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 es->s_checksum = ext4_superblock_csum(sb, es); 293 } 294 295 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 296 struct ext4_group_desc *bg) 297 { 298 return le32_to_cpu(bg->bg_block_bitmap_lo) | 299 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 300 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 301 } 302 303 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 304 struct ext4_group_desc *bg) 305 { 306 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 307 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 308 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 309 } 310 311 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 312 struct ext4_group_desc *bg) 313 { 314 return le32_to_cpu(bg->bg_inode_table_lo) | 315 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 316 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 317 } 318 319 __u32 ext4_free_group_clusters(struct super_block *sb, 320 struct ext4_group_desc *bg) 321 { 322 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 323 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 324 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 325 } 326 327 __u32 ext4_free_inodes_count(struct super_block *sb, 328 struct ext4_group_desc *bg) 329 { 330 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 331 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 332 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 333 } 334 335 __u32 ext4_used_dirs_count(struct super_block *sb, 336 struct ext4_group_desc *bg) 337 { 338 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 339 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 340 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 341 } 342 343 __u32 ext4_itable_unused_count(struct super_block *sb, 344 struct ext4_group_desc *bg) 345 { 346 return le16_to_cpu(bg->bg_itable_unused_lo) | 347 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 348 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 349 } 350 351 void ext4_block_bitmap_set(struct super_block *sb, 352 struct ext4_group_desc *bg, ext4_fsblk_t blk) 353 { 354 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 355 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 356 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 357 } 358 359 void ext4_inode_bitmap_set(struct super_block *sb, 360 struct ext4_group_desc *bg, ext4_fsblk_t blk) 361 { 362 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 363 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 364 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 365 } 366 367 void ext4_inode_table_set(struct super_block *sb, 368 struct ext4_group_desc *bg, ext4_fsblk_t blk) 369 { 370 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 371 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 372 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 373 } 374 375 void ext4_free_group_clusters_set(struct super_block *sb, 376 struct ext4_group_desc *bg, __u32 count) 377 { 378 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 379 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 380 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 381 } 382 383 void ext4_free_inodes_set(struct super_block *sb, 384 struct ext4_group_desc *bg, __u32 count) 385 { 386 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 387 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 388 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 389 } 390 391 void ext4_used_dirs_set(struct super_block *sb, 392 struct ext4_group_desc *bg, __u32 count) 393 { 394 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 395 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 396 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 397 } 398 399 void ext4_itable_unused_set(struct super_block *sb, 400 struct ext4_group_desc *bg, __u32 count) 401 { 402 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 403 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 404 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 405 } 406 407 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi) 408 { 409 time64_t now = ktime_get_real_seconds(); 410 411 now = clamp_val(now, 0, (1ull << 40) - 1); 412 413 *lo = cpu_to_le32(lower_32_bits(now)); 414 *hi = upper_32_bits(now); 415 } 416 417 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 418 { 419 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 420 } 421 #define ext4_update_tstamp(es, tstamp) \ 422 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 423 #define ext4_get_tstamp(es, tstamp) \ 424 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 425 426 static void __save_error_info(struct super_block *sb, int error, 427 __u32 ino, __u64 block, 428 const char *func, unsigned int line) 429 { 430 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 431 int err; 432 433 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 434 if (bdev_read_only(sb->s_bdev)) 435 return; 436 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 437 ext4_update_tstamp(es, s_last_error_time); 438 strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func)); 439 es->s_last_error_line = cpu_to_le32(line); 440 es->s_last_error_ino = cpu_to_le32(ino); 441 es->s_last_error_block = cpu_to_le64(block); 442 switch (error) { 443 case EIO: 444 err = EXT4_ERR_EIO; 445 break; 446 case ENOMEM: 447 err = EXT4_ERR_ENOMEM; 448 break; 449 case EFSBADCRC: 450 err = EXT4_ERR_EFSBADCRC; 451 break; 452 case 0: 453 case EFSCORRUPTED: 454 err = EXT4_ERR_EFSCORRUPTED; 455 break; 456 case ENOSPC: 457 err = EXT4_ERR_ENOSPC; 458 break; 459 case ENOKEY: 460 err = EXT4_ERR_ENOKEY; 461 break; 462 case EROFS: 463 err = EXT4_ERR_EROFS; 464 break; 465 case EFBIG: 466 err = EXT4_ERR_EFBIG; 467 break; 468 case EEXIST: 469 err = EXT4_ERR_EEXIST; 470 break; 471 case ERANGE: 472 err = EXT4_ERR_ERANGE; 473 break; 474 case EOVERFLOW: 475 err = EXT4_ERR_EOVERFLOW; 476 break; 477 case EBUSY: 478 err = EXT4_ERR_EBUSY; 479 break; 480 case ENOTDIR: 481 err = EXT4_ERR_ENOTDIR; 482 break; 483 case ENOTEMPTY: 484 err = EXT4_ERR_ENOTEMPTY; 485 break; 486 case ESHUTDOWN: 487 err = EXT4_ERR_ESHUTDOWN; 488 break; 489 case EFAULT: 490 err = EXT4_ERR_EFAULT; 491 break; 492 default: 493 err = EXT4_ERR_UNKNOWN; 494 } 495 es->s_last_error_errcode = err; 496 if (!es->s_first_error_time) { 497 es->s_first_error_time = es->s_last_error_time; 498 es->s_first_error_time_hi = es->s_last_error_time_hi; 499 strncpy(es->s_first_error_func, func, 500 sizeof(es->s_first_error_func)); 501 es->s_first_error_line = cpu_to_le32(line); 502 es->s_first_error_ino = es->s_last_error_ino; 503 es->s_first_error_block = es->s_last_error_block; 504 es->s_first_error_errcode = es->s_last_error_errcode; 505 } 506 /* 507 * Start the daily error reporting function if it hasn't been 508 * started already 509 */ 510 if (!es->s_error_count) 511 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ); 512 le32_add_cpu(&es->s_error_count, 1); 513 } 514 515 static void save_error_info(struct super_block *sb, int error, 516 __u32 ino, __u64 block, 517 const char *func, unsigned int line) 518 { 519 __save_error_info(sb, error, ino, block, func, line); 520 if (!bdev_read_only(sb->s_bdev)) 521 ext4_commit_super(sb, 1); 522 } 523 524 /* 525 * The del_gendisk() function uninitializes the disk-specific data 526 * structures, including the bdi structure, without telling anyone 527 * else. Once this happens, any attempt to call mark_buffer_dirty() 528 * (for example, by ext4_commit_super), will cause a kernel OOPS. 529 * This is a kludge to prevent these oops until we can put in a proper 530 * hook in del_gendisk() to inform the VFS and file system layers. 531 */ 532 static int block_device_ejected(struct super_block *sb) 533 { 534 struct inode *bd_inode = sb->s_bdev->bd_inode; 535 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 536 537 return bdi->dev == NULL; 538 } 539 540 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 541 { 542 struct super_block *sb = journal->j_private; 543 struct ext4_sb_info *sbi = EXT4_SB(sb); 544 int error = is_journal_aborted(journal); 545 struct ext4_journal_cb_entry *jce; 546 547 BUG_ON(txn->t_state == T_FINISHED); 548 549 ext4_process_freed_data(sb, txn->t_tid); 550 551 spin_lock(&sbi->s_md_lock); 552 while (!list_empty(&txn->t_private_list)) { 553 jce = list_entry(txn->t_private_list.next, 554 struct ext4_journal_cb_entry, jce_list); 555 list_del_init(&jce->jce_list); 556 spin_unlock(&sbi->s_md_lock); 557 jce->jce_func(sb, jce, error); 558 spin_lock(&sbi->s_md_lock); 559 } 560 spin_unlock(&sbi->s_md_lock); 561 } 562 563 /* 564 * This writepage callback for write_cache_pages() 565 * takes care of a few cases after page cleaning. 566 * 567 * write_cache_pages() already checks for dirty pages 568 * and calls clear_page_dirty_for_io(), which we want, 569 * to write protect the pages. 570 * 571 * However, we may have to redirty a page (see below.) 572 */ 573 static int ext4_journalled_writepage_callback(struct page *page, 574 struct writeback_control *wbc, 575 void *data) 576 { 577 transaction_t *transaction = (transaction_t *) data; 578 struct buffer_head *bh, *head; 579 struct journal_head *jh; 580 581 bh = head = page_buffers(page); 582 do { 583 /* 584 * We have to redirty a page in these cases: 585 * 1) If buffer is dirty, it means the page was dirty because it 586 * contains a buffer that needs checkpointing. So the dirty bit 587 * needs to be preserved so that checkpointing writes the buffer 588 * properly. 589 * 2) If buffer is not part of the committing transaction 590 * (we may have just accidentally come across this buffer because 591 * inode range tracking is not exact) or if the currently running 592 * transaction already contains this buffer as well, dirty bit 593 * needs to be preserved so that the buffer gets writeprotected 594 * properly on running transaction's commit. 595 */ 596 jh = bh2jh(bh); 597 if (buffer_dirty(bh) || 598 (jh && (jh->b_transaction != transaction || 599 jh->b_next_transaction))) { 600 redirty_page_for_writepage(wbc, page); 601 goto out; 602 } 603 } while ((bh = bh->b_this_page) != head); 604 605 out: 606 return AOP_WRITEPAGE_ACTIVATE; 607 } 608 609 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 610 { 611 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 612 struct writeback_control wbc = { 613 .sync_mode = WB_SYNC_ALL, 614 .nr_to_write = LONG_MAX, 615 .range_start = jinode->i_dirty_start, 616 .range_end = jinode->i_dirty_end, 617 }; 618 619 return write_cache_pages(mapping, &wbc, 620 ext4_journalled_writepage_callback, 621 jinode->i_transaction); 622 } 623 624 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 625 { 626 int ret; 627 628 if (ext4_should_journal_data(jinode->i_vfs_inode)) 629 ret = ext4_journalled_submit_inode_data_buffers(jinode); 630 else 631 ret = jbd2_journal_submit_inode_data_buffers(jinode); 632 633 return ret; 634 } 635 636 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 637 { 638 int ret = 0; 639 640 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 641 ret = jbd2_journal_finish_inode_data_buffers(jinode); 642 643 return ret; 644 } 645 646 static bool system_going_down(void) 647 { 648 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 649 || system_state == SYSTEM_RESTART; 650 } 651 652 /* Deal with the reporting of failure conditions on a filesystem such as 653 * inconsistencies detected or read IO failures. 654 * 655 * On ext2, we can store the error state of the filesystem in the 656 * superblock. That is not possible on ext4, because we may have other 657 * write ordering constraints on the superblock which prevent us from 658 * writing it out straight away; and given that the journal is about to 659 * be aborted, we can't rely on the current, or future, transactions to 660 * write out the superblock safely. 661 * 662 * We'll just use the jbd2_journal_abort() error code to record an error in 663 * the journal instead. On recovery, the journal will complain about 664 * that error until we've noted it down and cleared it. 665 */ 666 667 static void ext4_handle_error(struct super_block *sb) 668 { 669 if (test_opt(sb, WARN_ON_ERROR)) 670 WARN_ON_ONCE(1); 671 672 if (sb_rdonly(sb)) 673 return; 674 675 if (!test_opt(sb, ERRORS_CONT)) { 676 journal_t *journal = EXT4_SB(sb)->s_journal; 677 678 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 679 if (journal) 680 jbd2_journal_abort(journal, -EIO); 681 } 682 /* 683 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 684 * could panic during 'reboot -f' as the underlying device got already 685 * disabled. 686 */ 687 if (test_opt(sb, ERRORS_RO) || system_going_down()) { 688 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 689 /* 690 * Make sure updated value of ->s_mount_flags will be visible 691 * before ->s_flags update 692 */ 693 smp_wmb(); 694 sb->s_flags |= SB_RDONLY; 695 } else if (test_opt(sb, ERRORS_PANIC)) { 696 panic("EXT4-fs (device %s): panic forced after error\n", 697 sb->s_id); 698 } 699 } 700 701 #define ext4_error_ratelimit(sb) \ 702 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 703 "EXT4-fs error") 704 705 void __ext4_error(struct super_block *sb, const char *function, 706 unsigned int line, int error, __u64 block, 707 const char *fmt, ...) 708 { 709 struct va_format vaf; 710 va_list args; 711 712 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 713 return; 714 715 trace_ext4_error(sb, function, line); 716 if (ext4_error_ratelimit(sb)) { 717 va_start(args, fmt); 718 vaf.fmt = fmt; 719 vaf.va = &args; 720 printk(KERN_CRIT 721 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 722 sb->s_id, function, line, current->comm, &vaf); 723 va_end(args); 724 } 725 save_error_info(sb, error, 0, block, function, line); 726 ext4_handle_error(sb); 727 } 728 729 void __ext4_error_inode(struct inode *inode, const char *function, 730 unsigned int line, ext4_fsblk_t block, int error, 731 const char *fmt, ...) 732 { 733 va_list args; 734 struct va_format vaf; 735 736 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 737 return; 738 739 trace_ext4_error(inode->i_sb, function, line); 740 if (ext4_error_ratelimit(inode->i_sb)) { 741 va_start(args, fmt); 742 vaf.fmt = fmt; 743 vaf.va = &args; 744 if (block) 745 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 746 "inode #%lu: block %llu: comm %s: %pV\n", 747 inode->i_sb->s_id, function, line, inode->i_ino, 748 block, current->comm, &vaf); 749 else 750 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 751 "inode #%lu: comm %s: %pV\n", 752 inode->i_sb->s_id, function, line, inode->i_ino, 753 current->comm, &vaf); 754 va_end(args); 755 } 756 save_error_info(inode->i_sb, error, inode->i_ino, block, 757 function, line); 758 ext4_handle_error(inode->i_sb); 759 } 760 761 void __ext4_error_file(struct file *file, const char *function, 762 unsigned int line, ext4_fsblk_t block, 763 const char *fmt, ...) 764 { 765 va_list args; 766 struct va_format vaf; 767 struct inode *inode = file_inode(file); 768 char pathname[80], *path; 769 770 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 771 return; 772 773 trace_ext4_error(inode->i_sb, function, line); 774 if (ext4_error_ratelimit(inode->i_sb)) { 775 path = file_path(file, pathname, sizeof(pathname)); 776 if (IS_ERR(path)) 777 path = "(unknown)"; 778 va_start(args, fmt); 779 vaf.fmt = fmt; 780 vaf.va = &args; 781 if (block) 782 printk(KERN_CRIT 783 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 784 "block %llu: comm %s: path %s: %pV\n", 785 inode->i_sb->s_id, function, line, inode->i_ino, 786 block, current->comm, path, &vaf); 787 else 788 printk(KERN_CRIT 789 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 790 "comm %s: path %s: %pV\n", 791 inode->i_sb->s_id, function, line, inode->i_ino, 792 current->comm, path, &vaf); 793 va_end(args); 794 } 795 save_error_info(inode->i_sb, EFSCORRUPTED, inode->i_ino, block, 796 function, line); 797 ext4_handle_error(inode->i_sb); 798 } 799 800 const char *ext4_decode_error(struct super_block *sb, int errno, 801 char nbuf[16]) 802 { 803 char *errstr = NULL; 804 805 switch (errno) { 806 case -EFSCORRUPTED: 807 errstr = "Corrupt filesystem"; 808 break; 809 case -EFSBADCRC: 810 errstr = "Filesystem failed CRC"; 811 break; 812 case -EIO: 813 errstr = "IO failure"; 814 break; 815 case -ENOMEM: 816 errstr = "Out of memory"; 817 break; 818 case -EROFS: 819 if (!sb || (EXT4_SB(sb)->s_journal && 820 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 821 errstr = "Journal has aborted"; 822 else 823 errstr = "Readonly filesystem"; 824 break; 825 default: 826 /* If the caller passed in an extra buffer for unknown 827 * errors, textualise them now. Else we just return 828 * NULL. */ 829 if (nbuf) { 830 /* Check for truncated error codes... */ 831 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 832 errstr = nbuf; 833 } 834 break; 835 } 836 837 return errstr; 838 } 839 840 /* __ext4_std_error decodes expected errors from journaling functions 841 * automatically and invokes the appropriate error response. */ 842 843 void __ext4_std_error(struct super_block *sb, const char *function, 844 unsigned int line, int errno) 845 { 846 char nbuf[16]; 847 const char *errstr; 848 849 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 850 return; 851 852 /* Special case: if the error is EROFS, and we're not already 853 * inside a transaction, then there's really no point in logging 854 * an error. */ 855 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 856 return; 857 858 if (ext4_error_ratelimit(sb)) { 859 errstr = ext4_decode_error(sb, errno, nbuf); 860 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 861 sb->s_id, function, line, errstr); 862 } 863 864 save_error_info(sb, -errno, 0, 0, function, line); 865 ext4_handle_error(sb); 866 } 867 868 /* 869 * ext4_abort is a much stronger failure handler than ext4_error. The 870 * abort function may be used to deal with unrecoverable failures such 871 * as journal IO errors or ENOMEM at a critical moment in log management. 