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/cleancache.h> 43 #include <linux/uaccess.h> 44 #include <linux/iversion.h> 45 #include <linux/unicode.h> 46 #include <linux/part_stat.h> 47 #include <linux/kthread.h> 48 #include <linux/freezer.h> 49 50 #include "ext4.h" 51 #include "ext4_extents.h" /* Needed for trace points definition */ 52 #include "ext4_jbd2.h" 53 #include "xattr.h" 54 #include "acl.h" 55 #include "mballoc.h" 56 #include "fsmap.h" 57 58 #define CREATE_TRACE_POINTS 59 #include <trace/events/ext4.h> 60 61 static struct ext4_lazy_init *ext4_li_info; 62 static DEFINE_MUTEX(ext4_li_mtx); 63 static struct ratelimit_state ext4_mount_msg_ratelimit; 64 65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 66 unsigned long journal_devnum); 67 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 68 static void ext4_update_super(struct super_block *sb); 69 static int ext4_commit_super(struct super_block *sb); 70 static int ext4_mark_recovery_complete(struct super_block *sb, 71 struct ext4_super_block *es); 72 static int ext4_clear_journal_err(struct super_block *sb, 73 struct ext4_super_block *es); 74 static int ext4_sync_fs(struct super_block *sb, int wait); 75 static int ext4_remount(struct super_block *sb, int *flags, char *data); 76 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 77 static int ext4_unfreeze(struct super_block *sb); 78 static int ext4_freeze(struct super_block *sb); 79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 80 const char *dev_name, void *data); 81 static inline int ext2_feature_set_ok(struct super_block *sb); 82 static inline int ext3_feature_set_ok(struct super_block *sb); 83 static int ext4_feature_set_ok(struct super_block *sb, int readonly); 84 static void ext4_destroy_lazyinit_thread(void); 85 static void ext4_unregister_li_request(struct super_block *sb); 86 static void ext4_clear_request_list(void); 87 static struct inode *ext4_get_journal_inode(struct super_block *sb, 88 unsigned int journal_inum); 89 90 /* 91 * Lock ordering 92 * 93 * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and 94 * i_mmap_rwsem (inode->i_mmap_rwsem)! 95 * 96 * page fault path: 97 * mmap_lock -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start -> 98 * page lock -> i_data_sem (rw) 99 * 100 * buffered write path: 101 * sb_start_write -> i_mutex -> mmap_lock 102 * sb_start_write -> i_mutex -> transaction start -> page lock -> 103 * i_data_sem (rw) 104 * 105 * truncate: 106 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock 107 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start -> 108 * i_data_sem (rw) 109 * 110 * direct IO: 111 * sb_start_write -> i_mutex -> mmap_lock 112 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 113 * 114 * writepages: 115 * transaction start -> page lock(s) -> i_data_sem (rw) 116 */ 117 118 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 119 static struct file_system_type ext2_fs_type = { 120 .owner = THIS_MODULE, 121 .name = "ext2", 122 .mount = ext4_mount, 123 .kill_sb = kill_block_super, 124 .fs_flags = FS_REQUIRES_DEV, 125 }; 126 MODULE_ALIAS_FS("ext2"); 127 MODULE_ALIAS("ext2"); 128 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type) 129 #else 130 #define IS_EXT2_SB(sb) (0) 131 #endif 132 133 134 static struct file_system_type ext3_fs_type = { 135 .owner = THIS_MODULE, 136 .name = "ext3", 137 .mount = ext4_mount, 138 .kill_sb = kill_block_super, 139 .fs_flags = FS_REQUIRES_DEV, 140 }; 141 MODULE_ALIAS_FS("ext3"); 142 MODULE_ALIAS("ext3"); 143 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type) 144 145 146 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags, 147 bh_end_io_t *end_io) 148 { 149 /* 150 * buffer's verified bit is no longer valid after reading from 151 * disk again due to write out error, clear it to make sure we 152 * recheck the buffer contents. 153 */ 154 clear_buffer_verified(bh); 155 156 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 157 get_bh(bh); 158 submit_bh(REQ_OP_READ, op_flags, bh); 159 } 160 161 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags, 162 bh_end_io_t *end_io) 163 { 164 BUG_ON(!buffer_locked(bh)); 165 166 if (ext4_buffer_uptodate(bh)) { 167 unlock_buffer(bh); 168 return; 169 } 170 __ext4_read_bh(bh, op_flags, end_io); 171 } 172 173 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io) 174 { 175 BUG_ON(!buffer_locked(bh)); 176 177 if (ext4_buffer_uptodate(bh)) { 178 unlock_buffer(bh); 179 return 0; 180 } 181 182 __ext4_read_bh(bh, op_flags, end_io); 183 184 wait_on_buffer(bh); 185 if (buffer_uptodate(bh)) 186 return 0; 187 return -EIO; 188 } 189 190 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait) 191 { 192 if (trylock_buffer(bh)) { 193 if (wait) 194 return ext4_read_bh(bh, op_flags, NULL); 195 ext4_read_bh_nowait(bh, op_flags, NULL); 196 return 0; 197 } 198 if (wait) { 199 wait_on_buffer(bh); 200 if (buffer_uptodate(bh)) 201 return 0; 202 return -EIO; 203 } 204 return 0; 205 } 206 207 /* 208 * This works like __bread_gfp() except it uses ERR_PTR for error 209 * returns. Currently with sb_bread it's impossible to distinguish 210 * between ENOMEM and EIO situations (since both result in a NULL 211 * return. 212 */ 213 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 214 sector_t block, int op_flags, 215 gfp_t gfp) 216 { 217 struct buffer_head *bh; 218 int ret; 219 220 bh = sb_getblk_gfp(sb, block, gfp); 221 if (bh == NULL) 222 return ERR_PTR(-ENOMEM); 223 if (ext4_buffer_uptodate(bh)) 224 return bh; 225 226 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 227 if (ret) { 228 put_bh(bh); 229 return ERR_PTR(ret); 230 } 231 return bh; 232 } 233 234 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 235 int op_flags) 236 { 237 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 238 } 239 240 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 241 sector_t block) 242 { 243 return __ext4_sb_bread_gfp(sb, block, 0, 0); 244 } 245 246 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 247 { 248 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 249 250 if (likely(bh)) { 251 ext4_read_bh_lock(bh, REQ_RAHEAD, false); 252 brelse(bh); 253 } 254 } 255 256 static int ext4_verify_csum_type(struct super_block *sb, 257 struct ext4_super_block *es) 258 { 259 if (!ext4_has_feature_metadata_csum(sb)) 260 return 1; 261 262 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 263 } 264 265 static __le32 ext4_superblock_csum(struct super_block *sb, 266 struct ext4_super_block *es) 267 { 268 struct ext4_sb_info *sbi = EXT4_SB(sb); 269 int offset = offsetof(struct ext4_super_block, s_checksum); 270 __u32 csum; 271 272 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 273 274 return cpu_to_le32(csum); 275 } 276 277 static int ext4_superblock_csum_verify(struct super_block *sb, 278 struct ext4_super_block *es) 279 { 280 if (!ext4_has_metadata_csum(sb)) 281 return 1; 282 283 return es->s_checksum == ext4_superblock_csum(sb, es); 284 } 285 286 void ext4_superblock_csum_set(struct super_block *sb) 287 { 288 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 289 290 if (!ext4_has_metadata_csum(sb)) 291 return; 292 293 es->s_checksum = ext4_superblock_csum(sb, es); 294 } 295 296 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 297 struct ext4_group_desc *bg) 298 { 299 return le32_to_cpu(bg->bg_block_bitmap_lo) | 300 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 301 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 302 } 303 304 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 305 struct ext4_group_desc *bg) 306 { 307 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 308 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 309 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 310 } 311 312 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 313 struct ext4_group_desc *bg) 314 { 315 return le32_to_cpu(bg->bg_inode_table_lo) | 316 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 317 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 318 } 319 320 __u32 ext4_free_group_clusters(struct super_block *sb, 321 struct ext4_group_desc *bg) 322 { 323 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 324 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 325 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 326 } 327 328 __u32 ext4_free_inodes_count(struct super_block *sb, 329 struct ext4_group_desc *bg) 330 { 331 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 332 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 333 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 334 } 335 336 __u32 ext4_used_dirs_count(struct super_block *sb, 337 struct ext4_group_desc *bg) 338 { 339 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 340 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 341 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 342 } 343 344 __u32 ext4_itable_unused_count(struct super_block *sb, 345 struct ext4_group_desc *bg) 346 { 347 return le16_to_cpu(bg->bg_itable_unused_lo) | 348 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 349 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 350 } 351 352 void ext4_block_bitmap_set(struct super_block *sb, 353 struct ext4_group_desc *bg, ext4_fsblk_t blk) 354 { 355 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 356 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 357 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 358 } 359 360 void ext4_inode_bitmap_set(struct super_block *sb, 361 struct ext4_group_desc *bg, ext4_fsblk_t blk) 362 { 363 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 364 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 365 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 366 } 367 368 void ext4_inode_table_set(struct super_block *sb, 369 struct ext4_group_desc *bg, ext4_fsblk_t blk) 370 { 371 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 372 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 373 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 374 } 375 376 void ext4_free_group_clusters_set(struct super_block *sb, 377 struct ext4_group_desc *bg, __u32 count) 378 { 379 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 380 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 381 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 382 } 383 384 void ext4_free_inodes_set(struct super_block *sb, 385 struct ext4_group_desc *bg, __u32 count) 386 { 387 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 388 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 389 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 390 } 391 392 void ext4_used_dirs_set(struct super_block *sb, 393 struct ext4_group_desc *bg, __u32 count) 394 { 395 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 396 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 397 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 398 } 399 400 void ext4_itable_unused_set(struct super_block *sb, 401 struct ext4_group_desc *bg, __u32 count) 402 { 403 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 404 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 405 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 406 } 407 408 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 409 { 410 now = clamp_val(now, 0, (1ull << 40) - 1); 411 412 *lo = cpu_to_le32(lower_32_bits(now)); 413 *hi = upper_32_bits(now); 414 } 415 416 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 417 { 418 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 419 } 420 #define ext4_update_tstamp(es, tstamp) \ 421 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 422 ktime_get_real_seconds()) 423 #define ext4_get_tstamp(es, tstamp) \ 424 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 425 426 /* 427 * The del_gendisk() function uninitializes the disk-specific data 428 * structures, including the bdi structure, without telling anyone 429 * else. Once this happens, any attempt to call mark_buffer_dirty() 430 * (for example, by ext4_commit_super), will cause a kernel OOPS. 431 * This is a kludge to prevent these oops until we can put in a proper 432 * hook in del_gendisk() to inform the VFS and file system layers. 433 */ 434 static int block_device_ejected(struct super_block *sb) 435 { 436 struct inode *bd_inode = sb->s_bdev->bd_inode; 437 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 438 439 return bdi->dev == NULL; 440 } 441 442 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 443 { 444 struct super_block *sb = journal->j_private; 445 struct ext4_sb_info *sbi = EXT4_SB(sb); 446 int error = is_journal_aborted(journal); 447 struct ext4_journal_cb_entry *jce; 448 449 BUG_ON(txn->t_state == T_FINISHED); 450 451 ext4_process_freed_data(sb, txn->t_tid); 452 453 spin_lock(&sbi->s_md_lock); 454 while (!list_empty(&txn->t_private_list)) { 455 jce = list_entry(txn->t_private_list.next, 456 struct ext4_journal_cb_entry, jce_list); 457 list_del_init(&jce->jce_list); 458 spin_unlock(&sbi->s_md_lock); 459 jce->jce_func(sb, jce, error); 460 spin_lock(&sbi->s_md_lock); 461 } 462 spin_unlock(&sbi->s_md_lock); 463 } 464 465 /* 466 * This writepage callback for write_cache_pages() 467 * takes care of a few cases after page cleaning. 468 * 469 * write_cache_pages() already checks for dirty pages 470 * and calls clear_page_dirty_for_io(), which we want, 471 * to write protect the pages. 472 * 473 * However, we may have to redirty a page (see below.) 474 */ 475 static int ext4_journalled_writepage_callback(struct page *page, 476 struct writeback_control *wbc, 477 void *data) 478 { 479 transaction_t *transaction = (transaction_t *) data; 480 struct buffer_head *bh, *head; 481 struct journal_head *jh; 482 483 bh = head = page_buffers(page); 484 do { 485 /* 486 * We have to redirty a page in these cases: 487 * 1) If buffer is dirty, it means the page was dirty because it 488 * contains a buffer that needs checkpointing. So the dirty bit 489 * needs to be preserved so that checkpointing writes the buffer 490 * properly. 491 * 2) If buffer is not part of the committing transaction 492 * (we may have just accidentally come across this buffer because 493 * inode range tracking is not exact) or if the currently running 494 * transaction already contains this buffer as well, dirty bit 495 * needs to be preserved so that the buffer gets writeprotected 496 * properly on running transaction's commit. 497 */ 498 jh = bh2jh(bh); 499 if (buffer_dirty(bh) || 500 (jh && (jh->b_transaction != transaction || 501 jh->b_next_transaction))) { 502 redirty_page_for_writepage(wbc, page); 503 goto out; 504 } 505 } while ((bh = bh->b_this_page) != head); 506 507 out: 508 return AOP_WRITEPAGE_ACTIVATE; 509 } 510 511 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 512 { 513 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 514 struct writeback_control wbc = { 515 .sync_mode = WB_SYNC_ALL, 516 .nr_to_write = LONG_MAX, 517 .range_start = jinode->i_dirty_start, 518 .range_end = jinode->i_dirty_end, 519 }; 520 521 return write_cache_pages(mapping, &wbc, 522 ext4_journalled_writepage_callback, 523 jinode->i_transaction); 524 } 525 526 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 527 { 528 int ret; 529 530 if (ext4_should_journal_data(jinode->i_vfs_inode)) 531 ret = ext4_journalled_submit_inode_data_buffers(jinode); 532 else 533 ret = jbd2_journal_submit_inode_data_buffers(jinode); 534 535 return ret; 536 } 537 538 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 539 { 540 int ret = 0; 541 542 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 543 ret = jbd2_journal_finish_inode_data_buffers(jinode); 544 545 return ret; 546 } 547 548 static bool system_going_down(void) 549 { 550 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 551 || system_state == SYSTEM_RESTART; 552 } 553 554 struct ext4_err_translation { 555 int code; 556 int errno; 557 }; 558 559 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 560 561 static struct ext4_err_translation err_translation[] = { 562 EXT4_ERR_TRANSLATE(EIO), 563 EXT4_ERR_TRANSLATE(ENOMEM), 564 EXT4_ERR_TRANSLATE(EFSBADCRC), 565 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 566 EXT4_ERR_TRANSLATE(ENOSPC), 567 EXT4_ERR_TRANSLATE(ENOKEY), 568 EXT4_ERR_TRANSLATE(EROFS), 569 EXT4_ERR_TRANSLATE(EFBIG), 570 EXT4_ERR_TRANSLATE(EEXIST), 571 EXT4_ERR_TRANSLATE(ERANGE), 572 EXT4_ERR_TRANSLATE(EOVERFLOW), 573 EXT4_ERR_TRANSLATE(EBUSY), 574 EXT4_ERR_TRANSLATE(ENOTDIR), 575 EXT4_ERR_TRANSLATE(ENOTEMPTY), 576 EXT4_ERR_TRANSLATE(ESHUTDOWN), 577 EXT4_ERR_TRANSLATE(EFAULT), 578 }; 579 580 static int ext4_errno_to_code(int errno) 581 { 582 int i; 583 584 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 585 if (err_translation[i].errno == errno) 586 return err_translation[i].code; 587 return EXT4_ERR_UNKNOWN; 588 } 589 590 static void save_error_info(struct super_block *sb, int error, 591 __u32 ino, __u64 block, 592 const char *func, unsigned int line) 593 { 594 struct ext4_sb_info *sbi = EXT4_SB(sb); 595 596 /* We default to EFSCORRUPTED error... */ 597 if (error == 0) 598 error = EFSCORRUPTED; 599 600 spin_lock(&sbi->s_error_lock); 601 sbi->s_add_error_count++; 602 sbi->s_last_error_code = error; 603 sbi->s_last_error_line = line; 604 sbi->s_last_error_ino = ino; 605 sbi->s_last_error_block = block; 606 sbi->s_last_error_func = func; 607 sbi->s_last_error_time = ktime_get_real_seconds(); 608 if (!sbi->s_first_error_time) { 609 sbi->s_first_error_code = error; 610 sbi->s_first_error_line = line; 611 sbi->s_first_error_ino = ino; 612 sbi->s_first_error_block = block; 613 sbi->s_first_error_func = func; 614 sbi->s_first_error_time = sbi->s_last_error_time; 615 } 616 spin_unlock(&sbi->s_error_lock); 617 } 618 619 /* Deal with the reporting of failure conditions on a filesystem such as 620 * inconsistencies detected or read IO failures. 621 * 622 * On ext2, we can store the error state of the filesystem in the 623 * superblock. That is not possible on ext4, because we may have other 624 * write ordering constraints on the superblock which prevent us from 625 * writing it out straight away; and given that the journal is about to 626 * be aborted, we can't rely on the current, or future, transactions to 627 * write out the superblock safely. 628 * 629 * We'll just use the jbd2_journal_abort() error code to record an error in 630 * the journal instead. On recovery, the journal will complain about 631 * that error until we've noted it down and cleared it. 632 * 633 * If force_ro is set, we unconditionally force the filesystem into an 634 * ABORT|READONLY state, unless the error response on the fs has been set to 635 * panic in which case we take the easy way out and panic immediately. This is 636 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 637 * at a critical moment in log management. 638 */ 639 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 640 __u32 ino, __u64 block, 641 const char *func, unsigned int line) 642 { 643 journal_t *journal = EXT4_SB(sb)->s_journal; 644 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 645 646 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 647 if (test_opt(sb, WARN_ON_ERROR)) 648 WARN_ON_ONCE(1); 649 650 if (!continue_fs && !sb_rdonly(sb)) { 651 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 652 if (journal) 653 jbd2_journal_abort(journal, -EIO); 654 } 655 656 if (!bdev_read_only(sb->s_bdev)) { 657 save_error_info(sb, error, ino, block, func, line); 658 /* 659 * In case the fs should keep running, we need to writeout 660 * superblock through the journal. Due to lock ordering 661 * constraints, it may not be safe to do it right here so we 662 * defer superblock flushing to a workqueue. 663 */ 664 if (continue_fs) 665 schedule_work(&EXT4_SB(sb)->s_error_work); 666 else 667 ext4_commit_super(sb); 668 } 669 670 if (sb_rdonly(sb) || continue_fs) 671 return; 672 673 /* 674 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 675 * could panic during 'reboot -f' as the underlying device got already 676 * disabled. 677 */ 678 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 679 panic("EXT4-fs (device %s): panic forced after error\n", 680 sb->s_id); 681 } 682 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 683 /* 684 * Make sure updated value of ->s_mount_flags will be visible before 685 * ->s_flags update 686 */ 687 smp_wmb(); 688 sb->s_flags |= SB_RDONLY; 689 } 690 691 static void flush_stashed_error_work(struct work_struct *work) 692 { 693 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 694 s_error_work); 695 journal_t *journal = sbi->s_journal; 696 handle_t *handle; 697 698 /* 699 * If the journal is still running, we have to write out superblock 700 * through the journal to avoid collisions of other journalled sb 701 * updates. 702 * 703 * We use directly jbd2 functions here to avoid recursing back into 704 * ext4 error handling code during handling of previous errors. 705 */ 706 if (!sb_rdonly(sbi->s_sb) && journal) { 707 handle = jbd2_journal_start(journal, 1); 708 if (IS_ERR(handle)) 709 goto write_directly; 710 if (jbd2_journal_get_write_access(handle, sbi->s_sbh)) { 711 jbd2_journal_stop(handle); 712 goto write_directly; 713 } 714 ext4_update_super(sbi->s_sb); 715 if (jbd2_journal_dirty_metadata(handle, sbi->s_sbh)) { 716 jbd2_journal_stop(handle); 717 goto write_directly; 718 } 719 jbd2_journal_stop(handle); 720 return; 721 } 722 write_directly: 723 /* 724 * Write through journal failed. Write sb directly to get error info 725 * out and hope for the best. 726 */ 727 ext4_commit_super(sbi->s_sb); 728 } 729 730 #define ext4_error_ratelimit(sb) \ 731 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 732 "EXT4-fs error") 733 734 void __ext4_error(struct super_block *sb, const char *function, 735 unsigned int line, bool force_ro, int error, __u64 block, 736 const char *fmt, ...) 737 { 738 struct va_format vaf; 739 va_list args; 740 741 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 742 return; 743 744 trace_ext4_error(sb, function, line); 745 if (ext4_error_ratelimit(sb)) { 746 va_start(args, fmt); 747 vaf.fmt = fmt; 748 vaf.va = &args; 749 printk(KERN_CRIT 750 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 751 sb->s_id, function, line, current->comm, &vaf); 752 va_end(args); 753 } 754 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 755 } 756 757 void __ext4_error_inode(struct inode *inode, const char *function, 758 unsigned int line, ext4_fsblk_t block, int error, 759 const char *fmt, ...) 760 { 761 va_list args; 762 struct va_format vaf; 763 764 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 765 return; 766 767 trace_ext4_error(inode->i_sb, function, line); 768 if (ext4_error_ratelimit(inode->i_sb)) { 769 va_start(args, fmt); 770 vaf.fmt = fmt; 771 vaf.va = &args; 772 if (block) 773 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 774 "inode #%lu: block %llu: comm %s: %pV\n", 775 inode->i_sb->s_id, function, line, inode->i_ino, 776 block, current->comm, &vaf); 777 else 778 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 779 "inode #%lu: comm %s: %pV\n", 780 inode->i_sb->s_id, function, line, inode->i_ino, 781 current->comm, &vaf); 782 va_end(args); 783 } 784 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 785 function, line); 786 } 787 788 void __ext4_error_file(struct file *file, const char *function, 789 unsigned int line, ext4_fsblk_t block, 790 const char *fmt, ...) 791 { 792 va_list args; 793 struct va_format vaf; 794 struct inode *inode = file_inode(file); 795 char pathname[80], *path; 796 797 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 798 return; 799 800 trace_ext4_error(inode->i_sb, function, line); 801 if (ext4_error_ratelimit(inode->i_sb)) { 802 path = file_path(file, pathname, sizeof(pathname)); 803 if (IS_ERR(path)) 804 path = "(unknown)"; 805 va_start(args, fmt); 806 vaf.fmt = fmt; 807 vaf.va = &args; 808 if (block) 809 printk(KERN_CRIT 810 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 811 "block %llu: comm %s: path %s: %pV\n", 812 inode->i_sb->s_id, function, line, inode->i_ino, 813 block, current->comm, path, &vaf); 814 else 815 printk(KERN_CRIT 816 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 817 "comm %s: path %s: %pV\n", 818 inode->i_sb->s_id, function, line, inode->i_ino, 819 current->comm, path, &vaf); 820 va_end(args); 821 } 822 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 823 function, line); 824 } 825 826 const char *ext4_decode_error(struct super_block *sb, int errno, 827 char nbuf[16]) 828 { 829 char *errstr = NULL; 830 831 switch (errno) { 832 case -EFSCORRUPTED: 833 errstr = "Corrupt filesystem"; 834 break; 835 case -EFSBADCRC: 836 errstr = "Filesystem failed CRC"; 837 break; 838 case -EIO: 839 errstr = "IO failure"; 840 break; 841 case -ENOMEM: 842 errstr = "Out of memory"; 843 break; 844 case -EROFS: 845 if (!sb || (EXT4_SB(sb)->s_journal && 846 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 847 errstr = "Journal has aborted"; 848 else 849 errstr = "Readonly filesystem"; 850 break; 851 default: 852 /* If the caller passed in an extra buffer for unknown 853 * errors, textualise them now. Else we just return 854 * NULL. */ 855 if (nbuf) { 856 /* Check for truncated error codes... */ 857 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 858 errstr = nbuf; 859 } 860 break; 861 } 862 863 return errstr; 864 } 865 866 /* __ext4_std_error decodes expected errors from journaling functions 867 * automatically and invokes the appropriate error response. */ 868 869 void __ext4_std_error(struct super_block *sb, const char *function, 870 unsigned int line, int errno) 871 { 872 char nbuf[16]; 873 const char *errstr; 874 875 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 876 return; 877 878 /* Special case: if the error is EROFS, and we're not already 879 * inside a transaction, then there's really no point in logging 880 * an error. */ 881 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 882 return; 883 884 if (ext4_error_ratelimit(sb)) { 885 errstr = ext4_decode_error(sb, errno, nbuf); 886 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 887 sb->s_id, function, line, errstr); 888 } 889 890 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 891 } 892 893 void __ext4_msg(struct super_block *sb, 894 const char *prefix, const char *fmt, ...) 895 { 896 struct va_format vaf; 897 va_list args; 898 899 atomic_inc(&EXT4_SB(sb)->s_msg_count); 900 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs")) 901 return; 902 903 va_start(args, fmt); 904 vaf.fmt = fmt; 905 vaf.va = &args; 906 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 907 va_end(args); 908 } 909 910 static int ext4_warning_ratelimit(struct super_block *sb) 911 { 912 atomic_inc(&EXT4_SB(sb)->s_warning_count); 913 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 914 "EXT4-fs warning"); 915 } 916 917 void __ext4_warning(struct super_block *sb, const char *function, 918 unsigned int line, const char *fmt, ...) 