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, bool simu_fail) 165 { 166 if (simu_fail) { 167 clear_buffer_uptodate(bh); 168 unlock_buffer(bh); 169 return; 170 } 171 172 /* 173 * buffer's verified bit is no longer valid after reading from 174 * disk again due to write out error, clear it to make sure we 175 * recheck the buffer contents. 176 */ 177 clear_buffer_verified(bh); 178 179 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 180 get_bh(bh); 181 submit_bh(REQ_OP_READ | op_flags, bh); 182 } 183 184 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags, 185 bh_end_io_t *end_io, bool simu_fail) 186 { 187 BUG_ON(!buffer_locked(bh)); 188 189 if (ext4_buffer_uptodate(bh)) { 190 unlock_buffer(bh); 191 return; 192 } 193 __ext4_read_bh(bh, op_flags, end_io, simu_fail); 194 } 195 196 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 197 bh_end_io_t *end_io, bool simu_fail) 198 { 199 BUG_ON(!buffer_locked(bh)); 200 201 if (ext4_buffer_uptodate(bh)) { 202 unlock_buffer(bh); 203 return 0; 204 } 205 206 __ext4_read_bh(bh, op_flags, end_io, simu_fail); 207 208 wait_on_buffer(bh); 209 if (buffer_uptodate(bh)) 210 return 0; 211 return -EIO; 212 } 213 214 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 215 { 216 lock_buffer(bh); 217 if (!wait) { 218 ext4_read_bh_nowait(bh, op_flags, NULL, false); 219 return 0; 220 } 221 return ext4_read_bh(bh, op_flags, NULL, false); 222 } 223 224 /* 225 * This works like __bread_gfp() except it uses ERR_PTR for error 226 * returns. Currently with sb_bread it's impossible to distinguish 227 * between ENOMEM and EIO situations (since both result in a NULL 228 * return. 229 */ 230 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 231 sector_t block, 232 blk_opf_t op_flags, gfp_t gfp) 233 { 234 struct buffer_head *bh; 235 int ret; 236 237 bh = sb_getblk_gfp(sb, block, gfp); 238 if (bh == NULL) 239 return ERR_PTR(-ENOMEM); 240 if (ext4_buffer_uptodate(bh)) 241 return bh; 242 243 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 244 if (ret) { 245 put_bh(bh); 246 return ERR_PTR(ret); 247 } 248 return bh; 249 } 250 251 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 252 blk_opf_t op_flags) 253 { 254 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 255 } 256 257 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 258 sector_t block) 259 { 260 return __ext4_sb_bread_gfp(sb, block, 0, 0); 261 } 262 263 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 264 { 265 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 266 267 if (likely(bh)) { 268 if (trylock_buffer(bh)) 269 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL, false); 270 brelse(bh); 271 } 272 } 273 274 static int ext4_verify_csum_type(struct super_block *sb, 275 struct ext4_super_block *es) 276 { 277 if (!ext4_has_feature_metadata_csum(sb)) 278 return 1; 279 280 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 281 } 282 283 __le32 ext4_superblock_csum(struct super_block *sb, 284 struct ext4_super_block *es) 285 { 286 struct ext4_sb_info *sbi = EXT4_SB(sb); 287 int offset = offsetof(struct ext4_super_block, s_checksum); 288 __u32 csum; 289 290 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 291 292 return cpu_to_le32(csum); 293 } 294 295 static int ext4_superblock_csum_verify(struct super_block *sb, 296 struct ext4_super_block *es) 297 { 298 if (!ext4_has_metadata_csum(sb)) 299 return 1; 300 301 return es->s_checksum == ext4_superblock_csum(sb, es); 302 } 303 304 void ext4_superblock_csum_set(struct super_block *sb) 305 { 306 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 307 308 if (!ext4_has_metadata_csum(sb)) 309 return; 310 311 es->s_checksum = ext4_superblock_csum(sb, es); 312 } 313 314 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 315 struct ext4_group_desc *bg) 316 { 317 return le32_to_cpu(bg->bg_block_bitmap_lo) | 318 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 319 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 320 } 321 322 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 323 struct ext4_group_desc *bg) 324 { 325 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 326 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 327 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 328 } 329 330 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 331 struct ext4_group_desc *bg) 332 { 333 return le32_to_cpu(bg->bg_inode_table_lo) | 334 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 335 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 336 } 337 338 __u32 ext4_free_group_clusters(struct super_block *sb, 339 struct ext4_group_desc *bg) 340 { 341 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 342 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 343 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 344 } 345 346 __u32 ext4_free_inodes_count(struct super_block *sb, 347 struct ext4_group_desc *bg) 348 { 349 return le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_lo)) | 350 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 351 (__u32)le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_hi)) << 16 : 0); 352 } 353 354 __u32 ext4_used_dirs_count(struct super_block *sb, 355 struct ext4_group_desc *bg) 356 { 357 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 358 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 359 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 360 } 361 362 __u32 ext4_itable_unused_count(struct super_block *sb, 363 struct ext4_group_desc *bg) 364 { 365 return le16_to_cpu(bg->bg_itable_unused_lo) | 366 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 367 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 368 } 369 370 void ext4_block_bitmap_set(struct super_block *sb, 371 struct ext4_group_desc *bg, ext4_fsblk_t blk) 372 { 373 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 374 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 375 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 376 } 377 378 void ext4_inode_bitmap_set(struct super_block *sb, 379 struct ext4_group_desc *bg, ext4_fsblk_t blk) 380 { 381 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 382 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 383 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 384 } 385 386 void ext4_inode_table_set(struct super_block *sb, 387 struct ext4_group_desc *bg, ext4_fsblk_t blk) 388 { 389 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 390 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 391 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 392 } 393 394 void ext4_free_group_clusters_set(struct super_block *sb, 395 struct ext4_group_desc *bg, __u32 count) 396 { 397 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 398 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 399 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 400 } 401 402 void ext4_free_inodes_set(struct super_block *sb, 403 struct ext4_group_desc *bg, __u32 count) 404 { 405 WRITE_ONCE(bg->bg_free_inodes_count_lo, cpu_to_le16((__u16)count)); 406 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 407 WRITE_ONCE(bg->bg_free_inodes_count_hi, cpu_to_le16(count >> 16)); 408 } 409 410 void ext4_used_dirs_set(struct super_block *sb, 411 struct ext4_group_desc *bg, __u32 count) 412 { 413 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 414 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 415 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 416 } 417 418 void ext4_itable_unused_set(struct super_block *sb, 419 struct ext4_group_desc *bg, __u32 count) 420 { 421 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 422 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 423 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 424 } 425 426 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 427 { 428 now = clamp_val(now, 0, (1ull << 40) - 1); 429 430 *lo = cpu_to_le32(lower_32_bits(now)); 431 *hi = upper_32_bits(now); 432 } 433 434 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 435 { 436 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 437 } 438 #define ext4_update_tstamp(es, tstamp) \ 439 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 440 ktime_get_real_seconds()) 441 #define ext4_get_tstamp(es, tstamp) \ 442 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 443 444 #define EXT4_SB_REFRESH_INTERVAL_SEC (3600) /* seconds (1 hour) */ 445 #define EXT4_SB_REFRESH_INTERVAL_KB (16384) /* kilobytes (16MB) */ 446 447 /* 448 * The ext4_maybe_update_superblock() function checks and updates the 449 * superblock if needed. 450 * 451 * This function is designed to update the on-disk superblock only under 452 * certain conditions to prevent excessive disk writes and unnecessary 453 * waking of the disk from sleep. The superblock will be updated if: 454 * 1. More than an hour has passed since the last superblock update, and 455 * 2. More than 16MB have been written since the last superblock update. 456 * 457 * @sb: The superblock 458 */ 459 static void ext4_maybe_update_superblock(struct super_block *sb) 460 { 461 struct ext4_sb_info *sbi = EXT4_SB(sb); 462 struct ext4_super_block *es = sbi->s_es; 463 journal_t *journal = sbi->s_journal; 464 time64_t now; 465 __u64 last_update; 466 __u64 lifetime_write_kbytes; 467 __u64 diff_size; 468 469 if (sb_rdonly(sb) || !(sb->s_flags & SB_ACTIVE) || 470 !journal || (journal->j_flags & JBD2_UNMOUNT)) 471 return; 472 473 now = ktime_get_real_seconds(); 474 last_update = ext4_get_tstamp(es, s_wtime); 475 476 if (likely(now - last_update < EXT4_SB_REFRESH_INTERVAL_SEC)) 477 return; 478 479 lifetime_write_kbytes = sbi->s_kbytes_written + 480 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 481 sbi->s_sectors_written_start) >> 1); 482 483 /* Get the number of kilobytes not written to disk to account 484 * for statistics and compare with a multiple of 16 MB. This 485 * is used to determine when the next superblock commit should 486 * occur (i.e. not more often than once per 16MB if there was 487 * less written in an hour). 488 */ 489 diff_size = lifetime_write_kbytes - le64_to_cpu(es->s_kbytes_written); 490 491 if (diff_size > EXT4_SB_REFRESH_INTERVAL_KB) 492 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 493 } 494 495 /* 496 * The del_gendisk() function uninitializes the disk-specific data 497 * structures, including the bdi structure, without telling anyone 498 * else. Once this happens, any attempt to call mark_buffer_dirty() 499 * (for example, by ext4_commit_super), will cause a kernel OOPS. 500 * This is a kludge to prevent these oops until we can put in a proper 501 * hook in del_gendisk() to inform the VFS and file system layers. 502 */ 503 static int block_device_ejected(struct super_block *sb) 504 { 505 struct inode *bd_inode = sb->s_bdev->bd_inode; 506 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 507 508 return bdi->dev == NULL; 509 } 510 511 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 512 { 513 struct super_block *sb = journal->j_private; 514 struct ext4_sb_info *sbi = EXT4_SB(sb); 515 int error = is_journal_aborted(journal); 516 struct ext4_journal_cb_entry *jce; 517 518 BUG_ON(txn->t_state == T_FINISHED); 519 520 ext4_process_freed_data(sb, txn->t_tid); 521 ext4_maybe_update_superblock(sb); 522 523 spin_lock(&sbi->s_md_lock); 524 while (!list_empty(&txn->t_private_list)) { 525 jce = list_entry(txn->t_private_list.next, 526 struct ext4_journal_cb_entry, jce_list); 527 list_del_init(&jce->jce_list); 528 spin_unlock(&sbi->s_md_lock); 529 jce->jce_func(sb, jce, error); 530 spin_lock(&sbi->s_md_lock); 531 } 532 spin_unlock(&sbi->s_md_lock); 533 } 534 535 /* 536 * This writepage callback for write_cache_pages() 537 * takes care of a few cases after page cleaning. 538 * 539 * write_cache_pages() already checks for dirty pages 540 * and calls clear_page_dirty_for_io(), which we want, 541 * to write protect the pages. 542 * 543 * However, we may have to redirty a page (see below.) 544 */ 545 static int ext4_journalled_writepage_callback(struct folio *folio, 546 struct writeback_control *wbc, 547 void *data) 548 { 549 transaction_t *transaction = (transaction_t *) data; 550 struct buffer_head *bh, *head; 551 struct journal_head *jh; 552 553 bh = head = folio_buffers(folio); 554 do { 555 /* 556 * We have to redirty a page in these cases: 557 * 1) If buffer is dirty, it means the page was dirty because it 558 * contains a buffer that needs checkpointing. So the dirty bit 559 * needs to be preserved so that checkpointing writes the buffer 560 * properly. 561 * 2) If buffer is not part of the committing transaction 562 * (we may have just accidentally come across this buffer because 563 * inode range tracking is not exact) or if the currently running 564 * transaction already contains this buffer as well, dirty bit 565 * needs to be preserved so that the buffer gets writeprotected 566 * properly on running transaction's commit. 567 */ 568 jh = bh2jh(bh); 569 if (buffer_dirty(bh) || 570 (jh && (jh->b_transaction != transaction || 571 jh->b_next_transaction))) { 572 folio_redirty_for_writepage(wbc, folio); 573 goto out; 574 } 575 } while ((bh = bh->b_this_page) != head); 576 577 out: 578 return AOP_WRITEPAGE_ACTIVATE; 579 } 580 581 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 582 { 583 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 584 struct writeback_control wbc = { 585 .sync_mode = WB_SYNC_ALL, 586 .nr_to_write = LONG_MAX, 587 .range_start = jinode->i_dirty_start, 588 .range_end = jinode->i_dirty_end, 589 }; 590 591 return write_cache_pages(mapping, &wbc, 592 ext4_journalled_writepage_callback, 593 jinode->i_transaction); 594 } 595 596 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 597 { 598 int ret; 599 600 if (ext4_should_journal_data(jinode->i_vfs_inode)) 601 ret = ext4_journalled_submit_inode_data_buffers(jinode); 602 else 603 ret = ext4_normal_submit_inode_data_buffers(jinode); 604 return ret; 605 } 606 607 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 608 { 609 int ret = 0; 610 611 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 612 ret = jbd2_journal_finish_inode_data_buffers(jinode); 613 614 return ret; 615 } 616 617 static bool system_going_down(void) 618 { 619 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 620 || system_state == SYSTEM_RESTART; 621 } 622 623 struct ext4_err_translation { 624 int code; 625 int errno; 626 }; 627 628 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 629 630 static struct ext4_err_translation err_translation[] = { 631 EXT4_ERR_TRANSLATE(EIO), 632 EXT4_ERR_TRANSLATE(ENOMEM), 633 EXT4_ERR_TRANSLATE(EFSBADCRC), 634 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 635 EXT4_ERR_TRANSLATE(ENOSPC), 636 EXT4_ERR_TRANSLATE(ENOKEY), 637 EXT4_ERR_TRANSLATE(EROFS), 638 EXT4_ERR_TRANSLATE(EFBIG), 639 EXT4_ERR_TRANSLATE(EEXIST), 640 EXT4_ERR_TRANSLATE(ERANGE), 641 EXT4_ERR_TRANSLATE(EOVERFLOW), 642 EXT4_ERR_TRANSLATE(EBUSY), 643 EXT4_ERR_TRANSLATE(ENOTDIR), 644 EXT4_ERR_TRANSLATE(ENOTEMPTY), 645 EXT4_ERR_TRANSLATE(ESHUTDOWN), 646 EXT4_ERR_TRANSLATE(EFAULT), 647 }; 648 649 static int ext4_errno_to_code(int errno) 650 { 651 int i; 652 653 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 654 if (err_translation[i].errno == errno) 655 return err_translation[i].code; 656 return EXT4_ERR_UNKNOWN; 657 } 658 659 static void save_error_info(struct super_block *sb, int error, 660 __u32 ino, __u64 block, 661 const char *func, unsigned int line) 662 { 663 struct ext4_sb_info *sbi = EXT4_SB(sb); 664 665 /* We default to EFSCORRUPTED error... */ 666 if (error == 0) 667 error = EFSCORRUPTED; 668 669 spin_lock(&sbi->s_error_lock); 670 sbi->s_add_error_count++; 671 sbi->s_last_error_code = error; 672 sbi->s_last_error_line = line; 673 sbi->s_last_error_ino = ino; 674 sbi->s_last_error_block = block; 675 sbi->s_last_error_func = func; 676 sbi->s_last_error_time = ktime_get_real_seconds(); 677 if (!sbi->s_first_error_time) { 678 sbi->s_first_error_code = error; 679 sbi->s_first_error_line = line; 680 sbi->s_first_error_ino = ino; 681 sbi->s_first_error_block = block; 682 sbi->s_first_error_func = func; 683 sbi->s_first_error_time = sbi->s_last_error_time; 684 } 685 spin_unlock(&sbi->s_error_lock); 686 } 687 688 /* Deal with the reporting of failure conditions on a filesystem such as 689 * inconsistencies detected or read IO failures. 690 * 691 * On ext2, we can store the error state of the filesystem in the 692 * superblock. That is not possible on ext4, because we may have other 693 * write ordering constraints on the superblock which prevent us from 694 * writing it out straight away; and given that the journal is about to 695 * be aborted, we can't rely on the current, or future, transactions to 696 * write out the superblock safely. 697 * 698 * We'll just use the jbd2_journal_abort() error code to record an error in 699 * the journal instead. On recovery, the journal will complain about 700 * that error until we've noted it down and cleared it. 701 * 702 * If force_ro is set, we unconditionally force the filesystem into an 703 * ABORT|READONLY state, unless the error response on the fs has been set to 704 * panic in which case we take the easy way out and panic immediately. This is 705 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 706 * at a critical moment in log management. 707 */ 708 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 709 __u32 ino, __u64 block, 710 const char *func, unsigned int line) 711 { 712 journal_t *journal = EXT4_SB(sb)->s_journal; 713 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 714 715 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 716 if (test_opt(sb, WARN_ON_ERROR)) 717 WARN_ON_ONCE(1); 718 719 if (!continue_fs && !sb_rdonly(sb)) { 720 set_bit(EXT4_FLAGS_SHUTDOWN, &EXT4_SB(sb)->s_ext4_flags); 721 if (journal) 722 jbd2_journal_abort(journal, -EIO); 723 } 724 725 if (!bdev_read_only(sb->s_bdev)) { 726 save_error_info(sb, error, ino, block, func, line); 727 /* 728 * In case the fs should keep running, we need to writeout 729 * superblock through the journal. Due to lock ordering 730 * constraints, it may not be safe to do it right here so we 731 * defer superblock flushing to a workqueue. 732 */ 733 if (continue_fs && journal) 734 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 735 else 736 ext4_commit_super(sb); 737 } 738 739 /* 740 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 741 * could panic during 'reboot -f' as the underlying device got already 742 * disabled. 743 */ 744 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 745 panic("EXT4-fs (device %s): panic forced after error\n", 746 sb->s_id); 747 } 748 749 if (sb_rdonly(sb) || continue_fs) 750 return; 751 752 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 753 /* 754 * EXT4_FLAGS_SHUTDOWN was set which stops all filesystem 755 * modifications. We don't set SB_RDONLY because that requires 756 * sb->s_umount semaphore and setting it without proper remount 757 * procedure is confusing code such as freeze_super() leading to 758 * deadlocks and other problems. 759 */ 760 } 761 762 static void update_super_work(struct work_struct *work) 763 { 764 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 765 s_sb_upd_work); 766 journal_t *journal = sbi->s_journal; 767 handle_t *handle; 768 769 /* 770 * If the journal is still running, we have to write out superblock 771 * through the journal to avoid collisions of other journalled sb 772 * updates. 773 * 774 * We use directly jbd2 functions here to avoid recursing back into 775 * ext4 error handling code during handling of previous errors. 776 */ 777 if (!sb_rdonly(sbi->s_sb) && journal) { 778 struct buffer_head *sbh = sbi->s_sbh; 779 bool call_notify_err = false; 780 781 handle = jbd2_journal_start(journal, 1); 782 if (IS_ERR(handle)) 783 goto write_directly; 784 if (jbd2_journal_get_write_access(handle, sbh)) { 785 jbd2_journal_stop(handle); 786 goto write_directly; 787 } 788 789 if (sbi->s_add_error_count > 0) 790 call_notify_err = true; 791 792 ext4_update_super(sbi->s_sb); 793 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 794 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 795 "superblock detected"); 796 clear_buffer_write_io_error(sbh); 797 set_buffer_uptodate(sbh); 798 } 799 800 if (jbd2_journal_dirty_metadata(handle, sbh)) { 801 jbd2_journal_stop(handle); 802 goto write_directly; 803 } 804 jbd2_journal_stop(handle); 805 806 if (call_notify_err) 807 ext4_notify_error_sysfs(sbi); 808 809 return; 810 } 811 write_directly: 812 /* 813 * Write through journal failed. Write sb directly to get error info 814 * out and hope for the best. 815 */ 816 ext4_commit_super(sbi->s_sb); 817 ext4_notify_error_sysfs(sbi); 818 } 819 820 #define ext4_error_ratelimit(sb) \ 821 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 822 "EXT4-fs error") 823 824 void __ext4_error(struct super_block *sb, const char *function, 825 unsigned int line, bool force_ro, int error, __u64 block, 826 const char *fmt, ...) 827 { 828 struct va_format vaf; 829 va_list args; 830 831 if (unlikely(ext4_forced_shutdown(sb))) 832 return; 833 834 trace_ext4_error(sb, function, line); 835 if (ext4_error_ratelimit(sb)) { 836 va_start(args, fmt); 837 vaf.fmt = fmt; 838 vaf.va = &args; 839 printk(KERN_CRIT 840 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 841 sb->s_id, function, line, current->comm, &vaf); 842 va_end(args); 843 } 844 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED); 845 846 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 847 } 848 849 void __ext4_error_inode(struct inode *inode, const char *function, 850 unsigned int line, ext4_fsblk_t block, int error, 851 const char *fmt, ...) 852 { 853 va_list args; 854 struct va_format vaf; 855 856 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 857 return; 858 859 trace_ext4_error(inode->i_sb, function, line); 860 if (ext4_error_ratelimit(inode->i_sb)) { 861 va_start(args, fmt); 862 vaf.fmt = fmt; 863 vaf.va = &args; 864 if (block) 865 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 866 "inode #%lu: block %llu: comm %s: %pV\n", 867 inode->i_sb->s_id, function, line, inode->i_ino, 868 block, current->comm, &vaf); 869 else 870 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 871 "inode #%lu: comm %s: %pV\n", 872 inode->i_sb->s_id, function, line, inode->i_ino, 873 current->comm, &vaf); 874 va_end(args); 875 } 876 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED); 877 878 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 879 function, line); 880 } 881 882 void __ext4_error_file(struct file *file, const char *function, 883 unsigned int line, ext4_fsblk_t block, 884 const char *fmt, ...) 885 { 886 va_list args; 887 struct va_format vaf; 888 struct inode *inode = file_inode(file); 889 char pathname[80], *path; 890 891 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 892 return; 893 894 trace_ext4_error(inode->i_sb, function, line); 895 if (ext4_error_ratelimit(inode->i_sb)) { 896 path = file_path(file, pathname, sizeof(pathname)); 897 if (IS_ERR(path)) 898 path = "(unknown)"; 899 va_start(args, fmt); 900 vaf.fmt = fmt; 901 vaf.va = &args; 902 if (block) 903 printk(KERN_CRIT 904 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 905 "block %llu: comm %s: path %s: %pV\n", 906 inode->i_sb->s_id, function, line, inode->i_ino, 907 block, current->comm, path, &vaf); 908 else 909 printk(KERN_CRIT 910 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 911 "comm %s: path %s: %pV\n", 912 inode->i_sb->s_id, function, line, inode->i_ino, 913 current->comm, path, &vaf); 914 va_end(args); 915 } 916 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED); 917 918 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 919 function, line); 920 } 921 922 const char *ext4_decode_error(struct super_block *sb, int errno, 923 char nbuf[16]) 924 { 925 char *errstr = NULL; 926 927 switch (errno) { 928 case -EFSCORRUPTED: 929 errstr = "Corrupt filesystem"; 930 break; 931 case -EFSBADCRC: 932 errstr = "Filesystem failed CRC"; 933 break; 934 case -EIO: 935 errstr = "IO failure"; 936 break; 937 case -ENOMEM: 938 errstr = "Out of memory"; 939 break; 940 case -EROFS: 941 if (!sb || (EXT4_SB(sb)->s_journal && 942 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 943 errstr = "Journal has aborted"; 944 else 945 errstr = "Readonly filesystem"; 946 break; 947 default: 948 /* If the caller passed in an extra buffer for unknown 949 * errors, textualise them now. Else we just return 950 * NULL. */ 951 if (nbuf) { 952 /* Check for truncated error codes... */ 953 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 954 errstr = nbuf; 955 } 956 break; 957 } 958 959 return errstr; 960 } 961 962 /* __ext4_std_error decodes expected errors from journaling functions 963 * automatically and invokes the appropriate error response. */ 964 965 void __ext4_std_error(struct super_block *sb, const char *function, 966 unsigned int line, int errno) 967 { 968 char nbuf[16]; 969 const char *errstr; 970 971 if (unlikely(ext4_forced_shutdown(sb))) 972 return; 973 974 /* Special case: if the error is EROFS, and we're not already 975 * inside a transaction, then there's really no point in logging 976 * an error. */ 977 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 978 return; 979 980 if (ext4_error_ratelimit(sb)) { 981 errstr = ext4_decode_error(sb, errno, nbuf); 982 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 983 sb->s_id, function, line, errstr); 984 } 985 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED); 986 987 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 988 } 989 990 void __ext4_msg(struct super_block *sb, 991 const char *prefix, const char *fmt, ...) 992 { 993 struct va_format vaf; 994 va_list args; 995 996 if (sb) { 997 atomic_inc(&EXT4_SB(sb)->s_msg_count); 998 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 999 "EXT4-fs")) 1000 return; 1001 } 1002 1003 va_start(args, fmt); 1004 vaf.fmt = fmt; 1005 vaf.va = &args; 1006 if (sb) 1007 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 1008 else 1009 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 1010 va_end(args); 1011 } 1012 1013 static int ext4_warning_ratelimit(struct super_block *sb) 1014 { 1015 atomic_inc(&EXT4_SB(sb)->s_warning_count); 1016 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 1017 "EXT4-fs warning"); 1018 } 1019 1020 void __ext4_warning(struct super_block *sb, const char *function, 1021 unsigned int line, const char *fmt, ...) 1022 { 1023 struct va_format vaf; 1024 va_list args; 1025 1026 if (!ext4_warning_ratelimit(sb)) 1027 return; 1028 1029 va_start(args, fmt); 1030 vaf.fmt = fmt; 1031 vaf.va = &args; 1032 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 1033 sb->s_id, function, line, &vaf); 1034 va_end(args); 1035 } 1036 1037 void __ext4_warning_inode(const struct inode *inode, const char *function, 1038 unsigned int line, const char *fmt, ...) 1039 { 1040 struct va_format vaf; 1041 va_list args; 1042 1043 if (!ext4_warning_ratelimit(inode->i_sb)) 1044 return; 1045 1046 va_start(args, fmt); 1047 vaf.fmt = fmt; 1048 vaf.va = &args; 1049 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 1050 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 1051 function, line, inode->i_ino, current->comm, &vaf); 1052 va_end(args); 1053 } 1054 1055 void __ext4_grp_locked_error(const char *function, unsigned int line, 1056 struct super_block *sb, ext4_group_t grp, 1057 unsigned long ino, ext4_fsblk_t block, 1058 const char *fmt, ...) 