872 * 873 * We unconditionally force the filesystem into an ABORT|READONLY state, 874 * unless the error response on the fs has been set to panic in which 875 * case we take the easy way out and panic immediately. 876 */ 877 878 void __ext4_abort(struct super_block *sb, const char *function, 879 unsigned int line, int error, const char *fmt, ...) 880 { 881 struct va_format vaf; 882 va_list args; 883 884 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 885 return; 886 887 save_error_info(sb, error, 0, 0, function, line); 888 va_start(args, fmt); 889 vaf.fmt = fmt; 890 vaf.va = &args; 891 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n", 892 sb->s_id, function, line, &vaf); 893 va_end(args); 894 895 if (sb_rdonly(sb) == 0) { 896 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 897 if (EXT4_SB(sb)->s_journal) 898 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO); 899 900 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 901 /* 902 * Make sure updated value of ->s_mount_flags will be visible 903 * before ->s_flags update 904 */ 905 smp_wmb(); 906 sb->s_flags |= SB_RDONLY; 907 } 908 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) 909 panic("EXT4-fs panic from previous error\n"); 910 } 911 912 void __ext4_msg(struct super_block *sb, 913 const char *prefix, const char *fmt, ...) 914 { 915 struct va_format vaf; 916 va_list args; 917 918 atomic_inc(&EXT4_SB(sb)->s_msg_count); 919 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs")) 920 return; 921 922 va_start(args, fmt); 923 vaf.fmt = fmt; 924 vaf.va = &args; 925 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 926 va_end(args); 927 } 928 929 static int ext4_warning_ratelimit(struct super_block *sb) 930 { 931 atomic_inc(&EXT4_SB(sb)->s_warning_count); 932 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 933 "EXT4-fs warning"); 934 } 935 936 void __ext4_warning(struct super_block *sb, const char *function, 937 unsigned int line, const char *fmt, ...) 938 { 939 struct va_format vaf; 940 va_list args; 941 942 if (!ext4_warning_ratelimit(sb)) 943 return; 944 945 va_start(args, fmt); 946 vaf.fmt = fmt; 947 vaf.va = &args; 948 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 949 sb->s_id, function, line, &vaf); 950 va_end(args); 951 } 952 953 void __ext4_warning_inode(const struct inode *inode, const char *function, 954 unsigned int line, const char *fmt, ...) 955 { 956 struct va_format vaf; 957 va_list args; 958 959 if (!ext4_warning_ratelimit(inode->i_sb)) 960 return; 961 962 va_start(args, fmt); 963 vaf.fmt = fmt; 964 vaf.va = &args; 965 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 966 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 967 function, line, inode->i_ino, current->comm, &vaf); 968 va_end(args); 969 } 970 971 void __ext4_grp_locked_error(const char *function, unsigned int line, 972 struct super_block *sb, ext4_group_t grp, 973 unsigned long ino, ext4_fsblk_t block, 974 const char *fmt, ...) 975 __releases(bitlock) 976 __acquires(bitlock) 977 { 978 struct va_format vaf; 979 va_list args; 980 981 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 982 return; 983 984 trace_ext4_error(sb, function, line); 985 __save_error_info(sb, EFSCORRUPTED, ino, block, function, line); 986 987 if (ext4_error_ratelimit(sb)) { 988 va_start(args, fmt); 989 vaf.fmt = fmt; 990 vaf.va = &args; 991 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 992 sb->s_id, function, line, grp); 993 if (ino) 994 printk(KERN_CONT "inode %lu: ", ino); 995 if (block) 996 printk(KERN_CONT "block %llu:", 997 (unsigned long long) block); 998 printk(KERN_CONT "%pV\n", &vaf); 999 va_end(args); 1000 } 1001 1002 if (test_opt(sb, WARN_ON_ERROR)) 1003 WARN_ON_ONCE(1); 1004 1005 if (test_opt(sb, ERRORS_CONT)) { 1006 ext4_commit_super(sb, 0); 1007 return; 1008 } 1009 1010 ext4_unlock_group(sb, grp); 1011 ext4_commit_super(sb, 1); 1012 ext4_handle_error(sb); 1013 /* 1014 * We only get here in the ERRORS_RO case; relocking the group 1015 * may be dangerous, but nothing bad will happen since the 1016 * filesystem will have already been marked read/only and the 1017 * journal has been aborted. We return 1 as a hint to callers 1018 * who might what to use the return value from 1019 * ext4_grp_locked_error() to distinguish between the 1020 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1021 * aggressively from the ext4 function in question, with a 1022 * more appropriate error code. 1023 */ 1024 ext4_lock_group(sb, grp); 1025 return; 1026 } 1027 1028 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1029 ext4_group_t group, 1030 unsigned int flags) 1031 { 1032 struct ext4_sb_info *sbi = EXT4_SB(sb); 1033 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1034 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1035 int ret; 1036 1037 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1038 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1039 &grp->bb_state); 1040 if (!ret) 1041 percpu_counter_sub(&sbi->s_freeclusters_counter, 1042 grp->bb_free); 1043 } 1044 1045 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1046 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1047 &grp->bb_state); 1048 if (!ret && gdp) { 1049 int count; 1050 1051 count = ext4_free_inodes_count(sb, gdp); 1052 percpu_counter_sub(&sbi->s_freeinodes_counter, 1053 count); 1054 } 1055 } 1056 } 1057 1058 void ext4_update_dynamic_rev(struct super_block *sb) 1059 { 1060 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1061 1062 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1063 return; 1064 1065 ext4_warning(sb, 1066 "updating to rev %d because of new feature flag, " 1067 "running e2fsck is recommended", 1068 EXT4_DYNAMIC_REV); 1069 1070 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1071 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1072 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1073 /* leave es->s_feature_*compat flags alone */ 1074 /* es->s_uuid will be set by e2fsck if empty */ 1075 1076 /* 1077 * The rest of the superblock fields should be zero, and if not it 1078 * means they are likely already in use, so leave them alone. We 1079 * can leave it up to e2fsck to clean up any inconsistencies there. 1080 */ 1081 } 1082 1083 /* 1084 * Open the external journal device 1085 */ 1086 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 1087 { 1088 struct block_device *bdev; 1089 1090 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 1091 if (IS_ERR(bdev)) 1092 goto fail; 1093 return bdev; 1094 1095 fail: 1096 ext4_msg(sb, KERN_ERR, 1097 "failed to open journal device unknown-block(%u,%u) %ld", 1098 MAJOR(dev), MINOR(dev), PTR_ERR(bdev)); 1099 return NULL; 1100 } 1101 1102 /* 1103 * Release the journal device 1104 */ 1105 static void ext4_blkdev_put(struct block_device *bdev) 1106 { 1107 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1108 } 1109 1110 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 1111 { 1112 struct block_device *bdev; 1113 bdev = sbi->s_journal_bdev; 1114 if (bdev) { 1115 ext4_blkdev_put(bdev); 1116 sbi->s_journal_bdev = NULL; 1117 } 1118 } 1119 1120 static inline struct inode *orphan_list_entry(struct list_head *l) 1121 { 1122 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1123 } 1124 1125 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1126 { 1127 struct list_head *l; 1128 1129 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1130 le32_to_cpu(sbi->s_es->s_last_orphan)); 1131 1132 printk(KERN_ERR "sb_info orphan list:\n"); 1133 list_for_each(l, &sbi->s_orphan) { 1134 struct inode *inode = orphan_list_entry(l); 1135 printk(KERN_ERR " " 1136 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1137 inode->i_sb->s_id, inode->i_ino, inode, 1138 inode->i_mode, inode->i_nlink, 1139 NEXT_ORPHAN(inode)); 1140 } 1141 } 1142 1143 #ifdef CONFIG_QUOTA 1144 static int ext4_quota_off(struct super_block *sb, int type); 1145 1146 static inline void ext4_quota_off_umount(struct super_block *sb) 1147 { 1148 int type; 1149 1150 /* Use our quota_off function to clear inode flags etc. */ 1151 for (type = 0; type < EXT4_MAXQUOTAS; type++) 1152 ext4_quota_off(sb, type); 1153 } 1154 1155 /* 1156 * This is a helper function which is used in the mount/remount 1157 * codepaths (which holds s_umount) to fetch the quota file name. 1158 */ 1159 static inline char *get_qf_name(struct super_block *sb, 1160 struct ext4_sb_info *sbi, 1161 int type) 1162 { 1163 return rcu_dereference_protected(sbi->s_qf_names[type], 1164 lockdep_is_held(&sb->s_umount)); 1165 } 1166 #else 1167 static inline void ext4_quota_off_umount(struct super_block *sb) 1168 { 1169 } 1170 #endif 1171 1172 static void ext4_put_super(struct super_block *sb) 1173 { 1174 struct ext4_sb_info *sbi = EXT4_SB(sb); 1175 struct ext4_super_block *es = sbi->s_es; 1176 struct buffer_head **group_desc; 1177 struct flex_groups **flex_groups; 1178 int aborted = 0; 1179 int i, err; 1180 1181 ext4_unregister_li_request(sb); 1182 ext4_quota_off_umount(sb); 1183 1184 destroy_workqueue(sbi->rsv_conversion_wq); 1185 1186 /* 1187 * Unregister sysfs before destroying jbd2 journal. 1188 * Since we could still access attr_journal_task attribute via sysfs 1189 * path which could have sbi->s_journal->j_task as NULL 1190 */ 1191 ext4_unregister_sysfs(sb); 1192 1193 if (sbi->s_journal) { 1194 aborted = is_journal_aborted(sbi->s_journal); 1195 err = jbd2_journal_destroy(sbi->s_journal); 1196 sbi->s_journal = NULL; 1197 if ((err < 0) && !aborted) { 1198 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1199 } 1200 } 1201 1202 ext4_es_unregister_shrinker(sbi); 1203 del_timer_sync(&sbi->s_err_report); 1204 ext4_release_system_zone(sb); 1205 ext4_mb_release(sb); 1206 ext4_ext_release(sb); 1207 1208 if (!sb_rdonly(sb) && !aborted) { 1209 ext4_clear_feature_journal_needs_recovery(sb); 1210 es->s_state = cpu_to_le16(sbi->s_mount_state); 1211 } 1212 if (!sb_rdonly(sb)) 1213 ext4_commit_super(sb, 1); 1214 1215 rcu_read_lock(); 1216 group_desc = rcu_dereference(sbi->s_group_desc); 1217 for (i = 0; i < sbi->s_gdb_count; i++) 1218 brelse(group_desc[i]); 1219 kvfree(group_desc); 1220 flex_groups = rcu_dereference(sbi->s_flex_groups); 1221 if (flex_groups) { 1222 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1223 kvfree(flex_groups[i]); 1224 kvfree(flex_groups); 1225 } 1226 rcu_read_unlock(); 1227 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1228 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1229 percpu_counter_destroy(&sbi->s_dirs_counter); 1230 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1231 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1232 #ifdef CONFIG_QUOTA 1233 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1234 kfree(get_qf_name(sb, sbi, i)); 1235 #endif 1236 1237 /* Debugging code just in case the in-memory inode orphan list 1238 * isn't empty. The on-disk one can be non-empty if we've 1239 * detected an error and taken the fs readonly, but the 1240 * in-memory list had better be clean by this point. */ 1241 if (!list_empty(&sbi->s_orphan)) 1242 dump_orphan_list(sb, sbi); 1243 J_ASSERT(list_empty(&sbi->s_orphan)); 1244 1245 sync_blockdev(sb->s_bdev); 1246 invalidate_bdev(sb->s_bdev); 1247 if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) { 1248 /* 1249 * Invalidate the journal device's buffers. We don't want them 1250 * floating about in memory - the physical journal device may 1251 * hotswapped, and it breaks the `ro-after' testing code. 1252 */ 1253 sync_blockdev(sbi->s_journal_bdev); 1254 invalidate_bdev(sbi->s_journal_bdev); 1255 ext4_blkdev_remove(sbi); 1256 } 1257 1258 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1259 sbi->s_ea_inode_cache = NULL; 1260 1261 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1262 sbi->s_ea_block_cache = NULL; 1263 1264 if (sbi->s_mmp_tsk) 1265 kthread_stop(sbi->s_mmp_tsk); 1266 brelse(sbi->s_sbh); 1267 sb->s_fs_info = NULL; 1268 /* 1269 * Now that we are completely done shutting down the 1270 * superblock, we need to actually destroy the kobject. 1271 */ 1272 kobject_put(&sbi->s_kobj); 1273 wait_for_completion(&sbi->s_kobj_unregister); 1274 if (sbi->s_chksum_driver) 1275 crypto_free_shash(sbi->s_chksum_driver); 1276 kfree(sbi->s_blockgroup_lock); 1277 fs_put_dax(sbi->s_daxdev); 1278 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1279 #ifdef CONFIG_UNICODE 1280 utf8_unload(sb->s_encoding); 1281 #endif 1282 kfree(sbi); 1283 } 1284 1285 static struct kmem_cache *ext4_inode_cachep; 1286 1287 /* 1288 * Called inside transaction, so use GFP_NOFS 1289 */ 1290 static struct inode *ext4_alloc_inode(struct super_block *sb) 1291 { 1292 struct ext4_inode_info *ei; 1293 1294 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS); 1295 if (!ei) 1296 return NULL; 1297 1298 inode_set_iversion(&ei->vfs_inode, 1); 1299 spin_lock_init(&ei->i_raw_lock); 1300 INIT_LIST_HEAD(&ei->i_prealloc_list); 1301 atomic_set(&ei->i_prealloc_active, 0); 1302 spin_lock_init(&ei->i_prealloc_lock); 1303 ext4_es_init_tree(&ei->i_es_tree); 1304 rwlock_init(&ei->i_es_lock); 1305 INIT_LIST_HEAD(&ei->i_es_list); 1306 ei->i_es_all_nr = 0; 1307 ei->i_es_shk_nr = 0; 1308 ei->i_es_shrink_lblk = 0; 1309 ei->i_reserved_data_blocks = 0; 1310 spin_lock_init(&(ei->i_block_reservation_lock)); 1311 ext4_init_pending_tree(&ei->i_pending_tree); 1312 #ifdef CONFIG_QUOTA 1313 ei->i_reserved_quota = 0; 1314 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1315 #endif 1316 ei->jinode = NULL; 1317 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1318 spin_lock_init(&ei->i_completed_io_lock); 1319 ei->i_sync_tid = 0; 1320 ei->i_datasync_tid = 0; 1321 atomic_set(&ei->i_unwritten, 0); 1322 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1323 ext4_fc_init_inode(&ei->vfs_inode); 1324 mutex_init(&ei->i_fc_lock); 1325 return &ei->vfs_inode; 1326 } 1327 1328 static int ext4_drop_inode(struct inode *inode) 1329 { 1330 int drop = generic_drop_inode(inode); 1331 1332 if (!drop) 1333 drop = fscrypt_drop_inode(inode); 1334 1335 trace_ext4_drop_inode(inode, drop); 1336 return drop; 1337 } 1338 1339 static void ext4_free_in_core_inode(struct inode *inode) 1340 { 1341 fscrypt_free_inode(inode); 1342 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1343 pr_warn("%s: inode %ld still in fc list", 1344 __func__, inode->i_ino); 1345 } 1346 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1347 } 1348 1349 static void ext4_destroy_inode(struct inode *inode) 1350 { 1351 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1352 ext4_msg(inode->i_sb, KERN_ERR, 1353 "Inode %lu (%p): orphan list check failed!", 1354 inode->i_ino, EXT4_I(inode)); 1355 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1356 EXT4_I(inode), sizeof(struct ext4_inode_info), 1357 true); 1358 dump_stack(); 1359 } 1360 } 1361 1362 static void init_once(void *foo) 1363 { 1364 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo; 1365 1366 INIT_LIST_HEAD(&ei->i_orphan); 1367 init_rwsem(&ei->xattr_sem); 1368 init_rwsem(&ei->i_data_sem); 1369 init_rwsem(&ei->i_mmap_sem); 1370 inode_init_once(&ei->vfs_inode); 1371 ext4_fc_init_inode(&ei->vfs_inode); 1372 } 1373 1374 static int __init init_inodecache(void) 1375 { 1376 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1377 sizeof(struct ext4_inode_info), 0, 1378 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1379 SLAB_ACCOUNT), 1380 offsetof(struct ext4_inode_info, i_data), 1381 sizeof_field(struct ext4_inode_info, i_data), 1382 init_once); 1383 if (ext4_inode_cachep == NULL) 1384 return -ENOMEM; 1385 return 0; 1386 } 1387 1388 static void destroy_inodecache(void) 1389 { 1390 /* 1391 * Make sure all delayed rcu free inodes are flushed before we 1392 * destroy cache. 