh); 7245 if (!bh) 7246 goto out; 7247 BUFFER_TRACE(bh, "get write access"); 7248 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7249 if (err) { 7250 brelse(bh); 7251 return err; 7252 } 7253 lock_buffer(bh); 7254 memcpy(bh->b_data+offset, data, len); 7255 flush_dcache_page(bh->b_page); 7256 unlock_buffer(bh); 7257 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7258 brelse(bh); 7259 out: 7260 if (inode->i_size < off + len) { 7261 i_size_write(inode, off + len); 7262 EXT4_I(inode)->i_disksize = inode->i_size; 7263 err2 = ext4_mark_inode_dirty(handle, inode); 7264 if (unlikely(err2 && !err)) 7265 err = err2; 7266 } 7267 return err ? err : len; 7268 } 7269 #endif 7270 7271 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7272 static inline void register_as_ext2(void) 7273 { 7274 int err = register_filesystem(&ext2_fs_type); 7275 if (err) 7276 printk(KERN_WARNING 7277 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7278 } 7279 7280 static inline void unregister_as_ext2(void) 7281 { 7282 unregister_filesystem(&ext2_fs_type); 7283 } 7284 7285 static inline int ext2_feature_set_ok(struct super_block *sb) 7286 { 7287 if (ext4_has_unknown_ext2_incompat_features(sb)) 7288 return 0; 7289 if (sb_rdonly(sb)) 7290 return 1; 7291 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7292 return 0; 7293 return 1; 7294 } 7295 #else 7296 static inline void register_as_ext2(void) { } 7297 static inline void unregister_as_ext2(void) { } 7298 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7299 #endif 7300 7301 static inline void register_as_ext3(void) 7302 { 7303 int err = register_filesystem(&ext3_fs_type); 7304 if (err) 7305 printk(KERN_WARNING 7306 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7307 } 7308 7309 static inline void unregister_as_ext3(void) 7310 { 7311 unregister_filesystem(&ext3_fs_type); 7312 } 7313 7314 static inline int ext3_feature_set_ok(struct super_block *sb) 7315 { 7316 if (ext4_has_unknown_ext3_incompat_features(sb)) 7317 return 0; 7318 if (!ext4_has_feature_journal(sb)) 7319 return 0; 7320 if (sb_rdonly(sb)) 7321 return 1; 7322 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7323 return 0; 7324 return 1; 7325 } 7326 7327 static void ext4_kill_sb(struct super_block *sb) 7328 { 7329 struct ext4_sb_info *sbi = EXT4_SB(sb); 7330 struct block_device *journal_bdev = sbi ? sbi->s_journal_bdev : NULL; 7331 7332 kill_block_super(sb); 7333 7334 if (journal_bdev) 7335 blkdev_put(journal_bdev, sb); 7336 } 7337 7338 static struct file_system_type ext4_fs_type = { 7339 .owner = THIS_MODULE, 7340 .name = "ext4", 7341 .init_fs_context = ext4_init_fs_context, 7342 .parameters = ext4_param_specs, 7343 .kill_sb = ext4_kill_sb, 7344 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 7345 }; 7346 MODULE_ALIAS_FS("ext4"); 7347 7348 /* Shared across all ext4 file systems */ 7349 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 7350 7351 static int __init ext4_init_fs(void) 7352 { 7353 int i, err; 7354 7355 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7356 ext4_li_info = NULL; 7357 7358 /* Build-time check for flags consistency */ 7359 ext4_check_flag_values(); 7360 7361 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 7362 init_waitqueue_head(&ext4__ioend_wq[i]); 7363 7364 err = ext4_init_es(); 7365 if (err) 7366 return err; 7367 7368 err = ext4_init_pending(); 7369 if (err) 7370 goto out7; 7371 7372 err = ext4_init_post_read_processing(); 7373 if (err) 7374 goto out6; 7375 7376 err = ext4_init_pageio(); 7377 if (err) 7378 goto out5; 7379 7380 err = ext4_init_system_zone(); 7381 if (err) 7382 goto out4; 7383 7384 err = ext4_init_sysfs(); 7385 if (err) 7386 goto out3; 7387 7388 err = ext4_init_mballoc(); 7389 if (err) 7390 goto out2; 7391 err = init_inodecache(); 7392 if (err) 7393 goto out1; 7394 7395 err = ext4_fc_init_dentry_cache(); 7396 if (err) 7397 goto out05; 7398 7399 register_as_ext3(); 7400 register_as_ext2(); 7401 err = register_filesystem(&ext4_fs_type); 7402 if (err) 7403 goto out; 7404 7405 return 0; 7406 out: 7407 unregister_as_ext2(); 7408 unregister_as_ext3(); 7409 ext4_fc_destroy_dentry_cache(); 7410 out05: 7411 destroy_inodecache(); 7412 out1: 7413 ext4_exit_mballoc(); 7414 out2: 7415 ext4_exit_sysfs(); 7416 out3: 7417 ext4_exit_system_zone(); 7418 out4: 7419 ext4_exit_pageio(); 7420 out5: 7421 ext4_exit_post_read_processing(); 7422 out6: 7423 ext4_exit_pending(); 7424 out7: 7425 ext4_exit_es(); 7426 7427 return err; 7428 } 7429 7430 static void __exit ext4_exit_fs(void) 7431 { 7432 ext4_destroy_lazyinit_thread(); 7433 unregister_as_ext2(); 7434 unregister_as_ext3(); 7435 unregister_filesystem(&ext4_fs_type); 7436 ext4_fc_destroy_dentry_cache(); 7437 destroy_inodecache(); 7438 ext4_exit_mballoc(); 7439 ext4_exit_sysfs(); 7440 ext4_exit_system_zone(); 7441 ext4_exit_pageio(); 7442 ext4_exit_post_read_processing(); 7443 ext4_exit_es(); 7444 ext4_exit_pending(); 7445 } 7446 7447 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7448 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7449 MODULE_LICENSE("GPL"); 7450 MODULE_SOFTDEP("pre: crc32c"); 7451 module_init(ext4_init_fs) 7452 module_exit(ext4_exit_fs) 7453