1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/super.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Big-endian to little-endian byte-swapping/bitmaps by 17 * David S. Miller (davem@caip.rutgers.edu), 1995 18 */ 19 20 #include <linux/module.h> 21 #include <linux/string.h> 22 #include <linux/fs.h> 23 #include <linux/time.h> 24 #include <linux/vmalloc.h> 25 #include <linux/slab.h> 26 #include <linux/init.h> 27 #include <linux/blkdev.h> 28 #include <linux/backing-dev.h> 29 #include <linux/parser.h> 30 #include <linux/buffer_head.h> 31 #include <linux/exportfs.h> 32 #include <linux/vfs.h> 33 #include <linux/random.h> 34 #include <linux/mount.h> 35 #include <linux/namei.h> 36 #include <linux/quotaops.h> 37 #include <linux/seq_file.h> 38 #include <linux/ctype.h> 39 #include <linux/log2.h> 40 #include <linux/crc16.h> 41 #include <linux/dax.h> 42 #include <linux/uaccess.h> 43 #include <linux/iversion.h> 44 #include <linux/unicode.h> 45 #include <linux/part_stat.h> 46 #include <linux/kthread.h> 47 #include <linux/freezer.h> 48 #include <linux/fsnotify.h> 49 #include <linux/fs_context.h> 50 #include <linux/fs_parser.h> 51 52 #include "ext4.h" 53 #include "ext4_extents.h" /* Needed for trace points definition */ 54 #include "ext4_jbd2.h" 55 #include "xattr.h" 56 #include "acl.h" 57 #include "mballoc.h" 58 #include "fsmap.h" 59 60 #define CREATE_TRACE_POINTS 61 #include <trace/events/ext4.h> 62 63 static struct ext4_lazy_init *ext4_li_info; 64 static DEFINE_MUTEX(ext4_li_mtx); 65 static struct ratelimit_state ext4_mount_msg_ratelimit; 66 67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 68 unsigned long journal_devnum); 69 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 70 static void ext4_update_super(struct super_block *sb); 71 static int ext4_commit_super(struct super_block *sb); 72 static int ext4_mark_recovery_complete(struct super_block *sb, 73 struct ext4_super_block *es); 74 static int ext4_clear_journal_err(struct super_block *sb, 75 struct ext4_super_block *es); 76 static int ext4_sync_fs(struct super_block *sb, int wait); 77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 78 static int ext4_unfreeze(struct super_block *sb); 79 static int ext4_freeze(struct super_block *sb); 80 static inline int ext2_feature_set_ok(struct super_block *sb); 81 static inline int ext3_feature_set_ok(struct super_block *sb); 82 static void ext4_destroy_lazyinit_thread(void); 83 static void ext4_unregister_li_request(struct super_block *sb); 84 static void ext4_clear_request_list(void); 85 static struct inode *ext4_get_journal_inode(struct super_block *sb, 86 unsigned int journal_inum); 87 static int ext4_validate_options(struct fs_context *fc); 88 static int ext4_check_opt_consistency(struct fs_context *fc, 89 struct super_block *sb); 90 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb); 91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param); 92 static int ext4_get_tree(struct fs_context *fc); 93 static int ext4_reconfigure(struct fs_context *fc); 94 static void ext4_fc_free(struct fs_context *fc); 95 static int ext4_init_fs_context(struct fs_context *fc); 96 static void ext4_kill_sb(struct super_block *sb); 97 static const struct fs_parameter_spec ext4_param_specs[]; 98 99 /* 100 * Lock ordering 101 * 102 * page fault path: 103 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start 104 * -> page lock -> i_data_sem (rw) 105 * 106 * buffered write path: 107 * sb_start_write -> i_mutex -> mmap_lock 108 * sb_start_write -> i_mutex -> transaction start -> page lock -> 109 * i_data_sem (rw) 110 * 111 * truncate: 112 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) -> 113 * page lock 114 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start -> 115 * i_data_sem (rw) 116 * 117 * direct IO: 118 * sb_start_write -> i_mutex -> mmap_lock 119 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 120 * 121 * writepages: 122 * transaction start -> page lock(s) -> i_data_sem (rw) 123 */ 124 125 static const struct fs_context_operations ext4_context_ops = { 126 .parse_param = ext4_parse_param, 127 .get_tree = ext4_get_tree, 128 .reconfigure = ext4_reconfigure, 129 .free = ext4_fc_free, 130 }; 131 132 133 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 134 static struct file_system_type ext2_fs_type = { 135 .owner = THIS_MODULE, 136 .name = "ext2", 137 .init_fs_context = ext4_init_fs_context, 138 .parameters = ext4_param_specs, 139 .kill_sb = ext4_kill_sb, 140 .fs_flags = FS_REQUIRES_DEV, 141 }; 142 MODULE_ALIAS_FS("ext2"); 143 MODULE_ALIAS("ext2"); 144 #define IS_EXT2_SB(sb) ((sb)->s_type == &ext2_fs_type) 145 #else 146 #define IS_EXT2_SB(sb) (0) 147 #endif 148 149 150 static struct file_system_type ext3_fs_type = { 151 .owner = THIS_MODULE, 152 .name = "ext3", 153 .init_fs_context = ext4_init_fs_context, 154 .parameters = ext4_param_specs, 155 .kill_sb = ext4_kill_sb, 156 .fs_flags = FS_REQUIRES_DEV, 157 }; 158 MODULE_ALIAS_FS("ext3"); 159 MODULE_ALIAS("ext3"); 160 #define IS_EXT3_SB(sb) ((sb)->s_type == &ext3_fs_type) 161 162 163 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 164 bh_end_io_t *end_io) 165 { 166 /* 167 * buffer's verified bit is no longer valid after reading from 168 * disk again due to write out error, clear it to make sure we 169 * recheck the buffer contents. 170 */ 171 clear_buffer_verified(bh); 172 173 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 174 get_bh(bh); 175 submit_bh(REQ_OP_READ | op_flags, bh); 176 } 177 178 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags, 179 bh_end_io_t *end_io) 180 { 181 BUG_ON(!buffer_locked(bh)); 182 183 if (ext4_buffer_uptodate(bh)) { 184 unlock_buffer(bh); 185 return; 186 } 187 __ext4_read_bh(bh, op_flags, end_io); 188 } 189 190 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io) 191 { 192 BUG_ON(!buffer_locked(bh)); 193 194 if (ext4_buffer_uptodate(bh)) { 195 unlock_buffer(bh); 196 return 0; 197 } 198 199 __ext4_read_bh(bh, op_flags, end_io); 200 201 wait_on_buffer(bh); 202 if (buffer_uptodate(bh)) 203 return 0; 204 return -EIO; 205 } 206 207 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 208 { 209 lock_buffer(bh); 210 if (!wait) { 211 ext4_read_bh_nowait(bh, op_flags, NULL); 212 return 0; 213 } 214 return ext4_read_bh(bh, op_flags, NULL); 215 } 216 217 /* 218 * This works like __bread_gfp() except it uses ERR_PTR for error 219 * returns. Currently with sb_bread it's impossible to distinguish 220 * between ENOMEM and EIO situations (since both result in a NULL 221 * return. 222 */ 223 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 224 sector_t block, 225 blk_opf_t op_flags, gfp_t gfp) 226 { 227 struct buffer_head *bh; 228 int ret; 229 230 bh = sb_getblk_gfp(sb, block, gfp); 231 if (bh == NULL) 232 return ERR_PTR(-ENOMEM); 233 if (ext4_buffer_uptodate(bh)) 234 return bh; 235 236 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 237 if (ret) { 238 put_bh(bh); 239 return ERR_PTR(ret); 240 } 241 return bh; 242 } 243 244 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 245 blk_opf_t op_flags) 246 { 247 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping, 248 ~__GFP_FS) | __GFP_MOVABLE; 249 250 return __ext4_sb_bread_gfp(sb, block, op_flags, gfp); 251 } 252 253 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 254 sector_t block) 255 { 256 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping, 257 ~__GFP_FS); 258 259 return __ext4_sb_bread_gfp(sb, block, 0, gfp); 260 } 261 262 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 263 { 264 struct buffer_head *bh = bdev_getblk(sb->s_bdev, block, 265 sb->s_blocksize, GFP_NOWAIT | __GFP_NOWARN); 266 267 if (likely(bh)) { 268 if (trylock_buffer(bh)) 269 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL); 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(bg->bg_free_inodes_count_lo) | 350 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 351 (__u32)le16_to_cpu(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 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 406 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 407 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 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 496 { 497 struct super_block *sb = journal->j_private; 498 struct ext4_sb_info *sbi = EXT4_SB(sb); 499 int error = is_journal_aborted(journal); 500 struct ext4_journal_cb_entry *jce; 501 502 BUG_ON(txn->t_state == T_FINISHED); 503 504 ext4_process_freed_data(sb, txn->t_tid); 505 ext4_maybe_update_superblock(sb); 506 507 spin_lock(&sbi->s_md_lock); 508 while (!list_empty(&txn->t_private_list)) { 509 jce = list_entry(txn->t_private_list.next, 510 struct ext4_journal_cb_entry, jce_list); 511 list_del_init(&jce->jce_list); 512 spin_unlock(&sbi->s_md_lock); 513 jce->jce_func(sb, jce, error); 514 spin_lock(&sbi->s_md_lock); 515 } 516 spin_unlock(&sbi->s_md_lock); 517 } 518 519 /* 520 * This writepage callback for write_cache_pages() 521 * takes care of a few cases after page cleaning. 522 * 523 * write_cache_pages() already checks for dirty pages 524 * and calls clear_page_dirty_for_io(), which we want, 525 * to write protect the pages. 526 * 527 * However, we may have to redirty a page (see below.) 528 */ 529 static int ext4_journalled_writepage_callback(struct folio *folio, 530 struct writeback_control *wbc, 531 void *data) 532 { 533 transaction_t *transaction = (transaction_t *) data; 534 struct buffer_head *bh, *head; 535 struct journal_head *jh; 536 537 bh = head = folio_buffers(folio); 538 do { 539 /* 540 * We have to redirty a page in these cases: 541 * 1) If buffer is dirty, it means the page was dirty because it 542 * contains a buffer that needs checkpointing. So the dirty bit 543 * needs to be preserved so that checkpointing writes the buffer 544 * properly. 545 * 2) If buffer is not part of the committing transaction 546 * (we may have just accidentally come across this buffer because 547 * inode range tracking is not exact) or if the currently running 548 * transaction already contains this buffer as well, dirty bit 549 * needs to be preserved so that the buffer gets writeprotected 550 * properly on running transaction's commit. 551 */ 552 jh = bh2jh(bh); 553 if (buffer_dirty(bh) || 554 (jh && (jh->b_transaction != transaction || 555 jh->b_next_transaction))) { 556 folio_redirty_for_writepage(wbc, folio); 557 goto out; 558 } 559 } while ((bh = bh->b_this_page) != head); 560 561 out: 562 return AOP_WRITEPAGE_ACTIVATE; 563 } 564 565 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 566 { 567 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 568 struct writeback_control wbc = { 569 .sync_mode = WB_SYNC_ALL, 570 .nr_to_write = LONG_MAX, 571 .range_start = jinode->i_dirty_start, 572 .range_end = jinode->i_dirty_end, 573 }; 574 575 return write_cache_pages(mapping, &wbc, 576 ext4_journalled_writepage_callback, 577 jinode->i_transaction); 578 } 579 580 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 581 { 582 int ret; 583 584 if (ext4_should_journal_data(jinode->i_vfs_inode)) 585 ret = ext4_journalled_submit_inode_data_buffers(jinode); 586 else 587 ret = ext4_normal_submit_inode_data_buffers(jinode); 588 return ret; 589 } 590 591 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 592 { 593 int ret = 0; 594 595 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 596 ret = jbd2_journal_finish_inode_data_buffers(jinode); 597 598 return ret; 599 } 600 601 static bool system_going_down(void) 602 { 603 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 604 || system_state == SYSTEM_RESTART; 605 } 606 607 struct ext4_err_translation { 608 int code; 609 int errno; 610 }; 611 612 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 613 614 static struct ext4_err_translation err_translation[] = { 615 EXT4_ERR_TRANSLATE(EIO), 616 EXT4_ERR_TRANSLATE(ENOMEM), 617 EXT4_ERR_TRANSLATE(EFSBADCRC), 618 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 619 EXT4_ERR_TRANSLATE(ENOSPC), 620 EXT4_ERR_TRANSLATE(ENOKEY), 621 EXT4_ERR_TRANSLATE(EROFS), 622 EXT4_ERR_TRANSLATE(EFBIG), 623 EXT4_ERR_TRANSLATE(EEXIST), 624 EXT4_ERR_TRANSLATE(ERANGE), 625 EXT4_ERR_TRANSLATE(EOVERFLOW), 626 EXT4_ERR_TRANSLATE(EBUSY), 627 EXT4_ERR_TRANSLATE(ENOTDIR), 628 EXT4_ERR_TRANSLATE(ENOTEMPTY), 629 EXT4_ERR_TRANSLATE(ESHUTDOWN), 630 EXT4_ERR_TRANSLATE(EFAULT), 631 }; 632 633 static int ext4_errno_to_code(int errno) 634 { 635 int i; 636 637 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 638 if (err_translation[i].errno == errno) 639 return err_translation[i].code; 640 return EXT4_ERR_UNKNOWN; 641 } 642 643 static void save_error_info(struct super_block *sb, int error, 644 __u32 ino, __u64 block, 645 const char *func, unsigned int line) 646 { 647 struct ext4_sb_info *sbi = EXT4_SB(sb); 648 649 /* We default to EFSCORRUPTED error... */ 650 if (error == 0) 651 error = EFSCORRUPTED; 652 653 spin_lock(&sbi->s_error_lock); 654 sbi->s_add_error_count++; 655 sbi->s_last_error_code = error; 656 sbi->s_last_error_line = line; 657 sbi->s_last_error_ino = ino; 658 sbi->s_last_error_block = block; 659 sbi->s_last_error_func = func; 660 sbi->s_last_error_time = ktime_get_real_seconds(); 661 if (!sbi->s_first_error_time) { 662 sbi->s_first_error_code = error; 663 sbi->s_first_error_line = line; 664 sbi->s_first_error_ino = ino; 665 sbi->s_first_error_block = block; 666 sbi->s_first_error_func = func; 667 sbi->s_first_error_time = sbi->s_last_error_time; 668 } 669 spin_unlock(&sbi->s_error_lock); 670 } 671 672 /* Deal with the reporting of failure conditions on a filesystem such as 673 * inconsistencies detected or read IO failures. 674 * 675 * On ext2, we can store the error state of the filesystem in the 676 * superblock. That is not possible on ext4, because we may have other 677 * write ordering constraints on the superblock which prevent us from 678 * writing it out straight away; and given that the journal is about to 679 * be aborted, we can't rely on the current, or future, transactions to 680 * write out the superblock safely. 681 * 682 * We'll just use the jbd2_journal_abort() error code to record an error in 683 * the journal instead. On recovery, the journal will complain about 684 * that error until we've noted it down and cleared it. 685 * 686 * If force_ro is set, we unconditionally force the filesystem into an 687 * ABORT|READONLY state, unless the error response on the fs has been set to 688 * panic in which case we take the easy way out and panic immediately. This is 689 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 690 * at a critical moment in log management. 691 */ 692 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 693 __u32 ino, __u64 block, 694 const char *func, unsigned int line) 695 { 696 journal_t *journal = EXT4_SB(sb)->s_journal; 697 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 698 699 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 700 if (test_opt(sb, WARN_ON_ERROR)) 701 WARN_ON_ONCE(1); 702 703 if (!continue_fs && !sb_rdonly(sb)) { 704 set_bit(EXT4_FLAGS_SHUTDOWN, &EXT4_SB(sb)->s_ext4_flags); 705 if (journal) 706 jbd2_journal_abort(journal, -EIO); 707 } 708 709 if (!bdev_read_only(sb->s_bdev)) { 710 save_error_info(sb, error, ino, block, func, line); 711 /* 712 * In case the fs should keep running, we need to writeout 713 * superblock through the journal. Due to lock ordering 714 * constraints, it may not be safe to do it right here so we 715 * defer superblock flushing to a workqueue. 716 */ 717 if (continue_fs && journal) 718 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 719 else 720 ext4_commit_super(sb); 721 } 722 723 /* 724 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 725 * could panic during 'reboot -f' as the underlying device got already 726 * disabled. 727 */ 728 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 729 panic("EXT4-fs (device %s): panic forced after error\n", 730 sb->s_id); 731 } 732 733 if (sb_rdonly(sb) || continue_fs) 734 return; 735 736 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 737 /* 738 * Make sure updated value of ->s_mount_flags will be visible before 739 * ->s_flags update 740 */ 741 smp_wmb(); 742 sb->s_flags |= SB_RDONLY; 743 } 744 745 static void update_super_work(struct work_struct *work) 746 { 747 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 748 s_sb_upd_work); 749 journal_t *journal = sbi->s_journal; 750 handle_t *handle; 751 752 /* 753 * If the journal is still running, we have to write out superblock 754 * through the journal to avoid collisions of other journalled sb 755 * updates. 756 * 757 * We use directly jbd2 functions here to avoid recursing back into 758 * ext4 error handling code during handling of previous errors. 759 */ 760 if (!sb_rdonly(sbi->s_sb) && journal) { 761 struct buffer_head *sbh = sbi->s_sbh; 762 bool call_notify_err = false; 763 764 handle = jbd2_journal_start(journal, 1); 765 if (IS_ERR(handle)) 766 goto write_directly; 767 if (jbd2_journal_get_write_access(handle, sbh)) { 768 jbd2_journal_stop(handle); 769 goto write_directly; 770 } 771 772 if (sbi->s_add_error_count > 0) 773 call_notify_err = true; 774 775 ext4_update_super(sbi->s_sb); 776 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 777 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 778 "superblock detected"); 779 clear_buffer_write_io_error(sbh); 780 set_buffer_uptodate(sbh); 781 } 782 783 if (jbd2_journal_dirty_metadata(handle, sbh)) { 784 jbd2_journal_stop(handle); 785 goto write_directly; 786 } 787 jbd2_journal_stop(handle); 788 789 if (call_notify_err) 790 ext4_notify_error_sysfs(sbi); 791 792 return; 793 } 794 write_directly: 795 /* 796 * Write through journal failed. Write sb directly to get error info 797 * out and hope for the best. 798 */ 799 ext4_commit_super(sbi->s_sb); 800 ext4_notify_error_sysfs(sbi); 801 } 802 803 #define ext4_error_ratelimit(sb) \ 804 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 805 "EXT4-fs error") 806 807 void __ext4_error(struct super_block *sb, const char *function, 808 unsigned int line, bool force_ro, int error, __u64 block, 809 const char *fmt, ...) 810 { 811 struct va_format vaf; 812 va_list args; 813 814 if (unlikely(ext4_forced_shutdown(sb))) 815 return; 816 817 trace_ext4_error(sb, function, line); 818 if (ext4_error_ratelimit(sb)) { 819 va_start(args, fmt); 820 vaf.fmt = fmt; 821 vaf.va = &args; 822 printk(KERN_CRIT 823 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 824 sb->s_id, function, line, current->comm, &vaf); 825 va_end(args); 826 } 827 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED); 828 829 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 830 } 831 832 void __ext4_error_inode(struct inode *inode, const char *function, 833 unsigned int line, ext4_fsblk_t block, int error, 834 const char *fmt, ...) 835 { 836 va_list args; 837 struct va_format vaf; 838 839 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 840 return; 841 842 trace_ext4_error(inode->i_sb, function, line); 843 if (ext4_error_ratelimit(inode->i_sb)) { 844 va_start(args, fmt); 845 vaf.fmt = fmt; 846 vaf.va = &args; 847 if (block) 848 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 849 "inode #%lu: block %llu: comm %s: %pV\n", 850 inode->i_sb->s_id, function, line, inode->i_ino, 851 block, current->comm, &vaf); 852 else 853 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 854 "inode #%lu: comm %s: %pV\n", 855 inode->i_sb->s_id, function, line, inode->i_ino, 856 current->comm, &vaf); 857 va_end(args); 858 } 859 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED); 860 861 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 862 function, line); 863 } 864 865 void __ext4_error_file(struct file *file, const char *function, 866 unsigned int line, ext4_fsblk_t block, 867 const char *fmt, ...) 868 { 869 va_list args; 870 struct va_format vaf; 871 struct inode *inode = file_inode(file); 872 char pathname[80], *path; 873 874 if (unlikely(ext4_forced_shutdown(inode->i_sb))) 875 return; 876 877 trace_ext4_error(inode->i_sb, function, line); 878 if (ext4_error_ratelimit(inode->i_sb)) { 879 path = file_path(file, pathname, sizeof(pathname)); 880 if (IS_ERR(path)) 881 path = "(unknown)"; 882 va_start(args, fmt); 883 vaf.fmt = fmt; 884 vaf.va = &args; 885 if (block) 886 printk(KERN_CRIT 887 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 888 "block %llu: comm %s: path %s: %pV\n", 889 inode->i_sb->s_id, function, line, inode->i_ino, 890 block, current->comm, path, &vaf); 891 else 892 printk(KERN_CRIT 893 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 894 "comm %s: path %s: %pV\n", 895 inode->i_sb->s_id, function, line, inode->i_ino, 896 current->comm, path, &vaf); 897 va_end(args); 898 } 899 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED); 900 901 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 902 function, line); 903 } 904 905 const char *ext4_decode_error(struct super_block *sb, int errno, 906 char nbuf[16]) 907 { 908 char *errstr = NULL; 909 910 switch (errno) { 911 case -EFSCORRUPTED: 912 errstr = "Corrupt filesystem"; 913 break; 914 case -EFSBADCRC: 915 errstr = "Filesystem failed CRC"; 916 break; 917 case -EIO: 918 errstr = "IO failure"; 919 break; 920 case -ENOMEM: 921 errstr = "Out of memory"; 922 break; 923 case -EROFS: 924 if (!sb || (EXT4_SB(sb)->s_journal && 925 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 926 errstr = "Journal has aborted"; 927 else 928 errstr = "Readonly filesystem"; 929 break; 930 default: 931 /* If the caller passed in an extra buffer for unknown 932 * errors, textualise them now. Else we just return 933 * NULL. */ 934 if (nbuf) { 935 /* Check for truncated error codes... */ 936 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 937 errstr = nbuf; 938 } 939 break; 940 } 941 942 return errstr; 943 } 944 945 /* __ext4_std_error decodes expected errors from journaling functions 946 * automatically and invokes the appropriate error response. */ 947 948 void __ext4_std_error(struct super_block *sb, const char *function, 949 unsigned int line, int errno) 950 { 951 char nbuf[16]; 952 const char *errstr; 953 954 if (unlikely(ext4_forced_shutdown(sb))) 955 return; 956 957 /* Special case: if the error is EROFS, and we're not already 958 * inside a transaction, then there's really no point in logging 959 * an error. */ 960 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 961 return; 962 963 if (ext4_error_ratelimit(sb)) { 964 errstr = ext4_decode_error(sb, errno, nbuf); 965 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 966 sb->s_id, function, line, errstr); 967 } 968 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED); 969 970 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 971 } 972 973 void __ext4_msg(struct super_block *sb, 974 const char *prefix, const char *fmt, ...) 975 { 976 struct va_format vaf; 977 va_list args; 978 979 if (sb) { 980 atomic_inc(&EXT4_SB(sb)->s_msg_count); 981 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 982 "EXT4-fs")) 983 return; 984 } 985 986 va_start(args, fmt); 987 vaf.fmt = fmt; 988 vaf.va = &args; 989 if (sb) 990 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 991 else 992 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 993 va_end(args); 994 } 995 996 static int ext4_warning_ratelimit(struct super_block *sb) 997 { 998 atomic_inc(&EXT4_SB(sb)->s_warning_count); 999 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 1000 "EXT4-fs warning"); 1001 } 1002 1003 void __ext4_warning(struct super_block *sb, const char *function, 1004 unsigned int line, const char *fmt, ...) 1005 { 1006 struct va_format vaf; 1007 va_list args; 1008 1009 if (!ext4_warning_ratelimit(sb)) 1010 return; 1011 1012 va_start(args, fmt); 1013 vaf.fmt = fmt; 1014 vaf.va = &args; 1015 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 1016 sb->s_id, function, line, &vaf); 1017 va_end(args); 1018 } 1019 1020 void __ext4_warning_inode(const struct inode *inode, 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(inode->i_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: " 1033 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 1034 function, line, inode->i_ino, current->comm, &vaf); 1035 va_end(args); 1036 } 1037 1038 void __ext4_grp_locked_error(const char *function, unsigned int line, 1039 struct super_block *sb, ext4_group_t grp, 1040 unsigned long ino, ext4_fsblk_t block, 1041 const char *fmt, ...) 1042 __releases(bitlock) 1043 __acquires(bitlock) 1044 { 1045 struct va_format vaf; 1046 va_list args; 1047 1048 if (unlikely(ext4_forced_shutdown(sb))) 1049 return; 1050 1051 trace_ext4_error(sb, function, line); 1052 if (ext4_error_ratelimit(sb)) { 1053 va_start(args, fmt); 1054 vaf.fmt = fmt; 1055 vaf.va = &args; 1056 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1057 sb->s_id, function, line, grp); 1058 if (ino) 1059 printk(KERN_CONT "inode %lu: ", ino); 1060 if (block) 1061 printk(KERN_CONT "block %llu:", 1062 (unsigned long long) block); 1063 printk(KERN_CONT "%pV\n", &vaf); 1064 va_end(args); 1065 } 1066 1067 if (test_opt(sb, ERRORS_CONT)) { 1068 if (test_opt(sb, WARN_ON_ERROR)) 1069 WARN_ON_ONCE(1); 1070 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1071 if (!bdev_read_only(sb->s_bdev)) { 1072 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1073 line); 1074 schedule_work(&EXT4_SB(sb)->s_sb_upd_work); 1075 } 1076 return; 1077 } 1078 ext4_unlock_group(sb, grp); 1079 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1080 /* 1081 * We only get here in the ERRORS_RO case; relocking the group 1082 * may be dangerous, but nothing bad will happen since the 1083 * filesystem will have already been marked read/only and the 1084 * journal has been aborted. We return 1 as a hint to callers 1085 * who might what to use the return value from 1086 * ext4_grp_locked_error() to distinguish between the 1087 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1088 * aggressively from the ext4 function in question, with a 1089 * more appropriate error code. 1090 */ 1091 ext4_lock_group(sb, grp); 1092 return; 1093 } 1094 1095 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1096 ext4_group_t group, 1097 unsigned int flags) 1098 { 1099 struct ext4_sb_info *sbi = EXT4_SB(sb); 1100 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1101 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1102 int ret; 1103 1104 if (!grp || !gdp) 1105 return; 1106 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1107 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1108 &grp->bb_state); 1109 if (!ret) 1110 percpu_counter_sub(&sbi->s_freeclusters_counter, 1111 grp->bb_free); 1112 } 1113 1114 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1115 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1116 &grp->bb_state); 1117 if (!ret && gdp) { 1118 int count; 1119 1120 count = ext4_free_inodes_count(sb, gdp); 1121 percpu_counter_sub(&sbi->s_freeinodes_counter, 1122 count); 1123 } 1124 } 1125 } 1126 1127 void ext4_update_dynamic_rev(struct super_block *sb) 1128 { 1129 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1130 1131 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1132 return; 1133 1134 ext4_warning(sb, 1135 "updating to rev %d because of new feature flag, " 1136 "running e2fsck is recommended", 1137 EXT4_DYNAMIC_REV); 1138 1139 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1140 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1141 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1142 /* leave es->s_feature_*compat flags alone */ 1143 /* es->s_uuid will be set by e2fsck if empty */ 1144 1145 /* 1146 * The rest of the superblock fields should be zero, and if not it 1147 * means they are likely already in use, so leave them alone. We 1148 * can leave it up to e2fsck to clean up any inconsistencies there. 