1393 */ 1394 rcu_barrier(); 1395 kmem_cache_destroy(ext4_inode_cachep); 1396 } 1397 1398 void ext4_clear_inode(struct inode *inode) 1399 { 1400 ext4_fc_del(inode); 1401 invalidate_inode_buffers(inode); 1402 clear_inode(inode); 1403 ext4_discard_preallocations(inode, 0); 1404 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1405 dquot_drop(inode); 1406 if (EXT4_I(inode)->jinode) { 1407 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1408 EXT4_I(inode)->jinode); 1409 jbd2_free_inode(EXT4_I(inode)->jinode); 1410 EXT4_I(inode)->jinode = NULL; 1411 } 1412 fscrypt_put_encryption_info(inode); 1413 fsverity_cleanup_inode(inode); 1414 } 1415 1416 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1417 u64 ino, u32 generation) 1418 { 1419 struct inode *inode; 1420 1421 /* 1422 * Currently we don't know the generation for parent directory, so 1423 * a generation of 0 means "accept any" 1424 */ 1425 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1426 if (IS_ERR(inode)) 1427 return ERR_CAST(inode); 1428 if (generation && inode->i_generation != generation) { 1429 iput(inode); 1430 return ERR_PTR(-ESTALE); 1431 } 1432 1433 return inode; 1434 } 1435 1436 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1437 int fh_len, int fh_type) 1438 { 1439 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1440 ext4_nfs_get_inode); 1441 } 1442 1443 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1444 int fh_len, int fh_type) 1445 { 1446 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1447 ext4_nfs_get_inode); 1448 } 1449 1450 static int ext4_nfs_commit_metadata(struct inode *inode) 1451 { 1452 struct writeback_control wbc = { 1453 .sync_mode = WB_SYNC_ALL 1454 }; 1455 1456 trace_ext4_nfs_commit_metadata(inode); 1457 return ext4_write_inode(inode, &wbc); 1458 } 1459 1460 /* 1461 * Try to release metadata pages (indirect blocks, directories) which are 1462 * mapped via the block device. Since these pages could have journal heads 1463 * which would prevent try_to_free_buffers() from freeing them, we must use 1464 * jbd2 layer's try_to_free_buffers() function to release them. 1465 */ 1466 static int bdev_try_to_free_page(struct super_block *sb, struct page *page, 1467 gfp_t wait) 1468 { 1469 journal_t *journal = EXT4_SB(sb)->s_journal; 1470 1471 WARN_ON(PageChecked(page)); 1472 if (!page_has_buffers(page)) 1473 return 0; 1474 if (journal) 1475 return jbd2_journal_try_to_free_buffers(journal, page); 1476 1477 return try_to_free_buffers(page); 1478 } 1479 1480 #ifdef CONFIG_FS_ENCRYPTION 1481 static int ext4_get_context(struct inode *inode, void *ctx, size_t len) 1482 { 1483 return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, 1484 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len); 1485 } 1486 1487 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len, 1488 void *fs_data) 1489 { 1490 handle_t *handle = fs_data; 1491 int res, res2, credits, retries = 0; 1492 1493 /* 1494 * Encrypting the root directory is not allowed because e2fsck expects 1495 * lost+found to exist and be unencrypted, and encrypting the root 1496 * directory would imply encrypting the lost+found directory as well as 1497 * the filename "lost+found" itself. 1498 */ 1499 if (inode->i_ino == EXT4_ROOT_INO) 1500 return -EPERM; 1501 1502 if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode))) 1503 return -EINVAL; 1504 1505 if (ext4_test_inode_flag(inode, EXT4_INODE_DAX)) 1506 return -EOPNOTSUPP; 1507 1508 res = ext4_convert_inline_data(inode); 1509 if (res) 1510 return res; 1511 1512 /* 1513 * If a journal handle was specified, then the encryption context is 1514 * being set on a new inode via inheritance and is part of a larger 1515 * transaction to create the inode. Otherwise the encryption context is 1516 * being set on an existing inode in its own transaction. Only in the 1517 * latter case should the "retry on ENOSPC" logic be used. 1518 */ 1519 1520 if (handle) { 1521 res = ext4_xattr_set_handle(handle, inode, 1522 EXT4_XATTR_INDEX_ENCRYPTION, 1523 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1524 ctx, len, 0); 1525 if (!res) { 1526 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1527 ext4_clear_inode_state(inode, 1528 EXT4_STATE_MAY_INLINE_DATA); 1529 /* 1530 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1531 * S_DAX may be disabled 1532 */ 1533 ext4_set_inode_flags(inode, false); 1534 } 1535 return res; 1536 } 1537 1538 res = dquot_initialize(inode); 1539 if (res) 1540 return res; 1541 retry: 1542 res = ext4_xattr_set_credits(inode, len, false /* is_create */, 1543 &credits); 1544 if (res) 1545 return res; 1546 1547 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); 1548 if (IS_ERR(handle)) 1549 return PTR_ERR(handle); 1550 1551 res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION, 1552 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1553 ctx, len, 0); 1554 if (!res) { 1555 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1556 /* 1557 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1558 * S_DAX may be disabled 1559 */ 1560 ext4_set_inode_flags(inode, false); 1561 res = ext4_mark_inode_dirty(handle, inode); 1562 if (res) 1563 EXT4_ERROR_INODE(inode, "Failed to mark inode dirty"); 1564 } 1565 res2 = ext4_journal_stop(handle); 1566 1567 if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 1568 goto retry; 1569 if (!res) 1570 res = res2; 1571 return res; 1572 } 1573 1574 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb) 1575 { 1576 return EXT4_SB(sb)->s_dummy_enc_policy.policy; 1577 } 1578 1579 static bool ext4_has_stable_inodes(struct super_block *sb) 1580 { 1581 return ext4_has_feature_stable_inodes(sb); 1582 } 1583 1584 static void ext4_get_ino_and_lblk_bits(struct super_block *sb, 1585 int *ino_bits_ret, int *lblk_bits_ret) 1586 { 1587 *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count); 1588 *lblk_bits_ret = 8 * sizeof(ext4_lblk_t); 1589 } 1590 1591 static const struct fscrypt_operations ext4_cryptops = { 1592 .key_prefix = "ext4:", 1593 .get_context = ext4_get_context, 1594 .set_context = ext4_set_context, 1595 .get_dummy_policy = ext4_get_dummy_policy, 1596 .empty_dir = ext4_empty_dir, 1597 .max_namelen = EXT4_NAME_LEN, 1598 .has_stable_inodes = ext4_has_stable_inodes, 1599 .get_ino_and_lblk_bits = ext4_get_ino_and_lblk_bits, 1600 }; 1601 #endif 1602 1603 #ifdef CONFIG_QUOTA 1604 static const char * const quotatypes[] = INITQFNAMES; 1605 #define QTYPE2NAME(t) (quotatypes[t]) 1606 1607 static int ext4_write_dquot(struct dquot *dquot); 1608 static int ext4_acquire_dquot(struct dquot *dquot); 1609 static int ext4_release_dquot(struct dquot *dquot); 1610 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1611 static int ext4_write_info(struct super_block *sb, int type); 1612 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1613 const struct path *path); 1614 static int ext4_quota_on_mount(struct super_block *sb, int type); 1615 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1616 size_t len, loff_t off); 1617 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1618 const char *data, size_t len, loff_t off); 1619 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1620 unsigned int flags); 1621 static int ext4_enable_quotas(struct super_block *sb); 1622 1623 static struct dquot **ext4_get_dquots(struct inode *inode) 1624 { 1625 return EXT4_I(inode)->i_dquot; 1626 } 1627 1628 static const struct dquot_operations ext4_quota_operations = { 1629 .get_reserved_space = ext4_get_reserved_space, 1630 .write_dquot = ext4_write_dquot, 1631 .acquire_dquot = ext4_acquire_dquot, 1632 .release_dquot = ext4_release_dquot, 1633 .mark_dirty = ext4_mark_dquot_dirty, 1634 .write_info = ext4_write_info, 1635 .alloc_dquot = dquot_alloc, 1636 .destroy_dquot = dquot_destroy, 1637 .get_projid = ext4_get_projid, 1638 .get_inode_usage = ext4_get_inode_usage, 1639 .get_next_id = dquot_get_next_id, 1640 }; 1641 1642 static const struct quotactl_ops ext4_qctl_operations = { 1643 .quota_on = ext4_quota_on, 1644 .quota_off = ext4_quota_off, 1645 .quota_sync = dquot_quota_sync, 1646 .get_state = dquot_get_state, 1647 .set_info = dquot_set_dqinfo, 1648 .get_dqblk = dquot_get_dqblk, 1649 .set_dqblk = dquot_set_dqblk, 1650 .get_nextdqblk = dquot_get_next_dqblk, 1651 }; 1652 #endif 1653 1654 static const struct super_operations ext4_sops = { 1655 .alloc_inode = ext4_alloc_inode, 1656 .free_inode = ext4_free_in_core_inode, 1657 .destroy_inode = ext4_destroy_inode, 1658 .write_inode = ext4_write_inode, 1659 .dirty_inode = ext4_dirty_inode, 1660 .drop_inode = ext4_drop_inode, 1661 .evict_inode = ext4_evict_inode, 1662 .put_super = ext4_put_super, 1663 .sync_fs = ext4_sync_fs, 1664 .freeze_fs = ext4_freeze, 1665 .unfreeze_fs = ext4_unfreeze, 1666 .statfs = ext4_statfs, 1667 .remount_fs = ext4_remount, 1668 .show_options = ext4_show_options, 1669 #ifdef CONFIG_QUOTA 1670 .quota_read = ext4_quota_read, 1671 .quota_write = ext4_quota_write, 1672 .get_dquots = ext4_get_dquots, 1673 #endif 1674 .bdev_try_to_free_page = bdev_try_to_free_page, 1675 }; 1676 1677 static const struct export_operations ext4_export_ops = { 1678 .fh_to_dentry = ext4_fh_to_dentry, 1679 .fh_to_parent = ext4_fh_to_parent, 1680 .get_parent = ext4_get_parent, 1681 .commit_metadata = ext4_nfs_commit_metadata, 1682 }; 1683 1684 enum { 1685 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1686 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro, 1687 Opt_nouid32, Opt_debug, Opt_removed, 1688 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, 1689 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1690 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1691 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1692 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1693 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1694 Opt_inlinecrypt, 1695 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota, 1696 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota, 1697 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1698 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, 1699 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1700 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1701 Opt_nowarn_on_error, Opt_mblk_io_submit, 1702 Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize, 1703 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1704 Opt_inode_readahead_blks, Opt_journal_ioprio, 1705 Opt_dioread_nolock, Opt_dioread_lock, 1706 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1707 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1708 Opt_prefetch_block_bitmaps, 1709 #ifdef CONFIG_EXT4_DEBUG 1710 Opt_fc_debug_max_replay, Opt_fc_debug_force 1711 #endif 1712 }; 1713 1714 static const match_table_t tokens = { 1715 {Opt_bsd_df, "bsddf"}, 1716 {Opt_minix_df, "minixdf"}, 1717 {Opt_grpid, "grpid"}, 1718 {Opt_grpid, "bsdgroups"}, 1719 {Opt_nogrpid, "nogrpid"}, 1720 {Opt_nogrpid, "sysvgroups"}, 1721 {Opt_resgid, "resgid=%u"}, 1722 {Opt_resuid, "resuid=%u"}, 1723 {Opt_sb, "sb=%u"}, 1724 {Opt_err_cont, "errors=continue"}, 1725 {Opt_err_panic, "errors=panic"}, 1726 {Opt_err_ro, "errors=remount-ro"}, 1727 {Opt_nouid32, "nouid32"}, 1728 {Opt_debug, "debug"}, 1729 {Opt_removed, "oldalloc"}, 1730 {Opt_removed, "orlov"}, 1731 {Opt_user_xattr, "user_xattr"}, 1732 {Opt_nouser_xattr, "nouser_xattr"}, 1733 {Opt_acl, "acl"}, 1734 {Opt_noacl, "noacl"}, 1735 {Opt_noload, "norecovery"}, 1736 {Opt_noload, "noload"}, 1737 {Opt_removed, "nobh"}, 1738 {Opt_removed, "bh"}, 1739 {Opt_commit, "commit=%u"}, 1740 {Opt_min_batch_time, "min_batch_time=%u"}, 1741 {Opt_max_batch_time, "max_batch_time=%u"}, 1742 {Opt_journal_dev, "journal_dev=%u"}, 1743 {Opt_journal_path, "journal_path=%s"}, 1744 {Opt_journal_checksum, "journal_checksum"}, 1745 {Opt_nojournal_checksum, "nojournal_checksum"}, 1746 {Opt_journal_async_commit, "journal_async_commit"}, 1747 {Opt_abort, "abort"}, 1748 {Opt_data_journal, "data=journal"}, 1749 {Opt_data_ordered, "data=ordered"}, 1750 {Opt_data_writeback, "data=writeback"}, 1751 {Opt_data_err_abort, "data_err=abort"}, 1752 {Opt_data_err_ignore, "data_err=ignore"}, 1753 {Opt_offusrjquota, "usrjquota="}, 1754 {Opt_usrjquota, "usrjquota=%s"}, 1755 {Opt_offgrpjquota, "grpjquota="}, 1756 {Opt_grpjquota, "grpjquota=%s"}, 1757 {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, 1758 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, 1759 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, 1760 {Opt_grpquota, "grpquota"}, 1761 {Opt_noquota, "noquota"}, 1762 {Opt_quota, "quota"}, 1763 {Opt_usrquota, "usrquota"}, 1764 {Opt_prjquota, "prjquota"}, 1765 {Opt_barrier, "barrier=%u"}, 1766 {Opt_barrier, "barrier"}, 1767 {Opt_nobarrier, "nobarrier"}, 1768 {Opt_i_version, "i_version"}, 1769 {Opt_dax, "dax"}, 1770 {Opt_dax_always, "dax=always"}, 1771 {Opt_dax_inode, "dax=inode"}, 1772 {Opt_dax_never, "dax=never"}, 1773 {Opt_stripe, "stripe=%u"}, 1774 {Opt_delalloc, "delalloc"}, 1775 {Opt_warn_on_error, "warn_on_error"}, 1776 {Opt_nowarn_on_error, "nowarn_on_error"}, 1777 {Opt_lazytime, "lazytime"}, 1778 {Opt_nolazytime, "nolazytime"}, 1779 {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"}, 1780 {Opt_nodelalloc, "nodelalloc"}, 1781 {Opt_removed, "mblk_io_submit"}, 1782 {Opt_removed, "nomblk_io_submit"}, 1783 {Opt_block_validity, "block_validity"}, 1784 {Opt_noblock_validity, "noblock_validity"}, 1785 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, 1786 {Opt_journal_ioprio, "journal_ioprio=%u"}, 1787 {Opt_auto_da_alloc, "auto_da_alloc=%u"}, 1788 {Opt_auto_da_alloc, "auto_da_alloc"}, 1789 {Opt_noauto_da_alloc, "noauto_da_alloc"}, 1790 {Opt_dioread_nolock, "dioread_nolock"}, 1791 {Opt_dioread_lock, "nodioread_nolock"}, 1792 {Opt_dioread_lock, "dioread_lock"}, 1793 {Opt_discard, "discard"}, 1794 {Opt_nodiscard, "nodiscard"}, 1795 {Opt_init_itable, "init_itable=%u"}, 1796 {Opt_init_itable, "init_itable"}, 1797 {Opt_noinit_itable, "noinit_itable"}, 1798 #ifdef CONFIG_EXT4_DEBUG 1799 {Opt_fc_debug_force, "fc_debug_force"}, 1800 {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"}, 1801 #endif 1802 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, 1803 {Opt_test_dummy_encryption, "test_dummy_encryption=%s"}, 1804 {Opt_test_dummy_encryption, "test_dummy_encryption"}, 1805 {Opt_inlinecrypt, "inlinecrypt"}, 1806 {Opt_nombcache, "nombcache"}, 1807 {Opt_nombcache, "no_mbcache"}, /* for backward compatibility */ 1808 {Opt_prefetch_block_bitmaps, "prefetch_block_bitmaps"}, 1809 {Opt_removed, "check=none"}, /* mount option from ext2/3 */ 1810 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */ 1811 {Opt_removed, "reservation"}, /* mount option from ext2/3 */ 1812 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */ 1813 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */ 1814 {Opt_err, NULL}, 1815 }; 1816 1817 static ext4_fsblk_t get_sb_block(void **data) 1818 { 1819 ext4_fsblk_t sb_block; 1820 char *options = (char *) *data; 1821 1822 if (!options || strncmp(options, "sb=", 3) != 0) 1823 return 1; /* Default location */ 1824 1825 options += 3; 1826 /* TODO: use simple_strtoll with >32bit ext4 */ 1827 sb_block = simple_strtoul(options, &options, 0); 1828 if (*options && *options != ',') { 1829 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n", 1830 (char *) *data); 1831 return 1; 1832 } 1833 if (*options == ',') 1834 options++; 1835 *data = (void *) options; 1836 1837 return sb_block; 1838 } 1839 1840 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1841 static const char deprecated_msg[] = 1842 "Mount option \"%s\" will be removed by %s\n" 1843 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1844 1845 #ifdef CONFIG_QUOTA 1846 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args) 1847 { 1848 struct ext4_sb_info *sbi = EXT4_SB(sb); 1849 char *qname, *old_qname = get_qf_name(sb, sbi, qtype); 1850 int ret = -1; 1851 1852 if (sb_any_quota_loaded(sb) && !old_qname) { 1853 ext4_msg(sb, KERN_ERR, 1854 "Cannot change journaled " 1855 "quota options when quota turned on"); 1856 return -1; 1857 } 1858 if (ext4_has_feature_quota(sb)) { 1859 ext4_msg(sb, KERN_INFO, "Journaled quota options " 1860 "ignored when QUOTA feature is enabled"); 1861 return 1; 1862 } 1863 qname = match_strdup(args); 1864 if (!qname) { 1865 ext4_msg(sb, KERN_ERR, 1866 "Not enough memory for storing quotafile name"); 1867 return -1; 1868 } 1869 if (old_qname) { 1870 if (strcmp(old_qname, qname) == 0) 1871 ret = 1; 1872 else 1873 ext4_msg(sb, KERN_ERR, 1874 "%s quota file already specified", 1875 QTYPE2NAME(qtype)); 1876 goto errout; 1877 } 1878 if (strchr(qname, '/')) { 1879 ext4_msg(sb, KERN_ERR, 1880 "quotafile must be on filesystem root"); 1881 goto errout; 1882 } 1883 rcu_assign_pointer(sbi->s_qf_names[qtype], qname); 1884 set_opt(sb, QUOTA); 1885 return 1; 1886 errout: 1887 kfree(qname); 1888 return ret; 1889 } 1890 1891 static int clear_qf_name(struct super_block *sb, int qtype) 1892 { 1893 1894 struct ext4_sb_info *sbi = EXT4_SB(sb); 1895 char *old_qname = get_qf_name(sb, sbi, qtype); 1896 1897 if (sb_any_quota_loaded(sb) && old_qname) { 1898 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options" 1899 " when quota turned on"); 1900 return -1; 1901 } 1902 rcu_assign_pointer(sbi->s_qf_names[qtype], NULL); 1903 synchronize_rcu(); 1904 kfree(old_qname); 1905 return 1; 1906 } 1907 #endif 1908 1909 #define MOPT_SET 0x0001 1910 #define MOPT_CLEAR 0x0002 1911 #define MOPT_NOSUPPORT 0x0004 1912 #define MOPT_EXPLICIT 0x0008 1913 #define MOPT_CLEAR_ERR 0x0010 1914 #define MOPT_GTE0 0x0020 1915 #ifdef CONFIG_QUOTA 1916 #define MOPT_Q 0 1917 #define MOPT_QFMT 0x0040 1918 #else 1919 #define MOPT_Q MOPT_NOSUPPORT 1920 #define MOPT_QFMT MOPT_NOSUPPORT 1921 #endif 1922 #define MOPT_DATAJ 0x0080 1923 #define MOPT_NO_EXT2 0x0100 1924 #define MOPT_NO_EXT3 0x0200 1925 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1926 #define MOPT_STRING 0x0400 1927 #define MOPT_SKIP 0x0800 1928 #define MOPT_2 0x1000 1929 1930 static const struct mount_opts { 1931 int token; 1932 int mount_opt; 1933 int flags; 1934 } ext4_mount_opts[] = { 1935 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1936 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1937 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1938 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1939 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1940 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1941 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1942 MOPT_EXT4_ONLY | MOPT_SET}, 1943 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1944 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1945 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1946 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1947 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1948 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1949 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1950 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1951 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1952 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1953 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1954 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1955 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1956 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1957 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1958 EXT4_MOUNT_JOURNAL_CHECKSUM), 1959 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1960 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1961 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR}, 1962 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR}, 1963 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR}, 1964 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, 1965 MOPT_NO_EXT2}, 1966 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, 1967 MOPT_NO_EXT2}, 1968 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1969 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1970 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1971 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1972 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1973 {Opt_commit, 0, MOPT_GTE0}, 1974 {Opt_max_batch_time, 0, MOPT_GTE0}, 1975 {Opt_min_batch_time, 0, MOPT_GTE0}, 1976 {Opt_inode_readahead_blks, 0, MOPT_GTE0}, 1977 {Opt_init_itable, 0, MOPT_GTE0}, 1978 {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP}, 1979 {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS, 1980 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1981 {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE, 1982 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1983 {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER, 1984 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1985 {Opt_stripe, 0, MOPT_GTE0}, 1986 {Opt_resuid, 0, MOPT_GTE0}, 1987 {Opt_resgid, 0, MOPT_GTE0}, 1988 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1989 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING}, 1990 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1991 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1992 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1993 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, 1994 MOPT_NO_EXT2 | MOPT_DATAJ}, 1995 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1996 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR}, 1997 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1998 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1999 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR}, 2000 #else 2001 {Opt_acl, 0, MOPT_NOSUPPORT}, 2002 {Opt_noacl, 0, MOPT_NOSUPPORT}, 2003 #endif 2004 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 2005 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 2006 {Opt_debug_want_extra_isize, 0, MOPT_GTE0}, 2007 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 2008 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 2009 MOPT_SET | MOPT_Q}, 2010 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 2011 MOPT_SET | MOPT_Q}, 2012 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 2013 MOPT_SET | MOPT_Q}, 2014 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2015 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 2016 MOPT_CLEAR | MOPT_Q}, 2017 {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING}, 2018 {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING}, 2019 {Opt_offusrjquota, 0, MOPT_Q}, 2020 {Opt_offgrpjquota, 0, MOPT_Q}, 2021 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT}, 2022 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT}, 2023 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT}, 2024 {Opt_max_dir_size_kb, 0, MOPT_GTE0}, 2025 {Opt_test_dummy_encryption, 0, MOPT_STRING}, 2026 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 2027 {Opt_prefetch_block_bitmaps, EXT4_MOUNT_PREFETCH_BLOCK_BITMAPS, 2028 MOPT_SET}, 2029 #ifdef CONFIG_EXT4_DEBUG 2030 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 2031 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 2032 {Opt_fc_debug_max_replay, 0, MOPT_GTE0}, 2033 #endif 2034 {Opt_err, 0, 0} 2035 }; 2036 2037 #ifdef CONFIG_UNICODE 2038 static const struct ext4_sb_encodings { 2039 __u16 magic; 2040 char *name; 2041 char *version; 2042 } ext4_sb_encoding_map[] = { 2043 {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"}, 2044 }; 2045 2046 static int ext4_sb_read_encoding(const struct ext4_super_block *es, 2047 const struct ext4_sb_encodings **encoding, 2048 __u16 *flags) 2049 { 2050 __u16 magic = le16_to_cpu(es->s_encoding); 2051 int i; 2052 2053 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 2054 if (magic == ext4_sb_encoding_map[i].magic) 2055 break; 2056 2057 if (i >= ARRAY_SIZE(ext4_sb_encoding_map)) 2058 return -EINVAL; 2059 2060 *encoding = &ext4_sb_encoding_map[i]; 2061 *flags = le16_to_cpu(es->s_encoding_flags); 2062 2063 return 0; 2064 } 2065 #endif 2066 2067 static int ext4_set_test_dummy_encryption(struct super_block *sb, 2068 const char *opt, 2069 const substring_t *arg, 2070 bool is_remount) 2071 { 2072 #ifdef CONFIG_FS_ENCRYPTION 2073 struct ext4_sb_info *sbi = EXT4_SB(sb); 2074 int err; 2075 2076 /* 2077 * This mount option is just for testing, and it's not worthwhile to 2078 * implement the extra complexity (e.g. RCU protection) that would be 2079 * needed to allow it to be set or changed during remount. We do allow 2080 * it to be specified during remount, but only if there is no change. 2081 */ 2082 if (is_remount && !sbi->s_dummy_enc_policy.policy) { 2083 ext4_msg(sb, KERN_WARNING, 2084 "Can't set test_dummy_encryption on remount"); 2085 return -1; 2086 } 2087 err = fscrypt_set_test_dummy_encryption(sb, arg->from, 2088 &sbi->s_dummy_enc_policy); 2089 if (err) { 2090 if (err == -EEXIST) 2091 ext4_msg(sb, KERN_WARNING, 2092 "Can't change test_dummy_encryption on remount"); 2093 else if (err == -EINVAL) 2094 ext4_msg(sb, KERN_WARNING, 2095 "Value of option \"%s\" is unrecognized", opt); 2096 else 2097 ext4_msg(sb, KERN_WARNING, 2098 "Error processing option \"%s\" [%d]", 2099 opt, err); 2100 return -1; 2101 } 2102 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2103 #else 2104 ext4_msg(sb, KERN_WARNING, 2105 "Test dummy encryption mount option ignored"); 2106 #endif 2107 return 1; 2108 } 2109 2110 static int handle_mount_opt(struct super_block *sb, char *opt, int token, 2111 substring_t *args, unsigned long *journal_devnum, 2112 unsigned int *journal_ioprio, int is_remount) 2113 { 2114 struct ext4_sb_info *sbi = EXT4_SB(sb); 2115 const struct mount_opts *m; 2116 kuid_t uid; 2117 kgid_t gid; 2118 int arg = 0; 2119 2120 #ifdef CONFIG_QUOTA 2121 if (token == Opt_usrjquota) 2122 return set_qf_name(sb, USRQUOTA, &args[0]); 2123 else if (token == Opt_grpjquota) 2124 return set_qf_name(sb, GRPQUOTA, &args[0]); 2125 else if (token == Opt_offusrjquota) 2126 return clear_qf_name(sb, USRQUOTA); 2127 else if (token == Opt_offgrpjquota) 2128 return clear_qf_name(sb, GRPQUOTA); 2129 #endif 2130 switch (token) { 2131 case Opt_noacl: 2132 case Opt_nouser_xattr: 2133 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5"); 2134 break; 2135 case Opt_sb: 2136 return 1; /* handled by get_sb_block() */ 2137 case Opt_removed: 2138 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt); 2139 return 1; 2140 case Opt_abort: 2141 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 2142 return 1; 2143 case Opt_i_version: 2144 sb->s_flags |= SB_I_VERSION; 2145 return 1; 2146 case Opt_lazytime: 2147 sb->s_flags |= SB_LAZYTIME; 2148 return 1; 2149 case Opt_nolazytime: 2150 sb->s_flags &= ~SB_LAZYTIME; 2151 return 1; 2152 case Opt_inlinecrypt: 2153 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2154 sb->s_flags |= SB_INLINECRYPT; 2155 #else 2156 ext4_msg(sb, KERN_ERR, "inline encryption not supported"); 2157 #endif 2158 return 1; 2159 } 2160 2161 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2162 if (token == m->token) 2163 break; 2164 2165 if (m->token == Opt_err) { 2166 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" " 2167 "or missing value", opt); 2168 return -1; 2169 } 2170 2171 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2172 ext4_msg(sb, KERN_ERR, 2173 "Mount option \"%s\" incompatible with ext2", opt); 2174 return -1; 2175 } 2176 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2177 ext4_msg(sb, KERN_ERR, 2178 "Mount option \"%s\" incompatible with ext3", opt); 2179 return -1; 2180 } 2181 2182 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg)) 2183 return -1; 2184 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0)) 2185 return -1; 2186 if (m->flags & MOPT_EXPLICIT) { 2187 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2188 set_opt2(sb, EXPLICIT_DELALLOC); 2189 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2190 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM); 2191 } else 2192 return -1; 2193 } 2194 if (m->flags & MOPT_CLEAR_ERR) 2195 clear_opt(sb, ERRORS_MASK); 2196 if (token == Opt_noquota && sb_any_quota_loaded(sb)) { 2197 ext4_msg(sb, KERN_ERR, "Cannot change quota " 2198 "options when quota turned on"); 2199 return -1; 2200 } 2201 2202 if (m->flags & MOPT_NOSUPPORT) { 2203 ext4_msg(sb, KERN_ERR, "%s option not supported", opt); 2204 } else if (token == Opt_commit) { 2205 if (arg == 0) 2206 arg = JBD2_DEFAULT_MAX_COMMIT_AGE; 2207 else if (arg > INT_MAX / HZ) { 2208 ext4_msg(sb, KERN_ERR, 2209 "Invalid commit interval %d, " 2210 "must be smaller than %d", 2211 arg, INT_MAX / HZ); 2212 return -1; 2213 } 2214 sbi->s_commit_interval = HZ * arg; 2215 } else if (token == Opt_debug_want_extra_isize) { 2216 if ((arg & 1) || 2217 (arg < 4) || 2218 (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) { 2219 ext4_msg(sb, KERN_ERR, 2220 "Invalid want_extra_isize %d", arg); 2221 return -1; 2222 } 2223 sbi->s_want_extra_isize = arg; 2224 } else if (token == Opt_max_batch_time) { 2225 sbi->s_max_batch_time = arg; 2226 } else if (token == Opt_min_batch_time) { 2227 sbi->s_min_batch_time = arg; 2228 } else if (token == Opt_inode_readahead_blks) { 2229 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) { 2230 ext4_msg(sb, KERN_ERR, 2231 "EXT4-fs: inode_readahead_blks must be " 2232 "0 or a power of 2 smaller than 2^31"); 2233 return -1; 2234 } 2235 sbi->s_inode_readahead_blks = arg; 2236 } else if (token == Opt_init_itable) { 2237 set_opt(sb, INIT_INODE_TABLE); 2238 if (!args->from) 2239 arg = EXT4_DEF_LI_WAIT_MULT; 2240 sbi->s_li_wait_mult = arg; 2241 } else if (token == Opt_max_dir_size_kb) { 2242 sbi->s_max_dir_size_kb = arg; 2243 #ifdef CONFIG_EXT4_DEBUG 2244 } else if (token == Opt_fc_debug_max_replay) { 2245 sbi->s_fc_debug_max_replay = arg; 2246 #endif 2247 } else if (token == Opt_stripe) { 2248 sbi->s_stripe = arg; 2249 } else if (token == Opt_resuid) { 2250 uid = make_kuid(current_user_ns(), arg); 2251 if (!uid_valid(uid)) { 2252 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg); 2253 return -1; 2254 } 2255 sbi->s_resuid = uid; 2256 } else if (token == Opt_resgid) { 2257 gid = make_kgid(current_user_ns(), arg); 2258 if (!gid_valid(gid)) { 2259 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg); 2260 return -1; 2261 } 2262 sbi->s_resgid = gid; 2263 } else if (token == Opt_journal_dev) { 2264 if (is_remount) { 2265 ext4_msg(sb, KERN_ERR, 2266 "Cannot specify journal on remount"); 2267 return -1; 2268 } 2269 *journal_devnum = arg; 2270 } else if (token == Opt_journal_path) { 2271 char *journal_path; 2272 struct inode *journal_inode; 2273 struct path path; 2274 int error; 2275 2276 if (is_remount) { 2277 ext4_msg(sb, KERN_ERR, 2278 "Cannot specify journal on remount"); 2279 return -1; 2280 } 2281 journal_path = match_strdup(&args[0]); 2282 if (!journal_path) { 2283 ext4_msg(sb, KERN_ERR, "error: could not dup " 2284 "journal device string"); 2285 return -1; 2286 } 2287 2288 error = kern_path(journal_path, LOOKUP_FOLLOW, &path); 2289 if (error) { 2290 ext4_msg(sb, KERN_ERR, "error: could not find " 2291 "journal device path: error %d", error); 2292 kfree(journal_path); 2293 return -1; 2294 } 2295 2296 journal_inode = d_inode(path.dentry); 2297 if (!S_ISBLK(journal_inode->i_mode)) { 2298 ext4_msg(sb, KERN_ERR, "error: journal path %s " 2299 "is not a block device", journal_path); 2300 path_put(&path); 2301 kfree(journal_path); 2302 return -1; 2303 } 2304 2305 *journal_devnum = new_encode_dev(journal_inode->i_rdev); 2306 path_put(&path); 2307 kfree(journal_path); 2308 } else if (token == Opt_journal_ioprio) { 2309 if (arg > 7) { 2310 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority" 2311 " (must be 0-7)"); 2312 return -1; 2313 } 2314 *journal_ioprio = 2315 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg); 2316 } else if (token == Opt_test_dummy_encryption) { 2317 return ext4_set_test_dummy_encryption(sb, opt, &args[0], 2318 is_remount); 2319 } else if (m->flags & MOPT_DATAJ) { 2320 if (is_remount) { 2321 if (!sbi->s_journal) 2322 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option"); 2323 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) { 2324 ext4_msg(sb, KERN_ERR, 2325 "Cannot change data mode on remount"); 2326 return -1; 2327 } 2328 } else { 2329 clear_opt(sb, DATA_FLAGS); 2330 sbi->s_mount_opt |= m->mount_opt; 2331 } 2332 #ifdef CONFIG_QUOTA 2333 } else if (m->flags & MOPT_QFMT) { 2334 if (sb_any_quota_loaded(sb) && 2335 sbi->s_jquota_fmt != m->mount_opt) { 