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 const struct fs_parameter_spec ext4_param_specs[]; 97 98 /* 99 * Lock ordering 100 * 101 * page fault path: 102 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start 103 * -> page lock -> i_data_sem (rw) 104 * 105 * buffered write path: 106 * sb_start_write -> i_mutex -> mmap_lock 107 * sb_start_write -> i_mutex -> transaction start -> page lock -> 108 * i_data_sem (rw) 109 * 110 * truncate: 111 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) -> 112 * page lock 113 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start -> 114 * i_data_sem (rw) 115 * 116 * direct IO: 117 * sb_start_write -> i_mutex -> mmap_lock 118 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 119 * 120 * writepages: 121 * transaction start -> page lock(s) -> i_data_sem (rw) 122 */ 123 124 static const struct fs_context_operations ext4_context_ops = { 125 .parse_param = ext4_parse_param, 126 .get_tree = ext4_get_tree, 127 .reconfigure = ext4_reconfigure, 128 .free = ext4_fc_free, 129 }; 130 131 132 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 133 static struct file_system_type ext2_fs_type = { 134 .owner = THIS_MODULE, 135 .name = "ext2", 136 .init_fs_context = ext4_init_fs_context, 137 .parameters = ext4_param_specs, 138 .kill_sb = kill_block_super, 139 .fs_flags = FS_REQUIRES_DEV, 140 }; 141 MODULE_ALIAS_FS("ext2"); 142 MODULE_ALIAS("ext2"); 143 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type) 144 #else 145 #define IS_EXT2_SB(sb) (0) 146 #endif 147 148 149 static struct file_system_type ext3_fs_type = { 150 .owner = THIS_MODULE, 151 .name = "ext3", 152 .init_fs_context = ext4_init_fs_context, 153 .parameters = ext4_param_specs, 154 .kill_sb = kill_block_super, 155 .fs_flags = FS_REQUIRES_DEV, 156 }; 157 MODULE_ALIAS_FS("ext3"); 158 MODULE_ALIAS("ext3"); 159 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type) 160 161 162 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 163 bh_end_io_t *end_io) 164 { 165 /* 166 * buffer's verified bit is no longer valid after reading from 167 * disk again due to write out error, clear it to make sure we 168 * recheck the buffer contents. 169 */ 170 clear_buffer_verified(bh); 171 172 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 173 get_bh(bh); 174 submit_bh(REQ_OP_READ | op_flags, bh); 175 } 176 177 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags, 178 bh_end_io_t *end_io) 179 { 180 BUG_ON(!buffer_locked(bh)); 181 182 if (ext4_buffer_uptodate(bh)) { 183 unlock_buffer(bh); 184 return; 185 } 186 __ext4_read_bh(bh, op_flags, end_io); 187 } 188 189 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io) 190 { 191 BUG_ON(!buffer_locked(bh)); 192 193 if (ext4_buffer_uptodate(bh)) { 194 unlock_buffer(bh); 195 return 0; 196 } 197 198 __ext4_read_bh(bh, op_flags, end_io); 199 200 wait_on_buffer(bh); 201 if (buffer_uptodate(bh)) 202 return 0; 203 return -EIO; 204 } 205 206 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 207 { 208 lock_buffer(bh); 209 if (!wait) { 210 ext4_read_bh_nowait(bh, op_flags, NULL); 211 return 0; 212 } 213 return ext4_read_bh(bh, op_flags, NULL); 214 } 215 216 /* 217 * This works like __bread_gfp() except it uses ERR_PTR for error 218 * returns. Currently with sb_bread it's impossible to distinguish 219 * between ENOMEM and EIO situations (since both result in a NULL 220 * return. 221 */ 222 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 223 sector_t block, 224 blk_opf_t op_flags, gfp_t gfp) 225 { 226 struct buffer_head *bh; 227 int ret; 228 229 bh = sb_getblk_gfp(sb, block, gfp); 230 if (bh == NULL) 231 return ERR_PTR(-ENOMEM); 232 if (ext4_buffer_uptodate(bh)) 233 return bh; 234 235 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 236 if (ret) { 237 put_bh(bh); 238 return ERR_PTR(ret); 239 } 240 return bh; 241 } 242 243 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 244 blk_opf_t op_flags) 245 { 246 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 247 } 248 249 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 250 sector_t block) 251 { 252 return __ext4_sb_bread_gfp(sb, block, 0, 0); 253 } 254 255 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 256 { 257 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 258 259 if (likely(bh)) { 260 if (trylock_buffer(bh)) 261 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL); 262 brelse(bh); 263 } 264 } 265 266 static int ext4_verify_csum_type(struct super_block *sb, 267 struct ext4_super_block *es) 268 { 269 if (!ext4_has_feature_metadata_csum(sb)) 270 return 1; 271 272 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 273 } 274 275 __le32 ext4_superblock_csum(struct super_block *sb, 276 struct ext4_super_block *es) 277 { 278 struct ext4_sb_info *sbi = EXT4_SB(sb); 279 int offset = offsetof(struct ext4_super_block, s_checksum); 280 __u32 csum; 281 282 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 283 284 return cpu_to_le32(csum); 285 } 286 287 static int ext4_superblock_csum_verify(struct super_block *sb, 288 struct ext4_super_block *es) 289 { 290 if (!ext4_has_metadata_csum(sb)) 291 return 1; 292 293 return es->s_checksum == ext4_superblock_csum(sb, es); 294 } 295 296 void ext4_superblock_csum_set(struct super_block *sb) 297 { 298 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 299 300 if (!ext4_has_metadata_csum(sb)) 301 return; 302 303 es->s_checksum = ext4_superblock_csum(sb, es); 304 } 305 306 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 307 struct ext4_group_desc *bg) 308 { 309 return le32_to_cpu(bg->bg_block_bitmap_lo) | 310 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 311 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 312 } 313 314 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 315 struct ext4_group_desc *bg) 316 { 317 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 318 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 319 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 320 } 321 322 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 323 struct ext4_group_desc *bg) 324 { 325 return le32_to_cpu(bg->bg_inode_table_lo) | 326 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 327 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 328 } 329 330 __u32 ext4_free_group_clusters(struct super_block *sb, 331 struct ext4_group_desc *bg) 332 { 333 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 334 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 335 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 336 } 337 338 __u32 ext4_free_inodes_count(struct super_block *sb, 339 struct ext4_group_desc *bg) 340 { 341 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 342 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 343 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 344 } 345 346 __u32 ext4_used_dirs_count(struct super_block *sb, 347 struct ext4_group_desc *bg) 348 { 349 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 350 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 351 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 352 } 353 354 __u32 ext4_itable_unused_count(struct super_block *sb, 355 struct ext4_group_desc *bg) 356 { 357 return le16_to_cpu(bg->bg_itable_unused_lo) | 358 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 359 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 360 } 361 362 void ext4_block_bitmap_set(struct super_block *sb, 363 struct ext4_group_desc *bg, ext4_fsblk_t blk) 364 { 365 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 366 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 367 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 368 } 369 370 void ext4_inode_bitmap_set(struct super_block *sb, 371 struct ext4_group_desc *bg, ext4_fsblk_t blk) 372 { 373 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 374 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 375 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 376 } 377 378 void ext4_inode_table_set(struct super_block *sb, 379 struct ext4_group_desc *bg, ext4_fsblk_t blk) 380 { 381 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 382 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 383 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 384 } 385 386 void ext4_free_group_clusters_set(struct super_block *sb, 387 struct ext4_group_desc *bg, __u32 count) 388 { 389 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 390 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 391 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 392 } 393 394 void ext4_free_inodes_set(struct super_block *sb, 395 struct ext4_group_desc *bg, __u32 count) 396 { 397 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 398 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 399 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 400 } 401 402 void ext4_used_dirs_set(struct super_block *sb, 403 struct ext4_group_desc *bg, __u32 count) 404 { 405 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 406 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 407 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 408 } 409 410 void ext4_itable_unused_set(struct super_block *sb, 411 struct ext4_group_desc *bg, __u32 count) 412 { 413 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 414 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 415 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 416 } 417 418 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 419 { 420 now = clamp_val(now, 0, (1ull << 40) - 1); 421 422 *lo = cpu_to_le32(lower_32_bits(now)); 423 *hi = upper_32_bits(now); 424 } 425 426 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 427 { 428 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 429 } 430 #define ext4_update_tstamp(es, tstamp) \ 431 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 432 ktime_get_real_seconds()) 433 #define ext4_get_tstamp(es, tstamp) \ 434 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 435 436 /* 437 * The del_gendisk() function uninitializes the disk-specific data 438 * structures, including the bdi structure, without telling anyone 439 * else. Once this happens, any attempt to call mark_buffer_dirty() 440 * (for example, by ext4_commit_super), will cause a kernel OOPS. 441 * This is a kludge to prevent these oops until we can put in a proper 442 * hook in del_gendisk() to inform the VFS and file system layers. 443 */ 444 static int block_device_ejected(struct super_block *sb) 445 { 446 struct inode *bd_inode = sb->s_bdev->bd_inode; 447 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 448 449 return bdi->dev == NULL; 450 } 451 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 453 { 454 struct super_block *sb = journal->j_private; 455 struct ext4_sb_info *sbi = EXT4_SB(sb); 456 int error = is_journal_aborted(journal); 457 struct ext4_journal_cb_entry *jce; 458 459 BUG_ON(txn->t_state == T_FINISHED); 460 461 ext4_process_freed_data(sb, txn->t_tid); 462 463 spin_lock(&sbi->s_md_lock); 464 while (!list_empty(&txn->t_private_list)) { 465 jce = list_entry(txn->t_private_list.next, 466 struct ext4_journal_cb_entry, jce_list); 467 list_del_init(&jce->jce_list); 468 spin_unlock(&sbi->s_md_lock); 469 jce->jce_func(sb, jce, error); 470 spin_lock(&sbi->s_md_lock); 471 } 472 spin_unlock(&sbi->s_md_lock); 473 } 474 475 /* 476 * This writepage callback for write_cache_pages() 477 * takes care of a few cases after page cleaning. 478 * 479 * write_cache_pages() already checks for dirty pages 480 * and calls clear_page_dirty_for_io(), which we want, 481 * to write protect the pages. 482 * 483 * However, we may have to redirty a page (see below.) 484 */ 485 static int ext4_journalled_writepage_callback(struct page *page, 486 struct writeback_control *wbc, 487 void *data) 488 { 489 transaction_t *transaction = (transaction_t *) data; 490 struct buffer_head *bh, *head; 491 struct journal_head *jh; 492 493 bh = head = page_buffers(page); 494 do { 495 /* 496 * We have to redirty a page in these cases: 497 * 1) If buffer is dirty, it means the page was dirty because it 498 * contains a buffer that needs checkpointing. So the dirty bit 499 * needs to be preserved so that checkpointing writes the buffer 500 * properly. 501 * 2) If buffer is not part of the committing transaction 502 * (we may have just accidentally come across this buffer because 503 * inode range tracking is not exact) or if the currently running 504 * transaction already contains this buffer as well, dirty bit 505 * needs to be preserved so that the buffer gets writeprotected 506 * properly on running transaction's commit. 507 */ 508 jh = bh2jh(bh); 509 if (buffer_dirty(bh) || 510 (jh && (jh->b_transaction != transaction || 511 jh->b_next_transaction))) { 512 redirty_page_for_writepage(wbc, page); 513 goto out; 514 } 515 } while ((bh = bh->b_this_page) != head); 516 517 out: 518 return AOP_WRITEPAGE_ACTIVATE; 519 } 520 521 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 522 { 523 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 524 struct writeback_control wbc = { 525 .sync_mode = WB_SYNC_ALL, 526 .nr_to_write = LONG_MAX, 527 .range_start = jinode->i_dirty_start, 528 .range_end = jinode->i_dirty_end, 529 }; 530 531 return write_cache_pages(mapping, &wbc, 532 ext4_journalled_writepage_callback, 533 jinode->i_transaction); 534 } 535 536 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 537 { 538 int ret; 539 540 if (ext4_should_journal_data(jinode->i_vfs_inode)) 541 ret = ext4_journalled_submit_inode_data_buffers(jinode); 542 else 543 ret = jbd2_journal_submit_inode_data_buffers(jinode); 544 545 return ret; 546 } 547 548 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 549 { 550 int ret = 0; 551 552 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 553 ret = jbd2_journal_finish_inode_data_buffers(jinode); 554 555 return ret; 556 } 557 558 static bool system_going_down(void) 559 { 560 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 561 || system_state == SYSTEM_RESTART; 562 } 563 564 struct ext4_err_translation { 565 int code; 566 int errno; 567 }; 568 569 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 570 571 static struct ext4_err_translation err_translation[] = { 572 EXT4_ERR_TRANSLATE(EIO), 573 EXT4_ERR_TRANSLATE(ENOMEM), 574 EXT4_ERR_TRANSLATE(EFSBADCRC), 575 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 576 EXT4_ERR_TRANSLATE(ENOSPC), 577 EXT4_ERR_TRANSLATE(ENOKEY), 578 EXT4_ERR_TRANSLATE(EROFS), 579 EXT4_ERR_TRANSLATE(EFBIG), 580 EXT4_ERR_TRANSLATE(EEXIST), 581 EXT4_ERR_TRANSLATE(ERANGE), 582 EXT4_ERR_TRANSLATE(EOVERFLOW), 583 EXT4_ERR_TRANSLATE(EBUSY), 584 EXT4_ERR_TRANSLATE(ENOTDIR), 585 EXT4_ERR_TRANSLATE(ENOTEMPTY), 586 EXT4_ERR_TRANSLATE(ESHUTDOWN), 587 EXT4_ERR_TRANSLATE(EFAULT), 588 }; 589 590 static int ext4_errno_to_code(int errno) 591 { 592 int i; 593 594 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 595 if (err_translation[i].errno == errno) 596 return err_translation[i].code; 597 return EXT4_ERR_UNKNOWN; 598 } 599 600 static void save_error_info(struct super_block *sb, int error, 601 __u32 ino, __u64 block, 602 const char *func, unsigned int line) 603 { 604 struct ext4_sb_info *sbi = EXT4_SB(sb); 605 606 /* We default to EFSCORRUPTED error... */ 607 if (error == 0) 608 error = EFSCORRUPTED; 609 610 spin_lock(&sbi->s_error_lock); 611 sbi->s_add_error_count++; 612 sbi->s_last_error_code = error; 613 sbi->s_last_error_line = line; 614 sbi->s_last_error_ino = ino; 615 sbi->s_last_error_block = block; 616 sbi->s_last_error_func = func; 617 sbi->s_last_error_time = ktime_get_real_seconds(); 618 if (!sbi->s_first_error_time) { 619 sbi->s_first_error_code = error; 620 sbi->s_first_error_line = line; 621 sbi->s_first_error_ino = ino; 622 sbi->s_first_error_block = block; 623 sbi->s_first_error_func = func; 624 sbi->s_first_error_time = sbi->s_last_error_time; 625 } 626 spin_unlock(&sbi->s_error_lock); 627 } 628 629 /* Deal with the reporting of failure conditions on a filesystem such as 630 * inconsistencies detected or read IO failures. 631 * 632 * On ext2, we can store the error state of the filesystem in the 633 * superblock. That is not possible on ext4, because we may have other 634 * write ordering constraints on the superblock which prevent us from 635 * writing it out straight away; and given that the journal is about to 636 * be aborted, we can't rely on the current, or future, transactions to 637 * write out the superblock safely. 638 * 639 * We'll just use the jbd2_journal_abort() error code to record an error in 640 * the journal instead. On recovery, the journal will complain about 641 * that error until we've noted it down and cleared it. 642 * 643 * If force_ro is set, we unconditionally force the filesystem into an 644 * ABORT|READONLY state, unless the error response on the fs has been set to 645 * panic in which case we take the easy way out and panic immediately. This is 646 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 647 * at a critical moment in log management. 648 */ 649 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 650 __u32 ino, __u64 block, 651 const char *func, unsigned int line) 652 { 653 journal_t *journal = EXT4_SB(sb)->s_journal; 654 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 655 656 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 657 if (test_opt(sb, WARN_ON_ERROR)) 658 WARN_ON_ONCE(1); 659 660 if (!continue_fs && !sb_rdonly(sb)) { 661 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 662 if (journal) 663 jbd2_journal_abort(journal, -EIO); 664 } 665 666 if (!bdev_read_only(sb->s_bdev)) { 667 save_error_info(sb, error, ino, block, func, line); 668 /* 669 * In case the fs should keep running, we need to writeout 670 * superblock through the journal. Due to lock ordering 671 * constraints, it may not be safe to do it right here so we 672 * defer superblock flushing to a workqueue. 673 */ 674 if (continue_fs && journal) 675 schedule_work(&EXT4_SB(sb)->s_error_work); 676 else 677 ext4_commit_super(sb); 678 } 679 680 /* 681 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 682 * could panic during 'reboot -f' as the underlying device got already 683 * disabled. 684 */ 685 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 686 panic("EXT4-fs (device %s): panic forced after error\n", 687 sb->s_id); 688 } 689 690 if (sb_rdonly(sb) || continue_fs) 691 return; 692 693 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 694 /* 695 * Make sure updated value of ->s_mount_flags will be visible before 696 * ->s_flags update 697 */ 698 smp_wmb(); 699 sb->s_flags |= SB_RDONLY; 700 } 701 702 static void flush_stashed_error_work(struct work_struct *work) 703 { 704 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 705 s_error_work); 706 journal_t *journal = sbi->s_journal; 707 handle_t *handle; 708 709 /* 710 * If the journal is still running, we have to write out superblock 711 * through the journal to avoid collisions of other journalled sb 712 * updates. 713 * 714 * We use directly jbd2 functions here to avoid recursing back into 715 * ext4 error handling code during handling of previous errors. 716 */ 717 if (!sb_rdonly(sbi->s_sb) && journal) { 718 struct buffer_head *sbh = sbi->s_sbh; 719 handle = jbd2_journal_start(journal, 1); 720 if (IS_ERR(handle)) 721 goto write_directly; 722 if (jbd2_journal_get_write_access(handle, sbh)) { 723 jbd2_journal_stop(handle); 724 goto write_directly; 725 } 726 ext4_update_super(sbi->s_sb); 727 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 728 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 729 "superblock detected"); 730 clear_buffer_write_io_error(sbh); 731 set_buffer_uptodate(sbh); 732 } 733 734 if (jbd2_journal_dirty_metadata(handle, sbh)) { 735 jbd2_journal_stop(handle); 736 goto write_directly; 737 } 738 jbd2_journal_stop(handle); 739 ext4_notify_error_sysfs(sbi); 740 return; 741 } 742 write_directly: 743 /* 744 * Write through journal failed. Write sb directly to get error info 745 * out and hope for the best. 746 */ 747 ext4_commit_super(sbi->s_sb); 748 ext4_notify_error_sysfs(sbi); 749 } 750 751 #define ext4_error_ratelimit(sb) \ 752 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 753 "EXT4-fs error") 754 755 void __ext4_error(struct super_block *sb, const char *function, 756 unsigned int line, bool force_ro, int error, __u64 block, 757 const char *fmt, ...) 758 { 759 struct va_format vaf; 760 va_list args; 761 762 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 763 return; 764 765 trace_ext4_error(sb, function, line); 766 if (ext4_error_ratelimit(sb)) { 767 va_start(args, fmt); 768 vaf.fmt = fmt; 769 vaf.va = &args; 770 printk(KERN_CRIT 771 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 772 sb->s_id, function, line, current->comm, &vaf); 773 va_end(args); 774 } 775 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED); 776 777 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 778 } 779 780 void __ext4_error_inode(struct inode *inode, const char *function, 781 unsigned int line, ext4_fsblk_t block, int error, 782 const char *fmt, ...) 783 { 784 va_list args; 785 struct va_format vaf; 786 787 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 788 return; 789 790 trace_ext4_error(inode->i_sb, function, line); 791 if (ext4_error_ratelimit(inode->i_sb)) { 792 va_start(args, fmt); 793 vaf.fmt = fmt; 794 vaf.va = &args; 795 if (block) 796 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 797 "inode #%lu: block %llu: comm %s: %pV\n", 798 inode->i_sb->s_id, function, line, inode->i_ino, 799 block, current->comm, &vaf); 800 else 801 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 802 "inode #%lu: comm %s: %pV\n", 803 inode->i_sb->s_id, function, line, inode->i_ino, 804 current->comm, &vaf); 805 va_end(args); 806 } 807 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED); 808 809 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 810 function, line); 811 } 812 813 void __ext4_error_file(struct file *file, const char *function, 814 unsigned int line, ext4_fsblk_t block, 815 const char *fmt, ...) 816 { 817 va_list args; 818 struct va_format vaf; 819 struct inode *inode = file_inode(file); 820 char pathname[80], *path; 821 822 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 823 return; 824 825 trace_ext4_error(inode->i_sb, function, line); 826 if (ext4_error_ratelimit(inode->i_sb)) { 827 path = file_path(file, pathname, sizeof(pathname)); 828 if (IS_ERR(path)) 829 path = "(unknown)"; 830 va_start(args, fmt); 831 vaf.fmt = fmt; 832 vaf.va = &args; 833 if (block) 834 printk(KERN_CRIT 835 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 836 "block %llu: comm %s: path %s: %pV\n", 837 inode->i_sb->s_id, function, line, inode->i_ino, 838 block, current->comm, path, &vaf); 839 else 840 printk(KERN_CRIT 841 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 842 "comm %s: path %s: %pV\n", 843 inode->i_sb->s_id, function, line, inode->i_ino, 844 current->comm, path, &vaf); 845 va_end(args); 846 } 847 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED); 848 849 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 850 function, line); 851 } 852 853 const char *ext4_decode_error(struct super_block *sb, int errno, 854 char nbuf[16]) 855 { 856 char *errstr = NULL; 857 858 switch (errno) { 859 case -EFSCORRUPTED: 860 errstr = "Corrupt filesystem"; 861 break; 862 case -EFSBADCRC: 863 errstr = "Filesystem failed CRC"; 864 break; 865 case -EIO: 866 errstr = "IO failure"; 867 break; 868 case -ENOMEM: 869 errstr = "Out of memory"; 870 break; 871 case -EROFS: 872 if (!sb || (EXT4_SB(sb)->s_journal && 873 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 874 errstr = "Journal has aborted"; 875 else 876 errstr = "Readonly filesystem"; 877 break; 878 default: 879 /* If the caller passed in an extra buffer for unknown 880 * errors, textualise them now. Else we just return 881 * NULL. */ 882 if (nbuf) { 883 /* Check for truncated error codes... */ 884 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 885 errstr = nbuf; 886 } 887 break; 888 } 889 890 return errstr; 891 } 892 893 /* __ext4_std_error decodes expected errors from journaling functions 894 * automatically and invokes the appropriate error response. */ 895 896 void __ext4_std_error(struct super_block *sb, const char *function, 897 unsigned int line, int errno) 898 { 899 char nbuf[16]; 900 const char *errstr; 901 902 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 903 return; 904 905 /* Special case: if the error is EROFS, and we're not already 906 * inside a transaction, then there's really no point in logging 907 * an error. */ 908 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 909 return; 910 911 if (ext4_error_ratelimit(sb)) { 912 errstr = ext4_decode_error(sb, errno, nbuf); 913 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 914 sb->s_id, function, line, errstr); 915 } 916 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED); 917 918 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 919 } 920 921 void __ext4_msg(struct super_block *sb, 922 const char *prefix, const char *fmt, ...) 923 { 924 struct va_format vaf; 925 va_list args; 926 927 if (sb) { 928 atomic_inc(&EXT4_SB(sb)->s_msg_count); 929 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 930 "EXT4-fs")) 931 return; 932 } 933 934 va_start(args, fmt); 935 vaf.fmt = fmt; 936 vaf.va = &args; 937 if (sb) 938 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 939 else 940 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 941 va_end(args); 942 } 943 944 static int ext4_warning_ratelimit(struct super_block *sb) 945 { 946 atomic_inc(&EXT4_SB(sb)->s_warning_count); 947 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 948 "EXT4-fs warning"); 949 } 950 951 void __ext4_warning(struct super_block *sb, const char *function, 952 unsigned int line, const char *fmt, ...) 953 { 954 struct va_format vaf; 955 va_list args; 956 957 if (!ext4_warning_ratelimit(sb)) 958 return; 959 960 va_start(args, fmt); 961 vaf.fmt = fmt; 962 vaf.va = &args; 963 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 964 sb->s_id, function, line, &vaf); 965 va_end(args); 966 } 967 968 void __ext4_warning_inode(const struct inode *inode, const char *function, 969 unsigned int line, const char *fmt, ...) 970 { 971 struct va_format vaf; 972 va_list args; 973 974 if (!ext4_warning_ratelimit(inode->i_sb)) 975 return; 976 977 va_start(args, fmt); 978 vaf.fmt = fmt; 979 vaf.va = &args; 980 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 981 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 982 function, line, inode->i_ino, current->comm, &vaf); 983 va_end(args); 984 } 985 986 void __ext4_grp_locked_error(const char *function, unsigned int line, 987 struct super_block *sb, ext4_group_t grp, 988 unsigned long ino, ext4_fsblk_t block, 989 const char *fmt, ...) 