1149 */ 1150 } 1151 1152 static inline struct inode *orphan_list_entry(struct list_head *l) 1153 { 1154 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1155 } 1156 1157 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1158 { 1159 struct list_head *l; 1160 1161 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1162 le32_to_cpu(sbi->s_es->s_last_orphan)); 1163 1164 printk(KERN_ERR "sb_info orphan list:\n"); 1165 list_for_each(l, &sbi->s_orphan) { 1166 struct inode *inode = orphan_list_entry(l); 1167 printk(KERN_ERR " " 1168 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1169 inode->i_sb->s_id, inode->i_ino, inode, 1170 inode->i_mode, inode->i_nlink, 1171 NEXT_ORPHAN(inode)); 1172 } 1173 } 1174 1175 #ifdef CONFIG_QUOTA 1176 static int ext4_quota_off(struct super_block *sb, int type); 1177 1178 static inline void ext4_quotas_off(struct super_block *sb, int type) 1179 { 1180 BUG_ON(type > EXT4_MAXQUOTAS); 1181 1182 /* Use our quota_off function to clear inode flags etc. */ 1183 for (type--; type >= 0; type--) 1184 ext4_quota_off(sb, type); 1185 } 1186 1187 /* 1188 * This is a helper function which is used in the mount/remount 1189 * codepaths (which holds s_umount) to fetch the quota file name. 1190 */ 1191 static inline char *get_qf_name(struct super_block *sb, 1192 struct ext4_sb_info *sbi, 1193 int type) 1194 { 1195 return rcu_dereference_protected(sbi->s_qf_names[type], 1196 lockdep_is_held(&sb->s_umount)); 1197 } 1198 #else 1199 static inline void ext4_quotas_off(struct super_block *sb, int type) 1200 { 1201 } 1202 #endif 1203 1204 static int ext4_percpu_param_init(struct ext4_sb_info *sbi) 1205 { 1206 ext4_fsblk_t block; 1207 int err; 1208 1209 block = ext4_count_free_clusters(sbi->s_sb); 1210 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block)); 1211 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 1212 GFP_KERNEL); 1213 if (!err) { 1214 unsigned long freei = ext4_count_free_inodes(sbi->s_sb); 1215 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 1216 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 1217 GFP_KERNEL); 1218 } 1219 if (!err) 1220 err = percpu_counter_init(&sbi->s_dirs_counter, 1221 ext4_count_dirs(sbi->s_sb), GFP_KERNEL); 1222 if (!err) 1223 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 1224 GFP_KERNEL); 1225 if (!err) 1226 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 1227 GFP_KERNEL); 1228 if (!err) 1229 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 1230 1231 if (err) 1232 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory"); 1233 1234 return err; 1235 } 1236 1237 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi) 1238 { 1239 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1240 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1241 percpu_counter_destroy(&sbi->s_dirs_counter); 1242 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1243 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1244 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1245 } 1246 1247 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 1248 { 1249 struct buffer_head **group_desc; 1250 int i; 1251 1252 rcu_read_lock(); 1253 group_desc = rcu_dereference(sbi->s_group_desc); 1254 for (i = 0; i < sbi->s_gdb_count; i++) 1255 brelse(group_desc[i]); 1256 kvfree(group_desc); 1257 rcu_read_unlock(); 1258 } 1259 1260 static void ext4_flex_groups_free(struct ext4_sb_info *sbi) 1261 { 1262 struct flex_groups **flex_groups; 1263 int i; 1264 1265 rcu_read_lock(); 1266 flex_groups = rcu_dereference(sbi->s_flex_groups); 1267 if (flex_groups) { 1268 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1269 kvfree(flex_groups[i]); 1270 kvfree(flex_groups); 1271 } 1272 rcu_read_unlock(); 1273 } 1274 1275 static void ext4_put_super(struct super_block *sb) 1276 { 1277 struct ext4_sb_info *sbi = EXT4_SB(sb); 1278 struct ext4_super_block *es = sbi->s_es; 1279 int aborted = 0; 1280 int err; 1281 1282 /* 1283 * Unregister sysfs before destroying jbd2 journal. 1284 * Since we could still access attr_journal_task attribute via sysfs 1285 * path which could have sbi->s_journal->j_task as NULL 1286 * Unregister sysfs before flush sbi->s_sb_upd_work. 1287 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1288 * read metadata verify failed then will queue error work. 1289 * update_super_work will call start_this_handle may trigger 1290 * BUG_ON. 1291 */ 1292 ext4_unregister_sysfs(sb); 1293 1294 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1295 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.", 1296 &sb->s_uuid); 1297 1298 ext4_unregister_li_request(sb); 1299 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 1300 1301 flush_work(&sbi->s_sb_upd_work); 1302 destroy_workqueue(sbi->rsv_conversion_wq); 1303 ext4_release_orphan_info(sb); 1304 1305 if (sbi->s_journal) { 1306 aborted = is_journal_aborted(sbi->s_journal); 1307 err = jbd2_journal_destroy(sbi->s_journal); 1308 sbi->s_journal = NULL; 1309 if ((err < 0) && !aborted) { 1310 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1311 } 1312 } 1313 1314 ext4_es_unregister_shrinker(sbi); 1315 timer_shutdown_sync(&sbi->s_err_report); 1316 ext4_release_system_zone(sb); 1317 ext4_mb_release(sb); 1318 ext4_ext_release(sb); 1319 1320 if (!sb_rdonly(sb) && !aborted) { 1321 ext4_clear_feature_journal_needs_recovery(sb); 1322 ext4_clear_feature_orphan_present(sb); 1323 es->s_state = cpu_to_le16(sbi->s_mount_state); 1324 } 1325 if (!sb_rdonly(sb)) 1326 ext4_commit_super(sb); 1327 1328 ext4_group_desc_free(sbi); 1329 ext4_flex_groups_free(sbi); 1330 ext4_percpu_param_destroy(sbi); 1331 #ifdef CONFIG_QUOTA 1332 for (int i = 0; i < EXT4_MAXQUOTAS; i++) 1333 kfree(get_qf_name(sb, sbi, i)); 1334 #endif 1335 1336 /* Debugging code just in case the in-memory inode orphan list 1337 * isn't empty. The on-disk one can be non-empty if we've 1338 * detected an error and taken the fs readonly, but the 1339 * in-memory list had better be clean by this point. */ 1340 if (!list_empty(&sbi->s_orphan)) 1341 dump_orphan_list(sb, sbi); 1342 ASSERT(list_empty(&sbi->s_orphan)); 1343 1344 sync_blockdev(sb->s_bdev); 1345 invalidate_bdev(sb->s_bdev); 1346 if (sbi->s_journal_bdev_file) { 1347 /* 1348 * Invalidate the journal device's buffers. We don't want them 1349 * floating about in memory - the physical journal device may 1350 * hotswapped, and it breaks the `ro-after' testing code. 1351 */ 1352 sync_blockdev(file_bdev(sbi->s_journal_bdev_file)); 1353 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 1354 } 1355 1356 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1357 sbi->s_ea_inode_cache = NULL; 1358 1359 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1360 sbi->s_ea_block_cache = NULL; 1361 1362 ext4_stop_mmpd(sbi); 1363 1364 brelse(sbi->s_sbh); 1365 sb->s_fs_info = NULL; 1366 /* 1367 * Now that we are completely done shutting down the 1368 * superblock, we need to actually destroy the kobject. 1369 */ 1370 kobject_put(&sbi->s_kobj); 1371 wait_for_completion(&sbi->s_kobj_unregister); 1372 if (sbi->s_chksum_driver) 1373 crypto_free_shash(sbi->s_chksum_driver); 1374 kfree(sbi->s_blockgroup_lock); 1375 fs_put_dax(sbi->s_daxdev, NULL); 1376 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1377 #if IS_ENABLED(CONFIG_UNICODE) 1378 utf8_unload(sb->s_encoding); 1379 #endif 1380 kfree(sbi); 1381 } 1382 1383 static struct kmem_cache *ext4_inode_cachep; 1384 1385 /* 1386 * Called inside transaction, so use GFP_NOFS 1387 */ 1388 static struct inode *ext4_alloc_inode(struct super_block *sb) 1389 { 1390 struct ext4_inode_info *ei; 1391 1392 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1393 if (!ei) 1394 return NULL; 1395 1396 inode_set_iversion(&ei->vfs_inode, 1); 1397 ei->i_flags = 0; 1398 spin_lock_init(&ei->i_raw_lock); 1399 ei->i_prealloc_node = RB_ROOT; 1400 atomic_set(&ei->i_prealloc_active, 0); 1401 rwlock_init(&ei->i_prealloc_lock); 1402 ext4_es_init_tree(&ei->i_es_tree); 1403 rwlock_init(&ei->i_es_lock); 1404 INIT_LIST_HEAD(&ei->i_es_list); 1405 ei->i_es_all_nr = 0; 1406 ei->i_es_shk_nr = 0; 1407 ei->i_es_shrink_lblk = 0; 1408 ei->i_reserved_data_blocks = 0; 1409 spin_lock_init(&(ei->i_block_reservation_lock)); 1410 ext4_init_pending_tree(&ei->i_pending_tree); 1411 #ifdef CONFIG_QUOTA 1412 ei->i_reserved_quota = 0; 1413 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1414 #endif 1415 ei->jinode = NULL; 1416 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1417 spin_lock_init(&ei->i_completed_io_lock); 1418 ei->i_sync_tid = 0; 1419 ei->i_datasync_tid = 0; 1420 atomic_set(&ei->i_unwritten, 0); 1421 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1422 ext4_fc_init_inode(&ei->vfs_inode); 1423 mutex_init(&ei->i_fc_lock); 1424 return &ei->vfs_inode; 1425 } 1426 1427 static int ext4_drop_inode(struct inode *inode) 1428 { 1429 int drop = generic_drop_inode(inode); 1430 1431 if (!drop) 1432 drop = fscrypt_drop_inode(inode); 1433 1434 trace_ext4_drop_inode(inode, drop); 1435 return drop; 1436 } 1437 1438 static void ext4_free_in_core_inode(struct inode *inode) 1439 { 1440 fscrypt_free_inode(inode); 1441 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1442 pr_warn("%s: inode %ld still in fc list", 1443 __func__, inode->i_ino); 1444 } 1445 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1446 } 1447 1448 static void ext4_destroy_inode(struct inode *inode) 1449 { 1450 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1451 ext4_msg(inode->i_sb, KERN_ERR, 1452 "Inode %lu (%p): orphan list check failed!", 1453 inode->i_ino, EXT4_I(inode)); 1454 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1455 EXT4_I(inode), sizeof(struct ext4_inode_info), 1456 true); 1457 dump_stack(); 1458 } 1459 1460 if (EXT4_I(inode)->i_reserved_data_blocks) 1461 ext4_msg(inode->i_sb, KERN_ERR, 1462 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!", 1463 inode->i_ino, EXT4_I(inode), 1464 EXT4_I(inode)->i_reserved_data_blocks); 1465 } 1466 1467 static void ext4_shutdown(struct super_block *sb) 1468 { 1469 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH); 1470 } 1471 1472 static void init_once(void *foo) 1473 { 1474 struct ext4_inode_info *ei = foo; 1475 1476 INIT_LIST_HEAD(&ei->i_orphan); 1477 init_rwsem(&ei->xattr_sem); 1478 init_rwsem(&ei->i_data_sem); 1479 inode_init_once(&ei->vfs_inode); 1480 ext4_fc_init_inode(&ei->vfs_inode); 1481 } 1482 1483 static int __init init_inodecache(void) 1484 { 1485 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1486 sizeof(struct ext4_inode_info), 0, 1487 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT, 1488 offsetof(struct ext4_inode_info, i_data), 1489 sizeof_field(struct ext4_inode_info, i_data), 1490 init_once); 1491 if (ext4_inode_cachep == NULL) 1492 return -ENOMEM; 1493 return 0; 1494 } 1495 1496 static void destroy_inodecache(void) 1497 { 1498 /* 1499 * Make sure all delayed rcu free inodes are flushed before we 1500 * destroy cache. 1501 */ 1502 rcu_barrier(); 1503 kmem_cache_destroy(ext4_inode_cachep); 1504 } 1505 1506 void ext4_clear_inode(struct inode *inode) 1507 { 1508 ext4_fc_del(inode); 1509 invalidate_inode_buffers(inode); 1510 clear_inode(inode); 1511 ext4_discard_preallocations(inode); 1512 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1513 dquot_drop(inode); 1514 if (EXT4_I(inode)->jinode) { 1515 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1516 EXT4_I(inode)->jinode); 1517 jbd2_free_inode(EXT4_I(inode)->jinode); 1518 EXT4_I(inode)->jinode = NULL; 1519 } 1520 fscrypt_put_encryption_info(inode); 1521 fsverity_cleanup_inode(inode); 1522 } 1523 1524 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1525 u64 ino, u32 generation) 1526 { 1527 struct inode *inode; 1528 1529 /* 1530 * Currently we don't know the generation for parent directory, so 1531 * a generation of 0 means "accept any" 1532 */ 1533 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1534 if (IS_ERR(inode)) 1535 return ERR_CAST(inode); 1536 if (generation && inode->i_generation != generation) { 1537 iput(inode); 1538 return ERR_PTR(-ESTALE); 1539 } 1540 1541 return inode; 1542 } 1543 1544 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1545 int fh_len, int fh_type) 1546 { 1547 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1548 ext4_nfs_get_inode); 1549 } 1550 1551 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1552 int fh_len, int fh_type) 1553 { 1554 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1555 ext4_nfs_get_inode); 1556 } 1557 1558 static int ext4_nfs_commit_metadata(struct inode *inode) 1559 { 1560 struct writeback_control wbc = { 1561 .sync_mode = WB_SYNC_ALL 1562 }; 1563 1564 trace_ext4_nfs_commit_metadata(inode); 1565 return ext4_write_inode(inode, &wbc); 1566 } 1567 1568 #ifdef CONFIG_QUOTA 1569 static const char * const quotatypes[] = INITQFNAMES; 1570 #define QTYPE2NAME(t) (quotatypes[t]) 1571 1572 static int ext4_write_dquot(struct dquot *dquot); 1573 static int ext4_acquire_dquot(struct dquot *dquot); 1574 static int ext4_release_dquot(struct dquot *dquot); 1575 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1576 static int ext4_write_info(struct super_block *sb, int type); 1577 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1578 const struct path *path); 1579 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1580 size_t len, loff_t off); 1581 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1582 const char *data, size_t len, loff_t off); 1583 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1584 unsigned int flags); 1585 1586 static struct dquot __rcu **ext4_get_dquots(struct inode *inode) 1587 { 1588 return EXT4_I(inode)->i_dquot; 1589 } 1590 1591 static const struct dquot_operations ext4_quota_operations = { 1592 .get_reserved_space = ext4_get_reserved_space, 1593 .write_dquot = ext4_write_dquot, 1594 .acquire_dquot = ext4_acquire_dquot, 1595 .release_dquot = ext4_release_dquot, 1596 .mark_dirty = ext4_mark_dquot_dirty, 1597 .write_info = ext4_write_info, 1598 .alloc_dquot = dquot_alloc, 1599 .destroy_dquot = dquot_destroy, 1600 .get_projid = ext4_get_projid, 1601 .get_inode_usage = ext4_get_inode_usage, 1602 .get_next_id = dquot_get_next_id, 1603 }; 1604 1605 static const struct quotactl_ops ext4_qctl_operations = { 1606 .quota_on = ext4_quota_on, 1607 .quota_off = ext4_quota_off, 1608 .quota_sync = dquot_quota_sync, 1609 .get_state = dquot_get_state, 1610 .set_info = dquot_set_dqinfo, 1611 .get_dqblk = dquot_get_dqblk, 1612 .set_dqblk = dquot_set_dqblk, 1613 .get_nextdqblk = dquot_get_next_dqblk, 1614 }; 1615 #endif 1616 1617 static const struct super_operations ext4_sops = { 1618 .alloc_inode = ext4_alloc_inode, 1619 .free_inode = ext4_free_in_core_inode, 1620 .destroy_inode = ext4_destroy_inode, 1621 .write_inode = ext4_write_inode, 1622 .dirty_inode = ext4_dirty_inode, 1623 .drop_inode = ext4_drop_inode, 1624 .evict_inode = ext4_evict_inode, 1625 .put_super = ext4_put_super, 1626 .sync_fs = ext4_sync_fs, 1627 .freeze_fs = ext4_freeze, 1628 .unfreeze_fs = ext4_unfreeze, 1629 .statfs = ext4_statfs, 1630 .show_options = ext4_show_options, 1631 .shutdown = ext4_shutdown, 1632 #ifdef CONFIG_QUOTA 1633 .quota_read = ext4_quota_read, 1634 .quota_write = ext4_quota_write, 1635 .get_dquots = ext4_get_dquots, 1636 #endif 1637 }; 1638 1639 static const struct export_operations ext4_export_ops = { 1640 .encode_fh = generic_encode_ino32_fh, 1641 .fh_to_dentry = ext4_fh_to_dentry, 1642 .fh_to_parent = ext4_fh_to_parent, 1643 .get_parent = ext4_get_parent, 1644 .commit_metadata = ext4_nfs_commit_metadata, 1645 }; 1646 1647 enum { 1648 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1649 Opt_resgid, Opt_resuid, Opt_sb, 1650 Opt_nouid32, Opt_debug, Opt_removed, 1651 Opt_user_xattr, Opt_acl, 1652 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1653 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1654 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1655 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1656 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1657 Opt_inlinecrypt, 1658 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1659 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1660 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1661 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1662 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1663 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1664 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1665 Opt_inode_readahead_blks, Opt_journal_ioprio, 1666 Opt_dioread_nolock, Opt_dioread_lock, 1667 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1668 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1669 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1670 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1671 #ifdef CONFIG_EXT4_DEBUG 1672 Opt_fc_debug_max_replay, Opt_fc_debug_force 1673 #endif 1674 }; 1675 1676 static const struct constant_table ext4_param_errors[] = { 1677 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1678 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1679 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1680 {} 1681 }; 1682 1683 static const struct constant_table ext4_param_data[] = { 1684 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1685 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1686 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1687 {} 1688 }; 1689 1690 static const struct constant_table ext4_param_data_err[] = { 1691 {"abort", Opt_data_err_abort}, 1692 {"ignore", Opt_data_err_ignore}, 1693 {} 1694 }; 1695 1696 static const struct constant_table ext4_param_jqfmt[] = { 1697 {"vfsold", QFMT_VFS_OLD}, 1698 {"vfsv0", QFMT_VFS_V0}, 1699 {"vfsv1", QFMT_VFS_V1}, 1700 {} 1701 }; 1702 1703 static const struct constant_table ext4_param_dax[] = { 1704 {"always", Opt_dax_always}, 1705 {"inode", Opt_dax_inode}, 1706 {"never", Opt_dax_never}, 1707 {} 1708 }; 1709 1710 /* 1711 * Mount option specification 1712 * We don't use fsparam_flag_no because of the way we set the 1713 * options and the way we show them in _ext4_show_options(). To 1714 * keep the changes to a minimum, let's keep the negative options 1715 * separate for now. 1716 */ 1717 static const struct fs_parameter_spec ext4_param_specs[] = { 1718 fsparam_flag ("bsddf", Opt_bsd_df), 1719 fsparam_flag ("minixdf", Opt_minix_df), 1720 fsparam_flag ("grpid", Opt_grpid), 1721 fsparam_flag ("bsdgroups", Opt_grpid), 1722 fsparam_flag ("nogrpid", Opt_nogrpid), 1723 fsparam_flag ("sysvgroups", Opt_nogrpid), 1724 fsparam_gid ("resgid", Opt_resgid), 1725 fsparam_uid ("resuid", Opt_resuid), 1726 fsparam_u32 ("sb", Opt_sb), 1727 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1728 fsparam_flag ("nouid32", Opt_nouid32), 1729 fsparam_flag ("debug", Opt_debug), 1730 fsparam_flag ("oldalloc", Opt_removed), 1731 fsparam_flag ("orlov", Opt_removed), 1732 fsparam_flag ("user_xattr", Opt_user_xattr), 1733 fsparam_flag ("acl", Opt_acl), 1734 fsparam_flag ("norecovery", Opt_noload), 1735 fsparam_flag ("noload", Opt_noload), 1736 fsparam_flag ("bh", Opt_removed), 1737 fsparam_flag ("nobh", Opt_removed), 1738 fsparam_u32 ("commit", Opt_commit), 1739 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1740 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1741 fsparam_u32 ("journal_dev", Opt_journal_dev), 1742 fsparam_bdev ("journal_path", Opt_journal_path), 1743 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1744 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1745 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1746 fsparam_flag ("abort", Opt_abort), 1747 fsparam_enum ("data", Opt_data, ext4_param_data), 1748 fsparam_enum ("data_err", Opt_data_err, 1749 ext4_param_data_err), 1750 fsparam_string_empty 1751 ("usrjquota", Opt_usrjquota), 1752 fsparam_string_empty 1753 ("grpjquota", Opt_grpjquota), 1754 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1755 fsparam_flag ("grpquota", Opt_grpquota), 1756 fsparam_flag ("quota", Opt_quota), 1757 fsparam_flag ("noquota", Opt_noquota), 1758 fsparam_flag ("usrquota", Opt_usrquota), 1759 fsparam_flag ("prjquota", Opt_prjquota), 1760 fsparam_flag ("barrier", Opt_barrier), 1761 fsparam_u32 ("barrier", Opt_barrier), 1762 fsparam_flag ("nobarrier", Opt_nobarrier), 1763 fsparam_flag ("i_version", Opt_removed), 1764 fsparam_flag ("dax", Opt_dax), 1765 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1766 fsparam_u32 ("stripe", Opt_stripe), 1767 fsparam_flag ("delalloc", Opt_delalloc), 1768 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1769 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1770 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1771 fsparam_u32 ("debug_want_extra_isize", 1772 Opt_debug_want_extra_isize), 1773 fsparam_flag ("mblk_io_submit", Opt_removed), 1774 fsparam_flag ("nomblk_io_submit", Opt_removed), 1775 fsparam_flag ("block_validity", Opt_block_validity), 1776 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1777 fsparam_u32 ("inode_readahead_blks", 1778 Opt_inode_readahead_blks), 1779 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1780 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1781 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1782 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1783 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1784 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1785 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1786 fsparam_flag ("discard", Opt_discard), 1787 fsparam_flag ("nodiscard", Opt_nodiscard), 1788 fsparam_u32 ("init_itable", Opt_init_itable), 1789 fsparam_flag ("init_itable", Opt_init_itable), 1790 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1791 #ifdef CONFIG_EXT4_DEBUG 1792 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1793 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1794 #endif 1795 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1796 fsparam_flag ("test_dummy_encryption", 1797 Opt_test_dummy_encryption), 1798 fsparam_string ("test_dummy_encryption", 1799 Opt_test_dummy_encryption), 1800 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1801 fsparam_flag ("nombcache", Opt_nombcache), 1802 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1803 fsparam_flag ("prefetch_block_bitmaps", 1804 Opt_removed), 1805 fsparam_flag ("no_prefetch_block_bitmaps", 1806 Opt_no_prefetch_block_bitmaps), 1807 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1808 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1809 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1810 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1811 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1812 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1813 {} 1814 }; 1815 1816 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1817 1818 #define MOPT_SET 0x0001 1819 #define MOPT_CLEAR 0x0002 1820 #define MOPT_NOSUPPORT 0x0004 1821 #define MOPT_EXPLICIT 0x0008 1822 #ifdef CONFIG_QUOTA 1823 #define MOPT_Q 0 1824 #define MOPT_QFMT 0x0010 1825 #else 1826 #define MOPT_Q MOPT_NOSUPPORT 1827 #define MOPT_QFMT MOPT_NOSUPPORT 1828 #endif 1829 #define MOPT_NO_EXT2 0x0020 1830 #define MOPT_NO_EXT3 0x0040 1831 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1832 #define MOPT_SKIP 0x0080 1833 #define MOPT_2 0x0100 1834 1835 static const struct mount_opts { 1836 int token; 1837 int mount_opt; 1838 int flags; 1839 } ext4_mount_opts[] = { 1840 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1841 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1842 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1843 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1844 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1845 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1846 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1847 MOPT_EXT4_ONLY | MOPT_SET}, 1848 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1849 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1850 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1851 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1852 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1853 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1854 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1855 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1856 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1857 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1858 {Opt_commit, 0, MOPT_NO_EXT2}, 1859 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1860 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1861 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1862 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1863 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1864 EXT4_MOUNT_JOURNAL_CHECKSUM), 1865 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1866 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1867 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1868 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1869 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1870 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1871 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1872 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1873 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1874 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1875 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1876 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1877 {Opt_data, 0, MOPT_NO_EXT2}, 1878 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1879 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1880 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1881 #else 1882 {Opt_acl, 0, MOPT_NOSUPPORT}, 1883 #endif 1884 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1885 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1886 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1887 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1888 MOPT_SET | MOPT_Q}, 1889 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1890 MOPT_SET | MOPT_Q}, 1891 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1892 MOPT_SET | MOPT_Q}, 1893 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1894 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1895 MOPT_CLEAR | MOPT_Q}, 1896 {Opt_usrjquota, 0, MOPT_Q}, 1897 {Opt_grpjquota, 0, MOPT_Q}, 1898 {Opt_jqfmt, 0, MOPT_QFMT}, 1899 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1900 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1901 MOPT_SET}, 1902 #ifdef CONFIG_EXT4_DEBUG 1903 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1904 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1905 #endif 1906 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2}, 1907 {Opt_err, 0, 0} 1908 }; 1909 1910 #if IS_ENABLED(CONFIG_UNICODE) 1911 static const struct ext4_sb_encodings { 1912 __u16 magic; 1913 char *name; 1914 unsigned int version; 1915 } ext4_sb_encoding_map[] = { 1916 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1917 }; 1918 1919 static const struct ext4_sb_encodings * 1920 ext4_sb_read_encoding(const struct ext4_super_block *es) 1921 { 1922 __u16 magic = le16_to_cpu(es->s_encoding); 1923 int i; 1924 1925 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1926 if (magic == ext4_sb_encoding_map[i].magic) 1927 return &ext4_sb_encoding_map[i]; 1928 1929 return NULL; 1930 } 1931 #endif 1932 1933 #define EXT4_SPEC_JQUOTA (1 << 0) 1934 #define EXT4_SPEC_JQFMT (1 << 1) 1935 #define EXT4_SPEC_DATAJ (1 << 2) 1936 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1937 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1938 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1939 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 1940 #define EXT4_SPEC_s_max_batch_time (1 << 8) 1941 #define EXT4_SPEC_s_min_batch_time (1 << 9) 1942 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 1943 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 1944 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 1945 #define EXT4_SPEC_s_stripe (1 << 13) 1946 #define EXT4_SPEC_s_resuid (1 << 14) 1947 #define EXT4_SPEC_s_resgid (1 << 15) 1948 #define EXT4_SPEC_s_commit_interval (1 << 16) 1949 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 1950 #define EXT4_SPEC_s_sb_block (1 << 18) 1951 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 1952 1953 struct ext4_fs_context { 1954 char *s_qf_names[EXT4_MAXQUOTAS]; 1955 struct fscrypt_dummy_policy dummy_enc_policy; 1956 int s_jquota_fmt; /* Format of quota to use */ 1957 #ifdef CONFIG_EXT4_DEBUG 1958 int s_fc_debug_max_replay; 1959 #endif 1960 unsigned short qname_spec; 1961 unsigned long vals_s_flags; /* Bits to set in s_flags */ 1962 unsigned long mask_s_flags; /* Bits changed in s_flags */ 1963 unsigned long journal_devnum; 1964 unsigned long s_commit_interval; 1965 unsigned long s_stripe; 1966 unsigned int s_inode_readahead_blks; 1967 unsigned int s_want_extra_isize; 1968 unsigned int s_li_wait_mult; 1969 unsigned int s_max_dir_size_kb; 1970 unsigned int journal_ioprio; 1971 unsigned int vals_s_mount_opt; 1972 unsigned int mask_s_mount_opt; 1973 unsigned int vals_s_mount_opt2; 1974 unsigned int mask_s_mount_opt2; 1975 unsigned int opt_flags; /* MOPT flags */ 1976 unsigned int spec; 1977 u32 s_max_batch_time; 1978 u32 s_min_batch_time; 1979 kuid_t s_resuid; 1980 kgid_t s_resgid; 1981 ext4_fsblk_t s_sb_block; 1982 }; 1983 1984 static void ext4_fc_free(struct fs_context *fc) 1985 { 1986 struct ext4_fs_context *ctx = fc->fs_private; 1987 int i; 1988 1989 if (!ctx) 1990 return; 1991 1992 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1993 kfree(ctx->s_qf_names[i]); 1994 1995 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 1996 kfree(ctx); 1997 } 1998 1999 int ext4_init_fs_context(struct fs_context *fc) 2000 { 2001 struct ext4_fs_context *ctx; 2002 2003 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2004 if (!ctx) 2005 return -ENOMEM; 2006 2007 fc->fs_private = ctx; 2008 fc->ops = &ext4_context_ops; 2009 2010 return 0; 2011 } 2012 2013 #ifdef CONFIG_QUOTA 2014 /* 2015 * Note the name of the specified quota file. 2016 */ 2017 static int note_qf_name(struct fs_context *fc, int qtype, 2018 struct fs_parameter *param) 2019 { 2020 struct ext4_fs_context *ctx = fc->fs_private; 2021 char *qname; 2022 2023 if (param->size < 1) { 2024 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 2025 return -EINVAL; 2026 } 2027 if (strchr(param->string, '/')) { 2028 ext4_msg(NULL, KERN_ERR, 2029 "quotafile must be on filesystem root"); 2030 return -EINVAL; 2031 } 2032 if (ctx->s_qf_names[qtype]) { 2033 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 2034 ext4_msg(NULL, KERN_ERR, 2035 "%s quota file already specified", 2036 QTYPE2NAME(qtype)); 2037 return -EINVAL; 2038 } 2039 return 0; 2040 } 2041 2042 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 2043 if (!qname) { 2044 ext4_msg(NULL, KERN_ERR, 2045 "Not enough memory for storing quotafile name"); 2046 return -ENOMEM; 2047 } 2048 ctx->s_qf_names[qtype] = qname; 2049 ctx->qname_spec |= 1 << qtype; 2050 ctx->spec |= EXT4_SPEC_JQUOTA; 2051 return 0; 2052 } 2053 2054 /* 2055 * Clear the name of the specified quota file. 2056 */ 2057 static int unnote_qf_name(struct fs_context *fc, int qtype) 2058 { 2059 struct ext4_fs_context *ctx = fc->fs_private; 2060 2061 kfree(ctx->s_qf_names[qtype]); 2062 2063 ctx->s_qf_names[qtype] = NULL; 2064 ctx->qname_spec |= 1 << qtype; 2065 ctx->spec |= EXT4_SPEC_JQUOTA; 2066 return 0; 2067 } 2068 #endif 2069 2070 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2071 struct ext4_fs_context *ctx) 2072 { 2073 int err; 2074 2075 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2076 ext4_msg(NULL, KERN_WARNING, 2077 "test_dummy_encryption option not supported"); 2078 return -EINVAL; 2079 } 2080 err = fscrypt_parse_test_dummy_encryption(param, 2081 &ctx->dummy_enc_policy); 2082 if (err == -EINVAL) { 2083 ext4_msg(NULL, KERN_WARNING, 2084 "Value of option \"%s\" is unrecognized", param->key); 2085 } else if (err == -EEXIST) { 2086 ext4_msg(NULL, KERN_WARNING, 2087 "Conflicting test_dummy_encryption options"); 2088 return -EINVAL; 2089 } 2090 return err; 2091 } 2092 2093 #define EXT4_SET_CTX(name) \ 2094 static inline void ctx_set_##name(struct ext4_fs_context *ctx, \ 2095 unsigned long flag) \ 2096 { \ 2097 ctx->mask_s_##name |= flag; \ 2098 ctx->vals_s_##name |= flag; \ 2099 } 2100 2101 #define EXT4_CLEAR_CTX(name) \ 2102 static inline void ctx_clear_##name(struct ext4_fs_context *ctx, \ 2103 unsigned long flag) \ 2104 { \ 2105 ctx->mask_s_##name |= flag; \ 2106 ctx->vals_s_##name &= ~flag; \ 2107 } 2108 2109 #define EXT4_TEST_CTX(name) \ 2110 static inline unsigned long \ 2111 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2112 { \ 2113 return (ctx->vals_s_##name & flag); \ 2114 } 2115 2116 EXT4_SET_CTX(flags); /* set only */ 2117 EXT4_SET_CTX(mount_opt); 2118 EXT4_CLEAR_CTX(mount_opt); 2119 EXT4_TEST_CTX(mount_opt); 2120 EXT4_SET_CTX(mount_opt2); 2121 EXT4_CLEAR_CTX(mount_opt2); 2122 EXT4_TEST_CTX(mount_opt2); 2123 2124 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2125 { 2126 struct ext4_fs_context *ctx = fc->fs_private; 2127 struct fs_parse_result result; 2128 const struct mount_opts *m; 2129 int is_remount; 2130 int token; 2131 2132 token = fs_parse(fc, ext4_param_specs, param, &result); 2133 if (token < 0) 2134 return token; 2135 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2136 2137 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2138 if (token == m->token) 2139 break; 2140 2141 ctx->opt_flags |= m->flags; 2142 2143 if (m->flags & MOPT_EXPLICIT) { 2144 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2145 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2146 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2147 ctx_set_mount_opt2(ctx, 2148 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2149 } else 2150 return -EINVAL; 2151 } 2152 2153 if (m->flags & MOPT_NOSUPPORT) { 2154 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2155 param->key); 2156 return 0; 2157 } 2158 2159 switch (token) { 2160 #ifdef CONFIG_QUOTA 2161 case Opt_usrjquota: 2162 if (!*param->string) 2163 return unnote_qf_name(fc, USRQUOTA); 2164 else 2165 return note_qf_name(fc, USRQUOTA, param); 2166 case Opt_grpjquota: 2167 if (!*param->string) 2168 return unnote_qf_name(fc, GRPQUOTA); 2169 else 2170 return note_qf_name(fc, GRPQUOTA, param); 2171 #endif 2172 case Opt_sb: 2173 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2174 ext4_msg(NULL, KERN_WARNING, 2175 "Ignoring %s option on remount", param->key); 2176 } else { 2177 ctx->s_sb_block = result.uint_32; 2178 ctx->spec |= EXT4_SPEC_s_sb_block; 2179 } 2180 return 0; 2181 case Opt_removed: 2182 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2183 param->key); 2184 return 0; 2185 case Opt_inlinecrypt: 2186 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2187 ctx_set_flags(ctx, SB_INLINECRYPT); 2188 #else 2189 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2190 #endif 2191 return 0; 2192 case Opt_errors: 2193 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2194 ctx_set_mount_opt(ctx, result.uint_32); 2195 return 0; 2196 #ifdef CONFIG_QUOTA 2197 case Opt_jqfmt: 2198 ctx->s_jquota_fmt = result.uint_32; 2199 ctx->spec |= EXT4_SPEC_JQFMT; 2200 return 0; 2201 #endif 2202 case Opt_data: 2203 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2204 ctx_set_mount_opt(ctx, result.uint_32); 2205 ctx->spec |= EXT4_SPEC_DATAJ; 2206 return 0; 2207 case Opt_commit: 2208 if (result.uint_32 == 0) 2209 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE; 2210 else if (result.uint_32 > INT_MAX / HZ) { 2211 ext4_msg(NULL, KERN_ERR, 2212 "Invalid commit interval %d, " 2213 "must be smaller than %d", 2214 result.uint_32, INT_MAX / HZ); 2215 return -EINVAL; 2216 } 2217 ctx->s_commit_interval = HZ * result.uint_32; 2218 ctx->spec |= EXT4_SPEC_s_commit_interval; 2219 return 0; 2220 case Opt_debug_want_extra_isize: 2221 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2222 ext4_msg(NULL, KERN_ERR, 2223 "Invalid want_extra_isize %d", result.uint_32); 2224 return -EINVAL; 2225 } 2226 ctx->s_want_extra_isize = result.uint_32; 2227 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2228 return 0; 2229 case Opt_max_batch_time: 2230 ctx->s_max_batch_time = result.uint_32; 2231 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2232 return 0; 2233 case Opt_min_batch_time: 2234 ctx->s_min_batch_time = result.uint_32; 2235 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2236 return 0; 2237 case Opt_inode_readahead_blks: 2238 if (result.uint_32 && 2239 (result.uint_32 > (1 << 30) || 2240 !is_power_of_2(result.uint_32))) { 2241 ext4_msg(NULL, KERN_ERR, 2242 "EXT4-fs: inode_readahead_blks must be " 2243 "0 or a power of 2 smaller than 2^31"); 2244 return -EINVAL; 2245 } 2246 ctx->s_inode_readahead_blks = result.uint_32; 2247 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2248 return 0; 2249 case Opt_init_itable: 2250 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2251 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2252 if (param->type == fs_value_is_string) 2253 ctx->s_li_wait_mult = result.uint_32; 2254 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2255 return 0; 2256 case Opt_max_dir_size_kb: 2257 ctx->s_max_dir_size_kb = result.uint_32; 2258 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2259 return 0; 2260 #ifdef CONFIG_EXT4_DEBUG 2261 case Opt_fc_debug_max_replay: 2262 ctx->s_fc_debug_max_replay = result.uint_32; 2263 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2264 return 0; 2265 #endif 2266 case Opt_stripe: 2267 ctx->s_stripe = result.uint_32; 2268 ctx->spec |= EXT4_SPEC_s_stripe; 2269 return 0; 2270 case Opt_resuid: 2271 ctx->s_resuid = result.uid; 2272 ctx->spec |= EXT4_SPEC_s_resuid; 2273 return 0; 2274 case Opt_resgid: 2275 ctx->s_resgid = result.gid; 2276 ctx->spec |= EXT4_SPEC_s_resgid; 2277 return 0; 2278 case Opt_journal_dev: 2279 if (is_remount) { 2280 ext4_msg(NULL, KERN_ERR, 2281 "Cannot specify journal on remount"); 2282 return -EINVAL; 2283 } 2284 ctx->journal_devnum = result.uint_32; 2285 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2286 return 0; 2287 case Opt_journal_path: 2288 { 2289 struct inode *journal_inode; 2290 struct path path; 2291 int error; 2292 2293 if (is_remount) { 2294 ext4_msg(NULL, KERN_ERR, 2295 "Cannot specify journal on remount"); 2296 return -EINVAL; 2297 } 2298 2299 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path); 2300 if (error) { 2301 ext4_msg(NULL, KERN_ERR, "error: could not find " 2302 "journal device path"); 2303 return -EINVAL; 2304 } 2305 2306 journal_inode = d_inode(path.dentry); 2307 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2308 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2309 path_put(&path); 2310 return 0; 2311 } 2312 case Opt_journal_ioprio: 2313 if (result.uint_32 > 7) { 2314 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2315 " (must be 0-7)"); 2316 return -EINVAL; 2317 } 2318 ctx->journal_ioprio = 2319 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2320 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2321 return 0; 2322 case Opt_test_dummy_encryption: 2323 return ext4_parse_test_dummy_encryption(param, ctx); 2324 case Opt_dax: 2325 case Opt_dax_type: 2326 #ifdef CONFIG_FS_DAX 2327 { 2328 int type = (token == Opt_dax) ? 2329 Opt_dax : result.uint_32; 2330 2331 switch (type) { 2332 case Opt_dax: 2333 case Opt_dax_always: 2334 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2335 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2336 break; 2337 case Opt_dax_never: 2338 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2339 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2340 break; 2341 case Opt_dax_inode: 2342 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2343 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2344 /* Strictly for printing options */ 2345 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2346 break; 2347 } 2348 return 0; 2349 } 2350 #else 2351 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2352 return -EINVAL; 2353 #endif 2354 case Opt_data_err: 2355 if (result.uint_32 == Opt_data_err_abort) 2356 ctx_set_mount_opt(ctx, m->mount_opt); 2357 else if (result.uint_32 == Opt_data_err_ignore) 2358 ctx_clear_mount_opt(ctx, m->mount_opt); 2359 return 0; 2360 case Opt_mb_optimize_scan: 2361 if (result.int_32 == 1) { 2362 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2363 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2364 } else if (result.int_32 == 0) { 2365 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2366 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2367 } else { 2368 ext4_msg(NULL, KERN_WARNING, 2369 "mb_optimize_scan should be set to 0 or 1."); 2370 return -EINVAL; 2371 } 2372 return 0; 2373 } 2374 2375 /* 2376 * At this point we should only be getting options requiring MOPT_SET, 2377 * or MOPT_CLEAR. Anything else is a bug 2378 */ 2379 if (m->token == Opt_err) { 2380 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2381 param->key); 2382 WARN_ON(1); 2383 return -EINVAL; 2384 } 2385 2386 else { 2387 unsigned int set = 0; 2388 2389 if ((param->type == fs_value_is_flag) || 2390 result.uint_32 > 0) 2391 set = 1; 2392 2393 if (m->flags & MOPT_CLEAR) 2394 set = !set; 2395 else if (unlikely(!(m->flags & MOPT_SET))) { 2396 ext4_msg(NULL, KERN_WARNING, 2397 "buggy handling of option %s", 2398 param->key); 2399 WARN_ON(1); 2400 return -EINVAL; 2401 } 2402 if (m->flags & MOPT_2) { 2403 if (set != 0) 2404 ctx_set_mount_opt2(ctx, m->mount_opt); 2405 else 2406 ctx_clear_mount_opt2(ctx, m->mount_opt); 2407 } else { 2408 if (set != 0) 2409 ctx_set_mount_opt(ctx, m->mount_opt); 2410 else 2411 ctx_clear_mount_opt(ctx, m->mount_opt); 2412 } 2413 } 2414 2415 return 0; 2416 } 2417 2418 static int parse_options(struct fs_context *fc, char *options) 2419 { 2420 struct fs_parameter param; 2421 int ret; 2422 char *key; 2423 2424 if (!options) 2425 return 0; 2426 2427 while ((key = strsep(&options, ",")) != NULL) { 2428 if (*key) { 2429 size_t v_len = 0; 2430 char *value = strchr(key, '='); 2431 2432 param.type = fs_value_is_flag; 2433 param.string = NULL; 2434 2435 if (value) { 2436 if (value == key) 2437 continue; 2438 2439 *value++ = 0; 2440 v_len = strlen(value); 2441 param.string = kmemdup_nul(value, v_len, 2442 GFP_KERNEL); 2443 if (!param.string) 2444 return -ENOMEM; 2445 param.type = fs_value_is_string; 2446 } 2447 2448 param.key = key; 2449 param.size = v_len; 2450 2451 ret = ext4_parse_param(fc, ¶m); 2452 kfree(param.string); 2453 if (ret < 0) 2454 return ret; 2455 } 2456 } 2457 2458 ret = ext4_validate_options(fc); 2459 if (ret < 0) 2460 return ret; 2461 2462 return 0; 2463 } 2464 2465 static int parse_apply_sb_mount_options(struct super_block *sb, 2466 struct ext4_fs_context *m_ctx) 2467 { 2468 struct ext4_sb_info *sbi = EXT4_SB(sb); 2469 char *s_mount_opts = NULL; 2470 struct ext4_fs_context *s_ctx = NULL; 2471 struct fs_context *fc = NULL; 2472 int ret = -ENOMEM; 2473 2474 if (!sbi->s_es->s_mount_opts[0]) 2475 return 0; 2476 2477 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2478 sizeof(sbi->s_es->s_mount_opts), 2479 GFP_KERNEL); 2480 if (!s_mount_opts) 2481 return ret; 2482 2483 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2484 if (!fc) 2485 goto out_free; 2486 2487 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2488 if (!s_ctx) 2489 goto out_free; 2490 2491 fc->fs_private = s_ctx; 2492 fc->s_fs_info = sbi; 2493 2494 ret = parse_options(fc, s_mount_opts); 2495 if (ret < 0) 2496 goto parse_failed; 2497 2498 ret = ext4_check_opt_consistency(fc, sb); 2499 if (ret < 0) { 2500 parse_failed: 2501 ext4_msg(sb, KERN_WARNING, 2502 "failed to parse options in superblock: %s", 2503 s_mount_opts); 2504 ret = 0; 2505 goto out_free; 2506 } 2507 2508 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2509 m_ctx->journal_devnum = s_ctx->journal_devnum; 2510 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2511 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2512 2513 ext4_apply_options(fc, sb); 2514 ret = 0; 2515 2516 out_free: 2517 if (fc) { 2518 ext4_fc_free(fc); 2519 kfree(fc); 2520 } 2521 kfree(s_mount_opts); 2522 return ret; 2523 } 2524 2525 static void ext4_apply_quota_options(struct fs_context *fc, 2526 struct super_block *sb) 2527 { 2528 #ifdef CONFIG_QUOTA 2529 bool quota_feature = ext4_has_feature_quota(sb); 2530 struct ext4_fs_context *ctx = fc->fs_private; 2531 struct ext4_sb_info *sbi = EXT4_SB(sb); 2532 char *qname; 2533 int i; 2534 2535 if (quota_feature) 2536 return; 2537 2538 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2539 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2540 if (!(ctx->qname_spec & (1 << i))) 2541 continue; 2542 2543 qname = ctx->s_qf_names[i]; /* May be NULL */ 2544 if (qname) 2545 set_opt(sb, QUOTA); 2546 ctx->s_qf_names[i] = NULL; 2547 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2548 lockdep_is_held(&sb->s_umount)); 2549 if (qname) 2550 kfree_rcu_mightsleep(qname); 2551 } 2552 } 2553 2554 if (ctx->spec & EXT4_SPEC_JQFMT) 2555 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2556 #endif 2557 } 2558 2559 /* 2560 * Check quota settings consistency. 2561 */ 2562 static int ext4_check_quota_consistency(struct fs_context *fc, 2563 struct super_block *sb) 2564 { 2565 #ifdef CONFIG_QUOTA 2566 struct ext4_fs_context *ctx = fc->fs_private; 2567 struct ext4_sb_info *sbi = EXT4_SB(sb); 2568 bool quota_feature = ext4_has_feature_quota(sb); 2569 bool quota_loaded = sb_any_quota_loaded(sb); 2570 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2571 int quota_flags, i; 2572 2573 /* 2574 * We do the test below only for project quotas. 'usrquota' and 2575 * 'grpquota' mount options are allowed even without quota feature 2576 * to support legacy quotas in quota files. 2577 */ 2578 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2579 !ext4_has_feature_project(sb)) { 2580 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2581 "Cannot enable project quota enforcement."); 2582 return -EINVAL; 2583 } 2584 2585 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2586 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2587 if (quota_loaded && 2588 ctx->mask_s_mount_opt & quota_flags && 2589 !ctx_test_mount_opt(ctx, quota_flags)) 2590 goto err_quota_change; 2591 2592 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2593 2594 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2595 if (!(ctx->qname_spec & (1 << i))) 2596 continue; 2597 2598 if (quota_loaded && 2599 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2600 goto err_jquota_change; 2601 2602 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2603 strcmp(get_qf_name(sb, sbi, i), 2604 ctx->s_qf_names[i]) != 0) 2605 goto err_jquota_specified; 2606 } 2607 2608 if (quota_feature) { 2609 ext4_msg(NULL, KERN_INFO, 2610 "Journaled quota options ignored when " 2611 "QUOTA feature is enabled"); 2612 return 0; 2613 } 2614 } 2615 2616 if (ctx->spec & EXT4_SPEC_JQFMT) { 2617 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2618 goto err_jquota_change; 2619 if (quota_feature) { 2620 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2621 "ignored when QUOTA feature is enabled"); 2622 return 0; 2623 } 2624 } 2625 2626 /* Make sure we don't mix old and new quota format */ 2627 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2628 ctx->s_qf_names[USRQUOTA]); 2629 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2630 ctx->s_qf_names[GRPQUOTA]); 2631 2632 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2633 test_opt(sb, USRQUOTA)); 2634 2635 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2636 test_opt(sb, GRPQUOTA)); 2637 2638 if (usr_qf_name) { 2639 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2640 usrquota = false; 2641 } 2642 if (grp_qf_name) { 2643 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2644 grpquota = false; 2645 } 2646 2647 if (usr_qf_name || grp_qf_name) { 2648 if (usrquota || grpquota) { 2649 ext4_msg(NULL, KERN_ERR, "old and new quota " 2650 "format mixing"); 2651 return -EINVAL; 2652 } 2653 2654 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2655 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2656 "not specified"); 2657 return -EINVAL; 2658 } 2659 } 2660 2661 return 0; 2662 2663 err_quota_change: 2664 ext4_msg(NULL, KERN_ERR, 2665 "Cannot change quota options when quota turned on"); 2666 return -EINVAL; 2667 err_jquota_change: 2668 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2669 "options when quota turned on"); 2670 return -EINVAL; 2671 err_jquota_specified: 2672 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2673 QTYPE2NAME(i)); 2674 return -EINVAL; 2675 #else 2676 return 0; 2677 #endif 2678 } 2679 2680 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2681 struct super_block *sb) 2682 { 2683 const struct ext4_fs_context *ctx = fc->fs_private; 2684 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2685 2686 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2687 return 0; 2688 2689 if (!ext4_has_feature_encrypt(sb)) { 2690 ext4_msg(NULL, KERN_WARNING, 2691 "test_dummy_encryption requires encrypt feature"); 2692 return -EINVAL; 2693 } 2694 /* 2695 * This mount option is just for testing, and it's not worthwhile to 2696 * implement the extra complexity (e.g. RCU protection) that would be 2697 * needed to allow it to be set or changed during remount. We do allow 2698 * it to be specified during remount, but only if there is no change. 2699 */ 2700 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2701 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2702 &ctx->dummy_enc_policy)) 2703 return 0; 2704 ext4_msg(NULL, KERN_WARNING, 2705 "Can't set or change test_dummy_encryption on remount"); 2706 return -EINVAL; 2707 } 2708 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2709 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2710 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2711 &ctx->dummy_enc_policy)) 2712 return 0; 2713 ext4_msg(NULL, KERN_WARNING, 2714 "Conflicting test_dummy_encryption options"); 2715 return -EINVAL; 2716 } 2717 return 0; 2718 } 2719 2720 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2721 struct super_block *sb) 2722 { 2723 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2724 /* if already set, it was already verified to be the same */ 2725 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2726 return; 2727 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2728 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2729 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2730 } 2731 2732 static int ext4_check_opt_consistency(struct fs_context *fc, 2733 struct super_block *sb) 2734 { 2735 struct ext4_fs_context *ctx = fc->fs_private; 2736 struct ext4_sb_info *sbi = fc->s_fs_info; 2737 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2738 int err; 2739 2740 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2741 ext4_msg(NULL, KERN_ERR, 2742 "Mount option(s) incompatible with ext2"); 2743 return -EINVAL; 2744 } 2745 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2746 ext4_msg(NULL, KERN_ERR, 2747 "Mount option(s) incompatible with ext3"); 2748 return -EINVAL; 2749 } 2750 2751 if (ctx->s_want_extra_isize > 2752 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2753 ext4_msg(NULL, KERN_ERR, 2754 "Invalid want_extra_isize %d", 2755 ctx->s_want_extra_isize); 2756 return -EINVAL; 2757 } 2758 2759 err = ext4_check_test_dummy_encryption(fc, sb); 2760 if (err) 2761 return err; 2762 2763 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2764 if (!sbi->s_journal) { 2765 ext4_msg(NULL, KERN_WARNING, 2766 "Remounting file system with no journal " 2767 "so ignoring journalled data option"); 2768 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2769 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2770 test_opt(sb, DATA_FLAGS)) { 2771 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2772 "on remount"); 2773 return -EINVAL; 2774 } 2775 } 2776 2777 if (is_remount) { 2778 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2779 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2780 ext4_msg(NULL, KERN_ERR, "can't mount with " 2781 "both data=journal and dax"); 2782 return -EINVAL; 2783 } 2784 2785 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2786 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2787 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2788 fail_dax_change_remount: 2789 ext4_msg(NULL, KERN_ERR, "can't change " 2790 "dax mount option while remounting"); 2791 return -EINVAL; 2792 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2793 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2794 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2795 goto fail_dax_change_remount; 2796 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2797 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2798 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2799 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2800 goto fail_dax_change_remount; 2801 } 2802 } 2803 2804 return ext4_check_quota_consistency(fc, sb); 2805 } 2806 2807 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2808 { 2809 struct ext4_fs_context *ctx = fc->fs_private; 2810 struct ext4_sb_info *sbi = fc->s_fs_info; 2811 2812 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2813 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2814 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2815 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2816 sb->s_flags &= ~ctx->mask_s_flags; 2817 sb->s_flags |= ctx->vals_s_flags; 2818 2819 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2820 APPLY(s_commit_interval); 2821 APPLY(s_stripe); 2822 APPLY(s_max_batch_time); 2823 APPLY(s_min_batch_time); 2824 APPLY(s_want_extra_isize); 2825 APPLY(s_inode_readahead_blks); 2826 APPLY(s_max_dir_size_kb); 2827 APPLY(s_li_wait_mult); 2828 APPLY(s_resgid); 2829 APPLY(s_resuid); 2830 2831 #ifdef CONFIG_EXT4_DEBUG 2832 APPLY(s_fc_debug_max_replay); 2833 #endif 2834 2835 ext4_apply_quota_options(fc, sb); 2836 ext4_apply_test_dummy_encryption(ctx, sb); 2837 } 2838 2839 2840 static int ext4_validate_options(struct fs_context *fc) 2841 { 2842 #ifdef CONFIG_QUOTA 2843 struct ext4_fs_context *ctx = fc->fs_private; 2844 char *usr_qf_name, *grp_qf_name; 2845 2846 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2847 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2848 2849 if (usr_qf_name || grp_qf_name) { 2850 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2851 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2852 2853 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2854 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2855 2856 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2857 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2858 ext4_msg(NULL, KERN_ERR, "old and new quota " 2859 "format mixing"); 2860 return -EINVAL; 2861 } 2862 } 2863 #endif 2864 return 1; 2865 } 2866 2867 static inline void ext4_show_quota_options(struct seq_file *seq, 2868 struct super_block *sb) 2869 { 2870 #if defined(CONFIG_QUOTA) 2871 struct ext4_sb_info *sbi = EXT4_SB(sb); 2872 char *usr_qf_name, *grp_qf_name; 2873 2874 if (sbi->s_jquota_fmt) { 2875 char *fmtname = ""; 2876 2877 switch (sbi->s_jquota_fmt) { 2878 case QFMT_VFS_OLD: 2879 fmtname = "vfsold"; 2880 break; 2881 case QFMT_VFS_V0: 2882 fmtname = "vfsv0"; 2883 break; 2884 case QFMT_VFS_V1: 2885 fmtname = "vfsv1"; 2886 break; 2887 } 2888 seq_printf(seq, ",jqfmt=%s", fmtname); 2889 } 2890 2891 rcu_read_lock(); 2892 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2893 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2894 if (usr_qf_name) 2895 seq_show_option(seq, "usrjquota", usr_qf_name); 2896 if (grp_qf_name) 2897 