1059 __releases(bitlock) 1060 __acquires(bitlock) 1061 { 1062 struct va_format vaf; 1063 va_list args; 1064 1065 if (unlikely(ext4_forced_shutdown(sb))) 1066 return; 1067 1068 trace_ext4_error(sb, function, line); 1069 if (ext4_error_ratelimit(sb)) { 1070 va_start(args, fmt); 1071 vaf.fmt = fmt; 1072 vaf.va = &args; 1073 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1074 sb->s_id, function, line, grp); 1075 if (ino) 1076 printk(KERN_CONT "inode %lu: ", ino); 1077 if (block) 1078 printk(KERN_CONT "block %llu:", 1079 (unsigned long long) block); 1080 printk(KERN_CONT "%pV\n", &vaf); 1081 va_end(args); 1082 } 1083 1084 if (test_opt(sb, ERRORS_CONT)) { 1085 if (test_opt(sb, WARN_ON_ERROR)) 1086 WARN_ON_ONCE(1); 1087 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1088 if (!bdev_read_only(sb->s_bdev)) { 1089 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1090 line); 1091 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 1092 } 1093 return; 1094 } 1095 ext4_unlock_group(sb, grp); 1096 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1097 /* 1098 * We only get here in the ERRORS_RO case; relocking the group 1099 * may be dangerous, but nothing bad will happen since the 1100 * filesystem will have already been marked read/only and the 1101 * journal has been aborted. We return 1 as a hint to callers 1102 * who might what to use the return value from 1103 * ext4_grp_locked_error() to distinguish between the 1104 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1105 * aggressively from the ext4 function in question, with a 1106 * more appropriate error code. 1107 */ 1108 ext4_lock_group(sb, grp); 1109 return; 1110 } 1111 1112 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1113 ext4_group_t group, 1114 unsigned int flags) 1115 { 1116 struct ext4_sb_info *sbi = EXT4_SB(sb); 1117 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1118 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1119 int ret; 1120 1121 if (!grp || !gdp) 1122 return; 1123 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1124 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1125 &grp->bb_state); 1126 if (!ret) 1127 percpu_counter_sub(&sbi->s_freeclusters_counter, 1128 grp->bb_free); 1129 } 1130 1131 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1132 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1133 &grp->bb_state); 1134 if (!ret && gdp) { 1135 int count; 1136 1137 count = ext4_free_inodes_count(sb, gdp); 1138 percpu_counter_sub(&sbi->s_freeinodes_counter, 1139 count); 1140 } 1141 } 1142 } 1143 1144 void ext4_update_dynamic_rev(struct super_block *sb) 1145 { 1146 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1147 1148 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1149 return; 1150 1151 ext4_warning(sb, 1152 "updating to rev %d because of new feature flag, " 1153 "running e2fsck is recommended", 1154 EXT4_DYNAMIC_REV); 1155 1156 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1157 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1158 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1159 /* leave es->s_feature_*compat flags alone */ 1160 /* es->s_uuid will be set by e2fsck if empty */ 1161 1162 /* 1163 * The rest of the superblock fields should be zero, and if not it 1164 * means they are likely already in use, so leave them alone. We 1165 * can leave it up to e2fsck to clean up any inconsistencies there. 1166 */ 1167 } 1168 1169 static inline struct inode *orphan_list_entry(struct list_head *l) 1170 { 1171 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1172 } 1173 1174 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1175 { 1176 struct list_head *l; 1177 1178 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1179 le32_to_cpu(sbi->s_es->s_last_orphan)); 1180 1181 printk(KERN_ERR "sb_info orphan list:\n"); 1182 list_for_each(l, &sbi->s_orphan) { 1183 struct inode *inode = orphan_list_entry(l); 1184 printk(KERN_ERR " " 1185 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1186 inode->i_sb->s_id, inode->i_ino, inode, 1187 inode->i_mode, inode->i_nlink, 1188 NEXT_ORPHAN(inode)); 1189 } 1190 } 1191 1192 #ifdef CONFIG_QUOTA 1193 static int ext4_quota_off(struct super_block *sb, int type); 1194 1195 static inline void ext4_quotas_off(struct super_block *sb, int type) 1196 { 1197 BUG_ON(type > EXT4_MAXQUOTAS); 1198 1199 /* Use our quota_off function to clear inode flags etc. */ 1200 for (type--; type >= 0; type--) 1201 ext4_quota_off(sb, type); 1202 } 1203 1204 /* 1205 * This is a helper function which is used in the mount/remount 1206 * codepaths (which holds s_umount) to fetch the quota file name. 1207 */ 1208 static inline char *get_qf_name(struct super_block *sb, 1209 struct ext4_sb_info *sbi, 1210 int type) 1211 { 1212 return rcu_dereference_protected(sbi->s_qf_names[type], 1213 lockdep_is_held(&sb->s_umount)); 1214 } 1215 #else 1216 static inline void ext4_quotas_off(struct super_block *sb, int type) 1217 { 1218 } 1219 #endif 1220 1221 static int ext4_percpu_param_init(struct ext4_sb_info *sbi) 1222 { 1223 ext4_fsblk_t block; 1224 int err; 1225 1226 block = ext4_count_free_clusters(sbi->s_sb); 1227 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block)); 1228 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 1229 GFP_KERNEL); 1230 if (!err) { 1231 unsigned long freei = ext4_count_free_inodes(sbi->s_sb); 1232 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 1233 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 1234 GFP_KERNEL); 1235 } 1236 if (!err) 1237 err = percpu_counter_init(&sbi->s_dirs_counter, 1238 ext4_count_dirs(sbi->s_sb), GFP_KERNEL); 1239 if (!err) 1240 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 1241 GFP_KERNEL); 1242 if (!err) 1243 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 1244 GFP_KERNEL); 1245 if (!err) 1246 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 1247 1248 if (err) 1249 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory"); 1250 1251 return err; 1252 } 1253 1254 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi) 1255 { 1256 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1257 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1258 percpu_counter_destroy(&sbi->s_dirs_counter); 1259 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1260 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1261 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1262 } 1263 1264 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 1265 { 1266 struct buffer_head **group_desc; 1267 int i; 1268 1269 rcu_read_lock(); 1270 group_desc = rcu_dereference(sbi->s_group_desc); 1271 for (i = 0; i < sbi->s_gdb_count; i++) 1272 brelse(group_desc[i]); 1273 kvfree(group_desc); 1274 rcu_read_unlock(); 1275 } 1276 1277 static void ext4_flex_groups_free(struct ext4_sb_info *sbi) 1278 { 1279 struct flex_groups **flex_groups; 1280 int i; 1281 1282 rcu_read_lock(); 1283 flex_groups = rcu_dereference(sbi->s_flex_groups); 1284 if (flex_groups) { 1285 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1286 kvfree(flex_groups[i]); 1287 kvfree(flex_groups); 1288 } 1289 rcu_read_unlock(); 1290 } 1291 1292 static void ext4_put_super(struct super_block *sb) 1293 { 1294 struct ext4_sb_info *sbi = EXT4_SB(sb); 1295 struct ext4_super_block *es = sbi->s_es; 1296 int aborted = 0; 1297 int err; 1298 1299 /* 1300 * Unregister sysfs before destroying jbd2 journal. 1301 * Since we could still access attr_journal_task attribute via sysfs 1302 * path which could have sbi->s_journal->j_task as NULL 1303 * Unregister sysfs before flush sbi->s_sb_upd_work. 1304 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1305 * read metadata verify failed then will queue error work. 1306 * update_super_work will call start_this_handle may trigger 1307 * BUG_ON. 1308 */ 1309 ext4_unregister_sysfs(sb); 1310 1311 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1312 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.", 1313 &sb->s_uuid); 1314 1315 ext4_unregister_li_request(sb); 1316 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 1317 1318 flush_work(&sbi->s_sb_upd_work); 1319 destroy_workqueue(sbi->rsv_conversion_wq); 1320 ext4_release_orphan_info(sb); 1321 1322 if (sbi->s_journal) { 1323 aborted = is_journal_aborted(sbi->s_journal); 1324 err = jbd2_journal_destroy(sbi->s_journal); 1325 sbi->s_journal = NULL; 1326 if ((err < 0) && !aborted) { 1327 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1328 } 1329 } 1330 1331 ext4_es_unregister_shrinker(sbi); 1332 timer_shutdown_sync(&sbi->s_err_report); 1333 ext4_release_system_zone(sb); 1334 ext4_mb_release(sb); 1335 ext4_ext_release(sb); 1336 1337 if (!sb_rdonly(sb) && !aborted) { 1338 ext4_clear_feature_journal_needs_recovery(sb); 1339 ext4_clear_feature_orphan_present(sb); 1340 es->s_state = cpu_to_le16(sbi->s_mount_state); 1341 } 1342 if (!sb_rdonly(sb)) 1343 ext4_commit_super(sb); 1344 1345 ext4_group_desc_free(sbi); 1346 ext4_flex_groups_free(sbi); 1347 ext4_percpu_param_destroy(sbi); 1348 #ifdef CONFIG_QUOTA 1349 for (int i = 0; i < EXT4_MAXQUOTAS; i++) 1350 kfree(get_qf_name(sb, sbi, i)); 1351 #endif 1352 1353 /* Debugging code just in case the in-memory inode orphan list 1354 * isn't empty. The on-disk one can be non-empty if we've 1355 * detected an error and taken the fs readonly, but the 1356 * in-memory list had better be clean by this point. */ 1357 if (!list_empty(&sbi->s_orphan)) 1358 dump_orphan_list(sb, sbi); 1359 ASSERT(list_empty(&sbi->s_orphan)); 1360 1361 sync_blockdev(sb->s_bdev); 1362 invalidate_bdev(sb->s_bdev); 1363 if (sbi->s_journal_bdev) { 1364 /* 1365 * Invalidate the journal device's buffers. We don't want them 1366 * floating about in memory - the physical journal device may 1367 * hotswapped, and it breaks the `ro-after' testing code. 1368 */ 1369 sync_blockdev(sbi->s_journal_bdev); 1370 invalidate_bdev(sbi->s_journal_bdev); 1371 } 1372 1373 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1374 sbi->s_ea_inode_cache = NULL; 1375 1376 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1377 sbi->s_ea_block_cache = NULL; 1378 1379 ext4_stop_mmpd(sbi); 1380 1381 brelse(sbi->s_sbh); 1382 sb->s_fs_info = NULL; 1383 /* 1384 * Now that we are completely done shutting down the 1385 * superblock, we need to actually destroy the kobject. 1386 */ 1387 kobject_put(&sbi->s_kobj); 1388 wait_for_completion(&sbi->s_kobj_unregister); 1389 if (sbi->s_chksum_driver) 1390 crypto_free_shash(sbi->s_chksum_driver); 1391 kfree(sbi->s_blockgroup_lock); 1392 fs_put_dax(sbi->s_daxdev, NULL); 1393 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1394 #if IS_ENABLED(CONFIG_UNICODE) 1395 utf8_unload(sb->s_encoding); 1396 #endif 1397 kfree(sbi); 1398 } 1399 1400 static struct kmem_cache *ext4_inode_cachep; 1401 1402 /* 1403 * Called inside transaction, so use GFP_NOFS 1404 */ 1405 static struct inode *ext4_alloc_inode(struct super_block *sb) 1406 { 1407 struct ext4_inode_info *ei; 1408 1409 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1410 if (!ei) 1411 return NULL; 1412 1413 inode_set_iversion(&ei->vfs_inode, 1); 1414 ei->i_flags = 0; 1415 spin_lock_init(&ei->i_raw_lock); 1416 ei->i_prealloc_node = RB_ROOT; 1417 atomic_set(&ei->i_prealloc_active, 0); 1418 rwlock_init(&ei->i_prealloc_lock); 1419 ext4_es_init_tree(&ei->i_es_tree); 1420 rwlock_init(&ei->i_es_lock); 1421 INIT_LIST_HEAD(&ei->i_es_list); 1422 ei->i_es_all_nr = 0; 1423 ei->i_es_shk_nr = 0; 1424 ei->i_es_shrink_lblk = 0; 1425 ei->i_reserved_data_blocks = 0; 1426 spin_lock_init(&(ei->i_block_reservation_lock)); 1427 ext4_init_pending_tree(&ei->i_pending_tree); 1428 #ifdef CONFIG_QUOTA 1429 ei->i_reserved_quota = 0; 1430 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1431 #endif 1432 ei->jinode = NULL; 1433 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1434 spin_lock_init(&ei->i_completed_io_lock); 1435 ei->i_sync_tid = 0; 1436 ei->i_datasync_tid = 0; 1437 atomic_set(&ei->i_unwritten, 0); 1438 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1439 ext4_fc_init_inode(&ei->vfs_inode); 1440 mutex_init(&ei->i_fc_lock); 1441 return &ei->vfs_inode; 1442 } 1443 1444 static int ext4_drop_inode(struct inode *inode) 1445 { 1446 int drop = generic_drop_inode(inode); 1447 1448 if (!drop) 1449 drop = fscrypt_drop_inode(inode); 1450 1451 trace_ext4_drop_inode(inode, drop); 1452 return drop; 1453 } 1454 1455 static void ext4_free_in_core_inode(struct inode *inode) 1456 { 1457 fscrypt_free_inode(inode); 1458 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1459 pr_warn("%s: inode %ld still in fc list", 1460 __func__, inode->i_ino); 1461 } 1462 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1463 } 1464 1465 static void ext4_destroy_inode(struct inode *inode) 1466 { 1467 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1468 ext4_msg(inode->i_sb, KERN_ERR, 1469 "Inode %lu (%p): orphan list check failed!", 1470 inode->i_ino, EXT4_I(inode)); 1471 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1472 EXT4_I(inode), sizeof(struct ext4_inode_info), 1473 true); 1474 dump_stack(); 1475 } 1476 1477 if (EXT4_I(inode)->i_reserved_data_blocks) 1478 ext4_msg(inode->i_sb, KERN_ERR, 1479 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!", 1480 inode->i_ino, EXT4_I(inode), 1481 EXT4_I(inode)->i_reserved_data_blocks); 1482 } 1483 1484 static void ext4_shutdown(struct super_block *sb) 1485 { 1486 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH); 1487 } 1488 1489 static void init_once(void *foo) 1490 { 1491 struct ext4_inode_info *ei = foo; 1492 1493 INIT_LIST_HEAD(&ei->i_orphan); 1494 init_rwsem(&ei->xattr_sem); 1495 init_rwsem(&ei->i_data_sem); 1496 inode_init_once(&ei->vfs_inode); 1497 ext4_fc_init_inode(&ei->vfs_inode); 1498 } 1499 1500 static int __init init_inodecache(void) 1501 { 1502 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1503 sizeof(struct ext4_inode_info), 0, 1504 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1505 SLAB_ACCOUNT), 1506 offsetof(struct ext4_inode_info, i_data), 1507 sizeof_field(struct ext4_inode_info, i_data), 1508 init_once); 1509 if (ext4_inode_cachep == NULL) 1510 return -ENOMEM; 1511 return 0; 1512 } 1513 1514 static void destroy_inodecache(void) 1515 { 1516 /* 1517 * Make sure all delayed rcu free inodes are flushed before we 1518 * destroy cache. 1519 */ 1520 rcu_barrier(); 1521 kmem_cache_destroy(ext4_inode_cachep); 1522 } 1523 1524 void ext4_clear_inode(struct inode *inode) 1525 { 1526 ext4_fc_del(inode); 1527 invalidate_inode_buffers(inode); 1528 clear_inode(inode); 1529 ext4_discard_preallocations(inode, 0); 1530 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1531 dquot_drop(inode); 1532 if (EXT4_I(inode)->jinode) { 1533 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1534 EXT4_I(inode)->jinode); 1535 jbd2_free_inode(EXT4_I(inode)->jinode); 1536 EXT4_I(inode)->jinode = NULL; 1537 } 1538 fscrypt_put_encryption_info(inode); 1539 fsverity_cleanup_inode(inode); 1540 } 1541 1542 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1543 u64 ino, u32 generation) 1544 { 1545 struct inode *inode; 1546 1547 /* 1548 * Currently we don't know the generation for parent directory, so 1549 * a generation of 0 means "accept any" 1550 */ 1551 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1552 if (IS_ERR(inode)) 1553 return ERR_CAST(inode); 1554 if (generation && inode->i_generation != generation) { 1555 iput(inode); 1556 return ERR_PTR(-ESTALE); 1557 } 1558 1559 return inode; 1560 } 1561 1562 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1563 int fh_len, int fh_type) 1564 { 1565 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1566 ext4_nfs_get_inode); 1567 } 1568 1569 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1570 int fh_len, int fh_type) 1571 { 1572 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1573 ext4_nfs_get_inode); 1574 } 1575 1576 static int ext4_nfs_commit_metadata(struct inode *inode) 1577 { 1578 struct writeback_control wbc = { 1579 .sync_mode = WB_SYNC_ALL 1580 }; 1581 1582 trace_ext4_nfs_commit_metadata(inode); 1583 return ext4_write_inode(inode, &wbc); 1584 } 1585 1586 #ifdef CONFIG_QUOTA 1587 static const char * const quotatypes[] = INITQFNAMES; 1588 #define QTYPE2NAME(t) (quotatypes[t]) 1589 1590 static int ext4_write_dquot(struct dquot *dquot); 1591 static int ext4_acquire_dquot(struct dquot *dquot); 1592 static int ext4_release_dquot(struct dquot *dquot); 1593 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1594 static int ext4_write_info(struct super_block *sb, int type); 1595 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1596 const struct path *path); 1597 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1598 size_t len, loff_t off); 1599 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1600 const char *data, size_t len, loff_t off); 1601 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1602 unsigned int flags); 1603 1604 static struct dquot __rcu **ext4_get_dquots(struct inode *inode) 1605 { 1606 return EXT4_I(inode)->i_dquot; 1607 } 1608 1609 static const struct dquot_operations ext4_quota_operations = { 1610 .get_reserved_space = ext4_get_reserved_space, 1611 .write_dquot = ext4_write_dquot, 1612 .acquire_dquot = ext4_acquire_dquot, 1613 .release_dquot = ext4_release_dquot, 1614 .mark_dirty = ext4_mark_dquot_dirty, 1615 .write_info = ext4_write_info, 1616 .alloc_dquot = dquot_alloc, 1617 .destroy_dquot = dquot_destroy, 1618 .get_projid = ext4_get_projid, 1619 .get_inode_usage = ext4_get_inode_usage, 1620 .get_next_id = dquot_get_next_id, 1621 }; 1622 1623 static const struct quotactl_ops ext4_qctl_operations = { 1624 .quota_on = ext4_quota_on, 1625 .quota_off = ext4_quota_off, 1626 .quota_sync = dquot_quota_sync, 1627 .get_state = dquot_get_state, 1628 .set_info = dquot_set_dqinfo, 1629 .get_dqblk = dquot_get_dqblk, 1630 .set_dqblk = dquot_set_dqblk, 1631 .get_nextdqblk = dquot_get_next_dqblk, 1632 }; 1633 #endif 1634 1635 static const struct super_operations ext4_sops = { 1636 .alloc_inode = ext4_alloc_inode, 1637 .free_inode = ext4_free_in_core_inode, 1638 .destroy_inode = ext4_destroy_inode, 1639 .write_inode = ext4_write_inode, 1640 .dirty_inode = ext4_dirty_inode, 1641 .drop_inode = ext4_drop_inode, 1642 .evict_inode = ext4_evict_inode, 1643 .put_super = ext4_put_super, 1644 .sync_fs = ext4_sync_fs, 1645 .freeze_fs = ext4_freeze, 1646 .unfreeze_fs = ext4_unfreeze, 1647 .statfs = ext4_statfs, 1648 .show_options = ext4_show_options, 1649 .shutdown = ext4_shutdown, 1650 #ifdef CONFIG_QUOTA 1651 .quota_read = ext4_quota_read, 1652 .quota_write = ext4_quota_write, 1653 .get_dquots = ext4_get_dquots, 1654 #endif 1655 }; 1656 1657 static const struct export_operations ext4_export_ops = { 1658 .fh_to_dentry = ext4_fh_to_dentry, 1659 .fh_to_parent = ext4_fh_to_parent, 1660 .get_parent = ext4_get_parent, 1661 .commit_metadata = ext4_nfs_commit_metadata, 1662 }; 1663 1664 enum { 1665 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1666 Opt_resgid, Opt_resuid, Opt_sb, 1667 Opt_nouid32, Opt_debug, Opt_removed, 1668 Opt_user_xattr, Opt_acl, 1669 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1670 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1671 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1672 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1673 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1674 Opt_inlinecrypt, 1675 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1676 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1677 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1678 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1679 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1680 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1681 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1682 Opt_inode_readahead_blks, Opt_journal_ioprio, 1683 Opt_dioread_nolock, Opt_dioread_lock, 1684 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1685 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1686 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1687 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1688 #ifdef CONFIG_EXT4_DEBUG 1689 Opt_fc_debug_max_replay, Opt_fc_debug_force 1690 #endif 1691 }; 1692 1693 static const struct constant_table ext4_param_errors[] = { 1694 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1695 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1696 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1697 {} 1698 }; 1699 1700 static const struct constant_table ext4_param_data[] = { 1701 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1702 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1703 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1704 {} 1705 }; 1706 1707 static const struct constant_table ext4_param_data_err[] = { 1708 {"abort", Opt_data_err_abort}, 1709 {"ignore", Opt_data_err_ignore}, 1710 {} 1711 }; 1712 1713 static const struct constant_table ext4_param_jqfmt[] = { 1714 {"vfsold", QFMT_VFS_OLD}, 1715 {"vfsv0", QFMT_VFS_V0}, 1716 {"vfsv1", QFMT_VFS_V1}, 1717 {} 1718 }; 1719 1720 static const struct constant_table ext4_param_dax[] = { 1721 {"always", Opt_dax_always}, 1722 {"inode", Opt_dax_inode}, 1723 {"never", Opt_dax_never}, 1724 {} 1725 }; 1726 1727 /* String parameter that allows empty argument */ 1728 #define fsparam_string_empty(NAME, OPT) \ 1729 __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL) 1730 1731 /* 1732 * Mount option specification 1733 * We don't use fsparam_flag_no because of the way we set the 1734 * options and the way we show them in _ext4_show_options(). To 1735 * keep the changes to a minimum, let's keep the negative options 1736 * separate for now. 1737 */ 1738 static const struct fs_parameter_spec ext4_param_specs[] = { 1739 fsparam_flag ("bsddf", Opt_bsd_df), 1740 fsparam_flag ("minixdf", Opt_minix_df), 1741 fsparam_flag ("grpid", Opt_grpid), 1742 fsparam_flag ("bsdgroups", Opt_grpid), 1743 fsparam_flag ("nogrpid", Opt_nogrpid), 1744 fsparam_flag ("sysvgroups", Opt_nogrpid), 1745 fsparam_u32 ("resgid", Opt_resgid), 1746 fsparam_u32 ("resuid", Opt_resuid), 1747 fsparam_u32 ("sb", Opt_sb), 1748 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1749 fsparam_flag ("nouid32", Opt_nouid32), 1750 fsparam_flag ("debug", Opt_debug), 1751 fsparam_flag ("oldalloc", Opt_removed), 1752 fsparam_flag ("orlov", Opt_removed), 1753 fsparam_flag ("user_xattr", Opt_user_xattr), 1754 fsparam_flag ("acl", Opt_acl), 1755 fsparam_flag ("norecovery", Opt_noload), 1756 fsparam_flag ("noload", Opt_noload), 1757 fsparam_flag ("bh", Opt_removed), 1758 fsparam_flag ("nobh", Opt_removed), 1759 fsparam_u32 ("commit", Opt_commit), 1760 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1761 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1762 fsparam_u32 ("journal_dev", Opt_journal_dev), 1763 fsparam_bdev ("journal_path", Opt_journal_path), 1764 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1765 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1766 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1767 fsparam_flag ("abort", Opt_abort), 1768 fsparam_enum ("data", Opt_data, ext4_param_data), 1769 fsparam_enum ("data_err", Opt_data_err, 1770 ext4_param_data_err), 1771 fsparam_string_empty 1772 ("usrjquota", Opt_usrjquota), 1773 fsparam_string_empty 1774 ("grpjquota", Opt_grpjquota), 1775 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1776 fsparam_flag ("grpquota", Opt_grpquota), 1777 fsparam_flag ("quota", Opt_quota), 1778 fsparam_flag ("noquota", Opt_noquota), 1779 fsparam_flag ("usrquota", Opt_usrquota), 1780 fsparam_flag ("prjquota", Opt_prjquota), 1781 fsparam_flag ("barrier", Opt_barrier), 1782 fsparam_u32 ("barrier", Opt_barrier), 1783 fsparam_flag ("nobarrier", Opt_nobarrier), 1784 fsparam_flag ("i_version", Opt_removed), 1785 fsparam_flag ("dax", Opt_dax), 1786 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1787 fsparam_u32 ("stripe", Opt_stripe), 1788 fsparam_flag ("delalloc", Opt_delalloc), 1789 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1790 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1791 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1792 fsparam_u32 ("debug_want_extra_isize", 1793 Opt_debug_want_extra_isize), 1794 fsparam_flag ("mblk_io_submit", Opt_removed), 1795 fsparam_flag ("nomblk_io_submit", Opt_removed), 1796 fsparam_flag ("block_validity", Opt_block_validity), 1797 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1798 fsparam_u32 ("inode_readahead_blks", 1799 Opt_inode_readahead_blks), 1800 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1801 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1802 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1803 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1804 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1805 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1806 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1807 fsparam_flag ("discard", Opt_discard), 1808 fsparam_flag ("nodiscard", Opt_nodiscard), 1809 fsparam_u32 ("init_itable", Opt_init_itable), 1810 fsparam_flag ("init_itable", Opt_init_itable), 1811 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1812 #ifdef CONFIG_EXT4_DEBUG 1813 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1814 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1815 #endif 1816 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1817 fsparam_flag ("test_dummy_encryption", 1818 Opt_test_dummy_encryption), 1819 fsparam_string ("test_dummy_encryption", 1820 Opt_test_dummy_encryption), 1821 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1822 fsparam_flag ("nombcache", Opt_nombcache), 1823 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1824 fsparam_flag ("prefetch_block_bitmaps", 1825 Opt_removed), 1826 fsparam_flag ("no_prefetch_block_bitmaps", 1827 Opt_no_prefetch_block_bitmaps), 1828 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1829 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1830 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1831 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1832 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1833 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1834 {} 1835 }; 1836 1837 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1838 1839 #define MOPT_SET 0x0001 1840 #define MOPT_CLEAR 0x0002 1841 #define MOPT_NOSUPPORT 0x0004 1842 #define MOPT_EXPLICIT 0x0008 1843 #ifdef CONFIG_QUOTA 1844 #define MOPT_Q 0 1845 #define MOPT_QFMT 0x0010 1846 #else 1847 #define MOPT_Q MOPT_NOSUPPORT 1848 #define MOPT_QFMT MOPT_NOSUPPORT 1849 #endif 1850 #define MOPT_NO_EXT2 0x0020 1851 #define MOPT_NO_EXT3 0x0040 1852 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1853 #define MOPT_SKIP 0x0080 1854 #define MOPT_2 0x0100 1855 1856 static const struct mount_opts { 1857 int token; 1858 int mount_opt; 1859 int flags; 1860 } ext4_mount_opts[] = { 1861 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1862 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1863 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1864 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1865 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1866 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1867 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1868 MOPT_EXT4_ONLY | MOPT_SET}, 1869 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1870 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1871 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1872 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1873 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1874 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1875 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1876 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1877 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1878 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1879 {Opt_commit, 0, MOPT_NO_EXT2}, 1880 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1881 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1882 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1883 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1884 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1885 EXT4_MOUNT_JOURNAL_CHECKSUM), 1886 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1887 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1888 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1889 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1890 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1891 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1892 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1893 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1894 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1895 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1896 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1897 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1898 {Opt_data, 0, MOPT_NO_EXT2}, 1899 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1900 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1901 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1902 #else 1903 {Opt_acl, 0, MOPT_NOSUPPORT}, 1904 #endif 1905 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1906 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1907 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1908 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1909 MOPT_SET | MOPT_Q}, 1910 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1911 MOPT_SET | MOPT_Q}, 1912 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1913 MOPT_SET | MOPT_Q}, 1914 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1915 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1916 MOPT_CLEAR | MOPT_Q}, 1917 {Opt_usrjquota, 0, MOPT_Q}, 1918 {Opt_grpjquota, 0, MOPT_Q}, 1919 {Opt_jqfmt, 0, MOPT_QFMT}, 1920 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1921 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1922 MOPT_SET}, 1923 #ifdef CONFIG_EXT4_DEBUG 1924 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1925 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1926 #endif 1927 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2}, 1928 {Opt_err, 0, 0} 1929 }; 1930 1931 #if IS_ENABLED(CONFIG_UNICODE) 1932 static const struct ext4_sb_encodings { 1933 __u16 magic; 1934 char *name; 1935 unsigned int version; 1936 } ext4_sb_encoding_map[] = { 1937 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1938 }; 1939 1940 static const struct ext4_sb_encodings * 1941 ext4_sb_read_encoding(const struct ext4_super_block *es) 1942 { 1943 __u16 magic = le16_to_cpu(es->s_encoding); 1944 int i; 1945 1946 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1947 if (magic == ext4_sb_encoding_map[i].magic) 1948 return &ext4_sb_encoding_map[i]; 1949 1950 return NULL; 1951 } 1952 #endif 1953 1954 #define EXT4_SPEC_JQUOTA (1 << 0) 1955 #define EXT4_SPEC_JQFMT (1 << 1) 1956 #define EXT4_SPEC_DATAJ (1 << 2) 1957 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1958 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1959 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1960 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 1961 #define EXT4_SPEC_s_max_batch_time (1 << 8) 1962 #define EXT4_SPEC_s_min_batch_time (1 << 9) 1963 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 1964 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 1965 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 1966 #define EXT4_SPEC_s_stripe (1 << 13) 1967 #define EXT4_SPEC_s_resuid (1 << 14) 1968 #define EXT4_SPEC_s_resgid (1 << 15) 1969 #define EXT4_SPEC_s_commit_interval (1 << 16) 1970 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 1971 #define EXT4_SPEC_s_sb_block (1 << 18) 1972 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 1973 1974 struct ext4_fs_context { 1975 char *s_qf_names[EXT4_MAXQUOTAS]; 1976 struct fscrypt_dummy_policy dummy_enc_policy; 1977 int s_jquota_fmt; /* Format of quota to use */ 1978 #ifdef CONFIG_EXT4_DEBUG 1979 int s_fc_debug_max_replay; 1980 #endif 1981 unsigned short qname_spec; 1982 unsigned long vals_s_flags; /* Bits to set in s_flags */ 1983 unsigned long mask_s_flags; /* Bits changed in s_flags */ 1984 unsigned long journal_devnum; 1985 unsigned long s_commit_interval; 1986 unsigned long s_stripe; 1987 unsigned int s_inode_readahead_blks; 1988 unsigned int s_want_extra_isize; 1989 unsigned int s_li_wait_mult; 1990 unsigned int s_max_dir_size_kb; 1991 unsigned int journal_ioprio; 1992 unsigned int vals_s_mount_opt; 1993 unsigned int mask_s_mount_opt; 1994 unsigned int vals_s_mount_opt2; 1995 unsigned int mask_s_mount_opt2; 1996 unsigned int opt_flags; /* MOPT flags */ 1997 unsigned int spec; 1998 u32 s_max_batch_time; 1999 u32 s_min_batch_time; 2000 kuid_t s_resuid; 2001 kgid_t s_resgid; 2002 ext4_fsblk_t s_sb_block; 2003 }; 2004 2005 static void ext4_fc_free(struct fs_context *fc) 2006 { 2007 struct ext4_fs_context *ctx = fc->fs_private; 2008 int i; 2009 2010 if (!ctx) 2011 return; 2012 2013 for (i = 0; i < EXT4_MAXQUOTAS; i++) 2014 kfree(ctx->s_qf_names[i]); 2015 2016 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 2017 kfree(ctx); 2018 } 2019 2020 int ext4_init_fs_context(struct fs_context *fc) 2021 { 2022 struct ext4_fs_context *ctx; 2023 2024 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2025 if (!ctx) 2026 return -ENOMEM; 2027 2028 fc->fs_private = ctx; 2029 fc->ops = &ext4_context_ops; 2030 2031 return 0; 2032 } 2033 2034 #ifdef CONFIG_QUOTA 2035 /* 2036 * Note the name of the specified quota file. 2037 */ 2038 static int note_qf_name(struct fs_context *fc, int qtype, 2039 struct fs_parameter *param) 2040 { 2041 struct ext4_fs_context *ctx = fc->fs_private; 2042 char *qname; 2043 2044 if (param->size < 1) { 2045 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 2046 return -EINVAL; 2047 } 2048 if (strchr(param->string, '/')) { 2049 ext4_msg(NULL, KERN_ERR, 2050 "quotafile must be on filesystem root"); 2051 return -EINVAL; 2052 } 2053 if (ctx->s_qf_names[qtype]) { 2054 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 2055 ext4_msg(NULL, KERN_ERR, 2056 "%s quota file already specified", 2057 QTYPE2NAME(qtype)); 2058 return -EINVAL; 2059 } 2060 return 0; 2061 } 2062 2063 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 2064 if (!qname) { 2065 ext4_msg(NULL, KERN_ERR, 2066 "Not enough memory for storing quotafile name"); 2067 return -ENOMEM; 2068 } 2069 ctx->s_qf_names[qtype] = qname; 2070 ctx->qname_spec |= 1 << qtype; 2071 ctx->spec |= EXT4_SPEC_JQUOTA; 2072 return 0; 2073 } 2074 2075 /* 2076 * Clear the name of the specified quota file. 2077 */ 2078 static int unnote_qf_name(struct fs_context *fc, int qtype) 2079 { 2080 struct ext4_fs_context *ctx = fc->fs_private; 2081 2082 if (ctx->s_qf_names[qtype]) 2083 kfree(ctx->s_qf_names[qtype]); 2084 2085 ctx->s_qf_names[qtype] = NULL; 2086 ctx->qname_spec |= 1 << qtype; 2087 ctx->spec |= EXT4_SPEC_JQUOTA; 2088 return 0; 2089 } 2090 #endif 2091 2092 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2093 struct ext4_fs_context *ctx) 2094 { 2095 int err; 2096 2097 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2098 ext4_msg(NULL, KERN_WARNING, 2099 "test_dummy_encryption option not supported"); 2100 return -EINVAL; 2101 } 2102 err = fscrypt_parse_test_dummy_encryption(param, 2103 &ctx->dummy_enc_policy); 2104 if (err == -EINVAL) { 2105 ext4_msg(NULL, KERN_WARNING, 2106 "Value of option \"%s\" is unrecognized", param->key); 2107 } else if (err == -EEXIST) { 2108 ext4_msg(NULL, KERN_WARNING, 2109 "Conflicting test_dummy_encryption options"); 2110 return -EINVAL; 2111 } 2112 return err; 2113 } 2114 2115 #define EXT4_SET_CTX(name) \ 2116 static inline void ctx_set_##name(struct ext4_fs_context *ctx, \ 2117 unsigned long flag) \ 2118 { \ 2119 ctx->mask_s_##name |= flag; \ 2120 ctx->vals_s_##name |= flag; \ 2121 } 2122 2123 #define EXT4_CLEAR_CTX(name) \ 2124 static inline void ctx_clear_##name(struct ext4_fs_context *ctx, \ 2125 unsigned long flag) \ 2126 { \ 2127 ctx->mask_s_##name |= flag; \ 2128 ctx->vals_s_##name &= ~flag; \ 2129 } 2130 2131 #define EXT4_TEST_CTX(name) \ 2132 static inline unsigned long \ 2133 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2134 { \ 2135 return (ctx->vals_s_##name & flag); \ 2136 } 2137 2138 EXT4_SET_CTX(flags); /* set only */ 2139 EXT4_SET_CTX(mount_opt); 2140 EXT4_CLEAR_CTX(mount_opt); 2141 EXT4_TEST_CTX(mount_opt); 2142 EXT4_SET_CTX(mount_opt2); 2143 EXT4_CLEAR_CTX(mount_opt2); 2144 EXT4_TEST_CTX(mount_opt2); 2145 2146 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2147 { 2148 struct ext4_fs_context *ctx = fc->fs_private; 2149 struct fs_parse_result result; 2150 const struct mount_opts *m; 2151 int is_remount; 2152 kuid_t uid; 2153 kgid_t gid; 2154 int token; 2155 2156 token = fs_parse(fc, ext4_param_specs, param, &result); 2157 if (token < 0) 2158 return token; 2159 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2160 2161 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2162 if (token == m->token) 2163 break; 2164 2165 ctx->opt_flags |= m->flags; 2166 2167 if (m->flags & MOPT_EXPLICIT) { 2168 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2169 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2170 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2171 ctx_set_mount_opt2(ctx, 2172 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2173 } else 2174 return -EINVAL; 2175 } 2176 2177 if (m->flags & MOPT_NOSUPPORT) { 2178 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2179 param->key); 2180 return 0; 2181 } 2182 2183 switch (token) { 2184 #ifdef CONFIG_QUOTA 2185 case Opt_usrjquota: 2186 if (!*param->string) 2187 return unnote_qf_name(fc, USRQUOTA); 2188 else 2189 return note_qf_name(fc, USRQUOTA, param); 2190 case Opt_grpjquota: 2191 if (!*param->string) 2192 return unnote_qf_name(fc, GRPQUOTA); 2193 else 2194 return note_qf_name(fc, GRPQUOTA, param); 2195 #endif 2196 case Opt_sb: 2197 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2198 ext4_msg(NULL, KERN_WARNING, 2199 "Ignoring %s option on remount", param->key); 2200 } else { 2201 ctx->s_sb_block = result.uint_32; 2202 ctx->spec |= EXT4_SPEC_s_sb_block; 2203 } 2204 return 0; 2205 case Opt_removed: 2206 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2207 param->key); 2208 return 0; 2209 case Opt_inlinecrypt: 2210 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2211 ctx_set_flags(ctx, SB_INLINECRYPT); 2212 #else 2213 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2214 #endif 2215 return 0; 2216 case Opt_errors: 2217 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2218 ctx_set_mount_opt(ctx, result.uint_32); 2219 return 0; 2220 #ifdef CONFIG_QUOTA 2221 case Opt_jqfmt: 2222 ctx->s_jquota_fmt = result.uint_32; 2223 ctx->spec |= EXT4_SPEC_JQFMT; 2224 return 0; 2225 #endif 2226 case Opt_data: 2227 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2228 ctx_set_mount_opt(ctx, result.uint_32); 2229 ctx->spec |= EXT4_SPEC_DATAJ; 2230 return 0; 2231 case Opt_commit: 2232 if (result.uint_32 == 0) 2233 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE; 2234 else if (result.uint_32 > INT_MAX / HZ) { 2235 ext4_msg(NULL, KERN_ERR, 2236 "Invalid commit interval %d, " 2237 "must be smaller than %d", 2238 result.uint_32, INT_MAX / HZ); 2239 return -EINVAL; 2240 } 2241 ctx->s_commit_interval = HZ * result.uint_32; 2242 ctx->spec |= EXT4_SPEC_s_commit_interval; 2243 return 0; 2244 case Opt_debug_want_extra_isize: 2245 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2246 ext4_msg(NULL, KERN_ERR, 2247 "Invalid want_extra_isize %d", result.uint_32); 2248 return -EINVAL; 2249 } 2250 ctx->s_want_extra_isize = result.uint_32; 2251 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2252 return 0; 2253 case Opt_max_batch_time: 2254 ctx->s_max_batch_time = result.uint_32; 2255 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2256 return 0; 2257 case Opt_min_batch_time: 2258 ctx->s_min_batch_time = result.uint_32; 2259 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2260 return 0; 2261 case Opt_inode_readahead_blks: 2262 if (result.uint_32 && 2263 (result.uint_32 > (1 << 30) || 2264 !is_power_of_2(result.uint_32))) { 2265 ext4_msg(NULL, KERN_ERR, 2266 "EXT4-fs: inode_readahead_blks must be " 2267 "0 or a power of 2 smaller than 2^31"); 2268 return -EINVAL; 2269 } 2270 ctx->s_inode_readahead_blks = result.uint_32; 2271 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2272 return 0; 2273 case Opt_init_itable: 2274 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2275 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2276 if (param->type == fs_value_is_string) 2277 ctx->s_li_wait_mult = result.uint_32; 2278 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2279 return 0; 2280 case Opt_max_dir_size_kb: 2281 ctx->s_max_dir_size_kb = result.uint_32; 2282 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2283 return 0; 2284 #ifdef CONFIG_EXT4_DEBUG 2285 case Opt_fc_debug_max_replay: 2286 ctx->s_fc_debug_max_replay = result.uint_32; 2287 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2288 return 0; 2289 #endif 2290 case Opt_stripe: 2291 ctx->s_stripe = result.uint_32; 2292 ctx->spec |= EXT4_SPEC_s_stripe; 2293 return 0; 2294 case Opt_resuid: 2295 uid = make_kuid(current_user_ns(), result.uint_32); 2296 if (!uid_valid(uid)) { 2297 ext4_msg(NULL, KERN_ERR, "Invalid uid value %d", 2298 result.uint_32); 2299 return -EINVAL; 2300 } 2301 ctx->s_resuid = uid; 2302 ctx->spec |= EXT4_SPEC_s_resuid; 2303 return 0; 2304 case Opt_resgid: 2305 gid = make_kgid(current_user_ns(), result.uint_32); 2306 if (!gid_valid(gid)) { 2307 ext4_msg(NULL, KERN_ERR, "Invalid gid value %d", 2308 result.uint_32); 2309 return -EINVAL; 2310 } 2311 ctx->s_resgid = gid; 2312 ctx->spec |= EXT4_SPEC_s_resgid; 2313 return 0; 2314 case Opt_journal_dev: 2315 if (is_remount) { 2316 ext4_msg(NULL, KERN_ERR, 2317 "Cannot specify journal on remount"); 2318 return -EINVAL; 2319 } 2320 ctx->journal_devnum = result.uint_32; 2321 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2322 return 0; 2323 case Opt_journal_path: 2324 { 2325 struct inode *journal_inode; 2326 struct path path; 2327 int error; 2328 2329 if (is_remount) { 2330 ext4_msg(NULL, KERN_ERR, 2331 "Cannot specify journal on remount"); 2332 return -EINVAL; 2333 } 2334 2335 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path); 2336 if (error) { 2337 ext4_msg(NULL, KERN_ERR, "error: could not find " 2338 "journal device path"); 2339 return -EINVAL; 2340 } 2341 2342 journal_inode = d_inode(path.dentry); 2343 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2344 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2345 path_put(&path); 2346 return 0; 2347 } 2348 case Opt_journal_ioprio: 2349 if (result.uint_32 > 7) { 2350 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2351 " (must be 0-7)"); 2352 return -EINVAL; 2353 } 2354 ctx->journal_ioprio = 2355 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2356 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2357 return 0; 2358 case Opt_test_dummy_encryption: 2359 return ext4_parse_test_dummy_encryption(param, ctx); 2360 case Opt_dax: 2361 case Opt_dax_type: 2362 #ifdef CONFIG_FS_DAX 2363 { 2364 int type = (token == Opt_dax) ? 2365 Opt_dax : result.uint_32; 2366 2367 switch (type) { 2368 case Opt_dax: 2369 case Opt_dax_always: 2370 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2371 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2372 break; 2373 case Opt_dax_never: 2374 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2375 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2376 break; 2377 case Opt_dax_inode: 2378 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2379 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2380 /* Strictly for printing options */ 2381 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2382 break; 2383 } 2384 return 0; 2385 } 2386 #else 2387 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2388 return -EINVAL; 2389 #endif 2390 case Opt_data_err: 2391 if (result.uint_32 == Opt_data_err_abort) 2392 ctx_set_mount_opt(ctx, m->mount_opt); 2393 else if (result.uint_32 == Opt_data_err_ignore) 2394 ctx_clear_mount_opt(ctx, m->mount_opt); 2395 return 0; 2396 case Opt_mb_optimize_scan: 2397 if (result.int_32 == 1) { 2398 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2399 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2400 } else if (result.int_32 == 0) { 2401 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2402 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2403 } else { 2404 ext4_msg(NULL, KERN_WARNING, 2405 "mb_optimize_scan should be set to 0 or 1."); 2406 return -EINVAL; 2407 } 2408 return 0; 2409 } 2410 2411 /* 2412 * At this point we should only be getting options requiring MOPT_SET, 2413 * or MOPT_CLEAR. Anything else is a bug 2414 */ 2415 if (m->token == Opt_err) { 2416 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2417 param->key); 2418 WARN_ON(1); 2419 return -EINVAL; 2420 } 2421 2422 else { 2423 unsigned int set = 0; 2424 2425 if ((param->type == fs_value_is_flag) || 2426 result.uint_32 > 0) 2427 set = 1; 2428 2429 if (m->flags & MOPT_CLEAR) 2430 set = !set; 2431 else if (unlikely(!(m->flags & MOPT_SET))) { 2432 ext4_msg(NULL, KERN_WARNING, 2433 "buggy handling of option %s", 2434 param->key); 2435 WARN_ON(1); 2436 return -EINVAL; 2437 } 2438 if (m->flags & MOPT_2) { 2439 if (set != 0) 2440 ctx_set_mount_opt2(ctx, m->mount_opt); 2441 else 2442 ctx_clear_mount_opt2(ctx, m->mount_opt); 2443 } else { 2444 if (set != 0) 2445 ctx_set_mount_opt(ctx, m->mount_opt); 2446 else 2447 ctx_clear_mount_opt(ctx, m->mount_opt); 2448 } 2449 } 2450 2451 return 0; 2452 } 2453 2454 static int parse_options(struct fs_context *fc, char *options) 2455 { 2456 struct fs_parameter param; 2457 int ret; 2458 char *key; 2459 2460 if (!options) 2461 return 0; 2462 2463 while ((key = strsep(&options, ",")) != NULL) { 2464 if (*key) { 2465 size_t v_len = 0; 2466 char *value = strchr(key, '='); 2467 2468 param.type = fs_value_is_flag; 2469 param.string = NULL; 2470 2471 if (value) { 2472 if (value == key) 2473 continue; 2474 2475 *value++ = 0; 2476 v_len = strlen(value); 2477 param.string = kmemdup_nul(value, v_len, 2478 GFP_KERNEL); 2479 if (!param.string) 2480 return -ENOMEM; 2481 param.type = fs_value_is_string; 2482 } 2483 2484 param.key = key; 2485 param.size = v_len; 2486 2487 ret = ext4_parse_param(fc, ¶m); 2488 if (param.string) 2489 kfree(param.string); 2490 if (ret < 0) 2491 return ret; 2492 } 2493 } 2494 2495 ret = ext4_validate_options(fc); 2496 if (ret < 0) 2497 return ret; 2498 2499 return 0; 2500 } 2501 2502 static int parse_apply_sb_mount_options(struct super_block *sb, 2503 struct ext4_fs_context *m_ctx) 2504 { 2505 struct ext4_sb_info *sbi = EXT4_SB(sb); 2506 char *s_mount_opts = NULL; 2507 struct ext4_fs_context *s_ctx = NULL; 2508 struct fs_context *fc = NULL; 2509 int ret = -ENOMEM; 2510 2511 if (!sbi->s_es->s_mount_opts[0]) 2512 return 0; 2513 2514 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2515 sizeof(sbi->s_es->s_mount_opts), 2516 GFP_KERNEL); 2517 if (!s_mount_opts) 2518 return ret; 2519 2520 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2521 if (!fc) 2522 goto out_free; 2523 2524 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2525 if (!s_ctx) 2526 goto out_free; 2527 2528 fc->fs_private = s_ctx; 2529 fc->s_fs_info = sbi; 2530 2531 ret = parse_options(fc, s_mount_opts); 2532 if (ret < 0) 2533 goto parse_failed; 2534 2535 ret = ext4_check_opt_consistency(fc, sb); 2536 if (ret < 0) { 2537 parse_failed: 2538 ext4_msg(sb, KERN_WARNING, 2539 "failed to parse options in superblock: %s", 2540 s_mount_opts); 2541 ret = 0; 2542 goto out_free; 2543 } 2544 2545 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2546 m_ctx->journal_devnum = s_ctx->journal_devnum; 2547 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2548 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2549 2550 ext4_apply_options(fc, sb); 2551 ret = 0; 2552 2553 out_free: 2554 if (fc) { 2555 ext4_fc_free(fc); 2556 kfree(fc); 2557 } 2558 kfree(s_mount_opts); 2559 return ret; 2560 } 2561 2562 static void ext4_apply_quota_options(struct fs_context *fc, 2563 struct super_block *sb) 2564 { 2565 #ifdef CONFIG_QUOTA 2566 bool quota_feature = ext4_has_feature_quota(sb); 2567 struct ext4_fs_context *ctx = fc->fs_private; 2568 struct ext4_sb_info *sbi = EXT4_SB(sb); 2569 char *qname; 2570 int i; 2571 2572 if (quota_feature) 2573 return; 2574 2575 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2576 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2577 if (!(ctx->qname_spec & (1 << i))) 2578 continue; 2579 2580 qname = ctx->s_qf_names[i]; /* May be NULL */ 2581 if (qname) 2582 set_opt(sb, QUOTA); 2583 ctx->s_qf_names[i] = NULL; 2584 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2585 lockdep_is_held(&sb->s_umount)); 2586 if (qname) 2587 kfree_rcu_mightsleep(qname); 2588 } 2589 } 2590 2591 if (ctx->spec & EXT4_SPEC_JQFMT) 2592 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2593 #endif 2594 } 2595 2596 /* 2597 * Check quota settings consistency. 2598 */ 2599 static int ext4_check_quota_consistency(struct fs_context *fc, 2600 struct super_block *sb) 2601 { 2602 #ifdef CONFIG_QUOTA 2603 struct ext4_fs_context *ctx = fc->fs_private; 2604 struct ext4_sb_info *sbi = EXT4_SB(sb); 2605 bool quota_feature = ext4_has_feature_quota(sb); 2606 bool quota_loaded = sb_any_quota_loaded(sb); 2607 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2608 int quota_flags, i; 2609 2610 /* 2611 * We do the test below only for project quotas. 'usrquota' and 2612 * 'grpquota' mount options are allowed even without quota feature 2613 * to support legacy quotas in quota files. 2614 */ 2615 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2616 !ext4_has_feature_project(sb)) { 2617 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2618 "Cannot enable project quota enforcement."); 2619 return -EINVAL; 2620 } 2621 2622 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2623 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2624 if (quota_loaded && 2625 ctx->mask_s_mount_opt & quota_flags && 2626 !ctx_test_mount_opt(ctx, quota_flags)) 2627 goto err_quota_change; 2628 2629 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2630 2631 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2632 if (!(ctx->qname_spec & (1 << i))) 2633 continue; 2634 2635 if (quota_loaded && 2636 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2637 goto err_jquota_change; 2638 2639 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2640 strcmp(get_qf_name(sb, sbi, i), 2641 ctx->s_qf_names[i]) != 0) 2642 goto err_jquota_specified; 2643 } 2644 2645 if (quota_feature) { 2646 ext4_msg(NULL, KERN_INFO, 2647 "Journaled quota options ignored when " 2648 "QUOTA feature is enabled"); 2649 return 0; 2650 } 2651 } 2652 2653 if (ctx->spec & EXT4_SPEC_JQFMT) { 2654 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2655 goto err_jquota_change; 2656 if (quota_feature) { 2657 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2658 "ignored when QUOTA feature is enabled"); 2659 return 0; 2660 } 2661 } 2662 2663 /* Make sure we don't mix old and new quota format */ 2664 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2665 ctx->s_qf_names[USRQUOTA]); 2666 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2667 ctx->s_qf_names[GRPQUOTA]); 2668 2669 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2670 test_opt(sb, USRQUOTA)); 2671 2672 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2673 test_opt(sb, GRPQUOTA)); 2674 2675 if (usr_qf_name) { 2676 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2677 usrquota = false; 2678 } 2679 if (grp_qf_name) { 2680 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2681 grpquota = false; 2682 } 2683 2684 if (usr_qf_name || grp_qf_name) { 2685 if (usrquota || grpquota) { 2686 ext4_msg(NULL, KERN_ERR, "old and new quota " 2687 "format mixing"); 2688 return -EINVAL; 2689 } 2690 2691 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2692 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2693 "not specified"); 2694 return -EINVAL; 2695 } 2696 } 2697 2698 return 0; 2699 2700 err_quota_change: 2701 ext4_msg(NULL, KERN_ERR, 2702 "Cannot change quota options when quota turned on"); 2703 return -EINVAL; 2704 err_jquota_change: 2705 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2706 "options when quota turned on"); 2707 return -EINVAL; 2708 err_jquota_specified: 2709 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2710 QTYPE2NAME(i)); 2711 return -EINVAL; 2712 #else 2713 return 0; 2714 #endif 2715 } 2716 2717 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2718 struct super_block *sb) 2719 { 2720 const struct ext4_fs_context *ctx = fc->fs_private; 2721 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2722 2723 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2724 return 0; 2725 2726 if (!ext4_has_feature_encrypt(sb)) { 2727 ext4_msg(NULL, KERN_WARNING, 2728 "test_dummy_encryption requires encrypt feature"); 2729 return -EINVAL; 2730 } 2731 /* 2732 * This mount option is just for testing, and it's not worthwhile to 2733 * implement the extra complexity (e.g. RCU protection) that would be 2734 * needed to allow it to be set or changed during remount. We do allow 2735 * it to be specified during remount, but only if there is no change. 2736 */ 2737 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 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 "Can't set or change test_dummy_encryption on remount"); 2743 return -EINVAL; 2744 } 2745 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2746 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2747 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2748 &ctx->dummy_enc_policy)) 2749 return 0; 2750 ext4_msg(NULL, KERN_WARNING, 2751 "Conflicting test_dummy_encryption options"); 2752 return -EINVAL; 2753 } 2754 return 0; 2755 } 2756 2757 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2758 struct super_block *sb) 2759 { 2760 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2761 /* if already set, it was already verified to be the same */ 2762 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2763 return; 2764 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2765 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2766 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2767 } 2768 2769 static int ext4_check_opt_consistency(struct fs_context *fc, 2770 struct super_block *sb) 2771 { 2772 struct ext4_fs_context *ctx = fc->fs_private; 2773 struct ext4_sb_info *sbi = fc->s_fs_info; 2774 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2775 int err; 2776 2777 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2778 ext4_msg(NULL, KERN_ERR, 2779 "Mount option(s) incompatible with ext2"); 2780 return -EINVAL; 2781 } 2782 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2783 ext4_msg(NULL, KERN_ERR, 2784 "Mount option(s) incompatible with ext3"); 2785 return -EINVAL; 2786 } 2787 2788 if (ctx->s_want_extra_isize > 2789 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2790 ext4_msg(NULL, KERN_ERR, 2791 "Invalid want_extra_isize %d", 2792 ctx->s_want_extra_isize); 2793 return -EINVAL; 2794 } 2795 2796 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) { 2797 int blocksize = 2798 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2799 if (blocksize < PAGE_SIZE) 2800 ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an " 2801 "experimental mount option 'dioread_nolock' " 2802 "for blocksize < PAGE_SIZE"); 2803 } 2804 2805 err = ext4_check_test_dummy_encryption(fc, sb); 2806 if (err) 2807 return err; 2808 2809 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2810 if (!sbi->s_journal) { 2811 ext4_msg(NULL, KERN_WARNING, 2812 "Remounting file system with no journal " 2813 "so ignoring journalled data option"); 2814 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2815 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2816 test_opt(sb, DATA_FLAGS)) { 2817 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2818 "on remount"); 2819 return -EINVAL; 2820 } 2821 } 2822 2823 if (is_remount) { 2824 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2825 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2826 ext4_msg(NULL, KERN_ERR, "can't mount with " 2827 "both data=journal and dax"); 2828 return -EINVAL; 2829 } 2830 2831 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2832 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2833 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2834 fail_dax_change_remount: 2835 ext4_msg(NULL, KERN_ERR, "can't change " 2836 "dax mount option while remounting"); 2837 return -EINVAL; 2838 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2839 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2840 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2841 goto fail_dax_change_remount; 2842 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2843 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2844 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2845 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2846 goto fail_dax_change_remount; 2847 } 2848 } 2849 2850 return ext4_check_quota_consistency(fc, sb); 2851 } 2852 2853 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2854 { 2855 struct ext4_fs_context *ctx = fc->fs_private; 2856 struct ext4_sb_info *sbi = fc->s_fs_info; 2857 2858 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2859 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2860 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2861 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2862 sb->s_flags &= ~ctx->mask_s_flags; 2863 sb->s_flags |= ctx->vals_s_flags; 2864 2865 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2866 