990 __releases(bitlock) 991 __acquires(bitlock) 992 { 993 struct va_format vaf; 994 va_list args; 995 996 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 997 return; 998 999 trace_ext4_error(sb, function, line); 1000 if (ext4_error_ratelimit(sb)) { 1001 va_start(args, fmt); 1002 vaf.fmt = fmt; 1003 vaf.va = &args; 1004 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1005 sb->s_id, function, line, grp); 1006 if (ino) 1007 printk(KERN_CONT "inode %lu: ", ino); 1008 if (block) 1009 printk(KERN_CONT "block %llu:", 1010 (unsigned long long) block); 1011 printk(KERN_CONT "%pV\n", &vaf); 1012 va_end(args); 1013 } 1014 1015 if (test_opt(sb, ERRORS_CONT)) { 1016 if (test_opt(sb, WARN_ON_ERROR)) 1017 WARN_ON_ONCE(1); 1018 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1019 if (!bdev_read_only(sb->s_bdev)) { 1020 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1021 line); 1022 schedule_work(&EXT4_SB(sb)->s_error_work); 1023 } 1024 return; 1025 } 1026 ext4_unlock_group(sb, grp); 1027 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1028 /* 1029 * We only get here in the ERRORS_RO case; relocking the group 1030 * may be dangerous, but nothing bad will happen since the 1031 * filesystem will have already been marked read/only and the 1032 * journal has been aborted. We return 1 as a hint to callers 1033 * who might what to use the return value from 1034 * ext4_grp_locked_error() to distinguish between the 1035 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1036 * aggressively from the ext4 function in question, with a 1037 * more appropriate error code. 1038 */ 1039 ext4_lock_group(sb, grp); 1040 return; 1041 } 1042 1043 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1044 ext4_group_t group, 1045 unsigned int flags) 1046 { 1047 struct ext4_sb_info *sbi = EXT4_SB(sb); 1048 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1049 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1050 int ret; 1051 1052 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1053 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1054 &grp->bb_state); 1055 if (!ret) 1056 percpu_counter_sub(&sbi->s_freeclusters_counter, 1057 grp->bb_free); 1058 } 1059 1060 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1061 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1062 &grp->bb_state); 1063 if (!ret && gdp) { 1064 int count; 1065 1066 count = ext4_free_inodes_count(sb, gdp); 1067 percpu_counter_sub(&sbi->s_freeinodes_counter, 1068 count); 1069 } 1070 } 1071 } 1072 1073 void ext4_update_dynamic_rev(struct super_block *sb) 1074 { 1075 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1076 1077 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1078 return; 1079 1080 ext4_warning(sb, 1081 "updating to rev %d because of new feature flag, " 1082 "running e2fsck is recommended", 1083 EXT4_DYNAMIC_REV); 1084 1085 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1086 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1087 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1088 /* leave es->s_feature_*compat flags alone */ 1089 /* es->s_uuid will be set by e2fsck if empty */ 1090 1091 /* 1092 * The rest of the superblock fields should be zero, and if not it 1093 * means they are likely already in use, so leave them alone. We 1094 * can leave it up to e2fsck to clean up any inconsistencies there. 1095 */ 1096 } 1097 1098 /* 1099 * Open the external journal device 1100 */ 1101 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 1102 { 1103 struct block_device *bdev; 1104 1105 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 1106 if (IS_ERR(bdev)) 1107 goto fail; 1108 return bdev; 1109 1110 fail: 1111 ext4_msg(sb, KERN_ERR, 1112 "failed to open journal device unknown-block(%u,%u) %ld", 1113 MAJOR(dev), MINOR(dev), PTR_ERR(bdev)); 1114 return NULL; 1115 } 1116 1117 /* 1118 * Release the journal device 1119 */ 1120 static void ext4_blkdev_put(struct block_device *bdev) 1121 { 1122 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1123 } 1124 1125 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 1126 { 1127 struct block_device *bdev; 1128 bdev = sbi->s_journal_bdev; 1129 if (bdev) { 1130 ext4_blkdev_put(bdev); 1131 sbi->s_journal_bdev = NULL; 1132 } 1133 } 1134 1135 static inline struct inode *orphan_list_entry(struct list_head *l) 1136 { 1137 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1138 } 1139 1140 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1141 { 1142 struct list_head *l; 1143 1144 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1145 le32_to_cpu(sbi->s_es->s_last_orphan)); 1146 1147 printk(KERN_ERR "sb_info orphan list:\n"); 1148 list_for_each(l, &sbi->s_orphan) { 1149 struct inode *inode = orphan_list_entry(l); 1150 printk(KERN_ERR " " 1151 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1152 inode->i_sb->s_id, inode->i_ino, inode, 1153 inode->i_mode, inode->i_nlink, 1154 NEXT_ORPHAN(inode)); 1155 } 1156 } 1157 1158 #ifdef CONFIG_QUOTA 1159 static int ext4_quota_off(struct super_block *sb, int type); 1160 1161 static inline void ext4_quota_off_umount(struct super_block *sb) 1162 { 1163 int type; 1164 1165 /* Use our quota_off function to clear inode flags etc. */ 1166 for (type = 0; type < EXT4_MAXQUOTAS; type++) 1167 ext4_quota_off(sb, type); 1168 } 1169 1170 /* 1171 * This is a helper function which is used in the mount/remount 1172 * codepaths (which holds s_umount) to fetch the quota file name. 1173 */ 1174 static inline char *get_qf_name(struct super_block *sb, 1175 struct ext4_sb_info *sbi, 1176 int type) 1177 { 1178 return rcu_dereference_protected(sbi->s_qf_names[type], 1179 lockdep_is_held(&sb->s_umount)); 1180 } 1181 #else 1182 static inline void ext4_quota_off_umount(struct super_block *sb) 1183 { 1184 } 1185 #endif 1186 1187 static void ext4_put_super(struct super_block *sb) 1188 { 1189 struct ext4_sb_info *sbi = EXT4_SB(sb); 1190 struct ext4_super_block *es = sbi->s_es; 1191 struct buffer_head **group_desc; 1192 struct flex_groups **flex_groups; 1193 int aborted = 0; 1194 int i, err; 1195 1196 /* 1197 * Unregister sysfs before destroying jbd2 journal. 1198 * Since we could still access attr_journal_task attribute via sysfs 1199 * path which could have sbi->s_journal->j_task as NULL 1200 * Unregister sysfs before flush sbi->s_error_work. 1201 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1202 * read metadata verify failed then will queue error work. 1203 * flush_stashed_error_work will call start_this_handle may trigger 1204 * BUG_ON. 1205 */ 1206 ext4_unregister_sysfs(sb); 1207 1208 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1209 ext4_msg(sb, KERN_INFO, "unmounting filesystem."); 1210 1211 ext4_unregister_li_request(sb); 1212 ext4_quota_off_umount(sb); 1213 1214 flush_work(&sbi->s_error_work); 1215 destroy_workqueue(sbi->rsv_conversion_wq); 1216 ext4_release_orphan_info(sb); 1217 1218 if (sbi->s_journal) { 1219 aborted = is_journal_aborted(sbi->s_journal); 1220 err = jbd2_journal_destroy(sbi->s_journal); 1221 sbi->s_journal = NULL; 1222 if ((err < 0) && !aborted) { 1223 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1224 } 1225 } 1226 1227 ext4_es_unregister_shrinker(sbi); 1228 del_timer_sync(&sbi->s_err_report); 1229 ext4_release_system_zone(sb); 1230 ext4_mb_release(sb); 1231 ext4_ext_release(sb); 1232 1233 if (!sb_rdonly(sb) && !aborted) { 1234 ext4_clear_feature_journal_needs_recovery(sb); 1235 ext4_clear_feature_orphan_present(sb); 1236 es->s_state = cpu_to_le16(sbi->s_mount_state); 1237 } 1238 if (!sb_rdonly(sb)) 1239 ext4_commit_super(sb); 1240 1241 rcu_read_lock(); 1242 group_desc = rcu_dereference(sbi->s_group_desc); 1243 for (i = 0; i < sbi->s_gdb_count; i++) 1244 brelse(group_desc[i]); 1245 kvfree(group_desc); 1246 flex_groups = rcu_dereference(sbi->s_flex_groups); 1247 if (flex_groups) { 1248 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1249 kvfree(flex_groups[i]); 1250 kvfree(flex_groups); 1251 } 1252 rcu_read_unlock(); 1253 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1254 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1255 percpu_counter_destroy(&sbi->s_dirs_counter); 1256 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1257 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1258 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1259 #ifdef CONFIG_QUOTA 1260 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1261 kfree(get_qf_name(sb, sbi, i)); 1262 #endif 1263 1264 /* Debugging code just in case the in-memory inode orphan list 1265 * isn't empty. The on-disk one can be non-empty if we've 1266 * detected an error and taken the fs readonly, but the 1267 * in-memory list had better be clean by this point. */ 1268 if (!list_empty(&sbi->s_orphan)) 1269 dump_orphan_list(sb, sbi); 1270 ASSERT(list_empty(&sbi->s_orphan)); 1271 1272 sync_blockdev(sb->s_bdev); 1273 invalidate_bdev(sb->s_bdev); 1274 if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) { 1275 /* 1276 * Invalidate the journal device's buffers. We don't want them 1277 * floating about in memory - the physical journal device may 1278 * hotswapped, and it breaks the `ro-after' testing code. 1279 */ 1280 sync_blockdev(sbi->s_journal_bdev); 1281 invalidate_bdev(sbi->s_journal_bdev); 1282 ext4_blkdev_remove(sbi); 1283 } 1284 1285 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1286 sbi->s_ea_inode_cache = NULL; 1287 1288 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1289 sbi->s_ea_block_cache = NULL; 1290 1291 ext4_stop_mmpd(sbi); 1292 1293 brelse(sbi->s_sbh); 1294 sb->s_fs_info = NULL; 1295 /* 1296 * Now that we are completely done shutting down the 1297 * superblock, we need to actually destroy the kobject. 1298 */ 1299 kobject_put(&sbi->s_kobj); 1300 wait_for_completion(&sbi->s_kobj_unregister); 1301 if (sbi->s_chksum_driver) 1302 crypto_free_shash(sbi->s_chksum_driver); 1303 kfree(sbi->s_blockgroup_lock); 1304 fs_put_dax(sbi->s_daxdev, NULL); 1305 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1306 #if IS_ENABLED(CONFIG_UNICODE) 1307 utf8_unload(sb->s_encoding); 1308 #endif 1309 kfree(sbi); 1310 } 1311 1312 static struct kmem_cache *ext4_inode_cachep; 1313 1314 /* 1315 * Called inside transaction, so use GFP_NOFS 1316 */ 1317 static struct inode *ext4_alloc_inode(struct super_block *sb) 1318 { 1319 struct ext4_inode_info *ei; 1320 1321 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1322 if (!ei) 1323 return NULL; 1324 1325 inode_set_iversion(&ei->vfs_inode, 1); 1326 spin_lock_init(&ei->i_raw_lock); 1327 INIT_LIST_HEAD(&ei->i_prealloc_list); 1328 atomic_set(&ei->i_prealloc_active, 0); 1329 spin_lock_init(&ei->i_prealloc_lock); 1330 ext4_es_init_tree(&ei->i_es_tree); 1331 rwlock_init(&ei->i_es_lock); 1332 INIT_LIST_HEAD(&ei->i_es_list); 1333 ei->i_es_all_nr = 0; 1334 ei->i_es_shk_nr = 0; 1335 ei->i_es_shrink_lblk = 0; 1336 ei->i_reserved_data_blocks = 0; 1337 spin_lock_init(&(ei->i_block_reservation_lock)); 1338 ext4_init_pending_tree(&ei->i_pending_tree); 1339 #ifdef CONFIG_QUOTA 1340 ei->i_reserved_quota = 0; 1341 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1342 #endif 1343 ei->jinode = NULL; 1344 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1345 spin_lock_init(&ei->i_completed_io_lock); 1346 ei->i_sync_tid = 0; 1347 ei->i_datasync_tid = 0; 1348 atomic_set(&ei->i_unwritten, 0); 1349 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1350 ext4_fc_init_inode(&ei->vfs_inode); 1351 mutex_init(&ei->i_fc_lock); 1352 return &ei->vfs_inode; 1353 } 1354 1355 static int ext4_drop_inode(struct inode *inode) 1356 { 1357 int drop = generic_drop_inode(inode); 1358 1359 if (!drop) 1360 drop = fscrypt_drop_inode(inode); 1361 1362 trace_ext4_drop_inode(inode, drop); 1363 return drop; 1364 } 1365 1366 static void ext4_free_in_core_inode(struct inode *inode) 1367 { 1368 fscrypt_free_inode(inode); 1369 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1370 pr_warn("%s: inode %ld still in fc list", 1371 __func__, inode->i_ino); 1372 } 1373 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1374 } 1375 1376 static void ext4_destroy_inode(struct inode *inode) 1377 { 1378 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1379 ext4_msg(inode->i_sb, KERN_ERR, 1380 "Inode %lu (%p): orphan list check failed!", 1381 inode->i_ino, EXT4_I(inode)); 1382 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1383 EXT4_I(inode), sizeof(struct ext4_inode_info), 1384 true); 1385 dump_stack(); 1386 } 1387 1388 if (EXT4_I(inode)->i_reserved_data_blocks) 1389 ext4_msg(inode->i_sb, KERN_ERR, 1390 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!", 1391 inode->i_ino, EXT4_I(inode), 1392 EXT4_I(inode)->i_reserved_data_blocks); 1393 } 1394 1395 static void init_once(void *foo) 1396 { 1397 struct ext4_inode_info *ei = foo; 1398 1399 INIT_LIST_HEAD(&ei->i_orphan); 1400 init_rwsem(&ei->xattr_sem); 1401 init_rwsem(&ei->i_data_sem); 1402 inode_init_once(&ei->vfs_inode); 1403 ext4_fc_init_inode(&ei->vfs_inode); 1404 } 1405 1406 static int __init init_inodecache(void) 1407 { 1408 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1409 sizeof(struct ext4_inode_info), 0, 1410 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1411 SLAB_ACCOUNT), 1412 offsetof(struct ext4_inode_info, i_data), 1413 sizeof_field(struct ext4_inode_info, i_data), 1414 init_once); 1415 if (ext4_inode_cachep == NULL) 1416 return -ENOMEM; 1417 return 0; 1418 } 1419 1420 static void destroy_inodecache(void) 1421 { 1422 /* 1423 * Make sure all delayed rcu free inodes are flushed before we 1424 * destroy cache. 1425 */ 1426 rcu_barrier(); 1427 kmem_cache_destroy(ext4_inode_cachep); 1428 } 1429 1430 void ext4_clear_inode(struct inode *inode) 1431 { 1432 ext4_fc_del(inode); 1433 invalidate_inode_buffers(inode); 1434 clear_inode(inode); 1435 ext4_discard_preallocations(inode, 0); 1436 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1437 dquot_drop(inode); 1438 if (EXT4_I(inode)->jinode) { 1439 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1440 EXT4_I(inode)->jinode); 1441 jbd2_free_inode(EXT4_I(inode)->jinode); 1442 EXT4_I(inode)->jinode = NULL; 1443 } 1444 fscrypt_put_encryption_info(inode); 1445 fsverity_cleanup_inode(inode); 1446 } 1447 1448 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1449 u64 ino, u32 generation) 1450 { 1451 struct inode *inode; 1452 1453 /* 1454 * Currently we don't know the generation for parent directory, so 1455 * a generation of 0 means "accept any" 1456 */ 1457 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1458 if (IS_ERR(inode)) 1459 return ERR_CAST(inode); 1460 if (generation && inode->i_generation != generation) { 1461 iput(inode); 1462 return ERR_PTR(-ESTALE); 1463 } 1464 1465 return inode; 1466 } 1467 1468 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1469 int fh_len, int fh_type) 1470 { 1471 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1472 ext4_nfs_get_inode); 1473 } 1474 1475 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1476 int fh_len, int fh_type) 1477 { 1478 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1479 ext4_nfs_get_inode); 1480 } 1481 1482 static int ext4_nfs_commit_metadata(struct inode *inode) 1483 { 1484 struct writeback_control wbc = { 1485 .sync_mode = WB_SYNC_ALL 1486 }; 1487 1488 trace_ext4_nfs_commit_metadata(inode); 1489 return ext4_write_inode(inode, &wbc); 1490 } 1491 1492 #ifdef CONFIG_QUOTA 1493 static const char * const quotatypes[] = INITQFNAMES; 1494 #define QTYPE2NAME(t) (quotatypes[t]) 1495 1496 static int ext4_write_dquot(struct dquot *dquot); 1497 static int ext4_acquire_dquot(struct dquot *dquot); 1498 static int ext4_release_dquot(struct dquot *dquot); 1499 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1500 static int ext4_write_info(struct super_block *sb, int type); 1501 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1502 const struct path *path); 1503 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1504 size_t len, loff_t off); 1505 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1506 const char *data, size_t len, loff_t off); 1507 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1508 unsigned int flags); 1509 1510 static struct dquot **ext4_get_dquots(struct inode *inode) 1511 { 1512 return EXT4_I(inode)->i_dquot; 1513 } 1514 1515 static const struct dquot_operations ext4_quota_operations = { 1516 .get_reserved_space = ext4_get_reserved_space, 1517 .write_dquot = ext4_write_dquot, 1518 .acquire_dquot = ext4_acquire_dquot, 1519 .release_dquot = ext4_release_dquot, 1520 .mark_dirty = ext4_mark_dquot_dirty, 1521 .write_info = ext4_write_info, 1522 .alloc_dquot = dquot_alloc, 1523 .destroy_dquot = dquot_destroy, 1524 .get_projid = ext4_get_projid, 1525 .get_inode_usage = ext4_get_inode_usage, 1526 .get_next_id = dquot_get_next_id, 1527 }; 1528 1529 static const struct quotactl_ops ext4_qctl_operations = { 1530 .quota_on = ext4_quota_on, 1531 .quota_off = ext4_quota_off, 1532 .quota_sync = dquot_quota_sync, 1533 .get_state = dquot_get_state, 1534 .set_info = dquot_set_dqinfo, 1535 .get_dqblk = dquot_get_dqblk, 1536 .set_dqblk = dquot_set_dqblk, 1537 .get_nextdqblk = dquot_get_next_dqblk, 1538 }; 1539 #endif 1540 1541 static const struct super_operations ext4_sops = { 1542 .alloc_inode = ext4_alloc_inode, 1543 .free_inode = ext4_free_in_core_inode, 1544 .destroy_inode = ext4_destroy_inode, 1545 .write_inode = ext4_write_inode, 1546 .dirty_inode = ext4_dirty_inode, 1547 .drop_inode = ext4_drop_inode, 1548 .evict_inode = ext4_evict_inode, 1549 .put_super = ext4_put_super, 1550 .sync_fs = ext4_sync_fs, 1551 .freeze_fs = ext4_freeze, 1552 .unfreeze_fs = ext4_unfreeze, 1553 .statfs = ext4_statfs, 1554 .show_options = ext4_show_options, 1555 #ifdef CONFIG_QUOTA 1556 .quota_read = ext4_quota_read, 1557 .quota_write = ext4_quota_write, 1558 .get_dquots = ext4_get_dquots, 1559 #endif 1560 }; 1561 1562 static const struct export_operations ext4_export_ops = { 1563 .fh_to_dentry = ext4_fh_to_dentry, 1564 .fh_to_parent = ext4_fh_to_parent, 1565 .get_parent = ext4_get_parent, 1566 .commit_metadata = ext4_nfs_commit_metadata, 1567 }; 1568 1569 enum { 1570 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1571 Opt_resgid, Opt_resuid, Opt_sb, 1572 Opt_nouid32, Opt_debug, Opt_removed, 1573 Opt_user_xattr, Opt_acl, 1574 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1575 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1576 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1577 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1578 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1579 Opt_inlinecrypt, 1580 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1581 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1582 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1583 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1584 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1585 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1586 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1587 Opt_inode_readahead_blks, Opt_journal_ioprio, 1588 Opt_dioread_nolock, Opt_dioread_lock, 1589 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1590 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1591 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1592 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1593 #ifdef CONFIG_EXT4_DEBUG 1594 Opt_fc_debug_max_replay, Opt_fc_debug_force 1595 #endif 1596 }; 1597 1598 static const struct constant_table ext4_param_errors[] = { 1599 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1600 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1601 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1602 {} 1603 }; 1604 1605 static const struct constant_table ext4_param_data[] = { 1606 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1607 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1608 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1609 {} 1610 }; 1611 1612 static const struct constant_table ext4_param_data_err[] = { 1613 {"abort", Opt_data_err_abort}, 1614 {"ignore", Opt_data_err_ignore}, 1615 {} 1616 }; 1617 1618 static const struct constant_table ext4_param_jqfmt[] = { 1619 {"vfsold", QFMT_VFS_OLD}, 1620 {"vfsv0", QFMT_VFS_V0}, 1621 {"vfsv1", QFMT_VFS_V1}, 1622 {} 1623 }; 1624 1625 static const struct constant_table ext4_param_dax[] = { 1626 {"always", Opt_dax_always}, 1627 {"inode", Opt_dax_inode}, 1628 {"never", Opt_dax_never}, 1629 {} 1630 }; 1631 1632 /* String parameter that allows empty argument */ 1633 #define fsparam_string_empty(NAME, OPT) \ 1634 __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL) 1635 1636 /* 1637 * Mount option specification 1638 * We don't use fsparam_flag_no because of the way we set the 1639 * options and the way we show them in _ext4_show_options(). To 1640 * keep the changes to a minimum, let's keep the negative options 1641 * separate for now. 1642 */ 1643 static const struct fs_parameter_spec ext4_param_specs[] = { 1644 fsparam_flag ("bsddf", Opt_bsd_df), 1645 fsparam_flag ("minixdf", Opt_minix_df), 1646 fsparam_flag ("grpid", Opt_grpid), 1647 fsparam_flag ("bsdgroups", Opt_grpid), 1648 fsparam_flag ("nogrpid", Opt_nogrpid), 1649 fsparam_flag ("sysvgroups", Opt_nogrpid), 1650 fsparam_u32 ("resgid", Opt_resgid), 1651 fsparam_u32 ("resuid", Opt_resuid), 1652 fsparam_u32 ("sb", Opt_sb), 1653 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1654 fsparam_flag ("nouid32", Opt_nouid32), 1655 fsparam_flag ("debug", Opt_debug), 1656 fsparam_flag ("oldalloc", Opt_removed), 1657 fsparam_flag ("orlov", Opt_removed), 1658 fsparam_flag ("user_xattr", Opt_user_xattr), 1659 fsparam_flag ("acl", Opt_acl), 1660 fsparam_flag ("norecovery", Opt_noload), 1661 fsparam_flag ("noload", Opt_noload), 1662 fsparam_flag ("bh", Opt_removed), 1663 fsparam_flag ("nobh", Opt_removed), 1664 fsparam_u32 ("commit", Opt_commit), 1665 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1666 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1667 fsparam_u32 ("journal_dev", Opt_journal_dev), 1668 fsparam_bdev ("journal_path", Opt_journal_path), 1669 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1670 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1671 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1672 fsparam_flag ("abort", Opt_abort), 1673 fsparam_enum ("data", Opt_data, ext4_param_data), 1674 fsparam_enum ("data_err", Opt_data_err, 1675 ext4_param_data_err), 1676 fsparam_string_empty 1677 ("usrjquota", Opt_usrjquota), 1678 fsparam_string_empty 1679 ("grpjquota", Opt_grpjquota), 1680 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1681 fsparam_flag ("grpquota", Opt_grpquota), 1682 fsparam_flag ("quota", Opt_quota), 1683 fsparam_flag ("noquota", Opt_noquota), 1684 fsparam_flag ("usrquota", Opt_usrquota), 1685 fsparam_flag ("prjquota", Opt_prjquota), 1686 fsparam_flag ("barrier", Opt_barrier), 1687 fsparam_u32 ("barrier", Opt_barrier), 1688 fsparam_flag ("nobarrier", Opt_nobarrier), 1689 fsparam_flag ("i_version", Opt_removed), 1690 fsparam_flag ("dax", Opt_dax), 1691 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1692 fsparam_u32 ("stripe", Opt_stripe), 1693 fsparam_flag ("delalloc", Opt_delalloc), 1694 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1695 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1696 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1697 fsparam_u32 ("debug_want_extra_isize", 1698 Opt_debug_want_extra_isize), 1699 fsparam_flag ("mblk_io_submit", Opt_removed), 1700 fsparam_flag ("nomblk_io_submit", Opt_removed), 1701 fsparam_flag ("block_validity", Opt_block_validity), 1702 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1703 fsparam_u32 ("inode_readahead_blks", 1704 Opt_inode_readahead_blks), 1705 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1706 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1707 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1708 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1709 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1710 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1711 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1712 fsparam_flag ("discard", Opt_discard), 1713 fsparam_flag ("nodiscard", Opt_nodiscard), 1714 fsparam_u32 ("init_itable", Opt_init_itable), 1715 fsparam_flag ("init_itable", Opt_init_itable), 1716 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1717 #ifdef CONFIG_EXT4_DEBUG 1718 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1719 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1720 #endif 1721 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1722 fsparam_flag ("test_dummy_encryption", 1723 Opt_test_dummy_encryption), 1724 fsparam_string ("test_dummy_encryption", 1725 Opt_test_dummy_encryption), 1726 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1727 fsparam_flag ("nombcache", Opt_nombcache), 1728 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1729 fsparam_flag ("prefetch_block_bitmaps", 1730 Opt_removed), 1731 fsparam_flag ("no_prefetch_block_bitmaps", 1732 Opt_no_prefetch_block_bitmaps), 1733 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1734 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1735 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1736 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1737 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1738 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1739 {} 1740 }; 1741 1742 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1743 1744 static const char deprecated_msg[] = 1745 "Mount option \"%s\" will be removed by %s\n" 1746 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1747 1748 #define MOPT_SET 0x0001 1749 #define MOPT_CLEAR 0x0002 1750 #define MOPT_NOSUPPORT 0x0004 1751 #define MOPT_EXPLICIT 0x0008 1752 #ifdef CONFIG_QUOTA 1753 #define MOPT_Q 0 1754 #define MOPT_QFMT 0x0010 1755 #else 1756 #define MOPT_Q MOPT_NOSUPPORT 1757 #define MOPT_QFMT MOPT_NOSUPPORT 1758 #endif 1759 #define MOPT_NO_EXT2 0x0020 1760 #define MOPT_NO_EXT3 0x0040 1761 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1762 #define MOPT_SKIP 0x0080 1763 #define MOPT_2 0x0100 1764 1765 static const struct mount_opts { 1766 int token; 1767 int mount_opt; 1768 int flags; 1769 } ext4_mount_opts[] = { 1770 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1771 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1772 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1773 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1774 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1775 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1776 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1777 MOPT_EXT4_ONLY | MOPT_SET}, 1778 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1779 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1780 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1781 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1782 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1783 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1784 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1785 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1786 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1787 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1788 {Opt_commit, 0, MOPT_NO_EXT2}, 1789 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1790 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1791 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1792 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1793 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1794 EXT4_MOUNT_JOURNAL_CHECKSUM), 1795 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1796 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1797 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1798 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1799 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1800 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1801 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1802 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1803 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1804 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1805 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1806 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1807 {Opt_data, 0, MOPT_NO_EXT2}, 1808 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1809 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1810 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1811 #else 1812 {Opt_acl, 0, MOPT_NOSUPPORT}, 1813 #endif 1814 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1815 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1816 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1817 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1818 MOPT_SET | MOPT_Q}, 1819 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1820 MOPT_SET | MOPT_Q}, 1821 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1822 MOPT_SET | MOPT_Q}, 1823 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1824 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1825 