seq_show_option(seq, "grpjquota", grp_qf_name); 2898 rcu_read_unlock(); 2899 #endif 2900 } 2901 2902 static const char *token2str(int token) 2903 { 2904 const struct fs_parameter_spec *spec; 2905 2906 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2907 if (spec->opt == token && !spec->type) 2908 break; 2909 return spec->name; 2910 } 2911 2912 /* 2913 * Show an option if 2914 * - it's set to a non-default value OR 2915 * - if the per-sb default is different from the global default 2916 */ 2917 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2918 int nodefs) 2919 { 2920 struct ext4_sb_info *sbi = EXT4_SB(sb); 2921 struct ext4_super_block *es = sbi->s_es; 2922 int def_errors; 2923 const struct mount_opts *m; 2924 char sep = nodefs ? '\n' : ','; 2925 2926 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2927 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2928 2929 if (sbi->s_sb_block != 1) 2930 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2931 2932 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2933 int want_set = m->flags & MOPT_SET; 2934 int opt_2 = m->flags & MOPT_2; 2935 unsigned int mount_opt, def_mount_opt; 2936 2937 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2938 m->flags & MOPT_SKIP) 2939 continue; 2940 2941 if (opt_2) { 2942 mount_opt = sbi->s_mount_opt2; 2943 def_mount_opt = sbi->s_def_mount_opt2; 2944 } else { 2945 mount_opt = sbi->s_mount_opt; 2946 def_mount_opt = sbi->s_def_mount_opt; 2947 } 2948 /* skip if same as the default */ 2949 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt))) 2950 continue; 2951 /* select Opt_noFoo vs Opt_Foo */ 2952 if ((want_set && 2953 (mount_opt & m->mount_opt) != m->mount_opt) || 2954 (!want_set && (mount_opt & m->mount_opt))) 2955 continue; 2956 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2957 } 2958 2959 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2960 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2961 SEQ_OPTS_PRINT("resuid=%u", 2962 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2963 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2964 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2965 SEQ_OPTS_PRINT("resgid=%u", 2966 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2967 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2968 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2969 SEQ_OPTS_PUTS("errors=remount-ro"); 2970 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2971 SEQ_OPTS_PUTS("errors=continue"); 2972 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2973 SEQ_OPTS_PUTS("errors=panic"); 2974 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2975 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2976 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2977 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2978 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2979 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2980 if (nodefs || sbi->s_stripe) 2981 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2982 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2983 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 2984 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2985 SEQ_OPTS_PUTS("data=journal"); 2986 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2987 SEQ_OPTS_PUTS("data=ordered"); 2988 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2989 SEQ_OPTS_PUTS("data=writeback"); 2990 } 2991 if (nodefs || 2992 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2993 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2994 sbi->s_inode_readahead_blks); 2995 2996 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2997 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2998 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2999 if (nodefs || sbi->s_max_dir_size_kb) 3000 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 3001 if (test_opt(sb, DATA_ERR_ABORT)) 3002 SEQ_OPTS_PUTS("data_err=abort"); 3003 3004 fscrypt_show_test_dummy_encryption(seq, sep, sb); 3005 3006 if (sb->s_flags & SB_INLINECRYPT) 3007 SEQ_OPTS_PUTS("inlinecrypt"); 3008 3009 if (test_opt(sb, DAX_ALWAYS)) { 3010 if (IS_EXT2_SB(sb)) 3011 SEQ_OPTS_PUTS("dax"); 3012 else 3013 SEQ_OPTS_PUTS("dax=always"); 3014 } else if (test_opt2(sb, DAX_NEVER)) { 3015 SEQ_OPTS_PUTS("dax=never"); 3016 } else if (test_opt2(sb, DAX_INODE)) { 3017 SEQ_OPTS_PUTS("dax=inode"); 3018 } 3019 3020 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3021 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3022 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 3023 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 3024 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 3025 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 3026 } 3027 3028 ext4_show_quota_options(seq, sb); 3029 return 0; 3030 } 3031 3032 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 3033 { 3034 return _ext4_show_options(seq, root->d_sb, 0); 3035 } 3036 3037 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3038 { 3039 struct super_block *sb = seq->private; 3040 int rc; 3041 3042 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3043 rc = _ext4_show_options(seq, sb, 1); 3044 seq_puts(seq, "\n"); 3045 return rc; 3046 } 3047 3048 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3049 int read_only) 3050 { 3051 struct ext4_sb_info *sbi = EXT4_SB(sb); 3052 int err = 0; 3053 3054 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3055 ext4_msg(sb, KERN_ERR, "revision level too high, " 3056 "forcing read-only mode"); 3057 err = -EROFS; 3058 goto done; 3059 } 3060 if (read_only) 3061 goto done; 3062 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3063 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3064 "running e2fsck is recommended"); 3065 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3066 ext4_msg(sb, KERN_WARNING, 3067 "warning: mounting fs with errors, " 3068 "running e2fsck is recommended"); 3069 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3070 le16_to_cpu(es->s_mnt_count) >= 3071 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3072 ext4_msg(sb, KERN_WARNING, 3073 "warning: maximal mount count reached, " 3074 "running e2fsck is recommended"); 3075 else if (le32_to_cpu(es->s_checkinterval) && 3076 (ext4_get_tstamp(es, s_lastcheck) + 3077 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3078 ext4_msg(sb, KERN_WARNING, 3079 "warning: checktime reached, " 3080 "running e2fsck is recommended"); 3081 if (!sbi->s_journal) 3082 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3083 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3084 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3085 le16_add_cpu(&es->s_mnt_count, 1); 3086 ext4_update_tstamp(es, s_mtime); 3087 if (sbi->s_journal) { 3088 ext4_set_feature_journal_needs_recovery(sb); 3089 if (ext4_has_feature_orphan_file(sb)) 3090 ext4_set_feature_orphan_present(sb); 3091 } 3092 3093 err = ext4_commit_super(sb); 3094 done: 3095 if (test_opt(sb, DEBUG)) 3096 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3097 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3098 sb->s_blocksize, 3099 sbi->s_groups_count, 3100 EXT4_BLOCKS_PER_GROUP(sb), 3101 EXT4_INODES_PER_GROUP(sb), 3102 sbi->s_mount_opt, sbi->s_mount_opt2); 3103 return err; 3104 } 3105 3106 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3107 { 3108 struct ext4_sb_info *sbi = EXT4_SB(sb); 3109 struct flex_groups **old_groups, **new_groups; 3110 int size, i, j; 3111 3112 if (!sbi->s_log_groups_per_flex) 3113 return 0; 3114 3115 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3116 if (size <= sbi->s_flex_groups_allocated) 3117 return 0; 3118 3119 new_groups = kvzalloc(roundup_pow_of_two(size * 3120 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3121 if (!new_groups) { 3122 ext4_msg(sb, KERN_ERR, 3123 "not enough memory for %d flex group pointers", size); 3124 return -ENOMEM; 3125 } 3126 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3127 new_groups[i] = kvzalloc(roundup_pow_of_two( 3128 sizeof(struct flex_groups)), 3129 GFP_KERNEL); 3130 if (!new_groups[i]) { 3131 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3132 kvfree(new_groups[j]); 3133 kvfree(new_groups); 3134 ext4_msg(sb, KERN_ERR, 3135 "not enough memory for %d flex groups", size); 3136 return -ENOMEM; 3137 } 3138 } 3139 rcu_read_lock(); 3140 old_groups = rcu_dereference(sbi->s_flex_groups); 3141 if (old_groups) 3142 memcpy(new_groups, old_groups, 3143 (sbi->s_flex_groups_allocated * 3144 sizeof(struct flex_groups *))); 3145 rcu_read_unlock(); 3146 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3147 sbi->s_flex_groups_allocated = size; 3148 if (old_groups) 3149 ext4_kvfree_array_rcu(old_groups); 3150 return 0; 3151 } 3152 3153 static int ext4_fill_flex_info(struct super_block *sb) 3154 { 3155 struct ext4_sb_info *sbi = EXT4_SB(sb); 3156 struct ext4_group_desc *gdp = NULL; 3157 struct flex_groups *fg; 3158 ext4_group_t flex_group; 3159 int i, err; 3160 3161 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3162 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3163 sbi->s_log_groups_per_flex = 0; 3164 return 1; 3165 } 3166 3167 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3168 if (err) 3169 goto failed; 3170 3171 for (i = 0; i < sbi->s_groups_count; i++) { 3172 gdp = ext4_get_group_desc(sb, i, NULL); 3173 3174 flex_group = ext4_flex_group(sbi, i); 3175 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3176 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3177 atomic64_add(ext4_free_group_clusters(sb, gdp), 3178 &fg->free_clusters); 3179 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3180 } 3181 3182 return 1; 3183 failed: 3184 return 0; 3185 } 3186 3187 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3188 struct ext4_group_desc *gdp) 3189 { 3190 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3191 __u16 crc = 0; 3192 __le32 le_group = cpu_to_le32(block_group); 3193 struct ext4_sb_info *sbi = EXT4_SB(sb); 3194 3195 if (ext4_has_metadata_csum(sbi->s_sb)) { 3196 /* Use new metadata_csum algorithm */ 3197 __u32 csum32; 3198 __u16 dummy_csum = 0; 3199 3200 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 3201 sizeof(le_group)); 3202 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 3203 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 3204 sizeof(dummy_csum)); 3205 offset += sizeof(dummy_csum); 3206 if (offset < sbi->s_desc_size) 3207 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 3208 sbi->s_desc_size - offset); 3209 3210 crc = csum32 & 0xFFFF; 3211 goto out; 3212 } 3213 3214 /* old crc16 code */ 3215 if (!ext4_has_feature_gdt_csum(sb)) 3216 return 0; 3217 3218 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3219 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3220 crc = crc16(crc, (__u8 *)gdp, offset); 3221 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3222 /* for checksum of struct ext4_group_desc do the rest...*/ 3223 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size) 3224 crc = crc16(crc, (__u8 *)gdp + offset, 3225 sbi->s_desc_size - offset); 3226 3227 out: 3228 return cpu_to_le16(crc); 3229 } 3230 3231 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3232 struct ext4_group_desc *gdp) 3233 { 3234 if (ext4_has_group_desc_csum(sb) && 3235 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3236 return 0; 3237 3238 return 1; 3239 } 3240 3241 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3242 struct ext4_group_desc *gdp) 3243 { 3244 if (!ext4_has_group_desc_csum(sb)) 3245 return; 3246 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3247 } 3248 3249 /* Called at mount-time, super-block is locked */ 3250 static int ext4_check_descriptors(struct super_block *sb, 3251 ext4_fsblk_t sb_block, 3252 ext4_group_t *first_not_zeroed) 3253 { 3254 struct ext4_sb_info *sbi = EXT4_SB(sb); 3255 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3256 ext4_fsblk_t last_block; 3257 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3258 ext4_fsblk_t block_bitmap; 3259 ext4_fsblk_t inode_bitmap; 3260 ext4_fsblk_t inode_table; 3261 int flexbg_flag = 0; 3262 ext4_group_t i, grp = sbi->s_groups_count; 3263 3264 if (ext4_has_feature_flex_bg(sb)) 3265 flexbg_flag = 1; 3266 3267 ext4_debug("Checking group descriptors"); 3268 3269 for (i = 0; i < sbi->s_groups_count; i++) { 3270 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3271 3272 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3273 last_block = ext4_blocks_count(sbi->s_es) - 1; 3274 else 3275 last_block = first_block + 3276 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3277 3278 if ((grp == sbi->s_groups_count) && 3279 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3280 grp = i; 3281 3282 block_bitmap = ext4_block_bitmap(sb, gdp); 3283 if (block_bitmap == sb_block) { 3284 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3285 "Block bitmap for group %u overlaps " 3286 "superblock", i); 3287 if (!sb_rdonly(sb)) 3288 return 0; 3289 } 3290 if (block_bitmap >= sb_block + 1 && 3291 block_bitmap <= last_bg_block) { 3292 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3293 "Block bitmap for group %u overlaps " 3294 "block group descriptors", i); 3295 if (!sb_rdonly(sb)) 3296 return 0; 3297 } 3298 if (block_bitmap < first_block || block_bitmap > last_block) { 3299 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3300 "Block bitmap for group %u not in group " 3301 "(block %llu)!", i, block_bitmap); 3302 return 0; 3303 } 3304 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3305 if (inode_bitmap == sb_block) { 3306 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3307 "Inode bitmap for group %u overlaps " 3308 "superblock", i); 3309 if (!sb_rdonly(sb)) 3310 return 0; 3311 } 3312 if (inode_bitmap >= sb_block + 1 && 3313 inode_bitmap <= last_bg_block) { 3314 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3315 "Inode bitmap for group %u overlaps " 3316 "block group descriptors", i); 3317 if (!sb_rdonly(sb)) 3318 return 0; 3319 } 3320 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3321 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3322 "Inode bitmap for group %u not in group " 3323 "(block %llu)!", i, inode_bitmap); 3324 return 0; 3325 } 3326 inode_table = ext4_inode_table(sb, gdp); 3327 if (inode_table == sb_block) { 3328 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3329 "Inode table for group %u overlaps " 3330 "superblock", i); 3331 if (!sb_rdonly(sb)) 3332 return 0; 3333 } 3334 if (inode_table >= sb_block + 1 && 3335 inode_table <= last_bg_block) { 3336 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3337 "Inode table for group %u overlaps " 3338 "block group descriptors", i); 3339 if (!sb_rdonly(sb)) 3340 return 0; 3341 } 3342 if (inode_table < first_block || 3343 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3344 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3345 "Inode table for group %u not in group " 3346 "(block %llu)!", i, inode_table); 3347 return 0; 3348 } 3349 ext4_lock_group(sb, i); 3350 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3351 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3352 "Checksum for group %u failed (%u!=%u)", 3353 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3354 gdp)), le16_to_cpu(gdp->bg_checksum)); 3355 if (!sb_rdonly(sb)) { 3356 ext4_unlock_group(sb, i); 3357 return 0; 3358 } 3359 } 3360 ext4_unlock_group(sb, i); 3361 if (!flexbg_flag) 3362 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3363 } 3364 if (NULL != first_not_zeroed) 3365 *first_not_zeroed = grp; 3366 return 1; 3367 } 3368 3369 /* 3370 * Maximal extent format file size. 3371 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3372 * extent format containers, within a sector_t, and within i_blocks 3373 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3374 * so that won't be a limiting factor. 3375 * 3376 * However there is other limiting factor. We do store extents in the form 3377 * of starting block and length, hence the resulting length of the extent 3378 * covering maximum file size must fit into on-disk format containers as 3379 * well. Given that length is always by 1 unit bigger than max unit (because 3380 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3381 * 3382 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3383 */ 3384 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3385 { 3386 loff_t res; 3387 loff_t upper_limit = MAX_LFS_FILESIZE; 3388 3389 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3390 3391 if (!has_huge_files) { 3392 upper_limit = (1LL << 32) - 1; 3393 3394 /* total blocks in file system block size */ 3395 upper_limit >>= (blkbits - 9); 3396 upper_limit <<= blkbits; 3397 } 3398 3399 /* 3400 * 32-bit extent-start container, ee_block. We lower the maxbytes 3401 * by one fs block, so ee_len can cover the extent of maximum file 3402 * size 3403 */ 3404 res = (1LL << 32) - 1; 3405 res <<= blkbits; 3406 3407 /* Sanity check against vm- & vfs- imposed limits */ 3408 if (res > upper_limit) 3409 res = upper_limit; 3410 3411 return res; 3412 } 3413 3414 /* 3415 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3416 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3417 * We need to be 1 filesystem block less than the 2^48 sector limit. 3418 */ 3419 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3420 { 3421 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3422 int meta_blocks; 3423 unsigned int ppb = 1 << (bits - 2); 3424 3425 /* 3426 * This is calculated to be the largest file size for a dense, block 3427 * mapped file such that the file's total number of 512-byte sectors, 3428 * including data and all indirect blocks, does not exceed (2^48 - 1). 3429 * 3430 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3431 * number of 512-byte sectors of the file. 3432 */ 3433 if (!has_huge_files) { 3434 /* 3435 * !has_huge_files or implies that the inode i_block field 3436 * represents total file blocks in 2^32 512-byte sectors == 3437 * size of vfs inode i_blocks * 8 3438 */ 3439 upper_limit = (1LL << 32) - 1; 3440 3441 /* total blocks in file system block size */ 3442 upper_limit >>= (bits - 9); 3443 3444 } else { 3445 /* 3446 * We use 48 bit ext4_inode i_blocks 3447 * With EXT4_HUGE_FILE_FL set the i_blocks 3448 * represent total number of blocks in 3449 * file system block size 3450 */ 3451 upper_limit = (1LL << 48) - 1; 3452 3453 } 3454 3455 /* Compute how many blocks we can address by block tree */ 3456 res += ppb; 3457 res += ppb * ppb; 3458 res += ((loff_t)ppb) * ppb * ppb; 3459 /* Compute how many metadata blocks are needed */ 3460 meta_blocks = 1; 3461 meta_blocks += 1 + ppb; 3462 meta_blocks += 1 + ppb + ppb * ppb; 3463 /* Does block tree limit file size? */ 3464 if (res + meta_blocks <= upper_limit) 3465 goto check_lfs; 3466 3467 res = upper_limit; 3468 /* How many metadata blocks are needed for addressing upper_limit? */ 3469 upper_limit -= EXT4_NDIR_BLOCKS; 3470 /* indirect blocks */ 3471 meta_blocks = 1; 3472 upper_limit -= ppb; 3473 /* double indirect blocks */ 3474 if (upper_limit < ppb * ppb) { 3475 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3476 res -= meta_blocks; 3477 goto check_lfs; 3478 } 3479 meta_blocks += 1 + ppb; 3480 upper_limit -= ppb * ppb; 3481 /* tripple indirect blocks for the rest */ 3482 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3483 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3484 res -= meta_blocks; 3485 check_lfs: 3486 res <<= bits; 3487 if (res > MAX_LFS_FILESIZE) 3488 res = MAX_LFS_FILESIZE; 3489 3490 return res; 3491 } 3492 3493 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3494 ext4_fsblk_t logical_sb_block, int nr) 3495 { 3496 struct ext4_sb_info *sbi = EXT4_SB(sb); 3497 ext4_group_t bg, first_meta_bg; 3498 int has_super = 0; 3499 3500 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3501 3502 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3503 return logical_sb_block + nr + 1; 3504 bg = sbi->s_desc_per_block * nr; 3505 if (ext4_bg_has_super(sb, bg)) 3506 has_super = 1; 3507 3508 /* 3509 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3510 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3511 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3512 * compensate. 3513 */ 3514 if (sb->s_blocksize == 1024 && nr == 0 && 3515 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3516 has_super++; 3517 3518 return (has_super + ext4_group_first_block_no(sb, bg)); 3519 } 3520 3521 /** 3522 * ext4_get_stripe_size: Get the stripe size. 3523 * @sbi: In memory super block info 3524 * 3525 * If we have specified it via mount option, then 3526 * use the mount option value. If the value specified at mount time is 3527 * greater than the blocks per group use the super block value. 3528 * If the super block value is greater than blocks per group return 0. 3529 * Allocator needs it be less than blocks per group. 3530 * 3531 */ 3532 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3533 { 3534 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3535 unsigned long stripe_width = 3536 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3537 int ret; 3538 3539 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3540 ret = sbi->s_stripe; 3541 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3542 ret = stripe_width; 3543 else if (stride && stride <= sbi->s_blocks_per_group) 3544 ret = stride; 3545 else 3546 ret = 0; 3547 3548 /* 3549 * If the stripe width is 1, this makes no sense and 3550 * we set it to 0 to turn off stripe handling code. 3551 */ 3552 if (ret <= 1) 3553 ret = 0; 3554 3555 return ret; 3556 } 3557 3558 /* 3559 * Check whether this filesystem can be mounted based on 3560 * the features present and the RDONLY/RDWR mount requested. 3561 * Returns 1 if this filesystem can be mounted as requested, 3562 * 0 if it cannot be. 3563 */ 3564 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3565 { 3566 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3567 ext4_msg(sb, KERN_ERR, 3568 "Couldn't mount because of " 3569 "unsupported optional features (%x)", 3570 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3571 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3572 return 0; 3573 } 3574 3575 if (!IS_ENABLED(CONFIG_UNICODE) && ext4_has_feature_casefold(sb)) { 3576 ext4_msg(sb, KERN_ERR, 3577 "Filesystem with casefold feature cannot be " 3578 "mounted without CONFIG_UNICODE"); 3579 return 0; 3580 } 3581 3582 if (readonly) 3583 return 1; 3584 3585 if (ext4_has_feature_readonly(sb)) { 3586 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3587 sb->s_flags |= SB_RDONLY; 3588 return 1; 3589 } 3590 3591 /* Check that feature set is OK for a read-write mount */ 3592 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3593 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3594 "unsupported optional features (%x)", 3595 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3596 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3597 return 0; 3598 } 3599 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3600 ext4_msg(sb, KERN_ERR, 3601 "Can't support bigalloc feature without " 3602 "extents feature\n"); 3603 return 0; 3604 } 3605 3606 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3607 if (!readonly && (ext4_has_feature_quota(sb) || 3608 ext4_has_feature_project(sb))) { 3609 ext4_msg(sb, KERN_ERR, 3610 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3611 return 0; 3612 } 3613 #endif /* CONFIG_QUOTA */ 3614 return 1; 3615 } 3616 3617 /* 3618 * This function is called once a day if we have errors logged 3619 * on the file system 3620 */ 3621 static void print_daily_error_info(struct timer_list *t) 3622 { 3623 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3624 struct super_block *sb = sbi->s_sb; 3625 struct ext4_super_block *es = sbi->s_es; 3626 3627 if (es->s_error_count) 3628 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3629 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3630 le32_to_cpu(es->s_error_count)); 3631 if (es->s_first_error_time) { 3632 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3633 sb->s_id, 3634 ext4_get_tstamp(es, s_first_error_time), 3635 (int) sizeof(es->s_first_error_func), 3636 es->s_first_error_func, 3637 le32_to_cpu(es->s_first_error_line)); 3638 if (es->s_first_error_ino) 3639 printk(KERN_CONT ": inode %u", 3640 le32_to_cpu(es->s_first_error_ino)); 3641 if (es->s_first_error_block) 3642 printk(KERN_CONT ": block %llu", (unsigned long long) 3643 le64_to_cpu(es->s_first_error_block)); 3644 printk(KERN_CONT "\n"); 3645 } 3646 if (es->s_last_error_time) { 3647 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3648 sb->s_id, 3649 ext4_get_tstamp(es, s_last_error_time), 3650 (int) sizeof(es->s_last_error_func), 3651 es->s_last_error_func, 3652 le32_to_cpu(es->s_last_error_line)); 3653 if (es->s_last_error_ino) 3654 printk(KERN_CONT ": inode %u", 3655 le32_to_cpu(es->s_last_error_ino)); 3656 if (es->s_last_error_block) 3657 printk(KERN_CONT ": block %llu", (unsigned long long) 3658 le64_to_cpu(es->s_last_error_block)); 3659 printk(KERN_CONT "\n"); 3660 } 3661 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3662 } 3663 3664 /* Find next suitable group and run ext4_init_inode_table */ 3665 static int ext4_run_li_request(struct ext4_li_request *elr) 3666 { 3667 struct ext4_group_desc *gdp = NULL; 3668 struct super_block *sb = elr->lr_super; 3669 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3670 ext4_group_t group = elr->lr_next_group; 3671 unsigned int prefetch_ios = 0; 3672 int ret = 0; 3673 int nr = EXT4_SB(sb)->s_mb_prefetch; 3674 u64 start_time; 3675 3676 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3677 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios); 3678 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr); 3679 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr); 3680 if (group >= elr->lr_next_group) { 3681 ret = 1; 3682 if (elr->lr_first_not_zeroed != ngroups && 3683 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3684 elr->lr_next_group = elr->lr_first_not_zeroed; 3685 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3686 ret = 0; 3687 } 3688 } 3689 return ret; 3690 } 3691 3692 for (; group < ngroups; group++) { 3693 gdp = ext4_get_group_desc(sb, group, NULL); 3694 if (!gdp) { 3695 ret = 1; 3696 break; 3697 } 3698 3699 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3700 break; 3701 } 3702 3703 if (group >= ngroups) 3704 ret = 1; 3705 3706 if (!ret) { 3707 start_time = ktime_get_real_ns(); 3708 ret = ext4_init_inode_table(sb, group, 3709 elr->lr_timeout ? 0 : 1); 3710 trace_ext4_lazy_itable_init(sb, group); 3711 if (elr->lr_timeout == 0) { 3712 elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) * 3713 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3714 } 3715 elr->lr_next_sched = jiffies + elr->lr_timeout; 3716 elr->lr_next_group = group + 1; 3717 } 3718 return ret; 3719 } 3720 3721 /* 3722 * Remove lr_request from the list_request and free the 3723 * request structure. Should be called with li_list_mtx held 3724 */ 3725 static void ext4_remove_li_request(struct ext4_li_request *elr) 3726 { 3727 if (!elr) 3728 return; 3729 3730 list_del(&elr->lr_request); 3731 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3732 kfree(elr); 3733 } 3734 3735 static void ext4_unregister_li_request(struct super_block *sb) 3736 { 3737 mutex_lock(&ext4_li_mtx); 3738 if (!ext4_li_info) { 3739 mutex_unlock(&ext4_li_mtx); 3740 return; 3741 } 3742 3743 mutex_lock(&ext4_li_info->li_list_mtx); 3744 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3745 mutex_unlock(&ext4_li_info->li_list_mtx); 3746 mutex_unlock(&ext4_li_mtx); 3747 } 3748 3749 static struct task_struct *ext4_lazyinit_task; 3750 3751 /* 3752 * This is the function where ext4lazyinit thread lives. It walks 3753 * through the request list searching for next scheduled filesystem. 