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 struct 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 int ext4_commit_super(struct super_block *sb, int sync); 69 static int ext4_mark_recovery_complete(struct super_block *sb, 70 struct ext4_super_block *es); 71 static int ext4_clear_journal_err(struct super_block *sb, 72 struct ext4_super_block *es); 73 static int ext4_sync_fs(struct super_block *sb, int wait); 74 static int ext4_remount(struct super_block *sb, int *flags, char *data); 75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 76 static int ext4_unfreeze(struct super_block *sb); 77 static int ext4_freeze(struct super_block *sb); 78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 79 const char *dev_name, void *data); 80 static inline int ext2_feature_set_ok(struct super_block *sb); 81 static inline int ext3_feature_set_ok(struct super_block *sb); 82 static int ext4_feature_set_ok(struct super_block *sb, int readonly); 83 static void ext4_destroy_lazyinit_thread(void); 84 static void ext4_unregister_li_request(struct super_block *sb); 85 static void ext4_clear_request_list(void); 86 static struct inode *ext4_get_journal_inode(struct super_block *sb, 87 unsigned int journal_inum); 88 89 /* 90 * Lock ordering 91 * 92 * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and 93 * i_mmap_rwsem (inode->i_mmap_rwsem)! 94 * 95 * page fault path: 96 * mmap_lock -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start -> 97 * page lock -> i_data_sem (rw) 98 * 99 * buffered write path: 100 * sb_start_write -> i_mutex -> mmap_lock 101 * sb_start_write -> i_mutex -> transaction start -> page lock -> 102 * i_data_sem (rw) 103 * 104 * truncate: 105 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock 106 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start -> 107 * i_data_sem (rw) 108 * 109 * direct IO: 110 * sb_start_write -> i_mutex -> mmap_lock 111 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 112 * 113 * writepages: 114 * transaction start -> page lock(s) -> i_data_sem (rw) 115 */ 116 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 118 static struct file_system_type ext2_fs_type = { 119 .owner = THIS_MODULE, 120 .name = "ext2", 121 .mount = ext4_mount, 122 .kill_sb = kill_block_super, 123 .fs_flags = FS_REQUIRES_DEV, 124 }; 125 MODULE_ALIAS_FS("ext2"); 126 MODULE_ALIAS("ext2"); 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type) 128 #else 129 #define IS_EXT2_SB(sb) (0) 130 #endif 131 132 133 static struct file_system_type ext3_fs_type = { 134 .owner = THIS_MODULE, 135 .name = "ext3", 136 .mount = ext4_mount, 137 .kill_sb = kill_block_super, 138 .fs_flags = FS_REQUIRES_DEV, 139 }; 140 MODULE_ALIAS_FS("ext3"); 141 MODULE_ALIAS("ext3"); 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type) 143 144 145 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags, 146 bh_end_io_t *end_io) 147 { 148 /* 149 * buffer's verified bit is no longer valid after reading from 150 * disk again due to write out error, clear it to make sure we 151 * recheck the buffer contents. 152 */ 153 clear_buffer_verified(bh); 154 155 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 156 get_bh(bh); 157 submit_bh(REQ_OP_READ, op_flags, bh); 158 } 159 160 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags, 161 bh_end_io_t *end_io) 162 { 163 BUG_ON(!buffer_locked(bh)); 164 165 if (ext4_buffer_uptodate(bh)) { 166 unlock_buffer(bh); 167 return; 168 } 169 __ext4_read_bh(bh, op_flags, end_io); 170 } 171 172 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io) 173 { 174 BUG_ON(!buffer_locked(bh)); 175 176 if (ext4_buffer_uptodate(bh)) { 177 unlock_buffer(bh); 178 return 0; 179 } 180 181 __ext4_read_bh(bh, op_flags, end_io); 182 183 wait_on_buffer(bh); 184 if (buffer_uptodate(bh)) 185 return 0; 186 return -EIO; 187 } 188 189 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait) 190 { 191 if (trylock_buffer(bh)) { 192 if (wait) 193 return ext4_read_bh(bh, op_flags, NULL); 194 ext4_read_bh_nowait(bh, op_flags, NULL); 195 return 0; 196 } 197 if (wait) { 198 wait_on_buffer(bh); 199 if (buffer_uptodate(bh)) 200 return 0; 201 return -EIO; 202 } 203 return 0; 204 } 205 206 /* 207 * This works like __bread_gfp() except it uses ERR_PTR for error 208 * returns. Currently with sb_bread it's impossible to distinguish 209 * between ENOMEM and EIO situations (since both result in a NULL 210 * return. 211 */ 212 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 213 sector_t block, int op_flags, 214 gfp_t gfp) 215 { 216 struct buffer_head *bh; 217 int ret; 218 219 bh = sb_getblk_gfp(sb, block, gfp); 220 if (bh == NULL) 221 return ERR_PTR(-ENOMEM); 222 if (ext4_buffer_uptodate(bh)) 223 return bh; 224 225 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 226 if (ret) { 227 put_bh(bh); 228 return ERR_PTR(ret); 229 } 230 return bh; 231 } 232 233 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 234 int op_flags) 235 { 236 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 237 } 238 239 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 240 sector_t block) 241 { 242 return __ext4_sb_bread_gfp(sb, block, 0, 0); 243 } 244 245 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 246 { 247 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 248 249 if (likely(bh)) { 250 ext4_read_bh_lock(bh, REQ_RAHEAD, false); 251 brelse(bh); 252 } 253 } 254 255 static int ext4_verify_csum_type(struct super_block *sb, 256 struct ext4_super_block *es) 257 { 258 if (!ext4_has_feature_metadata_csum(sb)) 259 return 1; 260 261 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 262 } 263 264 static __le32 ext4_superblock_csum(struct super_block *sb, 265 struct ext4_super_block *es) 266 { 267 struct ext4_sb_info *sbi = EXT4_SB(sb); 268 int offset = offsetof(struct ext4_super_block, s_checksum); 269 __u32 csum; 270 271 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 272 273 return cpu_to_le32(csum); 274 } 275 276 static int ext4_superblock_csum_verify(struct super_block *sb, 277 struct ext4_super_block *es) 278 { 279 if (!ext4_has_metadata_csum(sb)) 280 return 1; 281 282 return es->s_checksum == ext4_superblock_csum(sb, es); 283 } 284 285 void ext4_superblock_csum_set(struct super_block *sb) 286 { 287 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 288 289 if (!ext4_has_metadata_csum(sb)) 290 return; 291 292 /* 293 * Locking the superblock prevents the scenario 294 * where: 295 * 1) a first thread pauses during checksum calculation. 296 * 2) a second thread updates the superblock, recalculates 297 * the checksum, and updates s_checksum 298 * 3) the first thread resumes and finishes its checksum calculation 299 * and updates s_checksum with a potentially stale or torn value. 300 */ 301 lock_buffer(EXT4_SB(sb)->s_sbh); 302 es->s_checksum = ext4_superblock_csum(sb, es); 303 unlock_buffer(EXT4_SB(sb)->s_sbh); 304 } 305 306 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 307 struct ext4_group_desc *bg) 308 { 309 return le32_to_cpu(bg->bg_block_bitmap_lo) | 310 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 311 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 312 } 313 314 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 315 struct ext4_group_desc *bg) 316 { 317 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 318 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 319 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 320 } 321 322 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 323 struct ext4_group_desc *bg) 324 { 325 return le32_to_cpu(bg->bg_inode_table_lo) | 326 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 327 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 328 } 329 330 __u32 ext4_free_group_clusters(struct super_block *sb, 331 struct ext4_group_desc *bg) 332 { 333 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 334 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 335 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 336 } 337 338 __u32 ext4_free_inodes_count(struct super_block *sb, 339 struct ext4_group_desc *bg) 340 { 341 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 342 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 343 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 344 } 345 346 __u32 ext4_used_dirs_count(struct super_block *sb, 347 struct ext4_group_desc *bg) 348 { 349 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 350 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 351 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 352 } 353 354 __u32 ext4_itable_unused_count(struct super_block *sb, 355 struct ext4_group_desc *bg) 356 { 357 return le16_to_cpu(bg->bg_itable_unused_lo) | 358 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 359 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 360 } 361 362 void ext4_block_bitmap_set(struct super_block *sb, 363 struct ext4_group_desc *bg, ext4_fsblk_t blk) 364 { 365 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 366 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 367 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 368 } 369 370 void ext4_inode_bitmap_set(struct super_block *sb, 371 struct ext4_group_desc *bg, ext4_fsblk_t blk) 372 { 373 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 374 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 375 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 376 } 377 378 void ext4_inode_table_set(struct super_block *sb, 379 struct ext4_group_desc *bg, ext4_fsblk_t blk) 380 { 381 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 382 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 383 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 384 } 385 386 void ext4_free_group_clusters_set(struct super_block *sb, 387 struct ext4_group_desc *bg, __u32 count) 388 { 389 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 390 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 391 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 392 } 393 394 void ext4_free_inodes_set(struct super_block *sb, 395 struct ext4_group_desc *bg, __u32 count) 396 { 397 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 398 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 399 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 400 } 401 402 void ext4_used_dirs_set(struct super_block *sb, 403 struct ext4_group_desc *bg, __u32 count) 404 { 405 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 406 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 407 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 408 } 409 410 void ext4_itable_unused_set(struct super_block *sb, 411 struct ext4_group_desc *bg, __u32 count) 412 { 413 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 414 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 415 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 416 } 417 418 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi) 419 { 420 time64_t now = ktime_get_real_seconds(); 421 422 now = clamp_val(now, 0, (1ull << 40) - 1); 423 424 *lo = cpu_to_le32(lower_32_bits(now)); 425 *hi = upper_32_bits(now); 426 } 427 428 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 429 { 430 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 431 } 432 #define ext4_update_tstamp(es, tstamp) \ 433 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 434 #define ext4_get_tstamp(es, tstamp) \ 435 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 436 437 static void __save_error_info(struct super_block *sb, int error, 438 __u32 ino, __u64 block, 439 const char *func, unsigned int line) 440 { 441 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 442 int err; 443 444 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 445 if (bdev_read_only(sb->s_bdev)) 446 return; 447 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 448 ext4_update_tstamp(es, s_last_error_time); 449 strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func)); 450 es->s_last_error_line = cpu_to_le32(line); 451 es->s_last_error_ino = cpu_to_le32(ino); 452 es->s_last_error_block = cpu_to_le64(block); 453 switch (error) { 454 case EIO: 455 err = EXT4_ERR_EIO; 456 break; 457 case ENOMEM: 458 err = EXT4_ERR_ENOMEM; 459 break; 460 case EFSBADCRC: 461 err = EXT4_ERR_EFSBADCRC; 462 break; 463 case 0: 464 case EFSCORRUPTED: 465 err = EXT4_ERR_EFSCORRUPTED; 466 break; 467 case ENOSPC: 468 err = EXT4_ERR_ENOSPC; 469 break; 470 case ENOKEY: 471 err = EXT4_ERR_ENOKEY; 472 break; 473 case EROFS: 474 err = EXT4_ERR_EROFS; 475 break; 476 case EFBIG: 477 err = EXT4_ERR_EFBIG; 478 break; 479 case EEXIST: 480 err = EXT4_ERR_EEXIST; 481 break; 482 case ERANGE: 483 err = EXT4_ERR_ERANGE; 484 break; 485 case EOVERFLOW: 486 err = EXT4_ERR_EOVERFLOW; 487 break; 488 case EBUSY: 489 err = EXT4_ERR_EBUSY; 490 break; 491 case ENOTDIR: 492 err = EXT4_ERR_ENOTDIR; 493 break; 494 case ENOTEMPTY: 495 err = EXT4_ERR_ENOTEMPTY; 496 break; 497 case ESHUTDOWN: 498 err = EXT4_ERR_ESHUTDOWN; 499 break; 500 case EFAULT: 501 err = EXT4_ERR_EFAULT; 502 break; 503 default: 504 err = EXT4_ERR_UNKNOWN; 505 } 506 es->s_last_error_errcode = err; 507 if (!es->s_first_error_time) { 508 es->s_first_error_time = es->s_last_error_time; 509 es->s_first_error_time_hi = es->s_last_error_time_hi; 510 strncpy(es->s_first_error_func, func, 511 sizeof(es->s_first_error_func)); 512 es->s_first_error_line = cpu_to_le32(line); 513 es->s_first_error_ino = es->s_last_error_ino; 514 es->s_first_error_block = es->s_last_error_block; 515 es->s_first_error_errcode = es->s_last_error_errcode; 516 } 517 /* 518 * Start the daily error reporting function if it hasn't been 519 * started already 520 */ 521 if (!es->s_error_count) 522 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ); 523 le32_add_cpu(&es->s_error_count, 1); 524 } 525 526 static void save_error_info(struct super_block *sb, int error, 527 __u32 ino, __u64 block, 528 const char *func, unsigned int line) 529 { 530 __save_error_info(sb, error, ino, block, func, line); 531 if (!bdev_read_only(sb->s_bdev)) 532 ext4_commit_super(sb, 1); 533 } 534 535 /* 536 * The del_gendisk() function uninitializes the disk-specific data 537 * structures, including the bdi structure, without telling anyone 538 * else. Once this happens, any attempt to call mark_buffer_dirty() 539 * (for example, by ext4_commit_super), will cause a kernel OOPS. 540 * This is a kludge to prevent these oops until we can put in a proper 541 * hook in del_gendisk() to inform the VFS and file system layers. 542 */ 543 static int block_device_ejected(struct super_block *sb) 544 { 545 struct inode *bd_inode = sb->s_bdev->bd_inode; 546 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 547 548 return bdi->dev == NULL; 549 } 550 551 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 552 { 553 struct super_block *sb = journal->j_private; 554 struct ext4_sb_info *sbi = EXT4_SB(sb); 555 int error = is_journal_aborted(journal); 556 struct ext4_journal_cb_entry *jce; 557 558 BUG_ON(txn->t_state == T_FINISHED); 559 560 ext4_process_freed_data(sb, txn->t_tid); 561 562 spin_lock(&sbi->s_md_lock); 563 while (!list_empty(&txn->t_private_list)) { 564 jce = list_entry(txn->t_private_list.next, 565 struct ext4_journal_cb_entry, jce_list); 566 list_del_init(&jce->jce_list); 567 spin_unlock(&sbi->s_md_lock); 568 jce->jce_func(sb, jce, error); 569 spin_lock(&sbi->s_md_lock); 570 } 571 spin_unlock(&sbi->s_md_lock); 572 } 573 574 /* 575 * This writepage callback for write_cache_pages() 576 * takes care of a few cases after page cleaning. 577 * 578 * write_cache_pages() already checks for dirty pages 579 * and calls clear_page_dirty_for_io(), which we want, 580 * to write protect the pages. 581 * 582 * However, we may have to redirty a page (see below.) 583 */ 584 static int ext4_journalled_writepage_callback(struct page *page, 585 struct writeback_control *wbc, 586 void *data) 587 { 588 transaction_t *transaction = (transaction_t *) data; 589 struct buffer_head *bh, *head; 590 struct journal_head *jh; 591 592 bh = head = page_buffers(page); 593 do { 594 /* 595 * We have to redirty a page in these cases: 596 * 1) If buffer is dirty, it means the page was dirty because it 597 * contains a buffer that needs checkpointing. So the dirty bit 598 * needs to be preserved so that checkpointing writes the buffer 599 * properly. 600 * 2) If buffer is not part of the committing transaction 601 * (we may have just accidentally come across this buffer because 602 * inode range tracking is not exact) or if the currently running 603 * transaction already contains this buffer as well, dirty bit 604 * needs to be preserved so that the buffer gets writeprotected 605 * properly on running transaction's commit. 606 */ 607 jh = bh2jh(bh); 608 if (buffer_dirty(bh) || 609 (jh && (jh->b_transaction != transaction || 610 jh->b_next_transaction))) { 611 redirty_page_for_writepage(wbc, page); 612 goto out; 613 } 614 } while ((bh = bh->b_this_page) != head); 615 616 out: 617 return AOP_WRITEPAGE_ACTIVATE; 618 } 619 620 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 621 { 622 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 623 struct writeback_control wbc = { 624 .sync_mode = WB_SYNC_ALL, 625 .nr_to_write = LONG_MAX, 626 .range_start = jinode->i_dirty_start, 627 .range_end = jinode->i_dirty_end, 628 }; 629 630 return write_cache_pages(mapping, &wbc, 631 ext4_journalled_writepage_callback, 632 jinode->i_transaction); 633 } 634 635 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 636 { 637 int ret; 638 639 if (ext4_should_journal_data(jinode->i_vfs_inode)) 640 ret = ext4_journalled_submit_inode_data_buffers(jinode); 641 else 642 ret = jbd2_journal_submit_inode_data_buffers(jinode); 643 644 return ret; 645 } 646 647 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 648 { 649 int ret = 0; 650 651 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 652 ret = jbd2_journal_finish_inode_data_buffers(jinode); 653 654 return ret; 655 } 656 657 static bool system_going_down(void) 658 { 659 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 660 || system_state == SYSTEM_RESTART; 661 } 662 663 /* Deal with the reporting of failure conditions on a filesystem such as 664 * inconsistencies detected or read IO failures. 665 * 666 * On ext2, we can store the error state of the filesystem in the 667 * superblock. That is not possible on ext4, because we may have other 668 * write ordering constraints on the superblock which prevent us from 669 * writing it out straight away; and given that the journal is about to 670 * be aborted, we can't rely on the current, or future, transactions to 671 * write out the superblock safely. 672 * 673 * We'll just use the jbd2_journal_abort() error code to record an error in 674 * the journal instead. On recovery, the journal will complain about 675 * that error until we've noted it down and cleared it. 676 */ 677 678 static void ext4_handle_error(struct super_block *sb) 679 { 680 if (test_opt(sb, WARN_ON_ERROR)) 681 WARN_ON_ONCE(1); 682 683 if (sb_rdonly(sb)) 684 return; 685 686 if (!test_opt(sb, ERRORS_CONT)) { 687 journal_t *journal = EXT4_SB(sb)->s_journal; 688 689 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 690 if (journal) 691 jbd2_journal_abort(journal, -EIO); 692 } 693 /* 694 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 695 * could panic during 'reboot -f' as the underlying device got already 696 * disabled. 697 */ 698 if (test_opt(sb, ERRORS_RO) || system_going_down()) { 699 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 700 /* 701 * Make sure updated value of ->s_mount_flags will be visible 702 * before ->s_flags update 703 */ 704 smp_wmb(); 705 sb->s_flags |= SB_RDONLY; 706 } else if (test_opt(sb, ERRORS_PANIC)) { 707 panic("EXT4-fs (device %s): panic forced after error\n", 708 sb->s_id); 709 } 710 } 711 712 #define ext4_error_ratelimit(sb) \ 713 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 714 "EXT4-fs error") 715 716 void __ext4_error(struct super_block *sb, const char *function, 717 unsigned int line, int error, __u64 block, 718 const char *fmt, ...) 719 { 720 struct va_format vaf; 721 va_list args; 722 723 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 724 return; 725 726 trace_ext4_error(sb, function, line); 727 if (ext4_error_ratelimit(sb)) { 728 va_start(args, fmt); 729 vaf.fmt = fmt; 730 vaf.va = &args; 731 printk(KERN_CRIT 732 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 733 sb->s_id, function, line, current->comm, &vaf); 734 va_end(args); 735 } 736 save_error_info(sb, error, 0, block, function, line); 737 ext4_handle_error(sb); 738 } 739 740 void __ext4_error_inode(struct inode *inode, const char *function, 741 unsigned int line, ext4_fsblk_t block, int error, 742 const char *fmt, ...) 743 { 744 va_list args; 745 struct va_format vaf; 746 747 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 748 return; 749 750 trace_ext4_error(inode->i_sb, function, line); 751 if (ext4_error_ratelimit(inode->i_sb)) { 752 va_start(args, fmt); 753 vaf.fmt = fmt; 754 vaf.va = &args; 755 if (block) 756 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 757 "inode #%lu: block %llu: comm %s: %pV\n", 758 inode->i_sb->s_id, function, line, inode->i_ino, 759 block, current->comm, &vaf); 760 else 761 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 762 "inode #%lu: comm %s: %pV\n", 763 inode->i_sb->s_id, function, line, inode->i_ino, 764 current->comm, &vaf); 765 va_end(args); 766 } 767 save_error_info(inode->i_sb, error, inode->i_ino, block, 768 function, line); 769 ext4_handle_error(inode->i_sb); 770 } 771 772 void __ext4_error_file(struct file *file, const char *function, 773 unsigned int line, ext4_fsblk_t block, 774 const char *fmt, ...) 775 { 776 va_list args; 777 struct va_format vaf; 778 struct inode *inode = file_inode(file); 779 char pathname[80], *path; 780 781 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 782 return; 783 784 trace_ext4_error(inode->i_sb, function, line); 785 if (ext4_error_ratelimit(inode->i_sb)) { 786 path = file_path(file, pathname, sizeof(pathname)); 787 if (IS_ERR(path)) 788 path = "(unknown)"; 789 va_start(args, fmt); 790 vaf.fmt = fmt; 791 vaf.va = &args; 792 if (block) 793 printk(KERN_CRIT 794 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 795 "block %llu: comm %s: path %s: %pV\n", 796 inode->i_sb->s_id, function, line, inode->i_ino, 797 block, current->comm, path, &vaf); 798 else 799 printk(KERN_CRIT 800 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 801 "comm %s: path %s: %pV\n", 802 inode->i_sb->s_id, function, line, inode->i_ino, 803 current->comm, path, &vaf); 804 va_end(args); 805 } 806 save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block, 807 function, line); 808 ext4_handle_error(inode->i_sb); 809 } 810 811 const char *ext4_decode_error(struct super_block *sb, int errno, 812 char nbuf[16]) 813 { 814 char *errstr = NULL; 815 816 switch (errno) { 817 case -EFSCORRUPTED: 818 errstr = "Corrupt filesystem"; 819 break; 820 case -EFSBADCRC: 821 errstr = "Filesystem failed CRC"; 822 break; 823 case -EIO: 824 errstr = "IO failure"; 825 break; 826 case -ENOMEM: 827 errstr = "Out of memory"; 828 break; 829 case -EROFS: 830 if (!sb || (EXT4_SB(sb)->s_journal && 831 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 832 errstr = "Journal has aborted"; 833 else 834 errstr = "Readonly filesystem"; 835 break; 836 default: 837 /* If the caller passed in an extra buffer for unknown 838 * errors, textualise them now. Else we just return 839 * NULL. */ 840 if (nbuf) { 841 /* Check for truncated error codes... */ 842 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 843 errstr = nbuf; 844 } 845 break; 846 } 847 848 return errstr; 849 } 850 851 /* __ext4_std_error decodes expected errors from journaling functions 852 * automatically and invokes the appropriate error response. */ 853 854 void __ext4_std_error(struct super_block *sb, const char *function, 855 unsigned int line, int errno) 856 { 857 char nbuf[16]; 858 const char *errstr; 859 860 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 861 return; 862 863 /* Special case: if the error is EROFS, and we're not already 864 * inside a transaction, then there's really no point in logging 865 * an error. */ 866 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 867 return; 868 869 if (ext4_error_ratelimit(sb)) { 870 errstr = ext4_decode_error(sb, errno, nbuf); 871 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 872 sb->s_id, function, line, errstr); 873 } 874 875 save_error_info(sb, -errno, 0, 0, function, line); 876 ext4_handle_error(sb); 877 } 878 879 /* 880 * ext4_abort is a much stronger failure handler than ext4_error. The 881 * abort function may be used to deal with unrecoverable failures such 882 * as journal IO errors or ENOMEM at a critical moment in log management. 883 * 884 * We unconditionally force the filesystem into an ABORT|READONLY state, 885 * unless the error response on the fs has been set to panic in which 886 * case we take the easy way out and panic immediately. 887 */ 888 889 void __ext4_abort(struct super_block *sb, const char *function, 890 unsigned int line, int error, const char *fmt, ...) 891 { 892 struct va_format vaf; 893 va_list args; 894 895 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 896 return; 897 898 save_error_info(sb, error, 0, 0, function, line); 899 va_start(args, fmt); 900 vaf.fmt = fmt; 901 vaf.va = &args; 902 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n", 903 sb->s_id, function, line, &vaf); 904 va_end(args); 905 906 if (sb_rdonly(sb) == 0) { 907 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 908 if (EXT4_SB(sb)->s_journal) 909 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO); 910 911 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 912 /* 913 * Make sure updated value of ->s_mount_flags will be visible 914 * before ->s_flags update 915 */ 916 smp_wmb(); 917 sb->s_flags |= SB_RDONLY; 918 } 919 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) 920 panic("EXT4-fs panic from previous error\n"); 921 } 922 923 void __ext4_msg(struct super_block *sb, 924 const char *prefix, const char *fmt, ...) 