MOPT_CLEAR | MOPT_Q}, 1826 {Opt_usrjquota, 0, MOPT_Q}, 1827 {Opt_grpjquota, 0, MOPT_Q}, 1828 {Opt_jqfmt, 0, MOPT_QFMT}, 1829 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1830 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1831 MOPT_SET}, 1832 #ifdef CONFIG_EXT4_DEBUG 1833 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1834 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1835 #endif 1836 {Opt_err, 0, 0} 1837 }; 1838 1839 #if IS_ENABLED(CONFIG_UNICODE) 1840 static const struct ext4_sb_encodings { 1841 __u16 magic; 1842 char *name; 1843 unsigned int version; 1844 } ext4_sb_encoding_map[] = { 1845 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1846 }; 1847 1848 static const struct ext4_sb_encodings * 1849 ext4_sb_read_encoding(const struct ext4_super_block *es) 1850 { 1851 __u16 magic = le16_to_cpu(es->s_encoding); 1852 int i; 1853 1854 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1855 if (magic == ext4_sb_encoding_map[i].magic) 1856 return &ext4_sb_encoding_map[i]; 1857 1858 return NULL; 1859 } 1860 #endif 1861 1862 #define EXT4_SPEC_JQUOTA (1 << 0) 1863 #define EXT4_SPEC_JQFMT (1 << 1) 1864 #define EXT4_SPEC_DATAJ (1 << 2) 1865 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1866 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1867 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1868 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 1869 #define EXT4_SPEC_s_max_batch_time (1 << 8) 1870 #define EXT4_SPEC_s_min_batch_time (1 << 9) 1871 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 1872 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 1873 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 1874 #define EXT4_SPEC_s_stripe (1 << 13) 1875 #define EXT4_SPEC_s_resuid (1 << 14) 1876 #define EXT4_SPEC_s_resgid (1 << 15) 1877 #define EXT4_SPEC_s_commit_interval (1 << 16) 1878 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 1879 #define EXT4_SPEC_s_sb_block (1 << 18) 1880 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 1881 1882 struct ext4_fs_context { 1883 char *s_qf_names[EXT4_MAXQUOTAS]; 1884 struct fscrypt_dummy_policy dummy_enc_policy; 1885 int s_jquota_fmt; /* Format of quota to use */ 1886 #ifdef CONFIG_EXT4_DEBUG 1887 int s_fc_debug_max_replay; 1888 #endif 1889 unsigned short qname_spec; 1890 unsigned long vals_s_flags; /* Bits to set in s_flags */ 1891 unsigned long mask_s_flags; /* Bits changed in s_flags */ 1892 unsigned long journal_devnum; 1893 unsigned long s_commit_interval; 1894 unsigned long s_stripe; 1895 unsigned int s_inode_readahead_blks; 1896 unsigned int s_want_extra_isize; 1897 unsigned int s_li_wait_mult; 1898 unsigned int s_max_dir_size_kb; 1899 unsigned int journal_ioprio; 1900 unsigned int vals_s_mount_opt; 1901 unsigned int mask_s_mount_opt; 1902 unsigned int vals_s_mount_opt2; 1903 unsigned int mask_s_mount_opt2; 1904 unsigned long vals_s_mount_flags; 1905 unsigned long mask_s_mount_flags; 1906 unsigned int opt_flags; /* MOPT flags */ 1907 unsigned int spec; 1908 u32 s_max_batch_time; 1909 u32 s_min_batch_time; 1910 kuid_t s_resuid; 1911 kgid_t s_resgid; 1912 ext4_fsblk_t s_sb_block; 1913 }; 1914 1915 static void ext4_fc_free(struct fs_context *fc) 1916 { 1917 struct ext4_fs_context *ctx = fc->fs_private; 1918 int i; 1919 1920 if (!ctx) 1921 return; 1922 1923 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1924 kfree(ctx->s_qf_names[i]); 1925 1926 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 1927 kfree(ctx); 1928 } 1929 1930 int ext4_init_fs_context(struct fs_context *fc) 1931 { 1932 struct ext4_fs_context *ctx; 1933 1934 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 1935 if (!ctx) 1936 return -ENOMEM; 1937 1938 fc->fs_private = ctx; 1939 fc->ops = &ext4_context_ops; 1940 1941 return 0; 1942 } 1943 1944 #ifdef CONFIG_QUOTA 1945 /* 1946 * Note the name of the specified quota file. 1947 */ 1948 static int note_qf_name(struct fs_context *fc, int qtype, 1949 struct fs_parameter *param) 1950 { 1951 struct ext4_fs_context *ctx = fc->fs_private; 1952 char *qname; 1953 1954 if (param->size < 1) { 1955 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 1956 return -EINVAL; 1957 } 1958 if (strchr(param->string, '/')) { 1959 ext4_msg(NULL, KERN_ERR, 1960 "quotafile must be on filesystem root"); 1961 return -EINVAL; 1962 } 1963 if (ctx->s_qf_names[qtype]) { 1964 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 1965 ext4_msg(NULL, KERN_ERR, 1966 "%s quota file already specified", 1967 QTYPE2NAME(qtype)); 1968 return -EINVAL; 1969 } 1970 return 0; 1971 } 1972 1973 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 1974 if (!qname) { 1975 ext4_msg(NULL, KERN_ERR, 1976 "Not enough memory for storing quotafile name"); 1977 return -ENOMEM; 1978 } 1979 ctx->s_qf_names[qtype] = qname; 1980 ctx->qname_spec |= 1 << qtype; 1981 ctx->spec |= EXT4_SPEC_JQUOTA; 1982 return 0; 1983 } 1984 1985 /* 1986 * Clear the name of the specified quota file. 1987 */ 1988 static int unnote_qf_name(struct fs_context *fc, int qtype) 1989 { 1990 struct ext4_fs_context *ctx = fc->fs_private; 1991 1992 if (ctx->s_qf_names[qtype]) 1993 kfree(ctx->s_qf_names[qtype]); 1994 1995 ctx->s_qf_names[qtype] = NULL; 1996 ctx->qname_spec |= 1 << qtype; 1997 ctx->spec |= EXT4_SPEC_JQUOTA; 1998 return 0; 1999 } 2000 #endif 2001 2002 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 2003 struct ext4_fs_context *ctx) 2004 { 2005 int err; 2006 2007 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2008 ext4_msg(NULL, KERN_WARNING, 2009 "test_dummy_encryption option not supported"); 2010 return -EINVAL; 2011 } 2012 err = fscrypt_parse_test_dummy_encryption(param, 2013 &ctx->dummy_enc_policy); 2014 if (err == -EINVAL) { 2015 ext4_msg(NULL, KERN_WARNING, 2016 "Value of option \"%s\" is unrecognized", param->key); 2017 } else if (err == -EEXIST) { 2018 ext4_msg(NULL, KERN_WARNING, 2019 "Conflicting test_dummy_encryption options"); 2020 return -EINVAL; 2021 } 2022 return err; 2023 } 2024 2025 #define EXT4_SET_CTX(name) \ 2026 static inline void ctx_set_##name(struct ext4_fs_context *ctx, \ 2027 unsigned long flag) \ 2028 { \ 2029 ctx->mask_s_##name |= flag; \ 2030 ctx->vals_s_##name |= flag; \ 2031 } 2032 2033 #define EXT4_CLEAR_CTX(name) \ 2034 static inline void ctx_clear_##name(struct ext4_fs_context *ctx, \ 2035 unsigned long flag) \ 2036 { \ 2037 ctx->mask_s_##name |= flag; \ 2038 ctx->vals_s_##name &= ~flag; \ 2039 } 2040 2041 #define EXT4_TEST_CTX(name) \ 2042 static inline unsigned long \ 2043 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2044 { \ 2045 return (ctx->vals_s_##name & flag); \ 2046 } 2047 2048 EXT4_SET_CTX(flags); /* set only */ 2049 EXT4_SET_CTX(mount_opt); 2050 EXT4_CLEAR_CTX(mount_opt); 2051 EXT4_TEST_CTX(mount_opt); 2052 EXT4_SET_CTX(mount_opt2); 2053 EXT4_CLEAR_CTX(mount_opt2); 2054 EXT4_TEST_CTX(mount_opt2); 2055 2056 static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit) 2057 { 2058 set_bit(bit, &ctx->mask_s_mount_flags); 2059 set_bit(bit, &ctx->vals_s_mount_flags); 2060 } 2061 2062 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2063 { 2064 struct ext4_fs_context *ctx = fc->fs_private; 2065 struct fs_parse_result result; 2066 const struct mount_opts *m; 2067 int is_remount; 2068 kuid_t uid; 2069 kgid_t gid; 2070 int token; 2071 2072 token = fs_parse(fc, ext4_param_specs, param, &result); 2073 if (token < 0) 2074 return token; 2075 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2076 2077 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2078 if (token == m->token) 2079 break; 2080 2081 ctx->opt_flags |= m->flags; 2082 2083 if (m->flags & MOPT_EXPLICIT) { 2084 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2085 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2086 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2087 ctx_set_mount_opt2(ctx, 2088 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2089 } else 2090 return -EINVAL; 2091 } 2092 2093 if (m->flags & MOPT_NOSUPPORT) { 2094 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2095 param->key); 2096 return 0; 2097 } 2098 2099 switch (token) { 2100 #ifdef CONFIG_QUOTA 2101 case Opt_usrjquota: 2102 if (!*param->string) 2103 return unnote_qf_name(fc, USRQUOTA); 2104 else 2105 return note_qf_name(fc, USRQUOTA, param); 2106 case Opt_grpjquota: 2107 if (!*param->string) 2108 return unnote_qf_name(fc, GRPQUOTA); 2109 else 2110 return note_qf_name(fc, GRPQUOTA, param); 2111 #endif 2112 case Opt_sb: 2113 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2114 ext4_msg(NULL, KERN_WARNING, 2115 "Ignoring %s option on remount", param->key); 2116 } else { 2117 ctx->s_sb_block = result.uint_32; 2118 ctx->spec |= EXT4_SPEC_s_sb_block; 2119 } 2120 return 0; 2121 case Opt_removed: 2122 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2123 param->key); 2124 return 0; 2125 case Opt_abort: 2126 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED); 2127 return 0; 2128 case Opt_inlinecrypt: 2129 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2130 ctx_set_flags(ctx, SB_INLINECRYPT); 2131 #else 2132 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2133 #endif 2134 return 0; 2135 case Opt_errors: 2136 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2137 ctx_set_mount_opt(ctx, result.uint_32); 2138 return 0; 2139 #ifdef CONFIG_QUOTA 2140 case Opt_jqfmt: 2141 ctx->s_jquota_fmt = result.uint_32; 2142 ctx->spec |= EXT4_SPEC_JQFMT; 2143 return 0; 2144 #endif 2145 case Opt_data: 2146 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2147 ctx_set_mount_opt(ctx, result.uint_32); 2148 ctx->spec |= EXT4_SPEC_DATAJ; 2149 return 0; 2150 case Opt_commit: 2151 if (result.uint_32 == 0) 2152 ctx->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE; 2153 else if (result.uint_32 > INT_MAX / HZ) { 2154 ext4_msg(NULL, KERN_ERR, 2155 "Invalid commit interval %d, " 2156 "must be smaller than %d", 2157 result.uint_32, INT_MAX / HZ); 2158 return -EINVAL; 2159 } 2160 ctx->s_commit_interval = HZ * result.uint_32; 2161 ctx->spec |= EXT4_SPEC_s_commit_interval; 2162 return 0; 2163 case Opt_debug_want_extra_isize: 2164 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2165 ext4_msg(NULL, KERN_ERR, 2166 "Invalid want_extra_isize %d", result.uint_32); 2167 return -EINVAL; 2168 } 2169 ctx->s_want_extra_isize = result.uint_32; 2170 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2171 return 0; 2172 case Opt_max_batch_time: 2173 ctx->s_max_batch_time = result.uint_32; 2174 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2175 return 0; 2176 case Opt_min_batch_time: 2177 ctx->s_min_batch_time = result.uint_32; 2178 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2179 return 0; 2180 case Opt_inode_readahead_blks: 2181 if (result.uint_32 && 2182 (result.uint_32 > (1 << 30) || 2183 !is_power_of_2(result.uint_32))) { 2184 ext4_msg(NULL, KERN_ERR, 2185 "EXT4-fs: inode_readahead_blks must be " 2186 "0 or a power of 2 smaller than 2^31"); 2187 return -EINVAL; 2188 } 2189 ctx->s_inode_readahead_blks = result.uint_32; 2190 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2191 return 0; 2192 case Opt_init_itable: 2193 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2194 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2195 if (param->type == fs_value_is_string) 2196 ctx->s_li_wait_mult = result.uint_32; 2197 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2198 return 0; 2199 case Opt_max_dir_size_kb: 2200 ctx->s_max_dir_size_kb = result.uint_32; 2201 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2202 return 0; 2203 #ifdef CONFIG_EXT4_DEBUG 2204 case Opt_fc_debug_max_replay: 2205 ctx->s_fc_debug_max_replay = result.uint_32; 2206 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2207 return 0; 2208 #endif 2209 case Opt_stripe: 2210 ctx->s_stripe = result.uint_32; 2211 ctx->spec |= EXT4_SPEC_s_stripe; 2212 return 0; 2213 case Opt_resuid: 2214 uid = make_kuid(current_user_ns(), result.uint_32); 2215 if (!uid_valid(uid)) { 2216 ext4_msg(NULL, KERN_ERR, "Invalid uid value %d", 2217 result.uint_32); 2218 return -EINVAL; 2219 } 2220 ctx->s_resuid = uid; 2221 ctx->spec |= EXT4_SPEC_s_resuid; 2222 return 0; 2223 case Opt_resgid: 2224 gid = make_kgid(current_user_ns(), result.uint_32); 2225 if (!gid_valid(gid)) { 2226 ext4_msg(NULL, KERN_ERR, "Invalid gid value %d", 2227 result.uint_32); 2228 return -EINVAL; 2229 } 2230 ctx->s_resgid = gid; 2231 ctx->spec |= EXT4_SPEC_s_resgid; 2232 return 0; 2233 case Opt_journal_dev: 2234 if (is_remount) { 2235 ext4_msg(NULL, KERN_ERR, 2236 "Cannot specify journal on remount"); 2237 return -EINVAL; 2238 } 2239 ctx->journal_devnum = result.uint_32; 2240 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2241 return 0; 2242 case Opt_journal_path: 2243 { 2244 struct inode *journal_inode; 2245 struct path path; 2246 int error; 2247 2248 if (is_remount) { 2249 ext4_msg(NULL, KERN_ERR, 2250 "Cannot specify journal on remount"); 2251 return -EINVAL; 2252 } 2253 2254 error = fs_lookup_param(fc, param, 1, &path); 2255 if (error) { 2256 ext4_msg(NULL, KERN_ERR, "error: could not find " 2257 "journal device path"); 2258 return -EINVAL; 2259 } 2260 2261 journal_inode = d_inode(path.dentry); 2262 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2263 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2264 path_put(&path); 2265 return 0; 2266 } 2267 case Opt_journal_ioprio: 2268 if (result.uint_32 > 7) { 2269 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2270 " (must be 0-7)"); 2271 return -EINVAL; 2272 } 2273 ctx->journal_ioprio = 2274 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2275 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2276 return 0; 2277 case Opt_test_dummy_encryption: 2278 return ext4_parse_test_dummy_encryption(param, ctx); 2279 case Opt_dax: 2280 case Opt_dax_type: 2281 #ifdef CONFIG_FS_DAX 2282 { 2283 int type = (token == Opt_dax) ? 2284 Opt_dax : result.uint_32; 2285 2286 switch (type) { 2287 case Opt_dax: 2288 case Opt_dax_always: 2289 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2290 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2291 break; 2292 case Opt_dax_never: 2293 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2294 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2295 break; 2296 case Opt_dax_inode: 2297 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2298 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2299 /* Strictly for printing options */ 2300 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2301 break; 2302 } 2303 return 0; 2304 } 2305 #else 2306 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2307 return -EINVAL; 2308 #endif 2309 case Opt_data_err: 2310 if (result.uint_32 == Opt_data_err_abort) 2311 ctx_set_mount_opt(ctx, m->mount_opt); 2312 else if (result.uint_32 == Opt_data_err_ignore) 2313 ctx_clear_mount_opt(ctx, m->mount_opt); 2314 return 0; 2315 case Opt_mb_optimize_scan: 2316 if (result.int_32 == 1) { 2317 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2318 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2319 } else if (result.int_32 == 0) { 2320 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2321 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2322 } else { 2323 ext4_msg(NULL, KERN_WARNING, 2324 "mb_optimize_scan should be set to 0 or 1."); 2325 return -EINVAL; 2326 } 2327 return 0; 2328 } 2329 2330 /* 2331 * At this point we should only be getting options requiring MOPT_SET, 2332 * or MOPT_CLEAR. Anything else is a bug 2333 */ 2334 if (m->token == Opt_err) { 2335 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2336 param->key); 2337 WARN_ON(1); 2338 return -EINVAL; 2339 } 2340 2341 else { 2342 unsigned int set = 0; 2343 2344 if ((param->type == fs_value_is_flag) || 2345 result.uint_32 > 0) 2346 set = 1; 2347 2348 if (m->flags & MOPT_CLEAR) 2349 set = !set; 2350 else if (unlikely(!(m->flags & MOPT_SET))) { 2351 ext4_msg(NULL, KERN_WARNING, 2352 "buggy handling of option %s", 2353 param->key); 2354 WARN_ON(1); 2355 return -EINVAL; 2356 } 2357 if (m->flags & MOPT_2) { 2358 if (set != 0) 2359 ctx_set_mount_opt2(ctx, m->mount_opt); 2360 else 2361 ctx_clear_mount_opt2(ctx, m->mount_opt); 2362 } else { 2363 if (set != 0) 2364 ctx_set_mount_opt(ctx, m->mount_opt); 2365 else 2366 ctx_clear_mount_opt(ctx, m->mount_opt); 2367 } 2368 } 2369 2370 return 0; 2371 } 2372 2373 static int parse_options(struct fs_context *fc, char *options) 2374 { 2375 struct fs_parameter param; 2376 int ret; 2377 char *key; 2378 2379 if (!options) 2380 return 0; 2381 2382 while ((key = strsep(&options, ",")) != NULL) { 2383 if (*key) { 2384 size_t v_len = 0; 2385 char *value = strchr(key, '='); 2386 2387 param.type = fs_value_is_flag; 2388 param.string = NULL; 2389 2390 if (value) { 2391 if (value == key) 2392 continue; 2393 2394 *value++ = 0; 2395 v_len = strlen(value); 2396 param.string = kmemdup_nul(value, v_len, 2397 GFP_KERNEL); 2398 if (!param.string) 2399 return -ENOMEM; 2400 param.type = fs_value_is_string; 2401 } 2402 2403 param.key = key; 2404 param.size = v_len; 2405 2406 ret = ext4_parse_param(fc, ¶m); 2407 if (param.string) 2408 kfree(param.string); 2409 if (ret < 0) 2410 return ret; 2411 } 2412 } 2413 2414 ret = ext4_validate_options(fc); 2415 if (ret < 0) 2416 return ret; 2417 2418 return 0; 2419 } 2420 2421 static int parse_apply_sb_mount_options(struct super_block *sb, 2422 struct ext4_fs_context *m_ctx) 2423 { 2424 struct ext4_sb_info *sbi = EXT4_SB(sb); 2425 char *s_mount_opts = NULL; 2426 struct ext4_fs_context *s_ctx = NULL; 2427 struct fs_context *fc = NULL; 2428 int ret = -ENOMEM; 2429 2430 if (!sbi->s_es->s_mount_opts[0]) 2431 return 0; 2432 2433 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2434 sizeof(sbi->s_es->s_mount_opts), 2435 GFP_KERNEL); 2436 if (!s_mount_opts) 2437 return ret; 2438 2439 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2440 if (!fc) 2441 goto out_free; 2442 2443 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2444 if (!s_ctx) 2445 goto out_free; 2446 2447 fc->fs_private = s_ctx; 2448 fc->s_fs_info = sbi; 2449 2450 ret = parse_options(fc, s_mount_opts); 2451 if (ret < 0) 2452 goto parse_failed; 2453 2454 ret = ext4_check_opt_consistency(fc, sb); 2455 if (ret < 0) { 2456 parse_failed: 2457 ext4_msg(sb, KERN_WARNING, 2458 "failed to parse options in superblock: %s", 2459 s_mount_opts); 2460 ret = 0; 2461 goto out_free; 2462 } 2463 2464 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2465 m_ctx->journal_devnum = s_ctx->journal_devnum; 2466 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2467 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2468 2469 ext4_apply_options(fc, sb); 2470 ret = 0; 2471 2472 out_free: 2473 if (fc) { 2474 ext4_fc_free(fc); 2475 kfree(fc); 2476 } 2477 kfree(s_mount_opts); 2478 return ret; 2479 } 2480 2481 static void ext4_apply_quota_options(struct fs_context *fc, 2482 struct super_block *sb) 2483 { 2484 #ifdef CONFIG_QUOTA 2485 bool quota_feature = ext4_has_feature_quota(sb); 2486 struct ext4_fs_context *ctx = fc->fs_private; 2487 struct ext4_sb_info *sbi = EXT4_SB(sb); 2488 char *qname; 2489 int i; 2490 2491 if (quota_feature) 2492 return; 2493 2494 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2495 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2496 if (!(ctx->qname_spec & (1 << i))) 2497 continue; 2498 2499 qname = ctx->s_qf_names[i]; /* May be NULL */ 2500 if (qname) 2501 set_opt(sb, QUOTA); 2502 ctx->s_qf_names[i] = NULL; 2503 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2504 lockdep_is_held(&sb->s_umount)); 2505 if (qname) 2506 kfree_rcu(qname); 2507 } 2508 } 2509 2510 if (ctx->spec & EXT4_SPEC_JQFMT) 2511 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2512 #endif 2513 } 2514 2515 /* 2516 * Check quota settings consistency. 2517 */ 2518 static int ext4_check_quota_consistency(struct fs_context *fc, 2519 struct super_block *sb) 2520 { 2521 #ifdef CONFIG_QUOTA 2522 struct ext4_fs_context *ctx = fc->fs_private; 2523 struct ext4_sb_info *sbi = EXT4_SB(sb); 2524 bool quota_feature = ext4_has_feature_quota(sb); 2525 bool quota_loaded = sb_any_quota_loaded(sb); 2526 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2527 int quota_flags, i; 2528 2529 /* 2530 * We do the test below only for project quotas. 'usrquota' and 2531 * 'grpquota' mount options are allowed even without quota feature 2532 * to support legacy quotas in quota files. 2533 */ 2534 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2535 !ext4_has_feature_project(sb)) { 2536 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2537 "Cannot enable project quota enforcement."); 2538 return -EINVAL; 2539 } 2540 2541 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2542 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2543 if (quota_loaded && 2544 ctx->mask_s_mount_opt & quota_flags && 2545 !ctx_test_mount_opt(ctx, quota_flags)) 2546 goto err_quota_change; 2547 2548 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2549 2550 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2551 if (!(ctx->qname_spec & (1 << i))) 2552 continue; 2553 2554 if (quota_loaded && 2555 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2556 goto err_jquota_change; 2557 2558 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2559 strcmp(get_qf_name(sb, sbi, i), 2560 ctx->s_qf_names[i]) != 0) 2561 goto err_jquota_specified; 2562 } 2563 2564 if (quota_feature) { 2565 ext4_msg(NULL, KERN_INFO, 2566 "Journaled quota options ignored when " 2567 "QUOTA feature is enabled"); 2568 return 0; 2569 } 2570 } 2571 2572 if (ctx->spec & EXT4_SPEC_JQFMT) { 2573 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2574 goto err_jquota_change; 2575 if (quota_feature) { 2576 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2577 "ignored when QUOTA feature is enabled"); 2578 return 0; 2579 } 2580 } 2581 2582 /* Make sure we don't mix old and new quota format */ 2583 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2584 ctx->s_qf_names[USRQUOTA]); 2585 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2586 ctx->s_qf_names[GRPQUOTA]); 2587 2588 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2589 test_opt(sb, USRQUOTA)); 2590 2591 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2592 test_opt(sb, GRPQUOTA)); 2593 2594 if (usr_qf_name) { 2595 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2596 usrquota = false; 2597 } 2598 if (grp_qf_name) { 2599 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2600 grpquota = false; 2601 } 2602 2603 if (usr_qf_name || grp_qf_name) { 2604 if (usrquota || grpquota) { 2605 ext4_msg(NULL, KERN_ERR, "old and new quota " 2606 "format mixing"); 2607 return -EINVAL; 2608 } 2609 2610 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2611 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2612 "not specified"); 2613 return -EINVAL; 2614 } 2615 } 2616 2617 return 0; 2618 2619 err_quota_change: 2620 ext4_msg(NULL, KERN_ERR, 2621 "Cannot change quota options when quota turned on"); 2622 return -EINVAL; 2623 err_jquota_change: 2624 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2625 "options when quota turned on"); 2626 return -EINVAL; 2627 err_jquota_specified: 2628 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2629 QTYPE2NAME(i)); 2630 return -EINVAL; 2631 #else 2632 return 0; 2633 #endif 2634 } 2635 2636 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2637 struct super_block *sb) 2638 { 2639 const struct ext4_fs_context *ctx = fc->fs_private; 2640 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2641 int err; 2642 2643 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2644 return 0; 2645 2646 if (!ext4_has_feature_encrypt(sb)) { 2647 ext4_msg(NULL, KERN_WARNING, 2648 "test_dummy_encryption requires encrypt feature"); 2649 return -EINVAL; 2650 } 2651 /* 2652 * This mount option is just for testing, and it's not worthwhile to 2653 * implement the extra complexity (e.g. RCU protection) that would be 2654 * needed to allow it to be set or changed during remount. We do allow 2655 * it to be specified during remount, but only if there is no change. 