3754 * When such a fs is found, run the lazy initialization request 3755 * (ext4_rn_li_request) and keep track of the time spend in this 3756 * function. Based on that time we compute next schedule time of 3757 * the request. When walking through the list is complete, compute 3758 * next waking time and put itself into sleep. 3759 */ 3760 static int ext4_lazyinit_thread(void *arg) 3761 { 3762 struct ext4_lazy_init *eli = arg; 3763 struct list_head *pos, *n; 3764 struct ext4_li_request *elr; 3765 unsigned long next_wakeup, cur; 3766 3767 BUG_ON(NULL == eli); 3768 set_freezable(); 3769 3770 cont_thread: 3771 while (true) { 3772 next_wakeup = MAX_JIFFY_OFFSET; 3773 3774 mutex_lock(&eli->li_list_mtx); 3775 if (list_empty(&eli->li_request_list)) { 3776 mutex_unlock(&eli->li_list_mtx); 3777 goto exit_thread; 3778 } 3779 list_for_each_safe(pos, n, &eli->li_request_list) { 3780 int err = 0; 3781 int progress = 0; 3782 elr = list_entry(pos, struct ext4_li_request, 3783 lr_request); 3784 3785 if (time_before(jiffies, elr->lr_next_sched)) { 3786 if (time_before(elr->lr_next_sched, next_wakeup)) 3787 next_wakeup = elr->lr_next_sched; 3788 continue; 3789 } 3790 if (down_read_trylock(&elr->lr_super->s_umount)) { 3791 if (sb_start_write_trylock(elr->lr_super)) { 3792 progress = 1; 3793 /* 3794 * We hold sb->s_umount, sb can not 3795 * be removed from the list, it is 3796 * now safe to drop li_list_mtx 3797 */ 3798 mutex_unlock(&eli->li_list_mtx); 3799 err = ext4_run_li_request(elr); 3800 sb_end_write(elr->lr_super); 3801 mutex_lock(&eli->li_list_mtx); 3802 n = pos->next; 3803 } 3804 up_read((&elr->lr_super->s_umount)); 3805 } 3806 /* error, remove the lazy_init job */ 3807 if (err) { 3808 ext4_remove_li_request(elr); 3809 continue; 3810 } 3811 if (!progress) { 3812 elr->lr_next_sched = jiffies + 3813 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3814 } 3815 if (time_before(elr->lr_next_sched, next_wakeup)) 3816 next_wakeup = elr->lr_next_sched; 3817 } 3818 mutex_unlock(&eli->li_list_mtx); 3819 3820 try_to_freeze(); 3821 3822 cur = jiffies; 3823 if ((time_after_eq(cur, next_wakeup)) || 3824 (MAX_JIFFY_OFFSET == next_wakeup)) { 3825 cond_resched(); 3826 continue; 3827 } 3828 3829 schedule_timeout_interruptible(next_wakeup - cur); 3830 3831 if (kthread_should_stop()) { 3832 ext4_clear_request_list(); 3833 goto exit_thread; 3834 } 3835 } 3836 3837 exit_thread: 3838 /* 3839 * It looks like the request list is empty, but we need 3840 * to check it under the li_list_mtx lock, to prevent any 3841 * additions into it, and of course we should lock ext4_li_mtx 3842 * to atomically free the list and ext4_li_info, because at 3843 * this point another ext4 filesystem could be registering 3844 * new one. 3845 */ 3846 mutex_lock(&ext4_li_mtx); 3847 mutex_lock(&eli->li_list_mtx); 3848 if (!list_empty(&eli->li_request_list)) { 3849 mutex_unlock(&eli->li_list_mtx); 3850 mutex_unlock(&ext4_li_mtx); 3851 goto cont_thread; 3852 } 3853 mutex_unlock(&eli->li_list_mtx); 3854 kfree(ext4_li_info); 3855 ext4_li_info = NULL; 3856 mutex_unlock(&ext4_li_mtx); 3857 3858 return 0; 3859 } 3860 3861 static void ext4_clear_request_list(void) 3862 { 3863 struct list_head *pos, *n; 3864 struct ext4_li_request *elr; 3865 3866 mutex_lock(&ext4_li_info->li_list_mtx); 3867 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3868 elr = list_entry(pos, struct ext4_li_request, 3869 lr_request); 3870 ext4_remove_li_request(elr); 3871 } 3872 mutex_unlock(&ext4_li_info->li_list_mtx); 3873 } 3874 3875 static int ext4_run_lazyinit_thread(void) 3876 { 3877 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3878 ext4_li_info, "ext4lazyinit"); 3879 if (IS_ERR(ext4_lazyinit_task)) { 3880 int err = PTR_ERR(ext4_lazyinit_task); 3881 ext4_clear_request_list(); 3882 kfree(ext4_li_info); 3883 ext4_li_info = NULL; 3884 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3885 "initialization thread\n", 3886 err); 3887 return err; 3888 } 3889 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3890 return 0; 3891 } 3892 3893 /* 3894 * Check whether it make sense to run itable init. thread or not. 3895 * If there is at least one uninitialized inode table, return 3896 * corresponding group number, else the loop goes through all 3897 * groups and return total number of groups. 3898 */ 3899 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3900 { 3901 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3902 struct ext4_group_desc *gdp = NULL; 3903 3904 if (!ext4_has_group_desc_csum(sb)) 3905 return ngroups; 3906 3907 for (group = 0; group < ngroups; group++) { 3908 gdp = ext4_get_group_desc(sb, group, NULL); 3909 if (!gdp) 3910 continue; 3911 3912 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3913 break; 3914 } 3915 3916 return group; 3917 } 3918 3919 static int ext4_li_info_new(void) 3920 { 3921 struct ext4_lazy_init *eli = NULL; 3922 3923 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3924 if (!eli) 3925 return -ENOMEM; 3926 3927 INIT_LIST_HEAD(&eli->li_request_list); 3928 mutex_init(&eli->li_list_mtx); 3929 3930 eli->li_state |= EXT4_LAZYINIT_QUIT; 3931 3932 ext4_li_info = eli; 3933 3934 return 0; 3935 } 3936 3937 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3938 ext4_group_t start) 3939 { 3940 struct ext4_li_request *elr; 3941 3942 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3943 if (!elr) 3944 return NULL; 3945 3946 elr->lr_super = sb; 3947 elr->lr_first_not_zeroed = start; 3948 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3949 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3950 elr->lr_next_group = start; 3951 } else { 3952 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3953 } 3954 3955 /* 3956 * Randomize first schedule time of the request to 3957 * spread the inode table initialization requests 3958 * better. 3959 */ 3960 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3961 return elr; 3962 } 3963 3964 int ext4_register_li_request(struct super_block *sb, 3965 ext4_group_t first_not_zeroed) 3966 { 3967 struct ext4_sb_info *sbi = EXT4_SB(sb); 3968 struct ext4_li_request *elr = NULL; 3969 ext4_group_t ngroups = sbi->s_groups_count; 3970 int ret = 0; 3971 3972 mutex_lock(&ext4_li_mtx); 3973 if (sbi->s_li_request != NULL) { 3974 /* 3975 * Reset timeout so it can be computed again, because 3976 * s_li_wait_mult might have changed. 3977 */ 3978 sbi->s_li_request->lr_timeout = 0; 3979 goto out; 3980 } 3981 3982 if (sb_rdonly(sb) || 3983 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 3984 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 3985 goto out; 3986 3987 elr = ext4_li_request_new(sb, first_not_zeroed); 3988 if (!elr) { 3989 ret = -ENOMEM; 3990 goto out; 3991 } 3992 3993 if (NULL == ext4_li_info) { 3994 ret = ext4_li_info_new(); 3995 if (ret) 3996 goto out; 3997 } 3998 3999 mutex_lock(&ext4_li_info->li_list_mtx); 4000 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 4001 mutex_unlock(&ext4_li_info->li_list_mtx); 4002 4003 sbi->s_li_request = elr; 4004 /* 4005 * set elr to NULL here since it has been inserted to 4006 * the request_list and the removal and free of it is 4007 * handled by ext4_clear_request_list from now on. 4008 */ 4009 elr = NULL; 4010 4011 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 4012 ret = ext4_run_lazyinit_thread(); 4013 if (ret) 4014 goto out; 4015 } 4016 out: 4017 mutex_unlock(&ext4_li_mtx); 4018 if (ret) 4019 kfree(elr); 4020 return ret; 4021 } 4022 4023 /* 4024 * We do not need to lock anything since this is called on 4025 * module unload. 4026 */ 4027 static void ext4_destroy_lazyinit_thread(void) 4028 { 4029 /* 4030 * If thread exited earlier 4031 * there's nothing to be done. 4032 */ 4033 if (!ext4_li_info || !ext4_lazyinit_task) 4034 return; 4035 4036 kthread_stop(ext4_lazyinit_task); 4037 } 4038 4039 static int set_journal_csum_feature_set(struct super_block *sb) 4040 { 4041 int ret = 1; 4042 int compat, incompat; 4043 struct ext4_sb_info *sbi = EXT4_SB(sb); 4044 4045 if (ext4_has_metadata_csum(sb)) { 4046 /* journal checksum v3 */ 4047 compat = 0; 4048 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4049 } else { 4050 /* journal checksum v1 */ 4051 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4052 incompat = 0; 4053 } 4054 4055 jbd2_journal_clear_features(sbi->s_journal, 4056 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4057 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4058 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4059 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4060 ret = jbd2_journal_set_features(sbi->s_journal, 4061 compat, 0, 4062 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4063 incompat); 4064 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4065 ret = jbd2_journal_set_features(sbi->s_journal, 4066 compat, 0, 4067 incompat); 4068 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4069 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4070 } else { 4071 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4072 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4073 } 4074 4075 return ret; 4076 } 4077 4078 /* 4079 * Note: calculating the overhead so we can be compatible with 4080 * historical BSD practice is quite difficult in the face of 4081 * clusters/bigalloc. This is because multiple metadata blocks from 4082 * different block group can end up in the same allocation cluster. 4083 * Calculating the exact overhead in the face of clustered allocation 4084 * requires either O(all block bitmaps) in memory or O(number of block 4085 * groups**2) in time. We will still calculate the superblock for 4086 * older file systems --- and if we come across with a bigalloc file 4087 * system with zero in s_overhead_clusters the estimate will be close to 4088 * correct especially for very large cluster sizes --- but for newer 4089 * file systems, it's better to calculate this figure once at mkfs 4090 * time, and store it in the superblock. If the superblock value is 4091 * present (even for non-bigalloc file systems), we will use it. 4092 */ 4093 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4094 char *buf) 4095 { 4096 struct ext4_sb_info *sbi = EXT4_SB(sb); 4097 struct ext4_group_desc *gdp; 4098 ext4_fsblk_t first_block, last_block, b; 4099 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4100 int s, j, count = 0; 4101 int has_super = ext4_bg_has_super(sb, grp); 4102 4103 if (!ext4_has_feature_bigalloc(sb)) 4104 return (has_super + ext4_bg_num_gdb(sb, grp) + 4105 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4106 sbi->s_itb_per_group + 2); 4107 4108 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4109 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4110 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4111 for (i = 0; i < ngroups; i++) { 4112 gdp = ext4_get_group_desc(sb, i, NULL); 4113 b = ext4_block_bitmap(sb, gdp); 4114 if (b >= first_block && b <= last_block) { 4115 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4116 count++; 4117 } 4118 b = ext4_inode_bitmap(sb, gdp); 4119 if (b >= first_block && b <= last_block) { 4120 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4121 count++; 4122 } 4123 b = ext4_inode_table(sb, gdp); 4124 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4125 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4126 int c = EXT4_B2C(sbi, b - first_block); 4127 ext4_set_bit(c, buf); 4128 count++; 4129 } 4130 if (i != grp) 4131 continue; 4132 s = 0; 4133 if (ext4_bg_has_super(sb, grp)) { 4134 ext4_set_bit(s++, buf); 4135 count++; 4136 } 4137 j = ext4_bg_num_gdb(sb, grp); 4138 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4139 ext4_error(sb, "Invalid number of block group " 4140 "descriptor blocks: %d", j); 4141 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4142 } 4143 count += j; 4144 for (; j > 0; j--) 4145 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4146 } 4147 if (!count) 4148 return 0; 4149 return EXT4_CLUSTERS_PER_GROUP(sb) - 4150 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4151 } 4152 4153 /* 4154 * Compute the overhead and stash it in sbi->s_overhead 4155 */ 4156 int ext4_calculate_overhead(struct super_block *sb) 4157 { 4158 struct ext4_sb_info *sbi = EXT4_SB(sb); 4159 struct ext4_super_block *es = sbi->s_es; 4160 struct inode *j_inode; 4161 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4162 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4163 ext4_fsblk_t overhead = 0; 4164 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4165 4166 if (!buf) 4167 return -ENOMEM; 4168 4169 /* 4170 * Compute the overhead (FS structures). This is constant 4171 * for a given filesystem unless the number of block groups 4172 * changes so we cache the previous value until it does. 4173 */ 4174 4175 /* 4176 * All of the blocks before first_data_block are overhead 4177 */ 4178 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4179 4180 /* 4181 * Add the overhead found in each block group 4182 */ 4183 for (i = 0; i < ngroups; i++) { 4184 int blks; 4185 4186 blks = count_overhead(sb, i, buf); 4187 overhead += blks; 4188 if (blks) 4189 memset(buf, 0, PAGE_SIZE); 4190 cond_resched(); 4191 } 4192 4193 /* 4194 * Add the internal journal blocks whether the journal has been 4195 * loaded or not 4196 */ 4197 if (sbi->s_journal && !sbi->s_journal_bdev_file) 4198 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4199 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4200 /* j_inum for internal journal is non-zero */ 4201 j_inode = ext4_get_journal_inode(sb, j_inum); 4202 if (!IS_ERR(j_inode)) { 4203 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4204 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4205 iput(j_inode); 4206 } else { 4207 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4208 } 4209 } 4210 sbi->s_overhead = overhead; 4211 smp_wmb(); 4212 free_page((unsigned long) buf); 4213 return 0; 4214 } 4215 4216 static void ext4_set_resv_clusters(struct super_block *sb) 4217 { 4218 ext4_fsblk_t resv_clusters; 4219 struct ext4_sb_info *sbi = EXT4_SB(sb); 4220 4221 /* 4222 * There's no need to reserve anything when we aren't using extents. 4223 * The space estimates are exact, there are no unwritten extents, 4224 * hole punching doesn't need new metadata... This is needed especially 4225 * to keep ext2/3 backward compatibility. 4226 */ 4227 if (!ext4_has_feature_extents(sb)) 4228 return; 4229 /* 4230 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4231 * This should cover the situations where we can not afford to run 4232 * out of space like for example punch hole, or converting 4233 * unwritten extents in delalloc path. In most cases such 4234 * allocation would require 1, or 2 blocks, higher numbers are 4235 * very rare. 4236 */ 4237 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4238 sbi->s_cluster_bits); 4239 4240 do_div(resv_clusters, 50); 4241 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4242 4243 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4244 } 4245 4246 static const char *ext4_quota_mode(struct super_block *sb) 4247 { 4248 #ifdef CONFIG_QUOTA 4249 if (!ext4_quota_capable(sb)) 4250 return "none"; 4251 4252 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4253 return "journalled"; 4254 else 4255 return "writeback"; 4256 #else 4257 return "disabled"; 4258 #endif 4259 } 4260 4261 static void ext4_setup_csum_trigger(struct super_block *sb, 4262 enum ext4_journal_trigger_type type, 4263 void (*trigger)( 4264 struct jbd2_buffer_trigger_type *type, 4265 struct buffer_head *bh, 4266 void *mapped_data, 4267 size_t size)) 4268 { 4269 struct ext4_sb_info *sbi = EXT4_SB(sb); 4270 4271 sbi->s_journal_triggers[type].sb = sb; 4272 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4273 } 4274 4275 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4276 { 4277 if (!sbi) 4278 return; 4279 4280 kfree(sbi->s_blockgroup_lock); 4281 fs_put_dax(sbi->s_daxdev, NULL); 4282 kfree(sbi); 4283 } 4284 4285 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4286 { 4287 struct ext4_sb_info *sbi; 4288 4289 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4290 if (!sbi) 4291 return NULL; 4292 4293 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4294 NULL, NULL); 4295 4296 sbi->s_blockgroup_lock = 4297 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4298 4299 if (!sbi->s_blockgroup_lock) 4300 goto err_out; 4301 4302 sb->s_fs_info = sbi; 4303 sbi->s_sb = sb; 4304 return sbi; 4305 err_out: 4306 fs_put_dax(sbi->s_daxdev, NULL); 4307 kfree(sbi); 4308 return NULL; 4309 } 4310 4311 static void ext4_set_def_opts(struct super_block *sb, 4312 struct ext4_super_block *es) 4313 { 4314 unsigned long def_mount_opts; 4315 4316 /* Set defaults before we parse the mount options */ 4317 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4318 set_opt(sb, INIT_INODE_TABLE); 4319 if (def_mount_opts & EXT4_DEFM_DEBUG) 4320 set_opt(sb, DEBUG); 4321 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4322 set_opt(sb, GRPID); 4323 if (def_mount_opts & EXT4_DEFM_UID16) 4324 set_opt(sb, NO_UID32); 4325 /* xattr user namespace & acls are now defaulted on */ 4326 set_opt(sb, XATTR_USER); 4327 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4328 set_opt(sb, POSIX_ACL); 4329 #endif 4330 if (ext4_has_feature_fast_commit(sb)) 4331 set_opt2(sb, JOURNAL_FAST_COMMIT); 4332 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4333 if (ext4_has_metadata_csum(sb)) 4334 set_opt(sb, JOURNAL_CHECKSUM); 4335 4336 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4337 set_opt(sb, JOURNAL_DATA); 4338 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4339 set_opt(sb, ORDERED_DATA); 4340 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4341 set_opt(sb, WRITEBACK_DATA); 4342 4343 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4344 set_opt(sb, ERRORS_PANIC); 4345 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4346 set_opt(sb, ERRORS_CONT); 4347 else 4348 set_opt(sb, ERRORS_RO); 4349 /* block_validity enabled by default; disable with noblock_validity */ 4350 set_opt(sb, BLOCK_VALIDITY); 4351 if (def_mount_opts & EXT4_DEFM_DISCARD) 4352 set_opt(sb, DISCARD); 4353 4354 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4355 set_opt(sb, BARRIER); 4356 4357 /* 4358 * enable delayed allocation by default 4359 * Use -o nodelalloc to turn it off 4360 */ 4361 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4362 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4363 set_opt(sb, DELALLOC); 4364 4365 if (sb->s_blocksize <= PAGE_SIZE) 4366 set_opt(sb, DIOREAD_NOLOCK); 4367 } 4368 4369 static int ext4_handle_clustersize(struct super_block *sb) 4370 { 4371 struct ext4_sb_info *sbi = EXT4_SB(sb); 4372 struct ext4_super_block *es = sbi->s_es; 4373 int clustersize; 4374 4375 /* Handle clustersize */ 4376 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4377 if (ext4_has_feature_bigalloc(sb)) { 4378 if (clustersize < sb->s_blocksize) { 4379 ext4_msg(sb, KERN_ERR, 4380 "cluster size (%d) smaller than " 4381 "block size (%lu)", clustersize, sb->s_blocksize); 4382 return -EINVAL; 4383 } 4384 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4385 le32_to_cpu(es->s_log_block_size); 4386 } else { 4387 if (clustersize != sb->s_blocksize) { 4388 ext4_msg(sb, KERN_ERR, 4389 "fragment/cluster size (%d) != " 4390 "block size (%lu)", clustersize, sb->s_blocksize); 4391 return -EINVAL; 4392 } 4393 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4394 ext4_msg(sb, KERN_ERR, 4395 "#blocks per group too big: %lu", 4396 sbi->s_blocks_per_group); 4397 return -EINVAL; 4398 } 4399 sbi->s_cluster_bits = 0; 4400 } 4401 sbi->s_clusters_per_group = le32_to_cpu(es->s_clusters_per_group); 4402 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4403 ext4_msg(sb, KERN_ERR, "#clusters per group too big: %lu", 4404 sbi->s_clusters_per_group); 4405 return -EINVAL; 4406 } 4407 if (sbi->s_blocks_per_group != 4408 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4409 ext4_msg(sb, KERN_ERR, 4410 "blocks per group (%lu) and clusters per group (%lu) inconsistent", 4411 sbi->s_blocks_per_group, sbi->s_clusters_per_group); 4412 return -EINVAL; 4413 } 4414 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4415 4416 /* Do we have standard group size of clustersize * 8 blocks ? */ 4417 if (sbi->s_blocks_per_group == clustersize << 3) 4418 set_opt2(sb, STD_GROUP_SIZE); 4419 4420 return 0; 4421 } 4422 4423 static void ext4_fast_commit_init(struct super_block *sb) 4424 { 4425 struct ext4_sb_info *sbi = EXT4_SB(sb); 4426 4427 /* Initialize fast commit stuff */ 4428 atomic_set(&sbi->s_fc_subtid, 0); 4429 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4430 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4431 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4432 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4433 sbi->s_fc_bytes = 0; 4434 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4435 sbi->s_fc_ineligible_tid = 0; 4436 spin_lock_init(&sbi->s_fc_lock); 4437 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4438 sbi->s_fc_replay_state.fc_regions = NULL; 4439 sbi->s_fc_replay_state.fc_regions_size = 0; 4440 sbi->s_fc_replay_state.fc_regions_used = 0; 4441 sbi->s_fc_replay_state.fc_regions_valid = 0; 4442 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4443 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4444 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4445 } 4446 4447 static int ext4_inode_info_init(struct super_block *sb, 4448 struct ext4_super_block *es) 4449 { 4450 struct ext4_sb_info *sbi = EXT4_SB(sb); 4451 4452 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4453 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4454 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4455 } else { 4456 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4457 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4458 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4459 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4460 sbi->s_first_ino); 4461 return -EINVAL; 4462 } 4463 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4464 (!is_power_of_2(sbi->s_inode_size)) || 4465 (sbi->s_inode_size > sb->s_blocksize)) { 4466 ext4_msg(sb, KERN_ERR, 4467 "unsupported inode size: %d", 4468 sbi->s_inode_size); 4469 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4470 return -EINVAL; 4471 } 4472 /* 4473 * i_atime_extra is the last extra field available for 4474 * [acm]times in struct ext4_inode. Checking for that 4475 * field should suffice to ensure we have extra space 4476 * for all three. 4477 */ 4478 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4479 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4480 sb->s_time_gran = 1; 4481 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4482 } else { 4483 sb->s_time_gran = NSEC_PER_SEC; 4484 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4485 } 4486 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4487 } 4488 4489 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4490 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4491 EXT4_GOOD_OLD_INODE_SIZE; 4492 if (ext4_has_feature_extra_isize(sb)) { 4493 unsigned v, max = (sbi->s_inode_size - 4494 EXT4_GOOD_OLD_INODE_SIZE); 4495 4496 v = le16_to_cpu(es->s_want_extra_isize); 4497 if (v > max) { 4498 ext4_msg(sb, KERN_ERR, 4499 "bad s_want_extra_isize: %d", v); 4500 return -EINVAL; 4501 } 4502 if (sbi->s_want_extra_isize < v) 4503 sbi->s_want_extra_isize = v; 4504 4505 v = le16_to_cpu(es->s_min_extra_isize); 4506 if (v > max) { 4507 ext4_msg(sb, KERN_ERR, 4508 "bad s_min_extra_isize: %d", v); 4509 return -EINVAL; 4510 } 4511 if (sbi->s_want_extra_isize < v) 4512 sbi->s_want_extra_isize = v; 4513 } 4514 } 4515 4516 return 0; 4517 } 4518 4519 #if IS_ENABLED(CONFIG_UNICODE) 4520 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4521 { 4522 const struct ext4_sb_encodings *encoding_info; 4523 struct unicode_map *encoding; 4524 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4525 4526 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4527 return 0; 4528 4529 encoding_info = ext4_sb_read_encoding(es); 4530 if (!encoding_info) { 4531 ext4_msg(sb, KERN_ERR, 4532 "Encoding requested by superblock is unknown"); 4533 return -EINVAL; 4534 } 4535 4536 encoding = utf8_load(encoding_info->version); 4537 if (IS_ERR(encoding)) { 4538 ext4_msg(sb, KERN_ERR, 4539 "can't mount with superblock charset: %s-%u.%u.%u " 4540 "not supported by the kernel. flags: 0x%x.", 4541 encoding_info->name, 4542 unicode_major(encoding_info->version), 4543 unicode_minor(encoding_info->version), 4544 unicode_rev(encoding_info->version), 4545 encoding_flags); 4546 return -EINVAL; 4547 } 4548 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4549 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4550 unicode_major(encoding_info->version), 4551 unicode_minor(encoding_info->version), 4552 unicode_rev(encoding_info->version), 4553 encoding_flags); 4554 4555 sb->s_encoding = encoding; 4556 sb->s_encoding_flags = encoding_flags; 4557 4558 return 0; 4559 } 4560 #else 4561 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4562 { 4563 return 0; 4564 } 4565 #endif 4566 4567 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4568 { 4569 struct ext4_sb_info *sbi = EXT4_SB(sb); 4570 4571 /* Warn if metadata_csum and gdt_csum are both set. */ 4572 if (ext4_has_feature_metadata_csum(sb) && 4573 ext4_has_feature_gdt_csum(sb)) 4574 ext4_warning(sb, "metadata_csum and uninit_bg are " 4575 "redundant flags; please run fsck."); 4576 4577 /* Check for a known checksum algorithm */ 4578 if (!ext4_verify_csum_type(sb, es)) { 4579 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4580 "unknown checksum algorithm."); 4581 return -EINVAL; 4582 } 4583 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4584 ext4_orphan_file_block_trigger); 4585 4586 /* Load the checksum driver */ 4587 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4588 if (IS_ERR(sbi->s_chksum_driver)) { 4589 int ret = PTR_ERR(sbi->s_chksum_driver); 4590 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4591 sbi->s_chksum_driver = NULL; 4592 return ret; 4593 } 4594 4595 /* Check superblock checksum */ 4596 if (!ext4_superblock_csum_verify(sb, es)) { 4597 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4598 "invalid superblock checksum. Run e2fsck?"); 4599 return -EFSBADCRC; 4600 } 4601 4602 /* Precompute checksum seed for all metadata */ 4603 if (ext4_has_feature_csum_seed(sb)) 4604 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4605 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4606 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4607 sizeof(es->s_uuid)); 4608 return 0; 4609 } 4610 4611 static int ext4_check_feature_compatibility(struct super_block *sb, 4612 struct ext4_super_block *es, 4613 int silent) 4614 { 4615 struct ext4_sb_info *sbi = EXT4_SB(sb); 4616 4617 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4618 (ext4_has_compat_features(sb) || 4619 ext4_has_ro_compat_features(sb) || 4620 ext4_has_incompat_features(sb))) 4621 ext4_msg(sb, KERN_WARNING, 4622 "feature flags set on rev 0 fs, " 4623 "running e2fsck is recommended"); 4624 4625 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4626 set_opt2(sb, HURD_COMPAT); 4627 if (ext4_has_feature_64bit(sb)) { 4628 ext4_msg(sb, KERN_ERR, 4629 "The Hurd can't support 64-bit file systems"); 4630 return -EINVAL; 4631 } 4632 4633 /* 4634 * ea_inode feature uses l_i_version field which is not 4635 * available in HURD_COMPAT mode. 4636 */ 4637 if (ext4_has_feature_ea_inode(sb)) { 4638 ext4_msg(sb, KERN_ERR, 4639 "ea_inode feature is not supported for Hurd"); 4640 return -EINVAL; 4641 } 4642 } 4643 4644 if (IS_EXT2_SB(sb)) { 4645 if (ext2_feature_set_ok(sb)) 4646 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4647 "using the ext4 subsystem"); 4648 else { 4649 /* 4650 * If we're probing be silent, if this looks like 4651 * it's actually an ext[34] filesystem. 