925 { 926 struct va_format vaf; 927 va_list args; 928 929 atomic_inc(&EXT4_SB(sb)->s_msg_count); 930 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs")) 931 return; 932 933 va_start(args, fmt); 934 vaf.fmt = fmt; 935 vaf.va = &args; 936 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 937 va_end(args); 938 } 939 940 static int ext4_warning_ratelimit(struct super_block *sb) 941 { 942 atomic_inc(&EXT4_SB(sb)->s_warning_count); 943 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 944 "EXT4-fs warning"); 945 } 946 947 void __ext4_warning(struct super_block *sb, const char *function, 948 unsigned int line, const char *fmt, ...) 949 { 950 struct va_format vaf; 951 va_list args; 952 953 if (!ext4_warning_ratelimit(sb)) 954 return; 955 956 va_start(args, fmt); 957 vaf.fmt = fmt; 958 vaf.va = &args; 959 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 960 sb->s_id, function, line, &vaf); 961 va_end(args); 962 } 963 964 void __ext4_warning_inode(const struct inode *inode, const char *function, 965 unsigned int line, const char *fmt, ...) 966 { 967 struct va_format vaf; 968 va_list args; 969 970 if (!ext4_warning_ratelimit(inode->i_sb)) 971 return; 972 973 va_start(args, fmt); 974 vaf.fmt = fmt; 975 vaf.va = &args; 976 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 977 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 978 function, line, inode->i_ino, current->comm, &vaf); 979 va_end(args); 980 } 981 982 void __ext4_grp_locked_error(const char *function, unsigned int line, 983 struct super_block *sb, ext4_group_t grp, 984 unsigned long ino, ext4_fsblk_t block, 985 const char *fmt, ...) 986 __releases(bitlock) 987 __acquires(bitlock) 988 { 989 struct va_format vaf; 990 va_list args; 991 992 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 993 return; 994 995 trace_ext4_error(sb, function, line); 996 __save_error_info(sb, EFSCORRUPTED, ino, block, function, line); 997 998 if (ext4_error_ratelimit(sb)) { 999 va_start(args, fmt); 1000 vaf.fmt = fmt; 1001 vaf.va = &args; 1002 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1003 sb->s_id, function, line, grp); 1004 if (ino) 1005 printk(KERN_CONT "inode %lu: ", ino); 1006 if (block) 1007 printk(KERN_CONT "block %llu:", 1008 (unsigned long long) block); 1009 printk(KERN_CONT "%pV\n", &vaf); 1010 va_end(args); 1011 } 1012 1013 if (test_opt(sb, WARN_ON_ERROR)) 1014 WARN_ON_ONCE(1); 1015 1016 if (test_opt(sb, ERRORS_CONT)) { 1017 ext4_commit_super(sb, 0); 1018 return; 1019 } 1020 1021 ext4_unlock_group(sb, grp); 1022 ext4_commit_super(sb, 1); 1023 ext4_handle_error(sb); 1024 /* 1025 * We only get here in the ERRORS_RO case; relocking the group 1026 * may be dangerous, but nothing bad will happen since the 1027 * filesystem will have already been marked read/only and the 1028 * journal has been aborted. We return 1 as a hint to callers 1029 * who might what to use the return value from 1030 * ext4_grp_locked_error() to distinguish between the 1031 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1032 * aggressively from the ext4 function in question, with a 1033 * more appropriate error code. 1034 */ 1035 ext4_lock_group(sb, grp); 1036 return; 1037 } 1038 1039 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1040 ext4_group_t group, 1041 unsigned int flags) 1042 { 1043 struct ext4_sb_info *sbi = EXT4_SB(sb); 1044 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1045 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1046 int ret; 1047 1048 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1049 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1050 &grp->bb_state); 1051 if (!ret) 1052 percpu_counter_sub(&sbi->s_freeclusters_counter, 1053 grp->bb_free); 1054 } 1055 1056 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1057 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1058 &grp->bb_state); 1059 if (!ret && gdp) { 1060 int count; 1061 1062 count = ext4_free_inodes_count(sb, gdp); 1063 percpu_counter_sub(&sbi->s_freeinodes_counter, 1064 count); 1065 } 1066 } 1067 } 1068 1069 void ext4_update_dynamic_rev(struct super_block *sb) 1070 { 1071 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1072 1073 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1074 return; 1075 1076 ext4_warning(sb, 1077 "updating to rev %d because of new feature flag, " 1078 "running e2fsck is recommended", 1079 EXT4_DYNAMIC_REV); 1080 1081 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1082 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1083 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1084 /* leave es->s_feature_*compat flags alone */ 1085 /* es->s_uuid will be set by e2fsck if empty */ 1086 1087 /* 1088 * The rest of the superblock fields should be zero, and if not it 1089 * means they are likely already in use, so leave them alone. We 1090 * can leave it up to e2fsck to clean up any inconsistencies there. 1091 */ 1092 } 1093 1094 /* 1095 * Open the external journal device 1096 */ 1097 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 1098 { 1099 struct block_device *bdev; 1100 1101 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 1102 if (IS_ERR(bdev)) 1103 goto fail; 1104 return bdev; 1105 1106 fail: 1107 ext4_msg(sb, KERN_ERR, 1108 "failed to open journal device unknown-block(%u,%u) %ld", 1109 MAJOR(dev), MINOR(dev), PTR_ERR(bdev)); 1110 return NULL; 1111 } 1112 1113 /* 1114 * Release the journal device 1115 */ 1116 static void ext4_blkdev_put(struct block_device *bdev) 1117 { 1118 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1119 } 1120 1121 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 1122 { 1123 struct block_device *bdev; 1124 bdev = sbi->s_journal_bdev; 1125 if (bdev) { 1126 ext4_blkdev_put(bdev); 1127 sbi->s_journal_bdev = NULL; 1128 } 1129 } 1130 1131 static inline struct inode *orphan_list_entry(struct list_head *l) 1132 { 1133 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1134 } 1135 1136 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1137 { 1138 struct list_head *l; 1139 1140 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1141 le32_to_cpu(sbi->s_es->s_last_orphan)); 1142 1143 printk(KERN_ERR "sb_info orphan list:\n"); 1144 list_for_each(l, &sbi->s_orphan) { 1145 struct inode *inode = orphan_list_entry(l); 1146 printk(KERN_ERR " " 1147 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1148 inode->i_sb->s_id, inode->i_ino, inode, 1149 inode->i_mode, inode->i_nlink, 1150 NEXT_ORPHAN(inode)); 1151 } 1152 } 1153 1154 #ifdef CONFIG_QUOTA 1155 static int ext4_quota_off(struct super_block *sb, int type); 1156 1157 static inline void ext4_quota_off_umount(struct super_block *sb) 1158 { 1159 int type; 1160 1161 /* Use our quota_off function to clear inode flags etc. */ 1162 for (type = 0; type < EXT4_MAXQUOTAS; type++) 1163 ext4_quota_off(sb, type); 1164 } 1165 1166 /* 1167 * This is a helper function which is used in the mount/remount 1168 * codepaths (which holds s_umount) to fetch the quota file name. 1169 */ 1170 static inline char *get_qf_name(struct super_block *sb, 1171 struct ext4_sb_info *sbi, 1172 int type) 1173 { 1174 return rcu_dereference_protected(sbi->s_qf_names[type], 1175 lockdep_is_held(&sb->s_umount)); 1176 } 1177 #else 1178 static inline void ext4_quota_off_umount(struct super_block *sb) 1179 { 1180 } 1181 #endif 1182 1183 static void ext4_put_super(struct super_block *sb) 1184 { 1185 struct ext4_sb_info *sbi = EXT4_SB(sb); 1186 struct ext4_super_block *es = sbi->s_es; 1187 struct buffer_head **group_desc; 1188 struct flex_groups **flex_groups; 1189 int aborted = 0; 1190 int i, err; 1191 1192 ext4_unregister_li_request(sb); 1193 ext4_quota_off_umount(sb); 1194 1195 destroy_workqueue(sbi->rsv_conversion_wq); 1196 1197 /* 1198 * Unregister sysfs before destroying jbd2 journal. 1199 * Since we could still access attr_journal_task attribute via sysfs 1200 * path which could have sbi->s_journal->j_task as NULL 1201 */ 1202 ext4_unregister_sysfs(sb); 1203 1204 if (sbi->s_journal) { 1205 aborted = is_journal_aborted(sbi->s_journal); 1206 err = jbd2_journal_destroy(sbi->s_journal); 1207 sbi->s_journal = NULL; 1208 if ((err < 0) && !aborted) { 1209 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1210 } 1211 } 1212 1213 ext4_es_unregister_shrinker(sbi); 1214 del_timer_sync(&sbi->s_err_report); 1215 ext4_release_system_zone(sb); 1216 ext4_mb_release(sb); 1217 ext4_ext_release(sb); 1218 1219 if (!sb_rdonly(sb) && !aborted) { 1220 ext4_clear_feature_journal_needs_recovery(sb); 1221 es->s_state = cpu_to_le16(sbi->s_mount_state); 1222 } 1223 if (!sb_rdonly(sb)) 1224 ext4_commit_super(sb, 1); 1225 1226 rcu_read_lock(); 1227 group_desc = rcu_dereference(sbi->s_group_desc); 1228 for (i = 0; i < sbi->s_gdb_count; i++) 1229 brelse(group_desc[i]); 1230 kvfree(group_desc); 1231 flex_groups = rcu_dereference(sbi->s_flex_groups); 1232 if (flex_groups) { 1233 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1234 kvfree(flex_groups[i]); 1235 kvfree(flex_groups); 1236 } 1237 rcu_read_unlock(); 1238 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1239 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1240 percpu_counter_destroy(&sbi->s_dirs_counter); 1241 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1242 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1243 #ifdef CONFIG_QUOTA 1244 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1245 kfree(get_qf_name(sb, sbi, i)); 1246 #endif 1247 1248 /* Debugging code just in case the in-memory inode orphan list 1249 * isn't empty. The on-disk one can be non-empty if we've 1250 * detected an error and taken the fs readonly, but the 1251 * in-memory list had better be clean by this point. */ 1252 if (!list_empty(&sbi->s_orphan)) 1253 dump_orphan_list(sb, sbi); 1254 J_ASSERT(list_empty(&sbi->s_orphan)); 1255 1256 sync_blockdev(sb->s_bdev); 1257 invalidate_bdev(sb->s_bdev); 1258 if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) { 1259 /* 1260 * Invalidate the journal device's buffers. We don't want them 1261 * floating about in memory - the physical journal device may 1262 * hotswapped, and it breaks the `ro-after' testing code. 1263 */ 1264 sync_blockdev(sbi->s_journal_bdev); 1265 invalidate_bdev(sbi->s_journal_bdev); 1266 ext4_blkdev_remove(sbi); 1267 } 1268 1269 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1270 sbi->s_ea_inode_cache = NULL; 1271 1272 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1273 sbi->s_ea_block_cache = NULL; 1274 1275 if (sbi->s_mmp_tsk) 1276 kthread_stop(sbi->s_mmp_tsk); 1277 brelse(sbi->s_sbh); 1278 sb->s_fs_info = NULL; 1279 /* 1280 * Now that we are completely done shutting down the 1281 * superblock, we need to actually destroy the kobject. 1282 */ 1283 kobject_put(&sbi->s_kobj); 1284 wait_for_completion(&sbi->s_kobj_unregister); 1285 if (sbi->s_chksum_driver) 1286 crypto_free_shash(sbi->s_chksum_driver); 1287 kfree(sbi->s_blockgroup_lock); 1288 fs_put_dax(sbi->s_daxdev); 1289 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1290 #ifdef CONFIG_UNICODE 1291 utf8_unload(sb->s_encoding); 1292 #endif 1293 kfree(sbi); 1294 } 1295 1296 static struct kmem_cache *ext4_inode_cachep; 1297 1298 /* 1299 * Called inside transaction, so use GFP_NOFS 1300 */ 1301 static struct inode *ext4_alloc_inode(struct super_block *sb) 1302 { 1303 struct ext4_inode_info *ei; 1304 1305 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS); 1306 if (!ei) 1307 return NULL; 1308 1309 inode_set_iversion(&ei->vfs_inode, 1); 1310 spin_lock_init(&ei->i_raw_lock); 1311 INIT_LIST_HEAD(&ei->i_prealloc_list); 1312 atomic_set(&ei->i_prealloc_active, 0); 1313 spin_lock_init(&ei->i_prealloc_lock); 1314 ext4_es_init_tree(&ei->i_es_tree); 1315 rwlock_init(&ei->i_es_lock); 1316 INIT_LIST_HEAD(&ei->i_es_list); 1317 ei->i_es_all_nr = 0; 1318 ei->i_es_shk_nr = 0; 1319 ei->i_es_shrink_lblk = 0; 1320 ei->i_reserved_data_blocks = 0; 1321 spin_lock_init(&(ei->i_block_reservation_lock)); 1322 ext4_init_pending_tree(&ei->i_pending_tree); 1323 #ifdef CONFIG_QUOTA 1324 ei->i_reserved_quota = 0; 1325 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1326 #endif 1327 ei->jinode = NULL; 1328 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1329 spin_lock_init(&ei->i_completed_io_lock); 1330 ei->i_sync_tid = 0; 1331 ei->i_datasync_tid = 0; 1332 atomic_set(&ei->i_unwritten, 0); 1333 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1334 ext4_fc_init_inode(&ei->vfs_inode); 1335 mutex_init(&ei->i_fc_lock); 1336 return &ei->vfs_inode; 1337 } 1338 1339 static int ext4_drop_inode(struct inode *inode) 1340 { 1341 int drop = generic_drop_inode(inode); 1342 1343 if (!drop) 1344 drop = fscrypt_drop_inode(inode); 1345 1346 trace_ext4_drop_inode(inode, drop); 1347 return drop; 1348 } 1349 1350 static void ext4_free_in_core_inode(struct inode *inode) 1351 { 1352 fscrypt_free_inode(inode); 1353 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1354 pr_warn("%s: inode %ld still in fc list", 1355 __func__, inode->i_ino); 1356 } 1357 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1358 } 1359 1360 static void ext4_destroy_inode(struct inode *inode) 1361 { 1362 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1363 ext4_msg(inode->i_sb, KERN_ERR, 1364 "Inode %lu (%p): orphan list check failed!", 1365 inode->i_ino, EXT4_I(inode)); 1366 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1367 EXT4_I(inode), sizeof(struct ext4_inode_info), 1368 true); 1369 dump_stack(); 1370 } 1371 } 1372 1373 static void init_once(void *foo) 1374 { 1375 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo; 1376 1377 INIT_LIST_HEAD(&ei->i_orphan); 1378 init_rwsem(&ei->xattr_sem); 1379 init_rwsem(&ei->i_data_sem); 1380 init_rwsem(&ei->i_mmap_sem); 1381 inode_init_once(&ei->vfs_inode); 1382 ext4_fc_init_inode(&ei->vfs_inode); 1383 } 1384 1385 static int __init init_inodecache(void) 1386 { 1387 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1388 sizeof(struct ext4_inode_info), 0, 1389 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1390 SLAB_ACCOUNT), 1391 offsetof(struct ext4_inode_info, i_data), 1392 sizeof_field(struct ext4_inode_info, i_data), 1393 init_once); 1394 if (ext4_inode_cachep == NULL) 1395 return -ENOMEM; 1396 return 0; 1397 } 1398 1399 static void destroy_inodecache(void) 1400 { 1401 /* 1402 * Make sure all delayed rcu free inodes are flushed before we 1403 * destroy cache. 1404 */ 1405 rcu_barrier(); 1406 kmem_cache_destroy(ext4_inode_cachep); 1407 } 1408 1409 void ext4_clear_inode(struct inode *inode) 1410 { 1411 ext4_fc_del(inode); 1412 invalidate_inode_buffers(inode); 1413 clear_inode(inode); 1414 ext4_discard_preallocations(inode, 0); 1415 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1416 dquot_drop(inode); 1417 if (EXT4_I(inode)->jinode) { 1418 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1419 EXT4_I(inode)->jinode); 1420 jbd2_free_inode(EXT4_I(inode)->jinode); 1421 EXT4_I(inode)->jinode = NULL; 1422 } 1423 fscrypt_put_encryption_info(inode); 1424 fsverity_cleanup_inode(inode); 1425 } 1426 1427 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1428 u64 ino, u32 generation) 1429 { 1430 struct inode *inode; 1431 1432 /* 1433 * Currently we don't know the generation for parent directory, so 1434 * a generation of 0 means "accept any" 1435 */ 1436 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1437 if (IS_ERR(inode)) 1438 return ERR_CAST(inode); 1439 if (generation && inode->i_generation != generation) { 1440 iput(inode); 1441 return ERR_PTR(-ESTALE); 1442 } 1443 1444 return inode; 1445 } 1446 1447 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1448 int fh_len, int fh_type) 1449 { 1450 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1451 ext4_nfs_get_inode); 1452 } 1453 1454 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1455 int fh_len, int fh_type) 1456 { 1457 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1458 ext4_nfs_get_inode); 1459 } 1460 1461 static int ext4_nfs_commit_metadata(struct inode *inode) 1462 { 1463 struct writeback_control wbc = { 1464 .sync_mode = WB_SYNC_ALL 1465 }; 1466 1467 trace_ext4_nfs_commit_metadata(inode); 1468 return ext4_write_inode(inode, &wbc); 1469 } 1470 1471 /* 1472 * Try to release metadata pages (indirect blocks, directories) which are 1473 * mapped via the block device. Since these pages could have journal heads 1474 * which would prevent try_to_free_buffers() from freeing them, we must use 1475 * jbd2 layer's try_to_free_buffers() function to release them. 1476 */ 1477 static int bdev_try_to_free_page(struct super_block *sb, struct page *page, 1478 gfp_t wait) 1479 { 1480 journal_t *journal = EXT4_SB(sb)->s_journal; 1481 1482 WARN_ON(PageChecked(page)); 1483 if (!page_has_buffers(page)) 1484 return 0; 1485 if (journal) 1486 return jbd2_journal_try_to_free_buffers(journal, page); 1487 1488 return try_to_free_buffers(page); 1489 } 1490 1491 #ifdef CONFIG_FS_ENCRYPTION 1492 static int ext4_get_context(struct inode *inode, void *ctx, size_t len) 1493 { 1494 return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, 1495 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len); 1496 } 1497 1498 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len, 1499 void *fs_data) 1500 { 1501 handle_t *handle = fs_data; 1502 int res, res2, credits, retries = 0; 1503 1504 /* 1505 * Encrypting the root directory is not allowed because e2fsck expects 1506 * lost+found to exist and be unencrypted, and encrypting the root 1507 * directory would imply encrypting the lost+found directory as well as 1508 * the filename "lost+found" itself. 1509 */ 1510 if (inode->i_ino == EXT4_ROOT_INO) 1511 return -EPERM; 1512 1513 if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode))) 1514 return -EINVAL; 1515 1516 if (ext4_test_inode_flag(inode, EXT4_INODE_DAX)) 1517 return -EOPNOTSUPP; 1518 1519 res = ext4_convert_inline_data(inode); 1520 if (res) 1521 return res; 1522 1523 /* 1524 * If a journal handle was specified, then the encryption context is 1525 * being set on a new inode via inheritance and is part of a larger 1526 * transaction to create the inode. Otherwise the encryption context is 1527 * being set on an existing inode in its own transaction. Only in the 1528 * latter case should the "retry on ENOSPC" logic be used. 1529 */ 1530 1531 if (handle) { 1532 res = ext4_xattr_set_handle(handle, inode, 1533 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 ext4_clear_inode_state(inode, 1539 EXT4_STATE_MAY_INLINE_DATA); 1540 /* 1541 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1542 * S_DAX may be disabled 1543 */ 1544 ext4_set_inode_flags(inode, false); 1545 } 1546 return res; 1547 } 1548 1549 res = dquot_initialize(inode); 1550 if (res) 1551 return res; 1552 retry: 1553 res = ext4_xattr_set_credits(inode, len, false /* is_create */, 1554 &credits); 1555 if (res) 1556 return res; 1557 1558 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); 1559 if (IS_ERR(handle)) 1560 return PTR_ERR(handle); 1561 1562 res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION, 1563 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1564 ctx, len, 0); 1565 if (!res) { 1566 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1567 /* 1568 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1569 * S_DAX may be disabled 1570 */ 1571 ext4_set_inode_flags(inode, false); 1572 res = ext4_mark_inode_dirty(handle, inode); 1573 if (res) 1574 EXT4_ERROR_INODE(inode, "Failed to mark inode dirty"); 1575 } 1576 res2 = ext4_journal_stop(handle); 1577 1578 if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 1579 goto retry; 1580 if (!res) 1581 res = res2; 1582 return res; 1583 } 1584 1585 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb) 1586 { 1587 return EXT4_SB(sb)->s_dummy_enc_policy.policy; 1588 } 1589 1590 static bool ext4_has_stable_inodes(struct super_block *sb) 1591 { 1592 return ext4_has_feature_stable_inodes(sb); 1593 } 1594 1595 static void ext4_get_ino_and_lblk_bits(struct super_block *sb, 1596 int *ino_bits_ret, int *lblk_bits_ret) 1597 { 1598 *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count); 1599 *lblk_bits_ret = 8 * sizeof(ext4_lblk_t); 1600 } 1601 1602 static const struct fscrypt_operations ext4_cryptops = { 1603 .key_prefix = "ext4:", 1604 .get_context = ext4_get_context, 1605 .set_context = ext4_set_context, 1606 .get_dummy_policy = ext4_get_dummy_policy, 1607 .empty_dir = ext4_empty_dir, 1608 .max_namelen = EXT4_NAME_LEN, 1609 .has_stable_inodes = ext4_has_stable_inodes, 1610 .get_ino_and_lblk_bits = ext4_get_ino_and_lblk_bits, 1611 }; 1612 #endif 1613 1614 #ifdef CONFIG_QUOTA 1615 static const char * const quotatypes[] = INITQFNAMES; 1616 #define QTYPE2NAME(t) (quotatypes[t]) 1617 1618 static int ext4_write_dquot(struct dquot *dquot); 1619 static int ext4_acquire_dquot(struct dquot *dquot); 1620 static int ext4_release_dquot(struct dquot *dquot); 1621 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1622 static int ext4_write_info(struct super_block *sb, int type); 1623 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1624 const struct path *path); 1625 static int ext4_quota_on_mount(struct super_block *sb, int type); 1626 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1627 size_t len, loff_t off); 1628 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1629 const char *data, size_t len, loff_t off); 1630 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1631 unsigned int flags); 1632 static int ext4_enable_quotas(struct super_block *sb); 1633 1634 static struct dquot **ext4_get_dquots(struct inode *inode) 1635 { 1636 return EXT4_I(inode)->i_dquot; 1637 } 1638 1639 static const struct dquot_operations ext4_quota_operations = { 1640 .get_reserved_space = ext4_get_reserved_space, 1641 .write_dquot = ext4_write_dquot, 1642 .acquire_dquot = ext4_acquire_dquot, 1643 .release_dquot = ext4_release_dquot, 1644 .mark_dirty = ext4_mark_dquot_dirty, 1645 .write_info = ext4_write_info, 1646 .alloc_dquot = dquot_alloc, 1647 .destroy_dquot = dquot_destroy, 1648 .get_projid = ext4_get_projid, 1649 .get_inode_usage = ext4_get_inode_usage, 1650 .get_next_id = dquot_get_next_id, 1651 }; 1652 1653 static const struct quotactl_ops ext4_qctl_operations = { 1654 .quota_on = ext4_quota_on, 1655 .quota_off = ext4_quota_off, 1656 .quota_sync = dquot_quota_sync, 1657 .get_state = dquot_get_state, 1658 .set_info = dquot_set_dqinfo, 1659 .get_dqblk = dquot_get_dqblk, 1660 .set_dqblk = dquot_set_dqblk, 1661 .get_nextdqblk = dquot_get_next_dqblk, 1662 }; 1663 #endif 1664 1665 static const struct super_operations ext4_sops = { 1666 .alloc_inode = ext4_alloc_inode, 1667 .free_inode = ext4_free_in_core_inode, 1668 .destroy_inode = ext4_destroy_inode, 1669 .write_inode = ext4_write_inode, 1670 .dirty_inode = ext4_dirty_inode, 1671 .drop_inode = ext4_drop_inode, 1672 .evict_inode = ext4_evict_inode, 1673 .put_super = ext4_put_super, 1674 .sync_fs = ext4_sync_fs, 1675 .freeze_fs = ext4_freeze, 1676 .unfreeze_fs = ext4_unfreeze, 1677 .statfs = ext4_statfs, 1678 .remount_fs = ext4_remount, 1679 .show_options = ext4_show_options, 1680 #ifdef CONFIG_QUOTA 1681 .quota_read = ext4_quota_read, 1682 .quota_write = ext4_quota_write, 1683 .get_dquots = ext4_get_dquots, 1684 #endif 1685 .bdev_try_to_free_page = bdev_try_to_free_page, 1686 }; 1687 1688 static const struct export_operations ext4_export_ops = { 1689 .fh_to_dentry = ext4_fh_to_dentry, 1690 .fh_to_parent = ext4_fh_to_parent, 1691 .get_parent = ext4_get_parent, 1692 .commit_metadata = ext4_nfs_commit_metadata, 1693 }; 1694 1695 enum { 1696 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1697 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro, 1698 Opt_nouid32, Opt_debug, Opt_removed, 1699 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, 1700 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1701 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1702 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1703 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1704 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1705 Opt_inlinecrypt, 1706 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota, 1707 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota, 1708 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1709 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, 1710 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1711 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1712 Opt_nowarn_on_error, Opt_mblk_io_submit, 1713 Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize, 1714 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1715 Opt_inode_readahead_blks, Opt_journal_ioprio, 1716 Opt_dioread_nolock, Opt_dioread_lock, 1717 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1718 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1719 Opt_prefetch_block_bitmaps, 1720 #ifdef CONFIG_EXT4_DEBUG 1721 Opt_fc_debug_max_replay, Opt_fc_debug_force 1722 #endif 1723 }; 1724 1725 static const match_table_t tokens = { 1726 {Opt_bsd_df, "bsddf"}, 1727 {Opt_minix_df, "minixdf"}, 1728 {Opt_grpid, "grpid"}, 1729 {Opt_grpid, "bsdgroups"}, 1730 {Opt_nogrpid, "nogrpid"}, 1731 {Opt_nogrpid, "sysvgroups"}, 1732 {Opt_resgid, "resgid=%u"}, 1733 {Opt_resuid, "resuid=%u"}, 1734 {Opt_sb, "sb=%u"}, 1735 {Opt_err_cont, "errors=continue"}, 1736 {Opt_err_panic, "errors=panic"}, 1737 {Opt_err_ro, "errors=remount-ro"}, 1738 {Opt_nouid32, "nouid32"}, 1739 {Opt_debug, "debug"}, 1740 {Opt_removed, "oldalloc"}, 1741 {Opt_removed, "orlov"}, 1742 {Opt_user_xattr, "user_xattr"}, 1743 {Opt_nouser_xattr, "nouser_xattr"}, 1744 {Opt_acl, "acl"}, 1745 {Opt_noacl, "noacl"}, 1746 {Opt_noload, "norecovery"}, 1747 {Opt_noload, "noload"}, 1748 {Opt_removed, "nobh"}, 1749 {Opt_removed, "bh"}, 1750 {Opt_commit, "commit=%u"}, 1751 {Opt_min_batch_time, "min_batch_time=%u"}, 1752 {Opt_max_batch_time, "max_batch_time=%u"}, 1753 {Opt_journal_dev, "journal_dev=%u"}, 1754 {Opt_journal_path, "journal_path=%s"}, 1755 {Opt_journal_checksum, "journal_checksum"}, 1756 {Opt_nojournal_checksum, "nojournal_checksum"}, 1757 {Opt_journal_async_commit, "journal_async_commit"}, 1758 {Opt_abort, "abort"}, 1759 {Opt_data_journal, "data=journal"}, 1760 {Opt_data_ordered, "data=ordered"}, 1761 {Opt_data_writeback, "data=writeback"}, 1762 {Opt_data_err_abort, "data_err=abort"}, 1763 {Opt_data_err_ignore, "data_err=ignore"}, 1764 {Opt_offusrjquota, "usrjquota="}, 1765 {Opt_usrjquota, "usrjquota=%s"}, 1766 {Opt_offgrpjquota, "grpjquota="}, 1767 {Opt_grpjquota, "grpjquota=%s"}, 1768 {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, 1769 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, 1770 