919 { 920 struct va_format vaf; 921 va_list args; 922 923 if (!ext4_warning_ratelimit(sb)) 924 return; 925 926 va_start(args, fmt); 927 vaf.fmt = fmt; 928 vaf.va = &args; 929 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 930 sb->s_id, function, line, &vaf); 931 va_end(args); 932 } 933 934 void __ext4_warning_inode(const struct inode *inode, const char *function, 935 unsigned int line, const char *fmt, ...) 936 { 937 struct va_format vaf; 938 va_list args; 939 940 if (!ext4_warning_ratelimit(inode->i_sb)) 941 return; 942 943 va_start(args, fmt); 944 vaf.fmt = fmt; 945 vaf.va = &args; 946 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 947 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 948 function, line, inode->i_ino, current->comm, &vaf); 949 va_end(args); 950 } 951 952 void __ext4_grp_locked_error(const char *function, unsigned int line, 953 struct super_block *sb, ext4_group_t grp, 954 unsigned long ino, ext4_fsblk_t block, 955 const char *fmt, ...) 956 __releases(bitlock) 957 __acquires(bitlock) 958 { 959 struct va_format vaf; 960 va_list args; 961 962 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 963 return; 964 965 trace_ext4_error(sb, function, line); 966 if (ext4_error_ratelimit(sb)) { 967 va_start(args, fmt); 968 vaf.fmt = fmt; 969 vaf.va = &args; 970 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 971 sb->s_id, function, line, grp); 972 if (ino) 973 printk(KERN_CONT "inode %lu: ", ino); 974 if (block) 975 printk(KERN_CONT "block %llu:", 976 (unsigned long long) block); 977 printk(KERN_CONT "%pV\n", &vaf); 978 va_end(args); 979 } 980 981 if (test_opt(sb, ERRORS_CONT)) { 982 if (test_opt(sb, WARN_ON_ERROR)) 983 WARN_ON_ONCE(1); 984 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 985 if (!bdev_read_only(sb->s_bdev)) { 986 save_error_info(sb, EFSCORRUPTED, ino, block, function, 987 line); 988 schedule_work(&EXT4_SB(sb)->s_error_work); 989 } 990 return; 991 } 992 ext4_unlock_group(sb, grp); 993 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 994 /* 995 * We only get here in the ERRORS_RO case; relocking the group 996 * may be dangerous, but nothing bad will happen since the 997 * filesystem will have already been marked read/only and the 998 * journal has been aborted. We return 1 as a hint to callers 999 * who might what to use the return value from 1000 * ext4_grp_locked_error() to distinguish between the 1001 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1002 * aggressively from the ext4 function in question, with a 1003 * more appropriate error code. 1004 */ 1005 ext4_lock_group(sb, grp); 1006 return; 1007 } 1008 1009 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1010 ext4_group_t group, 1011 unsigned int flags) 1012 { 1013 struct ext4_sb_info *sbi = EXT4_SB(sb); 1014 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1015 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1016 int ret; 1017 1018 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1019 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1020 &grp->bb_state); 1021 if (!ret) 1022 percpu_counter_sub(&sbi->s_freeclusters_counter, 1023 grp->bb_free); 1024 } 1025 1026 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1027 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1028 &grp->bb_state); 1029 if (!ret && gdp) { 1030 int count; 1031 1032 count = ext4_free_inodes_count(sb, gdp); 1033 percpu_counter_sub(&sbi->s_freeinodes_counter, 1034 count); 1035 } 1036 } 1037 } 1038 1039 void ext4_update_dynamic_rev(struct super_block *sb) 1040 { 1041 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1042 1043 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1044 return; 1045 1046 ext4_warning(sb, 1047 "updating to rev %d because of new feature flag, " 1048 "running e2fsck is recommended", 1049 EXT4_DYNAMIC_REV); 1050 1051 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1052 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1053 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1054 /* leave es->s_feature_*compat flags alone */ 1055 /* es->s_uuid will be set by e2fsck if empty */ 1056 1057 /* 1058 * The rest of the superblock fields should be zero, and if not it 1059 * means they are likely already in use, so leave them alone. We 1060 * can leave it up to e2fsck to clean up any inconsistencies there. 1061 */ 1062 } 1063 1064 /* 1065 * Open the external journal device 1066 */ 1067 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 1068 { 1069 struct block_device *bdev; 1070 1071 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 1072 if (IS_ERR(bdev)) 1073 goto fail; 1074 return bdev; 1075 1076 fail: 1077 ext4_msg(sb, KERN_ERR, 1078 "failed to open journal device unknown-block(%u,%u) %ld", 1079 MAJOR(dev), MINOR(dev), PTR_ERR(bdev)); 1080 return NULL; 1081 } 1082 1083 /* 1084 * Release the journal device 1085 */ 1086 static void ext4_blkdev_put(struct block_device *bdev) 1087 { 1088 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1089 } 1090 1091 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 1092 { 1093 struct block_device *bdev; 1094 bdev = sbi->s_journal_bdev; 1095 if (bdev) { 1096 ext4_blkdev_put(bdev); 1097 sbi->s_journal_bdev = NULL; 1098 } 1099 } 1100 1101 static inline struct inode *orphan_list_entry(struct list_head *l) 1102 { 1103 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1104 } 1105 1106 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1107 { 1108 struct list_head *l; 1109 1110 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1111 le32_to_cpu(sbi->s_es->s_last_orphan)); 1112 1113 printk(KERN_ERR "sb_info orphan list:\n"); 1114 list_for_each(l, &sbi->s_orphan) { 1115 struct inode *inode = orphan_list_entry(l); 1116 printk(KERN_ERR " " 1117 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1118 inode->i_sb->s_id, inode->i_ino, inode, 1119 inode->i_mode, inode->i_nlink, 1120 NEXT_ORPHAN(inode)); 1121 } 1122 } 1123 1124 #ifdef CONFIG_QUOTA 1125 static int ext4_quota_off(struct super_block *sb, int type); 1126 1127 static inline void ext4_quota_off_umount(struct super_block *sb) 1128 { 1129 int type; 1130 1131 /* Use our quota_off function to clear inode flags etc. */ 1132 for (type = 0; type < EXT4_MAXQUOTAS; type++) 1133 ext4_quota_off(sb, type); 1134 } 1135 1136 /* 1137 * This is a helper function which is used in the mount/remount 1138 * codepaths (which holds s_umount) to fetch the quota file name. 1139 */ 1140 static inline char *get_qf_name(struct super_block *sb, 1141 struct ext4_sb_info *sbi, 1142 int type) 1143 { 1144 return rcu_dereference_protected(sbi->s_qf_names[type], 1145 lockdep_is_held(&sb->s_umount)); 1146 } 1147 #else 1148 static inline void ext4_quota_off_umount(struct super_block *sb) 1149 { 1150 } 1151 #endif 1152 1153 static void ext4_put_super(struct super_block *sb) 1154 { 1155 struct ext4_sb_info *sbi = EXT4_SB(sb); 1156 struct ext4_super_block *es = sbi->s_es; 1157 struct buffer_head **group_desc; 1158 struct flex_groups **flex_groups; 1159 int aborted = 0; 1160 int i, err; 1161 1162 ext4_unregister_li_request(sb); 1163 ext4_quota_off_umount(sb); 1164 1165 flush_work(&sbi->s_error_work); 1166 destroy_workqueue(sbi->rsv_conversion_wq); 1167 1168 /* 1169 * Unregister sysfs before destroying jbd2 journal. 1170 * Since we could still access attr_journal_task attribute via sysfs 1171 * path which could have sbi->s_journal->j_task as NULL 1172 */ 1173 ext4_unregister_sysfs(sb); 1174 1175 if (sbi->s_journal) { 1176 aborted = is_journal_aborted(sbi->s_journal); 1177 err = jbd2_journal_destroy(sbi->s_journal); 1178 sbi->s_journal = NULL; 1179 if ((err < 0) && !aborted) { 1180 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1181 } 1182 } 1183 1184 ext4_es_unregister_shrinker(sbi); 1185 del_timer_sync(&sbi->s_err_report); 1186 ext4_release_system_zone(sb); 1187 ext4_mb_release(sb); 1188 ext4_ext_release(sb); 1189 1190 if (!sb_rdonly(sb) && !aborted) { 1191 ext4_clear_feature_journal_needs_recovery(sb); 1192 es->s_state = cpu_to_le16(sbi->s_mount_state); 1193 } 1194 if (!sb_rdonly(sb)) 1195 ext4_commit_super(sb); 1196 1197 rcu_read_lock(); 1198 group_desc = rcu_dereference(sbi->s_group_desc); 1199 for (i = 0; i < sbi->s_gdb_count; i++) 1200 brelse(group_desc[i]); 1201 kvfree(group_desc); 1202 flex_groups = rcu_dereference(sbi->s_flex_groups); 1203 if (flex_groups) { 1204 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1205 kvfree(flex_groups[i]); 1206 kvfree(flex_groups); 1207 } 1208 rcu_read_unlock(); 1209 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1210 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1211 percpu_counter_destroy(&sbi->s_dirs_counter); 1212 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1213 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1214 #ifdef CONFIG_QUOTA 1215 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1216 kfree(get_qf_name(sb, sbi, i)); 1217 #endif 1218 1219 /* Debugging code just in case the in-memory inode orphan list 1220 * isn't empty. The on-disk one can be non-empty if we've 1221 * detected an error and taken the fs readonly, but the 1222 * in-memory list had better be clean by this point. */ 1223 if (!list_empty(&sbi->s_orphan)) 1224 dump_orphan_list(sb, sbi); 1225 ASSERT(list_empty(&sbi->s_orphan)); 1226 1227 sync_blockdev(sb->s_bdev); 1228 invalidate_bdev(sb->s_bdev); 1229 if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) { 1230 /* 1231 * Invalidate the journal device's buffers. We don't want them 1232 * floating about in memory - the physical journal device may 1233 * hotswapped, and it breaks the `ro-after' testing code. 1234 */ 1235 sync_blockdev(sbi->s_journal_bdev); 1236 invalidate_bdev(sbi->s_journal_bdev); 1237 ext4_blkdev_remove(sbi); 1238 } 1239 1240 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1241 sbi->s_ea_inode_cache = NULL; 1242 1243 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1244 sbi->s_ea_block_cache = NULL; 1245 1246 if (sbi->s_mmp_tsk) 1247 kthread_stop(sbi->s_mmp_tsk); 1248 brelse(sbi->s_sbh); 1249 sb->s_fs_info = NULL; 1250 /* 1251 * Now that we are completely done shutting down the 1252 * superblock, we need to actually destroy the kobject. 1253 */ 1254 kobject_put(&sbi->s_kobj); 1255 wait_for_completion(&sbi->s_kobj_unregister); 1256 if (sbi->s_chksum_driver) 1257 crypto_free_shash(sbi->s_chksum_driver); 1258 kfree(sbi->s_blockgroup_lock); 1259 fs_put_dax(sbi->s_daxdev); 1260 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1261 #ifdef CONFIG_UNICODE 1262 utf8_unload(sb->s_encoding); 1263 #endif 1264 kfree(sbi); 1265 } 1266 1267 static struct kmem_cache *ext4_inode_cachep; 1268 1269 /* 1270 * Called inside transaction, so use GFP_NOFS 1271 */ 1272 static struct inode *ext4_alloc_inode(struct super_block *sb) 1273 { 1274 struct ext4_inode_info *ei; 1275 1276 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS); 1277 if (!ei) 1278 return NULL; 1279 1280 inode_set_iversion(&ei->vfs_inode, 1); 1281 spin_lock_init(&ei->i_raw_lock); 1282 INIT_LIST_HEAD(&ei->i_prealloc_list); 1283 atomic_set(&ei->i_prealloc_active, 0); 1284 spin_lock_init(&ei->i_prealloc_lock); 1285 ext4_es_init_tree(&ei->i_es_tree); 1286 rwlock_init(&ei->i_es_lock); 1287 INIT_LIST_HEAD(&ei->i_es_list); 1288 ei->i_es_all_nr = 0; 1289 ei->i_es_shk_nr = 0; 1290 ei->i_es_shrink_lblk = 0; 1291 ei->i_reserved_data_blocks = 0; 1292 spin_lock_init(&(ei->i_block_reservation_lock)); 1293 ext4_init_pending_tree(&ei->i_pending_tree); 1294 #ifdef CONFIG_QUOTA 1295 ei->i_reserved_quota = 0; 1296 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1297 #endif 1298 ei->jinode = NULL; 1299 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1300 spin_lock_init(&ei->i_completed_io_lock); 1301 ei->i_sync_tid = 0; 1302 ei->i_datasync_tid = 0; 1303 atomic_set(&ei->i_unwritten, 0); 1304 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1305 ext4_fc_init_inode(&ei->vfs_inode); 1306 mutex_init(&ei->i_fc_lock); 1307 return &ei->vfs_inode; 1308 } 1309 1310 static int ext4_drop_inode(struct inode *inode) 1311 { 1312 int drop = generic_drop_inode(inode); 1313 1314 if (!drop) 1315 drop = fscrypt_drop_inode(inode); 1316 1317 trace_ext4_drop_inode(inode, drop); 1318 return drop; 1319 } 1320 1321 static void ext4_free_in_core_inode(struct inode *inode) 1322 { 1323 fscrypt_free_inode(inode); 1324 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1325 pr_warn("%s: inode %ld still in fc list", 1326 __func__, inode->i_ino); 1327 } 1328 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1329 } 1330 1331 static void ext4_destroy_inode(struct inode *inode) 1332 { 1333 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1334 ext4_msg(inode->i_sb, KERN_ERR, 1335 "Inode %lu (%p): orphan list check failed!", 1336 inode->i_ino, EXT4_I(inode)); 1337 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1338 EXT4_I(inode), sizeof(struct ext4_inode_info), 1339 true); 1340 dump_stack(); 1341 } 1342 } 1343 1344 static void init_once(void *foo) 1345 { 1346 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo; 1347 1348 INIT_LIST_HEAD(&ei->i_orphan); 1349 init_rwsem(&ei->xattr_sem); 1350 init_rwsem(&ei->i_data_sem); 1351 init_rwsem(&ei->i_mmap_sem); 1352 inode_init_once(&ei->vfs_inode); 1353 ext4_fc_init_inode(&ei->vfs_inode); 1354 } 1355 1356 static int __init init_inodecache(void) 1357 { 1358 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1359 sizeof(struct ext4_inode_info), 0, 1360 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1361 SLAB_ACCOUNT), 1362 offsetof(struct ext4_inode_info, i_data), 1363 sizeof_field(struct ext4_inode_info, i_data), 1364 init_once); 1365 if (ext4_inode_cachep == NULL) 1366 return -ENOMEM; 1367 return 0; 1368 } 1369 1370 static void destroy_inodecache(void) 1371 { 1372 /* 1373 * Make sure all delayed rcu free inodes are flushed before we 1374 * destroy cache. 1375 */ 1376 rcu_barrier(); 1377 kmem_cache_destroy(ext4_inode_cachep); 1378 } 1379 1380 void ext4_clear_inode(struct inode *inode) 1381 { 1382 ext4_fc_del(inode); 1383 invalidate_inode_buffers(inode); 1384 clear_inode(inode); 1385 ext4_discard_preallocations(inode, 0); 1386 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1387 dquot_drop(inode); 1388 if (EXT4_I(inode)->jinode) { 1389 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1390 EXT4_I(inode)->jinode); 1391 jbd2_free_inode(EXT4_I(inode)->jinode); 1392 EXT4_I(inode)->jinode = NULL; 1393 } 1394 fscrypt_put_encryption_info(inode); 1395 fsverity_cleanup_inode(inode); 1396 } 1397 1398 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1399 u64 ino, u32 generation) 1400 { 1401 struct inode *inode; 1402 1403 /* 1404 * Currently we don't know the generation for parent directory, so 1405 * a generation of 0 means "accept any" 1406 */ 1407 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1408 if (IS_ERR(inode)) 1409 return ERR_CAST(inode); 1410 if (generation && inode->i_generation != generation) { 1411 iput(inode); 1412 return ERR_PTR(-ESTALE); 1413 } 1414 1415 return inode; 1416 } 1417 1418 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1419 int fh_len, int fh_type) 1420 { 1421 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1422 ext4_nfs_get_inode); 1423 } 1424 1425 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1426 int fh_len, int fh_type) 1427 { 1428 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1429 ext4_nfs_get_inode); 1430 } 1431 1432 static int ext4_nfs_commit_metadata(struct inode *inode) 1433 { 1434 struct writeback_control wbc = { 1435 .sync_mode = WB_SYNC_ALL 1436 }; 1437 1438 trace_ext4_nfs_commit_metadata(inode); 1439 return ext4_write_inode(inode, &wbc); 1440 } 1441 1442 /* 1443 * Try to release metadata pages (indirect blocks, directories) which are 1444 * mapped via the block device. Since these pages could have journal heads 1445 * which would prevent try_to_free_buffers() from freeing them, we must use 1446 * jbd2 layer's try_to_free_buffers() function to release them. 1447 */ 1448 static int bdev_try_to_free_page(struct super_block *sb, struct page *page, 1449 gfp_t wait) 1450 { 1451 journal_t *journal = EXT4_SB(sb)->s_journal; 1452 1453 WARN_ON(PageChecked(page)); 1454 if (!page_has_buffers(page)) 1455 return 0; 1456 if (journal) 1457 return jbd2_journal_try_to_free_buffers(journal, page); 1458 1459 return try_to_free_buffers(page); 1460 } 1461 1462 #ifdef CONFIG_FS_ENCRYPTION 1463 static int ext4_get_context(struct inode *inode, void *ctx, size_t len) 1464 { 1465 return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, 1466 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len); 1467 } 1468 1469 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len, 1470 void *fs_data) 1471 { 1472 handle_t *handle = fs_data; 1473 int res, res2, credits, retries = 0; 1474 1475 /* 1476 * Encrypting the root directory is not allowed because e2fsck expects 1477 * lost+found to exist and be unencrypted, and encrypting the root 1478 * directory would imply encrypting the lost+found directory as well as 1479 * the filename "lost+found" itself. 1480 */ 1481 if (inode->i_ino == EXT4_ROOT_INO) 1482 return -EPERM; 1483 1484 if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode))) 1485 return -EINVAL; 1486 1487 if (ext4_test_inode_flag(inode, EXT4_INODE_DAX)) 1488 return -EOPNOTSUPP; 1489 1490 res = ext4_convert_inline_data(inode); 1491 if (res) 1492 return res; 1493 1494 /* 1495 * If a journal handle was specified, then the encryption context is 1496 * being set on a new inode via inheritance and is part of a larger 1497 * transaction to create the inode. Otherwise the encryption context is 1498 * being set on an existing inode in its own transaction. Only in the 1499 * latter case should the "retry on ENOSPC" logic be used. 1500 */ 1501 1502 if (handle) { 1503 res = ext4_xattr_set_handle(handle, inode, 1504 EXT4_XATTR_INDEX_ENCRYPTION, 1505 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1506 ctx, len, 0); 1507 if (!res) { 1508 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1509 ext4_clear_inode_state(inode, 1510 EXT4_STATE_MAY_INLINE_DATA); 1511 /* 1512 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1513 * S_DAX may be disabled 1514 */ 1515 ext4_set_inode_flags(inode, false); 1516 } 1517 return res; 1518 } 1519 1520 res = dquot_initialize(inode); 1521 if (res) 1522 return res; 1523 retry: 1524 res = ext4_xattr_set_credits(inode, len, false /* is_create */, 1525 &credits); 1526 if (res) 1527 return res; 1528 1529 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); 1530 if (IS_ERR(handle)) 1531 return PTR_ERR(handle); 1532 1533 res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION, 1534 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1535 ctx, len, 0); 1536 if (!res) { 1537 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1538 /* 1539 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1540 * S_DAX may be disabled 1541 */ 1542 ext4_set_inode_flags(inode, false); 1543 res = ext4_mark_inode_dirty(handle, inode); 1544 if (res) 1545 EXT4_ERROR_INODE(inode, "Failed to mark inode dirty"); 1546 } 1547 res2 = ext4_journal_stop(handle); 1548 1549 if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 1550 goto retry; 1551 if (!res) 1552 res = res2; 1553 return res; 1554 } 1555 1556 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb) 1557 { 1558 return EXT4_SB(sb)->s_dummy_enc_policy.policy; 1559 } 1560 1561 static bool ext4_has_stable_inodes(struct super_block *sb) 1562 { 1563 return ext4_has_feature_stable_inodes(sb); 1564 } 1565 1566 static void ext4_get_ino_and_lblk_bits(struct super_block *sb, 1567 int *ino_bits_ret, int *lblk_bits_ret) 1568 { 1569 *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count); 1570 *lblk_bits_ret = 8 * sizeof(ext4_lblk_t); 1571 } 1572 1573 static const struct fscrypt_operations ext4_cryptops = { 1574 .key_prefix = "ext4:", 1575 .get_context = ext4_get_context, 1576 .set_context = ext4_set_context, 1577 .get_dummy_policy = ext4_get_dummy_policy, 1578 .empty_dir = ext4_empty_dir, 1579 .max_namelen = EXT4_NAME_LEN, 1580 .has_stable_inodes = ext4_has_stable_inodes, 1581 .get_ino_and_lblk_bits = ext4_get_ino_and_lblk_bits, 1582 }; 1583 #endif 1584 1585 #ifdef CONFIG_QUOTA 1586 static const char * const quotatypes[] = INITQFNAMES; 1587 #define QTYPE2NAME(t) (quotatypes[t]) 1588 1589 static int ext4_write_dquot(struct dquot *dquot); 1590 static int ext4_acquire_dquot(struct dquot *dquot); 1591 static int ext4_release_dquot(struct dquot *dquot); 1592 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1593 static int ext4_write_info(struct super_block *sb, int type); 1594 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1595 const struct path *path); 1596 static int ext4_quota_on_mount(struct super_block *sb, int type); 1597 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1598 size_t len, loff_t off); 1599 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1600 const char *data, size_t len, loff_t off); 1601 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1602 unsigned int flags); 1603 static int ext4_enable_quotas(struct super_block *sb); 1604 1605 static struct dquot **ext4_get_dquots(struct inode *inode) 1606 { 1607 return EXT4_I(inode)->i_dquot; 1608 } 1609 1610 static const struct dquot_operations ext4_quota_operations = { 1611 .get_reserved_space = ext4_get_reserved_space, 1612 .write_dquot = ext4_write_dquot, 1613 .acquire_dquot = ext4_acquire_dquot, 1614 .release_dquot = ext4_release_dquot, 1615 .mark_dirty = ext4_mark_dquot_dirty, 1616 .write_info = ext4_write_info, 1617 .alloc_dquot = dquot_alloc, 1618 .destroy_dquot = dquot_destroy, 1619 .get_projid = ext4_get_projid, 1620 .get_inode_usage = ext4_get_inode_usage, 1621 .get_next_id = dquot_get_next_id, 1622 }; 1623 1624 static const struct quotactl_ops ext4_qctl_operations = { 1625 .quota_on = ext4_quota_on, 1626 .quota_off = ext4_quota_off, 1627 .quota_sync = dquot_quota_sync, 1628 .get_state = dquot_get_state, 1629 .set_info = dquot_set_dqinfo, 1630 .get_dqblk = dquot_get_dqblk, 1631 .set_dqblk = dquot_set_dqblk, 1632 .get_nextdqblk = dquot_get_next_dqblk, 1633 }; 1634 #endif 1635 1636 static const struct super_operations ext4_sops = { 1637 .alloc_inode = ext4_alloc_inode, 1638 .free_inode = ext4_free_in_core_inode, 1639 .destroy_inode = ext4_destroy_inode, 1640 .write_inode = ext4_write_inode, 1641 .dirty_inode = ext4_dirty_inode, 1642 .drop_inode = ext4_drop_inode, 1643 .evict_inode = ext4_evict_inode, 1644 .put_super = ext4_put_super, 1645 .sync_fs = ext4_sync_fs, 1646 .freeze_fs = ext4_freeze, 1647 .unfreeze_fs = ext4_unfreeze, 1648 .statfs = ext4_statfs, 1649 .remount_fs = ext4_remount, 1650 .show_options = ext4_show_options, 1651 #ifdef CONFIG_QUOTA 1652 .quota_read = ext4_quota_read, 1653 .quota_write = ext4_quota_write, 1654 .get_dquots = ext4_get_dquots, 1655 #endif 1656 .bdev_try_to_free_page = bdev_try_to_free_page, 1657 }; 1658 1659 static const struct export_operations ext4_export_ops = { 1660 .fh_to_dentry = ext4_fh_to_dentry, 1661 .fh_to_parent = ext4_fh_to_parent, 1662 .get_parent = ext4_get_parent, 1663 .commit_metadata = ext4_nfs_commit_metadata, 1664 }; 1665 1666 enum { 1667 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1668 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro, 1669 Opt_nouid32, Opt_debug, Opt_removed, 1670 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, 1671 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1672 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1673 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1674 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1675 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1676 Opt_inlinecrypt, 1677 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota, 1678 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota, 1679 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1680 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, 1681 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1682 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1683 Opt_nowarn_on_error, Opt_mblk_io_submit, 1684 Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize, 1685 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1686 Opt_inode_readahead_blks, Opt_journal_ioprio, 1687 Opt_dioread_nolock, Opt_dioread_lock, 1688 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1689 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1690 Opt_prefetch_block_bitmaps, 1691 #ifdef CONFIG_EXT4_DEBUG 1692 Opt_fc_debug_max_replay, Opt_fc_debug_force 1693 #endif 1694 }; 1695 1696 static const match_table_t tokens = { 1697 {Opt_bsd_df, "bsddf"}, 1698 {Opt_minix_df, "minixdf"}, 1699 {Opt_grpid, "grpid"}, 1700 {Opt_grpid, "bsdgroups"}, 1701 {Opt_nogrpid, "nogrpid"}, 1702 {Opt_nogrpid, "sysvgroups"}, 1703 {Opt_resgid, "resgid=%u"}, 1704 {Opt_resuid, "resuid=%u"}, 1705 {Opt_sb, "sb=%u"}, 1706 {Opt_err_cont, "errors=continue"}, 1707 {Opt_err_panic, "errors=panic"}, 1708 {Opt_err_ro, "errors=remount-ro"}, 1709 {Opt_nouid32, "nouid32"}, 1710 {Opt_debug, "debug"}, 1711 {Opt_removed, "oldalloc"}, 1712 {Opt_removed, "orlov"}, 1713 {Opt_user_xattr, "user_xattr"}, 1714 {Opt_nouser_xattr, "nouser_xattr"}, 1715 {Opt_acl, "acl"}, 1716 {Opt_noacl, "noacl"}, 1717 {Opt_noload, "norecovery"}, 1718 {Opt_noload, "noload"}, 1719 {Opt_removed, "nobh"}, 1720 {Opt_removed, "bh"}, 1721 {Opt_commit, "commit=%u"}, 1722 {Opt_min_batch_time, "min_batch_time=%u"}, 1723 {Opt_max_batch_time, "max_batch_time=%u"}, 1724 {Opt_journal_dev, "journal_dev=%u"}, 1725 {Opt_journal_path, "journal_path=%s"}, 1726 {Opt_journal_checksum, "journal_checksum"}, 1727 {Opt_nojournal_checksum, "nojournal_checksum"}, 1728 {Opt_journal_async_commit, "journal_async_commit"}, 1729 {Opt_abort, "abort"}, 1730 {Opt_data_journal, "data=journal"}, 1731 {Opt_data_ordered, "data=ordered"}, 1732 {Opt_data_writeback, "data=writeback"}, 1733 {Opt_data_err_abort, "data_err=abort"}, 1734 {Opt_data_err_ignore, "data_err=ignore"}, 1735 {Opt_offusrjquota, "usrjquota="}, 1736 {Opt_usrjquota, "usrjquota=%s"}, 1737 {Opt_offgrpjquota, "grpjquota="}, 1738 {Opt_grpjquota, "grpjquota=%s"}, 1739 {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, 1740 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, 1741 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, 1742 {Opt_grpquota, "grpquota"}, 1743 {Opt_noquota, "noquota"}, 1744 {Opt_quota, "quota"}, 