4652 */ 4653 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4654 return -EINVAL; 4655 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4656 "to feature incompatibilities"); 4657 return -EINVAL; 4658 } 4659 } 4660 4661 if (IS_EXT3_SB(sb)) { 4662 if (ext3_feature_set_ok(sb)) 4663 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4664 "using the ext4 subsystem"); 4665 else { 4666 /* 4667 * If we're probing be silent, if this looks like 4668 * it's actually an ext4 filesystem. 4669 */ 4670 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4671 return -EINVAL; 4672 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4673 "to feature incompatibilities"); 4674 return -EINVAL; 4675 } 4676 } 4677 4678 /* 4679 * Check feature flags regardless of the revision level, since we 4680 * previously didn't change the revision level when setting the flags, 4681 * so there is a chance incompat flags are set on a rev 0 filesystem. 4682 */ 4683 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4684 return -EINVAL; 4685 4686 if (sbi->s_daxdev) { 4687 if (sb->s_blocksize == PAGE_SIZE) 4688 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4689 else 4690 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 4691 } 4692 4693 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4694 if (ext4_has_feature_inline_data(sb)) { 4695 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4696 " that may contain inline data"); 4697 return -EINVAL; 4698 } 4699 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4700 ext4_msg(sb, KERN_ERR, 4701 "DAX unsupported by block device."); 4702 return -EINVAL; 4703 } 4704 } 4705 4706 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4707 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4708 es->s_encryption_level); 4709 return -EINVAL; 4710 } 4711 4712 return 0; 4713 } 4714 4715 static int ext4_check_geometry(struct super_block *sb, 4716 struct ext4_super_block *es) 4717 { 4718 struct ext4_sb_info *sbi = EXT4_SB(sb); 4719 __u64 blocks_count; 4720 int err; 4721 4722 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 4723 ext4_msg(sb, KERN_ERR, 4724 "Number of reserved GDT blocks insanely large: %d", 4725 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4726 return -EINVAL; 4727 } 4728 /* 4729 * Test whether we have more sectors than will fit in sector_t, 4730 * and whether the max offset is addressable by the page cache. 4731 */ 4732 err = generic_check_addressable(sb->s_blocksize_bits, 4733 ext4_blocks_count(es)); 4734 if (err) { 4735 ext4_msg(sb, KERN_ERR, "filesystem" 4736 " too large to mount safely on this system"); 4737 return err; 4738 } 4739 4740 /* check blocks count against device size */ 4741 blocks_count = sb_bdev_nr_blocks(sb); 4742 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4743 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4744 "exceeds size of device (%llu blocks)", 4745 ext4_blocks_count(es), blocks_count); 4746 return -EINVAL; 4747 } 4748 4749 /* 4750 * It makes no sense for the first data block to be beyond the end 4751 * of the filesystem. 4752 */ 4753 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4754 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4755 "block %u is beyond end of filesystem (%llu)", 4756 le32_to_cpu(es->s_first_data_block), 4757 ext4_blocks_count(es)); 4758 return -EINVAL; 4759 } 4760 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4761 (sbi->s_cluster_ratio == 1)) { 4762 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4763 "block is 0 with a 1k block and cluster size"); 4764 return -EINVAL; 4765 } 4766 4767 blocks_count = (ext4_blocks_count(es) - 4768 le32_to_cpu(es->s_first_data_block) + 4769 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4770 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4771 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4772 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4773 "(block count %llu, first data block %u, " 4774 "blocks per group %lu)", blocks_count, 4775 ext4_blocks_count(es), 4776 le32_to_cpu(es->s_first_data_block), 4777 EXT4_BLOCKS_PER_GROUP(sb)); 4778 return -EINVAL; 4779 } 4780 sbi->s_groups_count = blocks_count; 4781 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4782 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4783 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4784 le32_to_cpu(es->s_inodes_count)) { 4785 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4786 le32_to_cpu(es->s_inodes_count), 4787 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4788 return -EINVAL; 4789 } 4790 4791 return 0; 4792 } 4793 4794 static int ext4_group_desc_init(struct super_block *sb, 4795 struct ext4_super_block *es, 4796 ext4_fsblk_t logical_sb_block, 4797 ext4_group_t *first_not_zeroed) 4798 { 4799 struct ext4_sb_info *sbi = EXT4_SB(sb); 4800 unsigned int db_count; 4801 ext4_fsblk_t block; 4802 int i; 4803 4804 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4805 EXT4_DESC_PER_BLOCK(sb); 4806 if (ext4_has_feature_meta_bg(sb)) { 4807 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4808 ext4_msg(sb, KERN_WARNING, 4809 "first meta block group too large: %u " 4810 "(group descriptor block count %u)", 4811 le32_to_cpu(es->s_first_meta_bg), db_count); 4812 return -EINVAL; 4813 } 4814 } 4815 rcu_assign_pointer(sbi->s_group_desc, 4816 kvmalloc_array(db_count, 4817 sizeof(struct buffer_head *), 4818 GFP_KERNEL)); 4819 if (sbi->s_group_desc == NULL) { 4820 ext4_msg(sb, KERN_ERR, "not enough memory"); 4821 return -ENOMEM; 4822 } 4823 4824 bgl_lock_init(sbi->s_blockgroup_lock); 4825 4826 /* Pre-read the descriptors into the buffer cache */ 4827 for (i = 0; i < db_count; i++) { 4828 block = descriptor_loc(sb, logical_sb_block, i); 4829 ext4_sb_breadahead_unmovable(sb, block); 4830 } 4831 4832 for (i = 0; i < db_count; i++) { 4833 struct buffer_head *bh; 4834 4835 block = descriptor_loc(sb, logical_sb_block, i); 4836 bh = ext4_sb_bread_unmovable(sb, block); 4837 if (IS_ERR(bh)) { 4838 ext4_msg(sb, KERN_ERR, 4839 "can't read group descriptor %d", i); 4840 sbi->s_gdb_count = i; 4841 return PTR_ERR(bh); 4842 } 4843 rcu_read_lock(); 4844 rcu_dereference(sbi->s_group_desc)[i] = bh; 4845 rcu_read_unlock(); 4846 } 4847 sbi->s_gdb_count = db_count; 4848 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4849 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4850 return -EFSCORRUPTED; 4851 } 4852 4853 return 0; 4854 } 4855 4856 static int ext4_load_and_init_journal(struct super_block *sb, 4857 struct ext4_super_block *es, 4858 struct ext4_fs_context *ctx) 4859 { 4860 struct ext4_sb_info *sbi = EXT4_SB(sb); 4861 int err; 4862 4863 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4864 if (err) 4865 return err; 4866 4867 if (ext4_has_feature_64bit(sb) && 4868 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4869 JBD2_FEATURE_INCOMPAT_64BIT)) { 4870 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4871 goto out; 4872 } 4873 4874 if (!set_journal_csum_feature_set(sb)) { 4875 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4876 "feature set"); 4877 goto out; 4878 } 4879 4880 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4881 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4882 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4883 ext4_msg(sb, KERN_ERR, 4884 "Failed to set fast commit journal feature"); 4885 goto out; 4886 } 4887 4888 /* We have now updated the journal if required, so we can 4889 * validate the data journaling mode. */ 4890 switch (test_opt(sb, DATA_FLAGS)) { 4891 case 0: 4892 /* No mode set, assume a default based on the journal 4893 * capabilities: ORDERED_DATA if the journal can 4894 * cope, else JOURNAL_DATA 4895 */ 4896 if (jbd2_journal_check_available_features 4897 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4898 set_opt(sb, ORDERED_DATA); 4899 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4900 } else { 4901 set_opt(sb, JOURNAL_DATA); 4902 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4903 } 4904 break; 4905 4906 case EXT4_MOUNT_ORDERED_DATA: 4907 case EXT4_MOUNT_WRITEBACK_DATA: 4908 if (!jbd2_journal_check_available_features 4909 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4910 ext4_msg(sb, KERN_ERR, "Journal does not support " 4911 "requested data journaling mode"); 4912 goto out; 4913 } 4914 break; 4915 default: 4916 break; 4917 } 4918 4919 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4920 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4921 ext4_msg(sb, KERN_ERR, "can't mount with " 4922 "journal_async_commit in data=ordered mode"); 4923 goto out; 4924 } 4925 4926 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4927 4928 sbi->s_journal->j_submit_inode_data_buffers = 4929 ext4_journal_submit_inode_data_buffers; 4930 sbi->s_journal->j_finish_inode_data_buffers = 4931 ext4_journal_finish_inode_data_buffers; 4932 4933 return 0; 4934 4935 out: 4936 /* flush s_sb_upd_work before destroying the journal. */ 4937 flush_work(&sbi->s_sb_upd_work); 4938 jbd2_journal_destroy(sbi->s_journal); 4939 sbi->s_journal = NULL; 4940 return -EINVAL; 4941 } 4942 4943 static int ext4_check_journal_data_mode(struct super_block *sb) 4944 { 4945 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4946 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4947 "data=journal disables delayed allocation, " 4948 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 4949 /* can't mount with both data=journal and dioread_nolock. */ 4950 clear_opt(sb, DIOREAD_NOLOCK); 4951 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4952 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4953 ext4_msg(sb, KERN_ERR, "can't mount with " 4954 "both data=journal and delalloc"); 4955 return -EINVAL; 4956 } 4957 if (test_opt(sb, DAX_ALWAYS)) { 4958 ext4_msg(sb, KERN_ERR, "can't mount with " 4959 "both data=journal and dax"); 4960 return -EINVAL; 4961 } 4962 if (ext4_has_feature_encrypt(sb)) { 4963 ext4_msg(sb, KERN_WARNING, 4964 "encrypted files will use data=ordered " 4965 "instead of data journaling mode"); 4966 } 4967 if (test_opt(sb, DELALLOC)) 4968 clear_opt(sb, DELALLOC); 4969 } else { 4970 sb->s_iflags |= SB_I_CGROUPWB; 4971 } 4972 4973 return 0; 4974 } 4975 4976 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 4977 int silent) 4978 { 4979 struct ext4_sb_info *sbi = EXT4_SB(sb); 4980 struct ext4_super_block *es; 4981 ext4_fsblk_t logical_sb_block; 4982 unsigned long offset = 0; 4983 struct buffer_head *bh; 4984 int ret = -EINVAL; 4985 int blocksize; 4986 4987 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4988 if (!blocksize) { 4989 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4990 return -EINVAL; 4991 } 4992 4993 /* 4994 * The ext4 superblock will not be buffer aligned for other than 1kB 4995 * block sizes. We need to calculate the offset from buffer start. 4996 */ 4997 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4998 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 4999 offset = do_div(logical_sb_block, blocksize); 5000 } else { 5001 logical_sb_block = sbi->s_sb_block; 5002 } 5003 5004 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5005 if (IS_ERR(bh)) { 5006 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 5007 return PTR_ERR(bh); 5008 } 5009 /* 5010 * Note: s_es must be initialized as soon as possible because 5011 * some ext4 macro-instructions depend on its value 5012 */ 5013 es = (struct ext4_super_block *) (bh->b_data + offset); 5014 sbi->s_es = es; 5015 sb->s_magic = le16_to_cpu(es->s_magic); 5016 if (sb->s_magic != EXT4_SUPER_MAGIC) { 5017 if (!silent) 5018 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5019 goto out; 5020 } 5021 5022 if (le32_to_cpu(es->s_log_block_size) > 5023 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5024 ext4_msg(sb, KERN_ERR, 5025 "Invalid log block size: %u", 5026 le32_to_cpu(es->s_log_block_size)); 5027 goto out; 5028 } 5029 if (le32_to_cpu(es->s_log_cluster_size) > 5030 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 5031 ext4_msg(sb, KERN_ERR, 5032 "Invalid log cluster size: %u", 5033 le32_to_cpu(es->s_log_cluster_size)); 5034 goto out; 5035 } 5036 5037 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 5038 5039 /* 5040 * If the default block size is not the same as the real block size, 5041 * we need to reload it. 5042 */ 5043 if (sb->s_blocksize == blocksize) { 5044 *lsb = logical_sb_block; 5045 sbi->s_sbh = bh; 5046 return 0; 5047 } 5048 5049 /* 5050 * bh must be released before kill_bdev(), otherwise 5051 * it won't be freed and its page also. kill_bdev() 5052 * is called by sb_set_blocksize(). 5053 */ 5054 brelse(bh); 5055 /* Validate the filesystem blocksize */ 5056 if (!sb_set_blocksize(sb, blocksize)) { 5057 ext4_msg(sb, KERN_ERR, "bad block size %d", 5058 blocksize); 5059 bh = NULL; 5060 goto out; 5061 } 5062 5063 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5064 offset = do_div(logical_sb_block, blocksize); 5065 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5066 if (IS_ERR(bh)) { 5067 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5068 ret = PTR_ERR(bh); 5069 bh = NULL; 5070 goto out; 5071 } 5072 es = (struct ext4_super_block *)(bh->b_data + offset); 5073 sbi->s_es = es; 5074 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5075 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5076 goto out; 5077 } 5078 *lsb = logical_sb_block; 5079 sbi->s_sbh = bh; 5080 return 0; 5081 out: 5082 brelse(bh); 5083 return ret; 5084 } 5085 5086 static void ext4_hash_info_init(struct super_block *sb) 5087 { 5088 struct ext4_sb_info *sbi = EXT4_SB(sb); 5089 struct ext4_super_block *es = sbi->s_es; 5090 unsigned int i; 5091 5092 for (i = 0; i < 4; i++) 5093 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5094 5095 sbi->s_def_hash_version = es->s_def_hash_version; 5096 if (ext4_has_feature_dir_index(sb)) { 5097 i = le32_to_cpu(es->s_flags); 5098 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5099 sbi->s_hash_unsigned = 3; 5100 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5101 #ifdef __CHAR_UNSIGNED__ 5102 if (!sb_rdonly(sb)) 5103 es->s_flags |= 5104 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5105 sbi->s_hash_unsigned = 3; 5106 #else 5107 if (!sb_rdonly(sb)) 5108 es->s_flags |= 5109 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5110 #endif 5111 } 5112 } 5113 } 5114 5115 static int ext4_block_group_meta_init(struct super_block *sb, int silent) 5116 { 5117 struct ext4_sb_info *sbi = EXT4_SB(sb); 5118 struct ext4_super_block *es = sbi->s_es; 5119 int has_huge_files; 5120 5121 has_huge_files = ext4_has_feature_huge_file(sb); 5122 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5123 has_huge_files); 5124 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5125 5126 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5127 if (ext4_has_feature_64bit(sb)) { 5128 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5129 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5130 !is_power_of_2(sbi->s_desc_size)) { 5131 ext4_msg(sb, KERN_ERR, 5132 "unsupported descriptor size %lu", 5133 sbi->s_desc_size); 5134 return -EINVAL; 5135 } 5136 } else 5137 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5138 5139 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5140 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5141 5142 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5143 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5144 if (!silent) 5145 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5146 return -EINVAL; 5147 } 5148 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5149 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5150 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5151 sbi->s_inodes_per_group); 5152 return -EINVAL; 5153 } 5154 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5155 sbi->s_inodes_per_block; 5156 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5157 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5158 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5159 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5160 5161 return 0; 5162 } 5163 5164 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5165 { 5166 struct ext4_super_block *es = NULL; 5167 struct ext4_sb_info *sbi = EXT4_SB(sb); 5168 ext4_fsblk_t logical_sb_block; 5169 struct inode *root; 5170 int needs_recovery; 5171 int err; 5172 ext4_group_t first_not_zeroed; 5173 struct ext4_fs_context *ctx = fc->fs_private; 5174 int silent = fc->sb_flags & SB_SILENT; 5175 5176 /* Set defaults for the variables that will be set during parsing */ 5177 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5178 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5179 5180 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5181 sbi->s_sectors_written_start = 5182 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5183 5184 err = ext4_load_super(sb, &logical_sb_block, silent); 5185 if (err) 5186 goto out_fail; 5187 5188 es = sbi->s_es; 5189 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5190 5191 err = ext4_init_metadata_csum(sb, es); 5192 if (err) 5193 goto failed_mount; 5194 5195 ext4_set_def_opts(sb, es); 5196 5197 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5198 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5199 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5200 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5201 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5202 5203 /* 5204 * set default s_li_wait_mult for lazyinit, for the case there is 5205 * no mount option specified. 5206 */ 5207 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5208 5209 err = ext4_inode_info_init(sb, es); 5210 if (err) 5211 goto failed_mount; 5212 5213 err = parse_apply_sb_mount_options(sb, ctx); 5214 if (err < 0) 5215 goto failed_mount; 5216 5217 sbi->s_def_mount_opt = sbi->s_mount_opt; 5218 sbi->s_def_mount_opt2 = sbi->s_mount_opt2; 5219 5220 err = ext4_check_opt_consistency(fc, sb); 5221 if (err < 0) 5222 goto failed_mount; 5223 5224 ext4_apply_options(fc, sb); 5225 5226 err = ext4_encoding_init(sb, es); 5227 if (err) 5228 goto failed_mount; 5229 5230 err = ext4_check_journal_data_mode(sb); 5231 if (err) 5232 goto failed_mount; 5233 5234 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5235 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5236 5237 /* i_version is always enabled now */ 5238 sb->s_flags |= SB_I_VERSION; 5239 5240 err = ext4_check_feature_compatibility(sb, es, silent); 5241 if (err) 5242 goto failed_mount; 5243 5244 err = ext4_block_group_meta_init(sb, silent); 5245 if (err) 5246 goto failed_mount; 5247 5248 ext4_hash_info_init(sb); 5249 5250 err = ext4_handle_clustersize(sb); 5251 if (err) 5252 goto failed_mount; 5253 5254 err = ext4_check_geometry(sb, es); 5255 if (err) 5256 goto failed_mount; 5257 5258 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5259 spin_lock_init(&sbi->s_error_lock); 5260 INIT_WORK(&sbi->s_sb_upd_work, update_super_work); 5261 5262 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5263 if (err) 5264 goto failed_mount3; 5265 5266 err = ext4_es_register_shrinker(sbi); 5267 if (err) 5268 goto failed_mount3; 5269 5270 sbi->s_stripe = ext4_get_stripe_size(sbi); 5271 /* 5272 * It's hard to get stripe aligned blocks if stripe is not aligned with 5273 * cluster, just disable stripe and alert user to simpfy code and avoid 5274 * stripe aligned allocation which will rarely successes. 5275 */ 5276 if (sbi->s_stripe > 0 && sbi->s_cluster_ratio > 1 && 5277 sbi->s_stripe % sbi->s_cluster_ratio != 0) { 5278 ext4_msg(sb, KERN_WARNING, 5279 "stripe (%lu) is not aligned with cluster size (%u), " 5280 "stripe is disabled", 5281 sbi->s_stripe, sbi->s_cluster_ratio); 5282 sbi->s_stripe = 0; 5283 } 5284 sbi->s_extent_max_zeroout_kb = 32; 5285 5286 /* 5287 * set up enough so that it can read an inode 5288 */ 5289 sb->s_op = &ext4_sops; 5290 sb->s_export_op = &ext4_export_ops; 5291 sb->s_xattr = ext4_xattr_handlers; 5292 #ifdef CONFIG_FS_ENCRYPTION 5293 sb->s_cop = &ext4_cryptops; 5294 #endif 5295 #ifdef CONFIG_FS_VERITY 5296 sb->s_vop = &ext4_verityops; 5297 #endif 5298 #ifdef CONFIG_QUOTA 5299 sb->dq_op = &ext4_quota_operations; 5300 if (ext4_has_feature_quota(sb)) 5301 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5302 else 5303 sb->s_qcop = &ext4_qctl_operations; 5304 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5305 #endif 5306 super_set_uuid(sb, es->s_uuid, sizeof(es->s_uuid)); 5307 super_set_sysfs_name_bdev(sb); 5308 5309 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5310 mutex_init(&sbi->s_orphan_lock); 5311 5312 ext4_fast_commit_init(sb); 5313 5314 sb->s_root = NULL; 5315 5316 needs_recovery = (es->s_last_orphan != 0 || 5317 ext4_has_feature_orphan_present(sb) || 5318 ext4_has_feature_journal_needs_recovery(sb)); 5319 5320 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) { 5321 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)); 5322 if (err) 5323 goto failed_mount3a; 5324 } 5325 5326 err = -EINVAL; 5327 /* 5328 * The first inode we look at is the journal inode. Don't try 5329 * root first: it may be modified in the journal! 5330 */ 5331 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5332 err = ext4_load_and_init_journal(sb, es, ctx); 5333 if (err) 5334 goto failed_mount3a; 5335 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5336 ext4_has_feature_journal_needs_recovery(sb)) { 5337 ext4_msg(sb, KERN_ERR, "required journal recovery " 5338 "suppressed and not mounted read-only"); 5339 goto failed_mount3a; 5340 } else { 5341 /* Nojournal mode, all journal mount options are illegal */ 5342 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5343 ext4_msg(sb, KERN_ERR, "can't mount with " 5344 "journal_async_commit, fs mounted w/o journal"); 5345 goto failed_mount3a; 5346 } 5347 5348 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 5349 ext4_msg(sb, KERN_ERR, "can't mount with " 5350 "journal_checksum, fs mounted w/o journal"); 5351 goto failed_mount3a; 5352 } 5353 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 5354 ext4_msg(sb, KERN_ERR, "can't mount with " 5355 "commit=%lu, fs mounted w/o journal", 5356 sbi->s_commit_interval / HZ); 5357 goto failed_mount3a; 5358 } 5359 if (EXT4_MOUNT_DATA_FLAGS & 5360 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 5361 ext4_msg(sb, KERN_ERR, "can't mount with " 5362 "data=, fs mounted w/o journal"); 5363 goto failed_mount3a; 5364 } 5365 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5366 clear_opt(sb, JOURNAL_CHECKSUM); 5367 clear_opt(sb, DATA_FLAGS); 5368 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5369 sbi->s_journal = NULL; 5370 needs_recovery = 0; 5371 } 5372 5373 if (!test_opt(sb, NO_MBCACHE)) { 5374 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5375 if (!sbi->s_ea_block_cache) { 5376 ext4_msg(sb, KERN_ERR, 5377 "Failed to create ea_block_cache"); 5378 err = -EINVAL; 5379 goto failed_mount_wq; 5380 } 5381 5382 if (ext4_has_feature_ea_inode(sb)) { 5383 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5384 if (!sbi->s_ea_inode_cache) { 5385 ext4_msg(sb, KERN_ERR, 5386 "Failed to create ea_inode_cache"); 5387 err = -EINVAL; 5388 goto failed_mount_wq; 5389 } 5390 } 5391 } 5392 5393 /* 5394 * Get the # of file system overhead blocks from the 5395 * superblock if present. 5396 */ 5397 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5398 /* ignore the precalculated value if it is ridiculous */ 5399 if (sbi->s_overhead > ext4_blocks_count(es)) 5400 sbi->s_overhead = 0; 5401 /* 5402 * If the bigalloc feature is not enabled recalculating the 5403 * overhead doesn't take long, so we might as well just redo 5404 * it to make sure we are using the correct value. 5405 */ 5406 if (!ext4_has_feature_bigalloc(sb)) 5407 sbi->s_overhead = 0; 5408 if (sbi->s_overhead == 0) { 5409 err = ext4_calculate_overhead(sb); 5410 if (err) 5411 goto failed_mount_wq; 5412 } 5413 5414 /* 5415 * The maximum number of concurrent works can be high and 5416 * concurrency isn't really necessary. Limit it to 1. 5417 */ 5418 EXT4_SB(sb)->rsv_conversion_wq = 5419 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5420 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5421 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5422 err = -ENOMEM; 5423 goto failed_mount4; 5424 } 5425 5426 /* 5427 * The jbd2_journal_load will have done any necessary log recovery, 5428 * so we can safely mount the rest of the filesystem now. 5429 */ 5430 5431 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5432 if (IS_ERR(root)) { 5433 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5434 err = PTR_ERR(root); 5435 root = NULL; 5436 goto failed_mount4; 5437 } 5438 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5439 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5440 iput(root); 5441 err = -EFSCORRUPTED; 5442 goto failed_mount4; 5443 } 5444 5445 generic_set_sb_d_ops(sb); 5446 sb->s_root = d_make_root(root); 5447 if (!sb->s_root) { 5448 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5449 err = -ENOMEM; 5450 goto failed_mount4; 5451 } 5452 5453 err = ext4_setup_super(sb, es, sb_rdonly(sb)); 5454 if (err == -EROFS) { 5455 sb->s_flags |= SB_RDONLY; 5456 } else if (err) 5457 goto failed_mount4a; 5458 5459 ext4_set_resv_clusters(sb); 5460 5461 if (test_opt(sb, BLOCK_VALIDITY)) { 5462 err = ext4_setup_system_zone(sb); 5463 if (err) { 5464 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5465 "zone (%d)", err); 5466 goto failed_mount4a; 5467 } 5468 } 5469 ext4_fc_replay_cleanup(sb); 5470 5471 ext4_ext_init(sb); 5472 5473 /* 5474 * Enable optimize_scan if number of groups is > threshold. This can be 5475 * turned off by passing "mb_optimize_scan=0". This can also be 5476 * turned on forcefully by passing "mb_optimize_scan=1". 5477 */ 5478 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5479 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5480 set_opt2(sb, MB_OPTIMIZE_SCAN); 5481 else 5482 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5483 } 5484 5485 err = ext4_mb_init(sb); 5486 if (err) { 5487 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5488 err); 5489 goto failed_mount5; 5490 } 5491 5492 /* 5493 * We can only set up the journal commit callback once 5494 * mballoc is initialized 5495 */ 5496 if (sbi->s_journal) 5497 sbi->s_journal->j_commit_callback = 5498 ext4_journal_commit_callback; 5499 5500 err = ext4_percpu_param_init(sbi); 5501 if (err) 5502 goto failed_mount6; 5503 5504 if (ext4_has_feature_flex_bg(sb)) 5505 if (!ext4_fill_flex_info(sb)) { 5506 ext4_msg(sb, KERN_ERR, 5507 "unable to initialize " 5508 "flex_bg meta info!"); 5509 err = -ENOMEM; 5510 goto failed_mount6; 5511 } 5512 5513 err = ext4_register_li_request(sb, first_not_zeroed); 5514 if (err) 5515 goto failed_mount6; 5516 5517 err = ext4_init_orphan_info(sb); 5518 if (err) 5519 goto failed_mount7; 5520 #ifdef CONFIG_QUOTA 5521 /* Enable quota usage during mount. */ 5522 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5523 err = ext4_enable_quotas(sb); 5524 if (err) 5525 goto failed_mount8; 5526 } 5527 #endif /* CONFIG_QUOTA */ 5528 5529 /* 5530 * Save the original bdev mapping's wb_err value which could be 5531 * used to detect the metadata async write error. 