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, 1771 {Opt_grpquota, "grpquota"}, 1772 {Opt_noquota, "noquota"}, 1773 {Opt_quota, "quota"}, 1774 {Opt_usrquota, "usrquota"}, 1775 {Opt_prjquota, "prjquota"}, 1776 {Opt_barrier, "barrier=%u"}, 1777 {Opt_barrier, "barrier"}, 1778 {Opt_nobarrier, "nobarrier"}, 1779 {Opt_i_version, "i_version"}, 1780 {Opt_dax, "dax"}, 1781 {Opt_dax_always, "dax=always"}, 1782 {Opt_dax_inode, "dax=inode"}, 1783 {Opt_dax_never, "dax=never"}, 1784 {Opt_stripe, "stripe=%u"}, 1785 {Opt_delalloc, "delalloc"}, 1786 {Opt_warn_on_error, "warn_on_error"}, 1787 {Opt_nowarn_on_error, "nowarn_on_error"}, 1788 {Opt_lazytime, "lazytime"}, 1789 {Opt_nolazytime, "nolazytime"}, 1790 {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"}, 1791 {Opt_nodelalloc, "nodelalloc"}, 1792 {Opt_removed, "mblk_io_submit"}, 1793 {Opt_removed, "nomblk_io_submit"}, 1794 {Opt_block_validity, "block_validity"}, 1795 {Opt_noblock_validity, "noblock_validity"}, 1796 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, 1797 {Opt_journal_ioprio, "journal_ioprio=%u"}, 1798 {Opt_auto_da_alloc, "auto_da_alloc=%u"}, 1799 {Opt_auto_da_alloc, "auto_da_alloc"}, 1800 {Opt_noauto_da_alloc, "noauto_da_alloc"}, 1801 {Opt_dioread_nolock, "dioread_nolock"}, 1802 {Opt_dioread_lock, "nodioread_nolock"}, 1803 {Opt_dioread_lock, "dioread_lock"}, 1804 {Opt_discard, "discard"}, 1805 {Opt_nodiscard, "nodiscard"}, 1806 {Opt_init_itable, "init_itable=%u"}, 1807 {Opt_init_itable, "init_itable"}, 1808 {Opt_noinit_itable, "noinit_itable"}, 1809 #ifdef CONFIG_EXT4_DEBUG 1810 {Opt_fc_debug_force, "fc_debug_force"}, 1811 {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"}, 1812 #endif 1813 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, 1814 {Opt_test_dummy_encryption, "test_dummy_encryption=%s"}, 1815 {Opt_test_dummy_encryption, "test_dummy_encryption"}, 1816 {Opt_inlinecrypt, "inlinecrypt"}, 1817 {Opt_nombcache, "nombcache"}, 1818 {Opt_nombcache, "no_mbcache"}, /* for backward compatibility */ 1819 {Opt_prefetch_block_bitmaps, "prefetch_block_bitmaps"}, 1820 {Opt_removed, "check=none"}, /* mount option from ext2/3 */ 1821 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */ 1822 {Opt_removed, "reservation"}, /* mount option from ext2/3 */ 1823 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */ 1824 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */ 1825 {Opt_err, NULL}, 1826 }; 1827 1828 static ext4_fsblk_t get_sb_block(void **data) 1829 { 1830 ext4_fsblk_t sb_block; 1831 char *options = (char *) *data; 1832 1833 if (!options || strncmp(options, "sb=", 3) != 0) 1834 return 1; /* Default location */ 1835 1836 options += 3; 1837 /* TODO: use simple_strtoll with >32bit ext4 */ 1838 sb_block = simple_strtoul(options, &options, 0); 1839 if (*options && *options != ',') { 1840 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n", 1841 (char *) *data); 1842 return 1; 1843 } 1844 if (*options == ',') 1845 options++; 1846 *data = (void *) options; 1847 1848 return sb_block; 1849 } 1850 1851 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1852 static const char deprecated_msg[] = 1853 "Mount option \"%s\" will be removed by %s\n" 1854 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1855 1856 #ifdef CONFIG_QUOTA 1857 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args) 1858 { 1859 struct ext4_sb_info *sbi = EXT4_SB(sb); 1860 char *qname, *old_qname = get_qf_name(sb, sbi, qtype); 1861 int ret = -1; 1862 1863 if (sb_any_quota_loaded(sb) && !old_qname) { 1864 ext4_msg(sb, KERN_ERR, 1865 "Cannot change journaled " 1866 "quota options when quota turned on"); 1867 return -1; 1868 } 1869 if (ext4_has_feature_quota(sb)) { 1870 ext4_msg(sb, KERN_INFO, "Journaled quota options " 1871 "ignored when QUOTA feature is enabled"); 1872 return 1; 1873 } 1874 qname = match_strdup(args); 1875 if (!qname) { 1876 ext4_msg(sb, KERN_ERR, 1877 "Not enough memory for storing quotafile name"); 1878 return -1; 1879 } 1880 if (old_qname) { 1881 if (strcmp(old_qname, qname) == 0) 1882 ret = 1; 1883 else 1884 ext4_msg(sb, KERN_ERR, 1885 "%s quota file already specified", 1886 QTYPE2NAME(qtype)); 1887 goto errout; 1888 } 1889 if (strchr(qname, '/')) { 1890 ext4_msg(sb, KERN_ERR, 1891 "quotafile must be on filesystem root"); 1892 goto errout; 1893 } 1894 rcu_assign_pointer(sbi->s_qf_names[qtype], qname); 1895 set_opt(sb, QUOTA); 1896 return 1; 1897 errout: 1898 kfree(qname); 1899 return ret; 1900 } 1901 1902 static int clear_qf_name(struct super_block *sb, int qtype) 1903 { 1904 1905 struct ext4_sb_info *sbi = EXT4_SB(sb); 1906 char *old_qname = get_qf_name(sb, sbi, qtype); 1907 1908 if (sb_any_quota_loaded(sb) && old_qname) { 1909 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options" 1910 " when quota turned on"); 1911 return -1; 1912 } 1913 rcu_assign_pointer(sbi->s_qf_names[qtype], NULL); 1914 synchronize_rcu(); 1915 kfree(old_qname); 1916 return 1; 1917 } 1918 #endif 1919 1920 #define MOPT_SET 0x0001 1921 #define MOPT_CLEAR 0x0002 1922 #define MOPT_NOSUPPORT 0x0004 1923 #define MOPT_EXPLICIT 0x0008 1924 #define MOPT_CLEAR_ERR 0x0010 1925 #define MOPT_GTE0 0x0020 1926 #ifdef CONFIG_QUOTA 1927 #define MOPT_Q 0 1928 #define MOPT_QFMT 0x0040 1929 #else 1930 #define MOPT_Q MOPT_NOSUPPORT 1931 #define MOPT_QFMT MOPT_NOSUPPORT 1932 #endif 1933 #define MOPT_DATAJ 0x0080 1934 #define MOPT_NO_EXT2 0x0100 1935 #define MOPT_NO_EXT3 0x0200 1936 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1937 #define MOPT_STRING 0x0400 1938 #define MOPT_SKIP 0x0800 1939 #define MOPT_2 0x1000 1940 1941 static const struct mount_opts { 1942 int token; 1943 int mount_opt; 1944 int flags; 1945 } ext4_mount_opts[] = { 1946 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1947 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1948 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1949 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1950 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1951 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1952 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1953 MOPT_EXT4_ONLY | MOPT_SET}, 1954 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1955 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1956 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1957 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1958 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1959 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1960 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1961 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1962 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1963 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1964 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1965 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1966 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1967 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1968 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1969 EXT4_MOUNT_JOURNAL_CHECKSUM), 1970 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1971 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1972 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR}, 1973 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR}, 1974 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR}, 1975 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, 1976 MOPT_NO_EXT2}, 1977 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, 1978 MOPT_NO_EXT2}, 1979 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1980 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1981 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1982 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1983 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1984 {Opt_commit, 0, MOPT_GTE0}, 1985 {Opt_max_batch_time, 0, MOPT_GTE0}, 1986 {Opt_min_batch_time, 0, MOPT_GTE0}, 1987 {Opt_inode_readahead_blks, 0, MOPT_GTE0}, 1988 {Opt_init_itable, 0, MOPT_GTE0}, 1989 {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP}, 1990 {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS, 1991 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1992 {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE, 1993 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1994 {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER, 1995 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1996 {Opt_stripe, 0, MOPT_GTE0}, 1997 {Opt_resuid, 0, MOPT_GTE0}, 1998 {Opt_resgid, 0, MOPT_GTE0}, 1999 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 2000 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING}, 2001 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 2002 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 2003 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 2004 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, 2005 MOPT_NO_EXT2 | MOPT_DATAJ}, 2006 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 2007 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR}, 2008 #ifdef CONFIG_EXT4_FS_POSIX_ACL 2009 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 2010 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR}, 2011 #else 2012 {Opt_acl, 0, MOPT_NOSUPPORT}, 2013 {Opt_noacl, 0, MOPT_NOSUPPORT}, 2014 #endif 2015 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 2016 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 2017 {Opt_debug_want_extra_isize, 0, MOPT_GTE0}, 2018 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 2019 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 2020 MOPT_SET | MOPT_Q}, 2021 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 2022 MOPT_SET | MOPT_Q}, 2023 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 2024 MOPT_SET | MOPT_Q}, 2025 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2026 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 2027 MOPT_CLEAR | MOPT_Q}, 2028 {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING}, 2029 {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING}, 2030 {Opt_offusrjquota, 0, MOPT_Q}, 2031 {Opt_offgrpjquota, 0, MOPT_Q}, 2032 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT}, 2033 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT}, 2034 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT}, 2035 {Opt_max_dir_size_kb, 0, MOPT_GTE0}, 2036 {Opt_test_dummy_encryption, 0, MOPT_STRING}, 2037 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 2038 {Opt_prefetch_block_bitmaps, EXT4_MOUNT_PREFETCH_BLOCK_BITMAPS, 2039 MOPT_SET}, 2040 #ifdef CONFIG_EXT4_DEBUG 2041 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 2042 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 2043 {Opt_fc_debug_max_replay, 0, MOPT_GTE0}, 2044 #endif 2045 {Opt_err, 0, 0} 2046 }; 2047 2048 #ifdef CONFIG_UNICODE 2049 static const struct ext4_sb_encodings { 2050 __u16 magic; 2051 char *name; 2052 char *version; 2053 } ext4_sb_encoding_map[] = { 2054 {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"}, 2055 }; 2056 2057 static int ext4_sb_read_encoding(const struct ext4_super_block *es, 2058 const struct ext4_sb_encodings **encoding, 2059 __u16 *flags) 2060 { 2061 __u16 magic = le16_to_cpu(es->s_encoding); 2062 int i; 2063 2064 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 2065 if (magic == ext4_sb_encoding_map[i].magic) 2066 break; 2067 2068 if (i >= ARRAY_SIZE(ext4_sb_encoding_map)) 2069 return -EINVAL; 2070 2071 *encoding = &ext4_sb_encoding_map[i]; 2072 *flags = le16_to_cpu(es->s_encoding_flags); 2073 2074 return 0; 2075 } 2076 #endif 2077 2078 static int ext4_set_test_dummy_encryption(struct super_block *sb, 2079 const char *opt, 2080 const substring_t *arg, 2081 bool is_remount) 2082 { 2083 #ifdef CONFIG_FS_ENCRYPTION 2084 struct ext4_sb_info *sbi = EXT4_SB(sb); 2085 int err; 2086 2087 /* 2088 * This mount option is just for testing, and it's not worthwhile to 2089 * implement the extra complexity (e.g. RCU protection) that would be 2090 * needed to allow it to be set or changed during remount. We do allow 2091 * it to be specified during remount, but only if there is no change. 2092 */ 2093 if (is_remount && !sbi->s_dummy_enc_policy.policy) { 2094 ext4_msg(sb, KERN_WARNING, 2095 "Can't set test_dummy_encryption on remount"); 2096 return -1; 2097 } 2098 err = fscrypt_set_test_dummy_encryption(sb, arg->from, 2099 &sbi->s_dummy_enc_policy); 2100 if (err) { 2101 if (err == -EEXIST) 2102 ext4_msg(sb, KERN_WARNING, 2103 "Can't change test_dummy_encryption on remount"); 2104 else if (err == -EINVAL) 2105 ext4_msg(sb, KERN_WARNING, 2106 "Value of option \"%s\" is unrecognized", opt); 2107 else 2108 ext4_msg(sb, KERN_WARNING, 2109 "Error processing option \"%s\" [%d]", 2110 opt, err); 2111 return -1; 2112 } 2113 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2114 #else 2115 ext4_msg(sb, KERN_WARNING, 2116 "Test dummy encryption mount option ignored"); 2117 #endif 2118 return 1; 2119 } 2120 2121 static int handle_mount_opt(struct super_block *sb, char *opt, int token, 2122 substring_t *args, unsigned long *journal_devnum, 2123 unsigned int *journal_ioprio, int is_remount) 2124 { 2125 struct ext4_sb_info *sbi = EXT4_SB(sb); 2126 const struct mount_opts *m; 2127 kuid_t uid; 2128 kgid_t gid; 2129 int arg = 0; 2130 2131 #ifdef CONFIG_QUOTA 2132 if (token == Opt_usrjquota) 2133 return set_qf_name(sb, USRQUOTA, &args[0]); 2134 else if (token == Opt_grpjquota) 2135 return set_qf_name(sb, GRPQUOTA, &args[0]); 2136 else if (token == Opt_offusrjquota) 2137 return clear_qf_name(sb, USRQUOTA); 2138 else if (token == Opt_offgrpjquota) 2139 return clear_qf_name(sb, GRPQUOTA); 2140 #endif 2141 switch (token) { 2142 case Opt_noacl: 2143 case Opt_nouser_xattr: 2144 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5"); 2145 break; 2146 case Opt_sb: 2147 return 1; /* handled by get_sb_block() */ 2148 case Opt_removed: 2149 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt); 2150 return 1; 2151 case Opt_abort: 2152 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 2153 return 1; 2154 case Opt_i_version: 2155 sb->s_flags |= SB_I_VERSION; 2156 return 1; 2157 case Opt_lazytime: 2158 sb->s_flags |= SB_LAZYTIME; 2159 return 1; 2160 case Opt_nolazytime: 2161 sb->s_flags &= ~SB_LAZYTIME; 2162 return 1; 2163 case Opt_inlinecrypt: 2164 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2165 sb->s_flags |= SB_INLINECRYPT; 2166 #else 2167 ext4_msg(sb, KERN_ERR, "inline encryption not supported"); 2168 #endif 2169 return 1; 2170 } 2171 2172 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2173 if (token == m->token) 2174 break; 2175 2176 if (m->token == Opt_err) { 2177 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" " 2178 "or missing value", opt); 2179 return -1; 2180 } 2181 2182 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2183 ext4_msg(sb, KERN_ERR, 2184 "Mount option \"%s\" incompatible with ext2", opt); 2185 return -1; 2186 } 2187 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2188 ext4_msg(sb, KERN_ERR, 2189 "Mount option \"%s\" incompatible with ext3", opt); 2190 return -1; 2191 } 2192 2193 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg)) 2194 return -1; 2195 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0)) 2196 return -1; 2197 if (m->flags & MOPT_EXPLICIT) { 2198 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2199 set_opt2(sb, EXPLICIT_DELALLOC); 2200 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2201 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM); 2202 } else 2203 return -1; 2204 } 2205 if (m->flags & MOPT_CLEAR_ERR) 2206 clear_opt(sb, ERRORS_MASK); 2207 if (token == Opt_noquota && sb_any_quota_loaded(sb)) { 2208 ext4_msg(sb, KERN_ERR, "Cannot change quota " 2209 "options when quota turned on"); 2210 return -1; 2211 } 2212 2213 if (m->flags & MOPT_NOSUPPORT) { 2214 ext4_msg(sb, KERN_ERR, "%s option not supported", opt); 2215 } else if (token == Opt_commit) { 2216 if (arg == 0) 2217 arg = JBD2_DEFAULT_MAX_COMMIT_AGE; 2218 else if (arg > INT_MAX / HZ) { 2219 ext4_msg(sb, KERN_ERR, 2220 "Invalid commit interval %d, " 2221 "must be smaller than %d", 2222 arg, INT_MAX / HZ); 2223 return -1; 2224 } 2225 sbi->s_commit_interval = HZ * arg; 2226 } else if (token == Opt_debug_want_extra_isize) { 2227 if ((arg & 1) || 2228 (arg < 4) || 2229 (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) { 2230 ext4_msg(sb, KERN_ERR, 2231 "Invalid want_extra_isize %d", arg); 2232 return -1; 2233 } 2234 sbi->s_want_extra_isize = arg; 2235 } else if (token == Opt_max_batch_time) { 2236 sbi->s_max_batch_time = arg; 2237 } else if (token == Opt_min_batch_time) { 2238 sbi->s_min_batch_time = arg; 2239 } else if (token == Opt_inode_readahead_blks) { 2240 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) { 2241 ext4_msg(sb, KERN_ERR, 2242 "EXT4-fs: inode_readahead_blks must be " 2243 "0 or a power of 2 smaller than 2^31"); 2244 return -1; 2245 } 2246 sbi->s_inode_readahead_blks = arg; 2247 } else if (token == Opt_init_itable) { 2248 set_opt(sb, INIT_INODE_TABLE); 2249 if (!args->from) 2250 arg = EXT4_DEF_LI_WAIT_MULT; 2251 sbi->s_li_wait_mult = arg; 2252 } else if (token == Opt_max_dir_size_kb) { 2253 sbi->s_max_dir_size_kb = arg; 2254 #ifdef CONFIG_EXT4_DEBUG 2255 } else if (token == Opt_fc_debug_max_replay) { 2256 sbi->s_fc_debug_max_replay = arg; 2257 #endif 2258 } else if (token == Opt_stripe) { 2259 sbi->s_stripe = arg; 2260 } else if (token == Opt_resuid) { 2261 uid = make_kuid(current_user_ns(), arg); 2262 if (!uid_valid(uid)) { 2263 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg); 2264 return -1; 2265 } 2266 sbi->s_resuid = uid; 2267 } else if (token == Opt_resgid) { 2268 gid = make_kgid(current_user_ns(), arg); 2269 if (!gid_valid(gid)) { 2270 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg); 2271 return -1; 2272 } 2273 sbi->s_resgid = gid; 2274 } else if (token == Opt_journal_dev) { 2275 if (is_remount) { 2276 ext4_msg(sb, KERN_ERR, 2277 "Cannot specify journal on remount"); 2278 return -1; 2279 } 2280 *journal_devnum = arg; 2281 } else if (token == Opt_journal_path) { 2282 char *journal_path; 2283 struct inode *journal_inode; 2284 struct path path; 2285 int error; 2286 2287 if (is_remount) { 2288 ext4_msg(sb, KERN_ERR, 2289 "Cannot specify journal on remount"); 2290 return -1; 2291 } 2292 journal_path = match_strdup(&args[0]); 2293 if (!journal_path) { 2294 ext4_msg(sb, KERN_ERR, "error: could not dup " 2295 "journal device string"); 2296 return -1; 2297 } 2298 2299 error = kern_path(journal_path, LOOKUP_FOLLOW, &path); 2300 if (error) { 2301 ext4_msg(sb, KERN_ERR, "error: could not find " 2302 "journal device path: error %d", error); 2303 kfree(journal_path); 2304 return -1; 2305 } 2306 2307 journal_inode = d_inode(path.dentry); 2308 if (!S_ISBLK(journal_inode->i_mode)) { 2309 ext4_msg(sb, KERN_ERR, "error: journal path %s " 2310 "is not a block device", journal_path); 2311 path_put(&path); 2312 kfree(journal_path); 2313 return -1; 2314 } 2315 2316 *journal_devnum = new_encode_dev(journal_inode->i_rdev); 2317 path_put(&path); 2318 kfree(journal_path); 2319 } else if (token == Opt_journal_ioprio) { 2320 if (arg > 7) { 2321 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority" 2322 " (must be 0-7)"); 2323 return -1; 2324 } 2325 *journal_ioprio = 2326 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg); 2327 } else if (token == Opt_test_dummy_encryption) { 2328 return ext4_set_test_dummy_encryption(sb, opt, &args[0], 2329 is_remount); 2330 } else if (m->flags & MOPT_DATAJ) { 2331 if (is_remount) { 2332 if (!sbi->s_journal) 2333 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option"); 2334 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) { 2335 ext4_msg(sb, KERN_ERR, 2336 "Cannot change data mode on remount"); 2337 return -1; 2338 } 2339 } else { 2340 clear_opt(sb, DATA_FLAGS); 2341 sbi->s_mount_opt |= m->mount_opt; 2342 } 2343 #ifdef CONFIG_QUOTA 2344 } else if (m->flags & MOPT_QFMT) { 2345 if (sb_any_quota_loaded(sb) && 2346 sbi->s_jquota_fmt != m->mount_opt) { 2347 ext4_msg(sb, KERN_ERR, "Cannot change journaled " 2348 "quota options when quota turned on"); 2349 return -1; 2350 } 2351 if (ext4_has_feature_quota(sb)) { 2352 ext4_msg(sb, KERN_INFO, 2353 "Quota format mount options ignored " 2354 "when QUOTA feature is enabled"); 2355 return 1; 2356 } 2357 sbi->s_jquota_fmt = m->mount_opt; 2358 #endif 2359 } else if (token == Opt_dax || token == Opt_dax_always || 2360 token == Opt_dax_inode || token == Opt_dax_never) { 2361 #ifdef CONFIG_FS_DAX 2362 switch (token) { 2363 case Opt_dax: 2364 case Opt_dax_always: 2365 if (is_remount && 2366 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2367 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2368 fail_dax_change_remount: 2369 ext4_msg(sb, KERN_ERR, "can't change " 2370 "dax mount option while remounting"); 2371 return -1; 2372 } 2373 if (is_remount && 2374 (test_opt(sb, DATA_FLAGS) == 2375 EXT4_MOUNT_JOURNAL_DATA)) { 2376 ext4_msg(sb, KERN_ERR, "can't mount with " 2377 "both data=journal and dax"); 2378 return -1; 2379 } 2380 ext4_msg(sb, KERN_WARNING, 2381 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 2382 sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS; 2383 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2384 break; 2385 case Opt_dax_never: 2386 if (is_remount && 2387 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2388 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) 2389 goto fail_dax_change_remount; 2390 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2391 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2392 break; 2393 case Opt_dax_inode: 2394 if (is_remount && 2395 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2396 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2397 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) 2398 goto fail_dax_change_remount; 2399 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2400 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2401 /* Strictly for printing options */ 2402 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE; 2403 break; 2404 } 2405 #else 2406 ext4_msg(sb, KERN_INFO, "dax option not supported"); 2407 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2408 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2409 return -1; 2410 #endif 2411 } else if (token == Opt_data_err_abort) { 2412 sbi->s_mount_opt |= m->mount_opt; 2413 } else if (token == Opt_data_err_ignore) { 2414 sbi->s_mount_opt &= ~m->mount_opt; 2415 } else { 2416 if (!args->from) 2417 arg = 1; 2418 if (m->flags & MOPT_CLEAR) 2419 arg = !arg; 2420 else if (unlikely(!(m->flags & MOPT_SET))) { 2421 ext4_msg(sb, KERN_WARNING, 2422 "buggy handling of option %s", opt); 2423 WARN_ON(1); 2424 return -1; 2425 } 2426 if (m->flags & MOPT_2) { 2427 if (arg != 0) 2428 sbi->s_mount_opt2 |= m->mount_opt; 2429 else 2430 sbi->s_mount_opt2 &= ~m->mount_opt; 2431 } else { 2432 if (arg != 0) 2433 sbi->s_mount_opt |= m->mount_opt; 2434 else 2435 sbi->s_mount_opt &= ~m->mount_opt; 2436 } 2437 } 2438 return 1; 2439 } 2440 2441 static int parse_options(char *options, struct super_block *sb, 2442 unsigned long *journal_devnum, 2443 unsigned int *journal_ioprio, 2444 int is_remount) 2445 { 2446 struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb); 2447 char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name; 2448 substring_t args[MAX_OPT_ARGS]; 2449 int token; 2450 2451 if (!options) 2452 return 1; 2453 2454 while ((p = strsep(&options, ",")) != NULL) { 2455 if (!*p) 2456 continue; 2457 /* 2458 * Initialize args struct so we know whether arg was 2459 * found; some options take optional arguments. 2460 */ 2461 args[0].to = args[0].from = NULL; 2462 token = match_token(p, tokens, args); 2463 if (handle_mount_opt(sb, p, token, args, journal_devnum, 2464 journal_ioprio, is_remount) < 0) 2465 return 0; 2466 } 2467 #ifdef CONFIG_QUOTA 2468 /* 2469 * We do the test below only for project quotas. 'usrquota' and 2470 * 'grpquota' mount options are allowed even without quota feature 2471 * to support legacy quotas in quota files. 