1745 {Opt_usrquota, "usrquota"}, 1746 {Opt_prjquota, "prjquota"}, 1747 {Opt_barrier, "barrier=%u"}, 1748 {Opt_barrier, "barrier"}, 1749 {Opt_nobarrier, "nobarrier"}, 1750 {Opt_i_version, "i_version"}, 1751 {Opt_dax, "dax"}, 1752 {Opt_dax_always, "dax=always"}, 1753 {Opt_dax_inode, "dax=inode"}, 1754 {Opt_dax_never, "dax=never"}, 1755 {Opt_stripe, "stripe=%u"}, 1756 {Opt_delalloc, "delalloc"}, 1757 {Opt_warn_on_error, "warn_on_error"}, 1758 {Opt_nowarn_on_error, "nowarn_on_error"}, 1759 {Opt_lazytime, "lazytime"}, 1760 {Opt_nolazytime, "nolazytime"}, 1761 {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"}, 1762 {Opt_nodelalloc, "nodelalloc"}, 1763 {Opt_removed, "mblk_io_submit"}, 1764 {Opt_removed, "nomblk_io_submit"}, 1765 {Opt_block_validity, "block_validity"}, 1766 {Opt_noblock_validity, "noblock_validity"}, 1767 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, 1768 {Opt_journal_ioprio, "journal_ioprio=%u"}, 1769 {Opt_auto_da_alloc, "auto_da_alloc=%u"}, 1770 {Opt_auto_da_alloc, "auto_da_alloc"}, 1771 {Opt_noauto_da_alloc, "noauto_da_alloc"}, 1772 {Opt_dioread_nolock, "dioread_nolock"}, 1773 {Opt_dioread_lock, "nodioread_nolock"}, 1774 {Opt_dioread_lock, "dioread_lock"}, 1775 {Opt_discard, "discard"}, 1776 {Opt_nodiscard, "nodiscard"}, 1777 {Opt_init_itable, "init_itable=%u"}, 1778 {Opt_init_itable, "init_itable"}, 1779 {Opt_noinit_itable, "noinit_itable"}, 1780 #ifdef CONFIG_EXT4_DEBUG 1781 {Opt_fc_debug_force, "fc_debug_force"}, 1782 {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"}, 1783 #endif 1784 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, 1785 {Opt_test_dummy_encryption, "test_dummy_encryption=%s"}, 1786 {Opt_test_dummy_encryption, "test_dummy_encryption"}, 1787 {Opt_inlinecrypt, "inlinecrypt"}, 1788 {Opt_nombcache, "nombcache"}, 1789 {Opt_nombcache, "no_mbcache"}, /* for backward compatibility */ 1790 {Opt_prefetch_block_bitmaps, "prefetch_block_bitmaps"}, 1791 {Opt_removed, "check=none"}, /* mount option from ext2/3 */ 1792 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */ 1793 {Opt_removed, "reservation"}, /* mount option from ext2/3 */ 1794 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */ 1795 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */ 1796 {Opt_err, NULL}, 1797 }; 1798 1799 static ext4_fsblk_t get_sb_block(void **data) 1800 { 1801 ext4_fsblk_t sb_block; 1802 char *options = (char *) *data; 1803 1804 if (!options || strncmp(options, "sb=", 3) != 0) 1805 return 1; /* Default location */ 1806 1807 options += 3; 1808 /* TODO: use simple_strtoll with >32bit ext4 */ 1809 sb_block = simple_strtoul(options, &options, 0); 1810 if (*options && *options != ',') { 1811 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n", 1812 (char *) *data); 1813 return 1; 1814 } 1815 if (*options == ',') 1816 options++; 1817 *data = (void *) options; 1818 1819 return sb_block; 1820 } 1821 1822 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1823 static const char deprecated_msg[] = 1824 "Mount option \"%s\" will be removed by %s\n" 1825 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1826 1827 #ifdef CONFIG_QUOTA 1828 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args) 1829 { 1830 struct ext4_sb_info *sbi = EXT4_SB(sb); 1831 char *qname, *old_qname = get_qf_name(sb, sbi, qtype); 1832 int ret = -1; 1833 1834 if (sb_any_quota_loaded(sb) && !old_qname) { 1835 ext4_msg(sb, KERN_ERR, 1836 "Cannot change journaled " 1837 "quota options when quota turned on"); 1838 return -1; 1839 } 1840 if (ext4_has_feature_quota(sb)) { 1841 ext4_msg(sb, KERN_INFO, "Journaled quota options " 1842 "ignored when QUOTA feature is enabled"); 1843 return 1; 1844 } 1845 qname = match_strdup(args); 1846 if (!qname) { 1847 ext4_msg(sb, KERN_ERR, 1848 "Not enough memory for storing quotafile name"); 1849 return -1; 1850 } 1851 if (old_qname) { 1852 if (strcmp(old_qname, qname) == 0) 1853 ret = 1; 1854 else 1855 ext4_msg(sb, KERN_ERR, 1856 "%s quota file already specified", 1857 QTYPE2NAME(qtype)); 1858 goto errout; 1859 } 1860 if (strchr(qname, '/')) { 1861 ext4_msg(sb, KERN_ERR, 1862 "quotafile must be on filesystem root"); 1863 goto errout; 1864 } 1865 rcu_assign_pointer(sbi->s_qf_names[qtype], qname); 1866 set_opt(sb, QUOTA); 1867 return 1; 1868 errout: 1869 kfree(qname); 1870 return ret; 1871 } 1872 1873 static int clear_qf_name(struct super_block *sb, int qtype) 1874 { 1875 1876 struct ext4_sb_info *sbi = EXT4_SB(sb); 1877 char *old_qname = get_qf_name(sb, sbi, qtype); 1878 1879 if (sb_any_quota_loaded(sb) && old_qname) { 1880 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options" 1881 " when quota turned on"); 1882 return -1; 1883 } 1884 rcu_assign_pointer(sbi->s_qf_names[qtype], NULL); 1885 synchronize_rcu(); 1886 kfree(old_qname); 1887 return 1; 1888 } 1889 #endif 1890 1891 #define MOPT_SET 0x0001 1892 #define MOPT_CLEAR 0x0002 1893 #define MOPT_NOSUPPORT 0x0004 1894 #define MOPT_EXPLICIT 0x0008 1895 #define MOPT_CLEAR_ERR 0x0010 1896 #define MOPT_GTE0 0x0020 1897 #ifdef CONFIG_QUOTA 1898 #define MOPT_Q 0 1899 #define MOPT_QFMT 0x0040 1900 #else 1901 #define MOPT_Q MOPT_NOSUPPORT 1902 #define MOPT_QFMT MOPT_NOSUPPORT 1903 #endif 1904 #define MOPT_DATAJ 0x0080 1905 #define MOPT_NO_EXT2 0x0100 1906 #define MOPT_NO_EXT3 0x0200 1907 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1908 #define MOPT_STRING 0x0400 1909 #define MOPT_SKIP 0x0800 1910 #define MOPT_2 0x1000 1911 1912 static const struct mount_opts { 1913 int token; 1914 int mount_opt; 1915 int flags; 1916 } ext4_mount_opts[] = { 1917 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1918 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1919 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1920 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1921 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1922 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1923 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1924 MOPT_EXT4_ONLY | MOPT_SET}, 1925 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1926 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1927 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1928 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1929 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1930 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1931 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1932 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1933 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1934 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1935 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1936 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1937 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1938 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1939 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1940 EXT4_MOUNT_JOURNAL_CHECKSUM), 1941 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1942 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1943 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR}, 1944 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR}, 1945 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR}, 1946 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, 1947 MOPT_NO_EXT2}, 1948 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, 1949 MOPT_NO_EXT2}, 1950 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1951 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1952 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1953 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1954 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1955 {Opt_commit, 0, MOPT_GTE0}, 1956 {Opt_max_batch_time, 0, MOPT_GTE0}, 1957 {Opt_min_batch_time, 0, MOPT_GTE0}, 1958 {Opt_inode_readahead_blks, 0, MOPT_GTE0}, 1959 {Opt_init_itable, 0, MOPT_GTE0}, 1960 {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP}, 1961 {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS, 1962 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1963 {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE, 1964 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1965 {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER, 1966 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1967 {Opt_stripe, 0, MOPT_GTE0}, 1968 {Opt_resuid, 0, MOPT_GTE0}, 1969 {Opt_resgid, 0, MOPT_GTE0}, 1970 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1971 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING}, 1972 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1973 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1974 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1975 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, 1976 MOPT_NO_EXT2 | MOPT_DATAJ}, 1977 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1978 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR}, 1979 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1980 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1981 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR}, 1982 #else 1983 {Opt_acl, 0, MOPT_NOSUPPORT}, 1984 {Opt_noacl, 0, MOPT_NOSUPPORT}, 1985 #endif 1986 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1987 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1988 {Opt_debug_want_extra_isize, 0, MOPT_GTE0}, 1989 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1990 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1991 MOPT_SET | MOPT_Q}, 1992 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1993 MOPT_SET | MOPT_Q}, 1994 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1995 MOPT_SET | MOPT_Q}, 1996 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1997 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1998 MOPT_CLEAR | MOPT_Q}, 1999 {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING}, 2000 {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING}, 2001 {Opt_offusrjquota, 0, MOPT_Q}, 2002 {Opt_offgrpjquota, 0, MOPT_Q}, 2003 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT}, 2004 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT}, 2005 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT}, 2006 {Opt_max_dir_size_kb, 0, MOPT_GTE0}, 2007 {Opt_test_dummy_encryption, 0, MOPT_STRING}, 2008 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 2009 {Opt_prefetch_block_bitmaps, EXT4_MOUNT_PREFETCH_BLOCK_BITMAPS, 2010 MOPT_SET}, 2011 #ifdef CONFIG_EXT4_DEBUG 2012 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 2013 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 2014 {Opt_fc_debug_max_replay, 0, MOPT_GTE0}, 2015 #endif 2016 {Opt_err, 0, 0} 2017 }; 2018 2019 #ifdef CONFIG_UNICODE 2020 static const struct ext4_sb_encodings { 2021 __u16 magic; 2022 char *name; 2023 char *version; 2024 } ext4_sb_encoding_map[] = { 2025 {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"}, 2026 }; 2027 2028 static int ext4_sb_read_encoding(const struct ext4_super_block *es, 2029 const struct ext4_sb_encodings **encoding, 2030 __u16 *flags) 2031 { 2032 __u16 magic = le16_to_cpu(es->s_encoding); 2033 int i; 2034 2035 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 2036 if (magic == ext4_sb_encoding_map[i].magic) 2037 break; 2038 2039 if (i >= ARRAY_SIZE(ext4_sb_encoding_map)) 2040 return -EINVAL; 2041 2042 *encoding = &ext4_sb_encoding_map[i]; 2043 *flags = le16_to_cpu(es->s_encoding_flags); 2044 2045 return 0; 2046 } 2047 #endif 2048 2049 static int ext4_set_test_dummy_encryption(struct super_block *sb, 2050 const char *opt, 2051 const substring_t *arg, 2052 bool is_remount) 2053 { 2054 #ifdef CONFIG_FS_ENCRYPTION 2055 struct ext4_sb_info *sbi = EXT4_SB(sb); 2056 int err; 2057 2058 /* 2059 * This mount option is just for testing, and it's not worthwhile to 2060 * implement the extra complexity (e.g. RCU protection) that would be 2061 * needed to allow it to be set or changed during remount. We do allow 2062 * it to be specified during remount, but only if there is no change. 2063 */ 2064 if (is_remount && !sbi->s_dummy_enc_policy.policy) { 2065 ext4_msg(sb, KERN_WARNING, 2066 "Can't set test_dummy_encryption on remount"); 2067 return -1; 2068 } 2069 err = fscrypt_set_test_dummy_encryption(sb, arg->from, 2070 &sbi->s_dummy_enc_policy); 2071 if (err) { 2072 if (err == -EEXIST) 2073 ext4_msg(sb, KERN_WARNING, 2074 "Can't change test_dummy_encryption on remount"); 2075 else if (err == -EINVAL) 2076 ext4_msg(sb, KERN_WARNING, 2077 "Value of option \"%s\" is unrecognized", opt); 2078 else 2079 ext4_msg(sb, KERN_WARNING, 2080 "Error processing option \"%s\" [%d]", 2081 opt, err); 2082 return -1; 2083 } 2084 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2085 #else 2086 ext4_msg(sb, KERN_WARNING, 2087 "Test dummy encryption mount option ignored"); 2088 #endif 2089 return 1; 2090 } 2091 2092 static int handle_mount_opt(struct super_block *sb, char *opt, int token, 2093 substring_t *args, unsigned long *journal_devnum, 2094 unsigned int *journal_ioprio, int is_remount) 2095 { 2096 struct ext4_sb_info *sbi = EXT4_SB(sb); 2097 const struct mount_opts *m; 2098 kuid_t uid; 2099 kgid_t gid; 2100 int arg = 0; 2101 2102 #ifdef CONFIG_QUOTA 2103 if (token == Opt_usrjquota) 2104 return set_qf_name(sb, USRQUOTA, &args[0]); 2105 else if (token == Opt_grpjquota) 2106 return set_qf_name(sb, GRPQUOTA, &args[0]); 2107 else if (token == Opt_offusrjquota) 2108 return clear_qf_name(sb, USRQUOTA); 2109 else if (token == Opt_offgrpjquota) 2110 return clear_qf_name(sb, GRPQUOTA); 2111 #endif 2112 switch (token) { 2113 case Opt_noacl: 2114 case Opt_nouser_xattr: 2115 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5"); 2116 break; 2117 case Opt_sb: 2118 return 1; /* handled by get_sb_block() */ 2119 case Opt_removed: 2120 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt); 2121 return 1; 2122 case Opt_abort: 2123 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 2124 return 1; 2125 case Opt_i_version: 2126 sb->s_flags |= SB_I_VERSION; 2127 return 1; 2128 case Opt_lazytime: 2129 sb->s_flags |= SB_LAZYTIME; 2130 return 1; 2131 case Opt_nolazytime: 2132 sb->s_flags &= ~SB_LAZYTIME; 2133 return 1; 2134 case Opt_inlinecrypt: 2135 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2136 sb->s_flags |= SB_INLINECRYPT; 2137 #else 2138 ext4_msg(sb, KERN_ERR, "inline encryption not supported"); 2139 #endif 2140 return 1; 2141 } 2142 2143 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2144 if (token == m->token) 2145 break; 2146 2147 if (m->token == Opt_err) { 2148 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" " 2149 "or missing value", opt); 2150 return -1; 2151 } 2152 2153 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2154 ext4_msg(sb, KERN_ERR, 2155 "Mount option \"%s\" incompatible with ext2", opt); 2156 return -1; 2157 } 2158 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2159 ext4_msg(sb, KERN_ERR, 2160 "Mount option \"%s\" incompatible with ext3", opt); 2161 return -1; 2162 } 2163 2164 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg)) 2165 return -1; 2166 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0)) 2167 return -1; 2168 if (m->flags & MOPT_EXPLICIT) { 2169 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2170 set_opt2(sb, EXPLICIT_DELALLOC); 2171 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2172 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM); 2173 } else 2174 return -1; 2175 } 2176 if (m->flags & MOPT_CLEAR_ERR) 2177 clear_opt(sb, ERRORS_MASK); 2178 if (token == Opt_noquota && sb_any_quota_loaded(sb)) { 2179 ext4_msg(sb, KERN_ERR, "Cannot change quota " 2180 "options when quota turned on"); 2181 return -1; 2182 } 2183 2184 if (m->flags & MOPT_NOSUPPORT) { 2185 ext4_msg(sb, KERN_ERR, "%s option not supported", opt); 2186 } else if (token == Opt_commit) { 2187 if (arg == 0) 2188 arg = JBD2_DEFAULT_MAX_COMMIT_AGE; 2189 else if (arg > INT_MAX / HZ) { 2190 ext4_msg(sb, KERN_ERR, 2191 "Invalid commit interval %d, " 2192 "must be smaller than %d", 2193 arg, INT_MAX / HZ); 2194 return -1; 2195 } 2196 sbi->s_commit_interval = HZ * arg; 2197 } else if (token == Opt_debug_want_extra_isize) { 2198 if ((arg & 1) || 2199 (arg < 4) || 2200 (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) { 2201 ext4_msg(sb, KERN_ERR, 2202 "Invalid want_extra_isize %d", arg); 2203 return -1; 2204 } 2205 sbi->s_want_extra_isize = arg; 2206 } else if (token == Opt_max_batch_time) { 2207 sbi->s_max_batch_time = arg; 2208 } else if (token == Opt_min_batch_time) { 2209 sbi->s_min_batch_time = arg; 2210 } else if (token == Opt_inode_readahead_blks) { 2211 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) { 2212 ext4_msg(sb, KERN_ERR, 2213 "EXT4-fs: inode_readahead_blks must be " 2214 "0 or a power of 2 smaller than 2^31"); 2215 return -1; 2216 } 2217 sbi->s_inode_readahead_blks = arg; 2218 } else if (token == Opt_init_itable) { 2219 set_opt(sb, INIT_INODE_TABLE); 2220 if (!args->from) 2221 arg = EXT4_DEF_LI_WAIT_MULT; 2222 sbi->s_li_wait_mult = arg; 2223 } else if (token == Opt_max_dir_size_kb) { 2224 sbi->s_max_dir_size_kb = arg; 2225 #ifdef CONFIG_EXT4_DEBUG 2226 } else if (token == Opt_fc_debug_max_replay) { 2227 sbi->s_fc_debug_max_replay = arg; 2228 #endif 2229 } else if (token == Opt_stripe) { 2230 sbi->s_stripe = arg; 2231 } else if (token == Opt_resuid) { 2232 uid = make_kuid(current_user_ns(), arg); 2233 if (!uid_valid(uid)) { 2234 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg); 2235 return -1; 2236 } 2237 sbi->s_resuid = uid; 2238 } else if (token == Opt_resgid) { 2239 gid = make_kgid(current_user_ns(), arg); 2240 if (!gid_valid(gid)) { 2241 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg); 2242 return -1; 2243 } 2244 sbi->s_resgid = gid; 2245 } else if (token == Opt_journal_dev) { 2246 if (is_remount) { 2247 ext4_msg(sb, KERN_ERR, 2248 "Cannot specify journal on remount"); 2249 return -1; 2250 } 2251 *journal_devnum = arg; 2252 } else if (token == Opt_journal_path) { 2253 char *journal_path; 2254 struct inode *journal_inode; 2255 struct path path; 2256 int error; 2257 2258 if (is_remount) { 2259 ext4_msg(sb, KERN_ERR, 2260 "Cannot specify journal on remount"); 2261 return -1; 2262 } 2263 journal_path = match_strdup(&args[0]); 2264 if (!journal_path) { 2265 ext4_msg(sb, KERN_ERR, "error: could not dup " 2266 "journal device string"); 2267 return -1; 2268 } 2269 2270 error = kern_path(journal_path, LOOKUP_FOLLOW, &path); 2271 if (error) { 2272 ext4_msg(sb, KERN_ERR, "error: could not find " 2273 "journal device path: error %d", error); 2274 kfree(journal_path); 2275 return -1; 2276 } 2277 2278 journal_inode = d_inode(path.dentry); 2279 if (!S_ISBLK(journal_inode->i_mode)) { 2280 ext4_msg(sb, KERN_ERR, "error: journal path %s " 2281 "is not a block device", journal_path); 2282 path_put(&path); 2283 kfree(journal_path); 2284 return -1; 2285 } 2286 2287 *journal_devnum = new_encode_dev(journal_inode->i_rdev); 2288 path_put(&path); 2289 kfree(journal_path); 2290 } else if (token == Opt_journal_ioprio) { 2291 if (arg > 7) { 2292 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority" 2293 " (must be 0-7)"); 2294 return -1; 2295 } 2296 *journal_ioprio = 2297 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg); 2298 } else if (token == Opt_test_dummy_encryption) { 2299 return ext4_set_test_dummy_encryption(sb, opt, &args[0], 2300 is_remount); 2301 } else if (m->flags & MOPT_DATAJ) { 2302 if (is_remount) { 2303 if (!sbi->s_journal) 2304 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option"); 2305 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) { 2306 ext4_msg(sb, KERN_ERR, 2307 "Cannot change data mode on remount"); 2308 return -1; 2309 } 2310 } else { 2311 clear_opt(sb, DATA_FLAGS); 2312 sbi->s_mount_opt |= m->mount_opt; 2313 } 2314 #ifdef CONFIG_QUOTA 2315 } else if (m->flags & MOPT_QFMT) { 2316 if (sb_any_quota_loaded(sb) && 2317 sbi->s_jquota_fmt != m->mount_opt) { 2318 ext4_msg(sb, KERN_ERR, "Cannot change journaled " 2319 "quota options when quota turned on"); 2320 return -1; 2321 } 2322 if (ext4_has_feature_quota(sb)) { 2323 ext4_msg(sb, KERN_INFO, 2324 "Quota format mount options ignored " 2325 "when QUOTA feature is enabled"); 2326 return 1; 2327 } 2328 sbi->s_jquota_fmt = m->mount_opt; 2329 #endif 2330 } else if (token == Opt_dax || token == Opt_dax_always || 2331 token == Opt_dax_inode || token == Opt_dax_never) { 2332 #ifdef CONFIG_FS_DAX 2333 switch (token) { 2334 case Opt_dax: 2335 case Opt_dax_always: 2336 if (is_remount && 2337 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2338 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2339 fail_dax_change_remount: 2340 ext4_msg(sb, KERN_ERR, "can't change " 2341 "dax mount option while remounting"); 2342 return -1; 2343 } 2344 if (is_remount && 2345 (test_opt(sb, DATA_FLAGS) == 2346 EXT4_MOUNT_JOURNAL_DATA)) { 2347 ext4_msg(sb, KERN_ERR, "can't mount with " 2348 "both data=journal and dax"); 2349 return -1; 2350 } 2351 ext4_msg(sb, KERN_WARNING, 2352 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 2353 sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS; 2354 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2355 break; 2356 case Opt_dax_never: 2357 if (is_remount && 2358 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2359 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) 2360 goto fail_dax_change_remount; 2361 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2362 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2363 break; 2364 case Opt_dax_inode: 2365 if (is_remount && 2366 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2367 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2368 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) 2369 goto fail_dax_change_remount; 2370 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2371 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2372 /* Strictly for printing options */ 2373 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE; 2374 break; 2375 } 2376 #else 2377 ext4_msg(sb, KERN_INFO, "dax option not supported"); 2378 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2379 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2380 return -1; 2381 #endif 2382 } else if (token == Opt_data_err_abort) { 2383 sbi->s_mount_opt |= m->mount_opt; 2384 } else if (token == Opt_data_err_ignore) { 2385 sbi->s_mount_opt &= ~m->mount_opt; 2386 } else { 2387 if (!args->from) 2388 arg = 1; 2389 if (m->flags & MOPT_CLEAR) 2390 arg = !arg; 2391 else if (unlikely(!(m->flags & MOPT_SET))) { 2392 ext4_msg(sb, KERN_WARNING, 2393 "buggy handling of option %s", opt); 2394 WARN_ON(1); 2395 return -1; 2396 } 2397 if (m->flags & MOPT_2) { 2398 if (arg != 0) 2399 sbi->s_mount_opt2 |= m->mount_opt; 2400 else 2401 sbi->s_mount_opt2 &= ~m->mount_opt; 2402 } else { 2403 if (arg != 0) 2404 sbi->s_mount_opt |= m->mount_opt; 2405 else 2406 sbi->s_mount_opt &= ~m->mount_opt; 2407 } 2408 } 2409 return 1; 2410 } 2411 2412 static int parse_options(char *options, struct super_block *sb, 2413 unsigned long *journal_devnum, 2414 unsigned int *journal_ioprio, 2415 int is_remount) 2416 { 2417 struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb); 2418 char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name; 2419 substring_t args[MAX_OPT_ARGS]; 2420 int token; 2421 2422 if (!options) 2423 return 1; 2424 2425 while ((p = strsep(&options, ",")) != NULL) { 2426 if (!*p) 2427 continue; 2428 /* 2429 * Initialize args struct so we know whether arg was 2430 * found; some options take optional arguments. 2431 */ 2432 args[0].to = args[0].from = NULL; 2433 token = match_token(p, tokens, args); 2434 if (handle_mount_opt(sb, p, token, args, journal_devnum, 2435 journal_ioprio, is_remount) < 0) 2436 return 0; 2437 } 2438 #ifdef CONFIG_QUOTA 2439 /* 2440 * We do the test below only for project quotas. 'usrquota' and 2441 * 'grpquota' mount options are allowed even without quota feature 2442 * to support legacy quotas in quota files. 