APPLY(s_commit_interval); 2867 APPLY(s_stripe); 2868 APPLY(s_max_batch_time); 2869 APPLY(s_min_batch_time); 2870 APPLY(s_want_extra_isize); 2871 APPLY(s_inode_readahead_blks); 2872 APPLY(s_max_dir_size_kb); 2873 APPLY(s_li_wait_mult); 2874 APPLY(s_resgid); 2875 APPLY(s_resuid); 2876 2877 #ifdef CONFIG_EXT4_DEBUG 2878 APPLY(s_fc_debug_max_replay); 2879 #endif 2880 2881 ext4_apply_quota_options(fc, sb); 2882 ext4_apply_test_dummy_encryption(ctx, sb); 2883 } 2884 2885 2886 static int ext4_validate_options(struct fs_context *fc) 2887 { 2888 #ifdef CONFIG_QUOTA 2889 struct ext4_fs_context *ctx = fc->fs_private; 2890 char *usr_qf_name, *grp_qf_name; 2891 2892 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2893 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2894 2895 if (usr_qf_name || grp_qf_name) { 2896 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2897 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2898 2899 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2900 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2901 2902 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2903 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2904 ext4_msg(NULL, KERN_ERR, "old and new quota " 2905 "format mixing"); 2906 return -EINVAL; 2907 } 2908 } 2909 #endif 2910 return 1; 2911 } 2912 2913 static inline void ext4_show_quota_options(struct seq_file *seq, 2914 struct super_block *sb) 2915 { 2916 #if defined(CONFIG_QUOTA) 2917 struct ext4_sb_info *sbi = EXT4_SB(sb); 2918 char *usr_qf_name, *grp_qf_name; 2919 2920 if (sbi->s_jquota_fmt) { 2921 char *fmtname = ""; 2922 2923 switch (sbi->s_jquota_fmt) { 2924 case QFMT_VFS_OLD: 2925 fmtname = "vfsold"; 2926 break; 2927 case QFMT_VFS_V0: 2928 fmtname = "vfsv0"; 2929 break; 2930 case QFMT_VFS_V1: 2931 fmtname = "vfsv1"; 2932 break; 2933 } 2934 seq_printf(seq, ",jqfmt=%s", fmtname); 2935 } 2936 2937 rcu_read_lock(); 2938 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2939 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2940 if (usr_qf_name) 2941 seq_show_option(seq, "usrjquota", usr_qf_name); 2942 if (grp_qf_name) 2943 seq_show_option(seq, "grpjquota", grp_qf_name); 2944 rcu_read_unlock(); 2945 #endif 2946 } 2947 2948 static const char *token2str(int token) 2949 { 2950 const struct fs_parameter_spec *spec; 2951 2952 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2953 if (spec->opt == token && !spec->type) 2954 break; 2955 return spec->name; 2956 } 2957 2958 /* 2959 * Show an option if 2960 * - it's set to a non-default value OR 2961 * - if the per-sb default is different from the global default 2962 */ 2963 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2964 int nodefs) 2965 { 2966 struct ext4_sb_info *sbi = EXT4_SB(sb); 2967 struct ext4_super_block *es = sbi->s_es; 2968 int def_errors; 2969 const struct mount_opts *m; 2970 char sep = nodefs ? '\n' : ','; 2971 2972 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2973 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2974 2975 if (sbi->s_sb_block != 1) 2976 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2977 2978 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2979 int want_set = m->flags & MOPT_SET; 2980 int opt_2 = m->flags & MOPT_2; 2981 unsigned int mount_opt, def_mount_opt; 2982 2983 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2984 m->flags & MOPT_SKIP) 2985 continue; 2986 2987 if (opt_2) { 2988 mount_opt = sbi->s_mount_opt2; 2989 def_mount_opt = sbi->s_def_mount_opt2; 2990 } else { 2991 mount_opt = sbi->s_mount_opt; 2992 def_mount_opt = sbi->s_def_mount_opt; 2993 } 2994 /* skip if same as the default */ 2995 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt))) 2996 continue; 2997 /* select Opt_noFoo vs Opt_Foo */ 2998 if ((want_set && 2999 (mount_opt & m->mount_opt) != m->mount_opt) || 3000 (!want_set && (mount_opt & m->mount_opt))) 3001 continue; 3002 SEQ_OPTS_PRINT("%s", token2str(m->token)); 3003 } 3004 3005 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 3006 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 3007 SEQ_OPTS_PRINT("resuid=%u", 3008 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 3009 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 3010 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 3011 SEQ_OPTS_PRINT("resgid=%u", 3012 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 3013 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 3014 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 3015 SEQ_OPTS_PUTS("errors=remount-ro"); 3016 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 3017 SEQ_OPTS_PUTS("errors=continue"); 3018 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 3019 SEQ_OPTS_PUTS("errors=panic"); 3020 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 3021 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 3022 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 3023 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 3024 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 3025 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 3026 if (nodefs || sbi->s_stripe) 3027 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 3028 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 3029 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 3030 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 3031 SEQ_OPTS_PUTS("data=journal"); 3032 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 3033 SEQ_OPTS_PUTS("data=ordered"); 3034 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 3035 SEQ_OPTS_PUTS("data=writeback"); 3036 } 3037 if (nodefs || 3038 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 3039 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 3040 sbi->s_inode_readahead_blks); 3041 3042 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 3043 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 3044 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 3045 if (nodefs || sbi->s_max_dir_size_kb) 3046 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 3047 if (test_opt(sb, DATA_ERR_ABORT)) 3048 SEQ_OPTS_PUTS("data_err=abort"); 3049 3050 fscrypt_show_test_dummy_encryption(seq, sep, sb); 3051 3052 if (sb->s_flags & SB_INLINECRYPT) 3053 SEQ_OPTS_PUTS("inlinecrypt"); 3054 3055 if (test_opt(sb, DAX_ALWAYS)) { 3056 if (IS_EXT2_SB(sb)) 3057 SEQ_OPTS_PUTS("dax"); 3058 else 3059 SEQ_OPTS_PUTS("dax=always"); 3060 } else if (test_opt2(sb, DAX_NEVER)) { 3061 SEQ_OPTS_PUTS("dax=never"); 3062 } else if (test_opt2(sb, DAX_INODE)) { 3063 SEQ_OPTS_PUTS("dax=inode"); 3064 } 3065 3066 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3067 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3068 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 3069 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3070 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3071 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 3072 } 3073 3074 ext4_show_quota_options(seq, sb); 3075 return 0; 3076 } 3077 3078 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 3079 { 3080 return _ext4_show_options(seq, root->d_sb, 0); 3081 } 3082 3083 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3084 { 3085 struct super_block *sb = seq->private; 3086 int rc; 3087 3088 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3089 rc = _ext4_show_options(seq, sb, 1); 3090 seq_puts(seq, "\n"); 3091 return rc; 3092 } 3093 3094 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3095 int read_only) 3096 { 3097 struct ext4_sb_info *sbi = EXT4_SB(sb); 3098 int err = 0; 3099 3100 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3101 ext4_msg(sb, KERN_ERR, "revision level too high, " 3102 "forcing read-only mode"); 3103 err = -EROFS; 3104 goto done; 3105 } 3106 if (read_only) 3107 goto done; 3108 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3109 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3110 "running e2fsck is recommended"); 3111 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3112 ext4_msg(sb, KERN_WARNING, 3113 "warning: mounting fs with errors, " 3114 "running e2fsck is recommended"); 3115 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3116 le16_to_cpu(es->s_mnt_count) >= 3117 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3118 ext4_msg(sb, KERN_WARNING, 3119 "warning: maximal mount count reached, " 3120 "running e2fsck is recommended"); 3121 else if (le32_to_cpu(es->s_checkinterval) && 3122 (ext4_get_tstamp(es, s_lastcheck) + 3123 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3124 ext4_msg(sb, KERN_WARNING, 3125 "warning: checktime reached, " 3126 "running e2fsck is recommended"); 3127 if (!sbi->s_journal) 3128 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3129 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3130 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3131 le16_add_cpu(&es->s_mnt_count, 1); 3132 ext4_update_tstamp(es, s_mtime); 3133 if (sbi->s_journal) { 3134 ext4_set_feature_journal_needs_recovery(sb); 3135 if (ext4_has_feature_orphan_file(sb)) 3136 ext4_set_feature_orphan_present(sb); 3137 } 3138 3139 err = ext4_commit_super(sb); 3140 done: 3141 if (test_opt(sb, DEBUG)) 3142 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3143 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3144 sb->s_blocksize, 3145 sbi->s_groups_count, 3146 EXT4_BLOCKS_PER_GROUP(sb), 3147 EXT4_INODES_PER_GROUP(sb), 3148 sbi->s_mount_opt, sbi->s_mount_opt2); 3149 return err; 3150 } 3151 3152 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3153 { 3154 struct ext4_sb_info *sbi = EXT4_SB(sb); 3155 struct flex_groups **old_groups, **new_groups; 3156 int size, i, j; 3157 3158 if (!sbi->s_log_groups_per_flex) 3159 return 0; 3160 3161 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3162 if (size <= sbi->s_flex_groups_allocated) 3163 return 0; 3164 3165 new_groups = kvzalloc(roundup_pow_of_two(size * 3166 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3167 if (!new_groups) { 3168 ext4_msg(sb, KERN_ERR, 3169 "not enough memory for %d flex group pointers", size); 3170 return -ENOMEM; 3171 } 3172 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3173 new_groups[i] = kvzalloc(roundup_pow_of_two( 3174 sizeof(struct flex_groups)), 3175 GFP_KERNEL); 3176 if (!new_groups[i]) { 3177 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3178 kvfree(new_groups[j]); 3179 kvfree(new_groups); 3180 ext4_msg(sb, KERN_ERR, 3181 "not enough memory for %d flex groups", size); 3182 return -ENOMEM; 3183 } 3184 } 3185 rcu_read_lock(); 3186 old_groups = rcu_dereference(sbi->s_flex_groups); 3187 if (old_groups) 3188 memcpy(new_groups, old_groups, 3189 (sbi->s_flex_groups_allocated * 3190 sizeof(struct flex_groups *))); 3191 rcu_read_unlock(); 3192 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3193 sbi->s_flex_groups_allocated = size; 3194 if (old_groups) 3195 ext4_kvfree_array_rcu(old_groups); 3196 return 0; 3197 } 3198 3199 static int ext4_fill_flex_info(struct super_block *sb) 3200 { 3201 struct ext4_sb_info *sbi = EXT4_SB(sb); 3202 struct ext4_group_desc *gdp = NULL; 3203 struct flex_groups *fg; 3204 ext4_group_t flex_group; 3205 int i, err; 3206 3207 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3208 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3209 sbi->s_log_groups_per_flex = 0; 3210 return 1; 3211 } 3212 3213 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3214 if (err) 3215 goto failed; 3216 3217 for (i = 0; i < sbi->s_groups_count; i++) { 3218 gdp = ext4_get_group_desc(sb, i, NULL); 3219 3220 flex_group = ext4_flex_group(sbi, i); 3221 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3222 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3223 atomic64_add(ext4_free_group_clusters(sb, gdp), 3224 &fg->free_clusters); 3225 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3226 } 3227 3228 return 1; 3229 failed: 3230 return 0; 3231 } 3232 3233 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3234 struct ext4_group_desc *gdp) 3235 { 3236 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3237 __u16 crc = 0; 3238 __le32 le_group = cpu_to_le32(block_group); 3239 struct ext4_sb_info *sbi = EXT4_SB(sb); 3240 3241 if (ext4_has_metadata_csum(sbi->s_sb)) { 3242 /* Use new metadata_csum algorithm */ 3243 __u32 csum32; 3244 __u16 dummy_csum = 0; 3245 3246 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 3247 sizeof(le_group)); 3248 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 3249 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 3250 sizeof(dummy_csum)); 3251 offset += sizeof(dummy_csum); 3252 if (offset < sbi->s_desc_size) 3253 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 3254 sbi->s_desc_size - offset); 3255 3256 crc = csum32 & 0xFFFF; 3257 goto out; 3258 } 3259 3260 /* old crc16 code */ 3261 if (!ext4_has_feature_gdt_csum(sb)) 3262 return 0; 3263 3264 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3265 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3266 crc = crc16(crc, (__u8 *)gdp, offset); 3267 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3268 /* for checksum of struct ext4_group_desc do the rest...*/ 3269 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size) 3270 crc = crc16(crc, (__u8 *)gdp + offset, 3271 sbi->s_desc_size - offset); 3272 3273 out: 3274 return cpu_to_le16(crc); 3275 } 3276 3277 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3278 struct ext4_group_desc *gdp) 3279 { 3280 if (ext4_has_group_desc_csum(sb) && 3281 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3282 return 0; 3283 3284 return 1; 3285 } 3286 3287 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3288 struct ext4_group_desc *gdp) 3289 { 3290 if (!ext4_has_group_desc_csum(sb)) 3291 return; 3292 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3293 } 3294 3295 /* Called at mount-time, super-block is locked */ 3296 static int ext4_check_descriptors(struct super_block *sb, 3297 ext4_fsblk_t sb_block, 3298 ext4_group_t *first_not_zeroed) 3299 { 3300 struct ext4_sb_info *sbi = EXT4_SB(sb); 3301 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3302 ext4_fsblk_t last_block; 3303 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3304 ext4_fsblk_t block_bitmap; 3305 ext4_fsblk_t inode_bitmap; 3306 ext4_fsblk_t inode_table; 3307 int flexbg_flag = 0; 3308 ext4_group_t i, grp = sbi->s_groups_count; 3309 3310 if (ext4_has_feature_flex_bg(sb)) 3311 flexbg_flag = 1; 3312 3313 ext4_debug("Checking group descriptors"); 3314 3315 for (i = 0; i < sbi->s_groups_count; i++) { 3316 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3317 3318 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3319 last_block = ext4_blocks_count(sbi->s_es) - 1; 3320 else 3321 last_block = first_block + 3322 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3323 3324 if ((grp == sbi->s_groups_count) && 3325 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3326 grp = i; 3327 3328 block_bitmap = ext4_block_bitmap(sb, gdp); 3329 if (block_bitmap == sb_block) { 3330 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3331 "Block bitmap for group %u overlaps " 3332 "superblock", i); 3333 if (!sb_rdonly(sb)) 3334 return 0; 3335 } 3336 if (block_bitmap >= sb_block + 1 && 3337 block_bitmap <= last_bg_block) { 3338 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3339 "Block bitmap for group %u overlaps " 3340 "block group descriptors", i); 3341 if (!sb_rdonly(sb)) 3342 return 0; 3343 } 3344 if (block_bitmap < first_block || block_bitmap > last_block) { 3345 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3346 "Block bitmap for group %u not in group " 3347 "(block %llu)!", i, block_bitmap); 3348 return 0; 3349 } 3350 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3351 if (inode_bitmap == sb_block) { 3352 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3353 "Inode bitmap for group %u overlaps " 3354 "superblock", i); 3355 if (!sb_rdonly(sb)) 3356 return 0; 3357 } 3358 if (inode_bitmap >= sb_block + 1 && 3359 inode_bitmap <= last_bg_block) { 3360 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3361 "Inode bitmap for group %u overlaps " 3362 "block group descriptors", i); 3363 if (!sb_rdonly(sb)) 3364 return 0; 3365 } 3366 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3367 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3368 "Inode bitmap for group %u not in group " 3369 "(block %llu)!", i, inode_bitmap); 3370 return 0; 3371 } 3372 inode_table = ext4_inode_table(sb, gdp); 3373 if (inode_table == sb_block) { 3374 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3375 "Inode table for group %u overlaps " 3376 "superblock", i); 3377 if (!sb_rdonly(sb)) 3378 return 0; 3379 } 3380 if (inode_table >= sb_block + 1 && 3381 inode_table <= last_bg_block) { 3382 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3383 "Inode table for group %u overlaps " 3384 "block group descriptors", i); 3385 if (!sb_rdonly(sb)) 3386 return 0; 3387 } 3388 if (inode_table < first_block || 3389 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3390 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3391 "Inode table for group %u not in group " 3392 "(block %llu)!", i, inode_table); 3393 return 0; 3394 } 3395 ext4_lock_group(sb, i); 3396 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3397 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3398 "Checksum for group %u failed (%u!=%u)", 3399 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3400 gdp)), le16_to_cpu(gdp->bg_checksum)); 3401 if (!sb_rdonly(sb)) { 3402 ext4_unlock_group(sb, i); 3403 return 0; 3404 } 3405 } 3406 ext4_unlock_group(sb, i); 3407 if (!flexbg_flag) 3408 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3409 } 3410 if (NULL != first_not_zeroed) 3411 *first_not_zeroed = grp; 3412 return 1; 3413 } 3414 3415 /* 3416 * Maximal extent format file size. 3417 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3418 * extent format containers, within a sector_t, and within i_blocks 3419 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3420 * so that won't be a limiting factor. 3421 * 3422 * However there is other limiting factor. We do store extents in the form 3423 * of starting block and length, hence the resulting length of the extent 3424 * covering maximum file size must fit into on-disk format containers as 3425 * well. Given that length is always by 1 unit bigger than max unit (because 3426 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3427 * 3428 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3429 */ 3430 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3431 { 3432 loff_t res; 3433 loff_t upper_limit = MAX_LFS_FILESIZE; 3434 3435 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3436 3437 if (!has_huge_files) { 3438 upper_limit = (1LL << 32) - 1; 3439 3440 /* total blocks in file system block size */ 3441 upper_limit >>= (blkbits - 9); 3442 upper_limit <<= blkbits; 3443 } 3444 3445 /* 3446 * 32-bit extent-start container, ee_block. We lower the maxbytes 3447 * by one fs block, so ee_len can cover the extent of maximum file 3448 * size 3449 */ 3450 res = (1LL << 32) - 1; 3451 res <<= blkbits; 3452 3453 /* Sanity check against vm- & vfs- imposed limits */ 3454 if (res > upper_limit) 3455 res = upper_limit; 3456 3457 return res; 3458 } 3459 3460 /* 3461 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3462 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3463 * We need to be 1 filesystem block less than the 2^48 sector limit. 3464 */ 3465 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3466 { 3467 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3468 int meta_blocks; 3469 unsigned int ppb = 1 << (bits - 2); 3470 3471 /* 3472 * This is calculated to be the largest file size for a dense, block 3473 * mapped file such that the file's total number of 512-byte sectors, 3474 * including data and all indirect blocks, does not exceed (2^48 - 1). 3475 * 3476 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3477 * number of 512-byte sectors of the file. 3478 */ 3479 if (!has_huge_files) { 3480 /* 3481 * !has_huge_files or implies that the inode i_block field 3482 * represents total file blocks in 2^32 512-byte sectors == 3483 * size of vfs inode i_blocks * 8 3484 */ 3485 upper_limit = (1LL << 32) - 1; 3486 3487 /* total blocks in file system block size */ 3488 upper_limit >>= (bits - 9); 3489 3490 } else { 3491 /* 3492 * We use 48 bit ext4_inode i_blocks 3493 * With EXT4_HUGE_FILE_FL set the i_blocks 3494 * represent total number of blocks in 3495 * file system block size 3496 */ 3497 upper_limit = (1LL << 48) - 1; 3498 3499 } 3500 3501 /* Compute how many blocks we can address by block tree */ 3502 res += ppb; 3503 res += ppb * ppb; 3504 res += ((loff_t)ppb) * ppb * ppb; 3505 /* Compute how many metadata blocks are needed */ 3506 meta_blocks = 1; 3507 meta_blocks += 1 + ppb; 3508 meta_blocks += 1 + ppb + ppb * ppb; 3509 /* Does block tree limit file size? */ 3510 if (res + meta_blocks <= upper_limit) 3511 goto check_lfs; 3512 3513 res = upper_limit; 3514 /* How many metadata blocks are needed for addressing upper_limit? */ 3515 upper_limit -= EXT4_NDIR_BLOCKS; 3516 /* indirect blocks */ 3517 meta_blocks = 1; 3518 upper_limit -= ppb; 3519 /* double indirect blocks */ 3520 if (upper_limit < ppb * ppb) { 3521 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3522 res -= meta_blocks; 3523 goto check_lfs; 3524 } 3525 meta_blocks += 1 + ppb; 3526 upper_limit -= ppb * ppb; 3527 /* tripple indirect blocks for the rest */ 3528 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3529 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3530 res -= meta_blocks; 3531 check_lfs: 3532 res <<= bits; 3533 if (res > MAX_LFS_FILESIZE) 3534 res = MAX_LFS_FILESIZE; 3535 3536 return res; 3537 } 3538 3539 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3540 ext4_fsblk_t logical_sb_block, int nr) 3541 { 3542 struct ext4_sb_info *sbi = EXT4_SB(sb); 3543 ext4_group_t bg, first_meta_bg; 3544 int has_super = 0; 3545 3546 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3547 3548 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3549 return logical_sb_block + nr + 1; 3550 bg = sbi->s_desc_per_block * nr; 3551 if (ext4_bg_has_super(sb, bg)) 3552 has_super = 1; 3553 3554 /* 3555 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3556 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3557 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3558 * compensate. 3559 */ 3560 if (sb->s_blocksize == 1024 && nr == 0 && 3561 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3562 has_super++; 3563 3564 return (has_super + ext4_group_first_block_no(sb, bg)); 3565 } 3566 3567 /** 3568 * ext4_get_stripe_size: Get the stripe size. 3569 * @sbi: In memory super block info 3570 * 3571 * If we have specified it via mount option, then 3572 * use the mount option value. If the value specified at mount time is 3573 * greater than the blocks per group use the super block value. 3574 * If the super block value is greater than blocks per group return 0. 3575 * Allocator needs it be less than blocks per group. 3576 * 3577 */ 3578 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3579 { 3580 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3581 unsigned long stripe_width = 3582 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3583 int ret; 3584 3585 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3586 ret = sbi->s_stripe; 3587 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3588 ret = stripe_width; 3589 else if (stride && stride <= sbi->s_blocks_per_group) 3590 ret = stride; 3591 else 3592 ret = 0; 3593 3594 /* 3595 * If the stripe width is 1, this makes no sense and 3596 * we set it to 0 to turn off stripe handling code. 3597 */ 3598 if (ret <= 1) 3599 ret = 0; 3600 3601 return ret; 3602 } 3603 3604 /* 3605 * Check whether this filesystem can be mounted based on 3606 * the features present and the RDONLY/RDWR mount requested. 3607 * Returns 1 if this filesystem can be mounted as requested, 3608 * 0 if it cannot be. 3609 */ 3610 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3611 { 3612 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3613 ext4_msg(sb, KERN_ERR, 3614 "Couldn't mount because of " 3615 "unsupported optional features (%x)", 3616 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3617 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3618 return 0; 3619 } 3620 3621 #if !IS_ENABLED(CONFIG_UNICODE) 3622 if (ext4_has_feature_casefold(sb)) { 3623 ext4_msg(sb, KERN_ERR, 3624 "Filesystem with casefold feature cannot be " 3625 "mounted without CONFIG_UNICODE"); 3626 return 0; 3627 } 3628 #endif 3629 3630 if (readonly) 3631 return 1; 3632 3633 if (ext4_has_feature_readonly(sb)) { 3634 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3635 sb->s_flags |= SB_RDONLY; 3636 return 1; 3637 } 3638 3639 /* Check that feature set is OK for a read-write mount */ 3640 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3641 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3642 "unsupported optional features (%x)", 3643 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3644 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3645 return 0; 3646 } 3647 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3648 ext4_msg(sb, KERN_ERR, 3649 "Can't support bigalloc feature without " 3650 "extents feature\n"); 3651 return 0; 3652 } 3653 3654 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3655 if (!readonly && (ext4_has_feature_quota(sb) || 3656 ext4_has_feature_project(sb))) { 3657 ext4_msg(sb, KERN_ERR, 3658 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3659 return 0; 3660 } 3661 #endif /* CONFIG_QUOTA */ 3662 return 1; 3663 } 3664 3665 /* 3666 * This function is called once a day if we have errors logged 3667 * on the file system 3668 */ 3669 static void print_daily_error_info(struct timer_list *t) 3670 { 3671 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3672 struct super_block *sb = sbi->s_sb; 3673 struct ext4_super_block *es = sbi->s_es; 3674 3675 if (es->s_error_count) 3676 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3677 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3678 le32_to_cpu(es->s_error_count)); 3679 if (es->s_first_error_time) { 3680 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3681 sb->s_id, 3682 ext4_get_tstamp(es, s_first_error_time), 3683 (int) sizeof(es->s_first_error_func), 3684 es->s_first_error_func, 3685 le32_to_cpu(es->s_first_error_line)); 3686 if (es->s_first_error_ino) 3687 printk(KERN_CONT ": inode %u", 3688 le32_to_cpu(es->s_first_error_ino)); 3689 if (es->s_first_error_block) 3690 printk(KERN_CONT ": block %llu", (unsigned long long) 3691 le64_to_cpu(es->s_first_error_block)); 3692 printk(KERN_CONT "\n"); 3693 } 3694 if (es->s_last_error_time) { 3695 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3696 sb->s_id, 3697 ext4_get_tstamp(es, s_last_error_time), 3698 (int) sizeof(es->s_last_error_func), 3699 es->s_last_error_func, 3700 le32_to_cpu(es->s_last_error_line)); 3701 if (es->s_last_error_ino) 3702 printk(KERN_CONT ": inode %u", 3703 le32_to_cpu(es->s_last_error_ino)); 3704 if (es->s_last_error_block) 3705 printk(KERN_CONT ": block %llu", (unsigned long long) 3706 le64_to_cpu(es->s_last_error_block)); 3707 printk(KERN_CONT "\n"); 3708 } 3709 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3710 } 3711 3712 /* Find next suitable group and run ext4_init_inode_table */ 3713 static int ext4_run_li_request(struct ext4_li_request *elr) 3714 { 3715 struct ext4_group_desc *gdp = NULL; 3716 struct super_block *sb = elr->lr_super; 3717 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3718 ext4_group_t group = elr->lr_next_group; 3719 unsigned int prefetch_ios = 0; 3720 int ret = 0; 3721 int nr = EXT4_SB(sb)->s_mb_prefetch; 3722 u64 start_time; 3723 3724 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3725 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios); 3726 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr); 3727 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr); 3728 if (group >= elr->lr_next_group) { 3729 ret = 1; 3730 if (elr->lr_first_not_zeroed != ngroups && 3731 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3732 elr->lr_next_group = elr->lr_first_not_zeroed; 3733 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3734 ret = 0; 3735 } 3736 } 3737 return ret; 3738 } 3739 3740 for (; group < ngroups; group++) { 3741 gdp = ext4_get_group_desc(sb, group, NULL); 3742 if (!gdp) { 3743 ret = 1; 3744 break; 3745 } 3746 3747 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3748 break; 3749 } 3750 3751 if (group >= ngroups) 3752 ret = 1; 3753 3754 if (!ret) { 3755 start_time = ktime_get_real_ns(); 3756 ret = ext4_init_inode_table(sb, group, 3757 elr->lr_timeout ? 