5532 */ 5533 spin_lock_init(&sbi->s_bdev_wb_lock); 5534 errseq_check_and_advance(&sb->s_bdev->bd_mapping->wb_err, 5535 &sbi->s_bdev_wb_err); 5536 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5537 ext4_orphan_cleanup(sb, es); 5538 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5539 /* 5540 * Update the checksum after updating free space/inode counters and 5541 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5542 * checksum in the buffer cache until it is written out and 5543 * e2fsprogs programs trying to open a file system immediately 5544 * after it is mounted can fail. 5545 */ 5546 ext4_superblock_csum_set(sb); 5547 if (needs_recovery) { 5548 ext4_msg(sb, KERN_INFO, "recovery complete"); 5549 err = ext4_mark_recovery_complete(sb, es); 5550 if (err) 5551 goto failed_mount9; 5552 } 5553 5554 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) 5555 ext4_msg(sb, KERN_WARNING, 5556 "mounting with \"discard\" option, but the device does not support discard"); 5557 5558 if (es->s_error_count) 5559 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5560 5561 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5562 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5563 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5564 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5565 atomic_set(&sbi->s_warning_count, 0); 5566 atomic_set(&sbi->s_msg_count, 0); 5567 5568 /* Register sysfs after all initializations are complete. */ 5569 err = ext4_register_sysfs(sb); 5570 if (err) 5571 goto failed_mount9; 5572 5573 return 0; 5574 5575 failed_mount9: 5576 ext4_quotas_off(sb, EXT4_MAXQUOTAS); 5577 failed_mount8: __maybe_unused 5578 ext4_release_orphan_info(sb); 5579 failed_mount7: 5580 ext4_unregister_li_request(sb); 5581 failed_mount6: 5582 ext4_mb_release(sb); 5583 ext4_flex_groups_free(sbi); 5584 ext4_percpu_param_destroy(sbi); 5585 failed_mount5: 5586 ext4_ext_release(sb); 5587 ext4_release_system_zone(sb); 5588 failed_mount4a: 5589 dput(sb->s_root); 5590 sb->s_root = NULL; 5591 failed_mount4: 5592 ext4_msg(sb, KERN_ERR, "mount failed"); 5593 if (EXT4_SB(sb)->rsv_conversion_wq) 5594 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5595 failed_mount_wq: 5596 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5597 sbi->s_ea_inode_cache = NULL; 5598 5599 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5600 sbi->s_ea_block_cache = NULL; 5601 5602 if (sbi->s_journal) { 5603 /* flush s_sb_upd_work before journal destroy. */ 5604 flush_work(&sbi->s_sb_upd_work); 5605 jbd2_journal_destroy(sbi->s_journal); 5606 sbi->s_journal = NULL; 5607 } 5608 failed_mount3a: 5609 ext4_es_unregister_shrinker(sbi); 5610 failed_mount3: 5611 /* flush s_sb_upd_work before sbi destroy */ 5612 flush_work(&sbi->s_sb_upd_work); 5613 del_timer_sync(&sbi->s_err_report); 5614 ext4_stop_mmpd(sbi); 5615 ext4_group_desc_free(sbi); 5616 failed_mount: 5617 if (sbi->s_chksum_driver) 5618 crypto_free_shash(sbi->s_chksum_driver); 5619 5620 #if IS_ENABLED(CONFIG_UNICODE) 5621 utf8_unload(sb->s_encoding); 5622 #endif 5623 5624 #ifdef CONFIG_QUOTA 5625 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++) 5626 kfree(get_qf_name(sb, sbi, i)); 5627 #endif 5628 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5629 brelse(sbi->s_sbh); 5630 if (sbi->s_journal_bdev_file) { 5631 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file)); 5632 bdev_fput(sbi->s_journal_bdev_file); 5633 } 5634 out_fail: 5635 invalidate_bdev(sb->s_bdev); 5636 sb->s_fs_info = NULL; 5637 return err; 5638 } 5639 5640 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5641 { 5642 struct ext4_fs_context *ctx = fc->fs_private; 5643 struct ext4_sb_info *sbi; 5644 const char *descr; 5645 int ret; 5646 5647 sbi = ext4_alloc_sbi(sb); 5648 if (!sbi) 5649 return -ENOMEM; 5650 5651 fc->s_fs_info = sbi; 5652 5653 /* Cleanup superblock name */ 5654 strreplace(sb->s_id, '/', '!'); 5655 5656 sbi->s_sb_block = 1; /* Default super block location */ 5657 if (ctx->spec & EXT4_SPEC_s_sb_block) 5658 sbi->s_sb_block = ctx->s_sb_block; 5659 5660 ret = __ext4_fill_super(fc, sb); 5661 if (ret < 0) 5662 goto free_sbi; 5663 5664 if (sbi->s_journal) { 5665 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5666 descr = " journalled data mode"; 5667 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5668 descr = " ordered data mode"; 5669 else 5670 descr = " writeback data mode"; 5671 } else 5672 descr = "out journal"; 5673 5674 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5675 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. " 5676 "Quota mode: %s.", &sb->s_uuid, 5677 sb_rdonly(sb) ? "ro" : "r/w", descr, 5678 ext4_quota_mode(sb)); 5679 5680 /* Update the s_overhead_clusters if necessary */ 5681 ext4_update_overhead(sb, false); 5682 return 0; 5683 5684 free_sbi: 5685 ext4_free_sbi(sbi); 5686 fc->s_fs_info = NULL; 5687 return ret; 5688 } 5689 5690 static int ext4_get_tree(struct fs_context *fc) 5691 { 5692 return get_tree_bdev(fc, ext4_fill_super); 5693 } 5694 5695 /* 5696 * Setup any per-fs journal parameters now. We'll do this both on 5697 * initial mount, once the journal has been initialised but before we've 5698 * done any recovery; and again on any subsequent remount. 5699 */ 5700 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5701 { 5702 struct ext4_sb_info *sbi = EXT4_SB(sb); 5703 5704 journal->j_commit_interval = sbi->s_commit_interval; 5705 journal->j_min_batch_time = sbi->s_min_batch_time; 5706 journal->j_max_batch_time = sbi->s_max_batch_time; 5707 ext4_fc_init(sb, journal); 5708 5709 write_lock(&journal->j_state_lock); 5710 if (test_opt(sb, BARRIER)) 5711 journal->j_flags |= JBD2_BARRIER; 5712 else 5713 journal->j_flags &= ~JBD2_BARRIER; 5714 if (test_opt(sb, DATA_ERR_ABORT)) 5715 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5716 else 5717 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5718 /* 5719 * Always enable journal cycle record option, letting the journal 5720 * records log transactions continuously between each mount. 5721 */ 5722 journal->j_flags |= JBD2_CYCLE_RECORD; 5723 write_unlock(&journal->j_state_lock); 5724 } 5725 5726 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5727 unsigned int journal_inum) 5728 { 5729 struct inode *journal_inode; 5730 5731 /* 5732 * Test for the existence of a valid inode on disk. Bad things 5733 * happen if we iget() an unused inode, as the subsequent iput() 5734 * will try to delete it. 5735 */ 5736 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5737 if (IS_ERR(journal_inode)) { 5738 ext4_msg(sb, KERN_ERR, "no journal found"); 5739 return ERR_CAST(journal_inode); 5740 } 5741 if (!journal_inode->i_nlink) { 5742 make_bad_inode(journal_inode); 5743 iput(journal_inode); 5744 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5745 return ERR_PTR(-EFSCORRUPTED); 5746 } 5747 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) { 5748 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5749 iput(journal_inode); 5750 return ERR_PTR(-EFSCORRUPTED); 5751 } 5752 5753 ext4_debug("Journal inode found at %p: %lld bytes\n", 5754 journal_inode, journal_inode->i_size); 5755 return journal_inode; 5756 } 5757 5758 static int ext4_journal_bmap(journal_t *journal, sector_t *block) 5759 { 5760 struct ext4_map_blocks map; 5761 int ret; 5762 5763 if (journal->j_inode == NULL) 5764 return 0; 5765 5766 map.m_lblk = *block; 5767 map.m_len = 1; 5768 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0); 5769 if (ret <= 0) { 5770 ext4_msg(journal->j_inode->i_sb, KERN_CRIT, 5771 "journal bmap failed: block %llu ret %d\n", 5772 *block, ret); 5773 jbd2_journal_abort(journal, ret ? ret : -EIO); 5774 return ret; 5775 } 5776 *block = map.m_pblk; 5777 return 0; 5778 } 5779 5780 static journal_t *ext4_open_inode_journal(struct super_block *sb, 5781 unsigned int journal_inum) 5782 { 5783 struct inode *journal_inode; 5784 journal_t *journal; 5785 5786 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5787 if (IS_ERR(journal_inode)) 5788 return ERR_CAST(journal_inode); 5789 5790 journal = jbd2_journal_init_inode(journal_inode); 5791 if (IS_ERR(journal)) { 5792 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5793 iput(journal_inode); 5794 return ERR_CAST(journal); 5795 } 5796 journal->j_private = sb; 5797 journal->j_bmap = ext4_journal_bmap; 5798 ext4_init_journal_params(sb, journal); 5799 return journal; 5800 } 5801 5802 static struct file *ext4_get_journal_blkdev(struct super_block *sb, 5803 dev_t j_dev, ext4_fsblk_t *j_start, 5804 ext4_fsblk_t *j_len) 5805 { 5806 struct buffer_head *bh; 5807 struct block_device *bdev; 5808 struct file *bdev_file; 5809 int hblock, blocksize; 5810 ext4_fsblk_t sb_block; 5811 unsigned long offset; 5812 struct ext4_super_block *es; 5813 int errno; 5814 5815 bdev_file = bdev_file_open_by_dev(j_dev, 5816 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES, 5817 sb, &fs_holder_ops); 5818 if (IS_ERR(bdev_file)) { 5819 ext4_msg(sb, KERN_ERR, 5820 "failed to open journal device unknown-block(%u,%u) %ld", 5821 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev_file)); 5822 return bdev_file; 5823 } 5824 5825 bdev = file_bdev(bdev_file); 5826 blocksize = sb->s_blocksize; 5827 hblock = bdev_logical_block_size(bdev); 5828 if (blocksize < hblock) { 5829 ext4_msg(sb, KERN_ERR, 5830 "blocksize too small for journal device"); 5831 errno = -EINVAL; 5832 goto out_bdev; 5833 } 5834 5835 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5836 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5837 set_blocksize(bdev_file, blocksize); 5838 bh = __bread(bdev, sb_block, blocksize); 5839 if (!bh) { 5840 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5841 "external journal"); 5842 errno = -EINVAL; 5843 goto out_bdev; 5844 } 5845 5846 es = (struct ext4_super_block *) (bh->b_data + offset); 5847 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5848 !(le32_to_cpu(es->s_feature_incompat) & 5849 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5850 ext4_msg(sb, KERN_ERR, "external journal has bad superblock"); 5851 errno = -EFSCORRUPTED; 5852 goto out_bh; 5853 } 5854 5855 if ((le32_to_cpu(es->s_feature_ro_compat) & 5856 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5857 es->s_checksum != ext4_superblock_csum(sb, es)) { 5858 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock"); 5859 errno = -EFSCORRUPTED; 5860 goto out_bh; 5861 } 5862 5863 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5864 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5865 errno = -EFSCORRUPTED; 5866 goto out_bh; 5867 } 5868 5869 *j_start = sb_block + 1; 5870 *j_len = ext4_blocks_count(es); 5871 brelse(bh); 5872 return bdev_file; 5873 5874 out_bh: 5875 brelse(bh); 5876 out_bdev: 5877 bdev_fput(bdev_file); 5878 return ERR_PTR(errno); 5879 } 5880 5881 static journal_t *ext4_open_dev_journal(struct super_block *sb, 5882 dev_t j_dev) 5883 { 5884 journal_t *journal; 5885 ext4_fsblk_t j_start; 5886 ext4_fsblk_t j_len; 5887 struct file *bdev_file; 5888 int errno = 0; 5889 5890 bdev_file = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len); 5891 if (IS_ERR(bdev_file)) 5892 return ERR_CAST(bdev_file); 5893 5894 journal = jbd2_journal_init_dev(file_bdev(bdev_file), sb->s_bdev, j_start, 5895 j_len, sb->s_blocksize); 5896 if (IS_ERR(journal)) { 5897 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5898 errno = PTR_ERR(journal); 5899 goto out_bdev; 5900 } 5901 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5902 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5903 "user (unsupported) - %d", 5904 be32_to_cpu(journal->j_superblock->s_nr_users)); 5905 errno = -EINVAL; 5906 goto out_journal; 5907 } 5908 journal->j_private = sb; 5909 EXT4_SB(sb)->s_journal_bdev_file = bdev_file; 5910 ext4_init_journal_params(sb, journal); 5911 return journal; 5912 5913 out_journal: 5914 jbd2_journal_destroy(journal); 5915 out_bdev: 5916 bdev_fput(bdev_file); 5917 return ERR_PTR(errno); 5918 } 5919 5920 static int ext4_load_journal(struct super_block *sb, 5921 struct ext4_super_block *es, 5922 unsigned long journal_devnum) 5923 { 5924 journal_t *journal; 5925 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5926 dev_t journal_dev; 5927 int err = 0; 5928 int really_read_only; 5929 int journal_dev_ro; 5930 5931 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5932 return -EFSCORRUPTED; 5933 5934 if (journal_devnum && 5935 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5936 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5937 "numbers have changed"); 5938 journal_dev = new_decode_dev(journal_devnum); 5939 } else 5940 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5941 5942 if (journal_inum && journal_dev) { 5943 ext4_msg(sb, KERN_ERR, 5944 "filesystem has both journal inode and journal device!"); 5945 return -EINVAL; 5946 } 5947 5948 if (journal_inum) { 5949 journal = ext4_open_inode_journal(sb, journal_inum); 5950 if (IS_ERR(journal)) 5951 return PTR_ERR(journal); 5952 } else { 5953 journal = ext4_open_dev_journal(sb, journal_dev); 5954 if (IS_ERR(journal)) 5955 return PTR_ERR(journal); 5956 } 5957 5958 journal_dev_ro = bdev_read_only(journal->j_dev); 5959 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5960 5961 if (journal_dev_ro && !sb_rdonly(sb)) { 5962 ext4_msg(sb, KERN_ERR, 5963 "journal device read-only, try mounting with '-o ro'"); 5964 err = -EROFS; 5965 goto err_out; 5966 } 5967 5968 /* 5969 * Are we loading a blank journal or performing recovery after a 5970 * crash? For recovery, we need to check in advance whether we 5971 * can get read-write access to the device. 5972 */ 5973 if (ext4_has_feature_journal_needs_recovery(sb)) { 5974 if (sb_rdonly(sb)) { 5975 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5976 "required on readonly filesystem"); 5977 if (really_read_only) { 5978 ext4_msg(sb, KERN_ERR, "write access " 5979 "unavailable, cannot proceed " 5980 "(try mounting with noload)"); 5981 err = -EROFS; 5982 goto err_out; 5983 } 5984 ext4_msg(sb, KERN_INFO, "write access will " 5985 "be enabled during recovery"); 5986 } 5987 } 5988 5989 if (!(journal->j_flags & JBD2_BARRIER)) 5990 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5991 5992 if (!ext4_has_feature_journal_needs_recovery(sb)) 5993 err = jbd2_journal_wipe(journal, !really_read_only); 5994 if (!err) { 5995 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5996 __le16 orig_state; 5997 bool changed = false; 5998 5999 if (save) 6000 memcpy(save, ((char *) es) + 6001 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 6002 err = jbd2_journal_load(journal); 6003 if (save && memcmp(((char *) es) + EXT4_S_ERR_START, 6004 save, EXT4_S_ERR_LEN)) { 6005 memcpy(((char *) es) + EXT4_S_ERR_START, 6006 save, EXT4_S_ERR_LEN); 6007 changed = true; 6008 } 6009 kfree(save); 6010 orig_state = es->s_state; 6011 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state & 6012 EXT4_ERROR_FS); 6013 if (orig_state != es->s_state) 6014 changed = true; 6015 /* Write out restored error information to the superblock */ 6016 if (changed && !really_read_only) { 6017 int err2; 6018 err2 = ext4_commit_super(sb); 6019 err = err ? : err2; 6020 } 6021 } 6022 6023 if (err) { 6024 ext4_msg(sb, KERN_ERR, "error loading journal"); 6025 goto err_out; 6026 } 6027 6028 EXT4_SB(sb)->s_journal = journal; 6029 err = ext4_clear_journal_err(sb, es); 6030 if (err) { 6031 EXT4_SB(sb)->s_journal = NULL; 6032 jbd2_journal_destroy(journal); 6033 return err; 6034 } 6035 6036 if (!really_read_only && journal_devnum && 6037 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 6038 es->s_journal_dev = cpu_to_le32(journal_devnum); 6039 ext4_commit_super(sb); 6040 } 6041 if (!really_read_only && journal_inum && 6042 journal_inum != le32_to_cpu(es->s_journal_inum)) { 6043 es->s_journal_inum = cpu_to_le32(journal_inum); 6044 ext4_commit_super(sb); 6045 } 6046 6047 return 0; 6048 6049 err_out: 6050 jbd2_journal_destroy(journal); 6051 return err; 6052 } 6053 6054 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 6055 static void ext4_update_super(struct super_block *sb) 6056 { 6057 struct ext4_sb_info *sbi = EXT4_SB(sb); 6058 struct ext4_super_block *es = sbi->s_es; 6059 struct buffer_head *sbh = sbi->s_sbh; 6060 6061 lock_buffer(sbh); 6062 /* 6063 * If the file system is mounted read-only, don't update the 6064 * superblock write time. This avoids updating the superblock 6065 * write time when we are mounting the root file system 6066 * read/only but we need to replay the journal; at that point, 6067 * for people who are east of GMT and who make their clock 6068 * tick in localtime for Windows bug-for-bug compatibility, 6069 * the clock is set in the future, and this will cause e2fsck 6070 * to complain and force a full file system check. 6071 */ 6072 if (!sb_rdonly(sb)) 6073 ext4_update_tstamp(es, s_wtime); 6074 es->s_kbytes_written = 6075 cpu_to_le64(sbi->s_kbytes_written + 6076 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 6077 sbi->s_sectors_written_start) >> 1)); 6078 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 6079 ext4_free_blocks_count_set(es, 6080 EXT4_C2B(sbi, percpu_counter_sum_positive( 6081 &sbi->s_freeclusters_counter))); 6082 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 6083 es->s_free_inodes_count = 6084 cpu_to_le32(percpu_counter_sum_positive( 6085 &sbi->s_freeinodes_counter)); 6086 /* Copy error information to the on-disk superblock */ 6087 spin_lock(&sbi->s_error_lock); 6088 if (sbi->s_add_error_count > 0) { 6089 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6090 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6091 __ext4_update_tstamp(&es->s_first_error_time, 6092 &es->s_first_error_time_hi, 6093 sbi->s_first_error_time); 6094 strtomem_pad(es->s_first_error_func, 6095 sbi->s_first_error_func, 0); 6096 es->s_first_error_line = 6097 cpu_to_le32(sbi->s_first_error_line); 6098 es->s_first_error_ino = 6099 cpu_to_le32(sbi->s_first_error_ino); 6100 es->s_first_error_block = 6101 cpu_to_le64(sbi->s_first_error_block); 6102 es->s_first_error_errcode = 6103 ext4_errno_to_code(sbi->s_first_error_code); 6104 } 6105 __ext4_update_tstamp(&es->s_last_error_time, 6106 &es->s_last_error_time_hi, 6107 sbi->s_last_error_time); 6108 strtomem_pad(es->s_last_error_func, sbi->s_last_error_func, 0); 6109 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6110 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6111 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6112 es->s_last_error_errcode = 6113 ext4_errno_to_code(sbi->s_last_error_code); 6114 /* 6115 * Start the daily error reporting function if it hasn't been 6116 * started already 6117 */ 6118 if (!es->s_error_count) 6119 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6120 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6121 sbi->s_add_error_count = 0; 6122 } 6123 spin_unlock(&sbi->s_error_lock); 6124 6125 ext4_superblock_csum_set(sb); 6126 unlock_buffer(sbh); 6127 } 6128 6129 static int ext4_commit_super(struct super_block *sb) 6130 { 6131 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6132 6133 if (!sbh) 6134 return -EINVAL; 6135 6136 ext4_update_super(sb); 6137 6138 lock_buffer(sbh); 6139 /* Buffer got discarded which means block device got invalidated */ 6140 if (!buffer_mapped(sbh)) { 6141 unlock_buffer(sbh); 6142 return -EIO; 6143 } 6144 6145 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6146 /* 6147 * Oh, dear. A previous attempt to write the 6148 * superblock failed. This could happen because the 6149 * USB device was yanked out. Or it could happen to 6150 * be a transient write error and maybe the block will 6151 * be remapped. Nothing we can do but to retry the 6152 * write and hope for the best. 6153 */ 6154 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6155 "superblock detected"); 6156 clear_buffer_write_io_error(sbh); 6157 set_buffer_uptodate(sbh); 6158 } 6159 get_bh(sbh); 6160 /* Clear potential dirty bit if it was journalled update */ 6161 clear_buffer_dirty(sbh); 6162 sbh->b_end_io = end_buffer_write_sync; 6163 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6164 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6165 wait_on_buffer(sbh); 6166 if (buffer_write_io_error(sbh)) { 6167 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6168 "superblock"); 6169 clear_buffer_write_io_error(sbh); 6170 set_buffer_uptodate(sbh); 6171 return -EIO; 6172 } 6173 return 0; 6174 } 6175 6176 /* 6177 * Have we just finished recovery? If so, and if we are mounting (or 6178 * remounting) the filesystem readonly, then we will end up with a 6179 * consistent fs on disk. Record that fact. 6180 */ 6181 static int ext4_mark_recovery_complete(struct super_block *sb, 6182 struct ext4_super_block *es) 6183 { 6184 int err; 6185 journal_t *journal = EXT4_SB(sb)->s_journal; 6186 6187 if (!ext4_has_feature_journal(sb)) { 6188 if (journal != NULL) { 6189 ext4_error(sb, "Journal got removed while the fs was " 6190 "mounted!"); 6191 return -EFSCORRUPTED; 6192 } 6193 return 0; 6194 } 6195 jbd2_journal_lock_updates(journal); 6196 err = jbd2_journal_flush(journal, 0); 6197 if (err < 0) 6198 goto out; 6199 6200 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6201 ext4_has_feature_orphan_present(sb))) { 6202 if (!ext4_orphan_file_empty(sb)) { 6203 ext4_error(sb, "Orphan file not empty on read-only fs."); 6204 err = -EFSCORRUPTED; 6205 goto out; 6206 } 6207 ext4_clear_feature_journal_needs_recovery(sb); 6208 ext4_clear_feature_orphan_present(sb); 6209 ext4_commit_super(sb); 6210 } 6211 out: 6212 jbd2_journal_unlock_updates(journal); 6213 return err; 6214 } 6215 6216 /* 6217 * If we are mounting (or read-write remounting) a filesystem whose journal 6218 * has recorded an error from a previous lifetime, move that error to the 6219 * main filesystem now. 6220 */ 6221 static int ext4_clear_journal_err(struct super_block *sb, 6222 struct ext4_super_block *es) 6223 { 6224 journal_t *journal; 6225 int j_errno; 6226 const char *errstr; 6227 6228 if (!ext4_has_feature_journal(sb)) { 6229 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6230 return -EFSCORRUPTED; 6231 } 6232 6233 journal = EXT4_SB(sb)->s_journal; 6234 6235 /* 6236 * Now check for any error status which may have been recorded in the 6237 * journal by a prior ext4_error() or ext4_abort() 6238 */ 6239 6240 j_errno = jbd2_journal_errno(journal); 6241 if (j_errno) { 6242 char nbuf[16]; 6243 6244 errstr = ext4_decode_error(sb, j_errno, nbuf); 6245 ext4_warning(sb, "Filesystem error recorded " 6246 "from previous mount: %s", errstr); 6247 6248 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6249 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6250 j_errno = ext4_commit_super(sb); 6251 if (j_errno) 6252 return j_errno; 6253 ext4_warning(sb, "Marked fs in need of filesystem check."); 6254 6255 jbd2_journal_clear_err(journal); 6256 jbd2_journal_update_sb_errno(journal); 6257 } 6258 return 0; 6259 } 6260 6261 /* 6262 * Force the running and committing transactions to commit, 6263 * and wait on the commit. 6264 */ 6265 int ext4_force_commit(struct super_block *sb) 6266 { 6267 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal); 6268 } 6269 6270 static int ext4_sync_fs(struct super_block *sb, int wait) 6271 { 6272 int ret = 0; 6273 tid_t target; 6274 bool needs_barrier = false; 6275 struct ext4_sb_info *sbi = EXT4_SB(sb); 6276 6277 if (unlikely(ext4_forced_shutdown(sb))) 6278 return 0; 6279 6280 trace_ext4_sync_fs(sb, wait); 6281 flush_workqueue(sbi->rsv_conversion_wq); 6282 /* 6283 * Writeback quota in non-journalled quota case - journalled quota has 6284 * no dirty dquots 6285 */ 6286 dquot_writeback_dquots(sb, -1); 6287 /* 6288 * Data writeback is possible w/o journal transaction, so barrier must 6289 * being sent at the end of the function. But we can skip it if 6290 * transaction_commit will do it for us. 6291 */ 6292 if (sbi->s_journal) { 6293 target = jbd2_get_latest_transaction(sbi->s_journal); 6294 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6295 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6296 needs_barrier = true; 6297 6298 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6299 if (wait) 6300 ret = jbd2_log_wait_commit(sbi->s_journal, 6301 target); 6302 } 6303 } else if (wait && test_opt(sb, BARRIER)) 6304 needs_barrier = true; 6305 if (needs_barrier) { 6306 int err; 6307 err = blkdev_issue_flush(sb->s_bdev); 6308 if (!ret) 6309 ret = err; 6310 } 6311 6312 return ret; 6313 } 6314 6315 /* 6316 * LVM calls this function before a (read-only) snapshot is created. This 6317 * gives us a chance to flush the journal completely and mark the fs clean. 6318 * 6319 * Note that only this function cannot bring a filesystem to be in a clean 6320 * state independently. It relies on upper layer to stop all data & metadata 6321 * modifications. 6322 */ 6323 static int ext4_freeze(struct super_block *sb) 6324 { 6325 int error = 0; 6326 journal_t *journal = EXT4_SB(sb)->s_journal; 6327 6328 if (journal) { 6329 /* Now we set up the journal barrier. */ 6330 jbd2_journal_lock_updates(journal); 6331 6332 /* 6333 * Don't clear the needs_recovery flag if we failed to 6334 * flush the journal. 6335 */ 6336 error = jbd2_journal_flush(journal, 0); 6337 if (error < 0) 6338 goto out; 6339 6340 /* Journal blocked and flushed, clear needs_recovery flag. */ 6341 ext4_clear_feature_journal_needs_recovery(sb); 6342 if (ext4_orphan_file_empty(sb)) 6343 ext4_clear_feature_orphan_present(sb); 6344 } 6345 6346 error = ext4_commit_super(sb); 6347 out: 6348 if (journal) 6349 /* we rely on upper layer to stop further updates */ 6350 jbd2_journal_unlock_updates(journal); 6351 return error; 6352 } 6353 6354 /* 6355 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6356 * flag here, even though the filesystem is not technically dirty yet. 6357 */ 6358 static int ext4_unfreeze(struct super_block *sb) 6359 { 6360 if (ext4_forced_shutdown(sb)) 6361 return 0; 6362 6363 if (EXT4_SB(sb)->s_journal) { 6364 /* Reset the needs_recovery flag before the fs is unlocked. */ 6365 ext4_set_feature_journal_needs_recovery(sb); 6366 if (ext4_has_feature_orphan_file(sb)) 6367 ext4_set_feature_orphan_present(sb); 6368 } 6369 6370 ext4_commit_super(sb); 6371 return 0; 6372 } 6373 6374 /* 6375 * Structure to save mount options for ext4_remount's benefit 6376 */ 6377 struct ext4_mount_options { 6378 unsigned long s_mount_opt; 6379 unsigned long s_mount_opt2; 6380 kuid_t s_resuid; 6381 kgid_t s_resgid; 6382 unsigned long s_commit_interval; 6383 u32 s_min_batch_time, s_max_batch_time; 6384 #ifdef CONFIG_QUOTA 6385 int s_jquota_fmt; 6386 char *s_qf_names[EXT4_MAXQUOTAS]; 6387 #endif 6388 }; 6389 6390 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6391 { 6392 struct ext4_fs_context *ctx = fc->fs_private; 6393 struct ext4_super_block *es; 6394 struct ext4_sb_info *sbi = EXT4_SB(sb); 6395 unsigned long old_sb_flags; 6396 struct ext4_mount_options old_opts; 6397 ext4_group_t g; 6398 int err = 0; 6399 int alloc_ctx; 6400 #ifdef CONFIG_QUOTA 6401 int enable_quota = 0; 6402 int i, j; 6403 char *to_free[EXT4_MAXQUOTAS]; 6404 #endif 6405 6406 6407 /* Store the original options */ 6408 old_sb_flags = sb->s_flags; 6409 old_opts.s_mount_opt = sbi->s_mount_opt; 6410 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6411 old_opts.s_resuid = sbi->s_resuid; 6412 old_opts.s_resgid = sbi->s_resgid; 6413 old_opts.s_commit_interval = sbi->s_commit_interval; 6414 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6415 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6416 #ifdef CONFIG_QUOTA 6417 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6418 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6419 if (sbi->s_qf_names[i]) { 6420 char *qf_name = get_qf_name(sb, sbi, i); 6421 6422 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6423 if (!old_opts.s_qf_names[i]) { 6424 for (j = 0; j < i; j++) 6425 kfree(old_opts.s_qf_names[j]); 6426 return -ENOMEM; 6427 } 6428 } else 6429 old_opts.s_qf_names[i] = NULL; 6430 #endif 6431 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6432 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6433 ctx->journal_ioprio = 6434 sbi->s_journal->j_task->io_context->ioprio; 6435 else 6436 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 6437 6438 } 6439 6440 /* 6441 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause 6442 * two calls to ext4_should_dioread_nolock() to return inconsistent 6443 * values, triggering WARN_ON in ext4_add_complete_io(). we grab 6444 * here s_writepages_rwsem to avoid race between writepages ops and 6445 * remount. 