2472 */ 2473 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) { 2474 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. " 2475 "Cannot enable project quota enforcement."); 2476 return 0; 2477 } 2478 usr_qf_name = get_qf_name(sb, sbi, USRQUOTA); 2479 grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA); 2480 if (usr_qf_name || grp_qf_name) { 2481 if (test_opt(sb, USRQUOTA) && usr_qf_name) 2482 clear_opt(sb, USRQUOTA); 2483 2484 if (test_opt(sb, GRPQUOTA) && grp_qf_name) 2485 clear_opt(sb, GRPQUOTA); 2486 2487 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) { 2488 ext4_msg(sb, KERN_ERR, "old and new quota " 2489 "format mixing"); 2490 return 0; 2491 } 2492 2493 if (!sbi->s_jquota_fmt) { 2494 ext4_msg(sb, KERN_ERR, "journaled quota format " 2495 "not specified"); 2496 return 0; 2497 } 2498 } 2499 #endif 2500 if (test_opt(sb, DIOREAD_NOLOCK)) { 2501 int blocksize = 2502 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2503 if (blocksize < PAGE_SIZE) 2504 ext4_msg(sb, KERN_WARNING, "Warning: mounting with an " 2505 "experimental mount option 'dioread_nolock' " 2506 "for blocksize < PAGE_SIZE"); 2507 } 2508 return 1; 2509 } 2510 2511 static inline void ext4_show_quota_options(struct seq_file *seq, 2512 struct super_block *sb) 2513 { 2514 #if defined(CONFIG_QUOTA) 2515 struct ext4_sb_info *sbi = EXT4_SB(sb); 2516 char *usr_qf_name, *grp_qf_name; 2517 2518 if (sbi->s_jquota_fmt) { 2519 char *fmtname = ""; 2520 2521 switch (sbi->s_jquota_fmt) { 2522 case QFMT_VFS_OLD: 2523 fmtname = "vfsold"; 2524 break; 2525 case QFMT_VFS_V0: 2526 fmtname = "vfsv0"; 2527 break; 2528 case QFMT_VFS_V1: 2529 fmtname = "vfsv1"; 2530 break; 2531 } 2532 seq_printf(seq, ",jqfmt=%s", fmtname); 2533 } 2534 2535 rcu_read_lock(); 2536 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2537 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2538 if (usr_qf_name) 2539 seq_show_option(seq, "usrjquota", usr_qf_name); 2540 if (grp_qf_name) 2541 seq_show_option(seq, "grpjquota", grp_qf_name); 2542 rcu_read_unlock(); 2543 #endif 2544 } 2545 2546 static const char *token2str(int token) 2547 { 2548 const struct match_token *t; 2549 2550 for (t = tokens; t->token != Opt_err; t++) 2551 if (t->token == token && !strchr(t->pattern, '=')) 2552 break; 2553 return t->pattern; 2554 } 2555 2556 /* 2557 * Show an option if 2558 * - it's set to a non-default value OR 2559 * - if the per-sb default is different from the global default 2560 */ 2561 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2562 int nodefs) 2563 { 2564 struct ext4_sb_info *sbi = EXT4_SB(sb); 2565 struct ext4_super_block *es = sbi->s_es; 2566 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2567 const struct mount_opts *m; 2568 char sep = nodefs ? '\n' : ','; 2569 2570 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2571 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2572 2573 if (sbi->s_sb_block != 1) 2574 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2575 2576 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2577 int want_set = m->flags & MOPT_SET; 2578 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2579 (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP) 2580 continue; 2581 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2582 continue; /* skip if same as the default */ 2583 if ((want_set && 2584 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2585 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2586 continue; /* select Opt_noFoo vs Opt_Foo */ 2587 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2588 } 2589 2590 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2591 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2592 SEQ_OPTS_PRINT("resuid=%u", 2593 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2594 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2595 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2596 SEQ_OPTS_PRINT("resgid=%u", 2597 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2598 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2599 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2600 SEQ_OPTS_PUTS("errors=remount-ro"); 2601 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2602 SEQ_OPTS_PUTS("errors=continue"); 2603 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2604 SEQ_OPTS_PUTS("errors=panic"); 2605 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2606 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2607 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2608 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2609 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2610 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2611 if (sb->s_flags & SB_I_VERSION) 2612 SEQ_OPTS_PUTS("i_version"); 2613 if (nodefs || sbi->s_stripe) 2614 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2615 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2616 (sbi->s_mount_opt ^ def_mount_opt)) { 2617 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2618 SEQ_OPTS_PUTS("data=journal"); 2619 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2620 SEQ_OPTS_PUTS("data=ordered"); 2621 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2622 SEQ_OPTS_PUTS("data=writeback"); 2623 } 2624 if (nodefs || 2625 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2626 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2627 sbi->s_inode_readahead_blks); 2628 2629 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2630 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2631 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2632 if (nodefs || sbi->s_max_dir_size_kb) 2633 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2634 if (test_opt(sb, DATA_ERR_ABORT)) 2635 SEQ_OPTS_PUTS("data_err=abort"); 2636 2637 fscrypt_show_test_dummy_encryption(seq, sep, sb); 2638 2639 if (sb->s_flags & SB_INLINECRYPT) 2640 SEQ_OPTS_PUTS("inlinecrypt"); 2641 2642 if (test_opt(sb, DAX_ALWAYS)) { 2643 if (IS_EXT2_SB(sb)) 2644 SEQ_OPTS_PUTS("dax"); 2645 else 2646 SEQ_OPTS_PUTS("dax=always"); 2647 } else if (test_opt2(sb, DAX_NEVER)) { 2648 SEQ_OPTS_PUTS("dax=never"); 2649 } else if (test_opt2(sb, DAX_INODE)) { 2650 SEQ_OPTS_PUTS("dax=inode"); 2651 } 2652 2653 if (test_opt2(sb, JOURNAL_FAST_COMMIT)) 2654 SEQ_OPTS_PUTS("fast_commit"); 2655 2656 ext4_show_quota_options(seq, sb); 2657 return 0; 2658 } 2659 2660 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2661 { 2662 return _ext4_show_options(seq, root->d_sb, 0); 2663 } 2664 2665 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2666 { 2667 struct super_block *sb = seq->private; 2668 int rc; 2669 2670 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2671 rc = _ext4_show_options(seq, sb, 1); 2672 seq_puts(seq, "\n"); 2673 return rc; 2674 } 2675 2676 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2677 int read_only) 2678 { 2679 struct ext4_sb_info *sbi = EXT4_SB(sb); 2680 int err = 0; 2681 2682 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2683 ext4_msg(sb, KERN_ERR, "revision level too high, " 2684 "forcing read-only mode"); 2685 err = -EROFS; 2686 goto done; 2687 } 2688 if (read_only) 2689 goto done; 2690 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2691 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2692 "running e2fsck is recommended"); 2693 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2694 ext4_msg(sb, KERN_WARNING, 2695 "warning: mounting fs with errors, " 2696 "running e2fsck is recommended"); 2697 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2698 le16_to_cpu(es->s_mnt_count) >= 2699 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2700 ext4_msg(sb, KERN_WARNING, 2701 "warning: maximal mount count reached, " 2702 "running e2fsck is recommended"); 2703 else if (le32_to_cpu(es->s_checkinterval) && 2704 (ext4_get_tstamp(es, s_lastcheck) + 2705 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2706 ext4_msg(sb, KERN_WARNING, 2707 "warning: checktime reached, " 2708 "running e2fsck is recommended"); 2709 if (!sbi->s_journal) 2710 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2711 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2712 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2713 le16_add_cpu(&es->s_mnt_count, 1); 2714 ext4_update_tstamp(es, s_mtime); 2715 if (sbi->s_journal) 2716 ext4_set_feature_journal_needs_recovery(sb); 2717 2718 err = ext4_commit_super(sb, 1); 2719 done: 2720 if (test_opt(sb, DEBUG)) 2721 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2722 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2723 sb->s_blocksize, 2724 sbi->s_groups_count, 2725 EXT4_BLOCKS_PER_GROUP(sb), 2726 EXT4_INODES_PER_GROUP(sb), 2727 sbi->s_mount_opt, sbi->s_mount_opt2); 2728 2729 cleancache_init_fs(sb); 2730 return err; 2731 } 2732 2733 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2734 { 2735 struct ext4_sb_info *sbi = EXT4_SB(sb); 2736 struct flex_groups **old_groups, **new_groups; 2737 int size, i, j; 2738 2739 if (!sbi->s_log_groups_per_flex) 2740 return 0; 2741 2742 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2743 if (size <= sbi->s_flex_groups_allocated) 2744 return 0; 2745 2746 new_groups = kvzalloc(roundup_pow_of_two(size * 2747 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 2748 if (!new_groups) { 2749 ext4_msg(sb, KERN_ERR, 2750 "not enough memory for %d flex group pointers", size); 2751 return -ENOMEM; 2752 } 2753 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 2754 new_groups[i] = kvzalloc(roundup_pow_of_two( 2755 sizeof(struct flex_groups)), 2756 GFP_KERNEL); 2757 if (!new_groups[i]) { 2758 for (j = sbi->s_flex_groups_allocated; j < i; j++) 2759 kvfree(new_groups[j]); 2760 kvfree(new_groups); 2761 ext4_msg(sb, KERN_ERR, 2762 "not enough memory for %d flex groups", size); 2763 return -ENOMEM; 2764 } 2765 } 2766 rcu_read_lock(); 2767 old_groups = rcu_dereference(sbi->s_flex_groups); 2768 if (old_groups) 2769 memcpy(new_groups, old_groups, 2770 (sbi->s_flex_groups_allocated * 2771 sizeof(struct flex_groups *))); 2772 rcu_read_unlock(); 2773 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 2774 sbi->s_flex_groups_allocated = size; 2775 if (old_groups) 2776 ext4_kvfree_array_rcu(old_groups); 2777 return 0; 2778 } 2779 2780 static int ext4_fill_flex_info(struct super_block *sb) 2781 { 2782 struct ext4_sb_info *sbi = EXT4_SB(sb); 2783 struct ext4_group_desc *gdp = NULL; 2784 struct flex_groups *fg; 2785 ext4_group_t flex_group; 2786 int i, err; 2787 2788 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2789 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2790 sbi->s_log_groups_per_flex = 0; 2791 return 1; 2792 } 2793 2794 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2795 if (err) 2796 goto failed; 2797 2798 for (i = 0; i < sbi->s_groups_count; i++) { 2799 gdp = ext4_get_group_desc(sb, i, NULL); 2800 2801 flex_group = ext4_flex_group(sbi, i); 2802 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 2803 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 2804 atomic64_add(ext4_free_group_clusters(sb, gdp), 2805 &fg->free_clusters); 2806 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 2807 } 2808 2809 return 1; 2810 failed: 2811 return 0; 2812 } 2813 2814 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2815 struct ext4_group_desc *gdp) 2816 { 2817 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2818 __u16 crc = 0; 2819 __le32 le_group = cpu_to_le32(block_group); 2820 struct ext4_sb_info *sbi = EXT4_SB(sb); 2821 2822 if (ext4_has_metadata_csum(sbi->s_sb)) { 2823 /* Use new metadata_csum algorithm */ 2824 __u32 csum32; 2825 __u16 dummy_csum = 0; 2826 2827 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2828 sizeof(le_group)); 2829 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2830 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2831 sizeof(dummy_csum)); 2832 offset += sizeof(dummy_csum); 2833 if (offset < sbi->s_desc_size) 2834 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2835 sbi->s_desc_size - offset); 2836 2837 crc = csum32 & 0xFFFF; 2838 goto out; 2839 } 2840 2841 /* old crc16 code */ 2842 if (!ext4_has_feature_gdt_csum(sb)) 2843 return 0; 2844 2845 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2846 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2847 crc = crc16(crc, (__u8 *)gdp, offset); 2848 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2849 /* for checksum of struct ext4_group_desc do the rest...*/ 2850 if (ext4_has_feature_64bit(sb) && 2851 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2852 crc = crc16(crc, (__u8 *)gdp + offset, 2853 le16_to_cpu(sbi->s_es->s_desc_size) - 2854 offset); 2855 2856 out: 2857 return cpu_to_le16(crc); 2858 } 2859 2860 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2861 struct ext4_group_desc *gdp) 2862 { 2863 if (ext4_has_group_desc_csum(sb) && 2864 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2865 return 0; 2866 2867 return 1; 2868 } 2869 2870 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2871 struct ext4_group_desc *gdp) 2872 { 2873 if (!ext4_has_group_desc_csum(sb)) 2874 return; 2875 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2876 } 2877 2878 /* Called at mount-time, super-block is locked */ 2879 static int ext4_check_descriptors(struct super_block *sb, 2880 ext4_fsblk_t sb_block, 2881 ext4_group_t *first_not_zeroed) 2882 { 2883 struct ext4_sb_info *sbi = EXT4_SB(sb); 2884 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2885 ext4_fsblk_t last_block; 2886 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2887 ext4_fsblk_t block_bitmap; 2888 ext4_fsblk_t inode_bitmap; 2889 ext4_fsblk_t inode_table; 2890 int flexbg_flag = 0; 2891 ext4_group_t i, grp = sbi->s_groups_count; 2892 2893 if (ext4_has_feature_flex_bg(sb)) 2894 flexbg_flag = 1; 2895 2896 ext4_debug("Checking group descriptors"); 2897 2898 for (i = 0; i < sbi->s_groups_count; i++) { 2899 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2900 2901 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2902 last_block = ext4_blocks_count(sbi->s_es) - 1; 2903 else 2904 last_block = first_block + 2905 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2906 2907 if ((grp == sbi->s_groups_count) && 2908 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2909 grp = i; 2910 2911 block_bitmap = ext4_block_bitmap(sb, gdp); 2912 if (block_bitmap == sb_block) { 2913 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2914 "Block bitmap for group %u overlaps " 2915 "superblock", i); 2916 if (!sb_rdonly(sb)) 2917 return 0; 2918 } 2919 if (block_bitmap >= sb_block + 1 && 2920 block_bitmap <= last_bg_block) { 2921 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2922 "Block bitmap for group %u overlaps " 2923 "block group descriptors", i); 2924 if (!sb_rdonly(sb)) 2925 return 0; 2926 } 2927 if (block_bitmap < first_block || block_bitmap > last_block) { 2928 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2929 "Block bitmap for group %u not in group " 2930 "(block %llu)!", i, block_bitmap); 2931 return 0; 2932 } 2933 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2934 if (inode_bitmap == sb_block) { 2935 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2936 "Inode bitmap for group %u overlaps " 2937 "superblock", i); 2938 if (!sb_rdonly(sb)) 2939 return 0; 2940 } 2941 if (inode_bitmap >= sb_block + 1 && 2942 inode_bitmap <= last_bg_block) { 2943 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2944 "Inode bitmap for group %u overlaps " 2945 "block group descriptors", i); 2946 if (!sb_rdonly(sb)) 2947 return 0; 2948 } 2949 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2950 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2951 "Inode bitmap for group %u not in group " 2952 "(block %llu)!", i, inode_bitmap); 2953 return 0; 2954 } 2955 inode_table = ext4_inode_table(sb, gdp); 2956 if (inode_table == sb_block) { 2957 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2958 "Inode table for group %u overlaps " 2959 "superblock", i); 2960 if (!sb_rdonly(sb)) 2961 return 0; 2962 } 2963 if (inode_table >= sb_block + 1 && 2964 inode_table <= last_bg_block) { 2965 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2966 "Inode table for group %u overlaps " 2967 "block group descriptors", i); 2968 if (!sb_rdonly(sb)) 2969 return 0; 2970 } 2971 if (inode_table < first_block || 2972 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2973 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2974 "Inode table for group %u not in group " 2975 "(block %llu)!", i, inode_table); 2976 return 0; 2977 } 2978 ext4_lock_group(sb, i); 2979 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2980 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2981 "Checksum for group %u failed (%u!=%u)", 2982 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2983 gdp)), le16_to_cpu(gdp->bg_checksum)); 2984 if (!sb_rdonly(sb)) { 2985 ext4_unlock_group(sb, i); 2986 return 0; 2987 } 2988 } 2989 ext4_unlock_group(sb, i); 2990 if (!flexbg_flag) 2991 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2992 } 2993 if (NULL != first_not_zeroed) 2994 *first_not_zeroed = grp; 2995 return 1; 2996 } 2997 2998 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2999 * the superblock) which were deleted from all directories, but held open by 3000 * a process at the time of a crash. We walk the list and try to delete these 3001 * inodes at recovery time (only with a read-write filesystem). 3002 * 3003 * In order to keep the orphan inode chain consistent during traversal (in 3004 * case of crash during recovery), we link each inode into the superblock 3005 * orphan list_head and handle it the same way as an inode deletion during 3006 * normal operation (which journals the operations for us). 3007 * 3008 * We only do an iget() and an iput() on each inode, which is very safe if we 3009 * accidentally point at an in-use or already deleted inode. The worst that 3010 * can happen in this case is that we get a "bit already cleared" message from 3011 * ext4_free_inode(). The only reason we would point at a wrong inode is if 3012 * e2fsck was run on this filesystem, and it must have already done the orphan 3013 * inode cleanup for us, so we can safely abort without any further action. 3014 */ 3015 static void ext4_orphan_cleanup(struct super_block *sb, 3016 struct ext4_super_block *es) 3017 { 3018 unsigned int s_flags = sb->s_flags; 3019 int ret, nr_orphans = 0, nr_truncates = 0; 3020 #ifdef CONFIG_QUOTA 3021 int quota_update = 0; 3022 int i; 3023 #endif 3024 if (!es->s_last_orphan) { 3025 jbd_debug(4, "no orphan inodes to clean up\n"); 3026 return; 3027 } 3028 3029 if (bdev_read_only(sb->s_bdev)) { 3030 ext4_msg(sb, KERN_ERR, "write access " 3031 "unavailable, skipping orphan cleanup"); 3032 return; 3033 } 3034 3035 /* Check if feature set would not allow a r/w mount */ 3036 if (!ext4_feature_set_ok(sb, 0)) { 3037 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 3038 "unknown ROCOMPAT features"); 3039 return; 3040 } 3041 3042 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3043 /* don't clear list on RO mount w/ errors */ 3044 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 3045 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 3046 "clearing orphan list.\n"); 3047 es->s_last_orphan = 0; 3048 } 3049 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3050 return; 3051 } 3052 3053 if (s_flags & SB_RDONLY) { 3054 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 3055 sb->s_flags &= ~SB_RDONLY; 3056 } 3057 #ifdef CONFIG_QUOTA 3058 /* Needed for iput() to work correctly and not trash data */ 3059 sb->s_flags |= SB_ACTIVE; 3060 3061 /* 3062 * Turn on quotas which were not enabled for read-only mounts if 3063 * filesystem has quota feature, so that they are updated correctly. 3064 */ 3065 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 3066 int ret = ext4_enable_quotas(sb); 3067 3068 if (!ret) 3069 quota_update = 1; 3070 else 3071 ext4_msg(sb, KERN_ERR, 3072 "Cannot turn on quotas: error %d", ret); 3073 } 3074 3075 /* Turn on journaled quotas used for old sytle */ 3076 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3077 if (EXT4_SB(sb)->s_qf_names[i]) { 3078 int ret = ext4_quota_on_mount(sb, i); 3079 3080 if (!ret) 3081 quota_update = 1; 3082 else 3083 ext4_msg(sb, KERN_ERR, 3084 "Cannot turn on journaled " 3085 "quota: type %d: error %d", i, ret); 3086 } 3087 } 3088 #endif 3089 3090 while (es->s_last_orphan) { 3091 struct inode *inode; 3092 3093 /* 3094 * We may have encountered an error during cleanup; if 3095 * so, skip the rest. 3096 */ 3097 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3098 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3099 es->s_last_orphan = 0; 3100 break; 3101 } 3102 3103 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 3104 if (IS_ERR(inode)) { 3105 es->s_last_orphan = 0; 3106 break; 3107 } 3108 3109 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 3110 dquot_initialize(inode); 3111 if (inode->i_nlink) { 3112 if (test_opt(sb, DEBUG)) 3113 ext4_msg(sb, KERN_DEBUG, 3114 "%s: truncating inode %lu to %lld bytes", 3115 __func__, inode->i_ino, inode->i_size); 3116 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 3117 inode->i_ino, inode->i_size); 3118 inode_lock(inode); 3119 truncate_inode_pages(inode->i_mapping, inode->i_size); 3120 ret = ext4_truncate(inode); 3121 if (ret) 3122 ext4_std_error(inode->i_sb, ret); 3123 inode_unlock(inode); 3124 nr_truncates++; 3125 } else { 3126 if (test_opt(sb, DEBUG)) 3127 ext4_msg(sb, KERN_DEBUG, 3128 "%s: deleting unreferenced inode %lu", 3129 __func__, inode->i_ino); 3130 jbd_debug(2, "deleting unreferenced inode %lu\n", 3131 inode->i_ino); 3132 nr_orphans++; 3133 } 3134 iput(inode); /* The delete magic happens here! */ 3135 } 3136 3137 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 3138 3139 if (nr_orphans) 3140 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 3141 PLURAL(nr_orphans)); 3142 if (nr_truncates) 3143 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 3144 PLURAL(nr_truncates)); 3145 #ifdef CONFIG_QUOTA 3146 /* Turn off quotas if they were enabled for orphan cleanup */ 3147 if (quota_update) { 3148 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3149 if (sb_dqopt(sb)->files[i]) 3150 dquot_quota_off(sb, i); 3151 } 3152 } 3153 #endif 3154 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 3155 } 3156 3157 /* 3158 * Maximal extent format file size. 3159 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3160 * extent format containers, within a sector_t, and within i_blocks 3161 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3162 * so that won't be a limiting factor. 3163 * 3164 * However there is other limiting factor. We do store extents in the form 3165 * of starting block and length, hence the resulting length of the extent 3166 * covering maximum file size must fit into on-disk format containers as 3167 * well. Given that length is always by 1 unit bigger than max unit (because 3168 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3169 * 3170 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3171 */ 3172 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3173 { 3174 loff_t res; 3175 loff_t upper_limit = MAX_LFS_FILESIZE; 3176 3177 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3178 3179 if (!has_huge_files) { 3180 upper_limit = (1LL << 32) - 1; 3181 3182 /* total blocks in file system block size */ 3183 upper_limit >>= (blkbits - 9); 3184 upper_limit <<= blkbits; 3185 } 3186 3187 /* 3188 * 32-bit extent-start container, ee_block. We lower the maxbytes 3189 * by one fs block, so ee_len can cover the extent of maximum file 3190 * size 3191 */ 3192 res = (1LL << 32) - 1; 3193 res <<= blkbits; 3194 3195 /* Sanity check against vm- & vfs- imposed limits */ 3196 if (res > upper_limit) 3197 res = upper_limit; 3198 3199 return res; 3200 } 3201 3202 /* 3203 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3204 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3205 * We need to be 1 filesystem block less than the 2^48 sector limit. 3206 */ 3207 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3208 { 3209 loff_t res = EXT4_NDIR_BLOCKS; 3210 int meta_blocks; 3211 loff_t upper_limit; 3212 /* This is calculated to be the largest file size for a dense, block 3213 * mapped file such that the file's total number of 512-byte sectors, 3214 * including data and all indirect blocks, does not exceed (2^48 - 1). 3215 * 3216 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3217 * number of 512-byte sectors of the file. 3218 */ 3219 3220 if (!has_huge_files) { 3221 /* 3222 * !has_huge_files or implies that the inode i_block field 3223 * represents total file blocks in 2^32 512-byte sectors == 3224 * size of vfs inode i_blocks * 8 3225 */ 3226 upper_limit = (1LL << 32) - 1; 3227 3228 /* total blocks in file system block size */ 3229 upper_limit >>= (bits - 9); 3230 3231 } else { 3232 /* 3233 * We use 48 bit ext4_inode i_blocks 3234 * With EXT4_HUGE_FILE_FL set the i_blocks 3235 * represent total number of blocks in 3236 * file system block size 3237 */ 3238 upper_limit = (1LL << 48) - 1; 3239 3240 } 3241 3242 /* indirect blocks */ 3243 meta_blocks = 1; 3244 /* double indirect blocks */ 3245 meta_blocks += 1 + (1LL << (bits-2)); 3246 /* tripple indirect blocks */ 3247 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 3248 3249 upper_limit -= meta_blocks; 3250 upper_limit <<= bits; 3251 3252 res += 1LL << (bits-2); 3253 res += 1LL << (2*(bits-2)); 3254 res += 1LL << (3*(bits-2)); 3255 res <<= bits; 3256 if (res > upper_limit) 3257 res = upper_limit; 3258 3259 if (res > MAX_LFS_FILESIZE) 3260 res = MAX_LFS_FILESIZE; 3261 3262 return res; 3263 } 3264 3265 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3266 ext4_fsblk_t logical_sb_block, int nr) 3267 { 3268 struct ext4_sb_info *sbi = EXT4_SB(sb); 3269 ext4_group_t bg, first_meta_bg; 3270 int has_super = 0; 3271 3272 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3273 3274 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3275 return logical_sb_block + nr + 1; 3276 bg = sbi->s_desc_per_block * nr; 3277 if (ext4_bg_has_super(sb, bg)) 3278 has_super = 1; 3279 3280 /* 3281 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3282 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3283 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3284 * compensate. 3285 */ 3286 if (sb->s_blocksize == 1024 && nr == 0 && 3287 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3288 has_super++; 3289 3290 return (has_super + ext4_group_first_block_no(sb, bg)); 3291 } 3292 3293 /** 3294 * ext4_get_stripe_size: Get the stripe size. 3295 * @sbi: In memory super block info 3296 * 3297 * If we have specified it via mount option, then 3298 * use the mount option value. If the value specified at mount time is 3299 * greater than the blocks per group use the super block value. 3300 * If the super block value is greater than blocks per group return 0. 3301 * Allocator needs it be less than blocks per group. 3302 * 3303 */ 3304 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3305 { 3306 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3307 unsigned long stripe_width = 3308 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3309 int ret; 3310 3311 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3312 ret = sbi->s_stripe; 3313 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3314 ret = stripe_width; 3315 else if (stride && stride <= sbi->s_blocks_per_group) 3316 ret = stride; 3317 else 3318 ret = 0; 3319 3320 /* 3321 * If the stripe width is 1, this makes no sense and 3322 * we set it to 0 to turn off stripe handling code. 3323 */ 3324 if (ret <= 1) 3325 ret = 0; 3326 3327 return ret; 3328 } 3329 3330 /* 3331 * Check whether this filesystem can be mounted based on 3332 * the features present and the RDONLY/RDWR mount requested. 