2443 */ 2444 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) { 2445 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. " 2446 "Cannot enable project quota enforcement."); 2447 return 0; 2448 } 2449 usr_qf_name = get_qf_name(sb, sbi, USRQUOTA); 2450 grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA); 2451 if (usr_qf_name || grp_qf_name) { 2452 if (test_opt(sb, USRQUOTA) && usr_qf_name) 2453 clear_opt(sb, USRQUOTA); 2454 2455 if (test_opt(sb, GRPQUOTA) && grp_qf_name) 2456 clear_opt(sb, GRPQUOTA); 2457 2458 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) { 2459 ext4_msg(sb, KERN_ERR, "old and new quota " 2460 "format mixing"); 2461 return 0; 2462 } 2463 2464 if (!sbi->s_jquota_fmt) { 2465 ext4_msg(sb, KERN_ERR, "journaled quota format " 2466 "not specified"); 2467 return 0; 2468 } 2469 } 2470 #endif 2471 if (test_opt(sb, DIOREAD_NOLOCK)) { 2472 int blocksize = 2473 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2474 if (blocksize < PAGE_SIZE) 2475 ext4_msg(sb, KERN_WARNING, "Warning: mounting with an " 2476 "experimental mount option 'dioread_nolock' " 2477 "for blocksize < PAGE_SIZE"); 2478 } 2479 return 1; 2480 } 2481 2482 static inline void ext4_show_quota_options(struct seq_file *seq, 2483 struct super_block *sb) 2484 { 2485 #if defined(CONFIG_QUOTA) 2486 struct ext4_sb_info *sbi = EXT4_SB(sb); 2487 char *usr_qf_name, *grp_qf_name; 2488 2489 if (sbi->s_jquota_fmt) { 2490 char *fmtname = ""; 2491 2492 switch (sbi->s_jquota_fmt) { 2493 case QFMT_VFS_OLD: 2494 fmtname = "vfsold"; 2495 break; 2496 case QFMT_VFS_V0: 2497 fmtname = "vfsv0"; 2498 break; 2499 case QFMT_VFS_V1: 2500 fmtname = "vfsv1"; 2501 break; 2502 } 2503 seq_printf(seq, ",jqfmt=%s", fmtname); 2504 } 2505 2506 rcu_read_lock(); 2507 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2508 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2509 if (usr_qf_name) 2510 seq_show_option(seq, "usrjquota", usr_qf_name); 2511 if (grp_qf_name) 2512 seq_show_option(seq, "grpjquota", grp_qf_name); 2513 rcu_read_unlock(); 2514 #endif 2515 } 2516 2517 static const char *token2str(int token) 2518 { 2519 const struct match_token *t; 2520 2521 for (t = tokens; t->token != Opt_err; t++) 2522 if (t->token == token && !strchr(t->pattern, '=')) 2523 break; 2524 return t->pattern; 2525 } 2526 2527 /* 2528 * Show an option if 2529 * - it's set to a non-default value OR 2530 * - if the per-sb default is different from the global default 2531 */ 2532 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2533 int nodefs) 2534 { 2535 struct ext4_sb_info *sbi = EXT4_SB(sb); 2536 struct ext4_super_block *es = sbi->s_es; 2537 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2538 const struct mount_opts *m; 2539 char sep = nodefs ? '\n' : ','; 2540 2541 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2542 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2543 2544 if (sbi->s_sb_block != 1) 2545 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2546 2547 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2548 int want_set = m->flags & MOPT_SET; 2549 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2550 (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP) 2551 continue; 2552 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2553 continue; /* skip if same as the default */ 2554 if ((want_set && 2555 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2556 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2557 continue; /* select Opt_noFoo vs Opt_Foo */ 2558 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2559 } 2560 2561 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2562 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2563 SEQ_OPTS_PRINT("resuid=%u", 2564 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2565 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2566 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2567 SEQ_OPTS_PRINT("resgid=%u", 2568 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2569 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2570 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2571 SEQ_OPTS_PUTS("errors=remount-ro"); 2572 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2573 SEQ_OPTS_PUTS("errors=continue"); 2574 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2575 SEQ_OPTS_PUTS("errors=panic"); 2576 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2577 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2578 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2579 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2580 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2581 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2582 if (sb->s_flags & SB_I_VERSION) 2583 SEQ_OPTS_PUTS("i_version"); 2584 if (nodefs || sbi->s_stripe) 2585 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2586 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2587 (sbi->s_mount_opt ^ def_mount_opt)) { 2588 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2589 SEQ_OPTS_PUTS("data=journal"); 2590 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2591 SEQ_OPTS_PUTS("data=ordered"); 2592 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2593 SEQ_OPTS_PUTS("data=writeback"); 2594 } 2595 if (nodefs || 2596 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2597 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2598 sbi->s_inode_readahead_blks); 2599 2600 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2601 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2602 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2603 if (nodefs || sbi->s_max_dir_size_kb) 2604 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2605 if (test_opt(sb, DATA_ERR_ABORT)) 2606 SEQ_OPTS_PUTS("data_err=abort"); 2607 2608 fscrypt_show_test_dummy_encryption(seq, sep, sb); 2609 2610 if (sb->s_flags & SB_INLINECRYPT) 2611 SEQ_OPTS_PUTS("inlinecrypt"); 2612 2613 if (test_opt(sb, DAX_ALWAYS)) { 2614 if (IS_EXT2_SB(sb)) 2615 SEQ_OPTS_PUTS("dax"); 2616 else 2617 SEQ_OPTS_PUTS("dax=always"); 2618 } else if (test_opt2(sb, DAX_NEVER)) { 2619 SEQ_OPTS_PUTS("dax=never"); 2620 } else if (test_opt2(sb, DAX_INODE)) { 2621 SEQ_OPTS_PUTS("dax=inode"); 2622 } 2623 ext4_show_quota_options(seq, sb); 2624 return 0; 2625 } 2626 2627 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2628 { 2629 return _ext4_show_options(seq, root->d_sb, 0); 2630 } 2631 2632 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2633 { 2634 struct super_block *sb = seq->private; 2635 int rc; 2636 2637 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2638 rc = _ext4_show_options(seq, sb, 1); 2639 seq_puts(seq, "\n"); 2640 return rc; 2641 } 2642 2643 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2644 int read_only) 2645 { 2646 struct ext4_sb_info *sbi = EXT4_SB(sb); 2647 int err = 0; 2648 2649 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2650 ext4_msg(sb, KERN_ERR, "revision level too high, " 2651 "forcing read-only mode"); 2652 err = -EROFS; 2653 goto done; 2654 } 2655 if (read_only) 2656 goto done; 2657 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2658 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2659 "running e2fsck is recommended"); 2660 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2661 ext4_msg(sb, KERN_WARNING, 2662 "warning: mounting fs with errors, " 2663 "running e2fsck is recommended"); 2664 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2665 le16_to_cpu(es->s_mnt_count) >= 2666 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2667 ext4_msg(sb, KERN_WARNING, 2668 "warning: maximal mount count reached, " 2669 "running e2fsck is recommended"); 2670 else if (le32_to_cpu(es->s_checkinterval) && 2671 (ext4_get_tstamp(es, s_lastcheck) + 2672 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2673 ext4_msg(sb, KERN_WARNING, 2674 "warning: checktime reached, " 2675 "running e2fsck is recommended"); 2676 if (!sbi->s_journal) 2677 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2678 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2679 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2680 le16_add_cpu(&es->s_mnt_count, 1); 2681 ext4_update_tstamp(es, s_mtime); 2682 if (sbi->s_journal) 2683 ext4_set_feature_journal_needs_recovery(sb); 2684 2685 err = ext4_commit_super(sb); 2686 done: 2687 if (test_opt(sb, DEBUG)) 2688 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2689 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2690 sb->s_blocksize, 2691 sbi->s_groups_count, 2692 EXT4_BLOCKS_PER_GROUP(sb), 2693 EXT4_INODES_PER_GROUP(sb), 2694 sbi->s_mount_opt, sbi->s_mount_opt2); 2695 2696 cleancache_init_fs(sb); 2697 return err; 2698 } 2699 2700 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2701 { 2702 struct ext4_sb_info *sbi = EXT4_SB(sb); 2703 struct flex_groups **old_groups, **new_groups; 2704 int size, i, j; 2705 2706 if (!sbi->s_log_groups_per_flex) 2707 return 0; 2708 2709 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2710 if (size <= sbi->s_flex_groups_allocated) 2711 return 0; 2712 2713 new_groups = kvzalloc(roundup_pow_of_two(size * 2714 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 2715 if (!new_groups) { 2716 ext4_msg(sb, KERN_ERR, 2717 "not enough memory for %d flex group pointers", size); 2718 return -ENOMEM; 2719 } 2720 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 2721 new_groups[i] = kvzalloc(roundup_pow_of_two( 2722 sizeof(struct flex_groups)), 2723 GFP_KERNEL); 2724 if (!new_groups[i]) { 2725 for (j = sbi->s_flex_groups_allocated; j < i; j++) 2726 kvfree(new_groups[j]); 2727 kvfree(new_groups); 2728 ext4_msg(sb, KERN_ERR, 2729 "not enough memory for %d flex groups", size); 2730 return -ENOMEM; 2731 } 2732 } 2733 rcu_read_lock(); 2734 old_groups = rcu_dereference(sbi->s_flex_groups); 2735 if (old_groups) 2736 memcpy(new_groups, old_groups, 2737 (sbi->s_flex_groups_allocated * 2738 sizeof(struct flex_groups *))); 2739 rcu_read_unlock(); 2740 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 2741 sbi->s_flex_groups_allocated = size; 2742 if (old_groups) 2743 ext4_kvfree_array_rcu(old_groups); 2744 return 0; 2745 } 2746 2747 static int ext4_fill_flex_info(struct super_block *sb) 2748 { 2749 struct ext4_sb_info *sbi = EXT4_SB(sb); 2750 struct ext4_group_desc *gdp = NULL; 2751 struct flex_groups *fg; 2752 ext4_group_t flex_group; 2753 int i, err; 2754 2755 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2756 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2757 sbi->s_log_groups_per_flex = 0; 2758 return 1; 2759 } 2760 2761 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2762 if (err) 2763 goto failed; 2764 2765 for (i = 0; i < sbi->s_groups_count; i++) { 2766 gdp = ext4_get_group_desc(sb, i, NULL); 2767 2768 flex_group = ext4_flex_group(sbi, i); 2769 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 2770 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 2771 atomic64_add(ext4_free_group_clusters(sb, gdp), 2772 &fg->free_clusters); 2773 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 2774 } 2775 2776 return 1; 2777 failed: 2778 return 0; 2779 } 2780 2781 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2782 struct ext4_group_desc *gdp) 2783 { 2784 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2785 __u16 crc = 0; 2786 __le32 le_group = cpu_to_le32(block_group); 2787 struct ext4_sb_info *sbi = EXT4_SB(sb); 2788 2789 if (ext4_has_metadata_csum(sbi->s_sb)) { 2790 /* Use new metadata_csum algorithm */ 2791 __u32 csum32; 2792 __u16 dummy_csum = 0; 2793 2794 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2795 sizeof(le_group)); 2796 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2797 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2798 sizeof(dummy_csum)); 2799 offset += sizeof(dummy_csum); 2800 if (offset < sbi->s_desc_size) 2801 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2802 sbi->s_desc_size - offset); 2803 2804 crc = csum32 & 0xFFFF; 2805 goto out; 2806 } 2807 2808 /* old crc16 code */ 2809 if (!ext4_has_feature_gdt_csum(sb)) 2810 return 0; 2811 2812 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2813 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2814 crc = crc16(crc, (__u8 *)gdp, offset); 2815 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2816 /* for checksum of struct ext4_group_desc do the rest...*/ 2817 if (ext4_has_feature_64bit(sb) && 2818 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2819 crc = crc16(crc, (__u8 *)gdp + offset, 2820 le16_to_cpu(sbi->s_es->s_desc_size) - 2821 offset); 2822 2823 out: 2824 return cpu_to_le16(crc); 2825 } 2826 2827 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2828 struct ext4_group_desc *gdp) 2829 { 2830 if (ext4_has_group_desc_csum(sb) && 2831 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2832 return 0; 2833 2834 return 1; 2835 } 2836 2837 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2838 struct ext4_group_desc *gdp) 2839 { 2840 if (!ext4_has_group_desc_csum(sb)) 2841 return; 2842 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2843 } 2844 2845 /* Called at mount-time, super-block is locked */ 2846 static int ext4_check_descriptors(struct super_block *sb, 2847 ext4_fsblk_t sb_block, 2848 ext4_group_t *first_not_zeroed) 2849 { 2850 struct ext4_sb_info *sbi = EXT4_SB(sb); 2851 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2852 ext4_fsblk_t last_block; 2853 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2854 ext4_fsblk_t block_bitmap; 2855 ext4_fsblk_t inode_bitmap; 2856 ext4_fsblk_t inode_table; 2857 int flexbg_flag = 0; 2858 ext4_group_t i, grp = sbi->s_groups_count; 2859 2860 if (ext4_has_feature_flex_bg(sb)) 2861 flexbg_flag = 1; 2862 2863 ext4_debug("Checking group descriptors"); 2864 2865 for (i = 0; i < sbi->s_groups_count; i++) { 2866 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2867 2868 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2869 last_block = ext4_blocks_count(sbi->s_es) - 1; 2870 else 2871 last_block = first_block + 2872 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2873 2874 if ((grp == sbi->s_groups_count) && 2875 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2876 grp = i; 2877 2878 block_bitmap = ext4_block_bitmap(sb, gdp); 2879 if (block_bitmap == sb_block) { 2880 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2881 "Block bitmap for group %u overlaps " 2882 "superblock", i); 2883 if (!sb_rdonly(sb)) 2884 return 0; 2885 } 2886 if (block_bitmap >= sb_block + 1 && 2887 block_bitmap <= last_bg_block) { 2888 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2889 "Block bitmap for group %u overlaps " 2890 "block group descriptors", i); 2891 if (!sb_rdonly(sb)) 2892 return 0; 2893 } 2894 if (block_bitmap < first_block || block_bitmap > last_block) { 2895 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2896 "Block bitmap for group %u not in group " 2897 "(block %llu)!", i, block_bitmap); 2898 return 0; 2899 } 2900 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2901 if (inode_bitmap == sb_block) { 2902 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2903 "Inode bitmap for group %u overlaps " 2904 "superblock", i); 2905 if (!sb_rdonly(sb)) 2906 return 0; 2907 } 2908 if (inode_bitmap >= sb_block + 1 && 2909 inode_bitmap <= last_bg_block) { 2910 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2911 "Inode bitmap for group %u overlaps " 2912 "block group descriptors", i); 2913 if (!sb_rdonly(sb)) 2914 return 0; 2915 } 2916 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2917 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2918 "Inode bitmap for group %u not in group " 2919 "(block %llu)!", i, inode_bitmap); 2920 return 0; 2921 } 2922 inode_table = ext4_inode_table(sb, gdp); 2923 if (inode_table == sb_block) { 2924 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2925 "Inode table for group %u overlaps " 2926 "superblock", i); 2927 if (!sb_rdonly(sb)) 2928 return 0; 2929 } 2930 if (inode_table >= sb_block + 1 && 2931 inode_table <= last_bg_block) { 2932 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2933 "Inode table for group %u overlaps " 2934 "block group descriptors", i); 2935 if (!sb_rdonly(sb)) 2936 return 0; 2937 } 2938 if (inode_table < first_block || 2939 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2940 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2941 "Inode table for group %u not in group " 2942 "(block %llu)!", i, inode_table); 2943 return 0; 2944 } 2945 ext4_lock_group(sb, i); 2946 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2947 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2948 "Checksum for group %u failed (%u!=%u)", 2949 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2950 gdp)), le16_to_cpu(gdp->bg_checksum)); 2951 if (!sb_rdonly(sb)) { 2952 ext4_unlock_group(sb, i); 2953 return 0; 2954 } 2955 } 2956 ext4_unlock_group(sb, i); 2957 if (!flexbg_flag) 2958 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2959 } 2960 if (NULL != first_not_zeroed) 2961 *first_not_zeroed = grp; 2962 return 1; 2963 } 2964 2965 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2966 * the superblock) which were deleted from all directories, but held open by 2967 * a process at the time of a crash. We walk the list and try to delete these 2968 * inodes at recovery time (only with a read-write filesystem). 2969 * 2970 * In order to keep the orphan inode chain consistent during traversal (in 2971 * case of crash during recovery), we link each inode into the superblock 2972 * orphan list_head and handle it the same way as an inode deletion during 2973 * normal operation (which journals the operations for us). 2974 * 2975 * We only do an iget() and an iput() on each inode, which is very safe if we 2976 * accidentally point at an in-use or already deleted inode. The worst that 2977 * can happen in this case is that we get a "bit already cleared" message from 2978 * ext4_free_inode(). The only reason we would point at a wrong inode is if 2979 * e2fsck was run on this filesystem, and it must have already done the orphan 2980 * inode cleanup for us, so we can safely abort without any further action. 2981 */ 2982 static void ext4_orphan_cleanup(struct super_block *sb, 2983 struct ext4_super_block *es) 2984 { 2985 unsigned int s_flags = sb->s_flags; 2986 int ret, nr_orphans = 0, nr_truncates = 0; 2987 #ifdef CONFIG_QUOTA 2988 int quota_update = 0; 2989 int i; 2990 #endif 2991 if (!es->s_last_orphan) { 2992 jbd_debug(4, "no orphan inodes to clean up\n"); 2993 return; 2994 } 2995 2996 if (bdev_read_only(sb->s_bdev)) { 2997 ext4_msg(sb, KERN_ERR, "write access " 2998 "unavailable, skipping orphan cleanup"); 2999 return; 3000 } 3001 3002 /* Check if feature set would not allow a r/w mount */ 3003 if (!ext4_feature_set_ok(sb, 0)) { 3004 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 3005 "unknown ROCOMPAT features"); 3006 return; 3007 } 3008 3009 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3010 /* don't clear list on RO mount w/ errors */ 3011 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 3012 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 3013 "clearing orphan list.\n"); 3014 es->s_last_orphan = 0; 3015 } 3016 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3017 return; 3018 } 3019 3020 if (s_flags & SB_RDONLY) { 3021 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 3022 sb->s_flags &= ~SB_RDONLY; 3023 } 3024 #ifdef CONFIG_QUOTA 3025 /* Needed for iput() to work correctly and not trash data */ 3026 sb->s_flags |= SB_ACTIVE; 3027 3028 /* 3029 * Turn on quotas which were not enabled for read-only mounts if 3030 * filesystem has quota feature, so that they are updated correctly. 3031 */ 3032 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 3033 int ret = ext4_enable_quotas(sb); 3034 3035 if (!ret) 3036 quota_update = 1; 3037 else 3038 ext4_msg(sb, KERN_ERR, 3039 "Cannot turn on quotas: error %d", ret); 3040 } 3041 3042 /* Turn on journaled quotas used for old sytle */ 3043 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3044 if (EXT4_SB(sb)->s_qf_names[i]) { 3045 int ret = ext4_quota_on_mount(sb, i); 3046 3047 if (!ret) 3048 quota_update = 1; 3049 else 3050 ext4_msg(sb, KERN_ERR, 3051 "Cannot turn on journaled " 3052 "quota: type %d: error %d", i, ret); 3053 } 3054 } 3055 #endif 3056 3057 while (es->s_last_orphan) { 3058 struct inode *inode; 3059 3060 /* 3061 * We may have encountered an error during cleanup; if 3062 * so, skip the rest. 3063 */ 3064 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3065 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3066 es->s_last_orphan = 0; 3067 break; 3068 } 3069 3070 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 3071 if (IS_ERR(inode)) { 3072 es->s_last_orphan = 0; 3073 break; 3074 } 3075 3076 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 3077 dquot_initialize(inode); 3078 if (inode->i_nlink) { 3079 if (test_opt(sb, DEBUG)) 3080 ext4_msg(sb, KERN_DEBUG, 3081 "%s: truncating inode %lu to %lld bytes", 3082 __func__, inode->i_ino, inode->i_size); 3083 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 3084 inode->i_ino, inode->i_size); 3085 inode_lock(inode); 3086 truncate_inode_pages(inode->i_mapping, inode->i_size); 3087 ret = ext4_truncate(inode); 3088 if (ret) 3089 ext4_std_error(inode->i_sb, ret); 3090 inode_unlock(inode); 3091 nr_truncates++; 3092 } else { 3093 if (test_opt(sb, DEBUG)) 3094 ext4_msg(sb, KERN_DEBUG, 3095 "%s: deleting unreferenced inode %lu", 3096 __func__, inode->i_ino); 3097 jbd_debug(2, "deleting unreferenced inode %lu\n", 3098 inode->i_ino); 3099 nr_orphans++; 3100 } 3101 iput(inode); /* The delete magic happens here! */ 3102 } 3103 3104 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 3105 3106 if (nr_orphans) 3107 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 3108 PLURAL(nr_orphans)); 3109 if (nr_truncates) 3110 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 3111 PLURAL(nr_truncates)); 3112 #ifdef CONFIG_QUOTA 3113 /* Turn off quotas if they were enabled for orphan cleanup */ 3114 if (quota_update) { 3115 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3116 if (sb_dqopt(sb)->files[i]) 3117 dquot_quota_off(sb, i); 3118 } 3119 } 3120 #endif 3121 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 3122 } 3123 3124 /* 3125 * Maximal extent format file size. 3126 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3127 * extent format containers, within a sector_t, and within i_blocks 3128 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3129 * so that won't be a limiting factor. 3130 * 3131 * However there is other limiting factor. We do store extents in the form 3132 * of starting block and length, hence the resulting length of the extent 3133 * covering maximum file size must fit into on-disk format containers as 3134 * well. Given that length is always by 1 unit bigger than max unit (because 3135 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3136 * 3137 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3138 */ 3139 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3140 { 3141 loff_t res; 3142 loff_t upper_limit = MAX_LFS_FILESIZE; 3143 3144 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3145 3146 if (!has_huge_files) { 3147 upper_limit = (1LL << 32) - 1; 3148 3149 /* total blocks in file system block size */ 3150 upper_limit >>= (blkbits - 9); 3151 upper_limit <<= blkbits; 3152 } 3153 3154 /* 3155 * 32-bit extent-start container, ee_block. We lower the maxbytes 3156 * by one fs block, so ee_len can cover the extent of maximum file 3157 * size 3158 */ 3159 res = (1LL << 32) - 1; 3160 res <<= blkbits; 3161 3162 /* Sanity check against vm- & vfs- imposed limits */ 3163 if (res > upper_limit) 3164 res = upper_limit; 3165 3166 return res; 3167 } 3168 3169 /* 3170 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3171 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3172 * We need to be 1 filesystem block less than the 2^48 sector limit. 3173 */ 3174 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3175 { 3176 loff_t res = EXT4_NDIR_BLOCKS; 3177 int meta_blocks; 3178 loff_t upper_limit; 3179 /* This is calculated to be the largest file size for a dense, block 3180 * mapped file such that the file's total number of 512-byte sectors, 3181 * including data and all indirect blocks, does not exceed (2^48 - 1). 3182 * 3183 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3184 * number of 512-byte sectors of the file. 3185 */ 3186 3187 if (!has_huge_files) { 3188 /* 3189 * !has_huge_files or implies that the inode i_block field 3190 * represents total file blocks in 2^32 512-byte sectors == 3191 * size of vfs inode i_blocks * 8 3192 */ 3193 upper_limit = (1LL << 32) - 1; 3194 3195 /* total blocks in file system block size */ 3196 upper_limit >>= (bits - 9); 3197 3198 } else { 3199 /* 3200 * We use 48 bit ext4_inode i_blocks 3201 * With EXT4_HUGE_FILE_FL set the i_blocks 3202 * represent total number of blocks in 3203 * file system block size 3204 */ 3205 upper_limit = (1LL << 48) - 1; 3206 3207 } 3208 3209 /* indirect blocks */ 3210 meta_blocks = 1; 3211 /* double indirect blocks */ 3212 meta_blocks += 1 + (1LL << (bits-2)); 3213 /* tripple indirect blocks */ 3214 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 3215 3216 upper_limit -= meta_blocks; 3217 upper_limit <<= bits; 3218 3219 res += 1LL << (bits-2); 3220 res += 1LL << (2*(bits-2)); 3221 res += 1LL << (3*(bits-2)); 3222 res <<= bits; 3223 if (res > upper_limit) 3224 res = upper_limit; 3225 3226 if (res > MAX_LFS_FILESIZE) 3227 res = MAX_LFS_FILESIZE; 3228 3229 return res; 3230 } 3231 3232 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3233 ext4_fsblk_t logical_sb_block, int nr) 3234 { 3235 struct ext4_sb_info *sbi = EXT4_SB(sb); 3236 ext4_group_t bg, first_meta_bg; 3237 int has_super = 0; 3238 3239 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3240 3241 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3242 return logical_sb_block + nr + 1; 3243 bg = sbi->s_desc_per_block * nr; 3244 if (ext4_bg_has_super(sb, bg)) 3245 has_super = 1; 3246 3247 /* 3248 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3249 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3250 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3251 * compensate. 3252 */ 3253 if (sb->s_blocksize == 1024 && nr == 0 && 3254 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3255 has_super++; 3256 3257 return (has_super + ext4_group_first_block_no(sb, bg)); 3258 } 3259 3260 /** 3261 * ext4_get_stripe_size: Get the stripe size. 3262 * @sbi: In memory super block info 3263 * 3264 * If we have specified it via mount option, then 3265 * use the mount option value. If the value specified at mount time is 3266 * greater than the blocks per group use the super block value. 3267 * If the super block value is greater than blocks per group return 0. 3268 * Allocator needs it be less than blocks per group. 3269 * 3270 */ 3271 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3272 { 3273 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3274 unsigned long stripe_width = 3275 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3276 int ret; 3277 3278 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3279 ret = sbi->s_stripe; 3280 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3281 ret = stripe_width; 3282 else if (stride && stride <= sbi->s_blocks_per_group) 3283 ret = stride; 3284 else 3285 ret = 0; 3286 3287 /* 3288 * If the stripe width is 1, this makes no sense and 3289 * we set it to 0 to turn off stripe handling code. 3290 */ 3291 if (ret <= 1) 3292 ret = 0; 3293 3294 return ret; 3295 } 3296 3297 /* 3298 * Check whether this filesystem can be mounted based on 3299 * the features present and the RDONLY/RDWR mount requested. 