2656 */ 2657 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2658 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2659 &ctx->dummy_enc_policy)) 2660 return 0; 2661 ext4_msg(NULL, KERN_WARNING, 2662 "Can't set or change test_dummy_encryption on remount"); 2663 return -EINVAL; 2664 } 2665 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2666 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2667 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2668 &ctx->dummy_enc_policy)) 2669 return 0; 2670 ext4_msg(NULL, KERN_WARNING, 2671 "Conflicting test_dummy_encryption options"); 2672 return -EINVAL; 2673 } 2674 /* 2675 * fscrypt_add_test_dummy_key() technically changes the super_block, so 2676 * technically it should be delayed until ext4_apply_options() like the 2677 * other changes. But since we never get here for remounts (see above), 2678 * and this is the last chance to report errors, we do it here. 2679 */ 2680 err = fscrypt_add_test_dummy_key(sb, &ctx->dummy_enc_policy); 2681 if (err) 2682 ext4_msg(NULL, KERN_WARNING, 2683 "Error adding test dummy encryption key [%d]", err); 2684 return err; 2685 } 2686 2687 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2688 struct super_block *sb) 2689 { 2690 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2691 /* if already set, it was already verified to be the same */ 2692 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2693 return; 2694 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2695 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2696 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2697 } 2698 2699 static int ext4_check_opt_consistency(struct fs_context *fc, 2700 struct super_block *sb) 2701 { 2702 struct ext4_fs_context *ctx = fc->fs_private; 2703 struct ext4_sb_info *sbi = fc->s_fs_info; 2704 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2705 int err; 2706 2707 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2708 ext4_msg(NULL, KERN_ERR, 2709 "Mount option(s) incompatible with ext2"); 2710 return -EINVAL; 2711 } 2712 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2713 ext4_msg(NULL, KERN_ERR, 2714 "Mount option(s) incompatible with ext3"); 2715 return -EINVAL; 2716 } 2717 2718 if (ctx->s_want_extra_isize > 2719 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2720 ext4_msg(NULL, KERN_ERR, 2721 "Invalid want_extra_isize %d", 2722 ctx->s_want_extra_isize); 2723 return -EINVAL; 2724 } 2725 2726 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) { 2727 int blocksize = 2728 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2729 if (blocksize < PAGE_SIZE) 2730 ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an " 2731 "experimental mount option 'dioread_nolock' " 2732 "for blocksize < PAGE_SIZE"); 2733 } 2734 2735 err = ext4_check_test_dummy_encryption(fc, sb); 2736 if (err) 2737 return err; 2738 2739 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2740 if (!sbi->s_journal) { 2741 ext4_msg(NULL, KERN_WARNING, 2742 "Remounting file system with no journal " 2743 "so ignoring journalled data option"); 2744 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2745 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2746 test_opt(sb, DATA_FLAGS)) { 2747 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2748 "on remount"); 2749 return -EINVAL; 2750 } 2751 } 2752 2753 if (is_remount) { 2754 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2755 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2756 ext4_msg(NULL, KERN_ERR, "can't mount with " 2757 "both data=journal and dax"); 2758 return -EINVAL; 2759 } 2760 2761 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2762 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2763 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2764 fail_dax_change_remount: 2765 ext4_msg(NULL, KERN_ERR, "can't change " 2766 "dax mount option while remounting"); 2767 return -EINVAL; 2768 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2769 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2770 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2771 goto fail_dax_change_remount; 2772 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2773 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2774 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2775 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2776 goto fail_dax_change_remount; 2777 } 2778 } 2779 2780 return ext4_check_quota_consistency(fc, sb); 2781 } 2782 2783 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2784 { 2785 struct ext4_fs_context *ctx = fc->fs_private; 2786 struct ext4_sb_info *sbi = fc->s_fs_info; 2787 2788 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2789 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2790 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2791 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2792 sbi->s_mount_flags &= ~ctx->mask_s_mount_flags; 2793 sbi->s_mount_flags |= ctx->vals_s_mount_flags; 2794 sb->s_flags &= ~ctx->mask_s_flags; 2795 sb->s_flags |= ctx->vals_s_flags; 2796 2797 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2798 APPLY(s_commit_interval); 2799 APPLY(s_stripe); 2800 APPLY(s_max_batch_time); 2801 APPLY(s_min_batch_time); 2802 APPLY(s_want_extra_isize); 2803 APPLY(s_inode_readahead_blks); 2804 APPLY(s_max_dir_size_kb); 2805 APPLY(s_li_wait_mult); 2806 APPLY(s_resgid); 2807 APPLY(s_resuid); 2808 2809 #ifdef CONFIG_EXT4_DEBUG 2810 APPLY(s_fc_debug_max_replay); 2811 #endif 2812 2813 ext4_apply_quota_options(fc, sb); 2814 ext4_apply_test_dummy_encryption(ctx, sb); 2815 } 2816 2817 2818 static int ext4_validate_options(struct fs_context *fc) 2819 { 2820 #ifdef CONFIG_QUOTA 2821 struct ext4_fs_context *ctx = fc->fs_private; 2822 char *usr_qf_name, *grp_qf_name; 2823 2824 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2825 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2826 2827 if (usr_qf_name || grp_qf_name) { 2828 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2829 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2830 2831 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2832 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2833 2834 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2835 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2836 ext4_msg(NULL, KERN_ERR, "old and new quota " 2837 "format mixing"); 2838 return -EINVAL; 2839 } 2840 } 2841 #endif 2842 return 1; 2843 } 2844 2845 static inline void ext4_show_quota_options(struct seq_file *seq, 2846 struct super_block *sb) 2847 { 2848 #if defined(CONFIG_QUOTA) 2849 struct ext4_sb_info *sbi = EXT4_SB(sb); 2850 char *usr_qf_name, *grp_qf_name; 2851 2852 if (sbi->s_jquota_fmt) { 2853 char *fmtname = ""; 2854 2855 switch (sbi->s_jquota_fmt) { 2856 case QFMT_VFS_OLD: 2857 fmtname = "vfsold"; 2858 break; 2859 case QFMT_VFS_V0: 2860 fmtname = "vfsv0"; 2861 break; 2862 case QFMT_VFS_V1: 2863 fmtname = "vfsv1"; 2864 break; 2865 } 2866 seq_printf(seq, ",jqfmt=%s", fmtname); 2867 } 2868 2869 rcu_read_lock(); 2870 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2871 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2872 if (usr_qf_name) 2873 seq_show_option(seq, "usrjquota", usr_qf_name); 2874 if (grp_qf_name) 2875 seq_show_option(seq, "grpjquota", grp_qf_name); 2876 rcu_read_unlock(); 2877 #endif 2878 } 2879 2880 static const char *token2str(int token) 2881 { 2882 const struct fs_parameter_spec *spec; 2883 2884 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2885 if (spec->opt == token && !spec->type) 2886 break; 2887 return spec->name; 2888 } 2889 2890 /* 2891 * Show an option if 2892 * - it's set to a non-default value OR 2893 * - if the per-sb default is different from the global default 2894 */ 2895 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2896 int nodefs) 2897 { 2898 struct ext4_sb_info *sbi = EXT4_SB(sb); 2899 struct ext4_super_block *es = sbi->s_es; 2900 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2901 const struct mount_opts *m; 2902 char sep = nodefs ? '\n' : ','; 2903 2904 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2905 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2906 2907 if (sbi->s_sb_block != 1) 2908 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2909 2910 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2911 int want_set = m->flags & MOPT_SET; 2912 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2913 m->flags & MOPT_SKIP) 2914 continue; 2915 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2916 continue; /* skip if same as the default */ 2917 if ((want_set && 2918 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2919 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2920 continue; /* select Opt_noFoo vs Opt_Foo */ 2921 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2922 } 2923 2924 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2925 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2926 SEQ_OPTS_PRINT("resuid=%u", 2927 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2928 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2929 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2930 SEQ_OPTS_PRINT("resgid=%u", 2931 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2932 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2933 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2934 SEQ_OPTS_PUTS("errors=remount-ro"); 2935 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2936 SEQ_OPTS_PUTS("errors=continue"); 2937 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2938 SEQ_OPTS_PUTS("errors=panic"); 2939 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2940 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2941 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2942 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2943 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2944 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2945 if (nodefs || sbi->s_stripe) 2946 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2947 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2948 (sbi->s_mount_opt ^ def_mount_opt)) { 2949 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2950 SEQ_OPTS_PUTS("data=journal"); 2951 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2952 SEQ_OPTS_PUTS("data=ordered"); 2953 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2954 SEQ_OPTS_PUTS("data=writeback"); 2955 } 2956 if (nodefs || 2957 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2958 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2959 sbi->s_inode_readahead_blks); 2960 2961 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2962 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2963 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2964 if (nodefs || sbi->s_max_dir_size_kb) 2965 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2966 if (test_opt(sb, DATA_ERR_ABORT)) 2967 SEQ_OPTS_PUTS("data_err=abort"); 2968 2969 fscrypt_show_test_dummy_encryption(seq, sep, sb); 2970 2971 if (sb->s_flags & SB_INLINECRYPT) 2972 SEQ_OPTS_PUTS("inlinecrypt"); 2973 2974 if (test_opt(sb, DAX_ALWAYS)) { 2975 if (IS_EXT2_SB(sb)) 2976 SEQ_OPTS_PUTS("dax"); 2977 else 2978 SEQ_OPTS_PUTS("dax=always"); 2979 } else if (test_opt2(sb, DAX_NEVER)) { 2980 SEQ_OPTS_PUTS("dax=never"); 2981 } else if (test_opt2(sb, DAX_INODE)) { 2982 SEQ_OPTS_PUTS("dax=inode"); 2983 } 2984 2985 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 2986 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 2987 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 2988 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 2989 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 2990 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 2991 } 2992 2993 ext4_show_quota_options(seq, sb); 2994 return 0; 2995 } 2996 2997 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2998 { 2999 return _ext4_show_options(seq, root->d_sb, 0); 3000 } 3001 3002 int ext4_seq_options_show(struct seq_file *seq, void *offset) 3003 { 3004 struct super_block *sb = seq->private; 3005 int rc; 3006 3007 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3008 rc = _ext4_show_options(seq, sb, 1); 3009 seq_puts(seq, "\n"); 3010 return rc; 3011 } 3012 3013 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3014 int read_only) 3015 { 3016 struct ext4_sb_info *sbi = EXT4_SB(sb); 3017 int err = 0; 3018 3019 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3020 ext4_msg(sb, KERN_ERR, "revision level too high, " 3021 "forcing read-only mode"); 3022 err = -EROFS; 3023 goto done; 3024 } 3025 if (read_only) 3026 goto done; 3027 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3028 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3029 "running e2fsck is recommended"); 3030 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3031 ext4_msg(sb, KERN_WARNING, 3032 "warning: mounting fs with errors, " 3033 "running e2fsck is recommended"); 3034 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3035 le16_to_cpu(es->s_mnt_count) >= 3036 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3037 ext4_msg(sb, KERN_WARNING, 3038 "warning: maximal mount count reached, " 3039 "running e2fsck is recommended"); 3040 else if (le32_to_cpu(es->s_checkinterval) && 3041 (ext4_get_tstamp(es, s_lastcheck) + 3042 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3043 ext4_msg(sb, KERN_WARNING, 3044 "warning: checktime reached, " 3045 "running e2fsck is recommended"); 3046 if (!sbi->s_journal) 3047 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3048 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3049 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3050 le16_add_cpu(&es->s_mnt_count, 1); 3051 ext4_update_tstamp(es, s_mtime); 3052 if (sbi->s_journal) { 3053 ext4_set_feature_journal_needs_recovery(sb); 3054 if (ext4_has_feature_orphan_file(sb)) 3055 ext4_set_feature_orphan_present(sb); 3056 } 3057 3058 err = ext4_commit_super(sb); 3059 done: 3060 if (test_opt(sb, DEBUG)) 3061 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3062 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3063 sb->s_blocksize, 3064 sbi->s_groups_count, 3065 EXT4_BLOCKS_PER_GROUP(sb), 3066 EXT4_INODES_PER_GROUP(sb), 3067 sbi->s_mount_opt, sbi->s_mount_opt2); 3068 return err; 3069 } 3070 3071 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3072 { 3073 struct ext4_sb_info *sbi = EXT4_SB(sb); 3074 struct flex_groups **old_groups, **new_groups; 3075 int size, i, j; 3076 3077 if (!sbi->s_log_groups_per_flex) 3078 return 0; 3079 3080 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3081 if (size <= sbi->s_flex_groups_allocated) 3082 return 0; 3083 3084 new_groups = kvzalloc(roundup_pow_of_two(size * 3085 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3086 if (!new_groups) { 3087 ext4_msg(sb, KERN_ERR, 3088 "not enough memory for %d flex group pointers", size); 3089 return -ENOMEM; 3090 } 3091 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3092 new_groups[i] = kvzalloc(roundup_pow_of_two( 3093 sizeof(struct flex_groups)), 3094 GFP_KERNEL); 3095 if (!new_groups[i]) { 3096 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3097 kvfree(new_groups[j]); 3098 kvfree(new_groups); 3099 ext4_msg(sb, KERN_ERR, 3100 "not enough memory for %d flex groups", size); 3101 return -ENOMEM; 3102 } 3103 } 3104 rcu_read_lock(); 3105 old_groups = rcu_dereference(sbi->s_flex_groups); 3106 if (old_groups) 3107 memcpy(new_groups, old_groups, 3108 (sbi->s_flex_groups_allocated * 3109 sizeof(struct flex_groups *))); 3110 rcu_read_unlock(); 3111 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3112 sbi->s_flex_groups_allocated = size; 3113 if (old_groups) 3114 ext4_kvfree_array_rcu(old_groups); 3115 return 0; 3116 } 3117 3118 static int ext4_fill_flex_info(struct super_block *sb) 3119 { 3120 struct ext4_sb_info *sbi = EXT4_SB(sb); 3121 struct ext4_group_desc *gdp = NULL; 3122 struct flex_groups *fg; 3123 ext4_group_t flex_group; 3124 int i, err; 3125 3126 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3127 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3128 sbi->s_log_groups_per_flex = 0; 3129 return 1; 3130 } 3131 3132 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3133 if (err) 3134 goto failed; 3135 3136 for (i = 0; i < sbi->s_groups_count; i++) { 3137 gdp = ext4_get_group_desc(sb, i, NULL); 3138 3139 flex_group = ext4_flex_group(sbi, i); 3140 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3141 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3142 atomic64_add(ext4_free_group_clusters(sb, gdp), 3143 &fg->free_clusters); 3144 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3145 } 3146 3147 return 1; 3148 failed: 3149 return 0; 3150 } 3151 3152 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3153 struct ext4_group_desc *gdp) 3154 { 3155 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3156 __u16 crc = 0; 3157 __le32 le_group = cpu_to_le32(block_group); 3158 struct ext4_sb_info *sbi = EXT4_SB(sb); 3159 3160 if (ext4_has_metadata_csum(sbi->s_sb)) { 3161 /* Use new metadata_csum algorithm */ 3162 __u32 csum32; 3163 __u16 dummy_csum = 0; 3164 3165 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 3166 sizeof(le_group)); 3167 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 3168 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 3169 sizeof(dummy_csum)); 3170 offset += sizeof(dummy_csum); 3171 if (offset < sbi->s_desc_size) 3172 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 3173 sbi->s_desc_size - offset); 3174 3175 crc = csum32 & 0xFFFF; 3176 goto out; 3177 } 3178 3179 /* old crc16 code */ 3180 if (!ext4_has_feature_gdt_csum(sb)) 3181 return 0; 3182 3183 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3184 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3185 crc = crc16(crc, (__u8 *)gdp, offset); 3186 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3187 /* for checksum of struct ext4_group_desc do the rest...*/ 3188 if (ext4_has_feature_64bit(sb) && 3189 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 3190 crc = crc16(crc, (__u8 *)gdp + offset, 3191 le16_to_cpu(sbi->s_es->s_desc_size) - 3192 offset); 3193 3194 out: 3195 return cpu_to_le16(crc); 3196 } 3197 3198 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3199 struct ext4_group_desc *gdp) 3200 { 3201 if (ext4_has_group_desc_csum(sb) && 3202 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3203 return 0; 3204 3205 return 1; 3206 } 3207 3208 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3209 struct ext4_group_desc *gdp) 3210 { 3211 if (!ext4_has_group_desc_csum(sb)) 3212 return; 3213 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3214 } 3215 3216 /* Called at mount-time, super-block is locked */ 3217 static int ext4_check_descriptors(struct super_block *sb, 3218 ext4_fsblk_t sb_block, 3219 ext4_group_t *first_not_zeroed) 3220 { 3221 struct ext4_sb_info *sbi = EXT4_SB(sb); 3222 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3223 ext4_fsblk_t last_block; 3224 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3225 ext4_fsblk_t block_bitmap; 3226 ext4_fsblk_t inode_bitmap; 3227 ext4_fsblk_t inode_table; 3228 int flexbg_flag = 0; 3229 ext4_group_t i, grp = sbi->s_groups_count; 3230 3231 if (ext4_has_feature_flex_bg(sb)) 3232 flexbg_flag = 1; 3233 3234 ext4_debug("Checking group descriptors"); 3235 3236 for (i = 0; i < sbi->s_groups_count; i++) { 3237 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3238 3239 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3240 last_block = ext4_blocks_count(sbi->s_es) - 1; 3241 else 3242 last_block = first_block + 3243 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3244 3245 if ((grp == sbi->s_groups_count) && 3246 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3247 grp = i; 3248 3249 block_bitmap = ext4_block_bitmap(sb, gdp); 3250 if (block_bitmap == sb_block) { 3251 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3252 "Block bitmap for group %u overlaps " 3253 "superblock", i); 3254 if (!sb_rdonly(sb)) 3255 return 0; 3256 } 3257 if (block_bitmap >= sb_block + 1 && 3258 block_bitmap <= last_bg_block) { 3259 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3260 "Block bitmap for group %u overlaps " 3261 "block group descriptors", i); 3262 if (!sb_rdonly(sb)) 3263 return 0; 3264 } 3265 if (block_bitmap < first_block || block_bitmap > last_block) { 3266 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3267 "Block bitmap for group %u not in group " 3268 "(block %llu)!", i, block_bitmap); 3269 return 0; 3270 } 3271 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3272 if (inode_bitmap == sb_block) { 3273 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3274 "Inode bitmap for group %u overlaps " 3275 "superblock", i); 3276 if (!sb_rdonly(sb)) 3277 return 0; 3278 } 3279 if (inode_bitmap >= sb_block + 1 && 3280 inode_bitmap <= last_bg_block) { 3281 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3282 "Inode bitmap for group %u overlaps " 3283 "block group descriptors", i); 3284 if (!sb_rdonly(sb)) 3285 return 0; 3286 } 3287 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3288 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3289 "Inode bitmap for group %u not in group " 3290 "(block %llu)!", i, inode_bitmap); 3291 return 0; 3292 } 3293 inode_table = ext4_inode_table(sb, gdp); 3294 if (inode_table == sb_block) { 3295 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3296 "Inode table for group %u overlaps " 3297 "superblock", i); 3298 if (!sb_rdonly(sb)) 3299 return 0; 3300 } 3301 if (inode_table >= sb_block + 1 && 3302 inode_table <= last_bg_block) { 3303 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3304 "Inode table for group %u overlaps " 3305 "block group descriptors", i); 3306 if (!sb_rdonly(sb)) 3307 return 0; 3308 } 3309 if (inode_table < first_block || 3310 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3311 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3312 "Inode table for group %u not in group " 3313 "(block %llu)!", i, inode_table); 3314 return 0; 3315 } 3316 ext4_lock_group(sb, i); 3317 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3318 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3319 "Checksum for group %u failed (%u!=%u)", 3320 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3321 gdp)), le16_to_cpu(gdp->bg_checksum)); 3322 if (!sb_rdonly(sb)) { 3323 ext4_unlock_group(sb, i); 3324 return 0; 3325 } 3326 } 3327 ext4_unlock_group(sb, i); 3328 if (!flexbg_flag) 3329 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3330 } 3331 if (NULL != first_not_zeroed) 3332 *first_not_zeroed = grp; 3333 return 1; 3334 } 3335 3336 /* 3337 * Maximal extent format file size. 3338 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3339 * extent format containers, within a sector_t, and within i_blocks 3340 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3341 * so that won't be a limiting factor. 3342 * 3343 * However there is other limiting factor. We do store extents in the form 3344 * of starting block and length, hence the resulting length of the extent 3345 * covering maximum file size must fit into on-disk format containers as 3346 * well. Given that length is always by 1 unit bigger than max unit (because 3347 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3348 * 3349 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3350 */ 3351 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3352 { 3353 loff_t res; 3354 loff_t upper_limit = MAX_LFS_FILESIZE; 3355 3356 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3357 3358 if (!has_huge_files) { 3359 upper_limit = (1LL << 32) - 1; 3360 3361 /* total blocks in file system block size */ 3362 upper_limit >>= (blkbits - 9); 3363 upper_limit <<= blkbits; 3364 } 3365 3366 /* 3367 * 32-bit extent-start container, ee_block. We lower the maxbytes 3368 * by one fs block, so ee_len can cover the extent of maximum file 3369 * size 3370 */ 3371 res = (1LL << 32) - 1; 3372 res <<= blkbits; 3373 3374 /* Sanity check against vm- & vfs- imposed limits */ 3375 if (res > upper_limit) 3376 res = upper_limit; 3377 3378 return res; 3379 } 3380 3381 /* 3382 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3383 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3384 * We need to be 1 filesystem block less than the 2^48 sector limit. 3385 */ 3386 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3387 { 3388 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3389 int meta_blocks; 3390 unsigned int ppb = 1 << (bits - 2); 3391 3392 /* 3393 * This is calculated to be the largest file size for a dense, block 3394 * mapped file such that the file's total number of 512-byte sectors, 3395 * including data and all indirect blocks, does not exceed (2^48 - 1). 3396 * 3397 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3398 * number of 512-byte sectors of the file. 3399 */ 3400 if (!has_huge_files) { 3401 /* 3402 * !has_huge_files or implies that the inode i_block field 3403 * represents total file blocks in 2^32 512-byte sectors == 3404 * size of vfs inode i_blocks * 8 3405 */ 3406 upper_limit = (1LL << 32) - 1; 3407 3408 /* total blocks in file system block size */ 3409 upper_limit >>= (bits - 9); 3410 3411 } else { 3412 /* 3413 * We use 48 bit ext4_inode i_blocks 3414 * With EXT4_HUGE_FILE_FL set the i_blocks 3415 * represent total number of blocks in 3416 * file system block size 3417 */ 3418 upper_limit = (1LL << 48) - 1; 3419 3420 } 3421 3422 /* Compute how many blocks we can address by block tree */ 3423 res += ppb; 3424 res += ppb * ppb; 3425 res += ((loff_t)ppb) * ppb * ppb; 3426 /* Compute how many metadata blocks are needed */ 3427 meta_blocks = 1; 3428 meta_blocks += 1 + ppb; 3429 meta_blocks += 1 + ppb + ppb * ppb; 3430 /* Does block tree limit file size? */ 3431 if (res + meta_blocks <= upper_limit) 3432 goto check_lfs; 3433 3434 res = upper_limit; 3435 /* How many metadata blocks are needed for addressing upper_limit? */ 3436 upper_limit -= EXT4_NDIR_BLOCKS; 3437 /* indirect blocks */ 3438 meta_blocks = 1; 3439 upper_limit -= ppb; 3440 /* double indirect blocks */ 3441 if (upper_limit < ppb * ppb) { 3442 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3443 res -= meta_blocks; 3444 goto check_lfs; 3445 } 3446 meta_blocks += 1 + ppb; 3447 upper_limit -= ppb * ppb; 3448 /* tripple indirect blocks for the rest */ 3449 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3450 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3451 res -= meta_blocks; 3452 check_lfs: 3453 res <<= bits; 3454 if (res > MAX_LFS_FILESIZE) 3455 res = MAX_LFS_FILESIZE; 3456 3457 return res; 3458 } 3459 3460 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3461 ext4_fsblk_t logical_sb_block, int nr) 3462 { 3463 struct ext4_sb_info *sbi = EXT4_SB(sb); 3464 ext4_group_t bg, first_meta_bg; 3465 int has_super = 0; 3466 3467 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3468 3469 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3470 return logical_sb_block + nr + 1; 3471 bg = sbi->s_desc_per_block * nr; 3472 if (ext4_bg_has_super(sb, bg)) 3473 has_super = 1; 3474 3475 /* 3476 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3477 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3478 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3479 * compensate. 3480 */ 3481 if (sb->s_blocksize == 1024 && nr == 0 && 3482 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3483 has_super++; 3484 3485 return (has_super + ext4_group_first_block_no(sb, bg)); 3486 } 3487 3488 /** 3489 * ext4_get_stripe_size: Get the stripe size. 3490 * @sbi: In memory super block info 3491 * 3492 * If we have specified it via mount option, then 3493 * use the mount option value. If the value specified at mount time is 3494 * greater than the blocks per group use the super block value. 3495 * If the super block value is greater than blocks per group return 0. 3496 * Allocator needs it be less than blocks per group. 3497 * 3498 */ 3499 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3500 { 3501 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3502 unsigned long stripe_width = 3503 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3504 int ret; 3505 3506 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3507 ret = sbi->s_stripe; 3508 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3509 ret = stripe_width; 3510 else if (stride && stride <= sbi->s_blocks_per_group) 3511 ret = stride; 3512 else 3513 ret = 0; 3514 3515 /* 3516 * If the stripe width is 1, this makes no sense and 3517 * we set it to 0 to turn off stripe handling code. 3518 */ 3519 if (ret <= 1) 3520 ret = 0; 3521 3522 return ret; 3523 } 3524 3525 /* 3526 * Check whether this filesystem can be mounted based on 3527 * the features present and the RDONLY/RDWR mount requested. 3528 * Returns 1 if this filesystem can be mounted as requested, 3529 * 0 if it cannot be. 3530 */ 3531 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3532 { 3533 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3534 ext4_msg(sb, KERN_ERR, 3535 "Couldn't mount because of " 3536 "unsupported optional features (%x)", 3537 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3538 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3539 return 0; 3540 } 3541 3542 #if !IS_ENABLED(CONFIG_UNICODE) 3543 if (ext4_has_feature_casefold(sb)) { 3544 ext4_msg(sb, KERN_ERR, 3545 "Filesystem with casefold feature cannot be " 3546 "mounted without CONFIG_UNICODE"); 3547 return 0; 3548 } 3549 #endif 3550 3551 if (readonly) 3552 return 1; 3553 3554 if (ext4_has_feature_readonly(sb)) { 3555 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3556 sb->s_flags |= SB_RDONLY; 3557 return 1; 3558 } 3559 3560 /* Check that feature set is OK for a read-write mount */ 3561 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3562 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3563 "unsupported optional features (%x)", 3564 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3565 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3566 return 0; 3567 } 3568 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3569 ext4_msg(sb, KERN_ERR, 3570 "Can't support bigalloc feature without " 3571 "extents feature\n"); 3572 return 0; 3573 } 3574 3575 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3576 if (!readonly && (ext4_has_feature_quota(sb) || 3577 ext4_has_feature_project(sb))) { 3578 ext4_msg(sb, KERN_ERR, 3579 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3580 return 0; 3581 } 3582 #endif /* CONFIG_QUOTA */ 3583 return 1; 3584 } 3585 3586 /* 3587 * This function is called once a day if we have errors logged 3588 * on the file system 3589 */ 3590 static void print_daily_error_info(struct timer_list *t) 3591 { 3592 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3593 struct super_block *sb = sbi->s_sb; 3594 struct ext4_super_block *es = sbi->s_es; 3595 3596 if (es->s_error_count) 3597 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3598 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3599 le32_to_cpu(es->s_error_count)); 3600 if (es->s_first_error_time) { 3601 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3602 sb->s_id, 3603 ext4_get_tstamp(es, s_first_error_time), 3604 (int) sizeof(es->s_first_error_func), 3605 es->s_first_error_func, 3606 le32_to_cpu(es->s_first_error_line)); 3607 if (es->s_first_error_ino) 3608 printk(KERN_CONT ": inode %u", 3609 le32_to_cpu(es->s_first_error_ino)); 3610 if (es->s_first_error_block) 3611 printk(KERN_CONT ": block %llu", (unsigned long long) 3612 le64_to_cpu(es->s_first_error_block)); 3613 printk(KERN_CONT "\n"); 3614 } 3615 if (es->s_last_error_time) { 3616 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3617 sb->s_id, 3618 ext4_get_tstamp(es, s_last_error_time), 3619 (int) sizeof(es->s_last_error_func), 3620 es->s_last_error_func, 3621 le32_to_cpu(es->s_last_error_line)); 3622 if (es->s_last_error_ino) 3623 printk(KERN_CONT ": inode %u", 3624 le32_to_cpu(es->s_last_error_ino)); 3625 if (es->s_last_error_block) 3626 printk(KERN_CONT ": block %llu", (unsigned long long) 3627 le64_to_cpu(es->s_last_error_block)); 3628 printk(KERN_CONT "\n"); 3629 } 3630 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3631 } 3632 3633 /* Find next suitable group and run ext4_init_inode_table */ 3634 static int ext4_run_li_request(struct ext4_li_request *elr) 3635 { 3636 struct ext4_group_desc *gdp = NULL; 3637 struct super_block *sb = elr->lr_super; 3638 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3639 ext4_group_t group = elr->lr_next_group; 3640 unsigned int prefetch_ios = 0; 3641 int ret = 0; 3642 u64 start_time; 3643 3644 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3645 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3646 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3647 if (prefetch_ios) 3648 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3649 prefetch_ios); 3650 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3651 prefetch_ios); 3652 if (group >= elr->lr_next_group) { 3653 ret = 1; 3654 if (elr->lr_first_not_zeroed != ngroups && 3655 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3656 elr->lr_next_group = elr->lr_first_not_zeroed; 3657 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3658 ret = 0; 3659 } 3660 } 3661 return ret; 3662 } 3663 3664 for (; group < ngroups; group++) { 3665 gdp = ext4_get_group_desc(sb, group, NULL); 3666 if (!gdp) { 3667 ret = 1; 3668 break; 3669 } 3670 3671 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3672 break; 3673 } 3674 3675 if (group >= ngroups) 3676 ret = 1; 3677 3678 if (!ret) { 3679 start_time = ktime_get_real_ns(); 3680 ret = ext4_init_inode_table(sb, group, 3681 elr->lr_timeout ? 