2336 ext4_msg(sb, KERN_ERR, "Cannot change journaled " 2337 "quota options when quota turned on"); 2338 return -1; 2339 } 2340 if (ext4_has_feature_quota(sb)) { 2341 ext4_msg(sb, KERN_INFO, 2342 "Quota format mount options ignored " 2343 "when QUOTA feature is enabled"); 2344 return 1; 2345 } 2346 sbi->s_jquota_fmt = m->mount_opt; 2347 #endif 2348 } else if (token == Opt_dax || token == Opt_dax_always || 2349 token == Opt_dax_inode || token == Opt_dax_never) { 2350 #ifdef CONFIG_FS_DAX 2351 switch (token) { 2352 case Opt_dax: 2353 case Opt_dax_always: 2354 if (is_remount && 2355 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2356 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2357 fail_dax_change_remount: 2358 ext4_msg(sb, KERN_ERR, "can't change " 2359 "dax mount option while remounting"); 2360 return -1; 2361 } 2362 if (is_remount && 2363 (test_opt(sb, DATA_FLAGS) == 2364 EXT4_MOUNT_JOURNAL_DATA)) { 2365 ext4_msg(sb, KERN_ERR, "can't mount with " 2366 "both data=journal and dax"); 2367 return -1; 2368 } 2369 ext4_msg(sb, KERN_WARNING, 2370 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 2371 sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS; 2372 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2373 break; 2374 case Opt_dax_never: 2375 if (is_remount && 2376 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2377 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) 2378 goto fail_dax_change_remount; 2379 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2380 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2381 break; 2382 case Opt_dax_inode: 2383 if (is_remount && 2384 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2385 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2386 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) 2387 goto fail_dax_change_remount; 2388 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2389 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2390 /* Strictly for printing options */ 2391 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE; 2392 break; 2393 } 2394 #else 2395 ext4_msg(sb, KERN_INFO, "dax option not supported"); 2396 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2397 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2398 return -1; 2399 #endif 2400 } else if (token == Opt_data_err_abort) { 2401 sbi->s_mount_opt |= m->mount_opt; 2402 } else if (token == Opt_data_err_ignore) { 2403 sbi->s_mount_opt &= ~m->mount_opt; 2404 } else { 2405 if (!args->from) 2406 arg = 1; 2407 if (m->flags & MOPT_CLEAR) 2408 arg = !arg; 2409 else if (unlikely(!(m->flags & MOPT_SET))) { 2410 ext4_msg(sb, KERN_WARNING, 2411 "buggy handling of option %s", opt); 2412 WARN_ON(1); 2413 return -1; 2414 } 2415 if (m->flags & MOPT_2) { 2416 if (arg != 0) 2417 sbi->s_mount_opt2 |= m->mount_opt; 2418 else 2419 sbi->s_mount_opt2 &= ~m->mount_opt; 2420 } else { 2421 if (arg != 0) 2422 sbi->s_mount_opt |= m->mount_opt; 2423 else 2424 sbi->s_mount_opt &= ~m->mount_opt; 2425 } 2426 } 2427 return 1; 2428 } 2429 2430 static int parse_options(char *options, struct super_block *sb, 2431 unsigned long *journal_devnum, 2432 unsigned int *journal_ioprio, 2433 int is_remount) 2434 { 2435 struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb); 2436 char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name; 2437 substring_t args[MAX_OPT_ARGS]; 2438 int token; 2439 2440 if (!options) 2441 return 1; 2442 2443 while ((p = strsep(&options, ",")) != NULL) { 2444 if (!*p) 2445 continue; 2446 /* 2447 * Initialize args struct so we know whether arg was 2448 * found; some options take optional arguments. 2449 */ 2450 args[0].to = args[0].from = NULL; 2451 token = match_token(p, tokens, args); 2452 if (handle_mount_opt(sb, p, token, args, journal_devnum, 2453 journal_ioprio, is_remount) < 0) 2454 return 0; 2455 } 2456 #ifdef CONFIG_QUOTA 2457 /* 2458 * We do the test below only for project quotas. 'usrquota' and 2459 * 'grpquota' mount options are allowed even without quota feature 2460 * to support legacy quotas in quota files. 2461 */ 2462 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) { 2463 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. " 2464 "Cannot enable project quota enforcement."); 2465 return 0; 2466 } 2467 usr_qf_name = get_qf_name(sb, sbi, USRQUOTA); 2468 grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA); 2469 if (usr_qf_name || grp_qf_name) { 2470 if (test_opt(sb, USRQUOTA) && usr_qf_name) 2471 clear_opt(sb, USRQUOTA); 2472 2473 if (test_opt(sb, GRPQUOTA) && grp_qf_name) 2474 clear_opt(sb, GRPQUOTA); 2475 2476 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) { 2477 ext4_msg(sb, KERN_ERR, "old and new quota " 2478 "format mixing"); 2479 return 0; 2480 } 2481 2482 if (!sbi->s_jquota_fmt) { 2483 ext4_msg(sb, KERN_ERR, "journaled quota format " 2484 "not specified"); 2485 return 0; 2486 } 2487 } 2488 #endif 2489 if (test_opt(sb, DIOREAD_NOLOCK)) { 2490 int blocksize = 2491 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2492 if (blocksize < PAGE_SIZE) 2493 ext4_msg(sb, KERN_WARNING, "Warning: mounting with an " 2494 "experimental mount option 'dioread_nolock' " 2495 "for blocksize < PAGE_SIZE"); 2496 } 2497 return 1; 2498 } 2499 2500 static inline void ext4_show_quota_options(struct seq_file *seq, 2501 struct super_block *sb) 2502 { 2503 #if defined(CONFIG_QUOTA) 2504 struct ext4_sb_info *sbi = EXT4_SB(sb); 2505 char *usr_qf_name, *grp_qf_name; 2506 2507 if (sbi->s_jquota_fmt) { 2508 char *fmtname = ""; 2509 2510 switch (sbi->s_jquota_fmt) { 2511 case QFMT_VFS_OLD: 2512 fmtname = "vfsold"; 2513 break; 2514 case QFMT_VFS_V0: 2515 fmtname = "vfsv0"; 2516 break; 2517 case QFMT_VFS_V1: 2518 fmtname = "vfsv1"; 2519 break; 2520 } 2521 seq_printf(seq, ",jqfmt=%s", fmtname); 2522 } 2523 2524 rcu_read_lock(); 2525 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2526 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2527 if (usr_qf_name) 2528 seq_show_option(seq, "usrjquota", usr_qf_name); 2529 if (grp_qf_name) 2530 seq_show_option(seq, "grpjquota", grp_qf_name); 2531 rcu_read_unlock(); 2532 #endif 2533 } 2534 2535 static const char *token2str(int token) 2536 { 2537 const struct match_token *t; 2538 2539 for (t = tokens; t->token != Opt_err; t++) 2540 if (t->token == token && !strchr(t->pattern, '=')) 2541 break; 2542 return t->pattern; 2543 } 2544 2545 /* 2546 * Show an option if 2547 * - it's set to a non-default value OR 2548 * - if the per-sb default is different from the global default 2549 */ 2550 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2551 int nodefs) 2552 { 2553 struct ext4_sb_info *sbi = EXT4_SB(sb); 2554 struct ext4_super_block *es = sbi->s_es; 2555 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2556 const struct mount_opts *m; 2557 char sep = nodefs ? '\n' : ','; 2558 2559 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2560 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2561 2562 if (sbi->s_sb_block != 1) 2563 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2564 2565 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2566 int want_set = m->flags & MOPT_SET; 2567 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2568 (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP) 2569 continue; 2570 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2571 continue; /* skip if same as the default */ 2572 if ((want_set && 2573 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2574 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2575 continue; /* select Opt_noFoo vs Opt_Foo */ 2576 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2577 } 2578 2579 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2580 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2581 SEQ_OPTS_PRINT("resuid=%u", 2582 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2583 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2584 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2585 SEQ_OPTS_PRINT("resgid=%u", 2586 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2587 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2588 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2589 SEQ_OPTS_PUTS("errors=remount-ro"); 2590 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2591 SEQ_OPTS_PUTS("errors=continue"); 2592 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2593 SEQ_OPTS_PUTS("errors=panic"); 2594 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2595 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2596 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2597 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2598 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2599 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2600 if (sb->s_flags & SB_I_VERSION) 2601 SEQ_OPTS_PUTS("i_version"); 2602 if (nodefs || sbi->s_stripe) 2603 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2604 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2605 (sbi->s_mount_opt ^ def_mount_opt)) { 2606 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2607 SEQ_OPTS_PUTS("data=journal"); 2608 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2609 SEQ_OPTS_PUTS("data=ordered"); 2610 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2611 SEQ_OPTS_PUTS("data=writeback"); 2612 } 2613 if (nodefs || 2614 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2615 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2616 sbi->s_inode_readahead_blks); 2617 2618 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2619 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2620 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2621 if (nodefs || sbi->s_max_dir_size_kb) 2622 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2623 if (test_opt(sb, DATA_ERR_ABORT)) 2624 SEQ_OPTS_PUTS("data_err=abort"); 2625 2626 fscrypt_show_test_dummy_encryption(seq, sep, sb); 2627 2628 if (sb->s_flags & SB_INLINECRYPT) 2629 SEQ_OPTS_PUTS("inlinecrypt"); 2630 2631 if (test_opt(sb, DAX_ALWAYS)) { 2632 if (IS_EXT2_SB(sb)) 2633 SEQ_OPTS_PUTS("dax"); 2634 else 2635 SEQ_OPTS_PUTS("dax=always"); 2636 } else if (test_opt2(sb, DAX_NEVER)) { 2637 SEQ_OPTS_PUTS("dax=never"); 2638 } else if (test_opt2(sb, DAX_INODE)) { 2639 SEQ_OPTS_PUTS("dax=inode"); 2640 } 2641 2642 if (test_opt2(sb, JOURNAL_FAST_COMMIT)) 2643 SEQ_OPTS_PUTS("fast_commit"); 2644 2645 ext4_show_quota_options(seq, sb); 2646 return 0; 2647 } 2648 2649 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2650 { 2651 return _ext4_show_options(seq, root->d_sb, 0); 2652 } 2653 2654 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2655 { 2656 struct super_block *sb = seq->private; 2657 int rc; 2658 2659 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2660 rc = _ext4_show_options(seq, sb, 1); 2661 seq_puts(seq, "\n"); 2662 return rc; 2663 } 2664 2665 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2666 int read_only) 2667 { 2668 struct ext4_sb_info *sbi = EXT4_SB(sb); 2669 int err = 0; 2670 2671 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2672 ext4_msg(sb, KERN_ERR, "revision level too high, " 2673 "forcing read-only mode"); 2674 err = -EROFS; 2675 goto done; 2676 } 2677 if (read_only) 2678 goto done; 2679 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2680 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2681 "running e2fsck is recommended"); 2682 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2683 ext4_msg(sb, KERN_WARNING, 2684 "warning: mounting fs with errors, " 2685 "running e2fsck is recommended"); 2686 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2687 le16_to_cpu(es->s_mnt_count) >= 2688 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2689 ext4_msg(sb, KERN_WARNING, 2690 "warning: maximal mount count reached, " 2691 "running e2fsck is recommended"); 2692 else if (le32_to_cpu(es->s_checkinterval) && 2693 (ext4_get_tstamp(es, s_lastcheck) + 2694 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2695 ext4_msg(sb, KERN_WARNING, 2696 "warning: checktime reached, " 2697 "running e2fsck is recommended"); 2698 if (!sbi->s_journal) 2699 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2700 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2701 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2702 le16_add_cpu(&es->s_mnt_count, 1); 2703 ext4_update_tstamp(es, s_mtime); 2704 if (sbi->s_journal) 2705 ext4_set_feature_journal_needs_recovery(sb); 2706 2707 err = ext4_commit_super(sb, 1); 2708 done: 2709 if (test_opt(sb, DEBUG)) 2710 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2711 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2712 sb->s_blocksize, 2713 sbi->s_groups_count, 2714 EXT4_BLOCKS_PER_GROUP(sb), 2715 EXT4_INODES_PER_GROUP(sb), 2716 sbi->s_mount_opt, sbi->s_mount_opt2); 2717 2718 cleancache_init_fs(sb); 2719 return err; 2720 } 2721 2722 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2723 { 2724 struct ext4_sb_info *sbi = EXT4_SB(sb); 2725 struct flex_groups **old_groups, **new_groups; 2726 int size, i, j; 2727 2728 if (!sbi->s_log_groups_per_flex) 2729 return 0; 2730 2731 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2732 if (size <= sbi->s_flex_groups_allocated) 2733 return 0; 2734 2735 new_groups = kvzalloc(roundup_pow_of_two(size * 2736 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 2737 if (!new_groups) { 2738 ext4_msg(sb, KERN_ERR, 2739 "not enough memory for %d flex group pointers", size); 2740 return -ENOMEM; 2741 } 2742 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 2743 new_groups[i] = kvzalloc(roundup_pow_of_two( 2744 sizeof(struct flex_groups)), 2745 GFP_KERNEL); 2746 if (!new_groups[i]) { 2747 for (j = sbi->s_flex_groups_allocated; j < i; j++) 2748 kvfree(new_groups[j]); 2749 kvfree(new_groups); 2750 ext4_msg(sb, KERN_ERR, 2751 "not enough memory for %d flex groups", size); 2752 return -ENOMEM; 2753 } 2754 } 2755 rcu_read_lock(); 2756 old_groups = rcu_dereference(sbi->s_flex_groups); 2757 if (old_groups) 2758 memcpy(new_groups, old_groups, 2759 (sbi->s_flex_groups_allocated * 2760 sizeof(struct flex_groups *))); 2761 rcu_read_unlock(); 2762 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 2763 sbi->s_flex_groups_allocated = size; 2764 if (old_groups) 2765 ext4_kvfree_array_rcu(old_groups); 2766 return 0; 2767 } 2768 2769 static int ext4_fill_flex_info(struct super_block *sb) 2770 { 2771 struct ext4_sb_info *sbi = EXT4_SB(sb); 2772 struct ext4_group_desc *gdp = NULL; 2773 struct flex_groups *fg; 2774 ext4_group_t flex_group; 2775 int i, err; 2776 2777 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2778 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2779 sbi->s_log_groups_per_flex = 0; 2780 return 1; 2781 } 2782 2783 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2784 if (err) 2785 goto failed; 2786 2787 for (i = 0; i < sbi->s_groups_count; i++) { 2788 gdp = ext4_get_group_desc(sb, i, NULL); 2789 2790 flex_group = ext4_flex_group(sbi, i); 2791 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 2792 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 2793 atomic64_add(ext4_free_group_clusters(sb, gdp), 2794 &fg->free_clusters); 2795 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 2796 } 2797 2798 return 1; 2799 failed: 2800 return 0; 2801 } 2802 2803 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2804 struct ext4_group_desc *gdp) 2805 { 2806 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2807 __u16 crc = 0; 2808 __le32 le_group = cpu_to_le32(block_group); 2809 struct ext4_sb_info *sbi = EXT4_SB(sb); 2810 2811 if (ext4_has_metadata_csum(sbi->s_sb)) { 2812 /* Use new metadata_csum algorithm */ 2813 __u32 csum32; 2814 __u16 dummy_csum = 0; 2815 2816 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2817 sizeof(le_group)); 2818 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2819 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2820 sizeof(dummy_csum)); 2821 offset += sizeof(dummy_csum); 2822 if (offset < sbi->s_desc_size) 2823 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2824 sbi->s_desc_size - offset); 2825 2826 crc = csum32 & 0xFFFF; 2827 goto out; 2828 } 2829 2830 /* old crc16 code */ 2831 if (!ext4_has_feature_gdt_csum(sb)) 2832 return 0; 2833 2834 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2835 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2836 crc = crc16(crc, (__u8 *)gdp, offset); 2837 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2838 /* for checksum of struct ext4_group_desc do the rest...