3300 * Returns 1 if this filesystem can be mounted as requested, 3301 * 0 if it cannot be. 3302 */ 3303 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 3304 { 3305 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3306 ext4_msg(sb, KERN_ERR, 3307 "Couldn't mount because of " 3308 "unsupported optional features (%x)", 3309 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3310 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3311 return 0; 3312 } 3313 3314 #ifndef CONFIG_UNICODE 3315 if (ext4_has_feature_casefold(sb)) { 3316 ext4_msg(sb, KERN_ERR, 3317 "Filesystem with casefold feature cannot be " 3318 "mounted without CONFIG_UNICODE"); 3319 return 0; 3320 } 3321 #endif 3322 3323 if (readonly) 3324 return 1; 3325 3326 if (ext4_has_feature_readonly(sb)) { 3327 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3328 sb->s_flags |= SB_RDONLY; 3329 return 1; 3330 } 3331 3332 /* Check that feature set is OK for a read-write mount */ 3333 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3334 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3335 "unsupported optional features (%x)", 3336 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3337 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3338 return 0; 3339 } 3340 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3341 ext4_msg(sb, KERN_ERR, 3342 "Can't support bigalloc feature without " 3343 "extents feature\n"); 3344 return 0; 3345 } 3346 3347 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3348 if (!readonly && (ext4_has_feature_quota(sb) || 3349 ext4_has_feature_project(sb))) { 3350 ext4_msg(sb, KERN_ERR, 3351 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3352 return 0; 3353 } 3354 #endif /* CONFIG_QUOTA */ 3355 return 1; 3356 } 3357 3358 /* 3359 * This function is called once a day if we have errors logged 3360 * on the file system 3361 */ 3362 static void print_daily_error_info(struct timer_list *t) 3363 { 3364 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3365 struct super_block *sb = sbi->s_sb; 3366 struct ext4_super_block *es = sbi->s_es; 3367 3368 if (es->s_error_count) 3369 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3370 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3371 le32_to_cpu(es->s_error_count)); 3372 if (es->s_first_error_time) { 3373 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3374 sb->s_id, 3375 ext4_get_tstamp(es, s_first_error_time), 3376 (int) sizeof(es->s_first_error_func), 3377 es->s_first_error_func, 3378 le32_to_cpu(es->s_first_error_line)); 3379 if (es->s_first_error_ino) 3380 printk(KERN_CONT ": inode %u", 3381 le32_to_cpu(es->s_first_error_ino)); 3382 if (es->s_first_error_block) 3383 printk(KERN_CONT ": block %llu", (unsigned long long) 3384 le64_to_cpu(es->s_first_error_block)); 3385 printk(KERN_CONT "\n"); 3386 } 3387 if (es->s_last_error_time) { 3388 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3389 sb->s_id, 3390 ext4_get_tstamp(es, s_last_error_time), 3391 (int) sizeof(es->s_last_error_func), 3392 es->s_last_error_func, 3393 le32_to_cpu(es->s_last_error_line)); 3394 if (es->s_last_error_ino) 3395 printk(KERN_CONT ": inode %u", 3396 le32_to_cpu(es->s_last_error_ino)); 3397 if (es->s_last_error_block) 3398 printk(KERN_CONT ": block %llu", (unsigned long long) 3399 le64_to_cpu(es->s_last_error_block)); 3400 printk(KERN_CONT "\n"); 3401 } 3402 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3403 } 3404 3405 /* Find next suitable group and run ext4_init_inode_table */ 3406 static int ext4_run_li_request(struct ext4_li_request *elr) 3407 { 3408 struct ext4_group_desc *gdp = NULL; 3409 struct super_block *sb = elr->lr_super; 3410 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3411 ext4_group_t group = elr->lr_next_group; 3412 unsigned long timeout = 0; 3413 unsigned int prefetch_ios = 0; 3414 int ret = 0; 3415 3416 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3417 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3418 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3419 if (prefetch_ios) 3420 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3421 prefetch_ios); 3422 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3423 prefetch_ios); 3424 if (group >= elr->lr_next_group) { 3425 ret = 1; 3426 if (elr->lr_first_not_zeroed != ngroups && 3427 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3428 elr->lr_next_group = elr->lr_first_not_zeroed; 3429 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3430 ret = 0; 3431 } 3432 } 3433 return ret; 3434 } 3435 3436 for (; group < ngroups; group++) { 3437 gdp = ext4_get_group_desc(sb, group, NULL); 3438 if (!gdp) { 3439 ret = 1; 3440 break; 3441 } 3442 3443 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3444 break; 3445 } 3446 3447 if (group >= ngroups) 3448 ret = 1; 3449 3450 if (!ret) { 3451 timeout = jiffies; 3452 ret = ext4_init_inode_table(sb, group, 3453 elr->lr_timeout ? 0 : 1); 3454 trace_ext4_lazy_itable_init(sb, group); 3455 if (elr->lr_timeout == 0) { 3456 timeout = (jiffies - timeout) * 3457 EXT4_SB(elr->lr_super)->s_li_wait_mult; 3458 elr->lr_timeout = timeout; 3459 } 3460 elr->lr_next_sched = jiffies + elr->lr_timeout; 3461 elr->lr_next_group = group + 1; 3462 } 3463 return ret; 3464 } 3465 3466 /* 3467 * Remove lr_request from the list_request and free the 3468 * request structure. Should be called with li_list_mtx held 3469 */ 3470 static void ext4_remove_li_request(struct ext4_li_request *elr) 3471 { 3472 if (!elr) 3473 return; 3474 3475 list_del(&elr->lr_request); 3476 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3477 kfree(elr); 3478 } 3479 3480 static void ext4_unregister_li_request(struct super_block *sb) 3481 { 3482 mutex_lock(&ext4_li_mtx); 3483 if (!ext4_li_info) { 3484 mutex_unlock(&ext4_li_mtx); 3485 return; 3486 } 3487 3488 mutex_lock(&ext4_li_info->li_list_mtx); 3489 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3490 mutex_unlock(&ext4_li_info->li_list_mtx); 3491 mutex_unlock(&ext4_li_mtx); 3492 } 3493 3494 static struct task_struct *ext4_lazyinit_task; 3495 3496 /* 3497 * This is the function where ext4lazyinit thread lives. It walks 3498 * through the request list searching for next scheduled filesystem. 3499 * When such a fs is found, run the lazy initialization request 3500 * (ext4_rn_li_request) and keep track of the time spend in this 3501 * function. Based on that time we compute next schedule time of 3502 * the request. When walking through the list is complete, compute 3503 * next waking time and put itself into sleep. 3504 */ 3505 static int ext4_lazyinit_thread(void *arg) 3506 { 3507 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3508 struct list_head *pos, *n; 3509 struct ext4_li_request *elr; 3510 unsigned long next_wakeup, cur; 3511 3512 BUG_ON(NULL == eli); 3513 3514 cont_thread: 3515 while (true) { 3516 next_wakeup = MAX_JIFFY_OFFSET; 3517 3518 mutex_lock(&eli->li_list_mtx); 3519 if (list_empty(&eli->li_request_list)) { 3520 mutex_unlock(&eli->li_list_mtx); 3521 goto exit_thread; 3522 } 3523 list_for_each_safe(pos, n, &eli->li_request_list) { 3524 int err = 0; 3525 int progress = 0; 3526 elr = list_entry(pos, struct ext4_li_request, 3527 lr_request); 3528 3529 if (time_before(jiffies, elr->lr_next_sched)) { 3530 if (time_before(elr->lr_next_sched, next_wakeup)) 3531 next_wakeup = elr->lr_next_sched; 3532 continue; 3533 } 3534 if (down_read_trylock(&elr->lr_super->s_umount)) { 3535 if (sb_start_write_trylock(elr->lr_super)) { 3536 progress = 1; 3537 /* 3538 * We hold sb->s_umount, sb can not 3539 * be removed from the list, it is 3540 * now safe to drop li_list_mtx 3541 */ 3542 mutex_unlock(&eli->li_list_mtx); 3543 err = ext4_run_li_request(elr); 3544 sb_end_write(elr->lr_super); 3545 mutex_lock(&eli->li_list_mtx); 3546 n = pos->next; 3547 } 3548 up_read((&elr->lr_super->s_umount)); 3549 } 3550 /* error, remove the lazy_init job */ 3551 if (err) { 3552 ext4_remove_li_request(elr); 3553 continue; 3554 } 3555 if (!progress) { 3556 elr->lr_next_sched = jiffies + 3557 (prandom_u32() 3558 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3559 } 3560 if (time_before(elr->lr_next_sched, next_wakeup)) 3561 next_wakeup = elr->lr_next_sched; 3562 } 3563 mutex_unlock(&eli->li_list_mtx); 3564 3565 try_to_freeze(); 3566 3567 cur = jiffies; 3568 if ((time_after_eq(cur, next_wakeup)) || 3569 (MAX_JIFFY_OFFSET == next_wakeup)) { 3570 cond_resched(); 3571 continue; 3572 } 3573 3574 schedule_timeout_interruptible(next_wakeup - cur); 3575 3576 if (kthread_should_stop()) { 3577 ext4_clear_request_list(); 3578 goto exit_thread; 3579 } 3580 } 3581 3582 exit_thread: 3583 /* 3584 * It looks like the request list is empty, but we need 3585 * to check it under the li_list_mtx lock, to prevent any 3586 * additions into it, and of course we should lock ext4_li_mtx 3587 * to atomically free the list and ext4_li_info, because at 3588 * this point another ext4 filesystem could be registering 3589 * new one. 3590 */ 3591 mutex_lock(&ext4_li_mtx); 3592 mutex_lock(&eli->li_list_mtx); 3593 if (!list_empty(&eli->li_request_list)) { 3594 mutex_unlock(&eli->li_list_mtx); 3595 mutex_unlock(&ext4_li_mtx); 3596 goto cont_thread; 3597 } 3598 mutex_unlock(&eli->li_list_mtx); 3599 kfree(ext4_li_info); 3600 ext4_li_info = NULL; 3601 mutex_unlock(&ext4_li_mtx); 3602 3603 return 0; 3604 } 3605 3606 static void ext4_clear_request_list(void) 3607 { 3608 struct list_head *pos, *n; 3609 struct ext4_li_request *elr; 3610 3611 mutex_lock(&ext4_li_info->li_list_mtx); 3612 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3613 elr = list_entry(pos, struct ext4_li_request, 3614 lr_request); 3615 ext4_remove_li_request(elr); 3616 } 3617 mutex_unlock(&ext4_li_info->li_list_mtx); 3618 } 3619 3620 static int ext4_run_lazyinit_thread(void) 3621 { 3622 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3623 ext4_li_info, "ext4lazyinit"); 3624 if (IS_ERR(ext4_lazyinit_task)) { 3625 int err = PTR_ERR(ext4_lazyinit_task); 3626 ext4_clear_request_list(); 3627 kfree(ext4_li_info); 3628 ext4_li_info = NULL; 3629 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3630 "initialization thread\n", 3631 err); 3632 return err; 3633 } 3634 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3635 return 0; 3636 } 3637 3638 /* 3639 * Check whether it make sense to run itable init. thread or not. 3640 * If there is at least one uninitialized inode table, return 3641 * corresponding group number, else the loop goes through all 3642 * groups and return total number of groups. 3643 */ 3644 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3645 { 3646 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3647 struct ext4_group_desc *gdp = NULL; 3648 3649 if (!ext4_has_group_desc_csum(sb)) 3650 return ngroups; 3651 3652 for (group = 0; group < ngroups; group++) { 3653 gdp = ext4_get_group_desc(sb, group, NULL); 3654 if (!gdp) 3655 continue; 3656 3657 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3658 break; 3659 } 3660 3661 return group; 3662 } 3663 3664 static int ext4_li_info_new(void) 3665 { 3666 struct ext4_lazy_init *eli = NULL; 3667 3668 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3669 if (!eli) 3670 return -ENOMEM; 3671 3672 INIT_LIST_HEAD(&eli->li_request_list); 3673 mutex_init(&eli->li_list_mtx); 3674 3675 eli->li_state |= EXT4_LAZYINIT_QUIT; 3676 3677 ext4_li_info = eli; 3678 3679 return 0; 3680 } 3681 3682 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3683 ext4_group_t start) 3684 { 3685 struct ext4_li_request *elr; 3686 3687 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3688 if (!elr) 3689 return NULL; 3690 3691 elr->lr_super = sb; 3692 elr->lr_first_not_zeroed = start; 3693 if (test_opt(sb, PREFETCH_BLOCK_BITMAPS)) 3694 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3695 else { 3696 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3697 elr->lr_next_group = start; 3698 } 3699 3700 /* 3701 * Randomize first schedule time of the request to 3702 * spread the inode table initialization requests 3703 * better. 3704 */ 3705 elr->lr_next_sched = jiffies + (prandom_u32() % 3706 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3707 return elr; 3708 } 3709 3710 int ext4_register_li_request(struct super_block *sb, 3711 ext4_group_t first_not_zeroed) 3712 { 3713 struct ext4_sb_info *sbi = EXT4_SB(sb); 3714 struct ext4_li_request *elr = NULL; 3715 ext4_group_t ngroups = sbi->s_groups_count; 3716 int ret = 0; 3717 3718 mutex_lock(&ext4_li_mtx); 3719 if (sbi->s_li_request != NULL) { 3720 /* 3721 * Reset timeout so it can be computed again, because 3722 * s_li_wait_mult might have changed. 3723 */ 3724 sbi->s_li_request->lr_timeout = 0; 3725 goto out; 3726 } 3727 3728 if (!test_opt(sb, PREFETCH_BLOCK_BITMAPS) && 3729 (first_not_zeroed == ngroups || sb_rdonly(sb) || 3730 !test_opt(sb, INIT_INODE_TABLE))) 3731 goto out; 3732 3733 elr = ext4_li_request_new(sb, first_not_zeroed); 3734 if (!elr) { 3735 ret = -ENOMEM; 3736 goto out; 3737 } 3738 3739 if (NULL == ext4_li_info) { 3740 ret = ext4_li_info_new(); 3741 if (ret) 3742 goto out; 3743 } 3744 3745 mutex_lock(&ext4_li_info->li_list_mtx); 3746 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3747 mutex_unlock(&ext4_li_info->li_list_mtx); 3748 3749 sbi->s_li_request = elr; 3750 /* 3751 * set elr to NULL here since it has been inserted to 3752 * the request_list and the removal and free of it is 3753 * handled by ext4_clear_request_list from now on. 3754 */ 3755 elr = NULL; 3756 3757 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3758 ret = ext4_run_lazyinit_thread(); 3759 if (ret) 3760 goto out; 3761 } 3762 out: 3763 mutex_unlock(&ext4_li_mtx); 3764 if (ret) 3765 kfree(elr); 3766 return ret; 3767 } 3768 3769 /* 3770 * We do not need to lock anything since this is called on 3771 * module unload. 3772 */ 3773 static void ext4_destroy_lazyinit_thread(void) 3774 { 3775 /* 3776 * If thread exited earlier 3777 * there's nothing to be done. 3778 */ 3779 if (!ext4_li_info || !ext4_lazyinit_task) 3780 return; 3781 3782 kthread_stop(ext4_lazyinit_task); 3783 } 3784 3785 static int set_journal_csum_feature_set(struct super_block *sb) 3786 { 3787 int ret = 1; 3788 int compat, incompat; 3789 struct ext4_sb_info *sbi = EXT4_SB(sb); 3790 3791 if (ext4_has_metadata_csum(sb)) { 3792 /* journal checksum v3 */ 3793 compat = 0; 3794 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3795 } else { 3796 /* journal checksum v1 */ 3797 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3798 incompat = 0; 3799 } 3800 3801 jbd2_journal_clear_features(sbi->s_journal, 3802 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3803 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3804 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3805 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3806 ret = jbd2_journal_set_features(sbi->s_journal, 3807 compat, 0, 3808 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3809 incompat); 3810 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3811 ret = jbd2_journal_set_features(sbi->s_journal, 3812 compat, 0, 3813 incompat); 3814 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3815 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3816 } else { 3817 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3818 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3819 } 3820 3821 return ret; 3822 } 3823 3824 /* 3825 * Note: calculating the overhead so we can be compatible with 3826 * historical BSD practice is quite difficult in the face of 3827 * clusters/bigalloc. This is because multiple metadata blocks from 3828 * different block group can end up in the same allocation cluster. 3829 * Calculating the exact overhead in the face of clustered allocation 3830 * requires either O(all block bitmaps) in memory or O(number of block 3831 * groups**2) in time. We will still calculate the superblock for 3832 * older file systems --- and if we come across with a bigalloc file 3833 * system with zero in s_overhead_clusters the estimate will be close to 3834 * correct especially for very large cluster sizes --- but for newer 3835 * file systems, it's better to calculate this figure once at mkfs 3836 * time, and store it in the superblock. If the superblock value is 3837 * present (even for non-bigalloc file systems), we will use it. 3838 */ 3839 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3840 char *buf) 3841 { 3842 struct ext4_sb_info *sbi = EXT4_SB(sb); 3843 struct ext4_group_desc *gdp; 3844 ext4_fsblk_t first_block, last_block, b; 3845 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3846 int s, j, count = 0; 3847 3848 if (!ext4_has_feature_bigalloc(sb)) 3849 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3850 sbi->s_itb_per_group + 2); 3851 3852 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3853 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3854 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3855 for (i = 0; i < ngroups; i++) { 3856 gdp = ext4_get_group_desc(sb, i, NULL); 3857 b = ext4_block_bitmap(sb, gdp); 3858 if (b >= first_block && b <= last_block) { 3859 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3860 count++; 3861 } 3862 b = ext4_inode_bitmap(sb, gdp); 3863 if (b >= first_block && b <= last_block) { 3864 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3865 count++; 3866 } 3867 b = ext4_inode_table(sb, gdp); 3868 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3869 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3870 int c = EXT4_B2C(sbi, b - first_block); 3871 ext4_set_bit(c, buf); 3872 count++; 3873 } 3874 if (i != grp) 3875 continue; 3876 s = 0; 3877 if (ext4_bg_has_super(sb, grp)) { 3878 ext4_set_bit(s++, buf); 3879 count++; 3880 } 3881 j = ext4_bg_num_gdb(sb, grp); 3882 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3883 ext4_error(sb, "Invalid number of block group " 3884 "descriptor blocks: %d", j); 3885 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3886 } 3887 count += j; 3888 for (; j > 0; j--) 3889 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3890 } 3891 if (!count) 3892 return 0; 3893 return EXT4_CLUSTERS_PER_GROUP(sb) - 3894 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3895 } 3896 3897 /* 3898 * Compute the overhead and stash it in sbi->s_overhead 3899 */ 3900 int ext4_calculate_overhead(struct super_block *sb) 3901 { 3902 struct ext4_sb_info *sbi = EXT4_SB(sb); 3903 struct ext4_super_block *es = sbi->s_es; 3904 struct inode *j_inode; 3905 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3906 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3907 ext4_fsblk_t overhead = 0; 3908 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3909 3910 if (!buf) 3911 return -ENOMEM; 3912 3913 /* 3914 * Compute the overhead (FS structures). This is constant 3915 * for a given filesystem unless the number of block groups 3916 * changes so we cache the previous value until it does. 3917 */ 3918 3919 /* 3920 * All of the blocks before first_data_block are overhead 3921 */ 3922 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3923 3924 /* 3925 * Add the overhead found in each block group 3926 */ 3927 for (i = 0; i < ngroups; i++) { 3928 int blks; 3929 3930 blks = count_overhead(sb, i, buf); 3931 overhead += blks; 3932 if (blks) 3933 memset(buf, 0, PAGE_SIZE); 3934 cond_resched(); 3935 } 3936 3937 /* 3938 * Add the internal journal blocks whether the journal has been 3939 * loaded or not 3940 */ 3941 if (sbi->s_journal && !sbi->s_journal_bdev) 3942 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 3943 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 3944 /* j_inum for internal journal is non-zero */ 3945 j_inode = ext4_get_journal_inode(sb, j_inum); 3946 if (j_inode) { 3947 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3948 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3949 iput(j_inode); 3950 } else { 3951 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3952 } 3953 } 3954 sbi->s_overhead = overhead; 3955 smp_wmb(); 3956 free_page((unsigned long) buf); 3957 return 0; 3958 } 3959 3960 static void ext4_set_resv_clusters(struct super_block *sb) 3961 { 3962 ext4_fsblk_t resv_clusters; 3963 struct ext4_sb_info *sbi = EXT4_SB(sb); 3964 3965 /* 3966 * There's no need to reserve anything when we aren't using extents. 3967 * The space estimates are exact, there are no unwritten extents, 3968 * hole punching doesn't need new metadata... This is needed especially 3969 * to keep ext2/3 backward compatibility. 3970 */ 3971 if (!ext4_has_feature_extents(sb)) 3972 return; 3973 /* 3974 * By default we reserve 2% or 4096 clusters, whichever is smaller. 3975 * This should cover the situations where we can not afford to run 3976 * out of space like for example punch hole, or converting 3977 * unwritten extents in delalloc path. In most cases such 3978 * allocation would require 1, or 2 blocks, higher numbers are 3979 * very rare. 3980 */ 3981 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 3982 sbi->s_cluster_bits); 3983 3984 do_div(resv_clusters, 50); 3985 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 3986 3987 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 3988 } 3989 3990 static const char *ext4_quota_mode(struct super_block *sb) 3991 { 3992 #ifdef CONFIG_QUOTA 3993 if (!ext4_quota_capable(sb)) 3994 return "none"; 3995 3996 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 3997 return "journalled"; 3998 else 3999 return "writeback"; 4000 #else 4001 return "disabled"; 4002 #endif 4003 } 4004 4005 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 4006 { 4007 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 4008 char *orig_data = kstrdup(data, GFP_KERNEL); 4009 struct buffer_head *bh, **group_desc; 4010 struct ext4_super_block *es = NULL; 4011 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4012 struct flex_groups **flex_groups; 4013 ext4_fsblk_t block; 4014 ext4_fsblk_t sb_block = get_sb_block(&data); 4015 ext4_fsblk_t logical_sb_block; 4016 unsigned long offset = 0; 4017 unsigned long journal_devnum = 0; 4018 unsigned long def_mount_opts; 4019 struct inode *root; 4020 const char *descr; 4021 int ret = -ENOMEM; 4022 int blocksize, clustersize; 4023 unsigned int db_count; 4024 unsigned int i; 4025 int needs_recovery, has_huge_files; 4026 __u64 blocks_count; 4027 int err = 0; 4028 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 4029 ext4_group_t first_not_zeroed; 4030 4031 if ((data && !orig_data) || !sbi) 4032 goto out_free_base; 4033 4034 sbi->s_daxdev = dax_dev; 4035 sbi->s_blockgroup_lock = 4036 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4037 if (!sbi->s_blockgroup_lock) 4038 goto out_free_base; 4039 4040 sb->s_fs_info = sbi; 4041 sbi->s_sb = sb; 4042 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 4043 sbi->s_sb_block = sb_block; 4044 sbi->s_sectors_written_start = 4045 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 4046 4047 /* Cleanup superblock name */ 4048 strreplace(sb->s_id, '/', '!'); 4049 4050 /* -EINVAL is default */ 4051 ret = -EINVAL; 4052 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4053 if (!blocksize) { 4054 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4055 goto out_fail; 4056 } 4057 4058 /* 4059 * The ext4 superblock will not be buffer aligned for other than 1kB 4060 * block sizes. We need to calculate the offset from buffer start. 4061 */ 4062 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4063 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4064 offset = do_div(logical_sb_block, blocksize); 4065 } else { 4066 logical_sb_block = sb_block; 4067 } 4068 4069 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4070 if (IS_ERR(bh)) { 4071 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4072 ret = PTR_ERR(bh); 4073 goto out_fail; 4074 } 4075 /* 4076 * Note: s_es must be initialized as soon as possible because 4077 * some ext4 macro-instructions depend on its value 4078 */ 4079 es = (struct ext4_super_block *) (bh->b_data + offset); 4080 sbi->s_es = es; 4081 sb->s_magic = le16_to_cpu(es->s_magic); 4082 if (sb->s_magic != EXT4_SUPER_MAGIC) 4083 goto cantfind_ext4; 4084 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 4085 4086 /* Warn if metadata_csum and gdt_csum are both set. */ 4087 if (ext4_has_feature_metadata_csum(sb) && 4088 ext4_has_feature_gdt_csum(sb)) 4089 ext4_warning(sb, "metadata_csum and uninit_bg are " 4090 "redundant flags; please run fsck."); 4091 4092 /* Check for a known checksum algorithm */ 4093 if (!ext4_verify_csum_type(sb, es)) { 4094 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4095 "unknown checksum algorithm."); 4096 silent = 1; 4097 goto cantfind_ext4; 4098 } 4099 4100 /* Load the checksum driver */ 4101 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4102 if (IS_ERR(sbi->s_chksum_driver)) { 4103 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4104 ret = PTR_ERR(sbi->s_chksum_driver); 4105 sbi->s_chksum_driver = NULL; 4106 goto failed_mount; 4107 } 4108 4109 /* Check superblock checksum */ 4110 if (!ext4_superblock_csum_verify(sb, es)) { 4111 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4112 "invalid superblock checksum. Run e2fsck?"); 4113 silent = 1; 4114 ret = -EFSBADCRC; 4115 goto cantfind_ext4; 4116 } 4117 4118 /* Precompute checksum seed for all metadata */ 4119 if (ext4_has_feature_csum_seed(sb)) 4120 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4121 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4122 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4123 sizeof(es->s_uuid)); 4124 4125 /* Set defaults before we parse the mount options */ 4126 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4127 set_opt(sb, INIT_INODE_TABLE); 4128 if (def_mount_opts & EXT4_DEFM_DEBUG) 4129 set_opt(sb, DEBUG); 4130 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4131 set_opt(sb, GRPID); 4132 if (def_mount_opts & EXT4_DEFM_UID16) 4133 set_opt(sb, NO_UID32); 4134 /* xattr user namespace & acls are now defaulted on */ 4135 set_opt(sb, XATTR_USER); 4136 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4137 set_opt(sb, POSIX_ACL); 4138 #endif 4139 if (ext4_has_feature_fast_commit(sb)) 4140 set_opt2(sb, JOURNAL_FAST_COMMIT); 4141 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4142 if (ext4_has_metadata_csum(sb)) 4143 set_opt(sb, JOURNAL_CHECKSUM); 4144 4145 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4146 set_opt(sb, JOURNAL_DATA); 4147 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4148 set_opt(sb, ORDERED_DATA); 4149 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4150 set_opt(sb, WRITEBACK_DATA); 4151 4152 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 4153 set_opt(sb, ERRORS_PANIC); 4154 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 4155 set_opt(sb, ERRORS_CONT); 4156 else 4157 set_opt(sb, ERRORS_RO); 4158 /* block_validity enabled by default; disable with noblock_validity */ 4159 set_opt(sb, BLOCK_VALIDITY); 4160 if (def_mount_opts & EXT4_DEFM_DISCARD) 4161 set_opt(sb, DISCARD); 4162 4163 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 4164 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 4165 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 4166 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 4167 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 4168 4169 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4170 set_opt(sb, BARRIER); 4171 4172 /* 4173 * enable delayed allocation by default 4174 * Use -o nodelalloc to turn it off 4175 */ 4176 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4177 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4178 set_opt(sb, DELALLOC); 4179 4180 /* 4181 * set default s_li_wait_mult for lazyinit, for the case there is 4182 * no mount option specified. 