0 : 1); 3682 trace_ext4_lazy_itable_init(sb, group); 3683 if (elr->lr_timeout == 0) { 3684 elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) * 3685 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3686 } 3687 elr->lr_next_sched = jiffies + elr->lr_timeout; 3688 elr->lr_next_group = group + 1; 3689 } 3690 return ret; 3691 } 3692 3693 /* 3694 * Remove lr_request from the list_request and free the 3695 * request structure. Should be called with li_list_mtx held 3696 */ 3697 static void ext4_remove_li_request(struct ext4_li_request *elr) 3698 { 3699 if (!elr) 3700 return; 3701 3702 list_del(&elr->lr_request); 3703 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3704 kfree(elr); 3705 } 3706 3707 static void ext4_unregister_li_request(struct super_block *sb) 3708 { 3709 mutex_lock(&ext4_li_mtx); 3710 if (!ext4_li_info) { 3711 mutex_unlock(&ext4_li_mtx); 3712 return; 3713 } 3714 3715 mutex_lock(&ext4_li_info->li_list_mtx); 3716 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3717 mutex_unlock(&ext4_li_info->li_list_mtx); 3718 mutex_unlock(&ext4_li_mtx); 3719 } 3720 3721 static struct task_struct *ext4_lazyinit_task; 3722 3723 /* 3724 * This is the function where ext4lazyinit thread lives. It walks 3725 * through the request list searching for next scheduled filesystem. 3726 * When such a fs is found, run the lazy initialization request 3727 * (ext4_rn_li_request) and keep track of the time spend in this 3728 * function. Based on that time we compute next schedule time of 3729 * the request. When walking through the list is complete, compute 3730 * next waking time and put itself into sleep. 3731 */ 3732 static int ext4_lazyinit_thread(void *arg) 3733 { 3734 struct ext4_lazy_init *eli = arg; 3735 struct list_head *pos, *n; 3736 struct ext4_li_request *elr; 3737 unsigned long next_wakeup, cur; 3738 3739 BUG_ON(NULL == eli); 3740 set_freezable(); 3741 3742 cont_thread: 3743 while (true) { 3744 next_wakeup = MAX_JIFFY_OFFSET; 3745 3746 mutex_lock(&eli->li_list_mtx); 3747 if (list_empty(&eli->li_request_list)) { 3748 mutex_unlock(&eli->li_list_mtx); 3749 goto exit_thread; 3750 } 3751 list_for_each_safe(pos, n, &eli->li_request_list) { 3752 int err = 0; 3753 int progress = 0; 3754 elr = list_entry(pos, struct ext4_li_request, 3755 lr_request); 3756 3757 if (time_before(jiffies, elr->lr_next_sched)) { 3758 if (time_before(elr->lr_next_sched, next_wakeup)) 3759 next_wakeup = elr->lr_next_sched; 3760 continue; 3761 } 3762 if (down_read_trylock(&elr->lr_super->s_umount)) { 3763 if (sb_start_write_trylock(elr->lr_super)) { 3764 progress = 1; 3765 /* 3766 * We hold sb->s_umount, sb can not 3767 * be removed from the list, it is 3768 * now safe to drop li_list_mtx 3769 */ 3770 mutex_unlock(&eli->li_list_mtx); 3771 err = ext4_run_li_request(elr); 3772 sb_end_write(elr->lr_super); 3773 mutex_lock(&eli->li_list_mtx); 3774 n = pos->next; 3775 } 3776 up_read((&elr->lr_super->s_umount)); 3777 } 3778 /* error, remove the lazy_init job */ 3779 if (err) { 3780 ext4_remove_li_request(elr); 3781 continue; 3782 } 3783 if (!progress) { 3784 elr->lr_next_sched = jiffies + 3785 prandom_u32_max(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3786 } 3787 if (time_before(elr->lr_next_sched, next_wakeup)) 3788 next_wakeup = elr->lr_next_sched; 3789 } 3790 mutex_unlock(&eli->li_list_mtx); 3791 3792 try_to_freeze(); 3793 3794 cur = jiffies; 3795 if ((time_after_eq(cur, next_wakeup)) || 3796 (MAX_JIFFY_OFFSET == next_wakeup)) { 3797 cond_resched(); 3798 continue; 3799 } 3800 3801 schedule_timeout_interruptible(next_wakeup - cur); 3802 3803 if (kthread_should_stop()) { 3804 ext4_clear_request_list(); 3805 goto exit_thread; 3806 } 3807 } 3808 3809 exit_thread: 3810 /* 3811 * It looks like the request list is empty, but we need 3812 * to check it under the li_list_mtx lock, to prevent any 3813 * additions into it, and of course we should lock ext4_li_mtx 3814 * to atomically free the list and ext4_li_info, because at 3815 * this point another ext4 filesystem could be registering 3816 * new one. 3817 */ 3818 mutex_lock(&ext4_li_mtx); 3819 mutex_lock(&eli->li_list_mtx); 3820 if (!list_empty(&eli->li_request_list)) { 3821 mutex_unlock(&eli->li_list_mtx); 3822 mutex_unlock(&ext4_li_mtx); 3823 goto cont_thread; 3824 } 3825 mutex_unlock(&eli->li_list_mtx); 3826 kfree(ext4_li_info); 3827 ext4_li_info = NULL; 3828 mutex_unlock(&ext4_li_mtx); 3829 3830 return 0; 3831 } 3832 3833 static void ext4_clear_request_list(void) 3834 { 3835 struct list_head *pos, *n; 3836 struct ext4_li_request *elr; 3837 3838 mutex_lock(&ext4_li_info->li_list_mtx); 3839 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3840 elr = list_entry(pos, struct ext4_li_request, 3841 lr_request); 3842 ext4_remove_li_request(elr); 3843 } 3844 mutex_unlock(&ext4_li_info->li_list_mtx); 3845 } 3846 3847 static int ext4_run_lazyinit_thread(void) 3848 { 3849 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3850 ext4_li_info, "ext4lazyinit"); 3851 if (IS_ERR(ext4_lazyinit_task)) { 3852 int err = PTR_ERR(ext4_lazyinit_task); 3853 ext4_clear_request_list(); 3854 kfree(ext4_li_info); 3855 ext4_li_info = NULL; 3856 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3857 "initialization thread\n", 3858 err); 3859 return err; 3860 } 3861 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3862 return 0; 3863 } 3864 3865 /* 3866 * Check whether it make sense to run itable init. thread or not. 3867 * If there is at least one uninitialized inode table, return 3868 * corresponding group number, else the loop goes through all 3869 * groups and return total number of groups. 3870 */ 3871 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3872 { 3873 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3874 struct ext4_group_desc *gdp = NULL; 3875 3876 if (!ext4_has_group_desc_csum(sb)) 3877 return ngroups; 3878 3879 for (group = 0; group < ngroups; group++) { 3880 gdp = ext4_get_group_desc(sb, group, NULL); 3881 if (!gdp) 3882 continue; 3883 3884 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3885 break; 3886 } 3887 3888 return group; 3889 } 3890 3891 static int ext4_li_info_new(void) 3892 { 3893 struct ext4_lazy_init *eli = NULL; 3894 3895 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3896 if (!eli) 3897 return -ENOMEM; 3898 3899 INIT_LIST_HEAD(&eli->li_request_list); 3900 mutex_init(&eli->li_list_mtx); 3901 3902 eli->li_state |= EXT4_LAZYINIT_QUIT; 3903 3904 ext4_li_info = eli; 3905 3906 return 0; 3907 } 3908 3909 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3910 ext4_group_t start) 3911 { 3912 struct ext4_li_request *elr; 3913 3914 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3915 if (!elr) 3916 return NULL; 3917 3918 elr->lr_super = sb; 3919 elr->lr_first_not_zeroed = start; 3920 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3921 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3922 elr->lr_next_group = start; 3923 } else { 3924 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3925 } 3926 3927 /* 3928 * Randomize first schedule time of the request to 3929 * spread the inode table initialization requests 3930 * better. 3931 */ 3932 elr->lr_next_sched = jiffies + prandom_u32_max( 3933 EXT4_DEF_LI_MAX_START_DELAY * HZ); 3934 return elr; 3935 } 3936 3937 int ext4_register_li_request(struct super_block *sb, 3938 ext4_group_t first_not_zeroed) 3939 { 3940 struct ext4_sb_info *sbi = EXT4_SB(sb); 3941 struct ext4_li_request *elr = NULL; 3942 ext4_group_t ngroups = sbi->s_groups_count; 3943 int ret = 0; 3944 3945 mutex_lock(&ext4_li_mtx); 3946 if (sbi->s_li_request != NULL) { 3947 /* 3948 * Reset timeout so it can be computed again, because 3949 * s_li_wait_mult might have changed. 3950 */ 3951 sbi->s_li_request->lr_timeout = 0; 3952 goto out; 3953 } 3954 3955 if (sb_rdonly(sb) || 3956 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 3957 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 3958 goto out; 3959 3960 elr = ext4_li_request_new(sb, first_not_zeroed); 3961 if (!elr) { 3962 ret = -ENOMEM; 3963 goto out; 3964 } 3965 3966 if (NULL == ext4_li_info) { 3967 ret = ext4_li_info_new(); 3968 if (ret) 3969 goto out; 3970 } 3971 3972 mutex_lock(&ext4_li_info->li_list_mtx); 3973 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3974 mutex_unlock(&ext4_li_info->li_list_mtx); 3975 3976 sbi->s_li_request = elr; 3977 /* 3978 * set elr to NULL here since it has been inserted to 3979 * the request_list and the removal and free of it is 3980 * handled by ext4_clear_request_list from now on. 3981 */ 3982 elr = NULL; 3983 3984 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3985 ret = ext4_run_lazyinit_thread(); 3986 if (ret) 3987 goto out; 3988 } 3989 out: 3990 mutex_unlock(&ext4_li_mtx); 3991 if (ret) 3992 kfree(elr); 3993 return ret; 3994 } 3995 3996 /* 3997 * We do not need to lock anything since this is called on 3998 * module unload. 3999 */ 4000 static void ext4_destroy_lazyinit_thread(void) 4001 { 4002 /* 4003 * If thread exited earlier 4004 * there's nothing to be done. 4005 */ 4006 if (!ext4_li_info || !ext4_lazyinit_task) 4007 return; 4008 4009 kthread_stop(ext4_lazyinit_task); 4010 } 4011 4012 static int set_journal_csum_feature_set(struct super_block *sb) 4013 { 4014 int ret = 1; 4015 int compat, incompat; 4016 struct ext4_sb_info *sbi = EXT4_SB(sb); 4017 4018 if (ext4_has_metadata_csum(sb)) { 4019 /* journal checksum v3 */ 4020 compat = 0; 4021 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4022 } else { 4023 /* journal checksum v1 */ 4024 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4025 incompat = 0; 4026 } 4027 4028 jbd2_journal_clear_features(sbi->s_journal, 4029 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4030 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4031 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4032 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4033 ret = jbd2_journal_set_features(sbi->s_journal, 4034 compat, 0, 4035 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4036 incompat); 4037 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4038 ret = jbd2_journal_set_features(sbi->s_journal, 4039 compat, 0, 4040 incompat); 4041 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4042 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4043 } else { 4044 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4045 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4046 } 4047 4048 return ret; 4049 } 4050 4051 /* 4052 * Note: calculating the overhead so we can be compatible with 4053 * historical BSD practice is quite difficult in the face of 4054 * clusters/bigalloc. This is because multiple metadata blocks from 4055 * different block group can end up in the same allocation cluster. 4056 * Calculating the exact overhead in the face of clustered allocation 4057 * requires either O(all block bitmaps) in memory or O(number of block 4058 * groups**2) in time. We will still calculate the superblock for 4059 * older file systems --- and if we come across with a bigalloc file 4060 * system with zero in s_overhead_clusters the estimate will be close to 4061 * correct especially for very large cluster sizes --- but for newer 4062 * file systems, it's better to calculate this figure once at mkfs 4063 * time, and store it in the superblock. If the superblock value is 4064 * present (even for non-bigalloc file systems), we will use it. 4065 */ 4066 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4067 char *buf) 4068 { 4069 struct ext4_sb_info *sbi = EXT4_SB(sb); 4070 struct ext4_group_desc *gdp; 4071 ext4_fsblk_t first_block, last_block, b; 4072 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4073 int s, j, count = 0; 4074 int has_super = ext4_bg_has_super(sb, grp); 4075 4076 if (!ext4_has_feature_bigalloc(sb)) 4077 return (has_super + ext4_bg_num_gdb(sb, grp) + 4078 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4079 sbi->s_itb_per_group + 2); 4080 4081 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4082 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4083 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4084 for (i = 0; i < ngroups; i++) { 4085 gdp = ext4_get_group_desc(sb, i, NULL); 4086 b = ext4_block_bitmap(sb, gdp); 4087 if (b >= first_block && b <= last_block) { 4088 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4089 count++; 4090 } 4091 b = ext4_inode_bitmap(sb, gdp); 4092 if (b >= first_block && b <= last_block) { 4093 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4094 count++; 4095 } 4096 b = ext4_inode_table(sb, gdp); 4097 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4098 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4099 int c = EXT4_B2C(sbi, b - first_block); 4100 ext4_set_bit(c, buf); 4101 count++; 4102 } 4103 if (i != grp) 4104 continue; 4105 s = 0; 4106 if (ext4_bg_has_super(sb, grp)) { 4107 ext4_set_bit(s++, buf); 4108 count++; 4109 } 4110 j = ext4_bg_num_gdb(sb, grp); 4111 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4112 ext4_error(sb, "Invalid number of block group " 4113 "descriptor blocks: %d", j); 4114 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4115 } 4116 count += j; 4117 for (; j > 0; j--) 4118 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4119 } 4120 if (!count) 4121 return 0; 4122 return EXT4_CLUSTERS_PER_GROUP(sb) - 4123 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4124 } 4125 4126 /* 4127 * Compute the overhead and stash it in sbi->s_overhead 4128 */ 4129 int ext4_calculate_overhead(struct super_block *sb) 4130 { 4131 struct ext4_sb_info *sbi = EXT4_SB(sb); 4132 struct ext4_super_block *es = sbi->s_es; 4133 struct inode *j_inode; 4134 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4135 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4136 ext4_fsblk_t overhead = 0; 4137 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4138 4139 if (!buf) 4140 return -ENOMEM; 4141 4142 /* 4143 * Compute the overhead (FS structures). This is constant 4144 * for a given filesystem unless the number of block groups 4145 * changes so we cache the previous value until it does. 4146 */ 4147 4148 /* 4149 * All of the blocks before first_data_block are overhead 4150 */ 4151 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4152 4153 /* 4154 * Add the overhead found in each block group 4155 */ 4156 for (i = 0; i < ngroups; i++) { 4157 int blks; 4158 4159 blks = count_overhead(sb, i, buf); 4160 overhead += blks; 4161 if (blks) 4162 memset(buf, 0, PAGE_SIZE); 4163 cond_resched(); 4164 } 4165 4166 /* 4167 * Add the internal journal blocks whether the journal has been 4168 * loaded or not 4169 */ 4170 if (sbi->s_journal && !sbi->s_journal_bdev) 4171 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4172 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4173 /* j_inum for internal journal is non-zero */ 4174 j_inode = ext4_get_journal_inode(sb, j_inum); 4175 if (j_inode) { 4176 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4177 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4178 iput(j_inode); 4179 } else { 4180 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4181 } 4182 } 4183 sbi->s_overhead = overhead; 4184 smp_wmb(); 4185 free_page((unsigned long) buf); 4186 return 0; 4187 } 4188 4189 static void ext4_set_resv_clusters(struct super_block *sb) 4190 { 4191 ext4_fsblk_t resv_clusters; 4192 struct ext4_sb_info *sbi = EXT4_SB(sb); 4193 4194 /* 4195 * There's no need to reserve anything when we aren't using extents. 4196 * The space estimates are exact, there are no unwritten extents, 4197 * hole punching doesn't need new metadata... This is needed especially 4198 * to keep ext2/3 backward compatibility. 4199 */ 4200 if (!ext4_has_feature_extents(sb)) 4201 return; 4202 /* 4203 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4204 * This should cover the situations where we can not afford to run 4205 * out of space like for example punch hole, or converting 4206 * unwritten extents in delalloc path. In most cases such 4207 * allocation would require 1, or 2 blocks, higher numbers are 4208 * very rare. 4209 */ 4210 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4211 sbi->s_cluster_bits); 4212 4213 do_div(resv_clusters, 50); 4214 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4215 4216 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4217 } 4218 4219 static const char *ext4_quota_mode(struct super_block *sb) 4220 { 4221 #ifdef CONFIG_QUOTA 4222 if (!ext4_quota_capable(sb)) 4223 return "none"; 4224 4225 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4226 return "journalled"; 4227 else 4228 return "writeback"; 4229 #else 4230 return "disabled"; 4231 #endif 4232 } 4233 4234 static void ext4_setup_csum_trigger(struct super_block *sb, 4235 enum ext4_journal_trigger_type type, 4236 void (*trigger)( 4237 struct jbd2_buffer_trigger_type *type, 4238 struct buffer_head *bh, 4239 void *mapped_data, 4240 size_t size)) 4241 { 4242 struct ext4_sb_info *sbi = EXT4_SB(sb); 4243 4244 sbi->s_journal_triggers[type].sb = sb; 4245 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4246 } 4247 4248 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4249 { 4250 if (!sbi) 4251 return; 4252 4253 kfree(sbi->s_blockgroup_lock); 4254 fs_put_dax(sbi->s_daxdev, NULL); 4255 kfree(sbi); 4256 } 4257 4258 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4259 { 4260 struct ext4_sb_info *sbi; 4261 4262 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4263 if (!sbi) 4264 return NULL; 4265 4266 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4267 NULL, NULL); 4268 4269 sbi->s_blockgroup_lock = 4270 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4271 4272 if (!sbi->s_blockgroup_lock) 4273 goto err_out; 4274 4275 sb->s_fs_info = sbi; 4276 sbi->s_sb = sb; 4277 return sbi; 4278 err_out: 4279 fs_put_dax(sbi->s_daxdev, NULL); 4280 kfree(sbi); 4281 return NULL; 4282 } 4283 4284 static void ext4_set_def_opts(struct super_block *sb, 4285 struct ext4_super_block *es) 4286 { 4287 unsigned long def_mount_opts; 4288 4289 /* Set defaults before we parse the mount options */ 4290 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4291 set_opt(sb, INIT_INODE_TABLE); 4292 if (def_mount_opts & EXT4_DEFM_DEBUG) 4293 set_opt(sb, DEBUG); 4294 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4295 set_opt(sb, GRPID); 4296 if (def_mount_opts & EXT4_DEFM_UID16) 4297 set_opt(sb, NO_UID32); 4298 /* xattr user namespace & acls are now defaulted on */ 4299 set_opt(sb, XATTR_USER); 4300 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4301 set_opt(sb, POSIX_ACL); 4302 #endif 4303 if (ext4_has_feature_fast_commit(sb)) 4304 set_opt2(sb, JOURNAL_FAST_COMMIT); 4305 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4306 if (ext4_has_metadata_csum(sb)) 4307 set_opt(sb, JOURNAL_CHECKSUM); 4308 4309 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4310 set_opt(sb, JOURNAL_DATA); 4311 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4312 set_opt(sb, ORDERED_DATA); 4313 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4314 set_opt(sb, WRITEBACK_DATA); 4315 4316 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4317 set_opt(sb, ERRORS_PANIC); 4318 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4319 set_opt(sb, ERRORS_CONT); 4320 else 4321 set_opt(sb, ERRORS_RO); 4322 /* block_validity enabled by default; disable with noblock_validity */ 4323 set_opt(sb, BLOCK_VALIDITY); 4324 if (def_mount_opts & EXT4_DEFM_DISCARD) 4325 set_opt(sb, DISCARD); 4326 4327 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4328 set_opt(sb, BARRIER); 4329 4330 /* 4331 * enable delayed allocation by default 4332 * Use -o nodelalloc to turn it off 4333 */ 4334 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4335 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4336 set_opt(sb, DELALLOC); 4337 4338 if (sb->s_blocksize == PAGE_SIZE) 4339 set_opt(sb, DIOREAD_NOLOCK); 4340 } 4341 4342 static int ext4_handle_clustersize(struct super_block *sb) 4343 { 4344 struct ext4_sb_info *sbi = EXT4_SB(sb); 4345 struct ext4_super_block *es = sbi->s_es; 4346 int clustersize; 4347 4348 /* Handle clustersize */ 4349 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4350 if (ext4_has_feature_bigalloc(sb)) { 4351 if (clustersize < sb->s_blocksize) { 4352 ext4_msg(sb, KERN_ERR, 4353 "cluster size (%d) smaller than " 4354 "block size (%lu)", clustersize, sb->s_blocksize); 4355 return -EINVAL; 4356 } 4357 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4358 le32_to_cpu(es->s_log_block_size); 4359 sbi->s_clusters_per_group = 4360 le32_to_cpu(es->s_clusters_per_group); 4361 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4362 ext4_msg(sb, KERN_ERR, 4363 "#clusters per group too big: %lu", 4364 sbi->s_clusters_per_group); 4365 return -EINVAL; 4366 } 4367 if (sbi->s_blocks_per_group != 4368 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4369 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4370 "clusters per group (%lu) inconsistent", 4371 sbi->s_blocks_per_group, 4372 sbi->s_clusters_per_group); 4373 return -EINVAL; 4374 } 4375 } else { 4376 if (clustersize != sb->s_blocksize) { 4377 ext4_msg(sb, KERN_ERR, 4378 "fragment/cluster size (%d) != " 4379 "block size (%lu)", clustersize, sb->s_blocksize); 4380 return -EINVAL; 4381 } 4382 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4383 ext4_msg(sb, KERN_ERR, 4384 "#blocks per group too big: %lu", 4385 sbi->s_blocks_per_group); 4386 return -EINVAL; 4387 } 4388 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4389 sbi->s_cluster_bits = 0; 4390 } 4391 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4392 4393 /* Do we have standard group size of clustersize * 8 blocks ? */ 4394 if (sbi->s_blocks_per_group == clustersize << 3) 4395 set_opt2(sb, STD_GROUP_SIZE); 4396 4397 return 0; 4398 } 4399 4400 static void ext4_fast_commit_init(struct super_block *sb) 4401 { 4402 struct ext4_sb_info *sbi = EXT4_SB(sb); 4403 4404 /* Initialize fast commit stuff */ 4405 atomic_set(&sbi->s_fc_subtid, 0); 4406 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4407 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4408 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4409 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4410 sbi->s_fc_bytes = 0; 4411 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4412 sbi->s_fc_ineligible_tid = 0; 4413 spin_lock_init(&sbi->s_fc_lock); 4414 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4415 sbi->s_fc_replay_state.fc_regions = NULL; 4416 sbi->s_fc_replay_state.fc_regions_size = 0; 4417 sbi->s_fc_replay_state.fc_regions_used = 0; 4418 sbi->s_fc_replay_state.fc_regions_valid = 0; 4419 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4420 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4421 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4422 } 4423 4424 static int ext4_inode_info_init(struct super_block *sb, 4425 struct ext4_super_block *es) 4426 { 4427 struct ext4_sb_info *sbi = EXT4_SB(sb); 4428 4429 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4430 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4431 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4432 } else { 4433 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4434 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4435 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4436 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4437 sbi->s_first_ino); 4438 return -EINVAL; 4439 } 4440 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4441 (!is_power_of_2(sbi->s_inode_size)) || 4442 (sbi->s_inode_size > sb->s_blocksize)) { 4443 ext4_msg(sb, KERN_ERR, 4444 "unsupported inode size: %d", 4445 sbi->s_inode_size); 4446 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4447 return -EINVAL; 4448 } 4449 /* 4450 * i_atime_extra is the last extra field available for 4451 * [acm]times in struct ext4_inode. Checking for that 4452 * field should suffice to ensure we have extra space 4453 * for all three. 4454 */ 4455 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4456 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4457 sb->s_time_gran = 1; 4458 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4459 } else { 4460 sb->s_time_gran = NSEC_PER_SEC; 4461 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4462 } 4463 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4464 } 4465 4466 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4467 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4468 EXT4_GOOD_OLD_INODE_SIZE; 4469 if (ext4_has_feature_extra_isize(sb)) { 4470 unsigned v, max = (sbi->s_inode_size - 4471 EXT4_GOOD_OLD_INODE_SIZE); 4472 4473 v = le16_to_cpu(es->s_want_extra_isize); 4474 if (v > max) { 4475 ext4_msg(sb, KERN_ERR, 4476 "bad s_want_extra_isize: %d", v); 4477 return -EINVAL; 4478 } 4479 if (sbi->s_want_extra_isize < v) 4480 sbi->s_want_extra_isize = v; 4481 4482 v = le16_to_cpu(es->s_min_extra_isize); 4483 if (v > max) { 4484 ext4_msg(sb, KERN_ERR, 4485 "bad s_min_extra_isize: %d", v); 4486 return -EINVAL; 4487 } 4488 if (sbi->s_want_extra_isize < v) 4489 sbi->s_want_extra_isize = v; 4490 } 4491 } 4492 4493 return 0; 4494 } 4495 4496 #if IS_ENABLED(CONFIG_UNICODE) 4497 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4498 { 4499 const struct ext4_sb_encodings *encoding_info; 4500 struct unicode_map *encoding; 4501 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4502 4503 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4504 return 0; 4505 4506 encoding_info = ext4_sb_read_encoding(es); 4507 if (!encoding_info) { 4508 ext4_msg(sb, KERN_ERR, 4509 "Encoding requested by superblock is unknown"); 4510 return -EINVAL; 4511 } 4512 4513 encoding = utf8_load(encoding_info->version); 4514 if (IS_ERR(encoding)) { 4515 ext4_msg(sb, KERN_ERR, 4516 "can't mount with superblock charset: %s-%u.%u.