3333 * Returns 1 if this filesystem can be mounted as requested, 3334 * 0 if it cannot be. 3335 */ 3336 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 3337 { 3338 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3339 ext4_msg(sb, KERN_ERR, 3340 "Couldn't mount because of " 3341 "unsupported optional features (%x)", 3342 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3343 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3344 return 0; 3345 } 3346 3347 #ifndef CONFIG_UNICODE 3348 if (ext4_has_feature_casefold(sb)) { 3349 ext4_msg(sb, KERN_ERR, 3350 "Filesystem with casefold feature cannot be " 3351 "mounted without CONFIG_UNICODE"); 3352 return 0; 3353 } 3354 #endif 3355 3356 if (readonly) 3357 return 1; 3358 3359 if (ext4_has_feature_readonly(sb)) { 3360 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3361 sb->s_flags |= SB_RDONLY; 3362 return 1; 3363 } 3364 3365 /* Check that feature set is OK for a read-write mount */ 3366 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3367 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3368 "unsupported optional features (%x)", 3369 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3370 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3371 return 0; 3372 } 3373 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3374 ext4_msg(sb, KERN_ERR, 3375 "Can't support bigalloc feature without " 3376 "extents feature\n"); 3377 return 0; 3378 } 3379 3380 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3381 if (!readonly && (ext4_has_feature_quota(sb) || 3382 ext4_has_feature_project(sb))) { 3383 ext4_msg(sb, KERN_ERR, 3384 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3385 return 0; 3386 } 3387 #endif /* CONFIG_QUOTA */ 3388 return 1; 3389 } 3390 3391 /* 3392 * This function is called once a day if we have errors logged 3393 * on the file system 3394 */ 3395 static void print_daily_error_info(struct timer_list *t) 3396 { 3397 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3398 struct super_block *sb = sbi->s_sb; 3399 struct ext4_super_block *es = sbi->s_es; 3400 3401 if (es->s_error_count) 3402 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3403 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3404 le32_to_cpu(es->s_error_count)); 3405 if (es->s_first_error_time) { 3406 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3407 sb->s_id, 3408 ext4_get_tstamp(es, s_first_error_time), 3409 (int) sizeof(es->s_first_error_func), 3410 es->s_first_error_func, 3411 le32_to_cpu(es->s_first_error_line)); 3412 if (es->s_first_error_ino) 3413 printk(KERN_CONT ": inode %u", 3414 le32_to_cpu(es->s_first_error_ino)); 3415 if (es->s_first_error_block) 3416 printk(KERN_CONT ": block %llu", (unsigned long long) 3417 le64_to_cpu(es->s_first_error_block)); 3418 printk(KERN_CONT "\n"); 3419 } 3420 if (es->s_last_error_time) { 3421 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3422 sb->s_id, 3423 ext4_get_tstamp(es, s_last_error_time), 3424 (int) sizeof(es->s_last_error_func), 3425 es->s_last_error_func, 3426 le32_to_cpu(es->s_last_error_line)); 3427 if (es->s_last_error_ino) 3428 printk(KERN_CONT ": inode %u", 3429 le32_to_cpu(es->s_last_error_ino)); 3430 if (es->s_last_error_block) 3431 printk(KERN_CONT ": block %llu", (unsigned long long) 3432 le64_to_cpu(es->s_last_error_block)); 3433 printk(KERN_CONT "\n"); 3434 } 3435 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3436 } 3437 3438 /* Find next suitable group and run ext4_init_inode_table */ 3439 static int ext4_run_li_request(struct ext4_li_request *elr) 3440 { 3441 struct ext4_group_desc *gdp = NULL; 3442 struct super_block *sb = elr->lr_super; 3443 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3444 ext4_group_t group = elr->lr_next_group; 3445 unsigned long timeout = 0; 3446 unsigned int prefetch_ios = 0; 3447 int ret = 0; 3448 3449 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3450 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3451 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3452 if (prefetch_ios) 3453 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3454 prefetch_ios); 3455 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3456 prefetch_ios); 3457 if (group >= elr->lr_next_group) { 3458 ret = 1; 3459 if (elr->lr_first_not_zeroed != ngroups && 3460 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3461 elr->lr_next_group = elr->lr_first_not_zeroed; 3462 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3463 ret = 0; 3464 } 3465 } 3466 return ret; 3467 } 3468 3469 for (; group < ngroups; group++) { 3470 gdp = ext4_get_group_desc(sb, group, NULL); 3471 if (!gdp) { 3472 ret = 1; 3473 break; 3474 } 3475 3476 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3477 break; 3478 } 3479 3480 if (group >= ngroups) 3481 ret = 1; 3482 3483 if (!ret) { 3484 timeout = jiffies; 3485 ret = ext4_init_inode_table(sb, group, 3486 elr->lr_timeout ? 0 : 1); 3487 trace_ext4_lazy_itable_init(sb, group); 3488 if (elr->lr_timeout == 0) { 3489 timeout = (jiffies - timeout) * 3490 EXT4_SB(elr->lr_super)->s_li_wait_mult; 3491 elr->lr_timeout = timeout; 3492 } 3493 elr->lr_next_sched = jiffies + elr->lr_timeout; 3494 elr->lr_next_group = group + 1; 3495 } 3496 return ret; 3497 } 3498 3499 /* 3500 * Remove lr_request from the list_request and free the 3501 * request structure. Should be called with li_list_mtx held 3502 */ 3503 static void ext4_remove_li_request(struct ext4_li_request *elr) 3504 { 3505 if (!elr) 3506 return; 3507 3508 list_del(&elr->lr_request); 3509 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3510 kfree(elr); 3511 } 3512 3513 static void ext4_unregister_li_request(struct super_block *sb) 3514 { 3515 mutex_lock(&ext4_li_mtx); 3516 if (!ext4_li_info) { 3517 mutex_unlock(&ext4_li_mtx); 3518 return; 3519 } 3520 3521 mutex_lock(&ext4_li_info->li_list_mtx); 3522 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3523 mutex_unlock(&ext4_li_info->li_list_mtx); 3524 mutex_unlock(&ext4_li_mtx); 3525 } 3526 3527 static struct task_struct *ext4_lazyinit_task; 3528 3529 /* 3530 * This is the function where ext4lazyinit thread lives. It walks 3531 * through the request list searching for next scheduled filesystem. 3532 * When such a fs is found, run the lazy initialization request 3533 * (ext4_rn_li_request) and keep track of the time spend in this 3534 * function. Based on that time we compute next schedule time of 3535 * the request. When walking through the list is complete, compute 3536 * next waking time and put itself into sleep. 3537 */ 3538 static int ext4_lazyinit_thread(void *arg) 3539 { 3540 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3541 struct list_head *pos, *n; 3542 struct ext4_li_request *elr; 3543 unsigned long next_wakeup, cur; 3544 3545 BUG_ON(NULL == eli); 3546 3547 cont_thread: 3548 while (true) { 3549 next_wakeup = MAX_JIFFY_OFFSET; 3550 3551 mutex_lock(&eli->li_list_mtx); 3552 if (list_empty(&eli->li_request_list)) { 3553 mutex_unlock(&eli->li_list_mtx); 3554 goto exit_thread; 3555 } 3556 list_for_each_safe(pos, n, &eli->li_request_list) { 3557 int err = 0; 3558 int progress = 0; 3559 elr = list_entry(pos, struct ext4_li_request, 3560 lr_request); 3561 3562 if (time_before(jiffies, elr->lr_next_sched)) { 3563 if (time_before(elr->lr_next_sched, next_wakeup)) 3564 next_wakeup = elr->lr_next_sched; 3565 continue; 3566 } 3567 if (down_read_trylock(&elr->lr_super->s_umount)) { 3568 if (sb_start_write_trylock(elr->lr_super)) { 3569 progress = 1; 3570 /* 3571 * We hold sb->s_umount, sb can not 3572 * be removed from the list, it is 3573 * now safe to drop li_list_mtx 3574 */ 3575 mutex_unlock(&eli->li_list_mtx); 3576 err = ext4_run_li_request(elr); 3577 sb_end_write(elr->lr_super); 3578 mutex_lock(&eli->li_list_mtx); 3579 n = pos->next; 3580 } 3581 up_read((&elr->lr_super->s_umount)); 3582 } 3583 /* error, remove the lazy_init job */ 3584 if (err) { 3585 ext4_remove_li_request(elr); 3586 continue; 3587 } 3588 if (!progress) { 3589 elr->lr_next_sched = jiffies + 3590 (prandom_u32() 3591 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3592 } 3593 if (time_before(elr->lr_next_sched, next_wakeup)) 3594 next_wakeup = elr->lr_next_sched; 3595 } 3596 mutex_unlock(&eli->li_list_mtx); 3597 3598 try_to_freeze(); 3599 3600 cur = jiffies; 3601 if ((time_after_eq(cur, next_wakeup)) || 3602 (MAX_JIFFY_OFFSET == next_wakeup)) { 3603 cond_resched(); 3604 continue; 3605 } 3606 3607 schedule_timeout_interruptible(next_wakeup - cur); 3608 3609 if (kthread_should_stop()) { 3610 ext4_clear_request_list(); 3611 goto exit_thread; 3612 } 3613 } 3614 3615 exit_thread: 3616 /* 3617 * It looks like the request list is empty, but we need 3618 * to check it under the li_list_mtx lock, to prevent any 3619 * additions into it, and of course we should lock ext4_li_mtx 3620 * to atomically free the list and ext4_li_info, because at 3621 * this point another ext4 filesystem could be registering 3622 * new one. 3623 */ 3624 mutex_lock(&ext4_li_mtx); 3625 mutex_lock(&eli->li_list_mtx); 3626 if (!list_empty(&eli->li_request_list)) { 3627 mutex_unlock(&eli->li_list_mtx); 3628 mutex_unlock(&ext4_li_mtx); 3629 goto cont_thread; 3630 } 3631 mutex_unlock(&eli->li_list_mtx); 3632 kfree(ext4_li_info); 3633 ext4_li_info = NULL; 3634 mutex_unlock(&ext4_li_mtx); 3635 3636 return 0; 3637 } 3638 3639 static void ext4_clear_request_list(void) 3640 { 3641 struct list_head *pos, *n; 3642 struct ext4_li_request *elr; 3643 3644 mutex_lock(&ext4_li_info->li_list_mtx); 3645 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3646 elr = list_entry(pos, struct ext4_li_request, 3647 lr_request); 3648 ext4_remove_li_request(elr); 3649 } 3650 mutex_unlock(&ext4_li_info->li_list_mtx); 3651 } 3652 3653 static int ext4_run_lazyinit_thread(void) 3654 { 3655 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3656 ext4_li_info, "ext4lazyinit"); 3657 if (IS_ERR(ext4_lazyinit_task)) { 3658 int err = PTR_ERR(ext4_lazyinit_task); 3659 ext4_clear_request_list(); 3660 kfree(ext4_li_info); 3661 ext4_li_info = NULL; 3662 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3663 "initialization thread\n", 3664 err); 3665 return err; 3666 } 3667 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3668 return 0; 3669 } 3670 3671 /* 3672 * Check whether it make sense to run itable init. thread or not. 3673 * If there is at least one uninitialized inode table, return 3674 * corresponding group number, else the loop goes through all 3675 * groups and return total number of groups. 3676 */ 3677 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3678 { 3679 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3680 struct ext4_group_desc *gdp = NULL; 3681 3682 if (!ext4_has_group_desc_csum(sb)) 3683 return ngroups; 3684 3685 for (group = 0; group < ngroups; group++) { 3686 gdp = ext4_get_group_desc(sb, group, NULL); 3687 if (!gdp) 3688 continue; 3689 3690 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3691 break; 3692 } 3693 3694 return group; 3695 } 3696 3697 static int ext4_li_info_new(void) 3698 { 3699 struct ext4_lazy_init *eli = NULL; 3700 3701 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3702 if (!eli) 3703 return -ENOMEM; 3704 3705 INIT_LIST_HEAD(&eli->li_request_list); 3706 mutex_init(&eli->li_list_mtx); 3707 3708 eli->li_state |= EXT4_LAZYINIT_QUIT; 3709 3710 ext4_li_info = eli; 3711 3712 return 0; 3713 } 3714 3715 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3716 ext4_group_t start) 3717 { 3718 struct ext4_li_request *elr; 3719 3720 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3721 if (!elr) 3722 return NULL; 3723 3724 elr->lr_super = sb; 3725 elr->lr_first_not_zeroed = start; 3726 if (test_opt(sb, PREFETCH_BLOCK_BITMAPS)) 3727 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3728 else { 3729 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3730 elr->lr_next_group = start; 3731 } 3732 3733 /* 3734 * Randomize first schedule time of the request to 3735 * spread the inode table initialization requests 3736 * better. 3737 */ 3738 elr->lr_next_sched = jiffies + (prandom_u32() % 3739 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3740 return elr; 3741 } 3742 3743 int ext4_register_li_request(struct super_block *sb, 3744 ext4_group_t first_not_zeroed) 3745 { 3746 struct ext4_sb_info *sbi = EXT4_SB(sb); 3747 struct ext4_li_request *elr = NULL; 3748 ext4_group_t ngroups = sbi->s_groups_count; 3749 int ret = 0; 3750 3751 mutex_lock(&ext4_li_mtx); 3752 if (sbi->s_li_request != NULL) { 3753 /* 3754 * Reset timeout so it can be computed again, because 3755 * s_li_wait_mult might have changed. 3756 */ 3757 sbi->s_li_request->lr_timeout = 0; 3758 goto out; 3759 } 3760 3761 if (!test_opt(sb, PREFETCH_BLOCK_BITMAPS) && 3762 (first_not_zeroed == ngroups || sb_rdonly(sb) || 3763 !test_opt(sb, INIT_INODE_TABLE))) 3764 goto out; 3765 3766 elr = ext4_li_request_new(sb, first_not_zeroed); 3767 if (!elr) { 3768 ret = -ENOMEM; 3769 goto out; 3770 } 3771 3772 if (NULL == ext4_li_info) { 3773 ret = ext4_li_info_new(); 3774 if (ret) 3775 goto out; 3776 } 3777 3778 mutex_lock(&ext4_li_info->li_list_mtx); 3779 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3780 mutex_unlock(&ext4_li_info->li_list_mtx); 3781 3782 sbi->s_li_request = elr; 3783 /* 3784 * set elr to NULL here since it has been inserted to 3785 * the request_list and the removal and free of it is 3786 * handled by ext4_clear_request_list from now on. 3787 */ 3788 elr = NULL; 3789 3790 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3791 ret = ext4_run_lazyinit_thread(); 3792 if (ret) 3793 goto out; 3794 } 3795 out: 3796 mutex_unlock(&ext4_li_mtx); 3797 if (ret) 3798 kfree(elr); 3799 return ret; 3800 } 3801 3802 /* 3803 * We do not need to lock anything since this is called on 3804 * module unload. 3805 */ 3806 static void ext4_destroy_lazyinit_thread(void) 3807 { 3808 /* 3809 * If thread exited earlier 3810 * there's nothing to be done. 3811 */ 3812 if (!ext4_li_info || !ext4_lazyinit_task) 3813 return; 3814 3815 kthread_stop(ext4_lazyinit_task); 3816 } 3817 3818 static int set_journal_csum_feature_set(struct super_block *sb) 3819 { 3820 int ret = 1; 3821 int compat, incompat; 3822 struct ext4_sb_info *sbi = EXT4_SB(sb); 3823 3824 if (ext4_has_metadata_csum(sb)) { 3825 /* journal checksum v3 */ 3826 compat = 0; 3827 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3828 } else { 3829 /* journal checksum v1 */ 3830 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3831 incompat = 0; 3832 } 3833 3834 jbd2_journal_clear_features(sbi->s_journal, 3835 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3836 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3837 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3838 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3839 ret = jbd2_journal_set_features(sbi->s_journal, 3840 compat, 0, 3841 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3842 incompat); 3843 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3844 ret = jbd2_journal_set_features(sbi->s_journal, 3845 compat, 0, 3846 incompat); 3847 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3848 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3849 } else { 3850 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3851 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3852 } 3853 3854 return ret; 3855 } 3856 3857 /* 3858 * Note: calculating the overhead so we can be compatible with 3859 * historical BSD practice is quite difficult in the face of 3860 * clusters/bigalloc. This is because multiple metadata blocks from 3861 * different block group can end up in the same allocation cluster. 3862 * Calculating the exact overhead in the face of clustered allocation 3863 * requires either O(all block bitmaps) in memory or O(number of block 3864 * groups**2) in time. We will still calculate the superblock for 3865 * older file systems --- and if we come across with a bigalloc file 3866 * system with zero in s_overhead_clusters the estimate will be close to 3867 * correct especially for very large cluster sizes --- but for newer 3868 * file systems, it's better to calculate this figure once at mkfs 3869 * time, and store it in the superblock. If the superblock value is 3870 * present (even for non-bigalloc file systems), we will use it. 3871 */ 3872 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3873 char *buf) 3874 { 3875 struct ext4_sb_info *sbi = EXT4_SB(sb); 3876 struct ext4_group_desc *gdp; 3877 ext4_fsblk_t first_block, last_block, b; 3878 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3879 int s, j, count = 0; 3880 3881 if (!ext4_has_feature_bigalloc(sb)) 3882 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3883 sbi->s_itb_per_group + 2); 3884 3885 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3886 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3887 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3888 for (i = 0; i < ngroups; i++) { 3889 gdp = ext4_get_group_desc(sb, i, NULL); 3890 b = ext4_block_bitmap(sb, gdp); 3891 if (b >= first_block && b <= last_block) { 3892 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3893 count++; 3894 } 3895 b = ext4_inode_bitmap(sb, gdp); 3896 if (b >= first_block && b <= last_block) { 3897 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3898 count++; 3899 } 3900 b = ext4_inode_table(sb, gdp); 3901 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3902 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3903 int c = EXT4_B2C(sbi, b - first_block); 3904 ext4_set_bit(c, buf); 3905 count++; 3906 } 3907 if (i != grp) 3908 continue; 3909 s = 0; 3910 if (ext4_bg_has_super(sb, grp)) { 3911 ext4_set_bit(s++, buf); 3912 count++; 3913 } 3914 j = ext4_bg_num_gdb(sb, grp); 3915 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3916 ext4_error(sb, "Invalid number of block group " 3917 "descriptor blocks: %d", j); 3918 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3919 } 3920 count += j; 3921 for (; j > 0; j--) 3922 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3923 } 3924 if (!count) 3925 return 0; 3926 return EXT4_CLUSTERS_PER_GROUP(sb) - 3927 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3928 } 3929 3930 /* 3931 * Compute the overhead and stash it in sbi->s_overhead 3932 */ 3933 int ext4_calculate_overhead(struct super_block *sb) 3934 { 3935 struct ext4_sb_info *sbi = EXT4_SB(sb); 3936 struct ext4_super_block *es = sbi->s_es; 3937 struct inode *j_inode; 3938 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3939 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3940 ext4_fsblk_t overhead = 0; 3941 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3942 3943 if (!buf) 3944 return -ENOMEM; 3945 3946 /* 3947 * Compute the overhead (FS structures). This is constant 3948 * for a given filesystem unless the number of block groups 3949 * changes so we cache the previous value until it does. 3950 */ 3951 3952 /* 3953 * All of the blocks before first_data_block are overhead 3954 */ 3955 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3956 3957 /* 3958 * Add the overhead found in each block group 3959 */ 3960 for (i = 0; i < ngroups; i++) { 3961 int blks; 3962 3963 blks = count_overhead(sb, i, buf); 3964 overhead += blks; 3965 if (blks) 3966 memset(buf, 0, PAGE_SIZE); 3967 cond_resched(); 3968 } 3969 3970 /* 3971 * Add the internal journal blocks whether the journal has been 3972 * loaded or not 3973 */ 3974 if (sbi->s_journal && !sbi->s_journal_bdev) 3975 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 3976 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 3977 /* j_inum for internal journal is non-zero */ 3978 j_inode = ext4_get_journal_inode(sb, j_inum); 3979 if (j_inode) { 3980 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3981 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3982 iput(j_inode); 3983 } else { 3984 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3985 } 3986 } 3987 sbi->s_overhead = overhead; 3988 smp_wmb(); 3989 free_page((unsigned long) buf); 3990 return 0; 3991 } 3992 3993 static void ext4_set_resv_clusters(struct super_block *sb) 3994 { 3995 ext4_fsblk_t resv_clusters; 3996 struct ext4_sb_info *sbi = EXT4_SB(sb); 3997 3998 /* 3999 * There's no need to reserve anything when we aren't using extents. 4000 * The space estimates are exact, there are no unwritten extents, 4001 * hole punching doesn't need new metadata... This is needed especially 4002 * to keep ext2/3 backward compatibility. 4003 */ 4004 if (!ext4_has_feature_extents(sb)) 4005 return; 4006 /* 4007 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4008 * This should cover the situations where we can not afford to run 4009 * out of space like for example punch hole, or converting 4010 * unwritten extents in delalloc path. In most cases such 4011 * allocation would require 1, or 2 blocks, higher numbers are 4012 * very rare. 4013 */ 4014 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4015 sbi->s_cluster_bits); 4016 4017 do_div(resv_clusters, 50); 4018 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4019 4020 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4021 } 4022 4023 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 4024 { 4025 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 4026 char *orig_data = kstrdup(data, GFP_KERNEL); 4027 struct buffer_head *bh, **group_desc; 4028 struct ext4_super_block *es = NULL; 4029 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4030 struct flex_groups **flex_groups; 4031 ext4_fsblk_t block; 4032 ext4_fsblk_t sb_block = get_sb_block(&data); 4033 ext4_fsblk_t logical_sb_block; 4034 unsigned long offset = 0; 4035 unsigned long journal_devnum = 0; 4036 unsigned long def_mount_opts; 4037 struct inode *root; 4038 const char *descr; 4039 int ret = -ENOMEM; 4040 int blocksize, clustersize; 4041 unsigned int db_count; 4042 unsigned int i; 4043 int needs_recovery, has_huge_files; 4044 __u64 blocks_count; 4045 int err = 0; 4046 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 4047 ext4_group_t first_not_zeroed; 4048 4049 if ((data && !orig_data) || !sbi) 4050 goto out_free_base; 4051 4052 sbi->s_daxdev = dax_dev; 4053 sbi->s_blockgroup_lock = 4054 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4055 if (!sbi->s_blockgroup_lock) 4056 goto out_free_base; 4057 4058 sb->s_fs_info = sbi; 4059 sbi->s_sb = sb; 4060 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 4061 sbi->s_sb_block = sb_block; 4062 if (sb->s_bdev->bd_part) 4063 sbi->s_sectors_written_start = 4064 part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]); 4065 4066 /* Cleanup superblock name */ 4067 strreplace(sb->s_id, '/', '!'); 4068 4069 /* -EINVAL is default */ 4070 ret = -EINVAL; 4071 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4072 if (!blocksize) { 4073 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4074 goto out_fail; 4075 } 4076 4077 /* 4078 * The ext4 superblock will not be buffer aligned for other than 1kB 4079 * block sizes. We need to calculate the offset from buffer start. 4080 */ 4081 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4082 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4083 offset = do_div(logical_sb_block, blocksize); 4084 } else { 4085 logical_sb_block = sb_block; 4086 } 4087 4088 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4089 if (IS_ERR(bh)) { 4090 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4091 ret = PTR_ERR(bh); 4092 bh = NULL; 4093 goto out_fail; 4094 } 4095 /* 4096 * Note: s_es must be initialized as soon as possible because 4097 * some ext4 macro-instructions depend on its value 4098 */ 4099 es = (struct ext4_super_block *) (bh->b_data + offset); 4100 sbi->s_es = es; 4101 sb->s_magic = le16_to_cpu(es->s_magic); 4102 if (sb->s_magic != EXT4_SUPER_MAGIC) 4103 goto cantfind_ext4; 4104 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 4105 4106 /* Warn if metadata_csum and gdt_csum are both set. */ 4107 if (ext4_has_feature_metadata_csum(sb) && 4108 ext4_has_feature_gdt_csum(sb)) 4109 ext4_warning(sb, "metadata_csum and uninit_bg are " 4110 "redundant flags; please run fsck."); 4111 4112 /* Check for a known checksum algorithm */ 4113 if (!ext4_verify_csum_type(sb, es)) { 4114 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4115 "unknown checksum algorithm."); 4116 silent = 1; 4117 goto cantfind_ext4; 4118 } 4119 4120 /* Load the checksum driver */ 4121 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4122 if (IS_ERR(sbi->s_chksum_driver)) { 4123 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4124 ret = PTR_ERR(sbi->s_chksum_driver); 4125 sbi->s_chksum_driver = NULL; 4126 goto failed_mount; 4127 } 4128 4129 /* Check superblock checksum */ 4130 if (!ext4_superblock_csum_verify(sb, es)) { 4131 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4132 "invalid superblock checksum. Run e2fsck?"); 4133 silent = 1; 4134 ret = -EFSBADCRC; 4135 goto cantfind_ext4; 4136 } 4137 4138 /* Precompute checksum seed for all metadata */ 4139 if (ext4_has_feature_csum_seed(sb)) 4140 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4141 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4142 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4143 sizeof(es->s_uuid)); 4144 4145 /* Set defaults before we parse the mount options */ 4146 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4147 set_opt(sb, INIT_INODE_TABLE); 4148 if (def_mount_opts & EXT4_DEFM_DEBUG) 4149 set_opt(sb, DEBUG); 4150 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4151 set_opt(sb, GRPID); 4152 if (def_mount_opts & EXT4_DEFM_UID16) 4153 set_opt(sb, NO_UID32); 4154 /* xattr user namespace & acls are now defaulted on */ 4155 set_opt(sb, XATTR_USER); 4156 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4157 set_opt(sb, POSIX_ACL); 4158 #endif 4159 if (ext4_has_feature_fast_commit(sb)) 4160 set_opt2(sb, JOURNAL_FAST_COMMIT); 4161 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4162 if (ext4_has_metadata_csum(sb)) 4163 set_opt(sb, JOURNAL_CHECKSUM); 4164 4165 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4166 set_opt(sb, JOURNAL_DATA); 4167 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4168 set_opt(sb, ORDERED_DATA); 4169 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4170 set_opt(sb, WRITEBACK_DATA); 4171 4172 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 4173 set_opt(sb, ERRORS_PANIC); 4174 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 4175 set_opt(sb, ERRORS_CONT); 4176 else 4177 set_opt(sb, ERRORS_RO); 4178 /* block_validity enabled by default; disable with noblock_validity */ 4179 set_opt(sb, BLOCK_VALIDITY); 4180 if (def_mount_opts & EXT4_DEFM_DISCARD) 4181 set_opt(sb, DISCARD); 4182 4183 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 4184 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 4185 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 4186 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 4187 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 4188 4189 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4190 set_opt(sb, BARRIER); 4191 4192 /* 4193 * enable delayed allocation by default 4194 * Use -o nodelalloc to turn it off 4195 */ 4196 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4197 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4198 set_opt(sb, DELALLOC); 4199 4200 /* 4201 * set default s_li_wait_mult for lazyinit, for the case there is 4202 * no mount option specified. 