4183 */ 4184 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 4185 4186 if (le32_to_cpu(es->s_log_block_size) > 4187 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4188 ext4_msg(sb, KERN_ERR, 4189 "Invalid log block size: %u", 4190 le32_to_cpu(es->s_log_block_size)); 4191 goto failed_mount; 4192 } 4193 if (le32_to_cpu(es->s_log_cluster_size) > 4194 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4195 ext4_msg(sb, KERN_ERR, 4196 "Invalid log cluster size: %u", 4197 le32_to_cpu(es->s_log_cluster_size)); 4198 goto failed_mount; 4199 } 4200 4201 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4202 4203 if (blocksize == PAGE_SIZE) 4204 set_opt(sb, DIOREAD_NOLOCK); 4205 4206 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4207 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4208 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4209 } else { 4210 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4211 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4212 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4213 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4214 sbi->s_first_ino); 4215 goto failed_mount; 4216 } 4217 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4218 (!is_power_of_2(sbi->s_inode_size)) || 4219 (sbi->s_inode_size > blocksize)) { 4220 ext4_msg(sb, KERN_ERR, 4221 "unsupported inode size: %d", 4222 sbi->s_inode_size); 4223 ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize); 4224 goto failed_mount; 4225 } 4226 /* 4227 * i_atime_extra is the last extra field available for 4228 * [acm]times in struct ext4_inode. Checking for that 4229 * field should suffice to ensure we have extra space 4230 * for all three. 4231 */ 4232 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4233 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4234 sb->s_time_gran = 1; 4235 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4236 } else { 4237 sb->s_time_gran = NSEC_PER_SEC; 4238 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4239 } 4240 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4241 } 4242 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4243 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4244 EXT4_GOOD_OLD_INODE_SIZE; 4245 if (ext4_has_feature_extra_isize(sb)) { 4246 unsigned v, max = (sbi->s_inode_size - 4247 EXT4_GOOD_OLD_INODE_SIZE); 4248 4249 v = le16_to_cpu(es->s_want_extra_isize); 4250 if (v > max) { 4251 ext4_msg(sb, KERN_ERR, 4252 "bad s_want_extra_isize: %d", v); 4253 goto failed_mount; 4254 } 4255 if (sbi->s_want_extra_isize < v) 4256 sbi->s_want_extra_isize = v; 4257 4258 v = le16_to_cpu(es->s_min_extra_isize); 4259 if (v > max) { 4260 ext4_msg(sb, KERN_ERR, 4261 "bad s_min_extra_isize: %d", v); 4262 goto failed_mount; 4263 } 4264 if (sbi->s_want_extra_isize < v) 4265 sbi->s_want_extra_isize = v; 4266 } 4267 } 4268 4269 if (sbi->s_es->s_mount_opts[0]) { 4270 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 4271 sizeof(sbi->s_es->s_mount_opts), 4272 GFP_KERNEL); 4273 if (!s_mount_opts) 4274 goto failed_mount; 4275 if (!parse_options(s_mount_opts, sb, &journal_devnum, 4276 &journal_ioprio, 0)) { 4277 ext4_msg(sb, KERN_WARNING, 4278 "failed to parse options in superblock: %s", 4279 s_mount_opts); 4280 } 4281 kfree(s_mount_opts); 4282 } 4283 sbi->s_def_mount_opt = sbi->s_mount_opt; 4284 if (!parse_options((char *) data, sb, &journal_devnum, 4285 &journal_ioprio, 0)) 4286 goto failed_mount; 4287 4288 #ifdef CONFIG_UNICODE 4289 if (ext4_has_feature_casefold(sb) && !sb->s_encoding) { 4290 const struct ext4_sb_encodings *encoding_info; 4291 struct unicode_map *encoding; 4292 __u16 encoding_flags; 4293 4294 if (ext4_has_feature_encrypt(sb)) { 4295 ext4_msg(sb, KERN_ERR, 4296 "Can't mount with encoding and encryption"); 4297 goto failed_mount; 4298 } 4299 4300 if (ext4_sb_read_encoding(es, &encoding_info, 4301 &encoding_flags)) { 4302 ext4_msg(sb, KERN_ERR, 4303 "Encoding requested by superblock is unknown"); 4304 goto failed_mount; 4305 } 4306 4307 encoding = utf8_load(encoding_info->version); 4308 if (IS_ERR(encoding)) { 4309 ext4_msg(sb, KERN_ERR, 4310 "can't mount with superblock charset: %s-%s " 4311 "not supported by the kernel. flags: 0x%x.", 4312 encoding_info->name, encoding_info->version, 4313 encoding_flags); 4314 goto failed_mount; 4315 } 4316 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4317 "%s-%s with flags 0x%hx", encoding_info->name, 4318 encoding_info->version?:"\b", encoding_flags); 4319 4320 sb->s_encoding = encoding; 4321 sb->s_encoding_flags = encoding_flags; 4322 } 4323 #endif 4324 4325 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4326 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n"); 4327 /* can't mount with both data=journal and dioread_nolock. */ 4328 clear_opt(sb, DIOREAD_NOLOCK); 4329 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4330 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4331 ext4_msg(sb, KERN_ERR, "can't mount with " 4332 "both data=journal and delalloc"); 4333 goto failed_mount; 4334 } 4335 if (test_opt(sb, DAX_ALWAYS)) { 4336 ext4_msg(sb, KERN_ERR, "can't mount with " 4337 "both data=journal and dax"); 4338 goto failed_mount; 4339 } 4340 if (ext4_has_feature_encrypt(sb)) { 4341 ext4_msg(sb, KERN_WARNING, 4342 "encrypted files will use data=ordered " 4343 "instead of data journaling mode"); 4344 } 4345 if (test_opt(sb, DELALLOC)) 4346 clear_opt(sb, DELALLOC); 4347 } else { 4348 sb->s_iflags |= SB_I_CGROUPWB; 4349 } 4350 4351 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 4352 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 4353 4354 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4355 (ext4_has_compat_features(sb) || 4356 ext4_has_ro_compat_features(sb) || 4357 ext4_has_incompat_features(sb))) 4358 ext4_msg(sb, KERN_WARNING, 4359 "feature flags set on rev 0 fs, " 4360 "running e2fsck is recommended"); 4361 4362 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4363 set_opt2(sb, HURD_COMPAT); 4364 if (ext4_has_feature_64bit(sb)) { 4365 ext4_msg(sb, KERN_ERR, 4366 "The Hurd can't support 64-bit file systems"); 4367 goto failed_mount; 4368 } 4369 4370 /* 4371 * ea_inode feature uses l_i_version field which is not 4372 * available in HURD_COMPAT mode. 4373 */ 4374 if (ext4_has_feature_ea_inode(sb)) { 4375 ext4_msg(sb, KERN_ERR, 4376 "ea_inode feature is not supported for Hurd"); 4377 goto failed_mount; 4378 } 4379 } 4380 4381 if (IS_EXT2_SB(sb)) { 4382 if (ext2_feature_set_ok(sb)) 4383 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4384 "using the ext4 subsystem"); 4385 else { 4386 /* 4387 * If we're probing be silent, if this looks like 4388 * it's actually an ext[34] filesystem. 4389 */ 4390 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4391 goto failed_mount; 4392 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4393 "to feature incompatibilities"); 4394 goto failed_mount; 4395 } 4396 } 4397 4398 if (IS_EXT3_SB(sb)) { 4399 if (ext3_feature_set_ok(sb)) 4400 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4401 "using the ext4 subsystem"); 4402 else { 4403 /* 4404 * If we're probing be silent, if this looks like 4405 * it's actually an ext4 filesystem. 4406 */ 4407 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4408 goto failed_mount; 4409 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4410 "to feature incompatibilities"); 4411 goto failed_mount; 4412 } 4413 } 4414 4415 /* 4416 * Check feature flags regardless of the revision level, since we 4417 * previously didn't change the revision level when setting the flags, 4418 * so there is a chance incompat flags are set on a rev 0 filesystem. 4419 */ 4420 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4421 goto failed_mount; 4422 4423 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 4424 ext4_msg(sb, KERN_ERR, 4425 "Number of reserved GDT blocks insanely large: %d", 4426 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4427 goto failed_mount; 4428 } 4429 4430 if (bdev_dax_supported(sb->s_bdev, blocksize)) 4431 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4432 4433 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4434 if (ext4_has_feature_inline_data(sb)) { 4435 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4436 " that may contain inline data"); 4437 goto failed_mount; 4438 } 4439 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4440 ext4_msg(sb, KERN_ERR, 4441 "DAX unsupported by block device."); 4442 goto failed_mount; 4443 } 4444 } 4445 4446 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4447 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4448 es->s_encryption_level); 4449 goto failed_mount; 4450 } 4451 4452 if (sb->s_blocksize != blocksize) { 4453 /* Validate the filesystem blocksize */ 4454 if (!sb_set_blocksize(sb, blocksize)) { 4455 ext4_msg(sb, KERN_ERR, "bad block size %d", 4456 blocksize); 4457 goto failed_mount; 4458 } 4459 4460 brelse(bh); 4461 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4462 offset = do_div(logical_sb_block, blocksize); 4463 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4464 if (IS_ERR(bh)) { 4465 ext4_msg(sb, KERN_ERR, 4466 "Can't read superblock on 2nd try"); 4467 ret = PTR_ERR(bh); 4468 bh = NULL; 4469 goto failed_mount; 4470 } 4471 es = (struct ext4_super_block *)(bh->b_data + offset); 4472 sbi->s_es = es; 4473 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4474 ext4_msg(sb, KERN_ERR, 4475 "Magic mismatch, very weird!"); 4476 goto failed_mount; 4477 } 4478 } 4479 4480 has_huge_files = ext4_has_feature_huge_file(sb); 4481 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4482 has_huge_files); 4483 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4484 4485 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4486 if (ext4_has_feature_64bit(sb)) { 4487 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4488 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4489 !is_power_of_2(sbi->s_desc_size)) { 4490 ext4_msg(sb, KERN_ERR, 4491 "unsupported descriptor size %lu", 4492 sbi->s_desc_size); 4493 goto failed_mount; 4494 } 4495 } else 4496 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4497 4498 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4499 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4500 4501 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4502 if (sbi->s_inodes_per_block == 0) 4503 goto cantfind_ext4; 4504 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4505 sbi->s_inodes_per_group > blocksize * 8) { 4506 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4507 sbi->s_inodes_per_group); 4508 goto failed_mount; 4509 } 4510 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4511 sbi->s_inodes_per_block; 4512 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4513 sbi->s_sbh = bh; 4514 sbi->s_mount_state = le16_to_cpu(es->s_state); 4515 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4516 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4517 4518 for (i = 0; i < 4; i++) 4519 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4520 sbi->s_def_hash_version = es->s_def_hash_version; 4521 if (ext4_has_feature_dir_index(sb)) { 4522 i = le32_to_cpu(es->s_flags); 4523 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4524 sbi->s_hash_unsigned = 3; 4525 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4526 #ifdef __CHAR_UNSIGNED__ 4527 if (!sb_rdonly(sb)) 4528 es->s_flags |= 4529 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4530 sbi->s_hash_unsigned = 3; 4531 #else 4532 if (!sb_rdonly(sb)) 4533 es->s_flags |= 4534 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4535 #endif 4536 } 4537 } 4538 4539 /* Handle clustersize */ 4540 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4541 if (ext4_has_feature_bigalloc(sb)) { 4542 if (clustersize < blocksize) { 4543 ext4_msg(sb, KERN_ERR, 4544 "cluster size (%d) smaller than " 4545 "block size (%d)", clustersize, blocksize); 4546 goto failed_mount; 4547 } 4548 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4549 le32_to_cpu(es->s_log_block_size); 4550 sbi->s_clusters_per_group = 4551 le32_to_cpu(es->s_clusters_per_group); 4552 if (sbi->s_clusters_per_group > blocksize * 8) { 4553 ext4_msg(sb, KERN_ERR, 4554 "#clusters per group too big: %lu", 4555 sbi->s_clusters_per_group); 4556 goto failed_mount; 4557 } 4558 if (sbi->s_blocks_per_group != 4559 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4560 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4561 "clusters per group (%lu) inconsistent", 4562 sbi->s_blocks_per_group, 4563 sbi->s_clusters_per_group); 4564 goto failed_mount; 4565 } 4566 } else { 4567 if (clustersize != blocksize) { 4568 ext4_msg(sb, KERN_ERR, 4569 "fragment/cluster size (%d) != " 4570 "block size (%d)", clustersize, blocksize); 4571 goto failed_mount; 4572 } 4573 if (sbi->s_blocks_per_group > blocksize * 8) { 4574 ext4_msg(sb, KERN_ERR, 4575 "#blocks per group too big: %lu", 4576 sbi->s_blocks_per_group); 4577 goto failed_mount; 4578 } 4579 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4580 sbi->s_cluster_bits = 0; 4581 } 4582 sbi->s_cluster_ratio = clustersize / blocksize; 4583 4584 /* Do we have standard group size of clustersize * 8 blocks ? */ 4585 if (sbi->s_blocks_per_group == clustersize << 3) 4586 set_opt2(sb, STD_GROUP_SIZE); 4587 4588 /* 4589 * Test whether we have more sectors than will fit in sector_t, 4590 * and whether the max offset is addressable by the page cache. 4591 */ 4592 err = generic_check_addressable(sb->s_blocksize_bits, 4593 ext4_blocks_count(es)); 4594 if (err) { 4595 ext4_msg(sb, KERN_ERR, "filesystem" 4596 " too large to mount safely on this system"); 4597 goto failed_mount; 4598 } 4599 4600 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4601 goto cantfind_ext4; 4602 4603 /* check blocks count against device size */ 4604 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4605 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4606 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4607 "exceeds size of device (%llu blocks)", 4608 ext4_blocks_count(es), blocks_count); 4609 goto failed_mount; 4610 } 4611 4612 /* 4613 * It makes no sense for the first data block to be beyond the end 4614 * of the filesystem. 4615 */ 4616 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4617 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4618 "block %u is beyond end of filesystem (%llu)", 4619 le32_to_cpu(es->s_first_data_block), 4620 ext4_blocks_count(es)); 4621 goto failed_mount; 4622 } 4623 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4624 (sbi->s_cluster_ratio == 1)) { 4625 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4626 "block is 0 with a 1k block and cluster size"); 4627 goto failed_mount; 4628 } 4629 4630 blocks_count = (ext4_blocks_count(es) - 4631 le32_to_cpu(es->s_first_data_block) + 4632 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4633 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4634 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4635 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4636 "(block count %llu, first data block %u, " 4637 "blocks per group %lu)", blocks_count, 4638 ext4_blocks_count(es), 4639 le32_to_cpu(es->s_first_data_block), 4640 EXT4_BLOCKS_PER_GROUP(sb)); 4641 goto failed_mount; 4642 } 4643 sbi->s_groups_count = blocks_count; 4644 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4645 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4646 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4647 le32_to_cpu(es->s_inodes_count)) { 4648 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4649 le32_to_cpu(es->s_inodes_count), 4650 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4651 ret = -EINVAL; 4652 goto failed_mount; 4653 } 4654 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4655 EXT4_DESC_PER_BLOCK(sb); 4656 if (ext4_has_feature_meta_bg(sb)) { 4657 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4658 ext4_msg(sb, KERN_WARNING, 4659 "first meta block group too large: %u " 4660 "(group descriptor block count %u)", 4661 le32_to_cpu(es->s_first_meta_bg), db_count); 4662 goto failed_mount; 4663 } 4664 } 4665 rcu_assign_pointer(sbi->s_group_desc, 4666 kvmalloc_array(db_count, 4667 sizeof(struct buffer_head *), 4668 GFP_KERNEL)); 4669 if (sbi->s_group_desc == NULL) { 4670 ext4_msg(sb, KERN_ERR, "not enough memory"); 4671 ret = -ENOMEM; 4672 goto failed_mount; 4673 } 4674 4675 bgl_lock_init(sbi->s_blockgroup_lock); 4676 4677 /* Pre-read the descriptors into the buffer cache */ 4678 for (i = 0; i < db_count; i++) { 4679 block = descriptor_loc(sb, logical_sb_block, i); 4680 ext4_sb_breadahead_unmovable(sb, block); 4681 } 4682 4683 for (i = 0; i < db_count; i++) { 4684 struct buffer_head *bh; 4685 4686 block = descriptor_loc(sb, logical_sb_block, i); 4687 bh = ext4_sb_bread_unmovable(sb, block); 4688 if (IS_ERR(bh)) { 4689 ext4_msg(sb, KERN_ERR, 4690 "can't read group descriptor %d", i); 4691 db_count = i; 4692 ret = PTR_ERR(bh); 4693 goto failed_mount2; 4694 } 4695 rcu_read_lock(); 4696 rcu_dereference(sbi->s_group_desc)[i] = bh; 4697 rcu_read_unlock(); 4698 } 4699 sbi->s_gdb_count = db_count; 4700 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4701 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4702 ret = -EFSCORRUPTED; 4703 goto failed_mount2; 4704 } 4705 4706 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4707 spin_lock_init(&sbi->s_error_lock); 4708 INIT_WORK(&sbi->s_error_work, flush_stashed_error_work); 4709 4710 /* Register extent status tree shrinker */ 4711 if (ext4_es_register_shrinker(sbi)) 4712 goto failed_mount3; 4713 4714 sbi->s_stripe = ext4_get_stripe_size(sbi); 4715 sbi->s_extent_max_zeroout_kb = 32; 4716 4717 /* 4718 * set up enough so that it can read an inode 4719 */ 4720 sb->s_op = &ext4_sops; 4721 sb->s_export_op = &ext4_export_ops; 4722 sb->s_xattr = ext4_xattr_handlers; 4723 #ifdef CONFIG_FS_ENCRYPTION 4724 sb->s_cop = &ext4_cryptops; 4725 #endif 4726 #ifdef CONFIG_FS_VERITY 4727 sb->s_vop = &ext4_verityops; 4728 #endif 4729 #ifdef CONFIG_QUOTA 4730 sb->dq_op = &ext4_quota_operations; 4731 if (ext4_has_feature_quota(sb)) 4732 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4733 else 4734 sb->s_qcop = &ext4_qctl_operations; 4735 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4736 #endif 4737 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4738 4739 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4740 mutex_init(&sbi->s_orphan_lock); 4741 4742 /* Initialize fast commit stuff */ 4743 atomic_set(&sbi->s_fc_subtid, 0); 4744 atomic_set(&sbi->s_fc_ineligible_updates, 0); 4745 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4746 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4747 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4748 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4749 sbi->s_fc_bytes = 0; 4750 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4751 ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING); 4752 spin_lock_init(&sbi->s_fc_lock); 4753 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4754 sbi->s_fc_replay_state.fc_regions = NULL; 4755 sbi->s_fc_replay_state.fc_regions_size = 0; 4756 sbi->s_fc_replay_state.fc_regions_used = 0; 4757 sbi->s_fc_replay_state.fc_regions_valid = 0; 4758 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4759 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4760 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4761 4762 sb->s_root = NULL; 4763 4764 needs_recovery = (es->s_last_orphan != 0 || 4765 ext4_has_feature_journal_needs_recovery(sb)); 4766 4767 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4768 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4769 goto failed_mount3a; 4770 4771 /* 4772 * The first inode we look at is the journal inode. Don't try 4773 * root first: it may be modified in the journal! 4774 */ 4775 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4776 err = ext4_load_journal(sb, es, journal_devnum); 4777 if (err) 4778 goto failed_mount3a; 4779 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4780 ext4_has_feature_journal_needs_recovery(sb)) { 4781 ext4_msg(sb, KERN_ERR, "required journal recovery " 4782 "suppressed and not mounted read-only"); 4783 goto failed_mount_wq; 4784 } else { 4785 /* Nojournal mode, all journal mount options are illegal */ 4786 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4787 ext4_msg(sb, KERN_ERR, "can't mount with " 4788 "journal_checksum, fs mounted w/o journal"); 4789 goto failed_mount_wq; 4790 } 4791 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4792 ext4_msg(sb, KERN_ERR, "can't mount with " 4793 "journal_async_commit, fs mounted w/o journal"); 4794 goto failed_mount_wq; 4795 } 4796 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4797 ext4_msg(sb, KERN_ERR, "can't mount with " 4798 "commit=%lu, fs mounted w/o journal", 4799 sbi->s_commit_interval / HZ); 4800 goto failed_mount_wq; 4801 } 4802 if (EXT4_MOUNT_DATA_FLAGS & 4803 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4804 ext4_msg(sb, KERN_ERR, "can't mount with " 4805 "data=, fs mounted w/o journal"); 4806 goto failed_mount_wq; 4807 } 4808 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4809 clear_opt(sb, JOURNAL_CHECKSUM); 4810 clear_opt(sb, DATA_FLAGS); 4811 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4812 sbi->s_journal = NULL; 4813 needs_recovery = 0; 4814 goto no_journal; 4815 } 4816 4817 if (ext4_has_feature_64bit(sb) && 4818 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4819 JBD2_FEATURE_INCOMPAT_64BIT)) { 4820 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4821 goto failed_mount_wq; 4822 } 4823 4824 if (!set_journal_csum_feature_set(sb)) { 4825 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4826 "feature set"); 4827 goto failed_mount_wq; 4828 } 4829 4830 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4831 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4832 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4833 ext4_msg(sb, KERN_ERR, 4834 "Failed to set fast commit journal feature"); 4835 goto failed_mount_wq; 4836 } 4837 4838 /* We have now updated the journal if required, so we can 4839 * validate the data journaling mode. */ 4840 switch (test_opt(sb, DATA_FLAGS)) { 4841 case 0: 4842 /* No mode set, assume a default based on the journal 4843 * capabilities: ORDERED_DATA if the journal can 4844 * cope, else JOURNAL_DATA 4845 */ 4846 if (jbd2_journal_check_available_features 4847 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4848 set_opt(sb, ORDERED_DATA); 4849 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4850 } else { 4851 set_opt(sb, JOURNAL_DATA); 4852 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4853 } 4854 break; 4855 4856 case EXT4_MOUNT_ORDERED_DATA: 4857 case EXT4_MOUNT_WRITEBACK_DATA: 4858 if (!jbd2_journal_check_available_features 4859 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4860 ext4_msg(sb, KERN_ERR, "Journal does not support " 4861 "requested data journaling mode"); 4862 goto failed_mount_wq; 4863 } 4864 break; 4865 default: 4866 break; 4867 } 4868 4869 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4870 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4871 ext4_msg(sb, KERN_ERR, "can't mount with " 4872 "journal_async_commit in data=ordered mode"); 4873 goto failed_mount_wq; 4874 } 4875 4876 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 4877 4878 sbi->s_journal->j_submit_inode_data_buffers = 4879 ext4_journal_submit_inode_data_buffers; 4880 sbi->s_journal->j_finish_inode_data_buffers = 4881 ext4_journal_finish_inode_data_buffers; 4882 4883 no_journal: 4884 if (!test_opt(sb, NO_MBCACHE)) { 4885 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4886 if (!sbi->s_ea_block_cache) { 4887 ext4_msg(sb, KERN_ERR, 4888 "Failed to create ea_block_cache"); 4889 goto failed_mount_wq; 4890 } 4891 4892 if (ext4_has_feature_ea_inode(sb)) { 4893 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4894 if (!sbi->s_ea_inode_cache) { 4895 ext4_msg(sb, KERN_ERR, 4896 "Failed to create ea_inode_cache"); 4897 goto failed_mount_wq; 4898 } 4899 } 4900 } 4901 4902 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4903 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4904 goto failed_mount_wq; 4905 } 4906 4907 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4908 !ext4_has_feature_encrypt(sb)) { 4909 ext4_set_feature_encrypt(sb); 4910 ext4_commit_super(sb); 4911 } 4912 4913 /* 4914 * Get the # of file system overhead blocks from the 4915 * superblock if present. 4916 */ 4917 if (es->s_overhead_clusters) 4918 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4919 else { 4920 err = ext4_calculate_overhead(sb); 4921 if (err) 4922 goto failed_mount_wq; 4923 } 4924 4925 /* 4926 * The maximum number of concurrent works can be high and 4927 * concurrency isn't really necessary. Limit it to 1. 4928 */ 4929 EXT4_SB(sb)->rsv_conversion_wq = 4930 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4931 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4932 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4933 ret = -ENOMEM; 4934 goto failed_mount4; 4935 } 4936 4937 /* 4938 * The jbd2_journal_load will have done any necessary log recovery, 4939 * so we can safely mount the rest of the filesystem now. 4940 */ 4941 4942 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4943 if (IS_ERR(root)) { 4944 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4945 ret = PTR_ERR(root); 4946 root = NULL; 4947 goto failed_mount4; 4948 } 4949 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4950 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4951 iput(root); 4952 goto failed_mount4; 4953 } 4954 4955 sb->s_root = d_make_root(root); 4956 if (!sb->s_root) { 4957 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4958 ret = -ENOMEM; 4959 goto failed_mount4; 4960 } 4961 4962 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4963 if (ret == -EROFS) { 4964 sb->s_flags |= SB_RDONLY; 4965 ret = 0; 4966 } else if (ret) 4967 goto failed_mount4a; 4968 4969 ext4_set_resv_clusters(sb); 4970 4971 if (test_opt(sb, BLOCK_VALIDITY)) { 4972 err = ext4_setup_system_zone(sb); 4973 if (err) { 4974 ext4_msg(sb, KERN_ERR, "failed to initialize system " 4975 "zone (%d)", err); 4976 goto failed_mount4a; 4977 } 4978 } 4979 ext4_fc_replay_cleanup(sb); 4980 4981 ext4_ext_init(sb); 4982 err = ext4_mb_init(sb); 4983 if (err) { 4984 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 4985 err); 4986 goto failed_mount5; 4987 } 4988 4989 /* 4990 * We can only set up the journal commit callback once 4991 * mballoc is initialized 4992 */ 4993 if (sbi->s_journal) 4994 sbi->s_journal->j_commit_callback = 4995 ext4_journal_commit_callback; 4996 4997 block = ext4_count_free_clusters(sb); 4998 ext4_free_blocks_count_set(sbi->s_es, 4999 EXT4_C2B(sbi, block)); 5000 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5001 GFP_KERNEL); 5002 if (!err) { 5003 unsigned long freei = ext4_count_free_inodes(sb); 5004 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5005 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5006 GFP_KERNEL); 5007 } 5008 if (!err) 5009 err = percpu_counter_init(&sbi->s_dirs_counter, 5010 ext4_count_dirs(sb), GFP_KERNEL); 5011 if (!err) 5012 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5013 GFP_KERNEL); 5014 if (!err) 5015 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5016 5017 if (err) { 5018 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5019 goto failed_mount6; 5020 } 5021 5022 if (ext4_has_feature_flex_bg(sb)) 5023 if (!ext4_fill_flex_info(sb)) { 5024 ext4_msg(sb, KERN_ERR, 5025 "unable to initialize " 5026 "flex_bg meta info!"); 5027 goto failed_mount6; 5028 } 5029 5030 err = ext4_register_li_request(sb, first_not_zeroed); 5031 if (err) 5032 goto failed_mount6; 5033 5034 err = ext4_register_sysfs(sb); 5035 if (err) 5036 goto failed_mount7; 5037 5038 #ifdef CONFIG_QUOTA 5039 /* Enable quota usage during mount. */ 5040 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5041 err = ext4_enable_quotas(sb); 5042 if (err) 5043 goto failed_mount8; 5044 } 5045 #endif /* CONFIG_QUOTA */ 5046 5047 /* 5048 * Save the original bdev mapping's wb_err value which could be 5049 * used to detect the metadata async write error. 