0 : 1); 3758 trace_ext4_lazy_itable_init(sb, group); 3759 if (elr->lr_timeout == 0) { 3760 elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) * 3761 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3762 } 3763 elr->lr_next_sched = jiffies + elr->lr_timeout; 3764 elr->lr_next_group = group + 1; 3765 } 3766 return ret; 3767 } 3768 3769 /* 3770 * Remove lr_request from the list_request and free the 3771 * request structure. Should be called with li_list_mtx held 3772 */ 3773 static void ext4_remove_li_request(struct ext4_li_request *elr) 3774 { 3775 if (!elr) 3776 return; 3777 3778 list_del(&elr->lr_request); 3779 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3780 kfree(elr); 3781 } 3782 3783 static void ext4_unregister_li_request(struct super_block *sb) 3784 { 3785 mutex_lock(&ext4_li_mtx); 3786 if (!ext4_li_info) { 3787 mutex_unlock(&ext4_li_mtx); 3788 return; 3789 } 3790 3791 mutex_lock(&ext4_li_info->li_list_mtx); 3792 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3793 mutex_unlock(&ext4_li_info->li_list_mtx); 3794 mutex_unlock(&ext4_li_mtx); 3795 } 3796 3797 static struct task_struct *ext4_lazyinit_task; 3798 3799 /* 3800 * This is the function where ext4lazyinit thread lives. It walks 3801 * through the request list searching for next scheduled filesystem. 3802 * When such a fs is found, run the lazy initialization request 3803 * (ext4_rn_li_request) and keep track of the time spend in this 3804 * function. Based on that time we compute next schedule time of 3805 * the request. When walking through the list is complete, compute 3806 * next waking time and put itself into sleep. 3807 */ 3808 static int ext4_lazyinit_thread(void *arg) 3809 { 3810 struct ext4_lazy_init *eli = arg; 3811 struct list_head *pos, *n; 3812 struct ext4_li_request *elr; 3813 unsigned long next_wakeup, cur; 3814 3815 BUG_ON(NULL == eli); 3816 set_freezable(); 3817 3818 cont_thread: 3819 while (true) { 3820 next_wakeup = MAX_JIFFY_OFFSET; 3821 3822 mutex_lock(&eli->li_list_mtx); 3823 if (list_empty(&eli->li_request_list)) { 3824 mutex_unlock(&eli->li_list_mtx); 3825 goto exit_thread; 3826 } 3827 list_for_each_safe(pos, n, &eli->li_request_list) { 3828 int err = 0; 3829 int progress = 0; 3830 elr = list_entry(pos, struct ext4_li_request, 3831 lr_request); 3832 3833 if (time_before(jiffies, elr->lr_next_sched)) { 3834 if (time_before(elr->lr_next_sched, next_wakeup)) 3835 next_wakeup = elr->lr_next_sched; 3836 continue; 3837 } 3838 if (down_read_trylock(&elr->lr_super->s_umount)) { 3839 if (sb_start_write_trylock(elr->lr_super)) { 3840 progress = 1; 3841 /* 3842 * We hold sb->s_umount, sb can not 3843 * be removed from the list, it is 3844 * now safe to drop li_list_mtx 3845 */ 3846 mutex_unlock(&eli->li_list_mtx); 3847 err = ext4_run_li_request(elr); 3848 sb_end_write(elr->lr_super); 3849 mutex_lock(&eli->li_list_mtx); 3850 n = pos->next; 3851 } 3852 up_read((&elr->lr_super->s_umount)); 3853 } 3854 /* error, remove the lazy_init job */ 3855 if (err) { 3856 ext4_remove_li_request(elr); 3857 continue; 3858 } 3859 if (!progress) { 3860 elr->lr_next_sched = jiffies + 3861 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3862 } 3863 if (time_before(elr->lr_next_sched, next_wakeup)) 3864 next_wakeup = elr->lr_next_sched; 3865 } 3866 mutex_unlock(&eli->li_list_mtx); 3867 3868 try_to_freeze(); 3869 3870 cur = jiffies; 3871 if ((time_after_eq(cur, next_wakeup)) || 3872 (MAX_JIFFY_OFFSET == next_wakeup)) { 3873 cond_resched(); 3874 continue; 3875 } 3876 3877 schedule_timeout_interruptible(next_wakeup - cur); 3878 3879 if (kthread_should_stop()) { 3880 ext4_clear_request_list(); 3881 goto exit_thread; 3882 } 3883 } 3884 3885 exit_thread: 3886 /* 3887 * It looks like the request list is empty, but we need 3888 * to check it under the li_list_mtx lock, to prevent any 3889 * additions into it, and of course we should lock ext4_li_mtx 3890 * to atomically free the list and ext4_li_info, because at 3891 * this point another ext4 filesystem could be registering 3892 * new one. 3893 */ 3894 mutex_lock(&ext4_li_mtx); 3895 mutex_lock(&eli->li_list_mtx); 3896 if (!list_empty(&eli->li_request_list)) { 3897 mutex_unlock(&eli->li_list_mtx); 3898 mutex_unlock(&ext4_li_mtx); 3899 goto cont_thread; 3900 } 3901 mutex_unlock(&eli->li_list_mtx); 3902 kfree(ext4_li_info); 3903 ext4_li_info = NULL; 3904 mutex_unlock(&ext4_li_mtx); 3905 3906 return 0; 3907 } 3908 3909 static void ext4_clear_request_list(void) 3910 { 3911 struct list_head *pos, *n; 3912 struct ext4_li_request *elr; 3913 3914 mutex_lock(&ext4_li_info->li_list_mtx); 3915 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3916 elr = list_entry(pos, struct ext4_li_request, 3917 lr_request); 3918 ext4_remove_li_request(elr); 3919 } 3920 mutex_unlock(&ext4_li_info->li_list_mtx); 3921 } 3922 3923 static int ext4_run_lazyinit_thread(void) 3924 { 3925 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3926 ext4_li_info, "ext4lazyinit"); 3927 if (IS_ERR(ext4_lazyinit_task)) { 3928 int err = PTR_ERR(ext4_lazyinit_task); 3929 ext4_clear_request_list(); 3930 kfree(ext4_li_info); 3931 ext4_li_info = NULL; 3932 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3933 "initialization thread\n", 3934 err); 3935 return err; 3936 } 3937 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3938 return 0; 3939 } 3940 3941 /* 3942 * Check whether it make sense to run itable init. thread or not. 3943 * If there is at least one uninitialized inode table, return 3944 * corresponding group number, else the loop goes through all 3945 * groups and return total number of groups. 3946 */ 3947 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3948 { 3949 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3950 struct ext4_group_desc *gdp = NULL; 3951 3952 if (!ext4_has_group_desc_csum(sb)) 3953 return ngroups; 3954 3955 for (group = 0; group < ngroups; group++) { 3956 gdp = ext4_get_group_desc(sb, group, NULL); 3957 if (!gdp) 3958 continue; 3959 3960 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3961 break; 3962 } 3963 3964 return group; 3965 } 3966 3967 static int ext4_li_info_new(void) 3968 { 3969 struct ext4_lazy_init *eli = NULL; 3970 3971 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3972 if (!eli) 3973 return -ENOMEM; 3974 3975 INIT_LIST_HEAD(&eli->li_request_list); 3976 mutex_init(&eli->li_list_mtx); 3977 3978 eli->li_state |= EXT4_LAZYINIT_QUIT; 3979 3980 ext4_li_info = eli; 3981 3982 return 0; 3983 } 3984 3985 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3986 ext4_group_t start) 3987 { 3988 struct ext4_li_request *elr; 3989 3990 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3991 if (!elr) 3992 return NULL; 3993 3994 elr->lr_super = sb; 3995 elr->lr_first_not_zeroed = start; 3996 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3997 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3998 elr->lr_next_group = start; 3999 } else { 4000 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 4001 } 4002 4003 /* 4004 * Randomize first schedule time of the request to 4005 * spread the inode table initialization requests 4006 * better. 4007 */ 4008 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 4009 return elr; 4010 } 4011 4012 int ext4_register_li_request(struct super_block *sb, 4013 ext4_group_t first_not_zeroed) 4014 { 4015 struct ext4_sb_info *sbi = EXT4_SB(sb); 4016 struct ext4_li_request *elr = NULL; 4017 ext4_group_t ngroups = sbi->s_groups_count; 4018 int ret = 0; 4019 4020 mutex_lock(&ext4_li_mtx); 4021 if (sbi->s_li_request != NULL) { 4022 /* 4023 * Reset timeout so it can be computed again, because 4024 * s_li_wait_mult might have changed. 4025 */ 4026 sbi->s_li_request->lr_timeout = 0; 4027 goto out; 4028 } 4029 4030 if (sb_rdonly(sb) || 4031 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 4032 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 4033 goto out; 4034 4035 elr = ext4_li_request_new(sb, first_not_zeroed); 4036 if (!elr) { 4037 ret = -ENOMEM; 4038 goto out; 4039 } 4040 4041 if (NULL == ext4_li_info) { 4042 ret = ext4_li_info_new(); 4043 if (ret) 4044 goto out; 4045 } 4046 4047 mutex_lock(&ext4_li_info->li_list_mtx); 4048 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 4049 mutex_unlock(&ext4_li_info->li_list_mtx); 4050 4051 sbi->s_li_request = elr; 4052 /* 4053 * set elr to NULL here since it has been inserted to 4054 * the request_list and the removal and free of it is 4055 * handled by ext4_clear_request_list from now on. 4056 */ 4057 elr = NULL; 4058 4059 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 4060 ret = ext4_run_lazyinit_thread(); 4061 if (ret) 4062 goto out; 4063 } 4064 out: 4065 mutex_unlock(&ext4_li_mtx); 4066 if (ret) 4067 kfree(elr); 4068 return ret; 4069 } 4070 4071 /* 4072 * We do not need to lock anything since this is called on 4073 * module unload. 4074 */ 4075 static void ext4_destroy_lazyinit_thread(void) 4076 { 4077 /* 4078 * If thread exited earlier 4079 * there's nothing to be done. 4080 */ 4081 if (!ext4_li_info || !ext4_lazyinit_task) 4082 return; 4083 4084 kthread_stop(ext4_lazyinit_task); 4085 } 4086 4087 static int set_journal_csum_feature_set(struct super_block *sb) 4088 { 4089 int ret = 1; 4090 int compat, incompat; 4091 struct ext4_sb_info *sbi = EXT4_SB(sb); 4092 4093 if (ext4_has_metadata_csum(sb)) { 4094 /* journal checksum v3 */ 4095 compat = 0; 4096 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4097 } else { 4098 /* journal checksum v1 */ 4099 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4100 incompat = 0; 4101 } 4102 4103 jbd2_journal_clear_features(sbi->s_journal, 4104 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4105 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4106 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4107 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4108 ret = jbd2_journal_set_features(sbi->s_journal, 4109 compat, 0, 4110 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4111 incompat); 4112 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4113 ret = jbd2_journal_set_features(sbi->s_journal, 4114 compat, 0, 4115 incompat); 4116 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4117 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4118 } else { 4119 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4120 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4121 } 4122 4123 return ret; 4124 } 4125 4126 /* 4127 * Note: calculating the overhead so we can be compatible with 4128 * historical BSD practice is quite difficult in the face of 4129 * clusters/bigalloc. This is because multiple metadata blocks from 4130 * different block group can end up in the same allocation cluster. 4131 * Calculating the exact overhead in the face of clustered allocation 4132 * requires either O(all block bitmaps) in memory or O(number of block 4133 * groups**2) in time. We will still calculate the superblock for 4134 * older file systems --- and if we come across with a bigalloc file 4135 * system with zero in s_overhead_clusters the estimate will be close to 4136 * correct especially for very large cluster sizes --- but for newer 4137 * file systems, it's better to calculate this figure once at mkfs 4138 * time, and store it in the superblock. If the superblock value is 4139 * present (even for non-bigalloc file systems), we will use it. 4140 */ 4141 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4142 char *buf) 4143 { 4144 struct ext4_sb_info *sbi = EXT4_SB(sb); 4145 struct ext4_group_desc *gdp; 4146 ext4_fsblk_t first_block, last_block, b; 4147 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4148 int s, j, count = 0; 4149 int has_super = ext4_bg_has_super(sb, grp); 4150 4151 if (!ext4_has_feature_bigalloc(sb)) 4152 return (has_super + ext4_bg_num_gdb(sb, grp) + 4153 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4154 sbi->s_itb_per_group + 2); 4155 4156 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4157 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4158 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4159 for (i = 0; i < ngroups; i++) { 4160 gdp = ext4_get_group_desc(sb, i, NULL); 4161 b = ext4_block_bitmap(sb, gdp); 4162 if (b >= first_block && b <= last_block) { 4163 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4164 count++; 4165 } 4166 b = ext4_inode_bitmap(sb, gdp); 4167 if (b >= first_block && b <= last_block) { 4168 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4169 count++; 4170 } 4171 b = ext4_inode_table(sb, gdp); 4172 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4173 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4174 int c = EXT4_B2C(sbi, b - first_block); 4175 ext4_set_bit(c, buf); 4176 count++; 4177 } 4178 if (i != grp) 4179 continue; 4180 s = 0; 4181 if (ext4_bg_has_super(sb, grp)) { 4182 ext4_set_bit(s++, buf); 4183 count++; 4184 } 4185 j = ext4_bg_num_gdb(sb, grp); 4186 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4187 ext4_error(sb, "Invalid number of block group " 4188 "descriptor blocks: %d", j); 4189 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4190 } 4191 count += j; 4192 for (; j > 0; j--) 4193 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4194 } 4195 if (!count) 4196 return 0; 4197 return EXT4_CLUSTERS_PER_GROUP(sb) - 4198 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4199 } 4200 4201 /* 4202 * Compute the overhead and stash it in sbi->s_overhead 4203 */ 4204 int ext4_calculate_overhead(struct super_block *sb) 4205 { 4206 struct ext4_sb_info *sbi = EXT4_SB(sb); 4207 struct ext4_super_block *es = sbi->s_es; 4208 struct inode *j_inode; 4209 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4210 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4211 ext4_fsblk_t overhead = 0; 4212 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4213 4214 if (!buf) 4215 return -ENOMEM; 4216 4217 /* 4218 * Compute the overhead (FS structures). This is constant 4219 * for a given filesystem unless the number of block groups 4220 * changes so we cache the previous value until it does. 4221 */ 4222 4223 /* 4224 * All of the blocks before first_data_block are overhead 4225 */ 4226 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4227 4228 /* 4229 * Add the overhead found in each block group 4230 */ 4231 for (i = 0; i < ngroups; i++) { 4232 int blks; 4233 4234 blks = count_overhead(sb, i, buf); 4235 overhead += blks; 4236 if (blks) 4237 memset(buf, 0, PAGE_SIZE); 4238 cond_resched(); 4239 } 4240 4241 /* 4242 * Add the internal journal blocks whether the journal has been 4243 * loaded or not 4244 */ 4245 if (sbi->s_journal && !sbi->s_journal_bdev) 4246 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4247 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4248 /* j_inum for internal journal is non-zero */ 4249 j_inode = ext4_get_journal_inode(sb, j_inum); 4250 if (!IS_ERR(j_inode)) { 4251 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4252 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4253 iput(j_inode); 4254 } else { 4255 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4256 } 4257 } 4258 sbi->s_overhead = overhead; 4259 smp_wmb(); 4260 free_page((unsigned long) buf); 4261 return 0; 4262 } 4263 4264 static void ext4_set_resv_clusters(struct super_block *sb) 4265 { 4266 ext4_fsblk_t resv_clusters; 4267 struct ext4_sb_info *sbi = EXT4_SB(sb); 4268 4269 /* 4270 * There's no need to reserve anything when we aren't using extents. 4271 * The space estimates are exact, there are no unwritten extents, 4272 * hole punching doesn't need new metadata... This is needed especially 4273 * to keep ext2/3 backward compatibility. 4274 */ 4275 if (!ext4_has_feature_extents(sb)) 4276 return; 4277 /* 4278 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4279 * This should cover the situations where we can not afford to run 4280 * out of space like for example punch hole, or converting 4281 * unwritten extents in delalloc path. In most cases such 4282 * allocation would require 1, or 2 blocks, higher numbers are 4283 * very rare. 4284 */ 4285 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4286 sbi->s_cluster_bits); 4287 4288 do_div(resv_clusters, 50); 4289 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4290 4291 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4292 } 4293 4294 static const char *ext4_quota_mode(struct super_block *sb) 4295 { 4296 #ifdef CONFIG_QUOTA 4297 if (!ext4_quota_capable(sb)) 4298 return "none"; 4299 4300 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4301 return "journalled"; 4302 else 4303 return "writeback"; 4304 #else 4305 return "disabled"; 4306 #endif 4307 } 4308 4309 static void ext4_setup_csum_trigger(struct super_block *sb, 4310 enum ext4_journal_trigger_type type, 4311 void (*trigger)( 4312 struct jbd2_buffer_trigger_type *type, 4313 struct buffer_head *bh, 4314 void *mapped_data, 4315 size_t size)) 4316 { 4317 struct ext4_sb_info *sbi = EXT4_SB(sb); 4318 4319 sbi->s_journal_triggers[type].sb = sb; 4320 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4321 } 4322 4323 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4324 { 4325 if (!sbi) 4326 return; 4327 4328 kfree(sbi->s_blockgroup_lock); 4329 fs_put_dax(sbi->s_daxdev, NULL); 4330 kfree(sbi); 4331 } 4332 4333 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4334 { 4335 struct ext4_sb_info *sbi; 4336 4337 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4338 if (!sbi) 4339 return NULL; 4340 4341 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4342 NULL, NULL); 4343 4344 sbi->s_blockgroup_lock = 4345 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4346 4347 if (!sbi->s_blockgroup_lock) 4348 goto err_out; 4349 4350 sb->s_fs_info = sbi; 4351 sbi->s_sb = sb; 4352 return sbi; 4353 err_out: 4354 fs_put_dax(sbi->s_daxdev, NULL); 4355 kfree(sbi); 4356 return NULL; 4357 } 4358 4359 static void ext4_set_def_opts(struct super_block *sb, 4360 struct ext4_super_block *es) 4361 { 4362 unsigned long def_mount_opts; 4363 4364 /* Set defaults before we parse the mount options */ 4365 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4366 set_opt(sb, INIT_INODE_TABLE); 4367 if (def_mount_opts & EXT4_DEFM_DEBUG) 4368 set_opt(sb, DEBUG); 4369 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4370 set_opt(sb, GRPID); 4371 if (def_mount_opts & EXT4_DEFM_UID16) 4372 set_opt(sb, NO_UID32); 4373 /* xattr user namespace & acls are now defaulted on */ 4374 set_opt(sb, XATTR_USER); 4375 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4376 set_opt(sb, POSIX_ACL); 4377 #endif 4378 if (ext4_has_feature_fast_commit(sb)) 4379 set_opt2(sb, JOURNAL_FAST_COMMIT); 4380 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4381 if (ext4_has_metadata_csum(sb)) 4382 set_opt(sb, JOURNAL_CHECKSUM); 4383 4384 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4385 set_opt(sb, JOURNAL_DATA); 4386 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4387 set_opt(sb, ORDERED_DATA); 4388 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4389 set_opt(sb, WRITEBACK_DATA); 4390 4391 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4392 set_opt(sb, ERRORS_PANIC); 4393 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4394 set_opt(sb, ERRORS_CONT); 4395 else 4396 set_opt(sb, ERRORS_RO); 4397 /* block_validity enabled by default; disable with noblock_validity */ 4398 set_opt(sb, BLOCK_VALIDITY); 4399 if (def_mount_opts & EXT4_DEFM_DISCARD) 4400 set_opt(sb, DISCARD); 4401 4402 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4403 set_opt(sb, BARRIER); 4404 4405 /* 4406 * enable delayed allocation by default 4407 * Use -o nodelalloc to turn it off 4408 */ 4409 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4410 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4411 set_opt(sb, DELALLOC); 4412 4413 if (sb->s_blocksize == PAGE_SIZE) 4414 set_opt(sb, DIOREAD_NOLOCK); 4415 } 4416 4417 static int ext4_handle_clustersize(struct super_block *sb) 4418 { 4419 struct ext4_sb_info *sbi = EXT4_SB(sb); 4420 struct ext4_super_block *es = sbi->s_es; 4421 int clustersize; 4422 4423 /* Handle clustersize */ 4424 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4425 if (ext4_has_feature_bigalloc(sb)) { 4426 if (clustersize < sb->s_blocksize) { 4427 ext4_msg(sb, KERN_ERR, 4428 "cluster size (%d) smaller than " 4429 "block size (%lu)", clustersize, sb->s_blocksize); 4430 return -EINVAL; 4431 } 4432 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4433 le32_to_cpu(es->s_log_block_size); 4434 sbi->s_clusters_per_group = 4435 le32_to_cpu(es->s_clusters_per_group); 4436 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4437 ext4_msg(sb, KERN_ERR, 4438 "#clusters per group too big: %lu", 4439 sbi->s_clusters_per_group); 4440 return -EINVAL; 4441 } 4442 if (sbi->s_blocks_per_group != 4443 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4444 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4445 "clusters per group (%lu) inconsistent", 4446 sbi->s_blocks_per_group, 4447 sbi->s_clusters_per_group); 4448 return -EINVAL; 4449 } 4450 } else { 4451 if (clustersize != sb->s_blocksize) { 4452 ext4_msg(sb, KERN_ERR, 4453 "fragment/cluster size (%d) != " 4454 "block size (%lu)", clustersize, sb->s_blocksize); 4455 return -EINVAL; 4456 } 4457 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4458 ext4_msg(sb, KERN_ERR, 4459 "#blocks per group too big: %lu", 4460 sbi->s_blocks_per_group); 4461 return -EINVAL; 4462 } 4463 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4464 sbi->s_cluster_bits = 0; 4465 } 4466 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4467 4468 /* Do we have standard group size of clustersize * 8 blocks ? */ 4469 if (sbi->s_blocks_per_group == clustersize << 3) 4470 set_opt2(sb, STD_GROUP_SIZE); 4471 4472 return 0; 4473 } 4474 4475 static void ext4_fast_commit_init(struct super_block *sb) 4476 { 4477 struct ext4_sb_info *sbi = EXT4_SB(sb); 4478 4479 /* Initialize fast commit stuff */ 4480 atomic_set(&sbi->s_fc_subtid, 0); 4481 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4482 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4483 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4484 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4485 sbi->s_fc_bytes = 0; 4486 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4487 sbi->s_fc_ineligible_tid = 0; 4488 spin_lock_init(&sbi->s_fc_lock); 4489 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4490 sbi->s_fc_replay_state.fc_regions = NULL; 4491 sbi->s_fc_replay_state.fc_regions_size = 0; 4492 sbi->s_fc_replay_state.fc_regions_used = 0; 4493 sbi->s_fc_replay_state.fc_regions_valid = 0; 4494 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4495 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4496 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4497 } 4498 4499 static int ext4_inode_info_init(struct super_block *sb, 4500 struct ext4_super_block *es) 4501 { 4502 struct ext4_sb_info *sbi = EXT4_SB(sb); 4503 4504 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4505 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4506 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4507 } else { 4508 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4509 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4510 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4511 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4512 sbi->s_first_ino); 4513 return -EINVAL; 4514 } 4515 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4516 (!is_power_of_2(sbi->s_inode_size)) || 4517 (sbi->s_inode_size > sb->s_blocksize)) { 4518 ext4_msg(sb, KERN_ERR, 4519 "unsupported inode size: %d", 4520 sbi->s_inode_size); 4521 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4522 return -EINVAL; 4523 } 4524 /* 4525 * i_atime_extra is the last extra field available for 4526 * [acm]times in struct ext4_inode. Checking for that 4527 * field should suffice to ensure we have extra space 4528 * for all three. 4529 */ 4530 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4531 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4532 sb->s_time_gran = 1; 4533 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4534 } else { 4535 sb->s_time_gran = NSEC_PER_SEC; 4536 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4537 } 4538 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4539 } 4540 4541 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4542 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4543 EXT4_GOOD_OLD_INODE_SIZE; 4544 if (ext4_has_feature_extra_isize(sb)) { 4545 unsigned v, max = (sbi->s_inode_size - 4546 EXT4_GOOD_OLD_INODE_SIZE); 4547 4548 v = le16_to_cpu(es->s_want_extra_isize); 4549 if (v > max) { 4550 ext4_msg(sb, KERN_ERR, 4551 "bad s_want_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 v = le16_to_cpu(es->s_min_extra_isize); 4558 if (v > max) { 4559 ext4_msg(sb, KERN_ERR, 4560 "bad s_min_extra_isize: %d", v); 4561 return -EINVAL; 4562 } 4563 if (sbi->s_want_extra_isize < v) 4564 sbi->s_want_extra_isize = v; 4565 } 4566 } 4567 4568 return 0; 4569 } 4570 4571 #if IS_ENABLED(CONFIG_UNICODE) 4572 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4573 { 4574 const struct ext4_sb_encodings *encoding_info; 4575 struct unicode_map *encoding; 4576 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4577 4578 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4579 return 0; 4580 4581 encoding_info = ext4_sb_read_encoding(es); 4582 if (!encoding_info) { 4583 ext4_msg(sb, KERN_ERR, 4584 "Encoding requested by superblock is unknown"); 4585 return -EINVAL; 4586 } 4587 4588 encoding = utf8_load(encoding_info->version); 4589 if (IS_ERR(encoding)) { 4590 ext4_msg(sb, KERN_ERR, 4591 "can't mount with superblock charset: %s-%u.%u.%u " 4592 "not supported by the kernel. flags: 0x%x.", 4593 encoding_info->name, 4594 unicode_major(encoding_info->version), 4595 unicode_minor(encoding_info->version), 4596 unicode_rev(encoding_info->version), 4597 encoding_flags); 4598 return -EINVAL; 4599 } 4600 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4601 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4602 unicode_major(encoding_info->version), 4603 unicode_minor(encoding_info->version), 4604 unicode_rev(encoding_info->version), 4605 encoding_flags); 4606 4607 sb->s_encoding = encoding; 4608 sb->s_encoding_flags = encoding_flags; 4609 4610 return 0; 4611 } 4612 #else 4613 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4614 { 4615 return 0; 4616 } 4617 #endif 4618 4619 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4620 { 4621 struct ext4_sb_info *sbi = EXT4_SB(sb); 4622 4623 /* Warn if metadata_csum and gdt_csum are both set. */ 4624 if (ext4_has_feature_metadata_csum(sb) && 4625 ext4_has_feature_gdt_csum(sb)) 4626 ext4_warning(sb, "metadata_csum and uninit_bg are " 4627 "redundant flags; please run fsck."); 4628 4629 /* Check for a known checksum algorithm */ 4630 if (!ext4_verify_csum_type(sb, es)) { 4631 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4632 "unknown checksum algorithm."); 4633 return -EINVAL; 4634 } 4635 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4636 ext4_orphan_file_block_trigger); 4637 4638 /* Load the checksum driver */ 4639 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4640 if (IS_ERR(sbi->s_chksum_driver)) { 4641 int ret = PTR_ERR(sbi->s_chksum_driver); 4642 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4643 sbi->s_chksum_driver = NULL; 4644 return ret; 4645 } 4646 4647 /* Check superblock checksum */ 4648 if (!ext4_superblock_csum_verify(sb, es)) { 4649 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4650 "invalid superblock checksum. Run e2fsck?"); 4651 return -EFSBADCRC; 4652 } 4653 4654 /* Precompute checksum seed for all metadata */ 4655 if (ext4_has_feature_csum_seed(sb)) 4656 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4657 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4658 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4659 sizeof(es->s_uuid)); 4660 return 0; 4661 } 4662 4663 static int ext4_check_feature_compatibility(struct super_block *sb, 4664 struct ext4_super_block *es, 4665 int silent) 4666 { 4667 struct ext4_sb_info *sbi = EXT4_SB(sb); 4668 4669 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4670 (ext4_has_compat_features(sb) || 4671 ext4_has_ro_compat_features(sb) || 4672 ext4_has_incompat_features(sb))) 4673 ext4_msg(sb, KERN_WARNING, 4674 "feature flags set on rev 0 fs, " 4675 "running e2fsck is recommended"); 4676 4677 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4678 set_opt2(sb, HURD_COMPAT); 4679 if (ext4_has_feature_64bit(sb)) { 4680 ext4_msg(sb, KERN_ERR, 4681 "The Hurd can't support 64-bit file systems"); 4682 return -EINVAL; 4683 } 4684 4685 /* 4686 * ea_inode feature uses l_i_version field which is not 4687 * available in HURD_COMPAT mode. 4688 */ 4689 if (ext4_has_feature_ea_inode(sb)) { 4690 ext4_msg(sb, KERN_ERR, 4691 "ea_inode feature is not supported for Hurd"); 4692 return -EINVAL; 4693 } 4694 } 4695 4696 if (IS_EXT2_SB(sb)) { 4697 if (ext2_feature_set_ok(sb)) 4698 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4699 "using the ext4 subsystem"); 4700 else { 4701 /* 4702 * If we're probing be silent, if this looks like 4703 * it's actually an ext[34] filesystem. 