4203 */ 4204 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 4205 4206 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4207 4208 if (blocksize == PAGE_SIZE) 4209 set_opt(sb, DIOREAD_NOLOCK); 4210 4211 if (blocksize < EXT4_MIN_BLOCK_SIZE || 4212 blocksize > EXT4_MAX_BLOCK_SIZE) { 4213 ext4_msg(sb, KERN_ERR, 4214 "Unsupported filesystem blocksize %d (%d log_block_size)", 4215 blocksize, le32_to_cpu(es->s_log_block_size)); 4216 goto failed_mount; 4217 } 4218 4219 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4220 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4221 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4222 } else { 4223 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4224 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4225 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4226 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4227 sbi->s_first_ino); 4228 goto failed_mount; 4229 } 4230 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4231 (!is_power_of_2(sbi->s_inode_size)) || 4232 (sbi->s_inode_size > blocksize)) { 4233 ext4_msg(sb, KERN_ERR, 4234 "unsupported inode size: %d", 4235 sbi->s_inode_size); 4236 ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize); 4237 goto failed_mount; 4238 } 4239 /* 4240 * i_atime_extra is the last extra field available for 4241 * [acm]times in struct ext4_inode. Checking for that 4242 * field should suffice to ensure we have extra space 4243 * for all three. 4244 */ 4245 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4246 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4247 sb->s_time_gran = 1; 4248 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4249 } else { 4250 sb->s_time_gran = NSEC_PER_SEC; 4251 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4252 } 4253 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4254 } 4255 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4256 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4257 EXT4_GOOD_OLD_INODE_SIZE; 4258 if (ext4_has_feature_extra_isize(sb)) { 4259 unsigned v, max = (sbi->s_inode_size - 4260 EXT4_GOOD_OLD_INODE_SIZE); 4261 4262 v = le16_to_cpu(es->s_want_extra_isize); 4263 if (v > max) { 4264 ext4_msg(sb, KERN_ERR, 4265 "bad s_want_extra_isize: %d", v); 4266 goto failed_mount; 4267 } 4268 if (sbi->s_want_extra_isize < v) 4269 sbi->s_want_extra_isize = v; 4270 4271 v = le16_to_cpu(es->s_min_extra_isize); 4272 if (v > max) { 4273 ext4_msg(sb, KERN_ERR, 4274 "bad s_min_extra_isize: %d", v); 4275 goto failed_mount; 4276 } 4277 if (sbi->s_want_extra_isize < v) 4278 sbi->s_want_extra_isize = v; 4279 } 4280 } 4281 4282 if (sbi->s_es->s_mount_opts[0]) { 4283 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 4284 sizeof(sbi->s_es->s_mount_opts), 4285 GFP_KERNEL); 4286 if (!s_mount_opts) 4287 goto failed_mount; 4288 if (!parse_options(s_mount_opts, sb, &journal_devnum, 4289 &journal_ioprio, 0)) { 4290 ext4_msg(sb, KERN_WARNING, 4291 "failed to parse options in superblock: %s", 4292 s_mount_opts); 4293 } 4294 kfree(s_mount_opts); 4295 } 4296 sbi->s_def_mount_opt = sbi->s_mount_opt; 4297 if (!parse_options((char *) data, sb, &journal_devnum, 4298 &journal_ioprio, 0)) 4299 goto failed_mount; 4300 4301 #ifdef CONFIG_UNICODE 4302 if (ext4_has_feature_casefold(sb) && !sb->s_encoding) { 4303 const struct ext4_sb_encodings *encoding_info; 4304 struct unicode_map *encoding; 4305 __u16 encoding_flags; 4306 4307 if (ext4_has_feature_encrypt(sb)) { 4308 ext4_msg(sb, KERN_ERR, 4309 "Can't mount with encoding and encryption"); 4310 goto failed_mount; 4311 } 4312 4313 if (ext4_sb_read_encoding(es, &encoding_info, 4314 &encoding_flags)) { 4315 ext4_msg(sb, KERN_ERR, 4316 "Encoding requested by superblock is unknown"); 4317 goto failed_mount; 4318 } 4319 4320 encoding = utf8_load(encoding_info->version); 4321 if (IS_ERR(encoding)) { 4322 ext4_msg(sb, KERN_ERR, 4323 "can't mount with superblock charset: %s-%s " 4324 "not supported by the kernel. flags: 0x%x.", 4325 encoding_info->name, encoding_info->version, 4326 encoding_flags); 4327 goto failed_mount; 4328 } 4329 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4330 "%s-%s with flags 0x%hx", encoding_info->name, 4331 encoding_info->version?:"\b", encoding_flags); 4332 4333 sb->s_encoding = encoding; 4334 sb->s_encoding_flags = encoding_flags; 4335 } 4336 #endif 4337 4338 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4339 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n"); 4340 /* can't mount with both data=journal and dioread_nolock. */ 4341 clear_opt(sb, DIOREAD_NOLOCK); 4342 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4343 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4344 ext4_msg(sb, KERN_ERR, "can't mount with " 4345 "both data=journal and delalloc"); 4346 goto failed_mount; 4347 } 4348 if (test_opt(sb, DAX_ALWAYS)) { 4349 ext4_msg(sb, KERN_ERR, "can't mount with " 4350 "both data=journal and dax"); 4351 goto failed_mount; 4352 } 4353 if (ext4_has_feature_encrypt(sb)) { 4354 ext4_msg(sb, KERN_WARNING, 4355 "encrypted files will use data=ordered " 4356 "instead of data journaling mode"); 4357 } 4358 if (test_opt(sb, DELALLOC)) 4359 clear_opt(sb, DELALLOC); 4360 } else { 4361 sb->s_iflags |= SB_I_CGROUPWB; 4362 } 4363 4364 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 4365 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 4366 4367 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4368 (ext4_has_compat_features(sb) || 4369 ext4_has_ro_compat_features(sb) || 4370 ext4_has_incompat_features(sb))) 4371 ext4_msg(sb, KERN_WARNING, 4372 "feature flags set on rev 0 fs, " 4373 "running e2fsck is recommended"); 4374 4375 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4376 set_opt2(sb, HURD_COMPAT); 4377 if (ext4_has_feature_64bit(sb)) { 4378 ext4_msg(sb, KERN_ERR, 4379 "The Hurd can't support 64-bit file systems"); 4380 goto failed_mount; 4381 } 4382 4383 /* 4384 * ea_inode feature uses l_i_version field which is not 4385 * available in HURD_COMPAT mode. 4386 */ 4387 if (ext4_has_feature_ea_inode(sb)) { 4388 ext4_msg(sb, KERN_ERR, 4389 "ea_inode feature is not supported for Hurd"); 4390 goto failed_mount; 4391 } 4392 } 4393 4394 if (IS_EXT2_SB(sb)) { 4395 if (ext2_feature_set_ok(sb)) 4396 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4397 "using the ext4 subsystem"); 4398 else { 4399 /* 4400 * If we're probing be silent, if this looks like 4401 * it's actually an ext[34] filesystem. 4402 */ 4403 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4404 goto failed_mount; 4405 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4406 "to feature incompatibilities"); 4407 goto failed_mount; 4408 } 4409 } 4410 4411 if (IS_EXT3_SB(sb)) { 4412 if (ext3_feature_set_ok(sb)) 4413 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4414 "using the ext4 subsystem"); 4415 else { 4416 /* 4417 * If we're probing be silent, if this looks like 4418 * it's actually an ext4 filesystem. 4419 */ 4420 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4421 goto failed_mount; 4422 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4423 "to feature incompatibilities"); 4424 goto failed_mount; 4425 } 4426 } 4427 4428 /* 4429 * Check feature flags regardless of the revision level, since we 4430 * previously didn't change the revision level when setting the flags, 4431 * so there is a chance incompat flags are set on a rev 0 filesystem. 4432 */ 4433 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4434 goto failed_mount; 4435 4436 if (le32_to_cpu(es->s_log_block_size) > 4437 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4438 ext4_msg(sb, KERN_ERR, 4439 "Invalid log block size: %u", 4440 le32_to_cpu(es->s_log_block_size)); 4441 goto failed_mount; 4442 } 4443 if (le32_to_cpu(es->s_log_cluster_size) > 4444 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4445 ext4_msg(sb, KERN_ERR, 4446 "Invalid log cluster size: %u", 4447 le32_to_cpu(es->s_log_cluster_size)); 4448 goto failed_mount; 4449 } 4450 4451 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 4452 ext4_msg(sb, KERN_ERR, 4453 "Number of reserved GDT blocks insanely large: %d", 4454 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4455 goto failed_mount; 4456 } 4457 4458 if (bdev_dax_supported(sb->s_bdev, blocksize)) 4459 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4460 4461 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4462 if (ext4_has_feature_inline_data(sb)) { 4463 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4464 " that may contain inline data"); 4465 goto failed_mount; 4466 } 4467 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4468 ext4_msg(sb, KERN_ERR, 4469 "DAX unsupported by block device."); 4470 goto failed_mount; 4471 } 4472 } 4473 4474 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4475 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4476 es->s_encryption_level); 4477 goto failed_mount; 4478 } 4479 4480 if (sb->s_blocksize != blocksize) { 4481 /* Validate the filesystem blocksize */ 4482 if (!sb_set_blocksize(sb, blocksize)) { 4483 ext4_msg(sb, KERN_ERR, "bad block size %d", 4484 blocksize); 4485 goto failed_mount; 4486 } 4487 4488 brelse(bh); 4489 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4490 offset = do_div(logical_sb_block, blocksize); 4491 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4492 if (IS_ERR(bh)) { 4493 ext4_msg(sb, KERN_ERR, 4494 "Can't read superblock on 2nd try"); 4495 ret = PTR_ERR(bh); 4496 bh = NULL; 4497 goto failed_mount; 4498 } 4499 es = (struct ext4_super_block *)(bh->b_data + offset); 4500 sbi->s_es = es; 4501 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4502 ext4_msg(sb, KERN_ERR, 4503 "Magic mismatch, very weird!"); 4504 goto failed_mount; 4505 } 4506 } 4507 4508 has_huge_files = ext4_has_feature_huge_file(sb); 4509 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4510 has_huge_files); 4511 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4512 4513 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4514 if (ext4_has_feature_64bit(sb)) { 4515 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4516 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4517 !is_power_of_2(sbi->s_desc_size)) { 4518 ext4_msg(sb, KERN_ERR, 4519 "unsupported descriptor size %lu", 4520 sbi->s_desc_size); 4521 goto failed_mount; 4522 } 4523 } else 4524 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4525 4526 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4527 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4528 4529 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4530 if (sbi->s_inodes_per_block == 0) 4531 goto cantfind_ext4; 4532 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4533 sbi->s_inodes_per_group > blocksize * 8) { 4534 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4535 sbi->s_inodes_per_group); 4536 goto failed_mount; 4537 } 4538 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4539 sbi->s_inodes_per_block; 4540 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4541 sbi->s_sbh = bh; 4542 sbi->s_mount_state = le16_to_cpu(es->s_state); 4543 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4544 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4545 4546 for (i = 0; i < 4; i++) 4547 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4548 sbi->s_def_hash_version = es->s_def_hash_version; 4549 if (ext4_has_feature_dir_index(sb)) { 4550 i = le32_to_cpu(es->s_flags); 4551 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4552 sbi->s_hash_unsigned = 3; 4553 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4554 #ifdef __CHAR_UNSIGNED__ 4555 if (!sb_rdonly(sb)) 4556 es->s_flags |= 4557 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4558 sbi->s_hash_unsigned = 3; 4559 #else 4560 if (!sb_rdonly(sb)) 4561 es->s_flags |= 4562 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4563 #endif 4564 } 4565 } 4566 4567 /* Handle clustersize */ 4568 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4569 if (ext4_has_feature_bigalloc(sb)) { 4570 if (clustersize < blocksize) { 4571 ext4_msg(sb, KERN_ERR, 4572 "cluster size (%d) smaller than " 4573 "block size (%d)", clustersize, blocksize); 4574 goto failed_mount; 4575 } 4576 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4577 le32_to_cpu(es->s_log_block_size); 4578 sbi->s_clusters_per_group = 4579 le32_to_cpu(es->s_clusters_per_group); 4580 if (sbi->s_clusters_per_group > blocksize * 8) { 4581 ext4_msg(sb, KERN_ERR, 4582 "#clusters per group too big: %lu", 4583 sbi->s_clusters_per_group); 4584 goto failed_mount; 4585 } 4586 if (sbi->s_blocks_per_group != 4587 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4588 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4589 "clusters per group (%lu) inconsistent", 4590 sbi->s_blocks_per_group, 4591 sbi->s_clusters_per_group); 4592 goto failed_mount; 4593 } 4594 } else { 4595 if (clustersize != blocksize) { 4596 ext4_msg(sb, KERN_ERR, 4597 "fragment/cluster size (%d) != " 4598 "block size (%d)", clustersize, blocksize); 4599 goto failed_mount; 4600 } 4601 if (sbi->s_blocks_per_group > blocksize * 8) { 4602 ext4_msg(sb, KERN_ERR, 4603 "#blocks per group too big: %lu", 4604 sbi->s_blocks_per_group); 4605 goto failed_mount; 4606 } 4607 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4608 sbi->s_cluster_bits = 0; 4609 } 4610 sbi->s_cluster_ratio = clustersize / blocksize; 4611 4612 /* Do we have standard group size of clustersize * 8 blocks ? */ 4613 if (sbi->s_blocks_per_group == clustersize << 3) 4614 set_opt2(sb, STD_GROUP_SIZE); 4615 4616 /* 4617 * Test whether we have more sectors than will fit in sector_t, 4618 * and whether the max offset is addressable by the page cache. 4619 */ 4620 err = generic_check_addressable(sb->s_blocksize_bits, 4621 ext4_blocks_count(es)); 4622 if (err) { 4623 ext4_msg(sb, KERN_ERR, "filesystem" 4624 " too large to mount safely on this system"); 4625 goto failed_mount; 4626 } 4627 4628 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4629 goto cantfind_ext4; 4630 4631 /* check blocks count against device size */ 4632 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4633 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4634 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4635 "exceeds size of device (%llu blocks)", 4636 ext4_blocks_count(es), blocks_count); 4637 goto failed_mount; 4638 } 4639 4640 /* 4641 * It makes no sense for the first data block to be beyond the end 4642 * of the filesystem. 4643 */ 4644 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4645 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4646 "block %u is beyond end of filesystem (%llu)", 4647 le32_to_cpu(es->s_first_data_block), 4648 ext4_blocks_count(es)); 4649 goto failed_mount; 4650 } 4651 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4652 (sbi->s_cluster_ratio == 1)) { 4653 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4654 "block is 0 with a 1k block and cluster size"); 4655 goto failed_mount; 4656 } 4657 4658 blocks_count = (ext4_blocks_count(es) - 4659 le32_to_cpu(es->s_first_data_block) + 4660 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4661 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4662 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4663 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4664 "(block count %llu, first data block %u, " 4665 "blocks per group %lu)", blocks_count, 4666 ext4_blocks_count(es), 4667 le32_to_cpu(es->s_first_data_block), 4668 EXT4_BLOCKS_PER_GROUP(sb)); 4669 goto failed_mount; 4670 } 4671 sbi->s_groups_count = blocks_count; 4672 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4673 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4674 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4675 le32_to_cpu(es->s_inodes_count)) { 4676 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4677 le32_to_cpu(es->s_inodes_count), 4678 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4679 ret = -EINVAL; 4680 goto failed_mount; 4681 } 4682 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4683 EXT4_DESC_PER_BLOCK(sb); 4684 if (ext4_has_feature_meta_bg(sb)) { 4685 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4686 ext4_msg(sb, KERN_WARNING, 4687 "first meta block group too large: %u " 4688 "(group descriptor block count %u)", 4689 le32_to_cpu(es->s_first_meta_bg), db_count); 4690 goto failed_mount; 4691 } 4692 } 4693 rcu_assign_pointer(sbi->s_group_desc, 4694 kvmalloc_array(db_count, 4695 sizeof(struct buffer_head *), 4696 GFP_KERNEL)); 4697 if (sbi->s_group_desc == NULL) { 4698 ext4_msg(sb, KERN_ERR, "not enough memory"); 4699 ret = -ENOMEM; 4700 goto failed_mount; 4701 } 4702 4703 bgl_lock_init(sbi->s_blockgroup_lock); 4704 4705 /* Pre-read the descriptors into the buffer cache */ 4706 for (i = 0; i < db_count; i++) { 4707 block = descriptor_loc(sb, logical_sb_block, i); 4708 ext4_sb_breadahead_unmovable(sb, block); 4709 } 4710 4711 for (i = 0; i < db_count; i++) { 4712 struct buffer_head *bh; 4713 4714 block = descriptor_loc(sb, logical_sb_block, i); 4715 bh = ext4_sb_bread_unmovable(sb, block); 4716 if (IS_ERR(bh)) { 4717 ext4_msg(sb, KERN_ERR, 4718 "can't read group descriptor %d", i); 4719 db_count = i; 4720 ret = PTR_ERR(bh); 4721 bh = NULL; 4722 goto failed_mount2; 4723 } 4724 rcu_read_lock(); 4725 rcu_dereference(sbi->s_group_desc)[i] = bh; 4726 rcu_read_unlock(); 4727 } 4728 sbi->s_gdb_count = db_count; 4729 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4730 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4731 ret = -EFSCORRUPTED; 4732 goto failed_mount2; 4733 } 4734 4735 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4736 4737 /* Register extent status tree shrinker */ 4738 if (ext4_es_register_shrinker(sbi)) 4739 goto failed_mount3; 4740 4741 sbi->s_stripe = ext4_get_stripe_size(sbi); 4742 sbi->s_extent_max_zeroout_kb = 32; 4743 4744 /* 4745 * set up enough so that it can read an inode 4746 */ 4747 sb->s_op = &ext4_sops; 4748 sb->s_export_op = &ext4_export_ops; 4749 sb->s_xattr = ext4_xattr_handlers; 4750 #ifdef CONFIG_FS_ENCRYPTION 4751 sb->s_cop = &ext4_cryptops; 4752 #endif 4753 #ifdef CONFIG_FS_VERITY 4754 sb->s_vop = &ext4_verityops; 4755 #endif 4756 #ifdef CONFIG_QUOTA 4757 sb->dq_op = &ext4_quota_operations; 4758 if (ext4_has_feature_quota(sb)) 4759 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4760 else 4761 sb->s_qcop = &ext4_qctl_operations; 4762 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4763 #endif 4764 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4765 4766 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4767 mutex_init(&sbi->s_orphan_lock); 4768 4769 /* Initialize fast commit stuff */ 4770 atomic_set(&sbi->s_fc_subtid, 0); 4771 atomic_set(&sbi->s_fc_ineligible_updates, 0); 4772 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4773 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4774 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4775 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4776 sbi->s_fc_bytes = 0; 4777 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4778 ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING); 4779 spin_lock_init(&sbi->s_fc_lock); 4780 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4781 sbi->s_fc_replay_state.fc_regions = NULL; 4782 sbi->s_fc_replay_state.fc_regions_size = 0; 4783 sbi->s_fc_replay_state.fc_regions_used = 0; 4784 sbi->s_fc_replay_state.fc_regions_valid = 0; 4785 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4786 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4787 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4788 4789 sb->s_root = NULL; 4790 4791 needs_recovery = (es->s_last_orphan != 0 || 4792 ext4_has_feature_journal_needs_recovery(sb)); 4793 4794 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4795 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4796 goto failed_mount3a; 4797 4798 /* 4799 * The first inode we look at is the journal inode. Don't try 4800 * root first: it may be modified in the journal! 4801 */ 4802 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4803 err = ext4_load_journal(sb, es, journal_devnum); 4804 if (err) 4805 goto failed_mount3a; 4806 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4807 ext4_has_feature_journal_needs_recovery(sb)) { 4808 ext4_msg(sb, KERN_ERR, "required journal recovery " 4809 "suppressed and not mounted read-only"); 4810 goto failed_mount_wq; 4811 } else { 4812 /* Nojournal mode, all journal mount options are illegal */ 4813 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4814 ext4_msg(sb, KERN_ERR, "can't mount with " 4815 "journal_checksum, fs mounted w/o journal"); 4816 goto failed_mount_wq; 4817 } 4818 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4819 ext4_msg(sb, KERN_ERR, "can't mount with " 4820 "journal_async_commit, fs mounted w/o journal"); 4821 goto failed_mount_wq; 4822 } 4823 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4824 ext4_msg(sb, KERN_ERR, "can't mount with " 4825 "commit=%lu, fs mounted w/o journal", 4826 sbi->s_commit_interval / HZ); 4827 goto failed_mount_wq; 4828 } 4829 if (EXT4_MOUNT_DATA_FLAGS & 4830 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4831 ext4_msg(sb, KERN_ERR, "can't mount with " 4832 "data=, fs mounted w/o journal"); 4833 goto failed_mount_wq; 4834 } 4835 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4836 clear_opt(sb, JOURNAL_CHECKSUM); 4837 clear_opt(sb, DATA_FLAGS); 4838 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4839 sbi->s_journal = NULL; 4840 needs_recovery = 0; 4841 goto no_journal; 4842 } 4843 4844 if (ext4_has_feature_64bit(sb) && 4845 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4846 JBD2_FEATURE_INCOMPAT_64BIT)) { 4847 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4848 goto failed_mount_wq; 4849 } 4850 4851 if (!set_journal_csum_feature_set(sb)) { 4852 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4853 "feature set"); 4854 goto failed_mount_wq; 4855 } 4856 4857 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4858 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4859 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4860 ext4_msg(sb, KERN_ERR, 4861 "Failed to set fast commit journal feature"); 4862 goto failed_mount_wq; 4863 } 4864 4865 /* We have now updated the journal if required, so we can 4866 * validate the data journaling mode. */ 4867 switch (test_opt(sb, DATA_FLAGS)) { 4868 case 0: 4869 /* No mode set, assume a default based on the journal 4870 * capabilities: ORDERED_DATA if the journal can 4871 * cope, else JOURNAL_DATA 4872 */ 4873 if (jbd2_journal_check_available_features 4874 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4875 set_opt(sb, ORDERED_DATA); 4876 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4877 } else { 4878 set_opt(sb, JOURNAL_DATA); 4879 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4880 } 4881 break; 4882 4883 case EXT4_MOUNT_ORDERED_DATA: 4884 case EXT4_MOUNT_WRITEBACK_DATA: 4885 if (!jbd2_journal_check_available_features 4886 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4887 ext4_msg(sb, KERN_ERR, "Journal does not support " 4888 "requested data journaling mode"); 4889 goto failed_mount_wq; 4890 } 4891 default: 4892 break; 4893 } 4894 4895 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4896 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4897 ext4_msg(sb, KERN_ERR, "can't mount with " 4898 "journal_async_commit in data=ordered mode"); 4899 goto failed_mount_wq; 4900 } 4901 4902 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 4903 4904 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback; 4905 sbi->s_journal->j_submit_inode_data_buffers = 4906 ext4_journal_submit_inode_data_buffers; 4907 sbi->s_journal->j_finish_inode_data_buffers = 4908 ext4_journal_finish_inode_data_buffers; 4909 4910 no_journal: 4911 if (!test_opt(sb, NO_MBCACHE)) { 4912 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4913 if (!sbi->s_ea_block_cache) { 4914 ext4_msg(sb, KERN_ERR, 4915 "Failed to create ea_block_cache"); 4916 goto failed_mount_wq; 4917 } 4918 4919 if (ext4_has_feature_ea_inode(sb)) { 4920 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4921 if (!sbi->s_ea_inode_cache) { 4922 ext4_msg(sb, KERN_ERR, 4923 "Failed to create ea_inode_cache"); 4924 goto failed_mount_wq; 4925 } 4926 } 4927 } 4928 4929 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4930 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4931 goto failed_mount_wq; 4932 } 4933 4934 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4935 !ext4_has_feature_encrypt(sb)) { 4936 ext4_set_feature_encrypt(sb); 4937 ext4_commit_super(sb, 1); 4938 } 4939 4940 /* 4941 * Get the # of file system overhead blocks from the 4942 * superblock if present. 4943 */ 4944 if (es->s_overhead_clusters) 4945 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4946 else { 4947 err = ext4_calculate_overhead(sb); 4948 if (err) 4949 goto failed_mount_wq; 4950 } 4951 4952 /* 4953 * The maximum number of concurrent works can be high and 4954 * concurrency isn't really necessary. Limit it to 1. 