5050 */ 5051 spin_lock_init(&sbi->s_bdev_wb_lock); 5052 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5053 &sbi->s_bdev_wb_err); 5054 sb->s_bdev->bd_super = sb; 5055 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5056 ext4_orphan_cleanup(sb, es); 5057 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5058 if (needs_recovery) { 5059 ext4_msg(sb, KERN_INFO, "recovery complete"); 5060 err = ext4_mark_recovery_complete(sb, es); 5061 if (err) 5062 goto failed_mount8; 5063 } 5064 if (EXT4_SB(sb)->s_journal) { 5065 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5066 descr = " journalled data mode"; 5067 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5068 descr = " ordered data mode"; 5069 else 5070 descr = " writeback data mode"; 5071 } else 5072 descr = "out journal"; 5073 5074 if (test_opt(sb, DISCARD)) { 5075 struct request_queue *q = bdev_get_queue(sb->s_bdev); 5076 if (!blk_queue_discard(q)) 5077 ext4_msg(sb, KERN_WARNING, 5078 "mounting with \"discard\" option, but " 5079 "the device does not support discard"); 5080 } 5081 5082 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5083 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 5084 "Opts: %.*s%s%s. Quota mode: %s.", descr, 5085 (int) sizeof(sbi->s_es->s_mount_opts), 5086 sbi->s_es->s_mount_opts, 5087 *sbi->s_es->s_mount_opts ? "; " : "", orig_data, 5088 ext4_quota_mode(sb)); 5089 5090 if (es->s_error_count) 5091 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5092 5093 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5094 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5095 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5096 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5097 atomic_set(&sbi->s_warning_count, 0); 5098 atomic_set(&sbi->s_msg_count, 0); 5099 5100 kfree(orig_data); 5101 return 0; 5102 5103 cantfind_ext4: 5104 if (!silent) 5105 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5106 goto failed_mount; 5107 5108 failed_mount8: 5109 ext4_unregister_sysfs(sb); 5110 kobject_put(&sbi->s_kobj); 5111 failed_mount7: 5112 ext4_unregister_li_request(sb); 5113 failed_mount6: 5114 ext4_mb_release(sb); 5115 rcu_read_lock(); 5116 flex_groups = rcu_dereference(sbi->s_flex_groups); 5117 if (flex_groups) { 5118 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 5119 kvfree(flex_groups[i]); 5120 kvfree(flex_groups); 5121 } 5122 rcu_read_unlock(); 5123 percpu_counter_destroy(&sbi->s_freeclusters_counter); 5124 percpu_counter_destroy(&sbi->s_freeinodes_counter); 5125 percpu_counter_destroy(&sbi->s_dirs_counter); 5126 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 5127 percpu_free_rwsem(&sbi->s_writepages_rwsem); 5128 failed_mount5: 5129 ext4_ext_release(sb); 5130 ext4_release_system_zone(sb); 5131 failed_mount4a: 5132 dput(sb->s_root); 5133 sb->s_root = NULL; 5134 failed_mount4: 5135 ext4_msg(sb, KERN_ERR, "mount failed"); 5136 if (EXT4_SB(sb)->rsv_conversion_wq) 5137 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5138 failed_mount_wq: 5139 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5140 sbi->s_ea_inode_cache = NULL; 5141 5142 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5143 sbi->s_ea_block_cache = NULL; 5144 5145 if (sbi->s_journal) { 5146 jbd2_journal_destroy(sbi->s_journal); 5147 sbi->s_journal = NULL; 5148 } 5149 failed_mount3a: 5150 ext4_es_unregister_shrinker(sbi); 5151 failed_mount3: 5152 del_timer_sync(&sbi->s_err_report); 5153 flush_work(&sbi->s_error_work); 5154 if (sbi->s_mmp_tsk) 5155 kthread_stop(sbi->s_mmp_tsk); 5156 failed_mount2: 5157 rcu_read_lock(); 5158 group_desc = rcu_dereference(sbi->s_group_desc); 5159 for (i = 0; i < db_count; i++) 5160 brelse(group_desc[i]); 5161 kvfree(group_desc); 5162 rcu_read_unlock(); 5163 failed_mount: 5164 if (sbi->s_chksum_driver) 5165 crypto_free_shash(sbi->s_chksum_driver); 5166 5167 #ifdef CONFIG_UNICODE 5168 utf8_unload(sb->s_encoding); 5169 #endif 5170 5171 #ifdef CONFIG_QUOTA 5172 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5173 kfree(get_qf_name(sb, sbi, i)); 5174 #endif 5175 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5176 ext4_blkdev_remove(sbi); 5177 brelse(bh); 5178 out_fail: 5179 sb->s_fs_info = NULL; 5180 kfree(sbi->s_blockgroup_lock); 5181 out_free_base: 5182 kfree(sbi); 5183 kfree(orig_data); 5184 fs_put_dax(dax_dev); 5185 return err ? err : ret; 5186 } 5187 5188 /* 5189 * Setup any per-fs journal parameters now. We'll do this both on 5190 * initial mount, once the journal has been initialised but before we've 5191 * done any recovery; and again on any subsequent remount. 5192 */ 5193 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5194 { 5195 struct ext4_sb_info *sbi = EXT4_SB(sb); 5196 5197 journal->j_commit_interval = sbi->s_commit_interval; 5198 journal->j_min_batch_time = sbi->s_min_batch_time; 5199 journal->j_max_batch_time = sbi->s_max_batch_time; 5200 ext4_fc_init(sb, journal); 5201 5202 write_lock(&journal->j_state_lock); 5203 if (test_opt(sb, BARRIER)) 5204 journal->j_flags |= JBD2_BARRIER; 5205 else 5206 journal->j_flags &= ~JBD2_BARRIER; 5207 if (test_opt(sb, DATA_ERR_ABORT)) 5208 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5209 else 5210 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5211 write_unlock(&journal->j_state_lock); 5212 } 5213 5214 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5215 unsigned int journal_inum) 5216 { 5217 struct inode *journal_inode; 5218 5219 /* 5220 * Test for the existence of a valid inode on disk. Bad things 5221 * happen if we iget() an unused inode, as the subsequent iput() 5222 * will try to delete it. 5223 */ 5224 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5225 if (IS_ERR(journal_inode)) { 5226 ext4_msg(sb, KERN_ERR, "no journal found"); 5227 return NULL; 5228 } 5229 if (!journal_inode->i_nlink) { 5230 make_bad_inode(journal_inode); 5231 iput(journal_inode); 5232 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5233 return NULL; 5234 } 5235 5236 jbd_debug(2, "Journal inode found at %p: %lld bytes\n", 5237 journal_inode, journal_inode->i_size); 5238 if (!S_ISREG(journal_inode->i_mode)) { 5239 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5240 iput(journal_inode); 5241 return NULL; 5242 } 5243 return journal_inode; 5244 } 5245 5246 static journal_t *ext4_get_journal(struct super_block *sb, 5247 unsigned int journal_inum) 5248 { 5249 struct inode *journal_inode; 5250 journal_t *journal; 5251 5252 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5253 return NULL; 5254 5255 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5256 if (!journal_inode) 5257 return NULL; 5258 5259 journal = jbd2_journal_init_inode(journal_inode); 5260 if (!journal) { 5261 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5262 iput(journal_inode); 5263 return NULL; 5264 } 5265 journal->j_private = sb; 5266 ext4_init_journal_params(sb, journal); 5267 return journal; 5268 } 5269 5270 static journal_t *ext4_get_dev_journal(struct super_block *sb, 5271 dev_t j_dev) 5272 { 5273 struct buffer_head *bh; 5274 journal_t *journal; 5275 ext4_fsblk_t start; 5276 ext4_fsblk_t len; 5277 int hblock, blocksize; 5278 ext4_fsblk_t sb_block; 5279 unsigned long offset; 5280 struct ext4_super_block *es; 5281 struct block_device *bdev; 5282 5283 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5284 return NULL; 5285 5286 bdev = ext4_blkdev_get(j_dev, sb); 5287 if (bdev == NULL) 5288 return NULL; 5289 5290 blocksize = sb->s_blocksize; 5291 hblock = bdev_logical_block_size(bdev); 5292 if (blocksize < hblock) { 5293 ext4_msg(sb, KERN_ERR, 5294 "blocksize too small for journal device"); 5295 goto out_bdev; 5296 } 5297 5298 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5299 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5300 set_blocksize(bdev, blocksize); 5301 if (!(bh = __bread(bdev, sb_block, blocksize))) { 5302 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5303 "external journal"); 5304 goto out_bdev; 5305 } 5306 5307 es = (struct ext4_super_block *) (bh->b_data + offset); 5308 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5309 !(le32_to_cpu(es->s_feature_incompat) & 5310 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5311 ext4_msg(sb, KERN_ERR, "external journal has " 5312 "bad superblock"); 5313 brelse(bh); 5314 goto out_bdev; 5315 } 5316 5317 if ((le32_to_cpu(es->s_feature_ro_compat) & 5318 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5319 es->s_checksum != ext4_superblock_csum(sb, es)) { 5320 ext4_msg(sb, KERN_ERR, "external journal has " 5321 "corrupt superblock"); 5322 brelse(bh); 5323 goto out_bdev; 5324 } 5325 5326 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5327 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5328 brelse(bh); 5329 goto out_bdev; 5330 } 5331 5332 len = ext4_blocks_count(es); 5333 start = sb_block + 1; 5334 brelse(bh); /* we're done with the superblock */ 5335 5336 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 5337 start, len, blocksize); 5338 if (!journal) { 5339 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5340 goto out_bdev; 5341 } 5342 journal->j_private = sb; 5343 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) { 5344 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 5345 goto out_journal; 5346 } 5347 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5348 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5349 "user (unsupported) - %d", 5350 be32_to_cpu(journal->j_superblock->s_nr_users)); 5351 goto out_journal; 5352 } 5353 EXT4_SB(sb)->s_journal_bdev = bdev; 5354 ext4_init_journal_params(sb, journal); 5355 return journal; 5356 5357 out_journal: 5358 jbd2_journal_destroy(journal); 5359 out_bdev: 5360 ext4_blkdev_put(bdev); 5361 return NULL; 5362 } 5363 5364 static int ext4_load_journal(struct super_block *sb, 5365 struct ext4_super_block *es, 5366 unsigned long journal_devnum) 5367 { 5368 journal_t *journal; 5369 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5370 dev_t journal_dev; 5371 int err = 0; 5372 int really_read_only; 5373 int journal_dev_ro; 5374 5375 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5376 return -EFSCORRUPTED; 5377 5378 if (journal_devnum && 5379 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5380 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5381 "numbers have changed"); 5382 journal_dev = new_decode_dev(journal_devnum); 5383 } else 5384 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5385 5386 if (journal_inum && journal_dev) { 5387 ext4_msg(sb, KERN_ERR, 5388 "filesystem has both journal inode and journal device!"); 5389 return -EINVAL; 5390 } 5391 5392 if (journal_inum) { 5393 journal = ext4_get_journal(sb, journal_inum); 5394 if (!journal) 5395 return -EINVAL; 5396 } else { 5397 journal = ext4_get_dev_journal(sb, journal_dev); 5398 if (!journal) 5399 return -EINVAL; 5400 } 5401 5402 journal_dev_ro = bdev_read_only(journal->j_dev); 5403 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5404 5405 if (journal_dev_ro && !sb_rdonly(sb)) { 5406 ext4_msg(sb, KERN_ERR, 5407 "journal device read-only, try mounting with '-o ro'"); 5408 err = -EROFS; 5409 goto err_out; 5410 } 5411 5412 /* 5413 * Are we loading a blank journal or performing recovery after a 5414 * crash? For recovery, we need to check in advance whether we 5415 * can get read-write access to the device. 5416 */ 5417 if (ext4_has_feature_journal_needs_recovery(sb)) { 5418 if (sb_rdonly(sb)) { 5419 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5420 "required on readonly filesystem"); 5421 if (really_read_only) { 5422 ext4_msg(sb, KERN_ERR, "write access " 5423 "unavailable, cannot proceed " 5424 "(try mounting with noload)"); 5425 err = -EROFS; 5426 goto err_out; 5427 } 5428 ext4_msg(sb, KERN_INFO, "write access will " 5429 "be enabled during recovery"); 5430 } 5431 } 5432 5433 if (!(journal->j_flags & JBD2_BARRIER)) 5434 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5435 5436 if (!ext4_has_feature_journal_needs_recovery(sb)) 5437 err = jbd2_journal_wipe(journal, !really_read_only); 5438 if (!err) { 5439 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5440 if (save) 5441 memcpy(save, ((char *) es) + 5442 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 5443 err = jbd2_journal_load(journal); 5444 if (save) 5445 memcpy(((char *) es) + EXT4_S_ERR_START, 5446 save, EXT4_S_ERR_LEN); 5447 kfree(save); 5448 } 5449 5450 if (err) { 5451 ext4_msg(sb, KERN_ERR, "error loading journal"); 5452 goto err_out; 5453 } 5454 5455 EXT4_SB(sb)->s_journal = journal; 5456 err = ext4_clear_journal_err(sb, es); 5457 if (err) { 5458 EXT4_SB(sb)->s_journal = NULL; 5459 jbd2_journal_destroy(journal); 5460 return err; 5461 } 5462 5463 if (!really_read_only && journal_devnum && 5464 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5465 es->s_journal_dev = cpu_to_le32(journal_devnum); 5466 5467 /* Make sure we flush the recovery flag to disk. */ 5468 ext4_commit_super(sb); 5469 } 5470 5471 return 0; 5472 5473 err_out: 5474 jbd2_journal_destroy(journal); 5475 return err; 5476 } 5477 5478 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 5479 static void ext4_update_super(struct super_block *sb) 5480 { 5481 struct ext4_sb_info *sbi = EXT4_SB(sb); 5482 struct ext4_super_block *es = sbi->s_es; 5483 struct buffer_head *sbh = sbi->s_sbh; 5484 5485 lock_buffer(sbh); 5486 /* 5487 * If the file system is mounted read-only, don't update the 5488 * superblock write time. This avoids updating the superblock 5489 * write time when we are mounting the root file system 5490 * read/only but we need to replay the journal; at that point, 5491 * for people who are east of GMT and who make their clock 5492 * tick in localtime for Windows bug-for-bug compatibility, 5493 * the clock is set in the future, and this will cause e2fsck 5494 * to complain and force a full file system check. 5495 */ 5496 if (!(sb->s_flags & SB_RDONLY)) 5497 ext4_update_tstamp(es, s_wtime); 5498 es->s_kbytes_written = 5499 cpu_to_le64(sbi->s_kbytes_written + 5500 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 5501 sbi->s_sectors_written_start) >> 1)); 5502 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 5503 ext4_free_blocks_count_set(es, 5504 EXT4_C2B(sbi, percpu_counter_sum_positive( 5505 &sbi->s_freeclusters_counter))); 5506 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 5507 es->s_free_inodes_count = 5508 cpu_to_le32(percpu_counter_sum_positive( 5509 &sbi->s_freeinodes_counter)); 5510 /* Copy error information to the on-disk superblock */ 5511 spin_lock(&sbi->s_error_lock); 5512 if (sbi->s_add_error_count > 0) { 5513 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5514 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 5515 __ext4_update_tstamp(&es->s_first_error_time, 5516 &es->s_first_error_time_hi, 5517 sbi->s_first_error_time); 5518 strncpy(es->s_first_error_func, sbi->s_first_error_func, 5519 sizeof(es->s_first_error_func)); 5520 es->s_first_error_line = 5521 cpu_to_le32(sbi->s_first_error_line); 5522 es->s_first_error_ino = 5523 cpu_to_le32(sbi->s_first_error_ino); 5524 es->s_first_error_block = 5525 cpu_to_le64(sbi->s_first_error_block); 5526 es->s_first_error_errcode = 5527 ext4_errno_to_code(sbi->s_first_error_code); 5528 } 5529 __ext4_update_tstamp(&es->s_last_error_time, 5530 &es->s_last_error_time_hi, 5531 sbi->s_last_error_time); 5532 strncpy(es->s_last_error_func, sbi->s_last_error_func, 5533 sizeof(es->s_last_error_func)); 5534 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 5535 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 5536 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 5537 es->s_last_error_errcode = 5538 ext4_errno_to_code(sbi->s_last_error_code); 5539 /* 5540 * Start the daily error reporting function if it hasn't been 5541 * started already 5542 */ 5543 if (!es->s_error_count) 5544 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 5545 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 5546 sbi->s_add_error_count = 0; 5547 } 5548 spin_unlock(&sbi->s_error_lock); 5549 5550 ext4_superblock_csum_set(sb); 5551 unlock_buffer(sbh); 5552 } 5553 5554 static int ext4_commit_super(struct super_block *sb) 5555 { 5556 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 5557 int error = 0; 5558 5559 if (!sbh || block_device_ejected(sb)) 5560 return error; 5561 5562 ext4_update_super(sb); 5563 5564 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 5565 /* 5566 * Oh, dear. A previous attempt to write the 5567 * superblock failed. This could happen because the 5568 * USB device was yanked out. Or it could happen to 5569 * be a transient write error and maybe the block will 5570 * be remapped. Nothing we can do but to retry the 5571 * write and hope for the best. 5572 */ 5573 ext4_msg(sb, KERN_ERR, "previous I/O error to " 5574 "superblock detected"); 5575 clear_buffer_write_io_error(sbh); 5576 set_buffer_uptodate(sbh); 5577 } 5578 BUFFER_TRACE(sbh, "marking dirty"); 5579 mark_buffer_dirty(sbh); 5580 error = __sync_dirty_buffer(sbh, 5581 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0)); 5582 if (buffer_write_io_error(sbh)) { 5583 ext4_msg(sb, KERN_ERR, "I/O error while writing " 5584 "superblock"); 5585 clear_buffer_write_io_error(sbh); 5586 set_buffer_uptodate(sbh); 5587 } 5588 return error; 5589 } 5590 5591 /* 5592 * Have we just finished recovery? If so, and if we are mounting (or 5593 * remounting) the filesystem readonly, then we will end up with a 5594 * consistent fs on disk. Record that fact. 5595 */ 5596 static int ext4_mark_recovery_complete(struct super_block *sb, 5597 struct ext4_super_block *es) 5598 { 5599 int err; 5600 journal_t *journal = EXT4_SB(sb)->s_journal; 5601 5602 if (!ext4_has_feature_journal(sb)) { 5603 if (journal != NULL) { 5604 ext4_error(sb, "Journal got removed while the fs was " 5605 "mounted!"); 5606 return -EFSCORRUPTED; 5607 } 5608 return 0; 5609 } 5610 jbd2_journal_lock_updates(journal); 5611 err = jbd2_journal_flush(journal); 5612 if (err < 0) 5613 goto out; 5614 5615 if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) { 5616 ext4_clear_feature_journal_needs_recovery(sb); 5617 ext4_commit_super(sb); 5618 } 5619 out: 5620 jbd2_journal_unlock_updates(journal); 5621 return err; 5622 } 5623 5624 /* 5625 * If we are mounting (or read-write remounting) a filesystem whose journal 5626 * has recorded an error from a previous lifetime, move that error to the 5627 * main filesystem now. 5628 */ 5629 static int ext4_clear_journal_err(struct super_block *sb, 5630 struct ext4_super_block *es) 5631 { 5632 journal_t *journal; 5633 int j_errno; 5634 const char *errstr; 5635 5636 if (!ext4_has_feature_journal(sb)) { 5637 ext4_error(sb, "Journal got removed while the fs was mounted!"); 5638 return -EFSCORRUPTED; 5639 } 5640 5641 journal = EXT4_SB(sb)->s_journal; 5642 5643 /* 5644 * Now check for any error status which may have been recorded in the 5645 * journal by a prior ext4_error() or ext4_abort() 5646 */ 5647 5648 j_errno = jbd2_journal_errno(journal); 5649 if (j_errno) { 5650 char nbuf[16]; 5651 5652 errstr = ext4_decode_error(sb, j_errno, nbuf); 5653 ext4_warning(sb, "Filesystem error recorded " 5654 "from previous mount: %s", errstr); 5655 ext4_warning(sb, "Marking fs in need of filesystem check."); 5656 5657 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 5658 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5659 ext4_commit_super(sb); 5660 5661 jbd2_journal_clear_err(journal); 5662 jbd2_journal_update_sb_errno(journal); 5663 } 5664 return 0; 5665 } 5666 5667 /* 5668 * Force the running and committing transactions to commit, 5669 * and wait on the commit. 5670 */ 5671 int ext4_force_commit(struct super_block *sb) 5672 { 5673 journal_t *journal; 5674 5675 if (sb_rdonly(sb)) 5676 return 0; 5677 5678 journal = EXT4_SB(sb)->s_journal; 5679 return ext4_journal_force_commit(journal); 5680 } 5681 5682 static int ext4_sync_fs(struct super_block *sb, int wait) 5683 { 5684 int ret = 0; 5685 tid_t target; 5686 bool needs_barrier = false; 5687 struct ext4_sb_info *sbi = EXT4_SB(sb); 5688 5689 if (unlikely(ext4_forced_shutdown(sbi))) 5690 return 0; 5691 5692 trace_ext4_sync_fs(sb, wait); 5693 flush_workqueue(sbi->rsv_conversion_wq); 5694 /* 5695 * Writeback quota in non-journalled quota case - journalled quota has 5696 * no dirty dquots 5697 */ 5698 dquot_writeback_dquots(sb, -1); 5699 /* 5700 * Data writeback is possible w/o journal transaction, so barrier must 5701 * being sent at the end of the function. But we can skip it if 5702 * transaction_commit will do it for us. 5703 */ 5704 if (sbi->s_journal) { 5705 target = jbd2_get_latest_transaction(sbi->s_journal); 5706 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 5707 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 5708 needs_barrier = true; 5709 5710 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 5711 if (wait) 5712 ret = jbd2_log_wait_commit(sbi->s_journal, 5713 target); 5714 } 5715 } else if (wait && test_opt(sb, BARRIER)) 5716 needs_barrier = true; 5717 if (needs_barrier) { 5718 int err; 5719 err = blkdev_issue_flush(sb->s_bdev); 5720 if (!ret) 5721 ret = err; 5722 } 5723 5724 return ret; 5725 } 5726 5727 /* 5728 * LVM calls this function before a (read-only) snapshot is created. This 5729 * gives us a chance to flush the journal completely and mark the fs clean. 5730 * 5731 * Note that only this function cannot bring a filesystem to be in a clean 5732 * state independently. It relies on upper layer to stop all data & metadata 5733 * modifications. 5734 */ 5735 static int ext4_freeze(struct super_block *sb) 5736 { 5737 int error = 0; 5738 journal_t *journal; 5739 5740 if (sb_rdonly(sb)) 5741 return 0; 5742 5743 journal = EXT4_SB(sb)->s_journal; 5744 5745 if (journal) { 5746 /* Now we set up the journal barrier. */ 5747 jbd2_journal_lock_updates(journal); 5748 5749 /* 5750 * Don't clear the needs_recovery flag if we failed to 5751 * flush the journal. 5752 */ 5753 error = jbd2_journal_flush(journal); 5754 if (error < 0) 5755 goto out; 5756 5757 /* Journal blocked and flushed, clear needs_recovery flag. */ 5758 ext4_clear_feature_journal_needs_recovery(sb); 5759 } 5760 5761 error = ext4_commit_super(sb); 5762 out: 5763 if (journal) 5764 /* we rely on upper layer to stop further updates */ 5765 jbd2_journal_unlock_updates(journal); 5766 return error; 5767 } 5768 5769 /* 5770 * Called by LVM after the snapshot is done. We need to reset the RECOVER 5771 * flag here, even though the filesystem is not technically dirty yet. 5772 */ 5773 static int ext4_unfreeze(struct super_block *sb) 5774 { 5775 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 5776 return 0; 5777 5778 if (EXT4_SB(sb)->s_journal) { 5779 /* Reset the needs_recovery flag before the fs is unlocked. */ 5780 ext4_set_feature_journal_needs_recovery(sb); 5781 } 5782 5783 ext4_commit_super(sb); 5784 return 0; 5785 } 5786 5787 /* 5788 * Structure to save mount options for ext4_remount's benefit 5789 */ 5790 struct ext4_mount_options { 5791 unsigned long s_mount_opt; 5792 unsigned long s_mount_opt2; 5793 kuid_t s_resuid; 5794 kgid_t s_resgid; 5795 unsigned long s_commit_interval; 5796 u32 s_min_batch_time, s_max_batch_time; 5797 #ifdef CONFIG_QUOTA 5798 int s_jquota_fmt; 5799 char *s_qf_names[EXT4_MAXQUOTAS]; 5800 #endif 5801 }; 5802 5803 static int ext4_remount(struct super_block *sb, int *flags, char *data) 5804 { 5805 struct ext4_super_block *es; 5806 struct ext4_sb_info *sbi = EXT4_SB(sb); 5807 unsigned long old_sb_flags, vfs_flags; 5808 struct ext4_mount_options old_opts; 5809 int enable_quota = 0; 5810 ext4_group_t g; 5811 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5812 int err = 0; 5813 #ifdef CONFIG_QUOTA 5814 int i, j; 5815 char *to_free[EXT4_MAXQUOTAS]; 5816 #endif 5817 char *orig_data = kstrdup(data, GFP_KERNEL); 5818 5819 if (data && !orig_data) 5820 return -ENOMEM; 5821 5822 /* Store the original options */ 5823 old_sb_flags = sb->s_flags; 5824 old_opts.s_mount_opt = sbi->s_mount_opt; 5825 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 5826 old_opts.s_resuid = sbi->s_resuid; 5827 old_opts.s_resgid = sbi->s_resgid; 5828 old_opts.s_commit_interval = sbi->s_commit_interval; 5829 old_opts.s_min_batch_time = sbi->s_min_batch_time; 5830 old_opts.s_max_batch_time = sbi->s_max_batch_time; 5831 #ifdef CONFIG_QUOTA 5832 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 5833 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5834 if (sbi->s_qf_names[i]) { 5835 char *qf_name = get_qf_name(sb, sbi, i); 5836 5837 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 5838 if (!old_opts.s_qf_names[i]) { 5839 for (j = 0; j < i; j++) 5840 kfree(old_opts.s_qf_names[j]); 5841 kfree(orig_data); 5842 return -ENOMEM; 5843 } 5844 } else 5845 old_opts.s_qf_names[i] = NULL; 5846 #endif 5847 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 5848 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio; 5849 5850 /* 5851 * Some options can be enabled by ext4 and/or by VFS mount flag 5852 * either way we need to make sure it matches in both *flags and 5853 * s_flags. Copy those selected flags from *flags to s_flags 5854 */ 5855 vfs_flags = SB_LAZYTIME | SB_I_VERSION; 5856 sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags); 5857 5858 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) { 5859 err = -EINVAL; 5860 goto restore_opts; 5861 } 5862 5863 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 5864 test_opt(sb, JOURNAL_CHECKSUM)) { 5865 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 5866 "during remount not supported; ignoring"); 5867 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 5868 } 5869 5870 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5871 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5872 ext4_msg(sb, KERN_ERR, "can't mount with " 5873 "both data=journal and delalloc"); 5874 err = -EINVAL; 5875 goto restore_opts; 5876 } 5877 if (test_opt(sb, DIOREAD_NOLOCK)) { 5878 ext4_msg(sb, KERN_ERR, "can't mount with " 5879 "both data=journal and dioread_nolock"); 5880 err = -EINVAL; 5881 goto restore_opts; 5882 } 5883 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 5884 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5885 ext4_msg(sb, KERN_ERR, "can't mount with " 5886 "journal_async_commit in data=ordered mode"); 5887 err = -EINVAL; 5888 goto restore_opts; 5889 } 5890 } 5891 5892 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 5893 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 5894 err = -EINVAL; 5895 goto restore_opts; 5896 } 5897 5898 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) 5899 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user"); 5900 5901 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5902 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5903 5904 es = sbi->s_es; 5905 5906 if (sbi->s_journal) { 5907 ext4_init_journal_params(sb, sbi->s_journal); 5908 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 5909 } 5910 5911 /* Flush outstanding errors before changing fs state */ 5912 flush_work(&sbi->s_error_work); 5913 5914 if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) { 5915 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) { 5916 err = -EROFS; 5917 goto restore_opts; 5918 } 5919 5920 if (*flags & SB_RDONLY) { 5921 err = sync_filesystem(sb); 5922 if (err < 0) 5923 goto restore_opts; 5924 err = dquot_suspend(sb, -1); 5925 if (err < 0) 5926 goto restore_opts; 5927 5928 /* 5929 * First of all, the unconditional stuff we have to do 5930 * to disable replay of the journal when we next remount 5931 */ 5932 sb->s_flags |= SB_RDONLY; 5933 5934 /* 5935 * OK, test if we are remounting a valid rw partition 5936 * readonly, and if so set the rdonly flag and then 5937 * mark the partition as valid again. 