4704 */ 4705 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4706 return -EINVAL; 4707 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4708 "to feature incompatibilities"); 4709 return -EINVAL; 4710 } 4711 } 4712 4713 if (IS_EXT3_SB(sb)) { 4714 if (ext3_feature_set_ok(sb)) 4715 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4716 "using the ext4 subsystem"); 4717 else { 4718 /* 4719 * If we're probing be silent, if this looks like 4720 * it's actually an ext4 filesystem. 4721 */ 4722 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4723 return -EINVAL; 4724 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4725 "to feature incompatibilities"); 4726 return -EINVAL; 4727 } 4728 } 4729 4730 /* 4731 * Check feature flags regardless of the revision level, since we 4732 * previously didn't change the revision level when setting the flags, 4733 * so there is a chance incompat flags are set on a rev 0 filesystem. 4734 */ 4735 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4736 return -EINVAL; 4737 4738 if (sbi->s_daxdev) { 4739 if (sb->s_blocksize == PAGE_SIZE) 4740 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4741 else 4742 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 4743 } 4744 4745 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4746 if (ext4_has_feature_inline_data(sb)) { 4747 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4748 " that may contain inline data"); 4749 return -EINVAL; 4750 } 4751 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4752 ext4_msg(sb, KERN_ERR, 4753 "DAX unsupported by block device."); 4754 return -EINVAL; 4755 } 4756 } 4757 4758 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4759 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4760 es->s_encryption_level); 4761 return -EINVAL; 4762 } 4763 4764 return 0; 4765 } 4766 4767 static int ext4_check_geometry(struct super_block *sb, 4768 struct ext4_super_block *es) 4769 { 4770 struct ext4_sb_info *sbi = EXT4_SB(sb); 4771 __u64 blocks_count; 4772 int err; 4773 4774 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 4775 ext4_msg(sb, KERN_ERR, 4776 "Number of reserved GDT blocks insanely large: %d", 4777 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4778 return -EINVAL; 4779 } 4780 /* 4781 * Test whether we have more sectors than will fit in sector_t, 4782 * and whether the max offset is addressable by the page cache. 4783 */ 4784 err = generic_check_addressable(sb->s_blocksize_bits, 4785 ext4_blocks_count(es)); 4786 if (err) { 4787 ext4_msg(sb, KERN_ERR, "filesystem" 4788 " too large to mount safely on this system"); 4789 return err; 4790 } 4791 4792 /* check blocks count against device size */ 4793 blocks_count = sb_bdev_nr_blocks(sb); 4794 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4795 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4796 "exceeds size of device (%llu blocks)", 4797 ext4_blocks_count(es), blocks_count); 4798 return -EINVAL; 4799 } 4800 4801 /* 4802 * It makes no sense for the first data block to be beyond the end 4803 * of the filesystem. 4804 */ 4805 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4806 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4807 "block %u is beyond end of filesystem (%llu)", 4808 le32_to_cpu(es->s_first_data_block), 4809 ext4_blocks_count(es)); 4810 return -EINVAL; 4811 } 4812 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4813 (sbi->s_cluster_ratio == 1)) { 4814 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4815 "block is 0 with a 1k block and cluster size"); 4816 return -EINVAL; 4817 } 4818 4819 blocks_count = (ext4_blocks_count(es) - 4820 le32_to_cpu(es->s_first_data_block) + 4821 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4822 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4823 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4824 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4825 "(block count %llu, first data block %u, " 4826 "blocks per group %lu)", blocks_count, 4827 ext4_blocks_count(es), 4828 le32_to_cpu(es->s_first_data_block), 4829 EXT4_BLOCKS_PER_GROUP(sb)); 4830 return -EINVAL; 4831 } 4832 sbi->s_groups_count = blocks_count; 4833 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4834 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4835 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4836 le32_to_cpu(es->s_inodes_count)) { 4837 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4838 le32_to_cpu(es->s_inodes_count), 4839 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4840 return -EINVAL; 4841 } 4842 4843 return 0; 4844 } 4845 4846 static int ext4_group_desc_init(struct super_block *sb, 4847 struct ext4_super_block *es, 4848 ext4_fsblk_t logical_sb_block, 4849 ext4_group_t *first_not_zeroed) 4850 { 4851 struct ext4_sb_info *sbi = EXT4_SB(sb); 4852 unsigned int db_count; 4853 ext4_fsblk_t block; 4854 int i; 4855 4856 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4857 EXT4_DESC_PER_BLOCK(sb); 4858 if (ext4_has_feature_meta_bg(sb)) { 4859 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4860 ext4_msg(sb, KERN_WARNING, 4861 "first meta block group too large: %u " 4862 "(group descriptor block count %u)", 4863 le32_to_cpu(es->s_first_meta_bg), db_count); 4864 return -EINVAL; 4865 } 4866 } 4867 rcu_assign_pointer(sbi->s_group_desc, 4868 kvmalloc_array(db_count, 4869 sizeof(struct buffer_head *), 4870 GFP_KERNEL)); 4871 if (sbi->s_group_desc == NULL) { 4872 ext4_msg(sb, KERN_ERR, "not enough memory"); 4873 return -ENOMEM; 4874 } 4875 4876 bgl_lock_init(sbi->s_blockgroup_lock); 4877 4878 /* Pre-read the descriptors into the buffer cache */ 4879 for (i = 0; i < db_count; i++) { 4880 block = descriptor_loc(sb, logical_sb_block, i); 4881 ext4_sb_breadahead_unmovable(sb, block); 4882 } 4883 4884 for (i = 0; i < db_count; i++) { 4885 struct buffer_head *bh; 4886 4887 block = descriptor_loc(sb, logical_sb_block, i); 4888 bh = ext4_sb_bread_unmovable(sb, block); 4889 if (IS_ERR(bh)) { 4890 ext4_msg(sb, KERN_ERR, 4891 "can't read group descriptor %d", i); 4892 sbi->s_gdb_count = i; 4893 return PTR_ERR(bh); 4894 } 4895 rcu_read_lock(); 4896 rcu_dereference(sbi->s_group_desc)[i] = bh; 4897 rcu_read_unlock(); 4898 } 4899 sbi->s_gdb_count = db_count; 4900 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4901 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4902 return -EFSCORRUPTED; 4903 } 4904 4905 return 0; 4906 } 4907 4908 static int ext4_load_and_init_journal(struct super_block *sb, 4909 struct ext4_super_block *es, 4910 struct ext4_fs_context *ctx) 4911 { 4912 struct ext4_sb_info *sbi = EXT4_SB(sb); 4913 int err; 4914 4915 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4916 if (err) 4917 return err; 4918 4919 if (ext4_has_feature_64bit(sb) && 4920 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4921 JBD2_FEATURE_INCOMPAT_64BIT)) { 4922 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4923 goto out; 4924 } 4925 4926 if (!set_journal_csum_feature_set(sb)) { 4927 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4928 "feature set"); 4929 goto out; 4930 } 4931 4932 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4933 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4934 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4935 ext4_msg(sb, KERN_ERR, 4936 "Failed to set fast commit journal feature"); 4937 goto out; 4938 } 4939 4940 /* We have now updated the journal if required, so we can 4941 * validate the data journaling mode. */ 4942 switch (test_opt(sb, DATA_FLAGS)) { 4943 case 0: 4944 /* No mode set, assume a default based on the journal 4945 * capabilities: ORDERED_DATA if the journal can 4946 * cope, else JOURNAL_DATA 4947 */ 4948 if (jbd2_journal_check_available_features 4949 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4950 set_opt(sb, ORDERED_DATA); 4951 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4952 } else { 4953 set_opt(sb, JOURNAL_DATA); 4954 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4955 } 4956 break; 4957 4958 case EXT4_MOUNT_ORDERED_DATA: 4959 case EXT4_MOUNT_WRITEBACK_DATA: 4960 if (!jbd2_journal_check_available_features 4961 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4962 ext4_msg(sb, KERN_ERR, "Journal does not support " 4963 "requested data journaling mode"); 4964 goto out; 4965 } 4966 break; 4967 default: 4968 break; 4969 } 4970 4971 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4972 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4973 ext4_msg(sb, KERN_ERR, "can't mount with " 4974 "journal_async_commit in data=ordered mode"); 4975 goto out; 4976 } 4977 4978 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4979 4980 sbi->s_journal->j_submit_inode_data_buffers = 4981 ext4_journal_submit_inode_data_buffers; 4982 sbi->s_journal->j_finish_inode_data_buffers = 4983 ext4_journal_finish_inode_data_buffers; 4984 4985 return 0; 4986 4987 out: 4988 /* flush s_sb_upd_work before destroying the journal. */ 4989 flush_work(&sbi->s_sb_upd_work); 4990 jbd2_journal_destroy(sbi->s_journal); 4991 sbi->s_journal = NULL; 4992 return -EINVAL; 4993 } 4994 4995 static int ext4_check_journal_data_mode(struct super_block *sb) 4996 { 4997 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4998 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4999 "data=journal disables delayed allocation, " 5000 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 5001 /* can't mount with both data=journal and dioread_nolock. */ 5002 clear_opt(sb, DIOREAD_NOLOCK); 5003 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5004 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5005 ext4_msg(sb, KERN_ERR, "can't mount with " 5006 "both data=journal and delalloc"); 5007 return -EINVAL; 5008 } 5009 if (test_opt(sb, DAX_ALWAYS)) { 5010 ext4_msg(sb, KERN_ERR, "can't mount with " 5011 "both data=journal and dax"); 5012 return -EINVAL; 5013 } 5014 if (ext4_has_feature_encrypt(sb)) { 5015 ext4_msg(sb, KERN_WARNING, 5016 "encrypted files will use data=ordered " 5017 "instead of data journaling mode"); 5018 } 5019 if (test_opt(sb, DELALLOC)) 5020 clear_opt(sb, DELALLOC); 5021 } else { 5022 sb->s_iflags |= SB_I_CGROUPWB; 5023 } 5024 5025 return 0; 5026 } 5027 5028 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 5029 int silent) 5030 { 5031 struct ext4_sb_info *sbi = EXT4_SB(sb); 5032 struct ext4_super_block *es; 5033 ext4_fsblk_t logical_sb_block; 5034 unsigned long offset = 0; 5035 struct buffer_head *bh; 5036 int ret = -EINVAL; 5037 int blocksize; 5038 5039 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 5040 if (!blocksize) { 5041 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 5042 return -EINVAL; 5043 } 5044 5045 /* 5046 * The ext4 superblock will not be buffer aligned for other than 1kB 5047 * block sizes. We need to calculate the offset from buffer start. 5048 */ 5049 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 5050 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5051 offset = do_div(logical_sb_block, blocksize); 5052 } else { 5053 logical_sb_block = sbi->s_sb_block; 5054 } 5055 5056 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5057 if (IS_ERR(bh)) { 5058 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 5059 return PTR_ERR(bh); 5060 } 5061 /* 5062 * Note: s_es must be initialized as soon as possible because 5063 * some ext4 macro-instructions depend on its value 5064 */ 5065 es = (struct ext4_super_block *) (bh->b_data + offset); 5066 sbi->s_es = es; 5067 sb->s_magic = le16_to_cpu(es->s_magic); 5068 if (sb->s_magic != EXT4_SUPER_MAGIC) { 5069 if (!silent) 5070 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5071 goto out; 5072 } 5073 5074 if (le32_to_cpu(es->s_log_block_size) > 5075 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5076 ext4_msg(sb, KERN_ERR, 5077 "Invalid log block size: %u", 5078 le32_to_cpu(es->s_log_block_size)); 5079 goto out; 5080 } 5081 if (le32_to_cpu(es->s_log_cluster_size) > 5082 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5083 ext4_msg(sb, KERN_ERR, 5084 "Invalid log cluster size: %u", 5085 le32_to_cpu(es->s_log_cluster_size)); 5086 goto out; 5087 } 5088 5089 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 5090 5091 /* 5092 * If the default block size is not the same as the real block size, 5093 * we need to reload it. 5094 */ 5095 if (sb->s_blocksize == blocksize) { 5096 *lsb = logical_sb_block; 5097 sbi->s_sbh = bh; 5098 return 0; 5099 } 5100 5101 /* 5102 * bh must be released before kill_bdev(), otherwise 5103 * it won't be freed and its page also. kill_bdev() 5104 * is called by sb_set_blocksize(). 5105 */ 5106 brelse(bh); 5107 /* Validate the filesystem blocksize */ 5108 if (!sb_set_blocksize(sb, blocksize)) { 5109 ext4_msg(sb, KERN_ERR, "bad block size %d", 5110 blocksize); 5111 bh = NULL; 5112 goto out; 5113 } 5114 5115 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5116 offset = do_div(logical_sb_block, blocksize); 5117 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5118 if (IS_ERR(bh)) { 5119 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5120 ret = PTR_ERR(bh); 5121 bh = NULL; 5122 goto out; 5123 } 5124 es = (struct ext4_super_block *)(bh->b_data + offset); 5125 sbi->s_es = es; 5126 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5127 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5128 goto out; 5129 } 5130 *lsb = logical_sb_block; 5131 sbi->s_sbh = bh; 5132 return 0; 5133 out: 5134 brelse(bh); 5135 return ret; 5136 } 5137 5138 static void ext4_hash_info_init(struct super_block *sb) 5139 { 5140 struct ext4_sb_info *sbi = EXT4_SB(sb); 5141 struct ext4_super_block *es = sbi->s_es; 5142 unsigned int i; 5143 5144 for (i = 0; i < 4; i++) 5145 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5146 5147 sbi->s_def_hash_version = es->s_def_hash_version; 5148 if (ext4_has_feature_dir_index(sb)) { 5149 i = le32_to_cpu(es->s_flags); 5150 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5151 sbi->s_hash_unsigned = 3; 5152 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5153 #ifdef __CHAR_UNSIGNED__ 5154 if (!sb_rdonly(sb)) 5155 es->s_flags |= 5156 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5157 sbi->s_hash_unsigned = 3; 5158 #else 5159 if (!sb_rdonly(sb)) 5160 es->s_flags |= 5161 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5162 #endif 5163 } 5164 } 5165 } 5166 5167 static int ext4_block_group_meta_init(struct super_block *sb, int silent) 5168 { 5169 struct ext4_sb_info *sbi = EXT4_SB(sb); 5170 struct ext4_super_block *es = sbi->s_es; 5171 int has_huge_files; 5172 5173 has_huge_files = ext4_has_feature_huge_file(sb); 5174 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5175 has_huge_files); 5176 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5177 5178 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5179 if (ext4_has_feature_64bit(sb)) { 5180 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5181 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5182 !is_power_of_2(sbi->s_desc_size)) { 5183 ext4_msg(sb, KERN_ERR, 5184 "unsupported descriptor size %lu", 5185 sbi->s_desc_size); 5186 return -EINVAL; 5187 } 5188 } else 5189 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5190 5191 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5192 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5193 5194 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5195 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5196 if (!silent) 5197 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5198 return -EINVAL; 5199 } 5200 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5201 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5202 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5203 sbi->s_inodes_per_group); 5204 return -EINVAL; 5205 } 5206 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5207 sbi->s_inodes_per_block; 5208 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5209 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5210 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5211 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5212 5213 return 0; 5214 } 5215 5216 /* 5217 * It's hard to get stripe aligned blocks if stripe is not aligned with 5218 * cluster, just disable stripe and alert user to simplify code and avoid 5219 * stripe aligned allocation which will rarely succeed. 5220 */ 5221 static bool ext4_is_stripe_incompatible(struct super_block *sb, unsigned long stripe) 5222 { 5223 struct ext4_sb_info *sbi = EXT4_SB(sb); 5224 return (stripe > 0 && sbi->s_cluster_ratio > 1 && 5225 stripe % sbi->s_cluster_ratio != 0); 5226 } 5227 5228 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5229 { 5230 struct ext4_super_block *es = NULL; 5231 struct ext4_sb_info *sbi = EXT4_SB(sb); 5232 ext4_fsblk_t logical_sb_block; 5233 struct inode *root; 5234 int needs_recovery; 5235 int err; 5236 ext4_group_t first_not_zeroed; 5237 struct ext4_fs_context *ctx = fc->fs_private; 5238 int silent = fc->sb_flags & SB_SILENT; 5239 5240 /* Set defaults for the variables that will be set during parsing */ 5241 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5242 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5243 5244 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5245 sbi->s_sectors_written_start = 5246 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5247 5248 err = ext4_load_super(sb, &logical_sb_block, silent); 5249 if (err) 5250 goto out_fail; 5251 5252 es = sbi->s_es; 5253 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5254 5255 err = ext4_init_metadata_csum(sb, es); 5256 if (err) 5257 goto failed_mount; 5258 5259 ext4_set_def_opts(sb, es); 5260 5261 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5262 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5263 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5264 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5265 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5266 5267 /* 5268 * set default s_li_wait_mult for lazyinit, for the case there is 5269 * no mount option specified. 5270 */ 5271 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5272 5273 err = ext4_inode_info_init(sb, es); 5274 if (err) 5275 goto failed_mount; 5276 5277 err = parse_apply_sb_mount_options(sb, ctx); 5278 if (err < 0) 5279 goto failed_mount; 5280 5281 sbi->s_def_mount_opt = sbi->s_mount_opt; 5282 sbi->s_def_mount_opt2 = sbi->s_mount_opt2; 5283 5284 err = ext4_check_opt_consistency(fc, sb); 5285 if (err < 0) 5286 goto failed_mount; 5287 5288 ext4_apply_options(fc, sb); 5289 5290 err = ext4_encoding_init(sb, es); 5291 if (err) 5292 goto failed_mount; 5293 5294 err = ext4_check_journal_data_mode(sb); 5295 if (err) 5296 goto failed_mount; 5297 5298 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5299 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5300 5301 /* i_version is always enabled now */ 5302 sb->s_flags |= SB_I_VERSION; 5303 5304 err = ext4_check_feature_compatibility(sb, es, silent); 5305 if (err) 5306 goto failed_mount; 5307 5308 err = ext4_block_group_meta_init(sb, silent); 5309 if (err) 5310 goto failed_mount; 5311 5312 ext4_hash_info_init(sb); 5313 5314 err = ext4_handle_clustersize(sb); 5315 if (err) 5316 goto failed_mount; 5317 5318 err = ext4_check_geometry(sb, es); 5319 if (err) 5320 goto failed_mount; 5321 5322 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5323 spin_lock_init(&sbi->s_error_lock); 5324 INIT_WORK(&sbi->s_sb_upd_work, update_super_work); 5325 5326 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5327 if (err) 5328 goto failed_mount3; 5329 5330 err = ext4_es_register_shrinker(sbi); 5331 if (err) 5332 goto failed_mount3; 5333 5334 sbi->s_stripe = ext4_get_stripe_size(sbi); 5335 if (ext4_is_stripe_incompatible(sb, sbi->s_stripe)) { 5336 ext4_msg(sb, KERN_WARNING, 5337 "stripe (%lu) is not aligned with cluster size (%u), " 5338 "stripe is disabled", 5339 sbi->s_stripe, sbi->s_cluster_ratio); 5340 sbi->s_stripe = 0; 5341 } 5342 sbi->s_extent_max_zeroout_kb = 32; 5343 5344 /* 5345 * set up enough so that it can read an inode 5346 */ 5347 sb->s_op = &ext4_sops; 5348 sb->s_export_op = &ext4_export_ops; 5349 sb->s_xattr = ext4_xattr_handlers; 5350 #ifdef CONFIG_FS_ENCRYPTION 5351 sb->s_cop = &ext4_cryptops; 5352 #endif 5353 #ifdef CONFIG_FS_VERITY 5354 sb->s_vop = &ext4_verityops; 5355 #endif 5356 #ifdef CONFIG_QUOTA 5357 sb->dq_op = &ext4_quota_operations; 5358 if (ext4_has_feature_quota(sb)) 5359 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5360 else 5361 sb->s_qcop = &ext4_qctl_operations; 5362 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5363 #endif 5364 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 5365 5366 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5367 mutex_init(&sbi->s_orphan_lock); 5368 5369 ext4_fast_commit_init(sb); 5370 5371 sb->s_root = NULL; 5372 5373 needs_recovery = (es->s_last_orphan != 0 || 5374 ext4_has_feature_orphan_present(sb) || 5375 ext4_has_feature_journal_needs_recovery(sb)); 5376 5377 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) { 5378 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)); 5379 if (err) 5380 goto failed_mount3a; 5381 } 5382 5383 err = -EINVAL; 5384 /* 5385 * The first inode we look at is the journal inode. Don't try 5386 * root first: it may be modified in the journal! 5387 */ 5388 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5389 err = ext4_load_and_init_journal(sb, es, ctx); 5390 if (err) 5391 goto failed_mount3a; 5392 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5393 ext4_has_feature_journal_needs_recovery(sb)) { 5394 ext4_msg(sb, KERN_ERR, "required journal recovery " 5395 "suppressed and not mounted read-only"); 5396 goto failed_mount3a; 5397 } else { 5398 /* Nojournal mode, all journal mount options are illegal */ 5399 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5400 ext4_msg(sb, KERN_ERR, "can't mount with " 5401 "journal_async_commit, fs mounted w/o journal"); 5402 goto failed_mount3a; 5403 } 5404 5405 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 5406 ext4_msg(sb, KERN_ERR, "can't mount with " 5407 "journal_checksum, fs mounted w/o journal"); 5408 goto failed_mount3a; 5409 } 5410 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 5411 ext4_msg(sb, KERN_ERR, "can't mount with " 5412 "commit=%lu, fs mounted w/o journal", 5413 sbi->s_commit_interval / HZ); 5414 goto failed_mount3a; 5415 } 5416 if (EXT4_MOUNT_DATA_FLAGS & 5417 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 5418 ext4_msg(sb, KERN_ERR, "can't mount with " 5419 "data=, fs mounted w/o journal"); 5420 goto failed_mount3a; 5421 } 5422 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5423 clear_opt(sb, JOURNAL_CHECKSUM); 5424 clear_opt(sb, DATA_FLAGS); 5425 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5426 sbi->s_journal = NULL; 5427 needs_recovery = 0; 5428 } 5429 5430 if (!test_opt(sb, NO_MBCACHE)) { 5431 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5432 if (!sbi->s_ea_block_cache) { 5433 ext4_msg(sb, KERN_ERR, 5434 "Failed to create ea_block_cache"); 5435 err = -EINVAL; 5436 goto failed_mount_wq; 5437 } 5438 5439 if (ext4_has_feature_ea_inode(sb)) { 5440 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5441 if (!sbi->s_ea_inode_cache) { 5442 ext4_msg(sb, KERN_ERR, 5443 "Failed to create ea_inode_cache"); 5444 err = -EINVAL; 5445 goto failed_mount_wq; 5446 } 5447 } 5448 } 5449 5450 /* 5451 * Get the # of file system overhead blocks from the 5452 * superblock if present. 5453 */ 5454 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5455 /* ignore the precalculated value if it is ridiculous */ 5456 if (sbi->s_overhead > ext4_blocks_count(es)) 5457 sbi->s_overhead = 0; 5458 /* 5459 * If the bigalloc feature is not enabled recalculating the 5460 * overhead doesn't take long, so we might as well just redo 5461 * it to make sure we are using the correct value. 5462 */ 5463 if (!ext4_has_feature_bigalloc(sb)) 5464 sbi->s_overhead = 0; 5465 if (sbi->s_overhead == 0) { 5466 err = ext4_calculate_overhead(sb); 5467 if (err) 5468 goto failed_mount_wq; 5469 } 5470 5471 /* 5472 * The maximum number of concurrent works can be high and 5473 * concurrency isn't really necessary. Limit it to 1. 5474 */ 5475 EXT4_SB(sb)->rsv_conversion_wq = 5476 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5477 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5478 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5479 err = -ENOMEM; 5480 goto failed_mount4; 5481 } 5482 5483 /* 5484 * The jbd2_journal_load will have done any necessary log recovery, 5485 * so we can safely mount the rest of the filesystem now. 5486 */ 5487 5488 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5489 if (IS_ERR(root)) { 5490 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5491 err = PTR_ERR(root); 5492 root = NULL; 5493 goto failed_mount4; 5494 } 5495 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5496 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5497 iput(root); 5498 err = -EFSCORRUPTED; 5499 goto failed_mount4; 5500 } 5501 5502 sb->s_root = d_make_root(root); 5503 if (!sb->s_root) { 5504 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5505 err = -ENOMEM; 5506 goto failed_mount4; 5507 } 5508 5509 err = ext4_setup_super(sb, es, sb_rdonly(sb)); 5510 if (err == -EROFS) { 5511 sb->s_flags |= SB_RDONLY; 5512 } else if (err) 5513 goto failed_mount4a; 5514 5515 ext4_set_resv_clusters(sb); 5516 5517 if (test_opt(sb, BLOCK_VALIDITY)) { 5518 err = ext4_setup_system_zone(sb); 5519 if (err) { 5520 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5521 "zone (%d)", err); 5522 goto failed_mount4a; 5523 } 5524 } 5525 ext4_fc_replay_cleanup(sb); 5526 5527 ext4_ext_init(sb); 5528 5529 /* 5530 * Enable optimize_scan if number of groups is > threshold. This can be 5531 * turned off by passing "mb_optimize_scan=0". This can also be 5532 * turned on forcefully by passing "mb_optimize_scan=1". 5533 */ 5534 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5535 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5536 set_opt2(sb, MB_OPTIMIZE_SCAN); 5537 else 5538 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5539 } 5540 5541 err = ext4_mb_init(sb); 5542 if (err) { 5543 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5544 err); 5545 goto failed_mount5; 5546 } 5547 5548 /* 5549 * We can only set up the journal commit callback once 5550 * mballoc is initialized 5551 */ 5552 if (sbi->s_journal) 5553 sbi->s_journal->j_commit_callback = 5554 ext4_journal_commit_callback; 5555 5556 err = ext4_percpu_param_init(sbi); 5557 if (err) 5558 goto failed_mount6; 5559 5560 if (ext4_has_feature_flex_bg(sb)) 5561 if (!ext4_fill_flex_info(sb)) { 5562 ext4_msg(sb, KERN_ERR, 5563 "unable to initialize " 5564 "flex_bg meta info!"); 5565 err = -ENOMEM; 5566 goto failed_mount6; 5567 } 5568 5569 err = ext4_register_li_request(sb, first_not_zeroed); 5570 if (err) 5571 goto failed_mount6; 5572 5573 err = ext4_init_orphan_info(sb); 5574 if (err) 5575 goto failed_mount7; 5576 #ifdef CONFIG_QUOTA 5577 /* Enable quota usage during mount. */ 5578 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5579 err = ext4_enable_quotas(sb); 5580 if (err) 5581 goto failed_mount8; 5582 } 5583 #endif /* CONFIG_QUOTA */ 5584 5585 /* 5586 * Save the original bdev mapping's wb_err value which could be 5587 * used to detect the metadata async write error. 