4955 */ 4956 EXT4_SB(sb)->rsv_conversion_wq = 4957 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4958 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4959 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4960 ret = -ENOMEM; 4961 goto failed_mount4; 4962 } 4963 4964 /* 4965 * The jbd2_journal_load will have done any necessary log recovery, 4966 * so we can safely mount the rest of the filesystem now. 4967 */ 4968 4969 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4970 if (IS_ERR(root)) { 4971 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4972 ret = PTR_ERR(root); 4973 root = NULL; 4974 goto failed_mount4; 4975 } 4976 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4977 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4978 iput(root); 4979 goto failed_mount4; 4980 } 4981 4982 #ifdef CONFIG_UNICODE 4983 if (sb->s_encoding) 4984 sb->s_d_op = &ext4_dentry_ops; 4985 #endif 4986 4987 sb->s_root = d_make_root(root); 4988 if (!sb->s_root) { 4989 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4990 ret = -ENOMEM; 4991 goto failed_mount4; 4992 } 4993 4994 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4995 if (ret == -EROFS) { 4996 sb->s_flags |= SB_RDONLY; 4997 ret = 0; 4998 } else if (ret) 4999 goto failed_mount4a; 5000 5001 ext4_set_resv_clusters(sb); 5002 5003 if (test_opt(sb, BLOCK_VALIDITY)) { 5004 err = ext4_setup_system_zone(sb); 5005 if (err) { 5006 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5007 "zone (%d)", err); 5008 goto failed_mount4a; 5009 } 5010 } 5011 ext4_fc_replay_cleanup(sb); 5012 5013 ext4_ext_init(sb); 5014 err = ext4_mb_init(sb); 5015 if (err) { 5016 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5017 err); 5018 goto failed_mount5; 5019 } 5020 5021 block = ext4_count_free_clusters(sb); 5022 ext4_free_blocks_count_set(sbi->s_es, 5023 EXT4_C2B(sbi, block)); 5024 ext4_superblock_csum_set(sb); 5025 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5026 GFP_KERNEL); 5027 if (!err) { 5028 unsigned long freei = ext4_count_free_inodes(sb); 5029 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5030 ext4_superblock_csum_set(sb); 5031 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5032 GFP_KERNEL); 5033 } 5034 if (!err) 5035 err = percpu_counter_init(&sbi->s_dirs_counter, 5036 ext4_count_dirs(sb), GFP_KERNEL); 5037 if (!err) 5038 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5039 GFP_KERNEL); 5040 if (!err) 5041 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5042 5043 if (err) { 5044 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5045 goto failed_mount6; 5046 } 5047 5048 if (ext4_has_feature_flex_bg(sb)) 5049 if (!ext4_fill_flex_info(sb)) { 5050 ext4_msg(sb, KERN_ERR, 5051 "unable to initialize " 5052 "flex_bg meta info!"); 5053 goto failed_mount6; 5054 } 5055 5056 err = ext4_register_li_request(sb, first_not_zeroed); 5057 if (err) 5058 goto failed_mount6; 5059 5060 err = ext4_register_sysfs(sb); 5061 if (err) 5062 goto failed_mount7; 5063 5064 #ifdef CONFIG_QUOTA 5065 /* Enable quota usage during mount. */ 5066 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5067 err = ext4_enable_quotas(sb); 5068 if (err) 5069 goto failed_mount8; 5070 } 5071 #endif /* CONFIG_QUOTA */ 5072 5073 /* 5074 * Save the original bdev mapping's wb_err value which could be 5075 * used to detect the metadata async write error. 5076 */ 5077 spin_lock_init(&sbi->s_bdev_wb_lock); 5078 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5079 &sbi->s_bdev_wb_err); 5080 sb->s_bdev->bd_super = sb; 5081 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5082 ext4_orphan_cleanup(sb, es); 5083 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5084 if (needs_recovery) { 5085 ext4_msg(sb, KERN_INFO, "recovery complete"); 5086 err = ext4_mark_recovery_complete(sb, es); 5087 if (err) 5088 goto failed_mount8; 5089 } 5090 if (EXT4_SB(sb)->s_journal) { 5091 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5092 descr = " journalled data mode"; 5093 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5094 descr = " ordered data mode"; 5095 else 5096 descr = " writeback data mode"; 5097 } else 5098 descr = "out journal"; 5099 5100 if (test_opt(sb, DISCARD)) { 5101 struct request_queue *q = bdev_get_queue(sb->s_bdev); 5102 if (!blk_queue_discard(q)) 5103 ext4_msg(sb, KERN_WARNING, 5104 "mounting with \"discard\" option, but " 5105 "the device does not support discard"); 5106 } 5107 5108 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5109 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 5110 "Opts: %.*s%s%s", descr, 5111 (int) sizeof(sbi->s_es->s_mount_opts), 5112 sbi->s_es->s_mount_opts, 5113 *sbi->s_es->s_mount_opts ? "; " : "", orig_data); 5114 5115 if (es->s_error_count) 5116 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5117 5118 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5119 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5120 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5121 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5122 atomic_set(&sbi->s_warning_count, 0); 5123 atomic_set(&sbi->s_msg_count, 0); 5124 5125 kfree(orig_data); 5126 return 0; 5127 5128 cantfind_ext4: 5129 if (!silent) 5130 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5131 goto failed_mount; 5132 5133 failed_mount8: 5134 ext4_unregister_sysfs(sb); 5135 kobject_put(&sbi->s_kobj); 5136 failed_mount7: 5137 ext4_unregister_li_request(sb); 5138 failed_mount6: 5139 ext4_mb_release(sb); 5140 rcu_read_lock(); 5141 flex_groups = rcu_dereference(sbi->s_flex_groups); 5142 if (flex_groups) { 5143 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 5144 kvfree(flex_groups[i]); 5145 kvfree(flex_groups); 5146 } 5147 rcu_read_unlock(); 5148 percpu_counter_destroy(&sbi->s_freeclusters_counter); 5149 percpu_counter_destroy(&sbi->s_freeinodes_counter); 5150 percpu_counter_destroy(&sbi->s_dirs_counter); 5151 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 5152 percpu_free_rwsem(&sbi->s_writepages_rwsem); 5153 failed_mount5: 5154 ext4_ext_release(sb); 5155 ext4_release_system_zone(sb); 5156 failed_mount4a: 5157 dput(sb->s_root); 5158 sb->s_root = NULL; 5159 failed_mount4: 5160 ext4_msg(sb, KERN_ERR, "mount failed"); 5161 if (EXT4_SB(sb)->rsv_conversion_wq) 5162 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5163 failed_mount_wq: 5164 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5165 sbi->s_ea_inode_cache = NULL; 5166 5167 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5168 sbi->s_ea_block_cache = NULL; 5169 5170 if (sbi->s_journal) { 5171 jbd2_journal_destroy(sbi->s_journal); 5172 sbi->s_journal = NULL; 5173 } 5174 failed_mount3a: 5175 ext4_es_unregister_shrinker(sbi); 5176 failed_mount3: 5177 del_timer_sync(&sbi->s_err_report); 5178 if (sbi->s_mmp_tsk) 5179 kthread_stop(sbi->s_mmp_tsk); 5180 failed_mount2: 5181 rcu_read_lock(); 5182 group_desc = rcu_dereference(sbi->s_group_desc); 5183 for (i = 0; i < db_count; i++) 5184 brelse(group_desc[i]); 5185 kvfree(group_desc); 5186 rcu_read_unlock(); 5187 failed_mount: 5188 if (sbi->s_chksum_driver) 5189 crypto_free_shash(sbi->s_chksum_driver); 5190 5191 #ifdef CONFIG_UNICODE 5192 utf8_unload(sb->s_encoding); 5193 #endif 5194 5195 #ifdef CONFIG_QUOTA 5196 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5197 kfree(get_qf_name(sb, sbi, i)); 5198 #endif 5199 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5200 ext4_blkdev_remove(sbi); 5201 brelse(bh); 5202 out_fail: 5203 sb->s_fs_info = NULL; 5204 kfree(sbi->s_blockgroup_lock); 5205 out_free_base: 5206 kfree(sbi); 5207 kfree(orig_data); 5208 fs_put_dax(dax_dev); 5209 return err ? err : ret; 5210 } 5211 5212 /* 5213 * Setup any per-fs journal parameters now. We'll do this both on 5214 * initial mount, once the journal has been initialised but before we've 5215 * done any recovery; and again on any subsequent remount. 5216 */ 5217 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5218 { 5219 struct ext4_sb_info *sbi = EXT4_SB(sb); 5220 5221 journal->j_commit_interval = sbi->s_commit_interval; 5222 journal->j_min_batch_time = sbi->s_min_batch_time; 5223 journal->j_max_batch_time = sbi->s_max_batch_time; 5224 ext4_fc_init(sb, journal); 5225 5226 write_lock(&journal->j_state_lock); 5227 if (test_opt(sb, BARRIER)) 5228 journal->j_flags |= JBD2_BARRIER; 5229 else 5230 journal->j_flags &= ~JBD2_BARRIER; 5231 if (test_opt(sb, DATA_ERR_ABORT)) 5232 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5233 else 5234 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5235 write_unlock(&journal->j_state_lock); 5236 } 5237 5238 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5239 unsigned int journal_inum) 5240 { 5241 struct inode *journal_inode; 5242 5243 /* 5244 * Test for the existence of a valid inode on disk. Bad things 5245 * happen if we iget() an unused inode, as the subsequent iput() 5246 * will try to delete it. 5247 */ 5248 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5249 if (IS_ERR(journal_inode)) { 5250 ext4_msg(sb, KERN_ERR, "no journal found"); 5251 return NULL; 5252 } 5253 if (!journal_inode->i_nlink) { 5254 make_bad_inode(journal_inode); 5255 iput(journal_inode); 5256 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5257 return NULL; 5258 } 5259 5260 jbd_debug(2, "Journal inode found at %p: %lld bytes\n", 5261 journal_inode, journal_inode->i_size); 5262 if (!S_ISREG(journal_inode->i_mode)) { 5263 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5264 iput(journal_inode); 5265 return NULL; 5266 } 5267 return journal_inode; 5268 } 5269 5270 static journal_t *ext4_get_journal(struct super_block *sb, 5271 unsigned int journal_inum) 5272 { 5273 struct inode *journal_inode; 5274 journal_t *journal; 5275 5276 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5277 return NULL; 5278 5279 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5280 if (!journal_inode) 5281 return NULL; 5282 5283 journal = jbd2_journal_init_inode(journal_inode); 5284 if (!journal) { 5285 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5286 iput(journal_inode); 5287 return NULL; 5288 } 5289 journal->j_private = sb; 5290 ext4_init_journal_params(sb, journal); 5291 return journal; 5292 } 5293 5294 static journal_t *ext4_get_dev_journal(struct super_block *sb, 5295 dev_t j_dev) 5296 { 5297 struct buffer_head *bh; 5298 journal_t *journal; 5299 ext4_fsblk_t start; 5300 ext4_fsblk_t len; 5301 int hblock, blocksize; 5302 ext4_fsblk_t sb_block; 5303 unsigned long offset; 5304 struct ext4_super_block *es; 5305 struct block_device *bdev; 5306 5307 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5308 return NULL; 5309 5310 bdev = ext4_blkdev_get(j_dev, sb); 5311 if (bdev == NULL) 5312 return NULL; 5313 5314 blocksize = sb->s_blocksize; 5315 hblock = bdev_logical_block_size(bdev); 5316 if (blocksize < hblock) { 5317 ext4_msg(sb, KERN_ERR, 5318 "blocksize too small for journal device"); 5319 goto out_bdev; 5320 } 5321 5322 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5323 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5324 set_blocksize(bdev, blocksize); 5325 if (!(bh = __bread(bdev, sb_block, blocksize))) { 5326 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5327 "external journal"); 5328 goto out_bdev; 5329 } 5330 5331 es = (struct ext4_super_block *) (bh->b_data + offset); 5332 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5333 !(le32_to_cpu(es->s_feature_incompat) & 5334 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5335 ext4_msg(sb, KERN_ERR, "external journal has " 5336 "bad superblock"); 5337 brelse(bh); 5338 goto out_bdev; 5339 } 5340 5341 if ((le32_to_cpu(es->s_feature_ro_compat) & 5342 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5343 es->s_checksum != ext4_superblock_csum(sb, es)) { 5344 ext4_msg(sb, KERN_ERR, "external journal has " 5345 "corrupt superblock"); 5346 brelse(bh); 5347 goto out_bdev; 5348 } 5349 5350 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5351 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5352 brelse(bh); 5353 goto out_bdev; 5354 } 5355 5356 len = ext4_blocks_count(es); 5357 start = sb_block + 1; 5358 brelse(bh); /* we're done with the superblock */ 5359 5360 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 5361 start, len, blocksize); 5362 if (!journal) { 5363 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5364 goto out_bdev; 5365 } 5366 journal->j_private = sb; 5367 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) { 5368 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 5369 goto out_journal; 5370 } 5371 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5372 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5373 "user (unsupported) - %d", 5374 be32_to_cpu(journal->j_superblock->s_nr_users)); 5375 goto out_journal; 5376 } 5377 EXT4_SB(sb)->s_journal_bdev = bdev; 5378 ext4_init_journal_params(sb, journal); 5379 return journal; 5380 5381 out_journal: 5382 jbd2_journal_destroy(journal); 5383 out_bdev: 5384 ext4_blkdev_put(bdev); 5385 return NULL; 5386 } 5387 5388 static int ext4_load_journal(struct super_block *sb, 5389 struct ext4_super_block *es, 5390 unsigned long journal_devnum) 5391 { 5392 journal_t *journal; 5393 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5394 dev_t journal_dev; 5395 int err = 0; 5396 int really_read_only; 5397 int journal_dev_ro; 5398 5399 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5400 return -EFSCORRUPTED; 5401 5402 if (journal_devnum && 5403 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5404 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5405 "numbers have changed"); 5406 journal_dev = new_decode_dev(journal_devnum); 5407 } else 5408 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5409 5410 if (journal_inum && journal_dev) { 5411 ext4_msg(sb, KERN_ERR, 5412 "filesystem has both journal inode and journal device!"); 5413 return -EINVAL; 5414 } 5415 5416 if (journal_inum) { 5417 journal = ext4_get_journal(sb, journal_inum); 5418 if (!journal) 5419 return -EINVAL; 5420 } else { 5421 journal = ext4_get_dev_journal(sb, journal_dev); 5422 if (!journal) 5423 return -EINVAL; 5424 } 5425 5426 journal_dev_ro = bdev_read_only(journal->j_dev); 5427 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5428 5429 if (journal_dev_ro && !sb_rdonly(sb)) { 5430 ext4_msg(sb, KERN_ERR, 5431 "journal device read-only, try mounting with '-o ro'"); 5432 err = -EROFS; 5433 goto err_out; 5434 } 5435 5436 /* 5437 * Are we loading a blank journal or performing recovery after a 5438 * crash? For recovery, we need to check in advance whether we 5439 * can get read-write access to the device. 5440 */ 5441 if (ext4_has_feature_journal_needs_recovery(sb)) { 5442 if (sb_rdonly(sb)) { 5443 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5444 "required on readonly filesystem"); 5445 if (really_read_only) { 5446 ext4_msg(sb, KERN_ERR, "write access " 5447 "unavailable, cannot proceed " 5448 "(try mounting with noload)"); 5449 err = -EROFS; 5450 goto err_out; 5451 } 5452 ext4_msg(sb, KERN_INFO, "write access will " 5453 "be enabled during recovery"); 5454 } 5455 } 5456 5457 if (!(journal->j_flags & JBD2_BARRIER)) 5458 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5459 5460 if (!ext4_has_feature_journal_needs_recovery(sb)) 5461 err = jbd2_journal_wipe(journal, !really_read_only); 5462 if (!err) { 5463 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5464 if (save) 5465 memcpy(save, ((char *) es) + 5466 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 5467 err = jbd2_journal_load(journal); 5468 if (save) 5469 memcpy(((char *) es) + EXT4_S_ERR_START, 5470 save, EXT4_S_ERR_LEN); 5471 kfree(save); 5472 } 5473 5474 if (err) { 5475 ext4_msg(sb, KERN_ERR, "error loading journal"); 5476 goto err_out; 5477 } 5478 5479 EXT4_SB(sb)->s_journal = journal; 5480 err = ext4_clear_journal_err(sb, es); 5481 if (err) { 5482 EXT4_SB(sb)->s_journal = NULL; 5483 jbd2_journal_destroy(journal); 5484 return err; 5485 } 5486 5487 if (!really_read_only && journal_devnum && 5488 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5489 es->s_journal_dev = cpu_to_le32(journal_devnum); 5490 5491 /* Make sure we flush the recovery flag to disk. */ 5492 ext4_commit_super(sb, 1); 5493 } 5494 5495 return 0; 5496 5497 err_out: 5498 jbd2_journal_destroy(journal); 5499 return err; 5500 } 5501 5502 static int ext4_commit_super(struct super_block *sb, int sync) 5503 { 5504 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 5505 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 5506 int error = 0; 5507 5508 if (!sbh || block_device_ejected(sb)) 5509 return error; 5510 5511 /* 5512 * If the file system is mounted read-only, don't update the 5513 * superblock write time. This avoids updating the superblock 5514 * write time when we are mounting the root file system 5515 * read/only but we need to replay the journal; at that point, 5516 * for people who are east of GMT and who make their clock 5517 * tick in localtime for Windows bug-for-bug compatibility, 5518 * the clock is set in the future, and this will cause e2fsck 5519 * to complain and force a full file system check. 5520 */ 5521 if (!(sb->s_flags & SB_RDONLY)) 5522 ext4_update_tstamp(es, s_wtime); 5523 if (sb->s_bdev->bd_part) 5524 es->s_kbytes_written = 5525 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 5526 ((part_stat_read(sb->s_bdev->bd_part, 5527 sectors[STAT_WRITE]) - 5528 EXT4_SB(sb)->s_sectors_written_start) >> 1)); 5529 else 5530 es->s_kbytes_written = 5531 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written); 5532 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter)) 5533 ext4_free_blocks_count_set(es, 5534 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive( 5535 &EXT4_SB(sb)->s_freeclusters_counter))); 5536 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter)) 5537 es->s_free_inodes_count = 5538 cpu_to_le32(percpu_counter_sum_positive( 5539 &EXT4_SB(sb)->s_freeinodes_counter)); 5540 BUFFER_TRACE(sbh, "marking dirty"); 5541 ext4_superblock_csum_set(sb); 5542 if (sync) 5543 lock_buffer(sbh); 5544 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 5545 /* 5546 * Oh, dear. A previous attempt to write the 5547 * superblock failed. This could happen because the 5548 * USB device was yanked out. Or it could happen to 5549 * be a transient write error and maybe the block will 5550 * be remapped. Nothing we can do but to retry the 5551 * write and hope for the best. 5552 */ 5553 ext4_msg(sb, KERN_ERR, "previous I/O error to " 5554 "superblock detected"); 5555 clear_buffer_write_io_error(sbh); 5556 set_buffer_uptodate(sbh); 5557 } 5558 mark_buffer_dirty(sbh); 5559 if (sync) { 5560 unlock_buffer(sbh); 5561 error = __sync_dirty_buffer(sbh, 5562 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0)); 5563 if (buffer_write_io_error(sbh)) { 5564 ext4_msg(sb, KERN_ERR, "I/O error while writing " 5565 "superblock"); 5566 clear_buffer_write_io_error(sbh); 5567 set_buffer_uptodate(sbh); 5568 } 5569 } 5570 return error; 5571 } 5572 5573 /* 5574 * Have we just finished recovery? If so, and if we are mounting (or 5575 * remounting) the filesystem readonly, then we will end up with a 5576 * consistent fs on disk. Record that fact. 5577 */ 5578 static int ext4_mark_recovery_complete(struct super_block *sb, 5579 struct ext4_super_block *es) 5580 { 5581 int err; 5582 journal_t *journal = EXT4_SB(sb)->s_journal; 5583 5584 if (!ext4_has_feature_journal(sb)) { 5585 if (journal != NULL) { 5586 ext4_error(sb, "Journal got removed while the fs was " 5587 "mounted!"); 5588 return -EFSCORRUPTED; 5589 } 5590 return 0; 5591 } 5592 jbd2_journal_lock_updates(journal); 5593 err = jbd2_journal_flush(journal); 5594 if (err < 0) 5595 goto out; 5596 5597 if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) { 5598 ext4_clear_feature_journal_needs_recovery(sb); 5599 ext4_commit_super(sb, 1); 5600 } 5601 out: 5602 jbd2_journal_unlock_updates(journal); 5603 return err; 5604 } 5605 5606 /* 5607 * If we are mounting (or read-write remounting) a filesystem whose journal 5608 * has recorded an error from a previous lifetime, move that error to the 5609 * main filesystem now. 5610 */ 5611 static int ext4_clear_journal_err(struct super_block *sb, 5612 struct ext4_super_block *es) 5613 { 5614 journal_t *journal; 5615 int j_errno; 5616 const char *errstr; 5617 5618 if (!ext4_has_feature_journal(sb)) { 5619 ext4_error(sb, "Journal got removed while the fs was mounted!"); 5620 return -EFSCORRUPTED; 5621 } 5622 5623 journal = EXT4_SB(sb)->s_journal; 5624 5625 /* 5626 * Now check for any error status which may have been recorded in the 5627 * journal by a prior ext4_error() or ext4_abort() 5628 */ 5629 5630 j_errno = jbd2_journal_errno(journal); 5631 if (j_errno) { 5632 char nbuf[16]; 5633 5634 errstr = ext4_decode_error(sb, j_errno, nbuf); 5635 ext4_warning(sb, "Filesystem error recorded " 5636 "from previous mount: %s", errstr); 5637 ext4_warning(sb, "Marking fs in need of filesystem check."); 5638 5639 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 5640 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5641 ext4_commit_super(sb, 1); 5642 5643 jbd2_journal_clear_err(journal); 5644 jbd2_journal_update_sb_errno(journal); 5645 } 5646 return 0; 5647 } 5648 5649 /* 5650 * Force the running and committing transactions to commit, 5651 * and wait on the commit. 5652 */ 5653 int ext4_force_commit(struct super_block *sb) 5654 { 5655 journal_t *journal; 5656 5657 if (sb_rdonly(sb)) 5658 return 0; 5659 5660 journal = EXT4_SB(sb)->s_journal; 5661 return ext4_journal_force_commit(journal); 5662 } 5663 5664 static int ext4_sync_fs(struct super_block *sb, int wait) 5665 { 5666 int ret = 0; 5667 tid_t target; 5668 bool needs_barrier = false; 5669 struct ext4_sb_info *sbi = EXT4_SB(sb); 5670 5671 if (unlikely(ext4_forced_shutdown(sbi))) 5672 return 0; 5673 5674 trace_ext4_sync_fs(sb, wait); 5675 flush_workqueue(sbi->rsv_conversion_wq); 5676 /* 5677 * Writeback quota in non-journalled quota case - journalled quota has 5678 * no dirty dquots 5679 */ 5680 dquot_writeback_dquots(sb, -1); 5681 /* 5682 * Data writeback is possible w/o journal transaction, so barrier must 5683 * being sent at the end of the function. But we can skip it if 5684 * transaction_commit will do it for us. 5685 */ 5686 if (sbi->s_journal) { 5687 target = jbd2_get_latest_transaction(sbi->s_journal); 5688 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 5689 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 5690 needs_barrier = true; 5691 5692 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 5693 if (wait) 5694 ret = jbd2_log_wait_commit(sbi->s_journal, 5695 target); 5696 } 5697 } else if (wait && test_opt(sb, BARRIER)) 5698 needs_barrier = true; 5699 if (needs_barrier) { 5700 int err; 5701 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL); 5702 if (!ret) 5703 ret = err; 5704 } 5705 5706 return ret; 5707 } 5708 5709 /* 5710 * LVM calls this function before a (read-only) snapshot is created. This 5711 * gives us a chance to flush the journal completely and mark the fs clean. 5712 * 5713 * Note that only this function cannot bring a filesystem to be in a clean 5714 * state independently. It relies on upper layer to stop all data & metadata 5715 * modifications. 5716 */ 5717 static int ext4_freeze(struct super_block *sb) 5718 { 5719 int error = 0; 5720 journal_t *journal; 5721 5722 if (sb_rdonly(sb)) 5723 return 0; 5724 5725 journal = EXT4_SB(sb)->s_journal; 5726 5727 if (journal) { 5728 /* Now we set up the journal barrier. */ 5729 jbd2_journal_lock_updates(journal); 5730 5731 /* 5732 * Don't clear the needs_recovery flag if we failed to 5733 * flush the journal. 5734 */ 5735 error = jbd2_journal_flush(journal); 5736 if (error < 0) 5737 goto out; 5738 5739 /* Journal blocked and flushed, clear needs_recovery flag. */ 5740 ext4_clear_feature_journal_needs_recovery(sb); 5741 } 5742 5743 error = ext4_commit_super(sb, 1); 5744 out: 5745 if (journal) 5746 /* we rely on upper layer to stop further updates */ 5747 jbd2_journal_unlock_updates(journal); 5748 return error; 5749 } 5750 5751 /* 5752 * Called by LVM after the snapshot is done. We need to reset the RECOVER 5753 * flag here, even though the filesystem is not technically dirty yet. 5754 */ 5755 static int ext4_unfreeze(struct super_block *sb) 5756 { 5757 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 5758 return 0; 5759 5760 if (EXT4_SB(sb)->s_journal) { 5761 /* Reset the needs_recovery flag before the fs is unlocked. */ 5762 ext4_set_feature_journal_needs_recovery(sb); 5763 } 5764 5765 ext4_commit_super(sb, 1); 5766 return 0; 5767 } 5768 5769 /* 5770 * Structure to save mount options for ext4_remount's benefit 5771 */ 5772 struct ext4_mount_options { 5773 unsigned long s_mount_opt; 5774 unsigned long s_mount_opt2; 5775 kuid_t s_resuid; 5776 kgid_t s_resgid; 5777 unsigned long s_commit_interval; 5778 u32 s_min_batch_time, s_max_batch_time; 5779 #ifdef CONFIG_QUOTA 5780 int s_jquota_fmt; 5781 char *s_qf_names[EXT4_MAXQUOTAS]; 5782 #endif 5783 }; 5784 5785 static int ext4_remount(struct super_block *sb, int *flags, char *data) 5786 { 5787 struct ext4_super_block *es; 5788 struct ext4_sb_info *sbi = EXT4_SB(sb); 5789 unsigned long old_sb_flags, vfs_flags; 5790 struct ext4_mount_options old_opts; 5791 int enable_quota = 0; 5792 ext4_group_t g; 5793 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5794 int err = 0; 5795 #ifdef CONFIG_QUOTA 5796 int i, j; 5797 char *to_free[EXT4_MAXQUOTAS]; 5798 #endif 5799 char *orig_data = kstrdup(data, GFP_KERNEL); 5800 5801 if (data && !orig_data) 5802 return -ENOMEM; 5803 5804 /* Store the original options */ 5805 old_sb_flags = sb->s_flags; 5806 old_opts.s_mount_opt = sbi->s_mount_opt; 5807 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 5808 old_opts.s_resuid = sbi->s_resuid; 5809 old_opts.s_resgid = sbi->s_resgid; 5810 old_opts.s_commit_interval = sbi->s_commit_interval; 5811 old_opts.s_min_batch_time = sbi->s_min_batch_time; 5812 old_opts.s_max_batch_time = sbi->s_max_batch_time; 5813 #ifdef CONFIG_QUOTA 5814 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 5815 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5816 if (sbi->s_qf_names[i]) { 5817 char *qf_name = get_qf_name(sb, sbi, i); 5818 5819 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 5820 if (!old_opts.s_qf_names[i]) { 5821 for (j = 0; j < i; j++) 5822 kfree(old_opts.s_qf_names[j]); 5823 kfree(orig_data); 5824 return -ENOMEM; 5825 } 5826 } else 5827 old_opts.s_qf_names[i] = NULL; 5828 #endif 5829 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 5830 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio; 5831 5832 /* 5833 * Some options can be enabled by ext4 and/or by VFS mount flag 5834 * either way we need to make sure it matches in both *flags and 5835 * s_flags. Copy those selected flags from *flags to s_flags 5836 */ 5837 vfs_flags = SB_LAZYTIME | SB_I_VERSION; 5838 sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags); 5839 5840 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) { 5841 err = -EINVAL; 5842 goto restore_opts; 5843 } 5844 5845 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 5846 test_opt(sb, JOURNAL_CHECKSUM)) { 5847 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 5848 "during remount not supported; ignoring"); 5849 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 5850 } 5851 5852 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5853 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5854 ext4_msg(sb, KERN_ERR, "can't mount with " 5855 "both data=journal and delalloc"); 5856 err = -EINVAL; 5857 goto restore_opts; 5858 } 5859 if (test_opt(sb, DIOREAD_NOLOCK)) { 5860 ext4_msg(sb, KERN_ERR, "can't mount with " 5861 "both data=journal and dioread_nolock"); 5862 err = -EINVAL; 5863 goto restore_opts; 5864 } 5865 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 5866 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5867 ext4_msg(sb, KERN_ERR, "can't mount with " 5868 "journal_async_commit in data=ordered mode"); 5869 err = -EINVAL; 5870 goto restore_opts; 5871 } 5872 } 5873 5874 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 5875 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 5876 err = -EINVAL; 5877 goto restore_opts; 5878 } 5879 5880 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) 5881 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user"); 5882 5883 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5884 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5885 5886 es = sbi->s_es; 5887 5888 if (sbi->s_journal) { 5889 ext4_init_journal_params(sb, sbi->s_journal); 5890 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 5891 } 5892 5893 if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) { 5894 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) { 5895 err = -EROFS; 5896 goto restore_opts; 5897 } 5898 5899 if (*flags & SB_RDONLY) { 5900 err = sync_filesystem(sb); 5901 if (err < 0) 5902 goto restore_opts; 5903 err = dquot_suspend(sb, -1); 5904 if (err < 0) 5905 goto restore_opts; 5906 5907 /* 5908 * First of all, the unconditional stuff we have to do 5909 * to disable replay of the journal when we next remount 5910 */ 5911 sb->s_flags |= SB_RDONLY; 5912 5913 /* 5914 * OK, test if we are remounting a valid rw partition 5915 * readonly, and if so set the rdonly flag and then 5916 * mark the partition as valid again. 