5938 */ 5939 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 5940 (sbi->s_mount_state & EXT4_VALID_FS)) 5941 es->s_state = cpu_to_le16(sbi->s_mount_state); 5942 5943 if (sbi->s_journal) { 5944 /* 5945 * We let remount-ro finish even if marking fs 5946 * as clean failed... 5947 */ 5948 ext4_mark_recovery_complete(sb, es); 5949 } 5950 if (sbi->s_mmp_tsk) 5951 kthread_stop(sbi->s_mmp_tsk); 5952 } else { 5953 /* Make sure we can mount this feature set readwrite */ 5954 if (ext4_has_feature_readonly(sb) || 5955 !ext4_feature_set_ok(sb, 0)) { 5956 err = -EROFS; 5957 goto restore_opts; 5958 } 5959 /* 5960 * Make sure the group descriptor checksums 5961 * are sane. If they aren't, refuse to remount r/w. 5962 */ 5963 for (g = 0; g < sbi->s_groups_count; g++) { 5964 struct ext4_group_desc *gdp = 5965 ext4_get_group_desc(sb, g, NULL); 5966 5967 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 5968 ext4_msg(sb, KERN_ERR, 5969 "ext4_remount: Checksum for group %u failed (%u!=%u)", 5970 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 5971 le16_to_cpu(gdp->bg_checksum)); 5972 err = -EFSBADCRC; 5973 goto restore_opts; 5974 } 5975 } 5976 5977 /* 5978 * If we have an unprocessed orphan list hanging 5979 * around from a previously readonly bdev mount, 5980 * require a full umount/remount for now. 5981 */ 5982 if (es->s_last_orphan) { 5983 ext4_msg(sb, KERN_WARNING, "Couldn't " 5984 "remount RDWR because of unprocessed " 5985 "orphan inode list. Please " 5986 "umount/remount instead"); 5987 err = -EINVAL; 5988 goto restore_opts; 5989 } 5990 5991 /* 5992 * Mounting a RDONLY partition read-write, so reread 5993 * and store the current valid flag. (It may have 5994 * been changed by e2fsck since we originally mounted 5995 * the partition.) 5996 */ 5997 if (sbi->s_journal) { 5998 err = ext4_clear_journal_err(sb, es); 5999 if (err) 6000 goto restore_opts; 6001 } 6002 sbi->s_mount_state = le16_to_cpu(es->s_state); 6003 6004 err = ext4_setup_super(sb, es, 0); 6005 if (err) 6006 goto restore_opts; 6007 6008 sb->s_flags &= ~SB_RDONLY; 6009 if (ext4_has_feature_mmp(sb)) 6010 if (ext4_multi_mount_protect(sb, 6011 le64_to_cpu(es->s_mmp_block))) { 6012 err = -EROFS; 6013 goto restore_opts; 6014 } 6015 enable_quota = 1; 6016 } 6017 } 6018 6019 /* 6020 * Reinitialize lazy itable initialization thread based on 6021 * current settings 6022 */ 6023 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6024 ext4_unregister_li_request(sb); 6025 else { 6026 ext4_group_t first_not_zeroed; 6027 first_not_zeroed = ext4_has_uninit_itable(sb); 6028 ext4_register_li_request(sb, first_not_zeroed); 6029 } 6030 6031 /* 6032 * Handle creation of system zone data early because it can fail. 6033 * Releasing of existing data is done when we are sure remount will 6034 * succeed. 6035 */ 6036 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6037 err = ext4_setup_system_zone(sb); 6038 if (err) 6039 goto restore_opts; 6040 } 6041 6042 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6043 err = ext4_commit_super(sb); 6044 if (err) 6045 goto restore_opts; 6046 } 6047 6048 #ifdef CONFIG_QUOTA 6049 /* Release old quota file names */ 6050 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6051 kfree(old_opts.s_qf_names[i]); 6052 if (enable_quota) { 6053 if (sb_any_quota_suspended(sb)) 6054 dquot_resume(sb, -1); 6055 else if (ext4_has_feature_quota(sb)) { 6056 err = ext4_enable_quotas(sb); 6057 if (err) 6058 goto restore_opts; 6059 } 6060 } 6061 #endif 6062 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6063 ext4_release_system_zone(sb); 6064 6065 /* 6066 * Some options can be enabled by ext4 and/or by VFS mount flag 6067 * either way we need to make sure it matches in both *flags and 6068 * s_flags. Copy those selected flags from s_flags to *flags 6069 */ 6070 *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags); 6071 6072 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s. Quota mode: %s.", 6073 orig_data, ext4_quota_mode(sb)); 6074 kfree(orig_data); 6075 return 0; 6076 6077 restore_opts: 6078 sb->s_flags = old_sb_flags; 6079 sbi->s_mount_opt = old_opts.s_mount_opt; 6080 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6081 sbi->s_resuid = old_opts.s_resuid; 6082 sbi->s_resgid = old_opts.s_resgid; 6083 sbi->s_commit_interval = old_opts.s_commit_interval; 6084 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6085 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6086 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6087 ext4_release_system_zone(sb); 6088 #ifdef CONFIG_QUOTA 6089 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6090 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6091 to_free[i] = get_qf_name(sb, sbi, i); 6092 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6093 } 6094 synchronize_rcu(); 6095 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6096 kfree(to_free[i]); 6097 #endif 6098 kfree(orig_data); 6099 return err; 6100 } 6101 6102 #ifdef CONFIG_QUOTA 6103 static int ext4_statfs_project(struct super_block *sb, 6104 kprojid_t projid, struct kstatfs *buf) 6105 { 6106 struct kqid qid; 6107 struct dquot *dquot; 6108 u64 limit; 6109 u64 curblock; 6110 6111 qid = make_kqid_projid(projid); 6112 dquot = dqget(sb, qid); 6113 if (IS_ERR(dquot)) 6114 return PTR_ERR(dquot); 6115 spin_lock(&dquot->dq_dqb_lock); 6116 6117 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6118 dquot->dq_dqb.dqb_bhardlimit); 6119 limit >>= sb->s_blocksize_bits; 6120 6121 if (limit && buf->f_blocks > limit) { 6122 curblock = (dquot->dq_dqb.dqb_curspace + 6123 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6124 buf->f_blocks = limit; 6125 buf->f_bfree = buf->f_bavail = 6126 (buf->f_blocks > curblock) ? 6127 (buf->f_blocks - curblock) : 0; 6128 } 6129 6130 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6131 dquot->dq_dqb.dqb_ihardlimit); 6132 if (limit && buf->f_files > limit) { 6133 buf->f_files = limit; 6134 buf->f_ffree = 6135 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6136 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6137 } 6138 6139 spin_unlock(&dquot->dq_dqb_lock); 6140 dqput(dquot); 6141 return 0; 6142 } 6143 #endif 6144 6145 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6146 { 6147 struct super_block *sb = dentry->d_sb; 6148 struct ext4_sb_info *sbi = EXT4_SB(sb); 6149 struct ext4_super_block *es = sbi->s_es; 6150 ext4_fsblk_t overhead = 0, resv_blocks; 6151 u64 fsid; 6152 s64 bfree; 6153 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6154 6155 if (!test_opt(sb, MINIX_DF)) 6156 overhead = sbi->s_overhead; 6157 6158 buf->f_type = EXT4_SUPER_MAGIC; 6159 buf->f_bsize = sb->s_blocksize; 6160 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6161 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6162 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6163 /* prevent underflow in case that few free space is available */ 6164 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6165 buf->f_bavail = buf->f_bfree - 6166 (ext4_r_blocks_count(es) + resv_blocks); 6167 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6168 buf->f_bavail = 0; 6169 buf->f_files = le32_to_cpu(es->s_inodes_count); 6170 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6171 buf->f_namelen = EXT4_NAME_LEN; 6172 fsid = le64_to_cpup((void *)es->s_uuid) ^ 6173 le64_to_cpup((void *)es->s_uuid + sizeof(u64)); 6174 buf->f_fsid = u64_to_fsid(fsid); 6175 6176 #ifdef CONFIG_QUOTA 6177 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6178 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6179 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6180 #endif 6181 return 0; 6182 } 6183 6184 6185 #ifdef CONFIG_QUOTA 6186 6187 /* 6188 * Helper functions so that transaction is started before we acquire dqio_sem 6189 * to keep correct lock ordering of transaction > dqio_sem 6190 */ 6191 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6192 { 6193 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6194 } 6195 6196 static int ext4_write_dquot(struct dquot *dquot) 6197 { 6198 int ret, err; 6199 handle_t *handle; 6200 struct inode *inode; 6201 6202 inode = dquot_to_inode(dquot); 6203 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6204 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6205 if (IS_ERR(handle)) 6206 return PTR_ERR(handle); 6207 ret = dquot_commit(dquot); 6208 err = ext4_journal_stop(handle); 6209 if (!ret) 6210 ret = err; 6211 return ret; 6212 } 6213 6214 static int ext4_acquire_dquot(struct dquot *dquot) 6215 { 6216 int ret, err; 6217 handle_t *handle; 6218 6219 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6220 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6221 if (IS_ERR(handle)) 6222 return PTR_ERR(handle); 6223 ret = dquot_acquire(dquot); 6224 err = ext4_journal_stop(handle); 6225 if (!ret) 6226 ret = err; 6227 return ret; 6228 } 6229 6230 static int ext4_release_dquot(struct dquot *dquot) 6231 { 6232 int ret, err; 6233 handle_t *handle; 6234 6235 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6236 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6237 if (IS_ERR(handle)) { 6238 /* Release dquot anyway to avoid endless cycle in dqput() */ 6239 dquot_release(dquot); 6240 return PTR_ERR(handle); 6241 } 6242 ret = dquot_release(dquot); 6243 err = ext4_journal_stop(handle); 6244 if (!ret) 6245 ret = err; 6246 return ret; 6247 } 6248 6249 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6250 { 6251 struct super_block *sb = dquot->dq_sb; 6252 6253 if (ext4_is_quota_journalled(sb)) { 6254 dquot_mark_dquot_dirty(dquot); 6255 return ext4_write_dquot(dquot); 6256 } else { 6257 return dquot_mark_dquot_dirty(dquot); 6258 } 6259 } 6260 6261 static int ext4_write_info(struct super_block *sb, int type) 6262 { 6263 int ret, err; 6264 handle_t *handle; 6265 6266 /* Data block + inode block */ 6267 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 6268 if (IS_ERR(handle)) 6269 return PTR_ERR(handle); 6270 ret = dquot_commit_info(sb, type); 6271 err = ext4_journal_stop(handle); 6272 if (!ret) 6273 ret = err; 6274 return ret; 6275 } 6276 6277 /* 6278 * Turn on quotas during mount time - we need to find 6279 * the quota file and such... 6280 */ 6281 static int ext4_quota_on_mount(struct super_block *sb, int type) 6282 { 6283 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 6284 EXT4_SB(sb)->s_jquota_fmt, type); 6285 } 6286 6287 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6288 { 6289 struct ext4_inode_info *ei = EXT4_I(inode); 6290 6291 /* The first argument of lockdep_set_subclass has to be 6292 * *exactly* the same as the argument to init_rwsem() --- in 6293 * this case, in init_once() --- or lockdep gets unhappy 6294 * because the name of the lock is set using the 6295 * stringification of the argument to init_rwsem(). 6296 */ 6297 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6298 lockdep_set_subclass(&ei->i_data_sem, subclass); 6299 } 6300 6301 /* 6302 * Standard function to be called on quota_on 6303 */ 6304 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6305 const struct path *path) 6306 { 6307 int err; 6308 6309 if (!test_opt(sb, QUOTA)) 6310 return -EINVAL; 6311 6312 /* Quotafile not on the same filesystem? */ 6313 if (path->dentry->d_sb != sb) 6314 return -EXDEV; 6315 6316 /* Quota already enabled for this file? */ 6317 if (IS_NOQUOTA(d_inode(path->dentry))) 6318 return -EBUSY; 6319 6320 /* Journaling quota? */ 6321 if (EXT4_SB(sb)->s_qf_names[type]) { 6322 /* Quotafile not in fs root? */ 6323 if (path->dentry->d_parent != sb->s_root) 6324 ext4_msg(sb, KERN_WARNING, 6325 "Quota file not on filesystem root. " 6326 "Journaled quota will not work"); 6327 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6328 } else { 6329 /* 6330 * Clear the flag just in case mount options changed since 6331 * last time. 6332 */ 6333 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6334 } 6335 6336 /* 6337 * When we journal data on quota file, we have to flush journal to see 6338 * all updates to the file when we bypass pagecache... 6339 */ 6340 if (EXT4_SB(sb)->s_journal && 6341 ext4_should_journal_data(d_inode(path->dentry))) { 6342 /* 6343 * We don't need to lock updates but journal_flush() could 6344 * otherwise be livelocked... 6345 */ 6346 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6347 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 6348 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6349 if (err) 6350 return err; 6351 } 6352 6353 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6354 err = dquot_quota_on(sb, type, format_id, path); 6355 if (err) { 6356 lockdep_set_quota_inode(path->dentry->d_inode, 6357 I_DATA_SEM_NORMAL); 6358 } else { 6359 struct inode *inode = d_inode(path->dentry); 6360 handle_t *handle; 6361 6362 /* 6363 * Set inode flags to prevent userspace from messing with quota 6364 * files. If this fails, we return success anyway since quotas 6365 * are already enabled and this is not a hard failure. 6366 */ 6367 inode_lock(inode); 6368 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6369 if (IS_ERR(handle)) 6370 goto unlock_inode; 6371 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6372 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6373 S_NOATIME | S_IMMUTABLE); 6374 err = ext4_mark_inode_dirty(handle, inode); 6375 ext4_journal_stop(handle); 6376 unlock_inode: 6377 inode_unlock(inode); 6378 } 6379 return err; 6380 } 6381 6382 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6383 unsigned int flags) 6384 { 6385 int err; 6386 struct inode *qf_inode; 6387 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6388 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6389 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6390 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6391 }; 6392 6393 BUG_ON(!ext4_has_feature_quota(sb)); 6394 6395 if (!qf_inums[type]) 6396 return -EPERM; 6397 6398 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6399 if (IS_ERR(qf_inode)) { 6400 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6401 return PTR_ERR(qf_inode); 6402 } 6403 6404 /* Don't account quota for quota files to avoid recursion */ 6405 qf_inode->i_flags |= S_NOQUOTA; 6406 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6407 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6408 if (err) 6409 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6410 iput(qf_inode); 6411 6412 return err; 6413 } 6414 6415 /* Enable usage tracking for all quota types. */ 6416 static int ext4_enable_quotas(struct super_block *sb) 6417 { 6418 int type, err = 0; 6419 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6420 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6421 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6422 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6423 }; 6424 bool quota_mopt[EXT4_MAXQUOTAS] = { 6425 test_opt(sb, USRQUOTA), 6426 test_opt(sb, GRPQUOTA), 6427 test_opt(sb, PRJQUOTA), 6428 }; 6429 6430 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6431 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6432 if (qf_inums[type]) { 6433 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6434 DQUOT_USAGE_ENABLED | 6435 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6436 if (err) { 6437 ext4_warning(sb, 6438 "Failed to enable quota tracking " 6439 "(type=%d, err=%d). Please run " 6440 "e2fsck to fix.", type, err); 6441 for (type--; type >= 0; type--) 6442 dquot_quota_off(sb, type); 6443 6444 return err; 6445 } 6446 } 6447 } 6448 return 0; 6449 } 6450 6451 static int ext4_quota_off(struct super_block *sb, int type) 6452 { 6453 struct inode *inode = sb_dqopt(sb)->files[type]; 6454 handle_t *handle; 6455 int err; 6456 6457 /* Force all delayed allocation blocks to be allocated. 6458 * Caller already holds s_umount sem */ 6459 if (test_opt(sb, DELALLOC)) 6460 sync_filesystem(sb); 6461 6462 if (!inode || !igrab(inode)) 6463 goto out; 6464 6465 err = dquot_quota_off(sb, type); 6466 if (err || ext4_has_feature_quota(sb)) 6467 goto out_put; 6468 6469 inode_lock(inode); 6470 /* 6471 * Update modification times of quota files when userspace can 6472 * start looking at them. If we fail, we return success anyway since 6473 * this is not a hard failure and quotas are already disabled. 6474 */ 6475 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6476 if (IS_ERR(handle)) { 6477 err = PTR_ERR(handle); 6478 goto out_unlock; 6479 } 6480 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 6481 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 6482 inode->i_mtime = inode->i_ctime = current_time(inode); 6483 err = ext4_mark_inode_dirty(handle, inode); 6484 ext4_journal_stop(handle); 6485 out_unlock: 6486 inode_unlock(inode); 6487 out_put: 6488 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 6489 iput(inode); 6490 return err; 6491 out: 6492 return dquot_quota_off(sb, type); 6493 } 6494 6495 /* Read data from quotafile - avoid pagecache and such because we cannot afford 6496 * acquiring the locks... As quota files are never truncated and quota code 6497 * itself serializes the operations (and no one else should touch the files) 6498 * we don't have to be afraid of races */ 6499 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 6500 size_t len, loff_t off) 6501 { 6502 struct inode *inode = sb_dqopt(sb)->files[type]; 6503 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6504 int offset = off & (sb->s_blocksize - 1); 6505 int tocopy; 6506 size_t toread; 6507 struct buffer_head *bh; 6508 loff_t i_size = i_size_read(inode); 6509 6510 if (off > i_size) 6511 return 0; 6512 if (off+len > i_size) 6513 len = i_size-off; 6514 toread = len; 6515 while (toread > 0) { 6516 tocopy = sb->s_blocksize - offset < toread ? 6517 sb->s_blocksize - offset : toread; 6518 bh = ext4_bread(NULL, inode, blk, 0); 6519 if (IS_ERR(bh)) 6520 return PTR_ERR(bh); 6521 if (!bh) /* A hole? */ 6522 memset(data, 0, tocopy); 6523 else 6524 memcpy(data, bh->b_data+offset, tocopy); 6525 brelse(bh); 6526 offset = 0; 6527 toread -= tocopy; 6528 data += tocopy; 6529 blk++; 6530 } 6531 return len; 6532 } 6533 6534 /* Write to quotafile (we know the transaction is already started and has 6535 * enough credits) */ 6536 static ssize_t ext4_quota_write(struct super_block *sb, int type, 6537 const char *data, size_t len, loff_t off) 6538 { 6539 struct inode *inode = sb_dqopt(sb)->files[type]; 6540 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6541 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 6542 int retries = 0; 6543 struct buffer_head *bh; 6544 handle_t *handle = journal_current_handle(); 6545 6546 if (EXT4_SB(sb)->s_journal && !handle) { 6547 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6548 " cancelled because transaction is not started", 6549 (unsigned long long)off, (unsigned long long)len); 6550 return -EIO; 6551 } 6552 /* 6553 * Since we account only one data block in transaction credits, 6554 * then it is impossible to cross a block boundary. 6555 */ 6556 if (sb->s_blocksize - offset < len) { 6557 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6558 " cancelled because not block aligned", 6559 (unsigned long long)off, (unsigned long long)len); 6560 return -EIO; 6561 } 6562 6563 do { 6564 bh = ext4_bread(handle, inode, blk, 6565 EXT4_GET_BLOCKS_CREATE | 6566 EXT4_GET_BLOCKS_METADATA_NOFAIL); 6567 } while (PTR_ERR(bh) == -ENOSPC && 6568 ext4_should_retry_alloc(inode->i_sb, &retries)); 6569 if (IS_ERR(bh)) 6570 return PTR_ERR(bh); 6571 if (!bh) 6572 goto out; 6573 BUFFER_TRACE(bh, "get write access"); 6574 err = ext4_journal_get_write_access(handle, bh); 6575 if (err) { 6576 brelse(bh); 6577 return err; 6578 } 6579 lock_buffer(bh); 6580 memcpy(bh->b_data+offset, data, len); 6581 flush_dcache_page(bh->b_page); 6582 unlock_buffer(bh); 6583 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6584 brelse(bh); 6585 out: 6586 if (inode->i_size < off + len) { 6587 i_size_write(inode, off + len); 6588 EXT4_I(inode)->i_disksize = inode->i_size; 6589 err2 = ext4_mark_inode_dirty(handle, inode); 6590 if (unlikely(err2 && !err)) 6591 err = err2; 6592 } 6593 return err ? err : len; 6594 } 6595 #endif 6596 6597 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6598 const char *dev_name, void *data) 6599 { 6600 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6601 } 6602 6603 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6604 static inline void register_as_ext2(void) 6605 { 6606 int err = register_filesystem(&ext2_fs_type); 6607 if (err) 6608 printk(KERN_WARNING 6609 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6610 } 6611 6612 static inline void unregister_as_ext2(void) 6613 { 6614 unregister_filesystem(&ext2_fs_type); 6615 } 6616 6617 static inline int ext2_feature_set_ok(struct super_block *sb) 6618 { 6619 if (ext4_has_unknown_ext2_incompat_features(sb)) 6620 return 0; 6621 if (sb_rdonly(sb)) 6622 return 1; 6623 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6624 return 0; 6625 return 1; 6626 } 6627 #else 6628 static inline void register_as_ext2(void) { } 6629 static inline void unregister_as_ext2(void) { } 6630 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6631 #endif 6632 6633 static inline void register_as_ext3(void) 6634 { 6635 int err = register_filesystem(&ext3_fs_type); 6636 if (err) 6637 printk(KERN_WARNING 6638 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6639 } 6640 6641 static inline void unregister_as_ext3(void) 6642 { 6643 unregister_filesystem(&ext3_fs_type); 6644 } 6645 6646 static inline int ext3_feature_set_ok(struct super_block *sb) 6647 { 6648 if (ext4_has_unknown_ext3_incompat_features(sb)) 6649 return 0; 6650 if (!ext4_has_feature_journal(sb)) 6651 return 0; 6652 if (sb_rdonly(sb)) 6653 return 1; 6654 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6655 return 0; 6656 return 1; 6657 } 6658 6659 static struct file_system_type ext4_fs_type = { 6660 .owner = THIS_MODULE, 6661 .name = "ext4", 6662 .mount = ext4_mount, 6663 .kill_sb = kill_block_super, 6664 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 6665 }; 6666 MODULE_ALIAS_FS("ext4"); 6667 6668 /* Shared across all ext4 file systems */ 6669 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6670 6671 static int __init ext4_init_fs(void) 6672 { 6673 int i, err; 6674 6675 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6676 ext4_li_info = NULL; 6677 6678 /* Build-time check for flags consistency */ 6679 ext4_check_flag_values(); 6680 6681 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6682 init_waitqueue_head(&ext4__ioend_wq[i]); 6683 6684 err = ext4_init_es(); 6685 if (err) 6686 return err; 6687 6688 err = ext4_init_pending(); 6689 if (err) 6690 goto out7; 6691 6692 err = ext4_init_post_read_processing(); 6693 if (err) 6694 goto out6; 6695 6696 err = ext4_init_pageio(); 6697 if (err) 6698 goto out5; 6699 6700 err = ext4_init_system_zone(); 6701 if (err) 6702 goto out4; 6703 6704 err = ext4_init_sysfs(); 6705 if (err) 6706 goto out3; 6707 6708 err = ext4_init_mballoc(); 6709 if (err) 6710 goto out2; 6711 err = init_inodecache(); 6712 if (err) 6713 goto out1; 6714 6715 err = ext4_fc_init_dentry_cache(); 6716 if (err) 6717 goto out05; 6718 6719 register_as_ext3(); 6720 register_as_ext2(); 6721 err = register_filesystem(&ext4_fs_type); 6722 if (err) 6723 goto out; 6724 6725 return 0; 6726 out: 6727 unregister_as_ext2(); 6728 unregister_as_ext3(); 6729 out05: 6730 destroy_inodecache(); 6731 out1: 6732 ext4_exit_mballoc(); 6733 out2: 6734 ext4_exit_sysfs(); 6735 out3: 6736 ext4_exit_system_zone(); 6737 out4: 6738 ext4_exit_pageio(); 6739 out5: 6740 ext4_exit_post_read_processing(); 6741 out6: 6742 ext4_exit_pending(); 6743 out7: 6744 ext4_exit_es(); 6745 6746 return err; 6747 } 6748 6749 static void __exit ext4_exit_fs(void) 6750 { 6751 ext4_destroy_lazyinit_thread(); 6752 unregister_as_ext2(); 6753 unregister_as_ext3(); 6754 unregister_filesystem(&ext4_fs_type); 6755 destroy_inodecache(); 6756 ext4_exit_mballoc(); 6757 ext4_exit_sysfs(); 6758 ext4_exit_system_zone(); 6759 ext4_exit_pageio(); 6760 ext4_exit_post_read_processing(); 6761 ext4_exit_es(); 6762 ext4_exit_pending(); 6763 } 6764 6765 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6766 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6767 MODULE_LICENSE("GPL"); 6768 MODULE_SOFTDEP("pre: crc32c"); 6769 module_init(ext4_init_fs) 6770 module_exit(ext4_exit_fs) 6771