%u " 4517 "not supported by the kernel. flags: 0x%x.", 4518 encoding_info->name, 4519 unicode_major(encoding_info->version), 4520 unicode_minor(encoding_info->version), 4521 unicode_rev(encoding_info->version), 4522 encoding_flags); 4523 return -EINVAL; 4524 } 4525 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4526 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4527 unicode_major(encoding_info->version), 4528 unicode_minor(encoding_info->version), 4529 unicode_rev(encoding_info->version), 4530 encoding_flags); 4531 4532 sb->s_encoding = encoding; 4533 sb->s_encoding_flags = encoding_flags; 4534 4535 return 0; 4536 } 4537 #else 4538 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4539 { 4540 return 0; 4541 } 4542 #endif 4543 4544 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4545 { 4546 struct ext4_sb_info *sbi = EXT4_SB(sb); 4547 4548 /* Warn if metadata_csum and gdt_csum are both set. */ 4549 if (ext4_has_feature_metadata_csum(sb) && 4550 ext4_has_feature_gdt_csum(sb)) 4551 ext4_warning(sb, "metadata_csum and uninit_bg are " 4552 "redundant flags; please run fsck."); 4553 4554 /* Check for a known checksum algorithm */ 4555 if (!ext4_verify_csum_type(sb, es)) { 4556 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4557 "unknown checksum algorithm."); 4558 return -EINVAL; 4559 } 4560 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4561 ext4_orphan_file_block_trigger); 4562 4563 /* Load the checksum driver */ 4564 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4565 if (IS_ERR(sbi->s_chksum_driver)) { 4566 int ret = PTR_ERR(sbi->s_chksum_driver); 4567 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4568 sbi->s_chksum_driver = NULL; 4569 return ret; 4570 } 4571 4572 /* Check superblock checksum */ 4573 if (!ext4_superblock_csum_verify(sb, es)) { 4574 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4575 "invalid superblock checksum. Run e2fsck?"); 4576 return -EFSBADCRC; 4577 } 4578 4579 /* Precompute checksum seed for all metadata */ 4580 if (ext4_has_feature_csum_seed(sb)) 4581 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4582 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4583 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4584 sizeof(es->s_uuid)); 4585 return 0; 4586 } 4587 4588 static int ext4_check_feature_compatibility(struct super_block *sb, 4589 struct ext4_super_block *es, 4590 int silent) 4591 { 4592 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4593 (ext4_has_compat_features(sb) || 4594 ext4_has_ro_compat_features(sb) || 4595 ext4_has_incompat_features(sb))) 4596 ext4_msg(sb, KERN_WARNING, 4597 "feature flags set on rev 0 fs, " 4598 "running e2fsck is recommended"); 4599 4600 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4601 set_opt2(sb, HURD_COMPAT); 4602 if (ext4_has_feature_64bit(sb)) { 4603 ext4_msg(sb, KERN_ERR, 4604 "The Hurd can't support 64-bit file systems"); 4605 return -EINVAL; 4606 } 4607 4608 /* 4609 * ea_inode feature uses l_i_version field which is not 4610 * available in HURD_COMPAT mode. 4611 */ 4612 if (ext4_has_feature_ea_inode(sb)) { 4613 ext4_msg(sb, KERN_ERR, 4614 "ea_inode feature is not supported for Hurd"); 4615 return -EINVAL; 4616 } 4617 } 4618 4619 if (IS_EXT2_SB(sb)) { 4620 if (ext2_feature_set_ok(sb)) 4621 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4622 "using the ext4 subsystem"); 4623 else { 4624 /* 4625 * If we're probing be silent, if this looks like 4626 * it's actually an ext[34] filesystem. 4627 */ 4628 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4629 return -EINVAL; 4630 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4631 "to feature incompatibilities"); 4632 return -EINVAL; 4633 } 4634 } 4635 4636 if (IS_EXT3_SB(sb)) { 4637 if (ext3_feature_set_ok(sb)) 4638 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4639 "using the ext4 subsystem"); 4640 else { 4641 /* 4642 * If we're probing be silent, if this looks like 4643 * it's actually an ext4 filesystem. 4644 */ 4645 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4646 return -EINVAL; 4647 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4648 "to feature incompatibilities"); 4649 return -EINVAL; 4650 } 4651 } 4652 4653 /* 4654 * Check feature flags regardless of the revision level, since we 4655 * previously didn't change the revision level when setting the flags, 4656 * so there is a chance incompat flags are set on a rev 0 filesystem. 4657 */ 4658 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4659 return -EINVAL; 4660 4661 return 0; 4662 } 4663 4664 static int ext4_geometry_check(struct super_block *sb, 4665 struct ext4_super_block *es) 4666 { 4667 struct ext4_sb_info *sbi = EXT4_SB(sb); 4668 __u64 blocks_count; 4669 4670 /* check blocks count against device size */ 4671 blocks_count = sb_bdev_nr_blocks(sb); 4672 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4673 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4674 "exceeds size of device (%llu blocks)", 4675 ext4_blocks_count(es), blocks_count); 4676 return -EINVAL; 4677 } 4678 4679 /* 4680 * It makes no sense for the first data block to be beyond the end 4681 * of the filesystem. 4682 */ 4683 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4684 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4685 "block %u is beyond end of filesystem (%llu)", 4686 le32_to_cpu(es->s_first_data_block), 4687 ext4_blocks_count(es)); 4688 return -EINVAL; 4689 } 4690 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4691 (sbi->s_cluster_ratio == 1)) { 4692 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4693 "block is 0 with a 1k block and cluster size"); 4694 return -EINVAL; 4695 } 4696 4697 blocks_count = (ext4_blocks_count(es) - 4698 le32_to_cpu(es->s_first_data_block) + 4699 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4700 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4701 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4702 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4703 "(block count %llu, first data block %u, " 4704 "blocks per group %lu)", blocks_count, 4705 ext4_blocks_count(es), 4706 le32_to_cpu(es->s_first_data_block), 4707 EXT4_BLOCKS_PER_GROUP(sb)); 4708 return -EINVAL; 4709 } 4710 sbi->s_groups_count = blocks_count; 4711 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4712 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4713 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4714 le32_to_cpu(es->s_inodes_count)) { 4715 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4716 le32_to_cpu(es->s_inodes_count), 4717 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4718 return -EINVAL; 4719 } 4720 4721 return 0; 4722 } 4723 4724 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 4725 { 4726 struct buffer_head **group_desc; 4727 int i; 4728 4729 rcu_read_lock(); 4730 group_desc = rcu_dereference(sbi->s_group_desc); 4731 for (i = 0; i < sbi->s_gdb_count; i++) 4732 brelse(group_desc[i]); 4733 kvfree(group_desc); 4734 rcu_read_unlock(); 4735 } 4736 4737 static int ext4_group_desc_init(struct super_block *sb, 4738 struct ext4_super_block *es, 4739 ext4_fsblk_t logical_sb_block, 4740 ext4_group_t *first_not_zeroed) 4741 { 4742 struct ext4_sb_info *sbi = EXT4_SB(sb); 4743 unsigned int db_count; 4744 ext4_fsblk_t block; 4745 int ret; 4746 int i; 4747 4748 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4749 EXT4_DESC_PER_BLOCK(sb); 4750 if (ext4_has_feature_meta_bg(sb)) { 4751 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4752 ext4_msg(sb, KERN_WARNING, 4753 "first meta block group too large: %u " 4754 "(group descriptor block count %u)", 4755 le32_to_cpu(es->s_first_meta_bg), db_count); 4756 return -EINVAL; 4757 } 4758 } 4759 rcu_assign_pointer(sbi->s_group_desc, 4760 kvmalloc_array(db_count, 4761 sizeof(struct buffer_head *), 4762 GFP_KERNEL)); 4763 if (sbi->s_group_desc == NULL) { 4764 ext4_msg(sb, KERN_ERR, "not enough memory"); 4765 return -ENOMEM; 4766 } 4767 4768 bgl_lock_init(sbi->s_blockgroup_lock); 4769 4770 /* Pre-read the descriptors into the buffer cache */ 4771 for (i = 0; i < db_count; i++) { 4772 block = descriptor_loc(sb, logical_sb_block, i); 4773 ext4_sb_breadahead_unmovable(sb, block); 4774 } 4775 4776 for (i = 0; i < db_count; i++) { 4777 struct buffer_head *bh; 4778 4779 block = descriptor_loc(sb, logical_sb_block, i); 4780 bh = ext4_sb_bread_unmovable(sb, block); 4781 if (IS_ERR(bh)) { 4782 ext4_msg(sb, KERN_ERR, 4783 "can't read group descriptor %d", i); 4784 sbi->s_gdb_count = i; 4785 ret = PTR_ERR(bh); 4786 goto out; 4787 } 4788 rcu_read_lock(); 4789 rcu_dereference(sbi->s_group_desc)[i] = bh; 4790 rcu_read_unlock(); 4791 } 4792 sbi->s_gdb_count = db_count; 4793 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4794 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4795 ret = -EFSCORRUPTED; 4796 goto out; 4797 } 4798 return 0; 4799 out: 4800 ext4_group_desc_free(sbi); 4801 return ret; 4802 } 4803 4804 static int ext4_load_and_init_journal(struct super_block *sb, 4805 struct ext4_super_block *es, 4806 struct ext4_fs_context *ctx) 4807 { 4808 struct ext4_sb_info *sbi = EXT4_SB(sb); 4809 int err; 4810 4811 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4812 if (err) 4813 return err; 4814 4815 if (ext4_has_feature_64bit(sb) && 4816 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4817 JBD2_FEATURE_INCOMPAT_64BIT)) { 4818 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4819 goto out; 4820 } 4821 4822 if (!set_journal_csum_feature_set(sb)) { 4823 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4824 "feature set"); 4825 goto out; 4826 } 4827 4828 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4829 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4830 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4831 ext4_msg(sb, KERN_ERR, 4832 "Failed to set fast commit journal feature"); 4833 goto out; 4834 } 4835 4836 /* We have now updated the journal if required, so we can 4837 * validate the data journaling mode. */ 4838 switch (test_opt(sb, DATA_FLAGS)) { 4839 case 0: 4840 /* No mode set, assume a default based on the journal 4841 * capabilities: ORDERED_DATA if the journal can 4842 * cope, else JOURNAL_DATA 4843 */ 4844 if (jbd2_journal_check_available_features 4845 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4846 set_opt(sb, ORDERED_DATA); 4847 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4848 } else { 4849 set_opt(sb, JOURNAL_DATA); 4850 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4851 } 4852 break; 4853 4854 case EXT4_MOUNT_ORDERED_DATA: 4855 case EXT4_MOUNT_WRITEBACK_DATA: 4856 if (!jbd2_journal_check_available_features 4857 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4858 ext4_msg(sb, KERN_ERR, "Journal does not support " 4859 "requested data journaling mode"); 4860 goto out; 4861 } 4862 break; 4863 default: 4864 break; 4865 } 4866 4867 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4868 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4869 ext4_msg(sb, KERN_ERR, "can't mount with " 4870 "journal_async_commit in data=ordered mode"); 4871 goto out; 4872 } 4873 4874 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4875 4876 sbi->s_journal->j_submit_inode_data_buffers = 4877 ext4_journal_submit_inode_data_buffers; 4878 sbi->s_journal->j_finish_inode_data_buffers = 4879 ext4_journal_finish_inode_data_buffers; 4880 4881 return 0; 4882 4883 out: 4884 /* flush s_error_work before journal destroy. */ 4885 flush_work(&sbi->s_error_work); 4886 jbd2_journal_destroy(sbi->s_journal); 4887 sbi->s_journal = NULL; 4888 return err; 4889 } 4890 4891 static int ext4_journal_data_mode_check(struct super_block *sb) 4892 { 4893 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4894 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4895 "data=journal disables delayed allocation, " 4896 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 4897 /* can't mount with both data=journal and dioread_nolock. */ 4898 clear_opt(sb, DIOREAD_NOLOCK); 4899 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4900 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4901 ext4_msg(sb, KERN_ERR, "can't mount with " 4902 "both data=journal and delalloc"); 4903 return -EINVAL; 4904 } 4905 if (test_opt(sb, DAX_ALWAYS)) { 4906 ext4_msg(sb, KERN_ERR, "can't mount with " 4907 "both data=journal and dax"); 4908 return -EINVAL; 4909 } 4910 if (ext4_has_feature_encrypt(sb)) { 4911 ext4_msg(sb, KERN_WARNING, 4912 "encrypted files will use data=ordered " 4913 "instead of data journaling mode"); 4914 } 4915 if (test_opt(sb, DELALLOC)) 4916 clear_opt(sb, DELALLOC); 4917 } else { 4918 sb->s_iflags |= SB_I_CGROUPWB; 4919 } 4920 4921 return 0; 4922 } 4923 4924 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 4925 int silent) 4926 { 4927 struct ext4_sb_info *sbi = EXT4_SB(sb); 4928 struct ext4_super_block *es; 4929 ext4_fsblk_t logical_sb_block; 4930 unsigned long offset = 0; 4931 struct buffer_head *bh; 4932 int ret = -EINVAL; 4933 int blocksize; 4934 4935 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4936 if (!blocksize) { 4937 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4938 return -EINVAL; 4939 } 4940 4941 /* 4942 * The ext4 superblock will not be buffer aligned for other than 1kB 4943 * block sizes. We need to calculate the offset from buffer start. 4944 */ 4945 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4946 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 4947 offset = do_div(logical_sb_block, blocksize); 4948 } else { 4949 logical_sb_block = sbi->s_sb_block; 4950 } 4951 4952 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4953 if (IS_ERR(bh)) { 4954 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4955 return PTR_ERR(bh); 4956 } 4957 /* 4958 * Note: s_es must be initialized as soon as possible because 4959 * some ext4 macro-instructions depend on its value 4960 */ 4961 es = (struct ext4_super_block *) (bh->b_data + offset); 4962 sbi->s_es = es; 4963 sb->s_magic = le16_to_cpu(es->s_magic); 4964 if (sb->s_magic != EXT4_SUPER_MAGIC) { 4965 if (!silent) 4966 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 4967 goto out; 4968 } 4969 4970 if (le32_to_cpu(es->s_log_block_size) > 4971 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4972 ext4_msg(sb, KERN_ERR, 4973 "Invalid log block size: %u", 4974 le32_to_cpu(es->s_log_block_size)); 4975 goto out; 4976 } 4977 if (le32_to_cpu(es->s_log_cluster_size) > 4978 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4979 ext4_msg(sb, KERN_ERR, 4980 "Invalid log cluster size: %u", 4981 le32_to_cpu(es->s_log_cluster_size)); 4982 goto out; 4983 } 4984 4985 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4986 4987 /* 4988 * If the default block size is not the same as the real block size, 4989 * we need to reload it. 4990 */ 4991 if (sb->s_blocksize == blocksize) { 4992 *lsb = logical_sb_block; 4993 sbi->s_sbh = bh; 4994 return 0; 4995 } 4996 4997 /* 4998 * bh must be released before kill_bdev(), otherwise 4999 * it won't be freed and its page also. kill_bdev() 5000 * is called by sb_set_blocksize(). 5001 */ 5002 brelse(bh); 5003 /* Validate the filesystem blocksize */ 5004 if (!sb_set_blocksize(sb, blocksize)) { 5005 ext4_msg(sb, KERN_ERR, "bad block size %d", 5006 blocksize); 5007 bh = NULL; 5008 goto out; 5009 } 5010 5011 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5012 offset = do_div(logical_sb_block, blocksize); 5013 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5014 if (IS_ERR(bh)) { 5015 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5016 ret = PTR_ERR(bh); 5017 bh = NULL; 5018 goto out; 5019 } 5020 es = (struct ext4_super_block *)(bh->b_data + offset); 5021 sbi->s_es = es; 5022 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5023 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5024 goto out; 5025 } 5026 *lsb = logical_sb_block; 5027 sbi->s_sbh = bh; 5028 return 0; 5029 out: 5030 brelse(bh); 5031 return ret; 5032 } 5033 5034 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5035 { 5036 struct ext4_super_block *es = NULL; 5037 struct ext4_sb_info *sbi = EXT4_SB(sb); 5038 struct flex_groups **flex_groups; 5039 ext4_fsblk_t block; 5040 ext4_fsblk_t logical_sb_block; 5041 struct inode *root; 5042 int ret = -ENOMEM; 5043 unsigned int i; 5044 int needs_recovery, has_huge_files; 5045 int err = 0; 5046 ext4_group_t first_not_zeroed; 5047 struct ext4_fs_context *ctx = fc->fs_private; 5048 int silent = fc->sb_flags & SB_SILENT; 5049 5050 /* Set defaults for the variables that will be set during parsing */ 5051 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5052 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5053 5054 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5055 sbi->s_sectors_written_start = 5056 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5057 5058 /* -EINVAL is default */ 5059 ret = -EINVAL; 5060 err = ext4_load_super(sb, &logical_sb_block, silent); 5061 if (err) 5062 goto out_fail; 5063 5064 es = sbi->s_es; 5065 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5066 5067 err = ext4_init_metadata_csum(sb, es); 5068 if (err) 5069 goto failed_mount; 5070 5071 ext4_set_def_opts(sb, es); 5072 5073 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5074 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5075 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5076 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5077 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5078 5079 /* 5080 * set default s_li_wait_mult for lazyinit, for the case there is 5081 * no mount option specified. 5082 */ 5083 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5084 5085 if (ext4_inode_info_init(sb, es)) 5086 goto failed_mount; 5087 5088 err = parse_apply_sb_mount_options(sb, ctx); 5089 if (err < 0) 5090 goto failed_mount; 5091 5092 sbi->s_def_mount_opt = sbi->s_mount_opt; 5093 5094 err = ext4_check_opt_consistency(fc, sb); 5095 if (err < 0) 5096 goto failed_mount; 5097 5098 ext4_apply_options(fc, sb); 5099 5100 if (ext4_encoding_init(sb, es)) 5101 goto failed_mount; 5102 5103 if (ext4_journal_data_mode_check(sb)) 5104 goto failed_mount; 5105 5106 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5107 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5108 5109 /* i_version is always enabled now */ 5110 sb->s_flags |= SB_I_VERSION; 5111 5112 if (ext4_check_feature_compatibility(sb, es, silent)) 5113 goto failed_mount; 5114 5115 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 5116 ext4_msg(sb, KERN_ERR, 5117 "Number of reserved GDT blocks insanely large: %d", 5118 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 5119 goto failed_mount; 5120 } 5121 5122 if (sbi->s_daxdev) { 5123 if (sb->s_blocksize == PAGE_SIZE) 5124 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 5125 else 5126 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 5127 } 5128 5129 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 5130 if (ext4_has_feature_inline_data(sb)) { 5131 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 5132 " that may contain inline data"); 5133 goto failed_mount; 5134 } 5135 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 5136 ext4_msg(sb, KERN_ERR, 5137 "DAX unsupported by block device."); 5138 goto failed_mount; 5139 } 5140 } 5141 5142 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 5143 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 5144 es->s_encryption_level); 5145 goto failed_mount; 5146 } 5147 5148 has_huge_files = ext4_has_feature_huge_file(sb); 5149 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5150 has_huge_files); 5151 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5152 5153 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5154 if (ext4_has_feature_64bit(sb)) { 5155 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5156 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5157 !is_power_of_2(sbi->s_desc_size)) { 5158 ext4_msg(sb, KERN_ERR, 5159 "unsupported descriptor size %lu", 5160 sbi->s_desc_size); 5161 goto failed_mount; 5162 } 5163 } else 5164 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5165 5166 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5167 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5168 5169 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5170 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5171 if (!silent) 5172 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5173 goto failed_mount; 5174 } 5175 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5176 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5177 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5178 sbi->s_inodes_per_group); 5179 goto failed_mount; 5180 } 5181 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5182 sbi->s_inodes_per_block; 5183 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5184 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5185 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5186 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5187 5188 for (i = 0; i < 4; i++) 5189 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5190 sbi->s_def_hash_version = es->s_def_hash_version; 5191 if (ext4_has_feature_dir_index(sb)) { 5192 i = le32_to_cpu(es->s_flags); 5193 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5194 sbi->s_hash_unsigned = 3; 5195 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5196 #ifdef __CHAR_UNSIGNED__ 5197 if (!sb_rdonly(sb)) 5198 es->s_flags |= 5199 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5200 sbi->s_hash_unsigned = 3; 5201 #else 5202 if (!sb_rdonly(sb)) 5203 es->s_flags |= 5204 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5205 #endif 5206 } 5207 } 5208 5209 if (ext4_handle_clustersize(sb)) 5210 goto failed_mount; 5211 5212 /* 5213 * Test whether we have more sectors than will fit in sector_t, 5214 * and whether the max offset is addressable by the page cache. 5215 */ 5216 err = generic_check_addressable(sb->s_blocksize_bits, 5217 ext4_blocks_count(es)); 5218 if (err) { 5219 ext4_msg(sb, KERN_ERR, "filesystem" 5220 " too large to mount safely on this system"); 5221 goto failed_mount; 5222 } 5223 5224 if (ext4_geometry_check(sb, es)) 5225 goto failed_mount; 5226 5227 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5228 if (err) 5229 goto failed_mount; 5230 5231 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5232 spin_lock_init(&sbi->s_error_lock); 5233 INIT_WORK(&sbi->s_error_work, flush_stashed_error_work); 5234 5235 /* Register extent status tree shrinker */ 5236 if (ext4_es_register_shrinker(sbi)) 5237 goto failed_mount3; 5238 5239 sbi->s_stripe = ext4_get_stripe_size(sbi); 5240 sbi->s_extent_max_zeroout_kb = 32; 5241 5242 /* 5243 * set up enough so that it can read an inode 5244 */ 5245 sb->s_op = &ext4_sops; 5246 sb->s_export_op = &ext4_export_ops; 5247 sb->s_xattr = ext4_xattr_handlers; 5248 #ifdef CONFIG_FS_ENCRYPTION 5249 sb->s_cop = &ext4_cryptops; 5250 #endif 5251 #ifdef CONFIG_FS_VERITY 5252 sb->s_vop = &ext4_verityops; 5253 #endif 5254 #ifdef CONFIG_QUOTA 5255 sb->dq_op = &ext4_quota_operations; 5256 if (ext4_has_feature_quota(sb)) 5257 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5258 else 5259 sb->s_qcop = &ext4_qctl_operations; 5260 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5261 #endif 5262 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 5263 5264 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5265 mutex_init(&sbi->s_orphan_lock); 5266 5267 ext4_fast_commit_init(sb); 5268 5269 sb->s_root = NULL; 5270 5271 needs_recovery = (es->s_last_orphan != 0 || 5272 ext4_has_feature_orphan_present(sb) || 5273 ext4_has_feature_journal_needs_recovery(sb)); 5274 5275 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 5276 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 5277 goto failed_mount3a; 5278 5279 /* 5280 * The first inode we look at is the journal inode. Don't try 5281 * root first: it may be modified in the journal! 5282 */ 5283 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5284 err = ext4_load_and_init_journal(sb, es, ctx); 5285 if (err) 5286 goto failed_mount3a; 5287 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5288 ext4_has_feature_journal_needs_recovery(sb)) { 5289 ext4_msg(sb, KERN_ERR, "required journal recovery " 5290 "suppressed and not mounted read-only"); 5291 goto failed_mount3a; 5292 } else { 5293 /* Nojournal mode, all journal mount options are illegal */ 5294 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 5295 ext4_msg(sb, KERN_ERR, "can't mount with " 5296 "journal_checksum, fs mounted w/o journal"); 5297 goto failed_mount3a; 5298 } 5299 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5300 ext4_msg(sb, KERN_ERR, "can't mount with " 5301 "journal_async_commit, fs mounted w/o journal"); 5302 goto failed_mount3a; 5303 } 5304 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 5305 ext4_msg(sb, KERN_ERR, "can't mount with " 5306 "commit=%lu, fs mounted w/o journal", 5307 sbi->s_commit_interval / HZ); 5308 goto failed_mount3a; 5309 } 5310 if (EXT4_MOUNT_DATA_FLAGS & 5311 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 5312 ext4_msg(sb, KERN_ERR, "can't mount with " 5313 "data=, fs mounted w/o journal"); 5314 goto failed_mount3a; 5315 } 5316 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5317 clear_opt(sb, JOURNAL_CHECKSUM); 5318 clear_opt(sb, DATA_FLAGS); 5319 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5320 sbi->s_journal = NULL; 5321 needs_recovery = 0; 5322 } 5323 5324 if (!test_opt(sb, NO_MBCACHE)) { 5325 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5326 if (!sbi->s_ea_block_cache) { 5327 ext4_msg(sb, KERN_ERR, 5328 "Failed to create ea_block_cache"); 5329 goto failed_mount_wq; 5330 } 5331 5332 if (ext4_has_feature_ea_inode(sb)) { 5333 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5334 if (!sbi->s_ea_inode_cache) { 5335 ext4_msg(sb, KERN_ERR, 5336 "Failed to create ea_inode_cache"); 5337 goto failed_mount_wq; 5338 } 5339 } 5340 } 5341 5342 if (ext4_has_feature_verity(sb) && sb->s_blocksize != PAGE_SIZE) { 5343 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 5344 goto failed_mount_wq; 5345 } 5346 5347 /* 5348 * Get the # of file system overhead blocks from the 5349 * superblock if present. 5350 */ 5351 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5352 /* ignore the precalculated value if it is ridiculous */ 5353 if (sbi->s_overhead > ext4_blocks_count(es)) 5354 sbi->s_overhead = 0; 5355 /* 5356 * If the bigalloc feature is not enabled recalculating the 5357 * overhead doesn't take long, so we might as well just redo 5358 * it to make sure we are using the correct value. 5359 */ 5360 if (!ext4_has_feature_bigalloc(sb)) 5361 sbi->s_overhead = 0; 5362 if (sbi->s_overhead == 0) { 5363 err = ext4_calculate_overhead(sb); 5364 if (err) 5365 goto failed_mount_wq; 5366 } 5367 5368 /* 5369 * The maximum number of concurrent works can be high and 5370 * concurrency isn't really necessary. Limit it to 1. 5371 */ 5372 EXT4_SB(sb)->rsv_conversion_wq = 5373 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5374 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5375 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5376 ret = -ENOMEM; 5377 goto failed_mount4; 5378 } 5379 5380 /* 5381 * The jbd2_journal_load will have done any necessary log recovery, 5382 * so we can safely mount the rest of the filesystem now. 5383 */ 5384 5385 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5386 if (IS_ERR(root)) { 5387 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5388 ret = PTR_ERR(root); 5389 root = NULL; 5390 goto failed_mount4; 5391 } 5392 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5393 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5394 iput(root); 5395 goto failed_mount4; 5396 } 5397 5398 sb->s_root = d_make_root(root); 5399 if (!sb->s_root) { 5400 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5401 ret = -ENOMEM; 5402 goto failed_mount4; 5403 } 5404 5405 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 5406 if (ret == -EROFS) { 5407 sb->s_flags |= SB_RDONLY; 5408 ret = 0; 5409 } else if (ret) 5410 goto failed_mount4a; 5411 5412 ext4_set_resv_clusters(sb); 5413 5414 if (test_opt(sb, BLOCK_VALIDITY)) { 5415 err = ext4_setup_system_zone(sb); 5416 if (err) { 5417 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5418 "zone (%d)", err); 5419 goto failed_mount4a; 5420 } 5421 } 5422 ext4_fc_replay_cleanup(sb); 5423 5424 ext4_ext_init(sb); 5425 5426 /* 5427 * Enable optimize_scan if number of groups is > threshold. This can be 5428 * turned off by passing "mb_optimize_scan=0". This can also be 5429 * turned on forcefully by passing "mb_optimize_scan=1". 