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