*/ 2839 if (ext4_has_feature_64bit(sb) && 2840 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2841 crc = crc16(crc, (__u8 *)gdp + offset, 2842 le16_to_cpu(sbi->s_es->s_desc_size) - 2843 offset); 2844 2845 out: 2846 return cpu_to_le16(crc); 2847 } 2848 2849 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2850 struct ext4_group_desc *gdp) 2851 { 2852 if (ext4_has_group_desc_csum(sb) && 2853 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2854 return 0; 2855 2856 return 1; 2857 } 2858 2859 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2860 struct ext4_group_desc *gdp) 2861 { 2862 if (!ext4_has_group_desc_csum(sb)) 2863 return; 2864 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2865 } 2866 2867 /* Called at mount-time, super-block is locked */ 2868 static int ext4_check_descriptors(struct super_block *sb, 2869 ext4_fsblk_t sb_block, 2870 ext4_group_t *first_not_zeroed) 2871 { 2872 struct ext4_sb_info *sbi = EXT4_SB(sb); 2873 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2874 ext4_fsblk_t last_block; 2875 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2876 ext4_fsblk_t block_bitmap; 2877 ext4_fsblk_t inode_bitmap; 2878 ext4_fsblk_t inode_table; 2879 int flexbg_flag = 0; 2880 ext4_group_t i, grp = sbi->s_groups_count; 2881 2882 if (ext4_has_feature_flex_bg(sb)) 2883 flexbg_flag = 1; 2884 2885 ext4_debug("Checking group descriptors"); 2886 2887 for (i = 0; i < sbi->s_groups_count; i++) { 2888 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2889 2890 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2891 last_block = ext4_blocks_count(sbi->s_es) - 1; 2892 else 2893 last_block = first_block + 2894 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2895 2896 if ((grp == sbi->s_groups_count) && 2897 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2898 grp = i; 2899 2900 block_bitmap = ext4_block_bitmap(sb, gdp); 2901 if (block_bitmap == sb_block) { 2902 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2903 "Block bitmap for group %u overlaps " 2904 "superblock", i); 2905 if (!sb_rdonly(sb)) 2906 return 0; 2907 } 2908 if (block_bitmap >= sb_block + 1 && 2909 block_bitmap <= last_bg_block) { 2910 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2911 "Block bitmap for group %u overlaps " 2912 "block group descriptors", i); 2913 if (!sb_rdonly(sb)) 2914 return 0; 2915 } 2916 if (block_bitmap < first_block || block_bitmap > last_block) { 2917 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2918 "Block bitmap for group %u not in group " 2919 "(block %llu)!", i, block_bitmap); 2920 return 0; 2921 } 2922 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2923 if (inode_bitmap == sb_block) { 2924 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2925 "Inode bitmap for group %u overlaps " 2926 "superblock", i); 2927 if (!sb_rdonly(sb)) 2928 return 0; 2929 } 2930 if (inode_bitmap >= sb_block + 1 && 2931 inode_bitmap <= last_bg_block) { 2932 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2933 "Inode bitmap for group %u overlaps " 2934 "block group descriptors", i); 2935 if (!sb_rdonly(sb)) 2936 return 0; 2937 } 2938 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2939 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2940 "Inode bitmap for group %u not in group " 2941 "(block %llu)!", i, inode_bitmap); 2942 return 0; 2943 } 2944 inode_table = ext4_inode_table(sb, gdp); 2945 if (inode_table == sb_block) { 2946 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2947 "Inode table for group %u overlaps " 2948 "superblock", i); 2949 if (!sb_rdonly(sb)) 2950 return 0; 2951 } 2952 if (inode_table >= sb_block + 1 && 2953 inode_table <= last_bg_block) { 2954 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2955 "Inode table for group %u overlaps " 2956 "block group descriptors", i); 2957 if (!sb_rdonly(sb)) 2958 return 0; 2959 } 2960 if (inode_table < first_block || 2961 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2962 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2963 "Inode table for group %u not in group " 2964 "(block %llu)!", i, inode_table); 2965 return 0; 2966 } 2967 ext4_lock_group(sb, i); 2968 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2969 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2970 "Checksum for group %u failed (%u!=%u)", 2971 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2972 gdp)), le16_to_cpu(gdp->bg_checksum)); 2973 if (!sb_rdonly(sb)) { 2974 ext4_unlock_group(sb, i); 2975 return 0; 2976 } 2977 } 2978 ext4_unlock_group(sb, i); 2979 if (!flexbg_flag) 2980 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2981 } 2982 if (NULL != first_not_zeroed) 2983 *first_not_zeroed = grp; 2984 return 1; 2985 } 2986 2987 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2988 * the superblock) which were deleted from all directories, but held open by 2989 * a process at the time of a crash. We walk the list and try to delete these 2990 * inodes at recovery time (only with a read-write filesystem). 2991 * 2992 * In order to keep the orphan inode chain consistent during traversal (in 2993 * case of crash during recovery), we link each inode into the superblock 2994 * orphan list_head and handle it the same way as an inode deletion during 2995 * normal operation (which journals the operations for us). 2996 * 2997 * We only do an iget() and an iput() on each inode, which is very safe if we 2998 * accidentally point at an in-use or already deleted inode. The worst that 2999 * can happen in this case is that we get a "bit already cleared" message from 3000 * ext4_free_inode(). The only reason we would point at a wrong inode is if 3001 * e2fsck was run on this filesystem, and it must have already done the orphan 3002 * inode cleanup for us, so we can safely abort without any further action. 3003 */ 3004 static void ext4_orphan_cleanup(struct super_block *sb, 3005 struct ext4_super_block *es) 3006 { 3007 unsigned int s_flags = sb->s_flags; 3008 int ret, nr_orphans = 0, nr_truncates = 0; 3009 #ifdef CONFIG_QUOTA 3010 int quota_update = 0; 3011 int i; 3012 #endif 3013 if (!es->s_last_orphan) { 3014 jbd_debug(4, "no orphan inodes to clean up\n"); 3015 return; 3016 } 3017 3018 if (bdev_read_only(sb->s_bdev)) { 3019 ext4_msg(sb, KERN_ERR, "write access " 3020 "unavailable, skipping orphan cleanup"); 3021 return; 3022 } 3023 3024 /* Check if feature set would not allow a r/w mount */ 3025 if (!ext4_feature_set_ok(sb, 0)) { 3026 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 3027 "unknown ROCOMPAT features"); 3028 return; 3029 } 3030 3031 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3032 /* don't clear list on RO mount w/ errors */ 3033 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 3034 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 3035 "clearing orphan list.\n"); 3036 es->s_last_orphan = 0; 3037 } 3038 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3039 return; 3040 } 3041 3042 if (s_flags & SB_RDONLY) { 3043 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 3044 sb->s_flags &= ~SB_RDONLY; 3045 } 3046 #ifdef CONFIG_QUOTA 3047 /* Needed for iput() to work correctly and not trash data */ 3048 sb->s_flags |= SB_ACTIVE; 3049 3050 /* 3051 * Turn on quotas which were not enabled for read-only mounts if 3052 * filesystem has quota feature, so that they are updated correctly. 3053 */ 3054 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 3055 int ret = ext4_enable_quotas(sb); 3056 3057 if (!ret) 3058 quota_update = 1; 3059 else 3060 ext4_msg(sb, KERN_ERR, 3061 "Cannot turn on quotas: error %d", ret); 3062 } 3063 3064 /* Turn on journaled quotas used for old sytle */ 3065 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3066 if (EXT4_SB(sb)->s_qf_names[i]) { 3067 int ret = ext4_quota_on_mount(sb, i); 3068 3069 if (!ret) 3070 quota_update = 1; 3071 else 3072 ext4_msg(sb, KERN_ERR, 3073 "Cannot turn on journaled " 3074 "quota: type %d: error %d", i, ret); 3075 } 3076 } 3077 #endif 3078 3079 while (es->s_last_orphan) { 3080 struct inode *inode; 3081 3082 /* 3083 * We may have encountered an error during cleanup; if 3084 * so, skip the rest. 3085 */ 3086 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3087 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3088 es->s_last_orphan = 0; 3089 break; 3090 } 3091 3092 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 3093 if (IS_ERR(inode)) { 3094 es->s_last_orphan = 0; 3095 break; 3096 } 3097 3098 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 3099 dquot_initialize(inode); 3100 if (inode->i_nlink) { 3101 if (test_opt(sb, DEBUG)) 3102 ext4_msg(sb, KERN_DEBUG, 3103 "%s: truncating inode %lu to %lld bytes", 3104 __func__, inode->i_ino, inode->i_size); 3105 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 3106 inode->i_ino, inode->i_size); 3107 inode_lock(inode); 3108 truncate_inode_pages(inode->i_mapping, inode->i_size); 3109 ret = ext4_truncate(inode); 3110 if (ret) 3111 ext4_std_error(inode->i_sb, ret); 3112 inode_unlock(inode); 3113 nr_truncates++; 3114 } else { 3115 if (test_opt(sb, DEBUG)) 3116 ext4_msg(sb, KERN_DEBUG, 3117 "%s: deleting unreferenced inode %lu", 3118 __func__, inode->i_ino); 3119 jbd_debug(2, "deleting unreferenced inode %lu\n", 3120 inode->i_ino); 3121 nr_orphans++; 3122 } 3123 iput(inode); /* The delete magic happens here! */ 3124 } 3125 3126 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 3127 3128 if (nr_orphans) 3129 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 3130 PLURAL(nr_orphans)); 3131 if (nr_truncates) 3132 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 3133 PLURAL(nr_truncates)); 3134 #ifdef CONFIG_QUOTA 3135 /* Turn off quotas if they were enabled for orphan cleanup */ 3136 if (quota_update) { 3137 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3138 if (sb_dqopt(sb)->files[i]) 3139 dquot_quota_off(sb, i); 3140 } 3141 } 3142 #endif 3143 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 3144 } 3145 3146 /* 3147 * Maximal extent format file size. 3148 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3149 * extent format containers, within a sector_t, and within i_blocks 3150 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3151 * so that won't be a limiting factor. 3152 * 3153 * However there is other limiting factor. We do store extents in the form 3154 * of starting block and length, hence the resulting length of the extent 3155 * covering maximum file size must fit into on-disk format containers as 3156 * well. Given that length is always by 1 unit bigger than max unit (because 3157 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3158 * 3159 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3160 */ 3161 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3162 { 3163 loff_t res; 3164 loff_t upper_limit = MAX_LFS_FILESIZE; 3165 3166 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3167 3168 if (!has_huge_files) { 3169 upper_limit = (1LL << 32) - 1; 3170 3171 /* total blocks in file system block size */ 3172 upper_limit >>= (blkbits - 9); 3173 upper_limit <<= blkbits; 3174 } 3175 3176 /* 3177 * 32-bit extent-start container, ee_block. We lower the maxbytes 3178 * by one fs block, so ee_len can cover the extent of maximum file 3179 * size 3180 */ 3181 res = (1LL << 32) - 1; 3182 res <<= blkbits; 3183 3184 /* Sanity check against vm- & vfs- imposed limits */ 3185 if (res > upper_limit) 3186 res = upper_limit; 3187 3188 return res; 3189 } 3190 3191 /* 3192 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3193 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3194 * We need to be 1 filesystem block less than the 2^48 sector limit. 3195 */ 3196 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3197 { 3198 loff_t res = EXT4_NDIR_BLOCKS; 3199 int meta_blocks; 3200 loff_t upper_limit; 3201 /* This is calculated to be the largest file size for a dense, block 3202 * mapped file such that the file's total number of 512-byte sectors, 3203 * including data and all indirect blocks, does not exceed (2^48 - 1). 3204 * 3205 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3206 * number of 512-byte sectors of the file. 3207 */ 3208 3209 if (!has_huge_files) { 3210 /* 3211 * !has_huge_files or implies that the inode i_block field 3212 * represents total file blocks in 2^32 512-byte sectors == 3213 * size of vfs inode i_blocks * 8 3214 */ 3215 upper_limit = (1LL << 32) - 1; 3216 3217 /* total blocks in file system block size */ 3218 upper_limit >>= (bits - 9); 3219 3220 } else { 3221 /* 3222 * We use 48 bit ext4_inode i_blocks 3223 * With EXT4_HUGE_FILE_FL set the i_blocks 3224 * represent total number of blocks in 3225 * file system block size 3226 */ 3227 upper_limit = (1LL << 48) - 1; 3228 3229 } 3230 3231 /* indirect blocks */ 3232 meta_blocks = 1; 3233 /* double indirect blocks */ 3234 meta_blocks += 1 + (1LL << (bits-2)); 3235 /* tripple indirect blocks */ 3236 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 3237 3238 upper_limit -= meta_blocks; 3239 upper_limit <<= bits; 3240 3241 res += 1LL << (bits-2); 3242 res += 1LL << (2*(bits-2)); 3243 res += 1LL << (3*(bits-2)); 3244 res <<= bits; 3245 if (res > upper_limit) 3246 res = upper_limit; 3247 3248 if (res > MAX_LFS_FILESIZE) 3249 res = MAX_LFS_FILESIZE; 3250 3251 return res; 3252 } 3253 3254 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3255 ext4_fsblk_t logical_sb_block, int nr) 3256 { 3257 struct ext4_sb_info *sbi = EXT4_SB(sb); 3258 ext4_group_t bg, first_meta_bg; 3259 int has_super = 0; 3260 3261 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3262 3263 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3264 return logical_sb_block + nr + 1; 3265 bg = sbi->s_desc_per_block * nr; 3266 if (ext4_bg_has_super(sb, bg)) 3267 has_super = 1; 3268 3269 /* 3270 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3271 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3272 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3273 * compensate. 3274 */ 3275 if (sb->s_blocksize == 1024 && nr == 0 && 3276 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3277 has_super++; 3278 3279 return (has_super + ext4_group_first_block_no(sb, bg)); 3280 } 3281 3282 /** 3283 * ext4_get_stripe_size: Get the stripe size. 3284 * @sbi: In memory super block info 3285 * 3286 * If we have specified it via mount option, then 3287 * use the mount option value. If the value specified at mount time is 3288 * greater than the blocks per group use the super block value. 3289 * If the super block value is greater than blocks per group return 0. 3290 * Allocator needs it be less than blocks per group. 3291 * 3292 */ 3293 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3294 { 3295 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3296 unsigned long stripe_width = 3297 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3298 int ret; 3299 3300 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3301 ret = sbi->s_stripe; 3302 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3303 ret = stripe_width; 3304 else if (stride && stride <= sbi->s_blocks_per_group) 3305 ret = stride; 3306 else 3307 ret = 0; 3308 3309 /* 3310 * If the stripe width is 1, this makes no sense and 3311 * we set it to 0 to turn off stripe handling code. 3312 */ 3313 if (ret <= 1) 3314 ret = 0; 3315 3316 return ret; 3317 } 3318 3319 /* 3320 * Check whether this filesystem can be mounted based on 3321 * the features present and the RDONLY/RDWR mount requested. 