5430 */ 5431 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5432 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5433 set_opt2(sb, MB_OPTIMIZE_SCAN); 5434 else 5435 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5436 } 5437 5438 err = ext4_mb_init(sb); 5439 if (err) { 5440 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5441 err); 5442 goto failed_mount5; 5443 } 5444 5445 /* 5446 * We can only set up the journal commit callback once 5447 * mballoc is initialized 5448 */ 5449 if (sbi->s_journal) 5450 sbi->s_journal->j_commit_callback = 5451 ext4_journal_commit_callback; 5452 5453 block = ext4_count_free_clusters(sb); 5454 ext4_free_blocks_count_set(sbi->s_es, 5455 EXT4_C2B(sbi, block)); 5456 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5457 GFP_KERNEL); 5458 if (!err) { 5459 unsigned long freei = ext4_count_free_inodes(sb); 5460 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5461 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5462 GFP_KERNEL); 5463 } 5464 if (!err) 5465 err = percpu_counter_init(&sbi->s_dirs_counter, 5466 ext4_count_dirs(sb), GFP_KERNEL); 5467 if (!err) 5468 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5469 GFP_KERNEL); 5470 if (!err) 5471 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 5472 GFP_KERNEL); 5473 if (!err) 5474 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5475 5476 if (err) { 5477 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5478 goto failed_mount6; 5479 } 5480 5481 if (ext4_has_feature_flex_bg(sb)) 5482 if (!ext4_fill_flex_info(sb)) { 5483 ext4_msg(sb, KERN_ERR, 5484 "unable to initialize " 5485 "flex_bg meta info!"); 5486 ret = -ENOMEM; 5487 goto failed_mount6; 5488 } 5489 5490 err = ext4_register_li_request(sb, first_not_zeroed); 5491 if (err) 5492 goto failed_mount6; 5493 5494 err = ext4_register_sysfs(sb); 5495 if (err) 5496 goto failed_mount7; 5497 5498 err = ext4_init_orphan_info(sb); 5499 if (err) 5500 goto failed_mount8; 5501 #ifdef CONFIG_QUOTA 5502 /* Enable quota usage during mount. */ 5503 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5504 err = ext4_enable_quotas(sb); 5505 if (err) 5506 goto failed_mount9; 5507 } 5508 #endif /* CONFIG_QUOTA */ 5509 5510 /* 5511 * Save the original bdev mapping's wb_err value which could be 5512 * used to detect the metadata async write error. 5513 */ 5514 spin_lock_init(&sbi->s_bdev_wb_lock); 5515 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5516 &sbi->s_bdev_wb_err); 5517 sb->s_bdev->bd_super = sb; 5518 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5519 ext4_orphan_cleanup(sb, es); 5520 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5521 /* 5522 * Update the checksum after updating free space/inode counters and 5523 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5524 * checksum in the buffer cache until it is written out and 5525 * e2fsprogs programs trying to open a file system immediately 5526 * after it is mounted can fail. 5527 */ 5528 ext4_superblock_csum_set(sb); 5529 if (needs_recovery) { 5530 ext4_msg(sb, KERN_INFO, "recovery complete"); 5531 err = ext4_mark_recovery_complete(sb, es); 5532 if (err) 5533 goto failed_mount9; 5534 } 5535 5536 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) 5537 ext4_msg(sb, KERN_WARNING, 5538 "mounting with \"discard\" option, but the device does not support discard"); 5539 5540 if (es->s_error_count) 5541 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5542 5543 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5544 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5545 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5546 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5547 atomic_set(&sbi->s_warning_count, 0); 5548 atomic_set(&sbi->s_msg_count, 0); 5549 5550 return 0; 5551 5552 failed_mount9: 5553 ext4_release_orphan_info(sb); 5554 failed_mount8: 5555 ext4_unregister_sysfs(sb); 5556 kobject_put(&sbi->s_kobj); 5557 failed_mount7: 5558 ext4_unregister_li_request(sb); 5559 failed_mount6: 5560 ext4_mb_release(sb); 5561 rcu_read_lock(); 5562 flex_groups = rcu_dereference(sbi->s_flex_groups); 5563 if (flex_groups) { 5564 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 5565 kvfree(flex_groups[i]); 5566 kvfree(flex_groups); 5567 } 5568 rcu_read_unlock(); 5569 percpu_counter_destroy(&sbi->s_freeclusters_counter); 5570 percpu_counter_destroy(&sbi->s_freeinodes_counter); 5571 percpu_counter_destroy(&sbi->s_dirs_counter); 5572 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 5573 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 5574 percpu_free_rwsem(&sbi->s_writepages_rwsem); 5575 failed_mount5: 5576 ext4_ext_release(sb); 5577 ext4_release_system_zone(sb); 5578 failed_mount4a: 5579 dput(sb->s_root); 5580 sb->s_root = NULL; 5581 failed_mount4: 5582 ext4_msg(sb, KERN_ERR, "mount failed"); 5583 if (EXT4_SB(sb)->rsv_conversion_wq) 5584 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5585 failed_mount_wq: 5586 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5587 sbi->s_ea_inode_cache = NULL; 5588 5589 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5590 sbi->s_ea_block_cache = NULL; 5591 5592 if (sbi->s_journal) { 5593 /* flush s_error_work before journal destroy. */ 5594 flush_work(&sbi->s_error_work); 5595 jbd2_journal_destroy(sbi->s_journal); 5596 sbi->s_journal = NULL; 5597 } 5598 failed_mount3a: 5599 ext4_es_unregister_shrinker(sbi); 5600 failed_mount3: 5601 /* flush s_error_work before sbi destroy */ 5602 flush_work(&sbi->s_error_work); 5603 del_timer_sync(&sbi->s_err_report); 5604 ext4_stop_mmpd(sbi); 5605 ext4_group_desc_free(sbi); 5606 failed_mount: 5607 if (sbi->s_chksum_driver) 5608 crypto_free_shash(sbi->s_chksum_driver); 5609 5610 #if IS_ENABLED(CONFIG_UNICODE) 5611 utf8_unload(sb->s_encoding); 5612 #endif 5613 5614 #ifdef CONFIG_QUOTA 5615 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5616 kfree(get_qf_name(sb, sbi, i)); 5617 #endif 5618 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5619 /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */ 5620 brelse(sbi->s_sbh); 5621 ext4_blkdev_remove(sbi); 5622 out_fail: 5623 sb->s_fs_info = NULL; 5624 return err ? err : ret; 5625 } 5626 5627 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5628 { 5629 struct ext4_fs_context *ctx = fc->fs_private; 5630 struct ext4_sb_info *sbi; 5631 const char *descr; 5632 int ret; 5633 5634 sbi = ext4_alloc_sbi(sb); 5635 if (!sbi) 5636 return -ENOMEM; 5637 5638 fc->s_fs_info = sbi; 5639 5640 /* Cleanup superblock name */ 5641 strreplace(sb->s_id, '/', '!'); 5642 5643 sbi->s_sb_block = 1; /* Default super block location */ 5644 if (ctx->spec & EXT4_SPEC_s_sb_block) 5645 sbi->s_sb_block = ctx->s_sb_block; 5646 5647 ret = __ext4_fill_super(fc, sb); 5648 if (ret < 0) 5649 goto free_sbi; 5650 5651 if (sbi->s_journal) { 5652 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5653 descr = " journalled data mode"; 5654 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5655 descr = " ordered data mode"; 5656 else 5657 descr = " writeback data mode"; 5658 } else 5659 descr = "out journal"; 5660 5661 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5662 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 5663 "Quota mode: %s.", descr, ext4_quota_mode(sb)); 5664 5665 /* Update the s_overhead_clusters if necessary */ 5666 ext4_update_overhead(sb, false); 5667 return 0; 5668 5669 free_sbi: 5670 ext4_free_sbi(sbi); 5671 fc->s_fs_info = NULL; 5672 return ret; 5673 } 5674 5675 static int ext4_get_tree(struct fs_context *fc) 5676 { 5677 return get_tree_bdev(fc, ext4_fill_super); 5678 } 5679 5680 /* 5681 * Setup any per-fs journal parameters now. We'll do this both on 5682 * initial mount, once the journal has been initialised but before we've 5683 * done any recovery; and again on any subsequent remount. 5684 */ 5685 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5686 { 5687 struct ext4_sb_info *sbi = EXT4_SB(sb); 5688 5689 journal->j_commit_interval = sbi->s_commit_interval; 5690 journal->j_min_batch_time = sbi->s_min_batch_time; 5691 journal->j_max_batch_time = sbi->s_max_batch_time; 5692 ext4_fc_init(sb, journal); 5693 5694 write_lock(&journal->j_state_lock); 5695 if (test_opt(sb, BARRIER)) 5696 journal->j_flags |= JBD2_BARRIER; 5697 else 5698 journal->j_flags &= ~JBD2_BARRIER; 5699 if (test_opt(sb, DATA_ERR_ABORT)) 5700 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5701 else 5702 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5703 write_unlock(&journal->j_state_lock); 5704 } 5705 5706 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5707 unsigned int journal_inum) 5708 { 5709 struct inode *journal_inode; 5710 5711 /* 5712 * Test for the existence of a valid inode on disk. Bad things 5713 * happen if we iget() an unused inode, as the subsequent iput() 5714 * will try to delete it. 5715 */ 5716 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5717 if (IS_ERR(journal_inode)) { 5718 ext4_msg(sb, KERN_ERR, "no journal found"); 5719 return NULL; 5720 } 5721 if (!journal_inode->i_nlink) { 5722 make_bad_inode(journal_inode); 5723 iput(journal_inode); 5724 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5725 return NULL; 5726 } 5727 5728 ext4_debug("Journal inode found at %p: %lld bytes\n", 5729 journal_inode, journal_inode->i_size); 5730 if (!S_ISREG(journal_inode->i_mode)) { 5731 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5732 iput(journal_inode); 5733 return NULL; 5734 } 5735 return journal_inode; 5736 } 5737 5738 static journal_t *ext4_get_journal(struct super_block *sb, 5739 unsigned int journal_inum) 5740 { 5741 struct inode *journal_inode; 5742 journal_t *journal; 5743 5744 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5745 return NULL; 5746 5747 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5748 if (!journal_inode) 5749 return NULL; 5750 5751 journal = jbd2_journal_init_inode(journal_inode); 5752 if (!journal) { 5753 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5754 iput(journal_inode); 5755 return NULL; 5756 } 5757 journal->j_private = sb; 5758 ext4_init_journal_params(sb, journal); 5759 return journal; 5760 } 5761 5762 static journal_t *ext4_get_dev_journal(struct super_block *sb, 5763 dev_t j_dev) 5764 { 5765 struct buffer_head *bh; 5766 journal_t *journal; 5767 ext4_fsblk_t start; 5768 ext4_fsblk_t len; 5769 int hblock, blocksize; 5770 ext4_fsblk_t sb_block; 5771 unsigned long offset; 5772 struct ext4_super_block *es; 5773 struct block_device *bdev; 5774 5775 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5776 return NULL; 5777 5778 bdev = ext4_blkdev_get(j_dev, sb); 5779 if (bdev == NULL) 5780 return NULL; 5781 5782 blocksize = sb->s_blocksize; 5783 hblock = bdev_logical_block_size(bdev); 5784 if (blocksize < hblock) { 5785 ext4_msg(sb, KERN_ERR, 5786 "blocksize too small for journal device"); 5787 goto out_bdev; 5788 } 5789 5790 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5791 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5792 set_blocksize(bdev, blocksize); 5793 if (!(bh = __bread(bdev, sb_block, blocksize))) { 5794 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5795 "external journal"); 5796 goto out_bdev; 5797 } 5798 5799 es = (struct ext4_super_block *) (bh->b_data + offset); 5800 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5801 !(le32_to_cpu(es->s_feature_incompat) & 5802 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5803 ext4_msg(sb, KERN_ERR, "external journal has " 5804 "bad superblock"); 5805 brelse(bh); 5806 goto out_bdev; 5807 } 5808 5809 if ((le32_to_cpu(es->s_feature_ro_compat) & 5810 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5811 es->s_checksum != ext4_superblock_csum(sb, es)) { 5812 ext4_msg(sb, KERN_ERR, "external journal has " 5813 "corrupt superblock"); 5814 brelse(bh); 5815 goto out_bdev; 5816 } 5817 5818 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5819 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5820 brelse(bh); 5821 goto out_bdev; 5822 } 5823 5824 len = ext4_blocks_count(es); 5825 start = sb_block + 1; 5826 brelse(bh); /* we're done with the superblock */ 5827 5828 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 5829 start, len, blocksize); 5830 if (!journal) { 5831 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5832 goto out_bdev; 5833 } 5834 journal->j_private = sb; 5835 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) { 5836 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 5837 goto out_journal; 5838 } 5839 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5840 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5841 "user (unsupported) - %d", 5842 be32_to_cpu(journal->j_superblock->s_nr_users)); 5843 goto out_journal; 5844 } 5845 EXT4_SB(sb)->s_journal_bdev = bdev; 5846 ext4_init_journal_params(sb, journal); 5847 return journal; 5848 5849 out_journal: 5850 jbd2_journal_destroy(journal); 5851 out_bdev: 5852 ext4_blkdev_put(bdev); 5853 return NULL; 5854 } 5855 5856 static int ext4_load_journal(struct super_block *sb, 5857 struct ext4_super_block *es, 5858 unsigned long journal_devnum) 5859 { 5860 journal_t *journal; 5861 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5862 dev_t journal_dev; 5863 int err = 0; 5864 int really_read_only; 5865 int journal_dev_ro; 5866 5867 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5868 return -EFSCORRUPTED; 5869 5870 if (journal_devnum && 5871 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5872 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5873 "numbers have changed"); 5874 journal_dev = new_decode_dev(journal_devnum); 5875 } else 5876 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5877 5878 if (journal_inum && journal_dev) { 5879 ext4_msg(sb, KERN_ERR, 5880 "filesystem has both journal inode and journal device!"); 5881 return -EINVAL; 5882 } 5883 5884 if (journal_inum) { 5885 journal = ext4_get_journal(sb, journal_inum); 5886 if (!journal) 5887 return -EINVAL; 5888 } else { 5889 journal = ext4_get_dev_journal(sb, journal_dev); 5890 if (!journal) 5891 return -EINVAL; 5892 } 5893 5894 journal_dev_ro = bdev_read_only(journal->j_dev); 5895 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5896 5897 if (journal_dev_ro && !sb_rdonly(sb)) { 5898 ext4_msg(sb, KERN_ERR, 5899 "journal device read-only, try mounting with '-o ro'"); 5900 err = -EROFS; 5901 goto err_out; 5902 } 5903 5904 /* 5905 * Are we loading a blank journal or performing recovery after a 5906 * crash? For recovery, we need to check in advance whether we 5907 * can get read-write access to the device. 5908 */ 5909 if (ext4_has_feature_journal_needs_recovery(sb)) { 5910 if (sb_rdonly(sb)) { 5911 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5912 "required on readonly filesystem"); 5913 if (really_read_only) { 5914 ext4_msg(sb, KERN_ERR, "write access " 5915 "unavailable, cannot proceed " 5916 "(try mounting with noload)"); 5917 err = -EROFS; 5918 goto err_out; 5919 } 5920 ext4_msg(sb, KERN_INFO, "write access will " 5921 "be enabled during recovery"); 5922 } 5923 } 5924 5925 if (!(journal->j_flags & JBD2_BARRIER)) 5926 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5927 5928 if (!ext4_has_feature_journal_needs_recovery(sb)) 5929 err = jbd2_journal_wipe(journal, !really_read_only); 5930 if (!err) { 5931 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5932 if (save) 5933 memcpy(save, ((char *) es) + 5934 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 5935 err = jbd2_journal_load(journal); 5936 if (save) 5937 memcpy(((char *) es) + EXT4_S_ERR_START, 5938 save, EXT4_S_ERR_LEN); 5939 kfree(save); 5940 } 5941 5942 if (err) { 5943 ext4_msg(sb, KERN_ERR, "error loading journal"); 5944 goto err_out; 5945 } 5946 5947 EXT4_SB(sb)->s_journal = journal; 5948 err = ext4_clear_journal_err(sb, es); 5949 if (err) { 5950 EXT4_SB(sb)->s_journal = NULL; 5951 jbd2_journal_destroy(journal); 5952 return err; 5953 } 5954 5955 if (!really_read_only && journal_devnum && 5956 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5957 es->s_journal_dev = cpu_to_le32(journal_devnum); 5958 5959 /* Make sure we flush the recovery flag to disk. */ 5960 ext4_commit_super(sb); 5961 } 5962 5963 return 0; 5964 5965 err_out: 5966 jbd2_journal_destroy(journal); 5967 return err; 5968 } 5969 5970 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 5971 static void ext4_update_super(struct super_block *sb) 5972 { 5973 struct ext4_sb_info *sbi = EXT4_SB(sb); 5974 struct ext4_super_block *es = sbi->s_es; 5975 struct buffer_head *sbh = sbi->s_sbh; 5976 5977 lock_buffer(sbh); 5978 /* 5979 * If the file system is mounted read-only, don't update the 5980 * superblock write time. This avoids updating the superblock 5981 * write time when we are mounting the root file system 5982 * read/only but we need to replay the journal; at that point, 5983 * for people who are east of GMT and who make their clock 5984 * tick in localtime for Windows bug-for-bug compatibility, 5985 * the clock is set in the future, and this will cause e2fsck 5986 * to complain and force a full file system check. 5987 */ 5988 if (!(sb->s_flags & SB_RDONLY)) 5989 ext4_update_tstamp(es, s_wtime); 5990 es->s_kbytes_written = 5991 cpu_to_le64(sbi->s_kbytes_written + 5992 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 5993 sbi->s_sectors_written_start) >> 1)); 5994 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 5995 ext4_free_blocks_count_set(es, 5996 EXT4_C2B(sbi, percpu_counter_sum_positive( 5997 &sbi->s_freeclusters_counter))); 5998 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 5999 es->s_free_inodes_count = 6000 cpu_to_le32(percpu_counter_sum_positive( 6001 &sbi->s_freeinodes_counter)); 6002 /* Copy error information to the on-disk superblock */ 6003 spin_lock(&sbi->s_error_lock); 6004 if (sbi->s_add_error_count > 0) { 6005 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6006 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6007 __ext4_update_tstamp(&es->s_first_error_time, 6008 &es->s_first_error_time_hi, 6009 sbi->s_first_error_time); 6010 strncpy(es->s_first_error_func, sbi->s_first_error_func, 6011 sizeof(es->s_first_error_func)); 6012 es->s_first_error_line = 6013 cpu_to_le32(sbi->s_first_error_line); 6014 es->s_first_error_ino = 6015 cpu_to_le32(sbi->s_first_error_ino); 6016 es->s_first_error_block = 6017 cpu_to_le64(sbi->s_first_error_block); 6018 es->s_first_error_errcode = 6019 ext4_errno_to_code(sbi->s_first_error_code); 6020 } 6021 __ext4_update_tstamp(&es->s_last_error_time, 6022 &es->s_last_error_time_hi, 6023 sbi->s_last_error_time); 6024 strncpy(es->s_last_error_func, sbi->s_last_error_func, 6025 sizeof(es->s_last_error_func)); 6026 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6027 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6028 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6029 es->s_last_error_errcode = 6030 ext4_errno_to_code(sbi->s_last_error_code); 6031 /* 6032 * Start the daily error reporting function if it hasn't been 6033 * started already 6034 */ 6035 if (!es->s_error_count) 6036 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6037 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6038 sbi->s_add_error_count = 0; 6039 } 6040 spin_unlock(&sbi->s_error_lock); 6041 6042 ext4_superblock_csum_set(sb); 6043 unlock_buffer(sbh); 6044 } 6045 6046 static int ext4_commit_super(struct super_block *sb) 6047 { 6048 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6049 6050 if (!sbh) 6051 return -EINVAL; 6052 if (block_device_ejected(sb)) 6053 return -ENODEV; 6054 6055 ext4_update_super(sb); 6056 6057 lock_buffer(sbh); 6058 /* Buffer got discarded which means block device got invalidated */ 6059 if (!buffer_mapped(sbh)) { 6060 unlock_buffer(sbh); 6061 return -EIO; 6062 } 6063 6064 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6065 /* 6066 * Oh, dear. A previous attempt to write the 6067 * superblock failed. This could happen because the 6068 * USB device was yanked out. Or it could happen to 6069 * be a transient write error and maybe the block will 6070 * be remapped. Nothing we can do but to retry the 6071 * write and hope for the best. 6072 */ 6073 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6074 "superblock detected"); 6075 clear_buffer_write_io_error(sbh); 6076 set_buffer_uptodate(sbh); 6077 } 6078 get_bh(sbh); 6079 /* Clear potential dirty bit if it was journalled update */ 6080 clear_buffer_dirty(sbh); 6081 sbh->b_end_io = end_buffer_write_sync; 6082 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6083 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6084 wait_on_buffer(sbh); 6085 if (buffer_write_io_error(sbh)) { 6086 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6087 "superblock"); 6088 clear_buffer_write_io_error(sbh); 6089 set_buffer_uptodate(sbh); 6090 return -EIO; 6091 } 6092 return 0; 6093 } 6094 6095 /* 6096 * Have we just finished recovery? If so, and if we are mounting (or 6097 * remounting) the filesystem readonly, then we will end up with a 6098 * consistent fs on disk. Record that fact. 6099 */ 6100 static int ext4_mark_recovery_complete(struct super_block *sb, 6101 struct ext4_super_block *es) 6102 { 6103 int err; 6104 journal_t *journal = EXT4_SB(sb)->s_journal; 6105 6106 if (!ext4_has_feature_journal(sb)) { 6107 if (journal != NULL) { 6108 ext4_error(sb, "Journal got removed while the fs was " 6109 "mounted!"); 6110 return -EFSCORRUPTED; 6111 } 6112 return 0; 6113 } 6114 jbd2_journal_lock_updates(journal); 6115 err = jbd2_journal_flush(journal, 0); 6116 if (err < 0) 6117 goto out; 6118 6119 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6120 ext4_has_feature_orphan_present(sb))) { 6121 if (!ext4_orphan_file_empty(sb)) { 6122 ext4_error(sb, "Orphan file not empty on read-only fs."); 6123 err = -EFSCORRUPTED; 6124 goto out; 6125 } 6126 ext4_clear_feature_journal_needs_recovery(sb); 6127 ext4_clear_feature_orphan_present(sb); 6128 ext4_commit_super(sb); 6129 } 6130 out: 6131 jbd2_journal_unlock_updates(journal); 6132 return err; 6133 } 6134 6135 /* 6136 * If we are mounting (or read-write remounting) a filesystem whose journal 6137 * has recorded an error from a previous lifetime, move that error to the 6138 * main filesystem now. 6139 */ 6140 static int ext4_clear_journal_err(struct super_block *sb, 6141 struct ext4_super_block *es) 6142 { 6143 journal_t *journal; 6144 int j_errno; 6145 const char *errstr; 6146 6147 if (!ext4_has_feature_journal(sb)) { 6148 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6149 return -EFSCORRUPTED; 6150 } 6151 6152 journal = EXT4_SB(sb)->s_journal; 6153 6154 /* 6155 * Now check for any error status which may have been recorded in the 6156 * journal by a prior ext4_error() or ext4_abort() 6157 */ 6158 6159 j_errno = jbd2_journal_errno(journal); 6160 if (j_errno) { 6161 char nbuf[16]; 6162 6163 errstr = ext4_decode_error(sb, j_errno, nbuf); 6164 ext4_warning(sb, "Filesystem error recorded " 6165 "from previous mount: %s", errstr); 6166 ext4_warning(sb, "Marking fs in need of filesystem check."); 6167 6168 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6169 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6170 ext4_commit_super(sb); 6171 6172 jbd2_journal_clear_err(journal); 6173 jbd2_journal_update_sb_errno(journal); 6174 } 6175 return 0; 6176 } 6177 6178 /* 6179 * Force the running and committing transactions to commit, 6180 * and wait on the commit. 6181 */ 6182 int ext4_force_commit(struct super_block *sb) 6183 { 6184 journal_t *journal; 6185 6186 if (sb_rdonly(sb)) 6187 return 0; 6188 6189 journal = EXT4_SB(sb)->s_journal; 6190 return ext4_journal_force_commit(journal); 6191 } 6192 6193 static int ext4_sync_fs(struct super_block *sb, int wait) 6194 { 6195 int ret = 0; 6196 tid_t target; 6197 bool needs_barrier = false; 6198 struct ext4_sb_info *sbi = EXT4_SB(sb); 6199 6200 if (unlikely(ext4_forced_shutdown(sbi))) 6201 return 0; 6202 6203 trace_ext4_sync_fs(sb, wait); 6204 flush_workqueue(sbi->rsv_conversion_wq); 6205 /* 6206 * Writeback quota in non-journalled quota case - journalled quota has 6207 * no dirty dquots 6208 */ 6209 dquot_writeback_dquots(sb, -1); 6210 /* 6211 * Data writeback is possible w/o journal transaction, so barrier must 6212 * being sent at the end of the function. But we can skip it if 6213 * transaction_commit will do it for us. 6214 */ 6215 if (sbi->s_journal) { 6216 target = jbd2_get_latest_transaction(sbi->s_journal); 6217 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6218 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6219 needs_barrier = true; 6220 6221 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6222 if (wait) 6223 ret = jbd2_log_wait_commit(sbi->s_journal, 6224 target); 6225 } 6226 } else if (wait && test_opt(sb, BARRIER)) 6227 needs_barrier = true; 6228 if (needs_barrier) { 6229 int err; 6230 err = blkdev_issue_flush(sb->s_bdev); 6231 if (!ret) 6232 ret = err; 6233 } 6234 6235 return ret; 6236 } 6237 6238 /* 6239 * LVM calls this function before a (read-only) snapshot is created. This 6240 * gives us a chance to flush the journal completely and mark the fs clean. 6241 * 6242 * Note that only this function cannot bring a filesystem to be in a clean 6243 * state independently. It relies on upper layer to stop all data & metadata 6244 * modifications. 6245 */ 6246 static int ext4_freeze(struct super_block *sb) 6247 { 6248 int error = 0; 6249 journal_t *journal; 6250 6251 if (sb_rdonly(sb)) 6252 return 0; 6253 6254 journal = EXT4_SB(sb)->s_journal; 6255 6256 if (journal) { 6257 /* Now we set up the journal barrier. */ 6258 jbd2_journal_lock_updates(journal); 6259 6260 /* 6261 * Don't clear the needs_recovery flag if we failed to 6262 * flush the journal. 