5588 */ 5589 spin_lock_init(&sbi->s_bdev_wb_lock); 5590 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5591 &sbi->s_bdev_wb_err); 5592 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5593 ext4_orphan_cleanup(sb, es); 5594 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5595 /* 5596 * Update the checksum after updating free space/inode counters and 5597 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5598 * checksum in the buffer cache until it is written out and 5599 * e2fsprogs programs trying to open a file system immediately 5600 * after it is mounted can fail. 5601 */ 5602 ext4_superblock_csum_set(sb); 5603 if (needs_recovery) { 5604 ext4_msg(sb, KERN_INFO, "recovery complete"); 5605 err = ext4_mark_recovery_complete(sb, es); 5606 if (err) 5607 goto failed_mount9; 5608 } 5609 5610 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) 5611 ext4_msg(sb, KERN_WARNING, 5612 "mounting with \"discard\" option, but the device does not support discard"); 5613 5614 if (es->s_error_count) 5615 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5616 5617 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5618 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5619 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5620 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5621 atomic_set(&sbi->s_warning_count, 0); 5622 atomic_set(&sbi->s_msg_count, 0); 5623 5624 /* Register sysfs after all initializations are complete. */ 5625 err = ext4_register_sysfs(sb); 5626 if (err) 5627 goto failed_mount9; 5628 5629 return 0; 5630 5631 failed_mount9: 5632 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 5633 failed_mount8: __maybe_unused 5634 ext4_release_orphan_info(sb); 5635 failed_mount7: 5636 ext4_unregister_li_request(sb); 5637 failed_mount6: 5638 ext4_mb_release(sb); 5639 ext4_flex_groups_free(sbi); 5640 ext4_percpu_param_destroy(sbi); 5641 failed_mount5: 5642 ext4_ext_release(sb); 5643 ext4_release_system_zone(sb); 5644 failed_mount4a: 5645 dput(sb->s_root); 5646 sb->s_root = NULL; 5647 failed_mount4: 5648 ext4_msg(sb, KERN_ERR, "mount failed"); 5649 if (EXT4_SB(sb)->rsv_conversion_wq) 5650 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5651 failed_mount_wq: 5652 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5653 sbi->s_ea_inode_cache = NULL; 5654 5655 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5656 sbi->s_ea_block_cache = NULL; 5657 5658 if (sbi->s_journal) { 5659 /* flush s_sb_upd_work before journal destroy. */ 5660 flush_work(&sbi->s_sb_upd_work); 5661 jbd2_journal_destroy(sbi->s_journal); 5662 sbi->s_journal = NULL; 5663 } 5664 failed_mount3a: 5665 ext4_es_unregister_shrinker(sbi); 5666 failed_mount3: 5667 /* flush s_sb_upd_work before sbi destroy */ 5668 flush_work(&sbi->s_sb_upd_work); 5669 ext4_stop_mmpd(sbi); 5670 del_timer_sync(&sbi->s_err_report); 5671 ext4_group_desc_free(sbi); 5672 failed_mount: 5673 if (sbi->s_chksum_driver) 5674 crypto_free_shash(sbi->s_chksum_driver); 5675 5676 #if IS_ENABLED(CONFIG_UNICODE) 5677 utf8_unload(sb->s_encoding); 5678 #endif 5679 5680 #ifdef CONFIG_QUOTA 5681 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++) 5682 kfree(get_qf_name(sb, sbi, i)); 5683 #endif 5684 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5685 brelse(sbi->s_sbh); 5686 if (sbi->s_journal_bdev) { 5687 invalidate_bdev(sbi->s_journal_bdev); 5688 blkdev_put(sbi->s_journal_bdev, sb); 5689 } 5690 out_fail: 5691 invalidate_bdev(sb->s_bdev); 5692 sb->s_fs_info = NULL; 5693 return err; 5694 } 5695 5696 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5697 { 5698 struct ext4_fs_context *ctx = fc->fs_private; 5699 struct ext4_sb_info *sbi; 5700 const char *descr; 5701 int ret; 5702 5703 sbi = ext4_alloc_sbi(sb); 5704 if (!sbi) 5705 return -ENOMEM; 5706 5707 fc->s_fs_info = sbi; 5708 5709 /* Cleanup superblock name */ 5710 strreplace(sb->s_id, '/', '!'); 5711 5712 sbi->s_sb_block = 1; /* Default super block location */ 5713 if (ctx->spec & EXT4_SPEC_s_sb_block) 5714 sbi->s_sb_block = ctx->s_sb_block; 5715 5716 ret = __ext4_fill_super(fc, sb); 5717 if (ret < 0) 5718 goto free_sbi; 5719 5720 if (sbi->s_journal) { 5721 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5722 descr = " journalled data mode"; 5723 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5724 descr = " ordered data mode"; 5725 else 5726 descr = " writeback data mode"; 5727 } else 5728 descr = "out journal"; 5729 5730 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5731 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. " 5732 "Quota mode: %s.", &sb->s_uuid, 5733 sb_rdonly(sb) ? "ro" : "r/w", descr, 5734 ext4_quota_mode(sb)); 5735 5736 /* Update the s_overhead_clusters if necessary */ 5737 ext4_update_overhead(sb, false); 5738 return 0; 5739 5740 free_sbi: 5741 ext4_free_sbi(sbi); 5742 fc->s_fs_info = NULL; 5743 return ret; 5744 } 5745 5746 static int ext4_get_tree(struct fs_context *fc) 5747 { 5748 return get_tree_bdev(fc, ext4_fill_super); 5749 } 5750 5751 /* 5752 * Setup any per-fs journal parameters now. We'll do this both on 5753 * initial mount, once the journal has been initialised but before we've 5754 * done any recovery; and again on any subsequent remount. 5755 */ 5756 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5757 { 5758 struct ext4_sb_info *sbi = EXT4_SB(sb); 5759 5760 journal->j_commit_interval = sbi->s_commit_interval; 5761 journal->j_min_batch_time = sbi->s_min_batch_time; 5762 journal->j_max_batch_time = sbi->s_max_batch_time; 5763 ext4_fc_init(sb, journal); 5764 5765 write_lock(&journal->j_state_lock); 5766 if (test_opt(sb, BARRIER)) 5767 journal->j_flags |= JBD2_BARRIER; 5768 else 5769 journal->j_flags &= ~JBD2_BARRIER; 5770 if (test_opt(sb, DATA_ERR_ABORT)) 5771 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5772 else 5773 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5774 /* 5775 * Always enable journal cycle record option, letting the journal 5776 * records log transactions continuously between each mount. 5777 */ 5778 journal->j_flags |= JBD2_CYCLE_RECORD; 5779 write_unlock(&journal->j_state_lock); 5780 } 5781 5782 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5783 unsigned int journal_inum) 5784 { 5785 struct inode *journal_inode; 5786 5787 /* 5788 * Test for the existence of a valid inode on disk. Bad things 5789 * happen if we iget() an unused inode, as the subsequent iput() 5790 * will try to delete it. 5791 */ 5792 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5793 if (IS_ERR(journal_inode)) { 5794 ext4_msg(sb, KERN_ERR, "no journal found"); 5795 return ERR_CAST(journal_inode); 5796 } 5797 if (!journal_inode->i_nlink) { 5798 make_bad_inode(journal_inode); 5799 iput(journal_inode); 5800 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5801 return ERR_PTR(-EFSCORRUPTED); 5802 } 5803 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) { 5804 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5805 iput(journal_inode); 5806 return ERR_PTR(-EFSCORRUPTED); 5807 } 5808 5809 ext4_debug("Journal inode found at %p: %lld bytes\n", 5810 journal_inode, journal_inode->i_size); 5811 return journal_inode; 5812 } 5813 5814 static int ext4_journal_bmap(journal_t *journal, sector_t *block) 5815 { 5816 struct ext4_map_blocks map; 5817 int ret; 5818 5819 if (journal->j_inode == NULL) 5820 return 0; 5821 5822 map.m_lblk = *block; 5823 map.m_len = 1; 5824 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0); 5825 if (ret <= 0) { 5826 ext4_msg(journal->j_inode->i_sb, KERN_CRIT, 5827 "journal bmap failed: block %llu ret %d\n", 5828 *block, ret); 5829 jbd2_journal_abort(journal, ret ? ret : -EIO); 5830 return ret; 5831 } 5832 *block = map.m_pblk; 5833 return 0; 5834 } 5835 5836 static journal_t *ext4_open_inode_journal(struct super_block *sb, 5837 unsigned int journal_inum) 5838 { 5839 struct inode *journal_inode; 5840 journal_t *journal; 5841 5842 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5843 if (IS_ERR(journal_inode)) 5844 return ERR_CAST(journal_inode); 5845 5846 journal = jbd2_journal_init_inode(journal_inode); 5847 if (IS_ERR(journal)) { 5848 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5849 iput(journal_inode); 5850 return ERR_CAST(journal); 5851 } 5852 journal->j_private = sb; 5853 journal->j_bmap = ext4_journal_bmap; 5854 ext4_init_journal_params(sb, journal); 5855 return journal; 5856 } 5857 5858 static struct block_device *ext4_get_journal_blkdev(struct super_block *sb, 5859 dev_t j_dev, ext4_fsblk_t *j_start, 5860 ext4_fsblk_t *j_len) 5861 { 5862 struct buffer_head *bh; 5863 struct block_device *bdev; 5864 int hblock, blocksize; 5865 ext4_fsblk_t sb_block; 5866 unsigned long offset; 5867 struct ext4_super_block *es; 5868 int errno; 5869 5870 /* see get_tree_bdev why this is needed and safe */ 5871 up_write(&sb->s_umount); 5872 bdev = blkdev_get_by_dev(j_dev, BLK_OPEN_READ | BLK_OPEN_WRITE, sb, 5873 &fs_holder_ops); 5874 down_write(&sb->s_umount); 5875 if (IS_ERR(bdev)) { 5876 ext4_msg(sb, KERN_ERR, 5877 "failed to open journal device unknown-block(%u,%u) %ld", 5878 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev)); 5879 return ERR_CAST(bdev); 5880 } 5881 5882 blocksize = sb->s_blocksize; 5883 hblock = bdev_logical_block_size(bdev); 5884 if (blocksize < hblock) { 5885 ext4_msg(sb, KERN_ERR, 5886 "blocksize too small for journal device"); 5887 errno = -EINVAL; 5888 goto out_bdev; 5889 } 5890 5891 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5892 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5893 set_blocksize(bdev, blocksize); 5894 bh = __bread(bdev, sb_block, blocksize); 5895 if (!bh) { 5896 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5897 "external journal"); 5898 errno = -EINVAL; 5899 goto out_bdev; 5900 } 5901 5902 es = (struct ext4_super_block *) (bh->b_data + offset); 5903 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5904 !(le32_to_cpu(es->s_feature_incompat) & 5905 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5906 ext4_msg(sb, KERN_ERR, "external journal has bad superblock"); 5907 errno = -EFSCORRUPTED; 5908 goto out_bh; 5909 } 5910 5911 if ((le32_to_cpu(es->s_feature_ro_compat) & 5912 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5913 es->s_checksum != ext4_superblock_csum(sb, es)) { 5914 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock"); 5915 errno = -EFSCORRUPTED; 5916 goto out_bh; 5917 } 5918 5919 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5920 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5921 errno = -EFSCORRUPTED; 5922 goto out_bh; 5923 } 5924 5925 *j_start = sb_block + 1; 5926 *j_len = ext4_blocks_count(es); 5927 brelse(bh); 5928 return bdev; 5929 5930 out_bh: 5931 brelse(bh); 5932 out_bdev: 5933 blkdev_put(bdev, sb); 5934 return ERR_PTR(errno); 5935 } 5936 5937 static journal_t *ext4_open_dev_journal(struct super_block *sb, 5938 dev_t j_dev) 5939 { 5940 journal_t *journal; 5941 ext4_fsblk_t j_start; 5942 ext4_fsblk_t j_len; 5943 struct block_device *journal_bdev; 5944 int errno = 0; 5945 5946 journal_bdev = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len); 5947 if (IS_ERR(journal_bdev)) 5948 return ERR_CAST(journal_bdev); 5949 5950 journal = jbd2_journal_init_dev(journal_bdev, sb->s_bdev, j_start, 5951 j_len, sb->s_blocksize); 5952 if (IS_ERR(journal)) { 5953 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5954 errno = PTR_ERR(journal); 5955 goto out_bdev; 5956 } 5957 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5958 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5959 "user (unsupported) - %d", 5960 be32_to_cpu(journal->j_superblock->s_nr_users)); 5961 errno = -EINVAL; 5962 goto out_journal; 5963 } 5964 journal->j_private = sb; 5965 EXT4_SB(sb)->s_journal_bdev = journal_bdev; 5966 ext4_init_journal_params(sb, journal); 5967 return journal; 5968 5969 out_journal: 5970 jbd2_journal_destroy(journal); 5971 out_bdev: 5972 blkdev_put(journal_bdev, sb); 5973 return ERR_PTR(errno); 5974 } 5975 5976 static int ext4_load_journal(struct super_block *sb, 5977 struct ext4_super_block *es, 5978 unsigned long journal_devnum) 5979 { 5980 journal_t *journal; 5981 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5982 dev_t journal_dev; 5983 int err = 0; 5984 int really_read_only; 5985 int journal_dev_ro; 5986 5987 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5988 return -EFSCORRUPTED; 5989 5990 if (journal_devnum && 5991 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5992 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5993 "numbers have changed"); 5994 journal_dev = new_decode_dev(journal_devnum); 5995 } else 5996 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5997 5998 if (journal_inum && journal_dev) { 5999 ext4_msg(sb, KERN_ERR, 6000 "filesystem has both journal inode and journal device!"); 6001 return -EINVAL; 6002 } 6003 6004 if (journal_inum) { 6005 journal = ext4_open_inode_journal(sb, journal_inum); 6006 if (IS_ERR(journal)) 6007 return PTR_ERR(journal); 6008 } else { 6009 journal = ext4_open_dev_journal(sb, journal_dev); 6010 if (IS_ERR(journal)) 6011 return PTR_ERR(journal); 6012 } 6013 6014 journal_dev_ro = bdev_read_only(journal->j_dev); 6015 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 6016 6017 if (journal_dev_ro && !sb_rdonly(sb)) { 6018 ext4_msg(sb, KERN_ERR, 6019 "journal device read-only, try mounting with '-o ro'"); 6020 err = -EROFS; 6021 goto err_out; 6022 } 6023 6024 /* 6025 * Are we loading a blank journal or performing recovery after a 6026 * crash? For recovery, we need to check in advance whether we 6027 * can get read-write access to the device. 6028 */ 6029 if (ext4_has_feature_journal_needs_recovery(sb)) { 6030 if (sb_rdonly(sb)) { 6031 ext4_msg(sb, KERN_INFO, "INFO: recovery " 6032 "required on readonly filesystem"); 6033 if (really_read_only) { 6034 ext4_msg(sb, KERN_ERR, "write access " 6035 "unavailable, cannot proceed " 6036 "(try mounting with noload)"); 6037 err = -EROFS; 6038 goto err_out; 6039 } 6040 ext4_msg(sb, KERN_INFO, "write access will " 6041 "be enabled during recovery"); 6042 } 6043 } 6044 6045 if (!(journal->j_flags & JBD2_BARRIER)) 6046 ext4_msg(sb, KERN_INFO, "barriers disabled"); 6047 6048 if (!ext4_has_feature_journal_needs_recovery(sb)) 6049 err = jbd2_journal_wipe(journal, !really_read_only); 6050 if (!err) { 6051 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 6052 __le16 orig_state; 6053 bool changed = false; 6054 6055 if (save) 6056 memcpy(save, ((char *) es) + 6057 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 6058 err = jbd2_journal_load(journal); 6059 if (save && memcmp(((char *) es) + EXT4_S_ERR_START, 6060 save, EXT4_S_ERR_LEN)) { 6061 memcpy(((char *) es) + EXT4_S_ERR_START, 6062 save, EXT4_S_ERR_LEN); 6063 changed = true; 6064 } 6065 kfree(save); 6066 orig_state = es->s_state; 6067 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state & 6068 EXT4_ERROR_FS); 6069 if (orig_state != es->s_state) 6070 changed = true; 6071 /* Write out restored error information to the superblock */ 6072 if (changed && !really_read_only) { 6073 int err2; 6074 err2 = ext4_commit_super(sb); 6075 err = err ? : err2; 6076 } 6077 } 6078 6079 if (err) { 6080 ext4_msg(sb, KERN_ERR, "error loading journal"); 6081 goto err_out; 6082 } 6083 6084 EXT4_SB(sb)->s_journal = journal; 6085 err = ext4_clear_journal_err(sb, es); 6086 if (err) { 6087 EXT4_SB(sb)->s_journal = NULL; 6088 jbd2_journal_destroy(journal); 6089 return err; 6090 } 6091 6092 if (!really_read_only && journal_devnum && 6093 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6094 es->s_journal_dev = cpu_to_le32(journal_devnum); 6095 ext4_commit_super(sb); 6096 } 6097 if (!really_read_only && journal_inum && 6098 journal_inum != le32_to_cpu(es->s_journal_inum)) { 6099 es->s_journal_inum = cpu_to_le32(journal_inum); 6100 ext4_commit_super(sb); 6101 } 6102 6103 return 0; 6104 6105 err_out: 6106 jbd2_journal_destroy(journal); 6107 return err; 6108 } 6109 6110 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 6111 static void ext4_update_super(struct super_block *sb) 6112 { 6113 struct ext4_sb_info *sbi = EXT4_SB(sb); 6114 struct ext4_super_block *es = sbi->s_es; 6115 struct buffer_head *sbh = sbi->s_sbh; 6116 6117 lock_buffer(sbh); 6118 /* 6119 * If the file system is mounted read-only, don't update the 6120 * superblock write time. This avoids updating the superblock 6121 * write time when we are mounting the root file system 6122 * read/only but we need to replay the journal; at that point, 6123 * for people who are east of GMT and who make their clock 6124 * tick in localtime for Windows bug-for-bug compatibility, 6125 * the clock is set in the future, and this will cause e2fsck 6126 * to complain and force a full file system check. 6127 */ 6128 if (!sb_rdonly(sb)) 6129 ext4_update_tstamp(es, s_wtime); 6130 es->s_kbytes_written = 6131 cpu_to_le64(sbi->s_kbytes_written + 6132 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 6133 sbi->s_sectors_written_start) >> 1)); 6134 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 6135 ext4_free_blocks_count_set(es, 6136 EXT4_C2B(sbi, percpu_counter_sum_positive( 6137 &sbi->s_freeclusters_counter))); 6138 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 6139 es->s_free_inodes_count = 6140 cpu_to_le32(percpu_counter_sum_positive( 6141 &sbi->s_freeinodes_counter)); 6142 /* Copy error information to the on-disk superblock */ 6143 spin_lock(&sbi->s_error_lock); 6144 if (sbi->s_add_error_count > 0) { 6145 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6146 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6147 __ext4_update_tstamp(&es->s_first_error_time, 6148 &es->s_first_error_time_hi, 6149 sbi->s_first_error_time); 6150 strncpy(es->s_first_error_func, sbi->s_first_error_func, 6151 sizeof(es->s_first_error_func)); 6152 es->s_first_error_line = 6153 cpu_to_le32(sbi->s_first_error_line); 6154 es->s_first_error_ino = 6155 cpu_to_le32(sbi->s_first_error_ino); 6156 es->s_first_error_block = 6157 cpu_to_le64(sbi->s_first_error_block); 6158 es->s_first_error_errcode = 6159 ext4_errno_to_code(sbi->s_first_error_code); 6160 } 6161 __ext4_update_tstamp(&es->s_last_error_time, 6162 &es->s_last_error_time_hi, 6163 sbi->s_last_error_time); 6164 strncpy(es->s_last_error_func, sbi->s_last_error_func, 6165 sizeof(es->s_last_error_func)); 6166 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6167 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6168 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6169 es->s_last_error_errcode = 6170 ext4_errno_to_code(sbi->s_last_error_code); 6171 /* 6172 * Start the daily error reporting function if it hasn't been 6173 * started already 6174 */ 6175 if (!es->s_error_count) 6176 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6177 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6178 sbi->s_add_error_count = 0; 6179 } 6180 spin_unlock(&sbi->s_error_lock); 6181 6182 ext4_superblock_csum_set(sb); 6183 unlock_buffer(sbh); 6184 } 6185 6186 static int ext4_commit_super(struct super_block *sb) 6187 { 6188 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6189 6190 if (!sbh) 6191 return -EINVAL; 6192 if (block_device_ejected(sb)) 6193 return -ENODEV; 6194 6195 ext4_update_super(sb); 6196 6197 lock_buffer(sbh); 6198 /* Buffer got discarded which means block device got invalidated */ 6199 if (!buffer_mapped(sbh)) { 6200 unlock_buffer(sbh); 6201 return -EIO; 6202 } 6203 6204 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6205 /* 6206 * Oh, dear. A previous attempt to write the 6207 * superblock failed. This could happen because the 6208 * USB device was yanked out. Or it could happen to 6209 * be a transient write error and maybe the block will 6210 * be remapped. Nothing we can do but to retry the 6211 * write and hope for the best. 6212 */ 6213 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6214 "superblock detected"); 6215 clear_buffer_write_io_error(sbh); 6216 set_buffer_uptodate(sbh); 6217 } 6218 get_bh(sbh); 6219 /* Clear potential dirty bit if it was journalled update */ 6220 clear_buffer_dirty(sbh); 6221 sbh->b_end_io = end_buffer_write_sync; 6222 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6223 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6224 wait_on_buffer(sbh); 6225 if (buffer_write_io_error(sbh)) { 6226 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6227 "superblock"); 6228 clear_buffer_write_io_error(sbh); 6229 set_buffer_uptodate(sbh); 6230 return -EIO; 6231 } 6232 return 0; 6233 } 6234 6235 /* 6236 * Have we just finished recovery? If so, and if we are mounting (or 6237 * remounting) the filesystem readonly, then we will end up with a 6238 * consistent fs on disk. Record that fact. 6239 */ 6240 static int ext4_mark_recovery_complete(struct super_block *sb, 6241 struct ext4_super_block *es) 6242 { 6243 int err; 6244 journal_t *journal = EXT4_SB(sb)->s_journal; 6245 6246 if (!ext4_has_feature_journal(sb)) { 6247 if (journal != NULL) { 6248 ext4_error(sb, "Journal got removed while the fs was " 6249 "mounted!"); 6250 return -EFSCORRUPTED; 6251 } 6252 return 0; 6253 } 6254 jbd2_journal_lock_updates(journal); 6255 err = jbd2_journal_flush(journal, 0); 6256 if (err < 0) 6257 goto out; 6258 6259 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6260 ext4_has_feature_orphan_present(sb))) { 6261 if (!ext4_orphan_file_empty(sb)) { 6262 ext4_error(sb, "Orphan file not empty on read-only fs."); 6263 err = -EFSCORRUPTED; 6264 goto out; 6265 } 6266 ext4_clear_feature_journal_needs_recovery(sb); 6267 ext4_clear_feature_orphan_present(sb); 6268 ext4_commit_super(sb); 6269 } 6270 out: 6271 jbd2_journal_unlock_updates(journal); 6272 return err; 6273 } 6274 6275 /* 6276 * If we are mounting (or read-write remounting) a filesystem whose journal 6277 * has recorded an error from a previous lifetime, move that error to the 6278 * main filesystem now. 6279 */ 6280 static int ext4_clear_journal_err(struct super_block *sb, 6281 struct ext4_super_block *es) 6282 { 6283 journal_t *journal; 6284 int j_errno; 6285 const char *errstr; 6286 6287 if (!ext4_has_feature_journal(sb)) { 6288 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6289 return -EFSCORRUPTED; 6290 } 6291 6292 journal = EXT4_SB(sb)->s_journal; 6293 6294 /* 6295 * Now check for any error status which may have been recorded in the 6296 * journal by a prior ext4_error() or ext4_abort() 6297 */ 6298 6299 j_errno = jbd2_journal_errno(journal); 6300 if (j_errno) { 6301 char nbuf[16]; 6302 6303 errstr = ext4_decode_error(sb, j_errno, nbuf); 6304 ext4_warning(sb, "Filesystem error recorded " 6305 "from previous mount: %s", errstr); 6306 6307 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6308 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6309 j_errno = ext4_commit_super(sb); 6310 if (j_errno) 6311 return j_errno; 6312 ext4_warning(sb, "Marked fs in need of filesystem check."); 6313 6314 jbd2_journal_clear_err(journal); 6315 jbd2_journal_update_sb_errno(journal); 6316 } 6317 return 0; 6318 } 6319 6320 /* 6321 * Force the running and committing transactions to commit, 6322 * and wait on the commit. 6323 */ 6324 int ext4_force_commit(struct super_block *sb) 6325 { 6326 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal); 6327 } 6328 6329 static int ext4_sync_fs(struct super_block *sb, int wait) 6330 { 6331 int ret = 0; 6332 tid_t target; 6333 bool needs_barrier = false; 6334 struct ext4_sb_info *sbi = EXT4_SB(sb); 6335 6336 if (unlikely(ext4_forced_shutdown(sb))) 6337 return 0; 6338 6339 trace_ext4_sync_fs(sb, wait); 6340 flush_workqueue(sbi->rsv_conversion_wq); 6341 /* 6342 * Writeback quota in non-journalled quota case - journalled quota has 6343 * no dirty dquots 6344 */ 6345 dquot_writeback_dquots(sb, -1); 6346 /* 6347 * Data writeback is possible w/o journal transaction, so barrier must 6348 * being sent at the end of the function. But we can skip it if 6349 * transaction_commit will do it for us. 6350 */ 6351 if (sbi->s_journal) { 6352 target = jbd2_get_latest_transaction(sbi->s_journal); 6353 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6354 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6355 needs_barrier = true; 6356 6357 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6358 if (wait) 6359 ret = jbd2_log_wait_commit(sbi->s_journal, 6360 target); 6361 } 6362 } else if (wait && test_opt(sb, BARRIER)) 6363 needs_barrier = true; 6364 if (needs_barrier) { 6365 int err; 6366 err = blkdev_issue_flush(sb->s_bdev); 6367 if (!ret) 6368 ret = err; 6369 } 6370 6371 return ret; 6372 } 6373 6374 /* 6375 * LVM calls this function before a (read-only) snapshot is created. This 6376 * gives us a chance to flush the journal completely and mark the fs clean. 6377 * 6378 * Note that only this function cannot bring a filesystem to be in a clean 6379 * state independently. It relies on upper layer to stop all data & metadata 6380 * modifications. 6381 */ 6382 static int ext4_freeze(struct super_block *sb) 6383 { 6384 int error = 0; 6385 journal_t *journal = EXT4_SB(sb)->s_journal; 6386 6387 if (journal) { 6388 /* Now we set up the journal barrier. */ 6389 jbd2_journal_lock_updates(journal); 6390 6391 /* 6392 * Don't clear the needs_recovery flag if we failed to 6393 * flush the journal. 6394 */ 6395 error = jbd2_journal_flush(journal, 0); 6396 if (error < 0) 6397 goto out; 6398 6399 /* Journal blocked and flushed, clear needs_recovery flag. */ 6400 ext4_clear_feature_journal_needs_recovery(sb); 6401 if (ext4_orphan_file_empty(sb)) 6402 ext4_clear_feature_orphan_present(sb); 6403 } 6404 6405 error = ext4_commit_super(sb); 6406 out: 6407 if (journal) 6408 /* we rely on upper layer to stop further updates */ 6409 jbd2_journal_unlock_updates(journal); 6410 return error; 6411 } 6412 6413 /* 6414 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6415 * flag here, even though the filesystem is not technically dirty yet. 6416 */ 6417 static int ext4_unfreeze(struct super_block *sb) 6418 { 6419 if (ext4_forced_shutdown(sb)) 6420 return 0; 6421 6422 if (EXT4_SB(sb)->s_journal) { 6423 /* Reset the needs_recovery flag before the fs is unlocked. */ 6424 ext4_set_feature_journal_needs_recovery(sb); 6425 if (ext4_has_feature_orphan_file(sb)) 6426 ext4_set_feature_orphan_present(sb); 6427 } 6428 6429 ext4_commit_super(sb); 6430 return 0; 6431 } 6432 6433 /* 6434 * Structure to save mount options for ext4_remount's benefit 6435 */ 6436 struct ext4_mount_options { 6437 unsigned long s_mount_opt; 6438 unsigned long s_mount_opt2; 6439 kuid_t s_resuid; 6440 kgid_t s_resgid; 6441 unsigned long s_commit_interval; 6442 u32 s_min_batch_time, s_max_batch_time; 6443 #ifdef CONFIG_QUOTA 6444 int s_jquota_fmt; 6445 char *s_qf_names[EXT4_MAXQUOTAS]; 6446 #endif 6447 }; 6448 6449 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6450 { 6451 struct ext4_fs_context *ctx = fc->fs_private; 6452 struct ext4_super_block *es; 6453 struct ext4_sb_info *sbi = EXT4_SB(sb); 6454 unsigned long old_sb_flags; 6455 struct ext4_mount_options old_opts; 6456 ext4_group_t g; 6457 int err = 0; 6458 int alloc_ctx; 6459 #ifdef CONFIG_QUOTA 6460 int enable_quota = 0; 6461 int i, j; 6462 char *to_free[EXT4_MAXQUOTAS]; 6463 #endif 6464 6465 6466 /* Store the original options */ 6467 old_sb_flags = sb->s_flags; 6468 old_opts.s_mount_opt = sbi->s_mount_opt; 6469 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6470 old_opts.s_resuid = sbi->s_resuid; 6471 old_opts.s_resgid = sbi->s_resgid; 6472 old_opts.s_commit_interval = sbi->s_commit_interval; 6473 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6474 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6475 #ifdef CONFIG_QUOTA 6476 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6477 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6478 if (sbi->s_qf_names[i]) { 6479 char *qf_name = get_qf_name(sb, sbi, i); 6480 6481 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6482 if (!old_opts.s_qf_names[i]) { 6483 for (j = 0; j < i; j++) 6484 kfree(old_opts.s_qf_names[j]); 6485 return -ENOMEM; 6486 } 6487 } else 6488 old_opts.s_qf_names[i] = NULL; 6489 #endif 6490 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6491 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6492 ctx->journal_ioprio = 6493 sbi->s_journal->j_task->io_context->ioprio; 6494 else 6495 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 6496 6497 } 6498 6499 if ((ctx->spec & EXT4_SPEC_s_stripe) && 6500 ext4_is_stripe_incompatible(sb, ctx->s_stripe)) { 6501 ext4_msg(sb, KERN_WARNING, 6502 "stripe (%lu) is not aligned with cluster size (%u), " 6503 "stripe is disabled", 6504 ctx->s_stripe, sbi->s_cluster_ratio); 6505 ctx->s_stripe = 0; 6506 } 6507 6508 /* 6509 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause 6510 * two calls to ext4_should_dioread_nolock() to return inconsistent 6511 * values, triggering WARN_ON in ext4_add_complete_io(). we grab 6512 * here s_writepages_rwsem to avoid race between writepages ops and 6513 * remount. 