3322 * Returns 1 if this filesystem can be mounted as requested, 3323 * 0 if it cannot be. 3324 */ 3325 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 3326 { 3327 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3328 ext4_msg(sb, KERN_ERR, 3329 "Couldn't mount because of " 3330 "unsupported optional features (%x)", 3331 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3332 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3333 return 0; 3334 } 3335 3336 #ifndef CONFIG_UNICODE 3337 if (ext4_has_feature_casefold(sb)) { 3338 ext4_msg(sb, KERN_ERR, 3339 "Filesystem with casefold feature cannot be " 3340 "mounted without CONFIG_UNICODE"); 3341 return 0; 3342 } 3343 #endif 3344 3345 if (readonly) 3346 return 1; 3347 3348 if (ext4_has_feature_readonly(sb)) { 3349 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3350 sb->s_flags |= SB_RDONLY; 3351 return 1; 3352 } 3353 3354 /* Check that feature set is OK for a read-write mount */ 3355 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3356 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3357 "unsupported optional features (%x)", 3358 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3359 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3360 return 0; 3361 } 3362 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3363 ext4_msg(sb, KERN_ERR, 3364 "Can't support bigalloc feature without " 3365 "extents feature\n"); 3366 return 0; 3367 } 3368 3369 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3370 if (!readonly && (ext4_has_feature_quota(sb) || 3371 ext4_has_feature_project(sb))) { 3372 ext4_msg(sb, KERN_ERR, 3373 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3374 return 0; 3375 } 3376 #endif /* CONFIG_QUOTA */ 3377 return 1; 3378 } 3379 3380 /* 3381 * This function is called once a day if we have errors logged 3382 * on the file system 3383 */ 3384 static void print_daily_error_info(struct timer_list *t) 3385 { 3386 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3387 struct super_block *sb = sbi->s_sb; 3388 struct ext4_super_block *es = sbi->s_es; 3389 3390 if (es->s_error_count) 3391 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3392 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3393 le32_to_cpu(es->s_error_count)); 3394 if (es->s_first_error_time) { 3395 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3396 sb->s_id, 3397 ext4_get_tstamp(es, s_first_error_time), 3398 (int) sizeof(es->s_first_error_func), 3399 es->s_first_error_func, 3400 le32_to_cpu(es->s_first_error_line)); 3401 if (es->s_first_error_ino) 3402 printk(KERN_CONT ": inode %u", 3403 le32_to_cpu(es->s_first_error_ino)); 3404 if (es->s_first_error_block) 3405 printk(KERN_CONT ": block %llu", (unsigned long long) 3406 le64_to_cpu(es->s_first_error_block)); 3407 printk(KERN_CONT "\n"); 3408 } 3409 if (es->s_last_error_time) { 3410 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3411 sb->s_id, 3412 ext4_get_tstamp(es, s_last_error_time), 3413 (int) sizeof(es->s_last_error_func), 3414 es->s_last_error_func, 3415 le32_to_cpu(es->s_last_error_line)); 3416 if (es->s_last_error_ino) 3417 printk(KERN_CONT ": inode %u", 3418 le32_to_cpu(es->s_last_error_ino)); 3419 if (es->s_last_error_block) 3420 printk(KERN_CONT ": block %llu", (unsigned long long) 3421 le64_to_cpu(es->s_last_error_block)); 3422 printk(KERN_CONT "\n"); 3423 } 3424 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3425 } 3426 3427 /* Find next suitable group and run ext4_init_inode_table */ 3428 static int ext4_run_li_request(struct ext4_li_request *elr) 3429 { 3430 struct ext4_group_desc *gdp = NULL; 3431 struct super_block *sb = elr->lr_super; 3432 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3433 ext4_group_t group = elr->lr_next_group; 3434 unsigned long timeout = 0; 3435 unsigned int prefetch_ios = 0; 3436 int ret = 0; 3437 3438 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3439 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3440 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3441 if (prefetch_ios) 3442 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3443 prefetch_ios); 3444 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3445 prefetch_ios); 3446 if (group >= elr->lr_next_group) { 3447 ret = 1; 3448 if (elr->lr_first_not_zeroed != ngroups && 3449 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3450 elr->lr_next_group = elr->lr_first_not_zeroed; 3451 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3452 ret = 0; 3453 } 3454 } 3455 return ret; 3456 } 3457 3458 for (; group < ngroups; group++) { 3459 gdp = ext4_get_group_desc(sb, group, NULL); 3460 if (!gdp) { 3461 ret = 1; 3462 break; 3463 } 3464 3465 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3466 break; 3467 } 3468 3469 if (group >= ngroups) 3470 ret = 1; 3471 3472 if (!ret) { 3473 timeout = jiffies; 3474 ret = ext4_init_inode_table(sb, group, 3475 elr->lr_timeout ? 0 : 1); 3476 trace_ext4_lazy_itable_init(sb, group); 3477 if (elr->lr_timeout == 0) { 3478 timeout = (jiffies - timeout) * 3479 EXT4_SB(elr->lr_super)->s_li_wait_mult; 3480 elr->lr_timeout = timeout; 3481 } 3482 elr->lr_next_sched = jiffies + elr->lr_timeout; 3483 elr->lr_next_group = group + 1; 3484 } 3485 return ret; 3486 } 3487 3488 /* 3489 * Remove lr_request from the list_request and free the 3490 * request structure. Should be called with li_list_mtx held 3491 */ 3492 static void ext4_remove_li_request(struct ext4_li_request *elr) 3493 { 3494 if (!elr) 3495 return; 3496 3497 list_del(&elr->lr_request); 3498 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3499 kfree(elr); 3500 } 3501 3502 static void ext4_unregister_li_request(struct super_block *sb) 3503 { 3504 mutex_lock(&ext4_li_mtx); 3505 if (!ext4_li_info) { 3506 mutex_unlock(&ext4_li_mtx); 3507 return; 3508 } 3509 3510 mutex_lock(&ext4_li_info->li_list_mtx); 3511 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3512 mutex_unlock(&ext4_li_info->li_list_mtx); 3513 mutex_unlock(&ext4_li_mtx); 3514 } 3515 3516 static struct task_struct *ext4_lazyinit_task; 3517 3518 /* 3519 * This is the function where ext4lazyinit thread lives. It walks 3520 * through the request list searching for next scheduled filesystem. 3521 * When such a fs is found, run the lazy initialization request 3522 * (ext4_rn_li_request) and keep track of the time spend in this 3523 * function. Based on that time we compute next schedule time of 3524 * the request. When walking through the list is complete, compute 3525 * next waking time and put itself into sleep. 3526 */ 3527 static int ext4_lazyinit_thread(void *arg) 3528 { 3529 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3530 struct list_head *pos, *n; 3531 struct ext4_li_request *elr; 3532 unsigned long next_wakeup, cur; 3533 3534 BUG_ON(NULL == eli); 3535 3536 cont_thread: 3537 while (true) { 3538 next_wakeup = MAX_JIFFY_OFFSET; 3539 3540 mutex_lock(&eli->li_list_mtx); 3541 if (list_empty(&eli->li_request_list)) { 3542 mutex_unlock(&eli->li_list_mtx); 3543 goto exit_thread; 3544 } 3545 list_for_each_safe(pos, n, &eli->li_request_list) { 3546 int err = 0; 3547 int progress = 0; 3548 elr = list_entry(pos, struct ext4_li_request, 3549 lr_request); 3550 3551 if (time_before(jiffies, elr->lr_next_sched)) { 3552 if (time_before(elr->lr_next_sched, next_wakeup)) 3553 next_wakeup = elr->lr_next_sched; 3554 continue; 3555 } 3556 if (down_read_trylock(&elr->lr_super->s_umount)) { 3557 if (sb_start_write_trylock(elr->lr_super)) { 3558 progress = 1; 3559 /* 3560 * We hold sb->s_umount, sb can not 3561 * be removed from the list, it is 3562 * now safe to drop li_list_mtx 3563 */ 3564 mutex_unlock(&eli->li_list_mtx); 3565 err = ext4_run_li_request(elr); 3566 sb_end_write(elr->lr_super); 3567 mutex_lock(&eli->li_list_mtx); 3568 n = pos->next; 3569 } 3570 up_read((&elr->lr_super->s_umount)); 3571 } 3572 /* error, remove the lazy_init job */ 3573 if (err) { 3574 ext4_remove_li_request(elr); 3575 continue; 3576 } 3577 if (!progress) { 3578 elr->lr_next_sched = jiffies + 3579 (prandom_u32() 3580 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3581 } 3582 if (time_before(elr->lr_next_sched, next_wakeup)) 3583 next_wakeup = elr->lr_next_sched; 3584 } 3585 mutex_unlock(&eli->li_list_mtx); 3586 3587 try_to_freeze(); 3588 3589 cur = jiffies; 3590 if ((time_after_eq(cur, next_wakeup)) || 3591 (MAX_JIFFY_OFFSET == next_wakeup)) { 3592 cond_resched(); 3593 continue; 3594 } 3595 3596 schedule_timeout_interruptible(next_wakeup - cur); 3597 3598 if (kthread_should_stop()) { 3599 ext4_clear_request_list(); 3600 goto exit_thread; 3601 } 3602 } 3603 3604 exit_thread: 3605 /* 3606 * It looks like the request list is empty, but we need 3607 * to check it under the li_list_mtx lock, to prevent any 3608 * additions into it, and of course we should lock ext4_li_mtx 3609 * to atomically free the list and ext4_li_info, because at 3610 * this point another ext4 filesystem could be registering 3611 * new one. 3612 */ 3613 mutex_lock(&ext4_li_mtx); 3614 mutex_lock(&eli->li_list_mtx); 3615 if (!list_empty(&eli->li_request_list)) { 3616 mutex_unlock(&eli->li_list_mtx); 3617 mutex_unlock(&ext4_li_mtx); 3618 goto cont_thread; 3619 } 3620 mutex_unlock(&eli->li_list_mtx); 3621 kfree(ext4_li_info); 3622 ext4_li_info = NULL; 3623 mutex_unlock(&ext4_li_mtx); 3624 3625 return 0; 3626 } 3627 3628 static void ext4_clear_request_list(void) 3629 { 3630 struct list_head *pos, *n; 3631 struct ext4_li_request *elr; 3632 3633 mutex_lock(&ext4_li_info->li_list_mtx); 3634 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3635 elr = list_entry(pos, struct ext4_li_request, 3636 lr_request); 3637 ext4_remove_li_request(elr); 3638 } 3639 mutex_unlock(&ext4_li_info->li_list_mtx); 3640 } 3641 3642 static int ext4_run_lazyinit_thread(void) 3643 { 3644 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3645 ext4_li_info, "ext4lazyinit"); 3646 if (IS_ERR(ext4_lazyinit_task)) { 3647 int err = PTR_ERR(ext4_lazyinit_task); 3648 ext4_clear_request_list(); 3649 kfree(ext4_li_info); 3650 ext4_li_info = NULL; 3651 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3652 "initialization thread\n", 3653 err); 3654 return err; 3655 } 3656 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3657 return 0; 3658 } 3659 3660 /* 3661 * Check whether it make sense to run itable init. thread or not. 3662 * If there is at least one uninitialized inode table, return 3663 * corresponding group number, else the loop goes through all 3664 * groups and return total number of groups. 3665 */ 3666 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3667 { 3668 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3669 struct ext4_group_desc *gdp = NULL; 3670 3671 if (!ext4_has_group_desc_csum(sb)) 3672 return ngroups; 3673 3674 for (group = 0; group < ngroups; group++) { 3675 gdp = ext4_get_group_desc(sb, group, NULL); 3676 if (!gdp) 3677 continue; 3678 3679 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3680 break; 3681 } 3682 3683 return group; 3684 } 3685 3686 static int ext4_li_info_new(void) 3687 { 3688 struct ext4_lazy_init *eli = NULL; 3689 3690 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3691 if (!eli) 3692 return -ENOMEM; 3693 3694 INIT_LIST_HEAD(&eli->li_request_list); 3695 mutex_init(&eli->li_list_mtx); 3696 3697 eli->li_state |= EXT4_LAZYINIT_QUIT; 3698 3699 ext4_li_info = eli; 3700 3701 return 0; 3702 } 3703 3704 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3705 ext4_group_t start) 3706 { 3707 struct ext4_li_request *elr; 3708 3709 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3710 if (!elr) 3711 return NULL; 3712 3713 elr->lr_super = sb; 3714 elr->lr_first_not_zeroed = start; 3715 if (test_opt(sb, PREFETCH_BLOCK_BITMAPS)) 3716 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3717 else { 3718 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3719 elr->lr_next_group = start; 3720 } 3721 3722 /* 3723 * Randomize first schedule time of the request to 3724 * spread the inode table initialization requests 3725 * better. 3726 */ 3727 elr->lr_next_sched = jiffies + (prandom_u32() % 3728 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3729 return elr; 3730 } 3731 3732 int ext4_register_li_request(struct super_block *sb, 3733 ext4_group_t first_not_zeroed) 3734 { 3735 struct ext4_sb_info *sbi = EXT4_SB(sb); 3736 struct ext4_li_request *elr = NULL; 3737 ext4_group_t ngroups = sbi->s_groups_count; 3738 int ret = 0; 3739 3740 mutex_lock(&ext4_li_mtx); 3741 if (sbi->s_li_request != NULL) { 3742 /* 3743 * Reset timeout so it can be computed again, because 3744 * s_li_wait_mult might have changed. 3745 */ 3746 sbi->s_li_request->lr_timeout = 0; 3747 goto out; 3748 } 3749 3750 if (!test_opt(sb, PREFETCH_BLOCK_BITMAPS) && 3751 (first_not_zeroed == ngroups || sb_rdonly(sb) || 3752 !test_opt(sb, INIT_INODE_TABLE))) 3753 goto out; 3754 3755 elr = ext4_li_request_new(sb, first_not_zeroed); 3756 if (!elr) { 3757 ret = -ENOMEM; 3758 goto out; 3759 } 3760 3761 if (NULL == ext4_li_info) { 3762 ret = ext4_li_info_new(); 3763 if (ret) 3764 goto out; 3765 } 3766 3767 mutex_lock(&ext4_li_info->li_list_mtx); 3768 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3769 mutex_unlock(&ext4_li_info->li_list_mtx); 3770 3771 sbi->s_li_request = elr; 3772 /* 3773 * set elr to NULL here since it has been inserted to 3774 * the request_list and the removal and free of it is 3775 * handled by ext4_clear_request_list from now on. 3776 */ 3777 elr = NULL; 3778 3779 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3780 ret = ext4_run_lazyinit_thread(); 3781 if (ret) 3782 goto out; 3783 } 3784 out: 3785 mutex_unlock(&ext4_li_mtx); 3786 if (ret) 3787 kfree(elr); 3788 return ret; 3789 } 3790 3791 /* 3792 * We do not need to lock anything since this is called on 3793 * module unload. 3794 */ 3795 static void ext4_destroy_lazyinit_thread(void) 3796 { 3797 /* 3798 * If thread exited earlier 3799 * there's nothing to be done. 3800 */ 3801 if (!ext4_li_info || !ext4_lazyinit_task) 3802 return; 3803 3804 kthread_stop(ext4_lazyinit_task); 3805 } 3806 3807 static int set_journal_csum_feature_set(struct super_block *sb) 3808 { 3809 int ret = 1; 3810 int compat, incompat; 3811 struct ext4_sb_info *sbi = EXT4_SB(sb); 3812 3813 if (ext4_has_metadata_csum(sb)) { 3814 /* journal checksum v3 */ 3815 compat = 0; 3816 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3817 } else { 3818 /* journal checksum v1 */ 3819 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3820 incompat = 0; 3821 } 3822 3823 jbd2_journal_clear_features(sbi->s_journal, 3824 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3825 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3826 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3827 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3828 ret = jbd2_journal_set_features(sbi->s_journal, 3829 compat, 0, 3830 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3831 incompat); 3832 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3833 ret = jbd2_journal_set_features(sbi->s_journal, 3834 compat, 0, 3835 incompat); 3836 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3837 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3838 } else { 3839 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3840 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3841 } 3842 3843 return ret; 3844 } 3845 3846 /* 3847 * Note: calculating the overhead so we can be compatible with 3848 * historical BSD practice is quite difficult in the face of 3849 * clusters/bigalloc. This is because multiple metadata blocks from 3850 * different block group can end up in the same allocation cluster. 