6263 */ 6264 error = jbd2_journal_flush(journal, 0); 6265 if (error < 0) 6266 goto out; 6267 6268 /* Journal blocked and flushed, clear needs_recovery flag. */ 6269 ext4_clear_feature_journal_needs_recovery(sb); 6270 if (ext4_orphan_file_empty(sb)) 6271 ext4_clear_feature_orphan_present(sb); 6272 } 6273 6274 error = ext4_commit_super(sb); 6275 out: 6276 if (journal) 6277 /* we rely on upper layer to stop further updates */ 6278 jbd2_journal_unlock_updates(journal); 6279 return error; 6280 } 6281 6282 /* 6283 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6284 * flag here, even though the filesystem is not technically dirty yet. 6285 */ 6286 static int ext4_unfreeze(struct super_block *sb) 6287 { 6288 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 6289 return 0; 6290 6291 if (EXT4_SB(sb)->s_journal) { 6292 /* Reset the needs_recovery flag before the fs is unlocked. */ 6293 ext4_set_feature_journal_needs_recovery(sb); 6294 if (ext4_has_feature_orphan_file(sb)) 6295 ext4_set_feature_orphan_present(sb); 6296 } 6297 6298 ext4_commit_super(sb); 6299 return 0; 6300 } 6301 6302 /* 6303 * Structure to save mount options for ext4_remount's benefit 6304 */ 6305 struct ext4_mount_options { 6306 unsigned long s_mount_opt; 6307 unsigned long s_mount_opt2; 6308 kuid_t s_resuid; 6309 kgid_t s_resgid; 6310 unsigned long s_commit_interval; 6311 u32 s_min_batch_time, s_max_batch_time; 6312 #ifdef CONFIG_QUOTA 6313 int s_jquota_fmt; 6314 char *s_qf_names[EXT4_MAXQUOTAS]; 6315 #endif 6316 }; 6317 6318 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6319 { 6320 struct ext4_fs_context *ctx = fc->fs_private; 6321 struct ext4_super_block *es; 6322 struct ext4_sb_info *sbi = EXT4_SB(sb); 6323 unsigned long old_sb_flags; 6324 struct ext4_mount_options old_opts; 6325 ext4_group_t g; 6326 int err = 0; 6327 #ifdef CONFIG_QUOTA 6328 int enable_quota = 0; 6329 int i, j; 6330 char *to_free[EXT4_MAXQUOTAS]; 6331 #endif 6332 6333 6334 /* Store the original options */ 6335 old_sb_flags = sb->s_flags; 6336 old_opts.s_mount_opt = sbi->s_mount_opt; 6337 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6338 old_opts.s_resuid = sbi->s_resuid; 6339 old_opts.s_resgid = sbi->s_resgid; 6340 old_opts.s_commit_interval = sbi->s_commit_interval; 6341 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6342 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6343 #ifdef CONFIG_QUOTA 6344 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6345 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6346 if (sbi->s_qf_names[i]) { 6347 char *qf_name = get_qf_name(sb, sbi, i); 6348 6349 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6350 if (!old_opts.s_qf_names[i]) { 6351 for (j = 0; j < i; j++) 6352 kfree(old_opts.s_qf_names[j]); 6353 return -ENOMEM; 6354 } 6355 } else 6356 old_opts.s_qf_names[i] = NULL; 6357 #endif 6358 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6359 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6360 ctx->journal_ioprio = 6361 sbi->s_journal->j_task->io_context->ioprio; 6362 else 6363 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 6364 6365 } 6366 6367 ext4_apply_options(fc, sb); 6368 6369 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6370 test_opt(sb, JOURNAL_CHECKSUM)) { 6371 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6372 "during remount not supported; ignoring"); 6373 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6374 } 6375 6376 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6377 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6378 ext4_msg(sb, KERN_ERR, "can't mount with " 6379 "both data=journal and delalloc"); 6380 err = -EINVAL; 6381 goto restore_opts; 6382 } 6383 if (test_opt(sb, DIOREAD_NOLOCK)) { 6384 ext4_msg(sb, KERN_ERR, "can't mount with " 6385 "both data=journal and dioread_nolock"); 6386 err = -EINVAL; 6387 goto restore_opts; 6388 } 6389 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6390 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6391 ext4_msg(sb, KERN_ERR, "can't mount with " 6392 "journal_async_commit in data=ordered mode"); 6393 err = -EINVAL; 6394 goto restore_opts; 6395 } 6396 } 6397 6398 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6399 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6400 err = -EINVAL; 6401 goto restore_opts; 6402 } 6403 6404 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) 6405 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6406 6407 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6408 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6409 6410 es = sbi->s_es; 6411 6412 if (sbi->s_journal) { 6413 ext4_init_journal_params(sb, sbi->s_journal); 6414 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6415 } 6416 6417 /* Flush outstanding errors before changing fs state */ 6418 flush_work(&sbi->s_error_work); 6419 6420 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6421 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) { 6422 err = -EROFS; 6423 goto restore_opts; 6424 } 6425 6426 if (fc->sb_flags & SB_RDONLY) { 6427 err = sync_filesystem(sb); 6428 if (err < 0) 6429 goto restore_opts; 6430 err = dquot_suspend(sb, -1); 6431 if (err < 0) 6432 goto restore_opts; 6433 6434 /* 6435 * First of all, the unconditional stuff we have to do 6436 * to disable replay of the journal when we next remount 6437 */ 6438 sb->s_flags |= SB_RDONLY; 6439 6440 /* 6441 * OK, test if we are remounting a valid rw partition 6442 * readonly, and if so set the rdonly flag and then 6443 * mark the partition as valid again. 6444 */ 6445 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6446 (sbi->s_mount_state & EXT4_VALID_FS)) 6447 es->s_state = cpu_to_le16(sbi->s_mount_state); 6448 6449 if (sbi->s_journal) { 6450 /* 6451 * We let remount-ro finish even if marking fs 6452 * as clean failed... 6453 */ 6454 ext4_mark_recovery_complete(sb, es); 6455 } 6456 } else { 6457 /* Make sure we can mount this feature set readwrite */ 6458 if (ext4_has_feature_readonly(sb) || 6459 !ext4_feature_set_ok(sb, 0)) { 6460 err = -EROFS; 6461 goto restore_opts; 6462 } 6463 /* 6464 * Make sure the group descriptor checksums 6465 * are sane. If they aren't, refuse to remount r/w. 6466 */ 6467 for (g = 0; g < sbi->s_groups_count; g++) { 6468 struct ext4_group_desc *gdp = 6469 ext4_get_group_desc(sb, g, NULL); 6470 6471 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6472 ext4_msg(sb, KERN_ERR, 6473 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6474 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6475 le16_to_cpu(gdp->bg_checksum)); 6476 err = -EFSBADCRC; 6477 goto restore_opts; 6478 } 6479 } 6480 6481 /* 6482 * If we have an unprocessed orphan list hanging 6483 * around from a previously readonly bdev mount, 6484 * require a full umount/remount for now. 6485 */ 6486 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6487 ext4_msg(sb, KERN_WARNING, "Couldn't " 6488 "remount RDWR because of unprocessed " 6489 "orphan inode list. Please " 6490 "umount/remount instead"); 6491 err = -EINVAL; 6492 goto restore_opts; 6493 } 6494 6495 /* 6496 * Mounting a RDONLY partition read-write, so reread 6497 * and store the current valid flag. (It may have 6498 * been changed by e2fsck since we originally mounted 6499 * the partition.) 6500 */ 6501 if (sbi->s_journal) { 6502 err = ext4_clear_journal_err(sb, es); 6503 if (err) 6504 goto restore_opts; 6505 } 6506 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6507 ~EXT4_FC_REPLAY); 6508 6509 err = ext4_setup_super(sb, es, 0); 6510 if (err) 6511 goto restore_opts; 6512 6513 sb->s_flags &= ~SB_RDONLY; 6514 if (ext4_has_feature_mmp(sb)) 6515 if (ext4_multi_mount_protect(sb, 6516 le64_to_cpu(es->s_mmp_block))) { 6517 err = -EROFS; 6518 goto restore_opts; 6519 } 6520 #ifdef CONFIG_QUOTA 6521 enable_quota = 1; 6522 #endif 6523 } 6524 } 6525 6526 /* 6527 * Reinitialize lazy itable initialization thread based on 6528 * current settings 6529 */ 6530 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6531 ext4_unregister_li_request(sb); 6532 else { 6533 ext4_group_t first_not_zeroed; 6534 first_not_zeroed = ext4_has_uninit_itable(sb); 6535 ext4_register_li_request(sb, first_not_zeroed); 6536 } 6537 6538 /* 6539 * Handle creation of system zone data early because it can fail. 6540 * Releasing of existing data is done when we are sure remount will 6541 * succeed. 6542 */ 6543 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6544 err = ext4_setup_system_zone(sb); 6545 if (err) 6546 goto restore_opts; 6547 } 6548 6549 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6550 err = ext4_commit_super(sb); 6551 if (err) 6552 goto restore_opts; 6553 } 6554 6555 #ifdef CONFIG_QUOTA 6556 /* Release old quota file names */ 6557 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6558 kfree(old_opts.s_qf_names[i]); 6559 if (enable_quota) { 6560 if (sb_any_quota_suspended(sb)) 6561 dquot_resume(sb, -1); 6562 else if (ext4_has_feature_quota(sb)) { 6563 err = ext4_enable_quotas(sb); 6564 if (err) 6565 goto restore_opts; 6566 } 6567 } 6568 #endif 6569 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6570 ext4_release_system_zone(sb); 6571 6572 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6573 ext4_stop_mmpd(sbi); 6574 6575 return 0; 6576 6577 restore_opts: 6578 sb->s_flags = old_sb_flags; 6579 sbi->s_mount_opt = old_opts.s_mount_opt; 6580 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6581 sbi->s_resuid = old_opts.s_resuid; 6582 sbi->s_resgid = old_opts.s_resgid; 6583 sbi->s_commit_interval = old_opts.s_commit_interval; 6584 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6585 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6586 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6587 ext4_release_system_zone(sb); 6588 #ifdef CONFIG_QUOTA 6589 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6590 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6591 to_free[i] = get_qf_name(sb, sbi, i); 6592 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6593 } 6594 synchronize_rcu(); 6595 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6596 kfree(to_free[i]); 6597 #endif 6598 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6599 ext4_stop_mmpd(sbi); 6600 return err; 6601 } 6602 6603 static int ext4_reconfigure(struct fs_context *fc) 6604 { 6605 struct super_block *sb = fc->root->d_sb; 6606 int ret; 6607 6608 fc->s_fs_info = EXT4_SB(sb); 6609 6610 ret = ext4_check_opt_consistency(fc, sb); 6611 if (ret < 0) 6612 return ret; 6613 6614 ret = __ext4_remount(fc, sb); 6615 if (ret < 0) 6616 return ret; 6617 6618 ext4_msg(sb, KERN_INFO, "re-mounted. Quota mode: %s.", 6619 ext4_quota_mode(sb)); 6620 6621 return 0; 6622 } 6623 6624 #ifdef CONFIG_QUOTA 6625 static int ext4_statfs_project(struct super_block *sb, 6626 kprojid_t projid, struct kstatfs *buf) 6627 { 6628 struct kqid qid; 6629 struct dquot *dquot; 6630 u64 limit; 6631 u64 curblock; 6632 6633 qid = make_kqid_projid(projid); 6634 dquot = dqget(sb, qid); 6635 if (IS_ERR(dquot)) 6636 return PTR_ERR(dquot); 6637 spin_lock(&dquot->dq_dqb_lock); 6638 6639 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6640 dquot->dq_dqb.dqb_bhardlimit); 6641 limit >>= sb->s_blocksize_bits; 6642 6643 if (limit && buf->f_blocks > limit) { 6644 curblock = (dquot->dq_dqb.dqb_curspace + 6645 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6646 buf->f_blocks = limit; 6647 buf->f_bfree = buf->f_bavail = 6648 (buf->f_blocks > curblock) ? 6649 (buf->f_blocks - curblock) : 0; 6650 } 6651 6652 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6653 dquot->dq_dqb.dqb_ihardlimit); 6654 if (limit && buf->f_files > limit) { 6655 buf->f_files = limit; 6656 buf->f_ffree = 6657 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6658 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6659 } 6660 6661 spin_unlock(&dquot->dq_dqb_lock); 6662 dqput(dquot); 6663 return 0; 6664 } 6665 #endif 6666 6667 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6668 { 6669 struct super_block *sb = dentry->d_sb; 6670 struct ext4_sb_info *sbi = EXT4_SB(sb); 6671 struct ext4_super_block *es = sbi->s_es; 6672 ext4_fsblk_t overhead = 0, resv_blocks; 6673 s64 bfree; 6674 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6675 6676 if (!test_opt(sb, MINIX_DF)) 6677 overhead = sbi->s_overhead; 6678 6679 buf->f_type = EXT4_SUPER_MAGIC; 6680 buf->f_bsize = sb->s_blocksize; 6681 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6682 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6683 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6684 /* prevent underflow in case that few free space is available */ 6685 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6686 buf->f_bavail = buf->f_bfree - 6687 (ext4_r_blocks_count(es) + resv_blocks); 6688 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6689 buf->f_bavail = 0; 6690 buf->f_files = le32_to_cpu(es->s_inodes_count); 6691 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6692 buf->f_namelen = EXT4_NAME_LEN; 6693 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6694 6695 #ifdef CONFIG_QUOTA 6696 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6697 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6698 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6699 #endif 6700 return 0; 6701 } 6702 6703 6704 #ifdef CONFIG_QUOTA 6705 6706 /* 6707 * Helper functions so that transaction is started before we acquire dqio_sem 6708 * to keep correct lock ordering of transaction > dqio_sem 6709 */ 6710 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6711 { 6712 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6713 } 6714 6715 static int ext4_write_dquot(struct dquot *dquot) 6716 { 6717 int ret, err; 6718 handle_t *handle; 6719 struct inode *inode; 6720 6721 inode = dquot_to_inode(dquot); 6722 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6723 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6724 if (IS_ERR(handle)) 6725 return PTR_ERR(handle); 6726 ret = dquot_commit(dquot); 6727 err = ext4_journal_stop(handle); 6728 if (!ret) 6729 ret = err; 6730 return ret; 6731 } 6732 6733 static int ext4_acquire_dquot(struct dquot *dquot) 6734 { 6735 int ret, err; 6736 handle_t *handle; 6737 6738 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6739 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6740 if (IS_ERR(handle)) 6741 return PTR_ERR(handle); 6742 ret = dquot_acquire(dquot); 6743 err = ext4_journal_stop(handle); 6744 if (!ret) 6745 ret = err; 6746 return ret; 6747 } 6748 6749 static int ext4_release_dquot(struct dquot *dquot) 6750 { 6751 int ret, err; 6752 handle_t *handle; 6753 6754 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6755 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6756 if (IS_ERR(handle)) { 6757 /* Release dquot anyway to avoid endless cycle in dqput() */ 6758 dquot_release(dquot); 6759 return PTR_ERR(handle); 6760 } 6761 ret = dquot_release(dquot); 6762 err = ext4_journal_stop(handle); 6763 if (!ret) 6764 ret = err; 6765 return ret; 6766 } 6767 6768 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6769 { 6770 struct super_block *sb = dquot->dq_sb; 6771 6772 if (ext4_is_quota_journalled(sb)) { 6773 dquot_mark_dquot_dirty(dquot); 6774 return ext4_write_dquot(dquot); 6775 } else { 6776 return dquot_mark_dquot_dirty(dquot); 6777 } 6778 } 6779 6780 static int ext4_write_info(struct super_block *sb, int type) 6781 { 6782 int ret, err; 6783 handle_t *handle; 6784 6785 /* Data block + inode block */ 6786 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 6787 if (IS_ERR(handle)) 6788 return PTR_ERR(handle); 6789 ret = dquot_commit_info(sb, type); 6790 err = ext4_journal_stop(handle); 6791 if (!ret) 6792 ret = err; 6793 return ret; 6794 } 6795 6796 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6797 { 6798 struct ext4_inode_info *ei = EXT4_I(inode); 6799 6800 /* The first argument of lockdep_set_subclass has to be 6801 * *exactly* the same as the argument to init_rwsem() --- in 6802 * this case, in init_once() --- or lockdep gets unhappy 6803 * because the name of the lock is set using the 6804 * stringification of the argument to init_rwsem(). 6805 */ 6806 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6807 lockdep_set_subclass(&ei->i_data_sem, subclass); 6808 } 6809 6810 /* 6811 * Standard function to be called on quota_on 6812 */ 6813 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6814 const struct path *path) 6815 { 6816 int err; 6817 6818 if (!test_opt(sb, QUOTA)) 6819 return -EINVAL; 6820 6821 /* Quotafile not on the same filesystem? */ 6822 if (path->dentry->d_sb != sb) 6823 return -EXDEV; 6824 6825 /* Quota already enabled for this file? */ 6826 if (IS_NOQUOTA(d_inode(path->dentry))) 6827 return -EBUSY; 6828 6829 /* Journaling quota? */ 6830 if (EXT4_SB(sb)->s_qf_names[type]) { 6831 /* Quotafile not in fs root? */ 6832 if (path->dentry->d_parent != sb->s_root) 6833 ext4_msg(sb, KERN_WARNING, 6834 "Quota file not on filesystem root. " 6835 "Journaled quota will not work"); 6836 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6837 } else { 6838 /* 6839 * Clear the flag just in case mount options changed since 6840 * last time. 6841 */ 6842 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6843 } 6844 6845 /* 6846 * When we journal data on quota file, we have to flush journal to see 6847 * all updates to the file when we bypass pagecache... 6848 */ 6849 if (EXT4_SB(sb)->s_journal && 6850 ext4_should_journal_data(d_inode(path->dentry))) { 6851 /* 6852 * We don't need to lock updates but journal_flush() could 6853 * otherwise be livelocked... 6854 */ 6855 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6856 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0); 6857 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6858 if (err) 6859 return err; 6860 } 6861 6862 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6863 err = dquot_quota_on(sb, type, format_id, path); 6864 if (!err) { 6865 struct inode *inode = d_inode(path->dentry); 6866 handle_t *handle; 6867 6868 /* 6869 * Set inode flags to prevent userspace from messing with quota 6870 * files. If this fails, we return success anyway since quotas 6871 * are already enabled and this is not a hard failure. 6872 */ 6873 inode_lock(inode); 6874 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6875 if (IS_ERR(handle)) 6876 goto unlock_inode; 6877 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6878 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6879 S_NOATIME | S_IMMUTABLE); 6880 err = ext4_mark_inode_dirty(handle, inode); 6881 ext4_journal_stop(handle); 6882 unlock_inode: 6883 inode_unlock(inode); 6884 if (err) 6885 dquot_quota_off(sb, type); 6886 } 6887 if (err) 6888 lockdep_set_quota_inode(path->dentry->d_inode, 6889 I_DATA_SEM_NORMAL); 6890 return err; 6891 } 6892 6893 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6894 unsigned int flags) 6895 { 6896 int err; 6897 struct inode *qf_inode; 6898 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6899 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6900 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6901 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6902 }; 6903 6904 BUG_ON(!ext4_has_feature_quota(sb)); 6905 6906 if (!qf_inums[type]) 6907 return -EPERM; 6908 6909 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6910 if (IS_ERR(qf_inode)) { 6911 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6912 return PTR_ERR(qf_inode); 6913 } 6914 6915 /* Don't account quota for quota files to avoid recursion */ 6916 qf_inode->i_flags |= S_NOQUOTA; 6917 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6918 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6919 if (err) 6920 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6921 iput(qf_inode); 6922 6923 return err; 6924 } 6925 6926 /* Enable usage tracking for all quota types. */ 6927 int ext4_enable_quotas(struct super_block *sb) 6928 { 6929 int type, err = 0; 6930 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6931 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6932 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6933 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6934 }; 6935 bool quota_mopt[EXT4_MAXQUOTAS] = { 6936 test_opt(sb, USRQUOTA), 6937 test_opt(sb, GRPQUOTA), 6938 test_opt(sb, PRJQUOTA), 6939 }; 6940 6941 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6942 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6943 if (qf_inums[type]) { 6944 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6945 DQUOT_USAGE_ENABLED | 6946 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6947 if (err) { 6948 ext4_warning(sb, 6949 "Failed to enable quota tracking " 6950 "(type=%d, err=%d). Please run " 6951 "e2fsck to fix.", type, err); 6952 for (type--; type >= 0; type--) { 6953 struct inode *inode; 6954 6955 inode = sb_dqopt(sb)->files[type]; 6956 if (inode) 6957 inode = igrab(inode); 6958 dquot_quota_off(sb, type); 6959 if (inode) { 6960 lockdep_set_quota_inode(inode, 6961 I_DATA_SEM_NORMAL); 6962 iput(inode); 6963 } 6964 } 6965 6966 return err; 6967 } 6968 } 6969 } 6970 return 0; 6971 } 6972 6973 static int ext4_quota_off(struct super_block *sb, int type) 6974 { 6975 struct inode *inode = sb_dqopt(sb)->files[type]; 6976 handle_t *handle; 6977 int err; 6978 6979 /* Force all delayed allocation blocks to be allocated. 6980 * Caller already holds s_umount sem */ 6981 if (test_opt(sb, DELALLOC)) 6982 sync_filesystem(sb); 6983 6984 if (!inode || !igrab(inode)) 6985 goto out; 6986 6987 err = dquot_quota_off(sb, type); 6988 if (err || ext4_has_feature_quota(sb)) 6989 goto out_put; 6990 6991 inode_lock(inode); 6992 /* 6993 * Update modification times of quota files when userspace can 6994 * start looking at them. If we fail, we return success anyway since 6995 * this is not a hard failure and quotas are already disabled. 6996 */ 6997 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6998 if (IS_ERR(handle)) { 6999 err = PTR_ERR(handle); 7000 goto out_unlock; 7001 } 7002 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 7003 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 7004 inode->i_mtime = inode->i_ctime = current_time(inode); 7005 err = ext4_mark_inode_dirty(handle, inode); 7006 ext4_journal_stop(handle); 7007 out_unlock: 7008 inode_unlock(inode); 7009 out_put: 7010 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7011 iput(inode); 7012 return err; 7013 out: 7014 return dquot_quota_off(sb, type); 7015 } 7016 7017 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7018 * acquiring the locks... As quota files are never truncated and quota code 7019 * itself serializes the operations (and no one else should touch the files) 7020 * we don't have to be afraid of races */ 7021 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7022 size_t len, loff_t off) 7023 { 7024 struct inode *inode = sb_dqopt(sb)->files[type]; 7025 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7026 int offset = off & (sb->s_blocksize - 1); 7027 int tocopy; 7028 size_t toread; 7029 struct buffer_head *bh; 7030 loff_t i_size = i_size_read(inode); 7031 7032 if (off > i_size) 7033 return 0; 7034 if (off+len > i_size) 7035 len = i_size-off; 7036 toread = len; 7037 while (toread > 0) { 7038 tocopy = sb->s_blocksize - offset < toread ? 7039 sb->s_blocksize - offset : toread; 7040 bh = ext4_bread(NULL, inode, blk, 0); 7041 if (IS_ERR(bh)) 7042 return PTR_ERR(bh); 7043 if (!bh) /* A hole? */ 7044 memset(data, 0, tocopy); 7045 else 7046 memcpy(data, bh->b_data+offset, tocopy); 7047 brelse(bh); 7048 offset = 0; 7049 toread -= tocopy; 7050 data += tocopy; 7051 blk++; 7052 } 7053 return len; 7054 } 7055 7056 /* Write to quotafile (we know the transaction is already started and has 7057 * enough credits) */ 7058 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7059 const char *data, size_t len, loff_t off) 7060 { 7061 struct inode *inode = sb_dqopt(sb)->files[type]; 7062 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7063 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7064 int retries = 0; 7065 struct buffer_head *bh; 7066 handle_t *handle = journal_current_handle(); 7067 7068 if (!handle) { 7069 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7070 " cancelled because transaction is not started", 7071 (unsigned long long)off, (unsigned long long)len); 7072 return -EIO; 7073 } 7074 /* 7075 * Since we account only one data block in transaction credits, 7076 * then it is impossible to cross a block boundary. 7077 */ 7078 if (sb->s_blocksize - offset < len) { 7079 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7080 " cancelled because not block aligned", 7081 (unsigned long long)off, (unsigned long long)len); 7082 return -EIO; 7083 } 7084 7085 do { 7086 bh = ext4_bread(handle, inode, blk, 7087 EXT4_GET_BLOCKS_CREATE | 7088 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7089 } while (PTR_ERR(bh) == -ENOSPC && 7090 ext4_should_retry_alloc(inode->i_sb, &retries)); 7091 if (IS_ERR(bh)) 7092 return PTR_ERR(bh); 7093 if (!bh) 7094 goto out; 7095 BUFFER_TRACE(bh, "get write access"); 7096 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7097 if (err) { 7098 brelse(bh); 7099 return err; 7100 } 7101 lock_buffer(bh); 7102 memcpy(bh->b_data+offset, data, len); 7103 flush_dcache_page(bh->b_page); 7104 unlock_buffer(bh); 7105 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7106 brelse(bh); 7107 out: 7108 if (inode->i_size < off + len) { 7109 i_size_write(inode, off + len); 7110 EXT4_I(inode)->i_disksize = inode->i_size; 7111 err2 = ext4_mark_inode_dirty(handle, inode); 7112 if (unlikely(err2 && !err)) 7113 err = err2; 7114 } 7115 return err ? err : len; 7116 } 7117 #endif 7118 7119 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7120 static inline void register_as_ext2(void) 7121 { 7122 int err = register_filesystem(&ext2_fs_type); 7123 if (err) 7124 printk(KERN_WARNING 7125 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7126 } 7127 7128 static inline void unregister_as_ext2(void) 7129 { 7130 unregister_filesystem(&ext2_fs_type); 7131 } 7132 7133 static inline int ext2_feature_set_ok(struct super_block *sb) 7134 { 7135 if (ext4_has_unknown_ext2_incompat_features(sb)) 7136 return 0; 7137 if (sb_rdonly(sb)) 7138 return 1; 7139 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7140 return 0; 7141 return 1; 7142 } 7143 #else 7144 static inline void register_as_ext2(void) { } 7145 static inline void unregister_as_ext2(void) { } 7146 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7147 #endif 7148 7149 static inline void register_as_ext3(void) 7150 { 7151 int err = register_filesystem(&ext3_fs_type); 7152 if (err) 7153 printk(KERN_WARNING 7154 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7155 } 7156 7157 static inline void unregister_as_ext3(void) 7158 { 7159 unregister_filesystem(&ext3_fs_type); 7160 } 7161 7162 static inline int ext3_feature_set_ok(struct super_block *sb) 7163 { 7164 if (ext4_has_unknown_ext3_incompat_features(sb)) 7165 return 0; 7166 if (!ext4_has_feature_journal(sb)) 7167 return 0; 7168 if (sb_rdonly(sb)) 7169 return 1; 7170 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7171 return 0; 7172 return 1; 7173 } 7174 7175 static struct file_system_type ext4_fs_type = { 7176 .owner = THIS_MODULE, 7177 .name = "ext4", 7178 .init_fs_context = ext4_init_fs_context, 7179 .parameters = ext4_param_specs, 7180 .kill_sb = kill_block_super, 7181 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 7182 }; 7183 MODULE_ALIAS_FS("ext4"); 7184 7185 /* Shared across all ext4 file systems */ 7186 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 7187 7188 static int __init ext4_init_fs(void) 7189 { 7190 int i, err; 7191 7192 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7193 ext4_li_info = NULL; 7194 7195 /* Build-time check for flags consistency */ 7196 ext4_check_flag_values(); 7197 7198 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 7199 init_waitqueue_head(&ext4__ioend_wq[i]); 7200 7201 err = ext4_init_es(); 7202 if (err) 7203 return err; 7204 7205 err = ext4_init_pending(); 7206 if (err) 7207 goto out7; 7208 7209 err = ext4_init_post_read_processing(); 7210 if (err) 7211 goto out6; 7212 7213 err = ext4_init_pageio(); 7214 if (err) 7215 goto out5; 7216 7217 err = ext4_init_system_zone(); 7218 if (err) 7219 goto out4; 7220 7221 err = ext4_init_sysfs(); 7222 if (err) 7223 goto out3; 7224 7225 err = ext4_init_mballoc(); 7226 if (err) 7227 goto out2; 7228 err = init_inodecache(); 7229 if (err) 7230 goto out1; 7231 7232 err = ext4_fc_init_dentry_cache(); 7233 if (err) 7234 goto out05; 7235 7236 register_as_ext3(); 7237 register_as_ext2(); 7238 err = register_filesystem(&ext4_fs_type); 7239 if (err) 7240 goto out; 7241 7242 return 0; 7243 out: 7244 unregister_as_ext2(); 7245 unregister_as_ext3(); 7246 ext4_fc_destroy_dentry_cache(); 7247 out05: 7248 destroy_inodecache(); 7249 out1: 7250 ext4_exit_mballoc(); 7251 out2: 7252 ext4_exit_sysfs(); 7253 out3: 7254 ext4_exit_system_zone(); 7255 out4: 7256 ext4_exit_pageio(); 7257 out5: 7258 ext4_exit_post_read_processing(); 7259 out6: 7260 ext4_exit_pending(); 7261 out7: 7262 ext4_exit_es(); 7263 7264 return err; 7265 } 7266 7267 static void __exit ext4_exit_fs(void) 7268 { 7269 ext4_destroy_lazyinit_thread(); 7270 unregister_as_ext2(); 7271 unregister_as_ext3(); 7272 unregister_filesystem(&ext4_fs_type); 7273 ext4_fc_destroy_dentry_cache(); 7274 destroy_inodecache(); 7275 ext4_exit_mballoc(); 7276 ext4_exit_sysfs(); 7277 ext4_exit_system_zone(); 7278 ext4_exit_pageio(); 7279 ext4_exit_post_read_processing(); 7280 ext4_exit_es(); 7281 ext4_exit_pending(); 7282 } 7283 7284 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7285 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7286 MODULE_LICENSE("GPL"); 7287 MODULE_SOFTDEP("pre: crc32c"); 7288 module_init(ext4_init_fs) 7289 module_exit(ext4_exit_fs) 7290