6446 */ 6447 alloc_ctx = ext4_writepages_down_write(sb); 6448 ext4_apply_options(fc, sb); 6449 ext4_writepages_up_write(sb, alloc_ctx); 6450 6451 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6452 test_opt(sb, JOURNAL_CHECKSUM)) { 6453 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6454 "during remount not supported; ignoring"); 6455 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6456 } 6457 6458 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6459 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6460 ext4_msg(sb, KERN_ERR, "can't mount with " 6461 "both data=journal and delalloc"); 6462 err = -EINVAL; 6463 goto restore_opts; 6464 } 6465 if (test_opt(sb, DIOREAD_NOLOCK)) { 6466 ext4_msg(sb, KERN_ERR, "can't mount with " 6467 "both data=journal and dioread_nolock"); 6468 err = -EINVAL; 6469 goto restore_opts; 6470 } 6471 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6472 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6473 ext4_msg(sb, KERN_ERR, "can't mount with " 6474 "journal_async_commit in data=ordered mode"); 6475 err = -EINVAL; 6476 goto restore_opts; 6477 } 6478 } 6479 6480 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6481 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6482 err = -EINVAL; 6483 goto restore_opts; 6484 } 6485 6486 if (test_opt2(sb, ABORT)) 6487 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6488 6489 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6490 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6491 6492 es = sbi->s_es; 6493 6494 if (sbi->s_journal) { 6495 ext4_init_journal_params(sb, sbi->s_journal); 6496 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6497 } 6498 6499 /* Flush outstanding errors before changing fs state */ 6500 flush_work(&sbi->s_sb_upd_work); 6501 6502 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6503 if (ext4_forced_shutdown(sb)) { 6504 err = -EROFS; 6505 goto restore_opts; 6506 } 6507 6508 if (fc->sb_flags & SB_RDONLY) { 6509 err = sync_filesystem(sb); 6510 if (err < 0) 6511 goto restore_opts; 6512 err = dquot_suspend(sb, -1); 6513 if (err < 0) 6514 goto restore_opts; 6515 6516 /* 6517 * First of all, the unconditional stuff we have to do 6518 * to disable replay of the journal when we next remount 6519 */ 6520 sb->s_flags |= SB_RDONLY; 6521 6522 /* 6523 * OK, test if we are remounting a valid rw partition 6524 * readonly, and if so set the rdonly flag and then 6525 * mark the partition as valid again. 6526 */ 6527 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6528 (sbi->s_mount_state & EXT4_VALID_FS)) 6529 es->s_state = cpu_to_le16(sbi->s_mount_state); 6530 6531 if (sbi->s_journal) { 6532 /* 6533 * We let remount-ro finish even if marking fs 6534 * as clean failed... 6535 */ 6536 ext4_mark_recovery_complete(sb, es); 6537 } 6538 } else { 6539 /* Make sure we can mount this feature set readwrite */ 6540 if (ext4_has_feature_readonly(sb) || 6541 !ext4_feature_set_ok(sb, 0)) { 6542 err = -EROFS; 6543 goto restore_opts; 6544 } 6545 /* 6546 * Make sure the group descriptor checksums 6547 * are sane. If they aren't, refuse to remount r/w. 6548 */ 6549 for (g = 0; g < sbi->s_groups_count; g++) { 6550 struct ext4_group_desc *gdp = 6551 ext4_get_group_desc(sb, g, NULL); 6552 6553 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6554 ext4_msg(sb, KERN_ERR, 6555 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6556 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6557 le16_to_cpu(gdp->bg_checksum)); 6558 err = -EFSBADCRC; 6559 goto restore_opts; 6560 } 6561 } 6562 6563 /* 6564 * If we have an unprocessed orphan list hanging 6565 * around from a previously readonly bdev mount, 6566 * require a full umount/remount for now. 6567 */ 6568 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6569 ext4_msg(sb, KERN_WARNING, "Couldn't " 6570 "remount RDWR because of unprocessed " 6571 "orphan inode list. Please " 6572 "umount/remount instead"); 6573 err = -EINVAL; 6574 goto restore_opts; 6575 } 6576 6577 /* 6578 * Mounting a RDONLY partition read-write, so reread 6579 * and store the current valid flag. (It may have 6580 * been changed by e2fsck since we originally mounted 6581 * the partition.) 6582 */ 6583 if (sbi->s_journal) { 6584 err = ext4_clear_journal_err(sb, es); 6585 if (err) 6586 goto restore_opts; 6587 } 6588 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6589 ~EXT4_FC_REPLAY); 6590 6591 err = ext4_setup_super(sb, es, 0); 6592 if (err) 6593 goto restore_opts; 6594 6595 sb->s_flags &= ~SB_RDONLY; 6596 if (ext4_has_feature_mmp(sb)) { 6597 err = ext4_multi_mount_protect(sb, 6598 le64_to_cpu(es->s_mmp_block)); 6599 if (err) 6600 goto restore_opts; 6601 } 6602 #ifdef CONFIG_QUOTA 6603 enable_quota = 1; 6604 #endif 6605 } 6606 } 6607 6608 /* 6609 * Handle creation of system zone data early because it can fail. 6610 * Releasing of existing data is done when we are sure remount will 6611 * succeed. 6612 */ 6613 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6614 err = ext4_setup_system_zone(sb); 6615 if (err) 6616 goto restore_opts; 6617 } 6618 6619 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6620 err = ext4_commit_super(sb); 6621 if (err) 6622 goto restore_opts; 6623 } 6624 6625 #ifdef CONFIG_QUOTA 6626 if (enable_quota) { 6627 if (sb_any_quota_suspended(sb)) 6628 dquot_resume(sb, -1); 6629 else if (ext4_has_feature_quota(sb)) { 6630 err = ext4_enable_quotas(sb); 6631 if (err) 6632 goto restore_opts; 6633 } 6634 } 6635 /* Release old quota file names */ 6636 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6637 kfree(old_opts.s_qf_names[i]); 6638 #endif 6639 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6640 ext4_release_system_zone(sb); 6641 6642 /* 6643 * Reinitialize lazy itable initialization thread based on 6644 * current settings 6645 */ 6646 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6647 ext4_unregister_li_request(sb); 6648 else { 6649 ext4_group_t first_not_zeroed; 6650 first_not_zeroed = ext4_has_uninit_itable(sb); 6651 ext4_register_li_request(sb, first_not_zeroed); 6652 } 6653 6654 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6655 ext4_stop_mmpd(sbi); 6656 6657 return 0; 6658 6659 restore_opts: 6660 /* 6661 * If there was a failing r/w to ro transition, we may need to 6662 * re-enable quota 6663 */ 6664 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) && 6665 sb_any_quota_suspended(sb)) 6666 dquot_resume(sb, -1); 6667 6668 alloc_ctx = ext4_writepages_down_write(sb); 6669 sb->s_flags = old_sb_flags; 6670 sbi->s_mount_opt = old_opts.s_mount_opt; 6671 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6672 sbi->s_resuid = old_opts.s_resuid; 6673 sbi->s_resgid = old_opts.s_resgid; 6674 sbi->s_commit_interval = old_opts.s_commit_interval; 6675 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6676 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6677 ext4_writepages_up_write(sb, alloc_ctx); 6678 6679 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6680 ext4_release_system_zone(sb); 6681 #ifdef CONFIG_QUOTA 6682 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6683 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6684 to_free[i] = get_qf_name(sb, sbi, i); 6685 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6686 } 6687 synchronize_rcu(); 6688 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6689 kfree(to_free[i]); 6690 #endif 6691 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6692 ext4_stop_mmpd(sbi); 6693 return err; 6694 } 6695 6696 static int ext4_reconfigure(struct fs_context *fc) 6697 { 6698 struct super_block *sb = fc->root->d_sb; 6699 int ret; 6700 6701 fc->s_fs_info = EXT4_SB(sb); 6702 6703 ret = ext4_check_opt_consistency(fc, sb); 6704 if (ret < 0) 6705 return ret; 6706 6707 ret = __ext4_remount(fc, sb); 6708 if (ret < 0) 6709 return ret; 6710 6711 ext4_msg(sb, KERN_INFO, "re-mounted %pU %s. Quota mode: %s.", 6712 &sb->s_uuid, sb_rdonly(sb) ? "ro" : "r/w", 6713 ext4_quota_mode(sb)); 6714 6715 return 0; 6716 } 6717 6718 #ifdef CONFIG_QUOTA 6719 static int ext4_statfs_project(struct super_block *sb, 6720 kprojid_t projid, struct kstatfs *buf) 6721 { 6722 struct kqid qid; 6723 struct dquot *dquot; 6724 u64 limit; 6725 u64 curblock; 6726 6727 qid = make_kqid_projid(projid); 6728 dquot = dqget(sb, qid); 6729 if (IS_ERR(dquot)) 6730 return PTR_ERR(dquot); 6731 spin_lock(&dquot->dq_dqb_lock); 6732 6733 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6734 dquot->dq_dqb.dqb_bhardlimit); 6735 limit >>= sb->s_blocksize_bits; 6736 6737 if (limit && buf->f_blocks > limit) { 6738 curblock = (dquot->dq_dqb.dqb_curspace + 6739 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6740 buf->f_blocks = limit; 6741 buf->f_bfree = buf->f_bavail = 6742 (buf->f_blocks > curblock) ? 6743 (buf->f_blocks - curblock) : 0; 6744 } 6745 6746 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6747 dquot->dq_dqb.dqb_ihardlimit); 6748 if (limit && buf->f_files > limit) { 6749 buf->f_files = limit; 6750 buf->f_ffree = 6751 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6752 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6753 } 6754 6755 spin_unlock(&dquot->dq_dqb_lock); 6756 dqput(dquot); 6757 return 0; 6758 } 6759 #endif 6760 6761 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6762 { 6763 struct super_block *sb = dentry->d_sb; 6764 struct ext4_sb_info *sbi = EXT4_SB(sb); 6765 struct ext4_super_block *es = sbi->s_es; 6766 ext4_fsblk_t overhead = 0, resv_blocks; 6767 s64 bfree; 6768 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6769 6770 if (!test_opt(sb, MINIX_DF)) 6771 overhead = sbi->s_overhead; 6772 6773 buf->f_type = EXT4_SUPER_MAGIC; 6774 buf->f_bsize = sb->s_blocksize; 6775 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6776 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6777 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6778 /* prevent underflow in case that few free space is available */ 6779 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6780 buf->f_bavail = buf->f_bfree - 6781 (ext4_r_blocks_count(es) + resv_blocks); 6782 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6783 buf->f_bavail = 0; 6784 buf->f_files = le32_to_cpu(es->s_inodes_count); 6785 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6786 buf->f_namelen = EXT4_NAME_LEN; 6787 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6788 6789 #ifdef CONFIG_QUOTA 6790 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6791 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6792 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6793 #endif 6794 return 0; 6795 } 6796 6797 6798 #ifdef CONFIG_QUOTA 6799 6800 /* 6801 * Helper functions so that transaction is started before we acquire dqio_sem 6802 * to keep correct lock ordering of transaction > dqio_sem 6803 */ 6804 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6805 { 6806 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6807 } 6808 6809 static int ext4_write_dquot(struct dquot *dquot) 6810 { 6811 int ret, err; 6812 handle_t *handle; 6813 struct inode *inode; 6814 6815 inode = dquot_to_inode(dquot); 6816 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6817 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6818 if (IS_ERR(handle)) 6819 return PTR_ERR(handle); 6820 ret = dquot_commit(dquot); 6821 if (ret < 0) 6822 ext4_error_err(dquot->dq_sb, -ret, 6823 "Failed to commit dquot type %d", 6824 dquot->dq_id.type); 6825 err = ext4_journal_stop(handle); 6826 if (!ret) 6827 ret = err; 6828 return ret; 6829 } 6830 6831 static int ext4_acquire_dquot(struct dquot *dquot) 6832 { 6833 int ret, err; 6834 handle_t *handle; 6835 6836 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6837 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6838 if (IS_ERR(handle)) 6839 return PTR_ERR(handle); 6840 ret = dquot_acquire(dquot); 6841 if (ret < 0) 6842 ext4_error_err(dquot->dq_sb, -ret, 6843 "Failed to acquire dquot type %d", 6844 dquot->dq_id.type); 6845 err = ext4_journal_stop(handle); 6846 if (!ret) 6847 ret = err; 6848 return ret; 6849 } 6850 6851 static int ext4_release_dquot(struct dquot *dquot) 6852 { 6853 int ret, err; 6854 handle_t *handle; 6855 6856 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6857 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6858 if (IS_ERR(handle)) { 6859 /* Release dquot anyway to avoid endless cycle in dqput() */ 6860 dquot_release(dquot); 6861 return PTR_ERR(handle); 6862 } 6863 ret = dquot_release(dquot); 6864 if (ret < 0) 6865 ext4_error_err(dquot->dq_sb, -ret, 6866 "Failed to release dquot type %d", 6867 dquot->dq_id.type); 6868 err = ext4_journal_stop(handle); 6869 if (!ret) 6870 ret = err; 6871 return ret; 6872 } 6873 6874 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6875 { 6876 struct super_block *sb = dquot->dq_sb; 6877 6878 if (ext4_is_quota_journalled(sb)) { 6879 dquot_mark_dquot_dirty(dquot); 6880 return ext4_write_dquot(dquot); 6881 } else { 6882 return dquot_mark_dquot_dirty(dquot); 6883 } 6884 } 6885 6886 static int ext4_write_info(struct super_block *sb, int type) 6887 { 6888 int ret, err; 6889 handle_t *handle; 6890 6891 /* Data block + inode block */ 6892 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 6893 if (IS_ERR(handle)) 6894 return PTR_ERR(handle); 6895 ret = dquot_commit_info(sb, type); 6896 err = ext4_journal_stop(handle); 6897 if (!ret) 6898 ret = err; 6899 return ret; 6900 } 6901 6902 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6903 { 6904 struct ext4_inode_info *ei = EXT4_I(inode); 6905 6906 /* The first argument of lockdep_set_subclass has to be 6907 * *exactly* the same as the argument to init_rwsem() --- in 6908 * this case, in init_once() --- or lockdep gets unhappy 6909 * because the name of the lock is set using the 6910 * stringification of the argument to init_rwsem(). 6911 */ 6912 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6913 lockdep_set_subclass(&ei->i_data_sem, subclass); 6914 } 6915 6916 /* 6917 * Standard function to be called on quota_on 6918 */ 6919 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6920 const struct path *path) 6921 { 6922 int err; 6923 6924 if (!test_opt(sb, QUOTA)) 6925 return -EINVAL; 6926 6927 /* Quotafile not on the same filesystem? */ 6928 if (path->dentry->d_sb != sb) 6929 return -EXDEV; 6930 6931 /* Quota already enabled for this file? */ 6932 if (IS_NOQUOTA(d_inode(path->dentry))) 6933 return -EBUSY; 6934 6935 /* Journaling quota? */ 6936 if (EXT4_SB(sb)->s_qf_names[type]) { 6937 /* Quotafile not in fs root? */ 6938 if (path->dentry->d_parent != sb->s_root) 6939 ext4_msg(sb, KERN_WARNING, 6940 "Quota file not on filesystem root. " 6941 "Journaled quota will not work"); 6942 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6943 } else { 6944 /* 6945 * Clear the flag just in case mount options changed since 6946 * last time. 6947 */ 6948 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6949 } 6950 6951 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6952 err = dquot_quota_on(sb, type, format_id, path); 6953 if (!err) { 6954 struct inode *inode = d_inode(path->dentry); 6955 handle_t *handle; 6956 6957 /* 6958 * Set inode flags to prevent userspace from messing with quota 6959 * files. If this fails, we return success anyway since quotas 6960 * are already enabled and this is not a hard failure. 6961 */ 6962 inode_lock(inode); 6963 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6964 if (IS_ERR(handle)) 6965 goto unlock_inode; 6966 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6967 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6968 S_NOATIME | S_IMMUTABLE); 6969 err = ext4_mark_inode_dirty(handle, inode); 6970 ext4_journal_stop(handle); 6971 unlock_inode: 6972 inode_unlock(inode); 6973 if (err) 6974 dquot_quota_off(sb, type); 6975 } 6976 if (err) 6977 lockdep_set_quota_inode(path->dentry->d_inode, 6978 I_DATA_SEM_NORMAL); 6979 return err; 6980 } 6981 6982 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum) 6983 { 6984 switch (type) { 6985 case USRQUOTA: 6986 return qf_inum == EXT4_USR_QUOTA_INO; 6987 case GRPQUOTA: 6988 return qf_inum == EXT4_GRP_QUOTA_INO; 6989 case PRJQUOTA: 6990 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO; 6991 default: 6992 BUG(); 6993 } 6994 } 6995 6996 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6997 unsigned int flags) 6998 { 6999 int err; 7000 struct inode *qf_inode; 7001 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7002 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7003 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7004 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7005 }; 7006 7007 BUG_ON(!ext4_has_feature_quota(sb)); 7008 7009 if (!qf_inums[type]) 7010 return -EPERM; 7011 7012 if (!ext4_check_quota_inum(type, qf_inums[type])) { 7013 ext4_error(sb, "Bad quota inum: %lu, type: %d", 7014 qf_inums[type], type); 7015 return -EUCLEAN; 7016 } 7017 7018 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 7019 if (IS_ERR(qf_inode)) { 7020 ext4_error(sb, "Bad quota inode: %lu, type: %d", 7021 qf_inums[type], type); 7022 return PTR_ERR(qf_inode); 7023 } 7024 7025 /* Don't account quota for quota files to avoid recursion */ 7026 qf_inode->i_flags |= S_NOQUOTA; 7027 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 7028 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 7029 if (err) 7030 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 7031 iput(qf_inode); 7032 7033 return err; 7034 } 7035 7036 /* Enable usage tracking for all quota types. */ 7037 int ext4_enable_quotas(struct super_block *sb) 7038 { 7039 int type, err = 0; 7040 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 7041 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 7042 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 7043 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 7044 }; 7045 bool quota_mopt[EXT4_MAXQUOTAS] = { 7046 test_opt(sb, USRQUOTA), 7047 test_opt(sb, GRPQUOTA), 7048 test_opt(sb, PRJQUOTA), 7049 }; 7050 7051 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 7052 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 7053 if (qf_inums[type]) { 7054 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 7055 DQUOT_USAGE_ENABLED | 7056 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 7057 if (err) { 7058 ext4_warning(sb, 7059 "Failed to enable quota tracking " 7060 "(type=%d, err=%d, ino=%lu). " 7061 "Please run e2fsck to fix.", type, 7062 err, qf_inums[type]); 7063 7064 ext4_quotas_off(sb, type); 7065 return err; 7066 } 7067 } 7068 } 7069 return 0; 7070 } 7071 7072 static int ext4_quota_off(struct super_block *sb, int type) 7073 { 7074 struct inode *inode = sb_dqopt(sb)->files[type]; 7075 handle_t *handle; 7076 int err; 7077 7078 /* Force all delayed allocation blocks to be allocated. 7079 * Caller already holds s_umount sem */ 7080 if (test_opt(sb, DELALLOC)) 7081 sync_filesystem(sb); 7082 7083 if (!inode || !igrab(inode)) 7084 goto out; 7085 7086 err = dquot_quota_off(sb, type); 7087 if (err || ext4_has_feature_quota(sb)) 7088 goto out_put; 7089 /* 7090 * When the filesystem was remounted read-only first, we cannot cleanup 7091 * inode flags here. Bad luck but people should be using QUOTA feature 7092 * these days anyway. 7093 */ 7094 if (sb_rdonly(sb)) 7095 goto out_put; 7096 7097 inode_lock(inode); 7098 /* 7099 * Update modification times of quota files when userspace can 7100 * start looking at them. If we fail, we return success anyway since 7101 * this is not a hard failure and quotas are already disabled. 7102 */ 7103 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 7104 if (IS_ERR(handle)) { 7105 err = PTR_ERR(handle); 7106 goto out_unlock; 7107 } 7108 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7109 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7110 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 7111 err = ext4_mark_inode_dirty(handle, inode); 7112 ext4_journal_stop(handle); 7113 out_unlock: 7114 inode_unlock(inode); 7115 out_put: 7116 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7117 iput(inode); 7118 return err; 7119 out: 7120 return dquot_quota_off(sb, type); 7121 } 7122 7123 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7124 * acquiring the locks... As quota files are never truncated and quota code 7125 * itself serializes the operations (and no one else should touch the files) 7126 * we don't have to be afraid of races */ 7127 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7128 size_t len, loff_t off) 7129 { 7130 struct inode *inode = sb_dqopt(sb)->files[type]; 7131 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7132 int offset = off & (sb->s_blocksize - 1); 7133 int tocopy; 7134 size_t toread; 7135 struct buffer_head *bh; 7136 loff_t i_size = i_size_read(inode); 7137 7138 if (off > i_size) 7139 return 0; 7140 if (off+len > i_size) 7141 len = i_size-off; 7142 toread = len; 7143 while (toread > 0) { 7144 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread); 7145 bh = ext4_bread(NULL, inode, blk, 0); 7146 if (IS_ERR(bh)) 7147 return PTR_ERR(bh); 7148 if (!bh) /* A hole? */ 7149 memset(data, 0, tocopy); 7150 else 7151 memcpy(data, bh->b_data+offset, tocopy); 7152 brelse(bh); 7153 offset = 0; 7154 toread -= tocopy; 7155 data += tocopy; 7156 blk++; 7157 } 7158 return len; 7159 } 7160 7161 /* Write to quotafile (we know the transaction is already started and has 7162 * enough credits) */ 7163 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7164 const char *data, size_t len, loff_t off) 7165 { 7166 struct inode *inode = sb_dqopt(sb)->files[type]; 7167 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7168 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7169 int retries = 0; 7170 struct buffer_head *bh; 7171 handle_t *handle = journal_current_handle(); 7172 7173 if (!handle) { 7174 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7175 " cancelled because transaction is not started", 7176 (unsigned long long)off, (unsigned long long)len); 7177 return -EIO; 7178 } 7179 /* 7180 * Since we account only one data block in transaction credits, 7181 * then it is impossible to cross a block boundary. 7182 */ 7183 if (sb->s_blocksize - offset < len) { 7184 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7185 " cancelled because not block aligned", 7186 (unsigned long long)off, (unsigned long long)len); 7187 return -EIO; 7188 } 7189 7190 do { 7191 bh = ext4_bread(handle, inode, blk, 7192 EXT4_GET_BLOCKS_CREATE | 7193 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7194 } while (PTR_ERR(bh) == -ENOSPC && 7195 ext4_should_retry_alloc(inode->i_sb, &retries)); 7196 if (IS_ERR(bh)) 7197 return PTR_ERR(bh); 7198 if (!bh) 7199 goto out; 7200 BUFFER_TRACE(bh, "get write access"); 7201 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7202 if (err) { 7203 brelse(bh); 7204 return err; 7205 } 7206 lock_buffer(bh); 7207 memcpy(bh->b_data+offset, data, len); 7208 flush_dcache_page(bh->b_page); 7209 unlock_buffer(bh); 7210 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7211 brelse(bh); 7212 out: 7213 if (inode->i_size < off + len) { 7214 i_size_write(inode, off + len); 7215 EXT4_I(inode)->i_disksize = inode->i_size; 7216 err2 = ext4_mark_inode_dirty(handle, inode); 7217 if (unlikely(err2 && !err)) 7218 err = err2; 7219 } 7220 return err ? err : len; 7221 } 7222 #endif 7223 7224 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7225 static inline void register_as_ext2(void) 7226 { 7227 int err = register_filesystem(&ext2_fs_type); 7228 if (err) 7229 printk(KERN_WARNING 7230 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7231 } 7232 7233 static inline void unregister_as_ext2(void) 7234 { 7235 unregister_filesystem(&ext2_fs_type); 7236 } 7237 7238 static inline int ext2_feature_set_ok(struct super_block *sb) 7239 { 7240 if (ext4_has_unknown_ext2_incompat_features(sb)) 7241 return 0; 7242 if (sb_rdonly(sb)) 7243 return 1; 7244 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7245 return 0; 7246 return 1; 7247 } 7248 #else 7249 static inline void register_as_ext2(void) { } 7250 static inline void unregister_as_ext2(void) { } 7251 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7252 #endif 7253 7254 static inline void register_as_ext3(void) 7255 { 7256 int err = register_filesystem(&ext3_fs_type); 7257 if (err) 7258 printk(KERN_WARNING 7259 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7260 } 7261 7262 static inline void unregister_as_ext3(void) 7263 { 7264 unregister_filesystem(&ext3_fs_type); 7265 } 7266 7267 static inline int ext3_feature_set_ok(struct super_block *sb) 7268 { 7269 if (ext4_has_unknown_ext3_incompat_features(sb)) 7270 return 0; 7271 if (!ext4_has_feature_journal(sb)) 7272 return 0; 7273 if (sb_rdonly(sb)) 7274 return 1; 7275 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7276 return 0; 7277 return 1; 7278 } 7279 7280 static void ext4_kill_sb(struct super_block *sb) 7281 { 7282 struct ext4_sb_info *sbi = EXT4_SB(sb); 7283 struct file *bdev_file = sbi ? sbi->s_journal_bdev_file : NULL; 7284 7285 kill_block_super(sb); 7286 7287 if (bdev_file) 7288 bdev_fput(bdev_file); 7289 } 7290 7291 static struct file_system_type ext4_fs_type = { 7292 .owner = THIS_MODULE, 7293 .name = "ext4", 7294 .init_fs_context = ext4_init_fs_context, 7295 .parameters = ext4_param_specs, 7296 .kill_sb = ext4_kill_sb, 7297 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 7298 }; 7299 MODULE_ALIAS_FS("ext4"); 7300 7301 /* Shared across all ext4 file systems */ 7302 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 7303 7304 static int __init ext4_init_fs(void) 7305 { 7306 int i, err; 7307 7308 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7309 ext4_li_info = NULL; 7310 7311 /* Build-time check for flags consistency */ 7312 ext4_check_flag_values(); 7313 7314 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 7315 init_waitqueue_head(&ext4__ioend_wq[i]); 7316 7317 err = ext4_init_es(); 7318 if (err) 7319 return err; 7320 7321 err = ext4_init_pending(); 7322 if (err) 7323 goto out7; 7324 7325 err = ext4_init_post_read_processing(); 7326 if (err) 7327 goto out6; 7328 7329 err = ext4_init_pageio(); 7330 if (err) 7331 goto out5; 7332 7333 err = ext4_init_system_zone(); 7334 if (err) 7335 goto out4; 7336 7337 err = ext4_init_sysfs(); 7338 if (err) 7339 goto out3; 7340 7341 err = ext4_init_mballoc(); 7342 if (err) 7343 goto out2; 7344 err = init_inodecache(); 7345 if (err) 7346 goto out1; 7347 7348 err = ext4_fc_init_dentry_cache(); 7349 if (err) 7350 goto out05; 7351 7352 register_as_ext3(); 7353 register_as_ext2(); 7354 err = register_filesystem(&ext4_fs_type); 7355 if (err) 7356 goto out; 7357 7358 return 0; 7359 out: 7360 unregister_as_ext2(); 7361 unregister_as_ext3(); 7362 ext4_fc_destroy_dentry_cache(); 7363 out05: 7364 destroy_inodecache(); 7365 out1: 7366 ext4_exit_mballoc(); 7367 out2: 7368 ext4_exit_sysfs(); 7369 out3: 7370 ext4_exit_system_zone(); 7371 out4: 7372 ext4_exit_pageio(); 7373 out5: 7374 ext4_exit_post_read_processing(); 7375 out6: 7376 ext4_exit_pending(); 7377 out7: 7378 ext4_exit_es(); 7379 7380 return err; 7381 } 7382 7383 static void __exit ext4_exit_fs(void) 7384 { 7385 ext4_destroy_lazyinit_thread(); 7386 unregister_as_ext2(); 7387 unregister_as_ext3(); 7388 unregister_filesystem(&ext4_fs_type); 7389 ext4_fc_destroy_dentry_cache(); 7390 destroy_inodecache(); 7391 ext4_exit_mballoc(); 7392 ext4_exit_sysfs(); 7393 ext4_exit_system_zone(); 7394 ext4_exit_pageio(); 7395 ext4_exit_post_read_processing(); 7396 ext4_exit_es(); 7397 ext4_exit_pending(); 7398 } 7399 7400 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7401 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7402 MODULE_LICENSE("GPL"); 7403 MODULE_SOFTDEP("pre: crc32c"); 7404 module_init(ext4_init_fs) 7405 module_exit(ext4_exit_fs) 7406