5917 */ 5918 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 5919 (sbi->s_mount_state & EXT4_VALID_FS)) 5920 es->s_state = cpu_to_le16(sbi->s_mount_state); 5921 5922 if (sbi->s_journal) { 5923 /* 5924 * We let remount-ro finish even if marking fs 5925 * as clean failed... 5926 */ 5927 ext4_mark_recovery_complete(sb, es); 5928 } 5929 if (sbi->s_mmp_tsk) 5930 kthread_stop(sbi->s_mmp_tsk); 5931 } else { 5932 /* Make sure we can mount this feature set readwrite */ 5933 if (ext4_has_feature_readonly(sb) || 5934 !ext4_feature_set_ok(sb, 0)) { 5935 err = -EROFS; 5936 goto restore_opts; 5937 } 5938 /* 5939 * Make sure the group descriptor checksums 5940 * are sane. If they aren't, refuse to remount r/w. 5941 */ 5942 for (g = 0; g < sbi->s_groups_count; g++) { 5943 struct ext4_group_desc *gdp = 5944 ext4_get_group_desc(sb, g, NULL); 5945 5946 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 5947 ext4_msg(sb, KERN_ERR, 5948 "ext4_remount: Checksum for group %u failed (%u!=%u)", 5949 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 5950 le16_to_cpu(gdp->bg_checksum)); 5951 err = -EFSBADCRC; 5952 goto restore_opts; 5953 } 5954 } 5955 5956 /* 5957 * If we have an unprocessed orphan list hanging 5958 * around from a previously readonly bdev mount, 5959 * require a full umount/remount for now. 5960 */ 5961 if (es->s_last_orphan) { 5962 ext4_msg(sb, KERN_WARNING, "Couldn't " 5963 "remount RDWR because of unprocessed " 5964 "orphan inode list. Please " 5965 "umount/remount instead"); 5966 err = -EINVAL; 5967 goto restore_opts; 5968 } 5969 5970 /* 5971 * Mounting a RDONLY partition read-write, so reread 5972 * and store the current valid flag. (It may have 5973 * been changed by e2fsck since we originally mounted 5974 * the partition.) 5975 */ 5976 if (sbi->s_journal) { 5977 err = ext4_clear_journal_err(sb, es); 5978 if (err) 5979 goto restore_opts; 5980 } 5981 sbi->s_mount_state = le16_to_cpu(es->s_state); 5982 5983 err = ext4_setup_super(sb, es, 0); 5984 if (err) 5985 goto restore_opts; 5986 5987 sb->s_flags &= ~SB_RDONLY; 5988 if (ext4_has_feature_mmp(sb)) 5989 if (ext4_multi_mount_protect(sb, 5990 le64_to_cpu(es->s_mmp_block))) { 5991 err = -EROFS; 5992 goto restore_opts; 5993 } 5994 enable_quota = 1; 5995 } 5996 } 5997 5998 /* 5999 * Reinitialize lazy itable initialization thread based on 6000 * current settings 6001 */ 6002 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6003 ext4_unregister_li_request(sb); 6004 else { 6005 ext4_group_t first_not_zeroed; 6006 first_not_zeroed = ext4_has_uninit_itable(sb); 6007 ext4_register_li_request(sb, first_not_zeroed); 6008 } 6009 6010 /* 6011 * Handle creation of system zone data early because it can fail. 6012 * Releasing of existing data is done when we are sure remount will 6013 * succeed. 6014 */ 6015 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6016 err = ext4_setup_system_zone(sb); 6017 if (err) 6018 goto restore_opts; 6019 } 6020 6021 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6022 err = ext4_commit_super(sb, 1); 6023 if (err) 6024 goto restore_opts; 6025 } 6026 6027 #ifdef CONFIG_QUOTA 6028 /* Release old quota file names */ 6029 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6030 kfree(old_opts.s_qf_names[i]); 6031 if (enable_quota) { 6032 if (sb_any_quota_suspended(sb)) 6033 dquot_resume(sb, -1); 6034 else if (ext4_has_feature_quota(sb)) { 6035 err = ext4_enable_quotas(sb); 6036 if (err) 6037 goto restore_opts; 6038 } 6039 } 6040 #endif 6041 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6042 ext4_release_system_zone(sb); 6043 6044 /* 6045 * Some options can be enabled by ext4 and/or by VFS mount flag 6046 * either way we need to make sure it matches in both *flags and 6047 * s_flags. Copy those selected flags from s_flags to *flags 6048 */ 6049 *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags); 6050 6051 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data); 6052 kfree(orig_data); 6053 return 0; 6054 6055 restore_opts: 6056 sb->s_flags = old_sb_flags; 6057 sbi->s_mount_opt = old_opts.s_mount_opt; 6058 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6059 sbi->s_resuid = old_opts.s_resuid; 6060 sbi->s_resgid = old_opts.s_resgid; 6061 sbi->s_commit_interval = old_opts.s_commit_interval; 6062 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6063 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6064 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6065 ext4_release_system_zone(sb); 6066 #ifdef CONFIG_QUOTA 6067 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6068 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6069 to_free[i] = get_qf_name(sb, sbi, i); 6070 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6071 } 6072 synchronize_rcu(); 6073 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6074 kfree(to_free[i]); 6075 #endif 6076 kfree(orig_data); 6077 return err; 6078 } 6079 6080 #ifdef CONFIG_QUOTA 6081 static int ext4_statfs_project(struct super_block *sb, 6082 kprojid_t projid, struct kstatfs *buf) 6083 { 6084 struct kqid qid; 6085 struct dquot *dquot; 6086 u64 limit; 6087 u64 curblock; 6088 6089 qid = make_kqid_projid(projid); 6090 dquot = dqget(sb, qid); 6091 if (IS_ERR(dquot)) 6092 return PTR_ERR(dquot); 6093 spin_lock(&dquot->dq_dqb_lock); 6094 6095 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6096 dquot->dq_dqb.dqb_bhardlimit); 6097 limit >>= sb->s_blocksize_bits; 6098 6099 if (limit && buf->f_blocks > limit) { 6100 curblock = (dquot->dq_dqb.dqb_curspace + 6101 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6102 buf->f_blocks = limit; 6103 buf->f_bfree = buf->f_bavail = 6104 (buf->f_blocks > curblock) ? 6105 (buf->f_blocks - curblock) : 0; 6106 } 6107 6108 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6109 dquot->dq_dqb.dqb_ihardlimit); 6110 if (limit && buf->f_files > limit) { 6111 buf->f_files = limit; 6112 buf->f_ffree = 6113 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6114 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6115 } 6116 6117 spin_unlock(&dquot->dq_dqb_lock); 6118 dqput(dquot); 6119 return 0; 6120 } 6121 #endif 6122 6123 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6124 { 6125 struct super_block *sb = dentry->d_sb; 6126 struct ext4_sb_info *sbi = EXT4_SB(sb); 6127 struct ext4_super_block *es = sbi->s_es; 6128 ext4_fsblk_t overhead = 0, resv_blocks; 6129 u64 fsid; 6130 s64 bfree; 6131 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6132 6133 if (!test_opt(sb, MINIX_DF)) 6134 overhead = sbi->s_overhead; 6135 6136 buf->f_type = EXT4_SUPER_MAGIC; 6137 buf->f_bsize = sb->s_blocksize; 6138 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6139 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6140 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6141 /* prevent underflow in case that few free space is available */ 6142 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6143 buf->f_bavail = buf->f_bfree - 6144 (ext4_r_blocks_count(es) + resv_blocks); 6145 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6146 buf->f_bavail = 0; 6147 buf->f_files = le32_to_cpu(es->s_inodes_count); 6148 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6149 buf->f_namelen = EXT4_NAME_LEN; 6150 fsid = le64_to_cpup((void *)es->s_uuid) ^ 6151 le64_to_cpup((void *)es->s_uuid + sizeof(u64)); 6152 buf->f_fsid = u64_to_fsid(fsid); 6153 6154 #ifdef CONFIG_QUOTA 6155 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6156 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6157 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6158 #endif 6159 return 0; 6160 } 6161 6162 6163 #ifdef CONFIG_QUOTA 6164 6165 /* 6166 * Helper functions so that transaction is started before we acquire dqio_sem 6167 * to keep correct lock ordering of transaction > dqio_sem 6168 */ 6169 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6170 { 6171 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6172 } 6173 6174 static int ext4_write_dquot(struct dquot *dquot) 6175 { 6176 int ret, err; 6177 handle_t *handle; 6178 struct inode *inode; 6179 6180 inode = dquot_to_inode(dquot); 6181 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6182 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6183 if (IS_ERR(handle)) 6184 return PTR_ERR(handle); 6185 ret = dquot_commit(dquot); 6186 err = ext4_journal_stop(handle); 6187 if (!ret) 6188 ret = err; 6189 return ret; 6190 } 6191 6192 static int ext4_acquire_dquot(struct dquot *dquot) 6193 { 6194 int ret, err; 6195 handle_t *handle; 6196 6197 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6198 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6199 if (IS_ERR(handle)) 6200 return PTR_ERR(handle); 6201 ret = dquot_acquire(dquot); 6202 err = ext4_journal_stop(handle); 6203 if (!ret) 6204 ret = err; 6205 return ret; 6206 } 6207 6208 static int ext4_release_dquot(struct dquot *dquot) 6209 { 6210 int ret, err; 6211 handle_t *handle; 6212 6213 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6214 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6215 if (IS_ERR(handle)) { 6216 /* Release dquot anyway to avoid endless cycle in dqput() */ 6217 dquot_release(dquot); 6218 return PTR_ERR(handle); 6219 } 6220 ret = dquot_release(dquot); 6221 err = ext4_journal_stop(handle); 6222 if (!ret) 6223 ret = err; 6224 return ret; 6225 } 6226 6227 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6228 { 6229 struct super_block *sb = dquot->dq_sb; 6230 struct ext4_sb_info *sbi = EXT4_SB(sb); 6231 6232 /* Are we journaling quotas? */ 6233 if (ext4_has_feature_quota(sb) || 6234 sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) { 6235 dquot_mark_dquot_dirty(dquot); 6236 return ext4_write_dquot(dquot); 6237 } else { 6238 return dquot_mark_dquot_dirty(dquot); 6239 } 6240 } 6241 6242 static int ext4_write_info(struct super_block *sb, int type) 6243 { 6244 int ret, err; 6245 handle_t *handle; 6246 6247 /* Data block + inode block */ 6248 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 6249 if (IS_ERR(handle)) 6250 return PTR_ERR(handle); 6251 ret = dquot_commit_info(sb, type); 6252 err = ext4_journal_stop(handle); 6253 if (!ret) 6254 ret = err; 6255 return ret; 6256 } 6257 6258 /* 6259 * Turn on quotas during mount time - we need to find 6260 * the quota file and such... 6261 */ 6262 static int ext4_quota_on_mount(struct super_block *sb, int type) 6263 { 6264 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 6265 EXT4_SB(sb)->s_jquota_fmt, type); 6266 } 6267 6268 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6269 { 6270 struct ext4_inode_info *ei = EXT4_I(inode); 6271 6272 /* The first argument of lockdep_set_subclass has to be 6273 * *exactly* the same as the argument to init_rwsem() --- in 6274 * this case, in init_once() --- or lockdep gets unhappy 6275 * because the name of the lock is set using the 6276 * stringification of the argument to init_rwsem(). 6277 */ 6278 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6279 lockdep_set_subclass(&ei->i_data_sem, subclass); 6280 } 6281 6282 /* 6283 * Standard function to be called on quota_on 6284 */ 6285 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6286 const struct path *path) 6287 { 6288 int err; 6289 6290 if (!test_opt(sb, QUOTA)) 6291 return -EINVAL; 6292 6293 /* Quotafile not on the same filesystem? */ 6294 if (path->dentry->d_sb != sb) 6295 return -EXDEV; 6296 6297 /* Quota already enabled for this file? */ 6298 if (IS_NOQUOTA(d_inode(path->dentry))) 6299 return -EBUSY; 6300 6301 /* Journaling quota? */ 6302 if (EXT4_SB(sb)->s_qf_names[type]) { 6303 /* Quotafile not in fs root? */ 6304 if (path->dentry->d_parent != sb->s_root) 6305 ext4_msg(sb, KERN_WARNING, 6306 "Quota file not on filesystem root. " 6307 "Journaled quota will not work"); 6308 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6309 } else { 6310 /* 6311 * Clear the flag just in case mount options changed since 6312 * last time. 6313 */ 6314 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6315 } 6316 6317 /* 6318 * When we journal data on quota file, we have to flush journal to see 6319 * all updates to the file when we bypass pagecache... 6320 */ 6321 if (EXT4_SB(sb)->s_journal && 6322 ext4_should_journal_data(d_inode(path->dentry))) { 6323 /* 6324 * We don't need to lock updates but journal_flush() could 6325 * otherwise be livelocked... 6326 */ 6327 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6328 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 6329 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6330 if (err) 6331 return err; 6332 } 6333 6334 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6335 err = dquot_quota_on(sb, type, format_id, path); 6336 if (err) { 6337 lockdep_set_quota_inode(path->dentry->d_inode, 6338 I_DATA_SEM_NORMAL); 6339 } else { 6340 struct inode *inode = d_inode(path->dentry); 6341 handle_t *handle; 6342 6343 /* 6344 * Set inode flags to prevent userspace from messing with quota 6345 * files. If this fails, we return success anyway since quotas 6346 * are already enabled and this is not a hard failure. 6347 */ 6348 inode_lock(inode); 6349 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6350 if (IS_ERR(handle)) 6351 goto unlock_inode; 6352 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6353 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6354 S_NOATIME | S_IMMUTABLE); 6355 err = ext4_mark_inode_dirty(handle, inode); 6356 ext4_journal_stop(handle); 6357 unlock_inode: 6358 inode_unlock(inode); 6359 } 6360 return err; 6361 } 6362 6363 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6364 unsigned int flags) 6365 { 6366 int err; 6367 struct inode *qf_inode; 6368 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6369 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6370 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6371 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6372 }; 6373 6374 BUG_ON(!ext4_has_feature_quota(sb)); 6375 6376 if (!qf_inums[type]) 6377 return -EPERM; 6378 6379 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6380 if (IS_ERR(qf_inode)) { 6381 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6382 return PTR_ERR(qf_inode); 6383 } 6384 6385 /* Don't account quota for quota files to avoid recursion */ 6386 qf_inode->i_flags |= S_NOQUOTA; 6387 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6388 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6389 if (err) 6390 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6391 iput(qf_inode); 6392 6393 return err; 6394 } 6395 6396 /* Enable usage tracking for all quota types. */ 6397 static int ext4_enable_quotas(struct super_block *sb) 6398 { 6399 int type, err = 0; 6400 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6401 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6402 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6403 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6404 }; 6405 bool quota_mopt[EXT4_MAXQUOTAS] = { 6406 test_opt(sb, USRQUOTA), 6407 test_opt(sb, GRPQUOTA), 6408 test_opt(sb, PRJQUOTA), 6409 }; 6410 6411 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6412 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6413 if (qf_inums[type]) { 6414 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6415 DQUOT_USAGE_ENABLED | 6416 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6417 if (err) { 6418 ext4_warning(sb, 6419 "Failed to enable quota tracking " 6420 "(type=%d, err=%d). Please run " 6421 "e2fsck to fix.", type, err); 6422 for (type--; type >= 0; type--) 6423 dquot_quota_off(sb, type); 6424 6425 return err; 6426 } 6427 } 6428 } 6429 return 0; 6430 } 6431 6432 static int ext4_quota_off(struct super_block *sb, int type) 6433 { 6434 struct inode *inode = sb_dqopt(sb)->files[type]; 6435 handle_t *handle; 6436 int err; 6437 6438 /* Force all delayed allocation blocks to be allocated. 6439 * Caller already holds s_umount sem */ 6440 if (test_opt(sb, DELALLOC)) 6441 sync_filesystem(sb); 6442 6443 if (!inode || !igrab(inode)) 6444 goto out; 6445 6446 err = dquot_quota_off(sb, type); 6447 if (err || ext4_has_feature_quota(sb)) 6448 goto out_put; 6449 6450 inode_lock(inode); 6451 /* 6452 * Update modification times of quota files when userspace can 6453 * start looking at them. If we fail, we return success anyway since 6454 * this is not a hard failure and quotas are already disabled. 6455 */ 6456 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6457 if (IS_ERR(handle)) { 6458 err = PTR_ERR(handle); 6459 goto out_unlock; 6460 } 6461 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 6462 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 6463 inode->i_mtime = inode->i_ctime = current_time(inode); 6464 err = ext4_mark_inode_dirty(handle, inode); 6465 ext4_journal_stop(handle); 6466 out_unlock: 6467 inode_unlock(inode); 6468 out_put: 6469 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 6470 iput(inode); 6471 return err; 6472 out: 6473 return dquot_quota_off(sb, type); 6474 } 6475 6476 /* Read data from quotafile - avoid pagecache and such because we cannot afford 6477 * acquiring the locks... As quota files are never truncated and quota code 6478 * itself serializes the operations (and no one else should touch the files) 6479 * we don't have to be afraid of races */ 6480 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 6481 size_t len, loff_t off) 6482 { 6483 struct inode *inode = sb_dqopt(sb)->files[type]; 6484 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6485 int offset = off & (sb->s_blocksize - 1); 6486 int tocopy; 6487 size_t toread; 6488 struct buffer_head *bh; 6489 loff_t i_size = i_size_read(inode); 6490 6491 if (off > i_size) 6492 return 0; 6493 if (off+len > i_size) 6494 len = i_size-off; 6495 toread = len; 6496 while (toread > 0) { 6497 tocopy = sb->s_blocksize - offset < toread ? 6498 sb->s_blocksize - offset : toread; 6499 bh = ext4_bread(NULL, inode, blk, 0); 6500 if (IS_ERR(bh)) 6501 return PTR_ERR(bh); 6502 if (!bh) /* A hole? */ 6503 memset(data, 0, tocopy); 6504 else 6505 memcpy(data, bh->b_data+offset, tocopy); 6506 brelse(bh); 6507 offset = 0; 6508 toread -= tocopy; 6509 data += tocopy; 6510 blk++; 6511 } 6512 return len; 6513 } 6514 6515 /* Write to quotafile (we know the transaction is already started and has 6516 * enough credits) */ 6517 static ssize_t ext4_quota_write(struct super_block *sb, int type, 6518 const char *data, size_t len, loff_t off) 6519 { 6520 struct inode *inode = sb_dqopt(sb)->files[type]; 6521 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6522 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 6523 int retries = 0; 6524 struct buffer_head *bh; 6525 handle_t *handle = journal_current_handle(); 6526 6527 if (EXT4_SB(sb)->s_journal && !handle) { 6528 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6529 " cancelled because transaction is not started", 6530 (unsigned long long)off, (unsigned long long)len); 6531 return -EIO; 6532 } 6533 /* 6534 * Since we account only one data block in transaction credits, 6535 * then it is impossible to cross a block boundary. 6536 */ 6537 if (sb->s_blocksize - offset < len) { 6538 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6539 " cancelled because not block aligned", 6540 (unsigned long long)off, (unsigned long long)len); 6541 return -EIO; 6542 } 6543 6544 do { 6545 bh = ext4_bread(handle, inode, blk, 6546 EXT4_GET_BLOCKS_CREATE | 6547 EXT4_GET_BLOCKS_METADATA_NOFAIL); 6548 } while (PTR_ERR(bh) == -ENOSPC && 6549 ext4_should_retry_alloc(inode->i_sb, &retries)); 6550 if (IS_ERR(bh)) 6551 return PTR_ERR(bh); 6552 if (!bh) 6553 goto out; 6554 BUFFER_TRACE(bh, "get write access"); 6555 err = ext4_journal_get_write_access(handle, bh); 6556 if (err) { 6557 brelse(bh); 6558 return err; 6559 } 6560 lock_buffer(bh); 6561 memcpy(bh->b_data+offset, data, len); 6562 flush_dcache_page(bh->b_page); 6563 unlock_buffer(bh); 6564 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6565 brelse(bh); 6566 out: 6567 if (inode->i_size < off + len) { 6568 i_size_write(inode, off + len); 6569 EXT4_I(inode)->i_disksize = inode->i_size; 6570 err2 = ext4_mark_inode_dirty(handle, inode); 6571 if (unlikely(err2 && !err)) 6572 err = err2; 6573 } 6574 return err ? err : len; 6575 } 6576 #endif 6577 6578 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6579 const char *dev_name, void *data) 6580 { 6581 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6582 } 6583 6584 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6585 static inline void register_as_ext2(void) 6586 { 6587 int err = register_filesystem(&ext2_fs_type); 6588 if (err) 6589 printk(KERN_WARNING 6590 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6591 } 6592 6593 static inline void unregister_as_ext2(void) 6594 { 6595 unregister_filesystem(&ext2_fs_type); 6596 } 6597 6598 static inline int ext2_feature_set_ok(struct super_block *sb) 6599 { 6600 if (ext4_has_unknown_ext2_incompat_features(sb)) 6601 return 0; 6602 if (sb_rdonly(sb)) 6603 return 1; 6604 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6605 return 0; 6606 return 1; 6607 } 6608 #else 6609 static inline void register_as_ext2(void) { } 6610 static inline void unregister_as_ext2(void) { } 6611 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6612 #endif 6613 6614 static inline void register_as_ext3(void) 6615 { 6616 int err = register_filesystem(&ext3_fs_type); 6617 if (err) 6618 printk(KERN_WARNING 6619 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6620 } 6621 6622 static inline void unregister_as_ext3(void) 6623 { 6624 unregister_filesystem(&ext3_fs_type); 6625 } 6626 6627 static inline int ext3_feature_set_ok(struct super_block *sb) 6628 { 6629 if (ext4_has_unknown_ext3_incompat_features(sb)) 6630 return 0; 6631 if (!ext4_has_feature_journal(sb)) 6632 return 0; 6633 if (sb_rdonly(sb)) 6634 return 1; 6635 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6636 return 0; 6637 return 1; 6638 } 6639 6640 static struct file_system_type ext4_fs_type = { 6641 .owner = THIS_MODULE, 6642 .name = "ext4", 6643 .mount = ext4_mount, 6644 .kill_sb = kill_block_super, 6645 .fs_flags = FS_REQUIRES_DEV, 6646 }; 6647 MODULE_ALIAS_FS("ext4"); 6648 6649 /* Shared across all ext4 file systems */ 6650 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6651 6652 static int __init ext4_init_fs(void) 6653 { 6654 int i, err; 6655 6656 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6657 ext4_li_info = NULL; 6658 mutex_init(&ext4_li_mtx); 6659 6660 /* Build-time check for flags consistency */ 6661 ext4_check_flag_values(); 6662 6663 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6664 init_waitqueue_head(&ext4__ioend_wq[i]); 6665 6666 err = ext4_init_es(); 6667 if (err) 6668 return err; 6669 6670 err = ext4_init_pending(); 6671 if (err) 6672 goto out7; 6673 6674 err = ext4_init_post_read_processing(); 6675 if (err) 6676 goto out6; 6677 6678 err = ext4_init_pageio(); 6679 if (err) 6680 goto out5; 6681 6682 err = ext4_init_system_zone(); 6683 if (err) 6684 goto out4; 6685 6686 err = ext4_init_sysfs(); 6687 if (err) 6688 goto out3; 6689 6690 err = ext4_init_mballoc(); 6691 if (err) 6692 goto out2; 6693 err = init_inodecache(); 6694 if (err) 6695 goto out1; 6696 6697 err = ext4_fc_init_dentry_cache(); 6698 if (err) 6699 goto out05; 6700 6701 register_as_ext3(); 6702 register_as_ext2(); 6703 err = register_filesystem(&ext4_fs_type); 6704 if (err) 6705 goto out; 6706 6707 return 0; 6708 out: 6709 unregister_as_ext2(); 6710 unregister_as_ext3(); 6711 out05: 6712 destroy_inodecache(); 6713 out1: 6714 ext4_exit_mballoc(); 6715 out2: 6716 ext4_exit_sysfs(); 6717 out3: 6718 ext4_exit_system_zone(); 6719 out4: 6720 ext4_exit_pageio(); 6721 out5: 6722 ext4_exit_post_read_processing(); 6723 out6: 6724 ext4_exit_pending(); 6725 out7: 6726 ext4_exit_es(); 6727 6728 return err; 6729 } 6730 6731 static void __exit ext4_exit_fs(void) 6732 { 6733 ext4_destroy_lazyinit_thread(); 6734 unregister_as_ext2(); 6735 unregister_as_ext3(); 6736 unregister_filesystem(&ext4_fs_type); 6737 destroy_inodecache(); 6738 ext4_exit_mballoc(); 6739 ext4_exit_sysfs(); 6740 ext4_exit_system_zone(); 6741 ext4_exit_pageio(); 6742 ext4_exit_post_read_processing(); 6743 ext4_exit_es(); 6744 ext4_exit_pending(); 6745 } 6746 6747 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6748 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6749 MODULE_LICENSE("GPL"); 6750 MODULE_SOFTDEP("pre: crc32c"); 6751 module_init(ext4_init_fs) 6752 module_exit(ext4_exit_fs) 6753