6514 */ 6515 alloc_ctx = ext4_writepages_down_write(sb); 6516 ext4_apply_options(fc, sb); 6517 ext4_writepages_up_write(sb, alloc_ctx); 6518 6519 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6520 test_opt(sb, JOURNAL_CHECKSUM)) { 6521 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6522 "during remount not supported; ignoring"); 6523 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6524 } 6525 6526 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6527 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6528 ext4_msg(sb, KERN_ERR, "can't mount with " 6529 "both data=journal and delalloc"); 6530 err = -EINVAL; 6531 goto restore_opts; 6532 } 6533 if (test_opt(sb, DIOREAD_NOLOCK)) { 6534 ext4_msg(sb, KERN_ERR, "can't mount with " 6535 "both data=journal and dioread_nolock"); 6536 err = -EINVAL; 6537 goto restore_opts; 6538 } 6539 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6540 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6541 ext4_msg(sb, KERN_ERR, "can't mount with " 6542 "journal_async_commit in data=ordered mode"); 6543 err = -EINVAL; 6544 goto restore_opts; 6545 } 6546 } 6547 6548 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6549 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6550 err = -EINVAL; 6551 goto restore_opts; 6552 } 6553 6554 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6555 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6556 6557 es = sbi->s_es; 6558 6559 if (sbi->s_journal) { 6560 ext4_init_journal_params(sb, sbi->s_journal); 6561 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6562 } 6563 6564 /* Flush outstanding errors before changing fs state */ 6565 flush_work(&sbi->s_sb_upd_work); 6566 6567 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6568 if (ext4_forced_shutdown(sb)) { 6569 err = -EROFS; 6570 goto restore_opts; 6571 } 6572 6573 if (fc->sb_flags & SB_RDONLY) { 6574 err = sync_filesystem(sb); 6575 if (err < 0) 6576 goto restore_opts; 6577 err = dquot_suspend(sb, -1); 6578 if (err < 0) 6579 goto restore_opts; 6580 6581 /* 6582 * First of all, the unconditional stuff we have to do 6583 * to disable replay of the journal when we next remount 6584 */ 6585 sb->s_flags |= SB_RDONLY; 6586 6587 /* 6588 * OK, test if we are remounting a valid rw partition 6589 * readonly, and if so set the rdonly flag and then 6590 * mark the partition as valid again. 6591 */ 6592 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6593 (sbi->s_mount_state & EXT4_VALID_FS)) 6594 es->s_state = cpu_to_le16(sbi->s_mount_state); 6595 6596 if (sbi->s_journal) { 6597 /* 6598 * We let remount-ro finish even if marking fs 6599 * as clean failed... 6600 */ 6601 ext4_mark_recovery_complete(sb, es); 6602 } 6603 } else { 6604 /* Make sure we can mount this feature set readwrite */ 6605 if (ext4_has_feature_readonly(sb) || 6606 !ext4_feature_set_ok(sb, 0)) { 6607 err = -EROFS; 6608 goto restore_opts; 6609 } 6610 /* 6611 * Make sure the group descriptor checksums 6612 * are sane. If they aren't, refuse to remount r/w. 6613 */ 6614 for (g = 0; g < sbi->s_groups_count; g++) { 6615 struct ext4_group_desc *gdp = 6616 ext4_get_group_desc(sb, g, NULL); 6617 6618 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6619 ext4_msg(sb, KERN_ERR, 6620 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6621 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6622 le16_to_cpu(gdp->bg_checksum)); 6623 err = -EFSBADCRC; 6624 goto restore_opts; 6625 } 6626 } 6627 6628 /* 6629 * If we have an unprocessed orphan list hanging 6630 * around from a previously readonly bdev mount, 6631 * require a full umount/remount for now. 6632 */ 6633 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6634 ext4_msg(sb, KERN_WARNING, "Couldn't " 6635 "remount RDWR because of unprocessed " 6636 "orphan inode list. Please " 6637 "umount/remount instead"); 6638 err = -EINVAL; 6639 goto restore_opts; 6640 } 6641 6642 /* 6643 * Mounting a RDONLY partition read-write, so reread 6644 * and store the current valid flag. (It may have 6645 * been changed by e2fsck since we originally mounted 6646 * the partition.) 6647 */ 6648 if (sbi->s_journal) { 6649 err = ext4_clear_journal_err(sb, es); 6650 if (err) 6651 goto restore_opts; 6652 } 6653 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6654 ~EXT4_FC_REPLAY); 6655 6656 err = ext4_setup_super(sb, es, 0); 6657 if (err) 6658 goto restore_opts; 6659 6660 sb->s_flags &= ~SB_RDONLY; 6661 if (ext4_has_feature_mmp(sb)) { 6662 err = ext4_multi_mount_protect(sb, 6663 le64_to_cpu(es->s_mmp_block)); 6664 if (err) 6665 goto restore_opts; 6666 } 6667 #ifdef CONFIG_QUOTA 6668 enable_quota = 1; 6669 #endif 6670 } 6671 } 6672 6673 /* 6674 * Handle creation of system zone data early because it can fail. 6675 * Releasing of existing data is done when we are sure remount will 6676 * succeed. 6677 */ 6678 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6679 err = ext4_setup_system_zone(sb); 6680 if (err) 6681 goto restore_opts; 6682 } 6683 6684 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6685 err = ext4_commit_super(sb); 6686 if (err) 6687 goto restore_opts; 6688 } 6689 6690 #ifdef CONFIG_QUOTA 6691 if (enable_quota) { 6692 if (sb_any_quota_suspended(sb)) 6693 dquot_resume(sb, -1); 6694 else if (ext4_has_feature_quota(sb)) { 6695 err = ext4_enable_quotas(sb); 6696 if (err) 6697 goto restore_opts; 6698 } 6699 } 6700 /* Release old quota file names */ 6701 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6702 kfree(old_opts.s_qf_names[i]); 6703 #endif 6704 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6705 ext4_release_system_zone(sb); 6706 6707 /* 6708 * Reinitialize lazy itable initialization thread based on 6709 * current settings 6710 */ 6711 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6712 ext4_unregister_li_request(sb); 6713 else { 6714 ext4_group_t first_not_zeroed; 6715 first_not_zeroed = ext4_has_uninit_itable(sb); 6716 ext4_register_li_request(sb, first_not_zeroed); 6717 } 6718 6719 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6720 ext4_stop_mmpd(sbi); 6721 6722 /* 6723 * Handle aborting the filesystem as the last thing during remount to 6724 * avoid obsure errors during remount when some option changes fail to 6725 * apply due to shutdown filesystem. 6726 */ 6727 if (test_opt2(sb, ABORT)) 6728 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6729 6730 return 0; 6731 6732 restore_opts: 6733 /* 6734 * If there was a failing r/w to ro transition, we may need to 6735 * re-enable quota 6736 */ 6737 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) && 6738 sb_any_quota_suspended(sb)) 6739 dquot_resume(sb, -1); 6740 6741 alloc_ctx = ext4_writepages_down_write(sb); 6742 sb->s_flags = old_sb_flags; 6743 sbi->s_mount_opt = old_opts.s_mount_opt; 6744 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6745 sbi->s_resuid = old_opts.s_resuid; 6746 sbi->s_resgid = old_opts.s_resgid; 6747 sbi->s_commit_interval = old_opts.s_commit_interval; 6748 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6749 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6750 ext4_writepages_up_write(sb, alloc_ctx); 6751 6752 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6753 ext4_release_system_zone(sb); 6754 #ifdef CONFIG_QUOTA 6755 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6756 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6757 to_free[i] = get_qf_name(sb, sbi, i); 6758 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6759 } 6760 synchronize_rcu(); 6761 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6762 kfree(to_free[i]); 6763 #endif 6764 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6765 ext4_stop_mmpd(sbi); 6766 return err; 6767 } 6768 6769 static int ext4_reconfigure(struct fs_context *fc) 6770 { 6771 struct super_block *sb = fc->root->d_sb; 6772 int ret; 6773 6774 fc->s_fs_info = EXT4_SB(sb); 6775 6776 ret = ext4_check_opt_consistency(fc, sb); 6777 if (ret < 0) 6778 return ret; 6779 6780 ret = __ext4_remount(fc, sb); 6781 if (ret < 0) 6782 return ret; 6783 6784 ext4_msg(sb, KERN_INFO, "re-mounted %pU %s. Quota mode: %s.", 6785 &sb->s_uuid, sb_rdonly(sb) ? "ro" : "r/w", 6786 ext4_quota_mode(sb)); 6787 6788 return 0; 6789 } 6790 6791 #ifdef CONFIG_QUOTA 6792 static int ext4_statfs_project(struct super_block *sb, 6793 kprojid_t projid, struct kstatfs *buf) 6794 { 6795 struct kqid qid; 6796 struct dquot *dquot; 6797 u64 limit; 6798 u64 curblock; 6799 6800 qid = make_kqid_projid(projid); 6801 dquot = dqget(sb, qid); 6802 if (IS_ERR(dquot)) 6803 return PTR_ERR(dquot); 6804 spin_lock(&dquot->dq_dqb_lock); 6805 6806 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6807 dquot->dq_dqb.dqb_bhardlimit); 6808 limit >>= sb->s_blocksize_bits; 6809 6810 if (limit && buf->f_blocks > limit) { 6811 curblock = (dquot->dq_dqb.dqb_curspace + 6812 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6813 buf->f_blocks = limit; 6814 buf->f_bfree = buf->f_bavail = 6815 (buf->f_blocks > curblock) ? 6816 (buf->f_blocks - curblock) : 0; 6817 } 6818 6819 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6820 dquot->dq_dqb.dqb_ihardlimit); 6821 if (limit && buf->f_files > limit) { 6822 buf->f_files = limit; 6823 buf->f_ffree = 6824 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6825 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6826 } 6827 6828 spin_unlock(&dquot->dq_dqb_lock); 6829 dqput(dquot); 6830 return 0; 6831 } 6832 #endif 6833 6834 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6835 { 6836 struct super_block *sb = dentry->d_sb; 6837 struct ext4_sb_info *sbi = EXT4_SB(sb); 6838 struct ext4_super_block *es = sbi->s_es; 6839 ext4_fsblk_t overhead = 0, resv_blocks; 6840 s64 bfree; 6841 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6842 6843 if (!test_opt(sb, MINIX_DF)) 6844 overhead = sbi->s_overhead; 6845 6846 buf->f_type = EXT4_SUPER_MAGIC; 6847 buf->f_bsize = sb->s_blocksize; 6848 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6849 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6850 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6851 /* prevent underflow in case that few free space is available */ 6852 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6853 buf->f_bavail = buf->f_bfree - 6854 (ext4_r_blocks_count(es) + resv_blocks); 6855 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6856 buf->f_bavail = 0; 6857 buf->f_files = le32_to_cpu(es->s_inodes_count); 6858 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6859 buf->f_namelen = EXT4_NAME_LEN; 6860 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6861 6862 #ifdef CONFIG_QUOTA 6863 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6864 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6865 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6866 #endif 6867 return 0; 6868 } 6869 6870 6871 #ifdef CONFIG_QUOTA 6872 6873 /* 6874 * Helper functions so that transaction is started before we acquire dqio_sem 6875 * to keep correct lock ordering of transaction > dqio_sem 6876 */ 6877 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6878 { 6879 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6880 } 6881 6882 static int ext4_write_dquot(struct dquot *dquot) 6883 { 6884 int ret, err; 6885 handle_t *handle; 6886 struct inode *inode; 6887 6888 inode = dquot_to_inode(dquot); 6889 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6890 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6891 if (IS_ERR(handle)) 6892 return PTR_ERR(handle); 6893 ret = dquot_commit(dquot); 6894 if (ret < 0) 6895 ext4_error_err(dquot->dq_sb, -ret, 6896 "Failed to commit dquot type %d", 6897 dquot->dq_id.type); 6898 err = ext4_journal_stop(handle); 6899 if (!ret) 6900 ret = err; 6901 return ret; 6902 } 6903 6904 static int ext4_acquire_dquot(struct dquot *dquot) 6905 { 6906 int ret, err; 6907 handle_t *handle; 6908 6909 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6910 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6911 if (IS_ERR(handle)) 6912 return PTR_ERR(handle); 6913 ret = dquot_acquire(dquot); 6914 if (ret < 0) 6915 ext4_error_err(dquot->dq_sb, -ret, 6916 "Failed to acquire dquot type %d", 6917 dquot->dq_id.type); 6918 err = ext4_journal_stop(handle); 6919 if (!ret) 6920 ret = err; 6921 return ret; 6922 } 6923 6924 static int ext4_release_dquot(struct dquot *dquot) 6925 { 6926 int ret, err; 6927 handle_t *handle; 6928 6929 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6930 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6931 if (IS_ERR(handle)) { 6932 /* Release dquot anyway to avoid endless cycle in dqput() */ 6933 dquot_release(dquot); 6934 return PTR_ERR(handle); 6935 } 6936 ret = dquot_release(dquot); 6937 if (ret < 0) 6938 ext4_error_err(dquot->dq_sb, -ret, 6939 "Failed to release dquot type %d", 6940 dquot->dq_id.type); 6941 err = ext4_journal_stop(handle); 6942 if (!ret) 6943 ret = err; 6944 return ret; 6945 } 6946 6947 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6948 { 6949 struct super_block *sb = dquot->dq_sb; 6950 6951 if (ext4_is_quota_journalled(sb)) { 6952 dquot_mark_dquot_dirty(dquot); 6953 return ext4_write_dquot(dquot); 6954 } else { 6955 return dquot_mark_dquot_dirty(dquot); 6956 } 6957 } 6958 6959 static int ext4_write_info(struct super_block *sb, int type) 6960 { 6961 int ret, err; 6962 handle_t *handle; 6963 6964 /* Data block + inode block */ 6965 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 6966 if (IS_ERR(handle)) 6967 return PTR_ERR(handle); 6968 ret = dquot_commit_info(sb, type); 6969 err = ext4_journal_stop(handle); 6970 if (!ret) 6971 ret = err; 6972 return ret; 6973 } 6974 6975 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6976 { 6977 struct ext4_inode_info *ei = EXT4_I(inode); 6978 6979 /* The first argument of lockdep_set_subclass has to be 6980 * *exactly* the same as the argument to init_rwsem() --- in 6981 * this case, in init_once() --- or lockdep gets unhappy 6982 * because the name of the lock is set using the 6983 * stringification of the argument to init_rwsem(). 6984 */ 6985 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6986 lockdep_set_subclass(&ei->i_data_sem, subclass); 6987 } 6988 6989 /* 6990 * Standard function to be called on quota_on 6991 */ 6992 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6993 const struct path *path) 6994 { 6995 int err; 6996 6997 if (!test_opt(sb, QUOTA)) 6998 return -EINVAL; 6999 7000 /* Quotafile not on the same filesystem? */ 7001 if (path->dentry->d_sb != sb) 7002 return -EXDEV; 7003 7004 /* Quota already enabled for this file? */ 7005 if (IS_NOQUOTA(d_inode(path->dentry))) 7006 return -EBUSY; 7007 7008 /* Journaling quota? */ 7009 if (EXT4_SB(sb)->s_qf_names[type]) { 7010 /* Quotafile not in fs root? */ 7011 if (path->dentry->d_parent != sb->s_root) 7012 ext4_msg(sb, KERN_WARNING, 7013 "Quota file not on filesystem root. " 7014 "Journaled quota will not work"); 7015 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 7016 } else { 7017 /* 7018 * Clear the flag just in case mount options changed since 7019 * last time. 7020 */ 7021 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 7022 } 7023 7024 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 7025 err = dquot_quota_on(sb, type, format_id, path); 7026 if (!err) { 7027 struct inode *inode = d_inode(path->dentry); 7028 handle_t *handle; 7029 7030 /* 7031 * Set inode flags to prevent userspace from messing with quota 7032 * files. If this fails, we return success anyway since quotas 7033 * are already enabled and this is not a hard failure. 7034 */ 7035 inode_lock(inode); 7036 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7037 if (IS_ERR(handle)) 7038 goto unlock_inode; 7039 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 7040 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 7041 S_NOATIME | S_IMMUTABLE); 7042 err = ext4_mark_inode_dirty(handle, inode); 7043 ext4_journal_stop(handle); 7044 unlock_inode: 7045 inode_unlock(inode); 7046 if (err) 7047 dquot_quota_off(sb, type); 7048 } 7049 if (err) 7050 lockdep_set_quota_inode(path->dentry->d_inode, 7051 I_DATA_SEM_NORMAL); 7052 return err; 7053 } 7054 7055 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum) 7056 { 7057 switch (type) { 7058 case USRQUOTA: 7059 return qf_inum == EXT4_USR_QUOTA_INO; 7060 case GRPQUOTA: 7061 return qf_inum == EXT4_GRP_QUOTA_INO; 7062 case PRJQUOTA: 7063 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO; 7064 default: 7065 BUG(); 7066 } 7067 } 7068 7069 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 7070 unsigned int flags) 7071 { 7072 int err; 7073 struct inode *qf_inode; 7074 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7075 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7076 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7077 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7078 }; 7079 7080 BUG_ON(!ext4_has_feature_quota(sb)); 7081 7082 if (!qf_inums[type]) 7083 return -EPERM; 7084 7085 if (!ext4_check_quota_inum(type, qf_inums[type])) { 7086 ext4_error(sb, "Bad quota inum: %lu, type: %d", 7087 qf_inums[type], type); 7088 return -EUCLEAN; 7089 } 7090 7091 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 7092 if (IS_ERR(qf_inode)) { 7093 ext4_error(sb, "Bad quota inode: %lu, type: %d", 7094 qf_inums[type], type); 7095 return PTR_ERR(qf_inode); 7096 } 7097 7098 /* Don't account quota for quota files to avoid recursion */ 7099 qf_inode->i_flags |= S_NOQUOTA; 7100 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 7101 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 7102 if (err) 7103 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 7104 iput(qf_inode); 7105 7106 return err; 7107 } 7108 7109 /* Enable usage tracking for all quota types. */ 7110 int ext4_enable_quotas(struct super_block *sb) 7111 { 7112 int type, err = 0; 7113 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7114 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7115 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7116 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7117 }; 7118 bool quota_mopt[EXT4_MAXQUOTAS] = { 7119 test_opt(sb, USRQUOTA), 7120 test_opt(sb, GRPQUOTA), 7121 test_opt(sb, PRJQUOTA), 7122 }; 7123 7124 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 7125 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 7126 if (qf_inums[type]) { 7127 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 7128 DQUOT_USAGE_ENABLED | 7129 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 7130 if (err) { 7131 ext4_warning(sb, 7132 "Failed to enable quota tracking " 7133 "(type=%d, err=%d, ino=%lu). " 7134 "Please run e2fsck to fix.", type, 7135 err, qf_inums[type]); 7136 7137 ext4_quotas_off(sb, type); 7138 return err; 7139 } 7140 } 7141 } 7142 return 0; 7143 } 7144 7145 static int ext4_quota_off(struct super_block *sb, int type) 7146 { 7147 struct inode *inode = sb_dqopt(sb)->files[type]; 7148 handle_t *handle; 7149 int err; 7150 7151 /* Force all delayed allocation blocks to be allocated. 7152 * Caller already holds s_umount sem */ 7153 if (test_opt(sb, DELALLOC)) 7154 sync_filesystem(sb); 7155 7156 if (!inode || !igrab(inode)) 7157 goto out; 7158 7159 err = dquot_quota_off(sb, type); 7160 if (err || ext4_has_feature_quota(sb)) 7161 goto out_put; 7162 /* 7163 * When the filesystem was remounted read-only first, we cannot cleanup 7164 * inode flags here. Bad luck but people should be using QUOTA feature 7165 * these days anyway. 7166 */ 7167 if (sb_rdonly(sb)) 7168 goto out_put; 7169 7170 inode_lock(inode); 7171 /* 7172 * Update modification times of quota files when userspace can 7173 * start looking at them. If we fail, we return success anyway since 7174 * this is not a hard failure and quotas are already disabled. 7175 */ 7176 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7177 if (IS_ERR(handle)) { 7178 err = PTR_ERR(handle); 7179 goto out_unlock; 7180 } 7181 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7182 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7183 inode->i_mtime = inode_set_ctime_current(inode); 7184 err = ext4_mark_inode_dirty(handle, inode); 7185 ext4_journal_stop(handle); 7186 out_unlock: 7187 inode_unlock(inode); 7188 out_put: 7189 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7190 iput(inode); 7191 return err; 7192 out: 7193 return dquot_quota_off(sb, type); 7194 } 7195 7196 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7197 * acquiring the locks... As quota files are never truncated and quota code 7198 * itself serializes the operations (and no one else should touch the files) 7199 * we don't have to be afraid of races */ 7200 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7201 size_t len, loff_t off) 7202 { 7203 struct inode *inode = sb_dqopt(sb)->files[type]; 7204 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7205 int offset = off & (sb->s_blocksize - 1); 7206 int tocopy; 7207 size_t toread; 7208 struct buffer_head *bh; 7209 loff_t i_size = i_size_read(inode); 7210 7211 if (off > i_size) 7212 return 0; 7213 if (off+len > i_size) 7214 len = i_size-off; 7215 toread = len; 7216 while (toread > 0) { 7217 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread); 7218 bh = ext4_bread(NULL, inode, blk, 0); 7219 if (IS_ERR(bh)) 7220 return PTR_ERR(bh); 7221 if (!bh) /* A hole? */ 7222 memset(data, 0, tocopy); 7223 else 7224 memcpy(data, bh->b_data+offset, tocopy); 7225 brelse(bh); 7226 offset = 0; 7227 toread -= tocopy; 7228 data += tocopy; 7229 blk++; 7230 } 7231 return len; 7232 } 7233 7234 /* Write to quotafile (we know the transaction is already started and has 7235 * enough credits) */ 7236 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7237 const char *data, size_t len, loff_t off) 7238 { 7239 struct inode *inode = sb_dqopt(sb)->files[type]; 7240 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7241 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7242 int retries = 0; 7243 struct buffer_head *bh; 7244 handle_t *handle = journal_current_handle(); 7245 7246 if (!handle) { 7247 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7248 " cancelled because transaction is not started", 7249 (unsigned long long)off, (unsigned long long)len); 7250 return -EIO; 7251 } 7252 /* 7253 * Since we account only one data block in transaction credits, 7254 * then it is impossible to cross a block boundary. 7255 */ 7256 if (sb->s_blocksize - offset < len) { 7257 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7258 " cancelled because not block aligned", 7259 (unsigned long long)off, (unsigned long long)len); 7260 return -EIO; 7261 } 7262 7263 do { 7264 bh = ext4_bread(handle, inode, blk, 7265 EXT4_GET_BLOCKS_CREATE | 7266 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7267 } while (PTR_ERR(bh) == -ENOSPC && 7268 ext4_should_retry_alloc(inode->i_sb, &retries)); 7269 if (IS_ERR(bh)) 7270 return PTR_ERR(bh); 7271 if (!bh) 7272 goto out; 7273 BUFFER_TRACE(bh, "get write access"); 7274 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7275 if (err) { 7276 brelse(bh); 7277 return err; 7278 } 7279 lock_buffer(bh); 7280 memcpy(bh->b_data+offset, data, len); 7281 flush_dcache_page(bh->b_page); 7282 unlock_buffer(bh); 7283 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7284 brelse(bh); 7285 out: 7286 if (inode->i_size < off + len) { 7287 i_size_write(inode, off + len); 7288 EXT4_I(inode)->i_disksize = inode->i_size; 7289 err2 = ext4_mark_inode_dirty(handle, inode); 7290 if (unlikely(err2 && !err)) 7291 err = err2; 7292 } 7293 return err ? err : len; 7294 } 7295 #endif 7296 7297 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7298 static inline void register_as_ext2(void) 7299 { 7300 int err = register_filesystem(&ext2_fs_type); 7301 if (err) 7302 printk(KERN_WARNING 7303 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7304 } 7305 7306 static inline void unregister_as_ext2(void) 7307 { 7308 unregister_filesystem(&ext2_fs_type); 7309 } 7310 7311 static inline int ext2_feature_set_ok(struct super_block *sb) 7312 { 7313 if (ext4_has_unknown_ext2_incompat_features(sb)) 7314 return 0; 7315 if (sb_rdonly(sb)) 7316 return 1; 7317 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7318 return 0; 7319 return 1; 7320 } 7321 #else 7322 static inline void register_as_ext2(void) { } 7323 static inline void unregister_as_ext2(void) { } 7324 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7325 #endif 7326 7327 static inline void register_as_ext3(void) 7328 { 7329 int err = register_filesystem(&ext3_fs_type); 7330 if (err) 7331 printk(KERN_WARNING 7332 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7333 } 7334 7335 static inline void unregister_as_ext3(void) 7336 { 7337 unregister_filesystem(&ext3_fs_type); 7338 } 7339 7340 static inline int ext3_feature_set_ok(struct super_block *sb) 7341 { 7342 if (ext4_has_unknown_ext3_incompat_features(sb)) 7343 return 0; 7344 if (!ext4_has_feature_journal(sb)) 7345 return 0; 7346 if (sb_rdonly(sb)) 7347 return 1; 7348 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7349 return 0; 7350 return 1; 7351 } 7352 7353 static void ext4_kill_sb(struct super_block *sb) 7354 { 7355 struct ext4_sb_info *sbi = EXT4_SB(sb); 7356 struct block_device *journal_bdev = sbi ? sbi->s_journal_bdev : NULL; 7357 7358 kill_block_super(sb); 7359 7360 if (journal_bdev) 7361 blkdev_put(journal_bdev, sb); 7362 } 7363 7364 static struct file_system_type ext4_fs_type = { 7365 .owner = THIS_MODULE, 7366 .name = "ext4", 7367 .init_fs_context = ext4_init_fs_context, 7368 .parameters = ext4_param_specs, 7369 .kill_sb = ext4_kill_sb, 7370 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 7371 }; 7372 MODULE_ALIAS_FS("ext4"); 7373 7374 /* Shared across all ext4 file systems */ 7375 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 7376 7377 static int __init ext4_init_fs(void) 7378 { 7379 int i, err; 7380 7381 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7382 ext4_li_info = NULL; 7383 7384 /* Build-time check for flags consistency */ 7385 ext4_check_flag_values(); 7386 7387 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 7388 init_waitqueue_head(&ext4__ioend_wq[i]); 7389 7390 err = ext4_init_es(); 7391 if (err) 7392 return err; 7393 7394 err = ext4_init_pending(); 7395 if (err) 7396 goto out7; 7397 7398 err = ext4_init_post_read_processing(); 7399 if (err) 7400 goto out6; 7401 7402 err = ext4_init_pageio(); 7403 if (err) 7404 goto out5; 7405 7406 err = ext4_init_system_zone(); 7407 if (err) 7408 goto out4; 7409 7410 err = ext4_init_sysfs(); 7411 if (err) 7412 goto out3; 7413 7414 err = ext4_init_mballoc(); 7415 if (err) 7416 goto out2; 7417 err = init_inodecache(); 7418 if (err) 7419 goto out1; 7420 7421 err = ext4_fc_init_dentry_cache(); 7422 if (err) 7423 goto out05; 7424 7425 register_as_ext3(); 7426 register_as_ext2(); 7427 err = register_filesystem(&ext4_fs_type); 7428 if (err) 7429 goto out; 7430 7431 return 0; 7432 out: 7433 unregister_as_ext2(); 7434 unregister_as_ext3(); 7435 ext4_fc_destroy_dentry_cache(); 7436 out05: 7437 destroy_inodecache(); 7438 out1: 7439 ext4_exit_mballoc(); 7440 out2: 7441 ext4_exit_sysfs(); 7442 out3: 7443 ext4_exit_system_zone(); 7444 out4: 7445 ext4_exit_pageio(); 7446 out5: 7447 ext4_exit_post_read_processing(); 7448 out6: 7449 ext4_exit_pending(); 7450 out7: 7451 ext4_exit_es(); 7452 7453 return err; 7454 } 7455 7456 static void __exit ext4_exit_fs(void) 7457 { 7458 ext4_destroy_lazyinit_thread(); 7459 unregister_as_ext2(); 7460 unregister_as_ext3(); 7461 unregister_filesystem(&ext4_fs_type); 7462 ext4_fc_destroy_dentry_cache(); 7463 destroy_inodecache(); 7464 ext4_exit_mballoc(); 7465 ext4_exit_sysfs(); 7466 ext4_exit_system_zone(); 7467 ext4_exit_pageio(); 7468 ext4_exit_post_read_processing(); 7469 ext4_exit_es(); 7470 ext4_exit_pending(); 7471 } 7472 7473 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7474 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7475 MODULE_LICENSE("GPL"); 7476 MODULE_SOFTDEP("pre: crc32c"); 7477 module_init(ext4_init_fs) 7478 module_exit(ext4_exit_fs) 7479