3851 * Calculating the exact overhead in the face of clustered allocation 3852 * requires either O(all block bitmaps) in memory or O(number of block 3853 * groups**2) in time. We will still calculate the superblock for 3854 * older file systems --- and if we come across with a bigalloc file 3855 * system with zero in s_overhead_clusters the estimate will be close to 3856 * correct especially for very large cluster sizes --- but for newer 3857 * file systems, it's better to calculate this figure once at mkfs 3858 * time, and store it in the superblock. If the superblock value is 3859 * present (even for non-bigalloc file systems), we will use it. 3860 */ 3861 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3862 char *buf) 3863 { 3864 struct ext4_sb_info *sbi = EXT4_SB(sb); 3865 struct ext4_group_desc *gdp; 3866 ext4_fsblk_t first_block, last_block, b; 3867 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3868 int s, j, count = 0; 3869 3870 if (!ext4_has_feature_bigalloc(sb)) 3871 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3872 sbi->s_itb_per_group + 2); 3873 3874 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3875 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3876 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3877 for (i = 0; i < ngroups; i++) { 3878 gdp = ext4_get_group_desc(sb, i, NULL); 3879 b = ext4_block_bitmap(sb, gdp); 3880 if (b >= first_block && b <= last_block) { 3881 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3882 count++; 3883 } 3884 b = ext4_inode_bitmap(sb, gdp); 3885 if (b >= first_block && b <= last_block) { 3886 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3887 count++; 3888 } 3889 b = ext4_inode_table(sb, gdp); 3890 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3891 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3892 int c = EXT4_B2C(sbi, b - first_block); 3893 ext4_set_bit(c, buf); 3894 count++; 3895 } 3896 if (i != grp) 3897 continue; 3898 s = 0; 3899 if (ext4_bg_has_super(sb, grp)) { 3900 ext4_set_bit(s++, buf); 3901 count++; 3902 } 3903 j = ext4_bg_num_gdb(sb, grp); 3904 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3905 ext4_error(sb, "Invalid number of block group " 3906 "descriptor blocks: %d", j); 3907 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3908 } 3909 count += j; 3910 for (; j > 0; j--) 3911 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3912 } 3913 if (!count) 3914 return 0; 3915 return EXT4_CLUSTERS_PER_GROUP(sb) - 3916 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3917 } 3918 3919 /* 3920 * Compute the overhead and stash it in sbi->s_overhead 3921 */ 3922 int ext4_calculate_overhead(struct super_block *sb) 3923 { 3924 struct ext4_sb_info *sbi = EXT4_SB(sb); 3925 struct ext4_super_block *es = sbi->s_es; 3926 struct inode *j_inode; 3927 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3928 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3929 ext4_fsblk_t overhead = 0; 3930 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3931 3932 if (!buf) 3933 return -ENOMEM; 3934 3935 /* 3936 * Compute the overhead (FS structures). This is constant 3937 * for a given filesystem unless the number of block groups 3938 * changes so we cache the previous value until it does. 3939 */ 3940 3941 /* 3942 * All of the blocks before first_data_block are overhead 3943 */ 3944 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3945 3946 /* 3947 * Add the overhead found in each block group 3948 */ 3949 for (i = 0; i < ngroups; i++) { 3950 int blks; 3951 3952 blks = count_overhead(sb, i, buf); 3953 overhead += blks; 3954 if (blks) 3955 memset(buf, 0, PAGE_SIZE); 3956 cond_resched(); 3957 } 3958 3959 /* 3960 * Add the internal journal blocks whether the journal has been 3961 * loaded or not 3962 */ 3963 if (sbi->s_journal && !sbi->s_journal_bdev) 3964 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 3965 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 3966 /* j_inum for internal journal is non-zero */ 3967 j_inode = ext4_get_journal_inode(sb, j_inum); 3968 if (j_inode) { 3969 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3970 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3971 iput(j_inode); 3972 } else { 3973 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3974 } 3975 } 3976 sbi->s_overhead = overhead; 3977 smp_wmb(); 3978 free_page((unsigned long) buf); 3979 return 0; 3980 } 3981 3982 static void ext4_set_resv_clusters(struct super_block *sb) 3983 { 3984 ext4_fsblk_t resv_clusters; 3985 struct ext4_sb_info *sbi = EXT4_SB(sb); 3986 3987 /* 3988 * There's no need to reserve anything when we aren't using extents. 3989 * The space estimates are exact, there are no unwritten extents, 3990 * hole punching doesn't need new metadata... This is needed especially 3991 * to keep ext2/3 backward compatibility. 3992 */ 3993 if (!ext4_has_feature_extents(sb)) 3994 return; 3995 /* 3996 * By default we reserve 2% or 4096 clusters, whichever is smaller. 3997 * This should cover the situations where we can not afford to run 3998 * out of space like for example punch hole, or converting 3999 * unwritten extents in delalloc path. In most cases such 4000 * allocation would require 1, or 2 blocks, higher numbers are 4001 * very rare. 4002 */ 4003 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4004 sbi->s_cluster_bits); 4005 4006 do_div(resv_clusters, 50); 4007 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4008 4009 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4010 } 4011 4012 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 4013 { 4014 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 4015 char *orig_data = kstrdup(data, GFP_KERNEL); 4016 struct buffer_head *bh, **group_desc; 4017 struct ext4_super_block *es = NULL; 4018 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4019 struct flex_groups **flex_groups; 4020 ext4_fsblk_t block; 4021 ext4_fsblk_t sb_block = get_sb_block(&data); 4022 ext4_fsblk_t logical_sb_block; 4023 unsigned long offset = 0; 4024 unsigned long journal_devnum = 0; 4025 unsigned long def_mount_opts; 4026 struct inode *root; 4027 const char *descr; 4028 int ret = -ENOMEM; 4029 int blocksize, clustersize; 4030 unsigned int db_count; 4031 unsigned int i; 4032 int needs_recovery, has_huge_files; 4033 __u64 blocks_count; 4034 int err = 0; 4035 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 4036 ext4_group_t first_not_zeroed; 4037 4038 if ((data && !orig_data) || !sbi) 4039 goto out_free_base; 4040 4041 sbi->s_daxdev = dax_dev; 4042 sbi->s_blockgroup_lock = 4043 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4044 if (!sbi->s_blockgroup_lock) 4045 goto out_free_base; 4046 4047 sb->s_fs_info = sbi; 4048 sbi->s_sb = sb; 4049 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 4050 sbi->s_sb_block = sb_block; 4051 if (sb->s_bdev->bd_part) 4052 sbi->s_sectors_written_start = 4053 part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]); 4054 4055 /* Cleanup superblock name */ 4056 strreplace(sb->s_id, '/', '!'); 4057 4058 /* -EINVAL is default */ 4059 ret = -EINVAL; 4060 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4061 if (!blocksize) { 4062 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4063 goto out_fail; 4064 } 4065 4066 /* 4067 * The ext4 superblock will not be buffer aligned for other than 1kB 4068 * block sizes. We need to calculate the offset from buffer start. 4069 */ 4070 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4071 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4072 offset = do_div(logical_sb_block, blocksize); 4073 } else { 4074 logical_sb_block = sb_block; 4075 } 4076 4077 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4078 if (IS_ERR(bh)) { 4079 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4080 ret = PTR_ERR(bh); 4081 bh = NULL; 4082 goto out_fail; 4083 } 4084 /* 4085 * Note: s_es must be initialized as soon as possible because 4086 * some ext4 macro-instructions depend on its value 4087 */ 4088 es = (struct ext4_super_block *) (bh->b_data + offset); 4089 sbi->s_es = es; 4090 sb->s_magic = le16_to_cpu(es->s_magic); 4091 if (sb->s_magic != EXT4_SUPER_MAGIC) 4092 goto cantfind_ext4; 4093 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 4094 4095 /* Warn if metadata_csum and gdt_csum are both set. */ 4096 if (ext4_has_feature_metadata_csum(sb) && 4097 ext4_has_feature_gdt_csum(sb)) 4098 ext4_warning(sb, "metadata_csum and uninit_bg are " 4099 "redundant flags; please run fsck."); 4100 4101 /* Check for a known checksum algorithm */ 4102 if (!ext4_verify_csum_type(sb, es)) { 4103 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4104 "unknown checksum algorithm."); 4105 silent = 1; 4106 goto cantfind_ext4; 4107 } 4108 4109 /* Load the checksum driver */ 4110 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4111 if (IS_ERR(sbi->s_chksum_driver)) { 4112 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4113 ret = PTR_ERR(sbi->s_chksum_driver); 4114 sbi->s_chksum_driver = NULL; 4115 goto failed_mount; 4116 } 4117 4118 /* Check superblock checksum */ 4119 if (!ext4_superblock_csum_verify(sb, es)) { 4120 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4121 "invalid superblock checksum. Run e2fsck?"); 4122 silent = 1; 4123 ret = -EFSBADCRC; 4124 goto cantfind_ext4; 4125 } 4126 4127 /* Precompute checksum seed for all metadata */ 4128 if (ext4_has_feature_csum_seed(sb)) 4129 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4130 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4131 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4132 sizeof(es->s_uuid)); 4133 4134 /* Set defaults before we parse the mount options */ 4135 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4136 set_opt(sb, INIT_INODE_TABLE); 4137 if (def_mount_opts & EXT4_DEFM_DEBUG) 4138 set_opt(sb, DEBUG); 4139 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4140 set_opt(sb, GRPID); 4141 if (def_mount_opts & EXT4_DEFM_UID16) 4142 set_opt(sb, NO_UID32); 4143 /* xattr user namespace & acls are now defaulted on */ 4144 set_opt(sb, XATTR_USER); 4145 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4146 set_opt(sb, POSIX_ACL); 4147 #endif 4148 if (ext4_has_feature_fast_commit(sb)) 4149 set_opt2(sb, JOURNAL_FAST_COMMIT); 4150 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4151 if (ext4_has_metadata_csum(sb)) 4152 set_opt(sb, JOURNAL_CHECKSUM); 4153 4154 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4155 set_opt(sb, JOURNAL_DATA); 4156 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4157 set_opt(sb, ORDERED_DATA); 4158 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4159 set_opt(sb, WRITEBACK_DATA); 4160 4161 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 4162 set_opt(sb, ERRORS_PANIC); 4163 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 4164 set_opt(sb, ERRORS_CONT); 4165 else 4166 set_opt(sb, ERRORS_RO); 4167 /* block_validity enabled by default; disable with noblock_validity */ 4168 set_opt(sb, BLOCK_VALIDITY); 4169 if (def_mount_opts & EXT4_DEFM_DISCARD) 4170 set_opt(sb, DISCARD); 4171 4172 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 4173 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 4174 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 4175 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 4176 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 4177 4178 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4179 set_opt(sb, BARRIER); 4180 4181 /* 4182 * enable delayed allocation by default 4183 * Use -o nodelalloc to turn it off 4184 */ 4185 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4186 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4187 set_opt(sb, DELALLOC); 4188 4189 /* 4190 * set default s_li_wait_mult for lazyinit, for the case there is 4191 * no mount option specified. 4192 */ 4193 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 4194 4195 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4196 4197 if (blocksize == PAGE_SIZE) 4198 set_opt(sb, DIOREAD_NOLOCK); 4199 4200 if (blocksize < EXT4_MIN_BLOCK_SIZE || 4201 blocksize > EXT4_MAX_BLOCK_SIZE) { 4202 ext4_msg(sb, KERN_ERR, 4203 "Unsupported filesystem blocksize %d (%d log_block_size)", 4204 blocksize, le32_to_cpu(es->s_log_block_size)); 4205 goto failed_mount; 4206 } 4207 4208 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4209 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4210 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4211 } else { 4212 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4213 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4214 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4215 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4216 sbi->s_first_ino); 4217 goto failed_mount; 4218 } 4219 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4220 (!is_power_of_2(sbi->s_inode_size)) || 4221 (sbi->s_inode_size > blocksize)) { 4222 ext4_msg(sb, KERN_ERR, 4223 "unsupported inode size: %d", 4224 sbi->s_inode_size); 4225 ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize); 4226 goto failed_mount; 4227 } 4228 /* 4229 * i_atime_extra is the last extra field available for 4230 * [acm]times in struct ext4_inode. Checking for that 4231 * field should suffice to ensure we have extra space 4232 * for all three. 4233 */ 4234 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4235 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4236 sb->s_time_gran = 1; 4237 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4238 } else { 4239 sb->s_time_gran = NSEC_PER_SEC; 4240 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4241 } 4242 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4243 } 4244 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4245 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4246 EXT4_GOOD_OLD_INODE_SIZE; 4247 if (ext4_has_feature_extra_isize(sb)) { 4248 unsigned v, max = (sbi->s_inode_size - 4249 EXT4_GOOD_OLD_INODE_SIZE); 4250 4251 v = le16_to_cpu(es->s_want_extra_isize); 4252 if (v > max) { 4253 ext4_msg(sb, KERN_ERR, 4254 "bad s_want_extra_isize: %d", v); 4255 goto failed_mount; 4256 } 4257 if (sbi->s_want_extra_isize < v) 4258 sbi->s_want_extra_isize = v; 4259 4260 v = le16_to_cpu(es->s_min_extra_isize); 4261 if (v > max) { 4262 ext4_msg(sb, KERN_ERR, 4263 "bad s_min_extra_isize: %d", v); 4264 goto failed_mount; 4265 } 4266 if (sbi->s_want_extra_isize < v) 4267 sbi->s_want_extra_isize = v; 4268 } 4269 } 4270 4271 if (sbi->s_es->s_mount_opts[0]) { 4272 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 4273 sizeof(sbi->s_es->s_mount_opts), 4274 GFP_KERNEL); 4275 if (!s_mount_opts) 4276 goto failed_mount; 4277 if (!parse_options(s_mount_opts, sb, &journal_devnum, 4278 &journal_ioprio, 0)) { 4279 ext4_msg(sb, KERN_WARNING, 4280 "failed to parse options in superblock: %s", 4281 s_mount_opts); 4282 } 4283 kfree(s_mount_opts); 4284 } 4285 sbi->s_def_mount_opt = sbi->s_mount_opt; 4286 if (!parse_options((char *) data, sb, &journal_devnum, 4287 &journal_ioprio, 0)) 4288 goto failed_mount; 4289 4290 #ifdef CONFIG_UNICODE 4291 if (ext4_has_feature_casefold(sb) && !sb->s_encoding) { 4292 const struct ext4_sb_encodings *encoding_info; 4293 struct unicode_map *encoding; 4294 __u16 encoding_flags; 4295 4296 if (ext4_has_feature_encrypt(sb)) { 4297 ext4_msg(sb, KERN_ERR, 4298 "Can't mount with encoding and encryption"); 4299 goto failed_mount; 4300 } 4301 4302 if (ext4_sb_read_encoding(es, &encoding_info, 4303 &encoding_flags)) { 4304 ext4_msg(sb, KERN_ERR, 4305 "Encoding requested by superblock is unknown"); 4306 goto failed_mount; 4307 } 4308 4309 encoding = utf8_load(encoding_info->version); 4310 if (IS_ERR(encoding)) { 4311 ext4_msg(sb, KERN_ERR, 4312 "can't mount with superblock charset: %s-%s " 4313 "not supported by the kernel. flags: 0x%x.", 4314 encoding_info->name, encoding_info->version, 4315 encoding_flags); 4316 goto failed_mount; 4317 } 4318 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4319 "%s-%s with flags 0x%hx", encoding_info->name, 4320 encoding_info->version?:"\b", encoding_flags); 4321 4322 sb->s_encoding = encoding; 4323 sb->s_encoding_flags = encoding_flags; 4324 } 4325 #endif 4326 4327 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4328 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n"); 4329 /* can't mount with both data=journal and dioread_nolock. */ 4330 clear_opt(sb, DIOREAD_NOLOCK); 4331 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4332 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4333 ext4_msg(sb, KERN_ERR, "can't mount with " 4334 "both data=journal and delalloc"); 4335 goto failed_mount; 4336 } 4337 if (test_opt(sb, DAX_ALWAYS)) { 4338 ext4_msg(sb, KERN_ERR, "can't mount with " 4339 "both data=journal and dax"); 4340 goto failed_mount; 4341 } 4342 if (ext4_has_feature_encrypt(sb)) { 4343 ext4_msg(sb, KERN_WARNING, 4344 "encrypted files will use data=ordered " 4345 "instead of data journaling mode"); 4346 } 4347 if (test_opt(sb, DELALLOC)) 4348 clear_opt(sb, DELALLOC); 4349 } else { 4350 sb->s_iflags |= SB_I_CGROUPWB; 4351 } 4352 4353 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 4354 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 4355 4356 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4357 (ext4_has_compat_features(sb) || 4358 ext4_has_ro_compat_features(sb) || 4359 ext4_has_incompat_features(sb))) 4360 ext4_msg(sb, KERN_WARNING, 4361 "feature flags set on rev 0 fs, " 4362 "running e2fsck is recommended"); 4363 4364 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4365 set_opt2(sb, HURD_COMPAT); 4366 if (ext4_has_feature_64bit(sb)) { 4367 ext4_msg(sb, KERN_ERR, 4368 "The Hurd can't support 64-bit file systems"); 4369 goto failed_mount; 4370 } 4371 4372 /* 4373 * ea_inode feature uses l_i_version field which is not 4374 * available in HURD_COMPAT mode. 4375 */ 4376 if (ext4_has_feature_ea_inode(sb)) { 4377 ext4_msg(sb, KERN_ERR, 4378 "ea_inode feature is not supported for Hurd"); 4379 goto failed_mount; 4380 } 4381 } 4382 4383 if (IS_EXT2_SB(sb)) { 4384 if (ext2_feature_set_ok(sb)) 4385 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4386 "using the ext4 subsystem"); 4387 else { 4388 /* 4389 * If we're probing be silent, if this looks like 4390 * it's actually an ext[34] filesystem. 4391 */ 4392 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4393 goto failed_mount; 4394 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4395 "to feature incompatibilities"); 4396 goto failed_mount; 4397 } 4398 } 4399 4400 if (IS_EXT3_SB(sb)) { 4401 if (ext3_feature_set_ok(sb)) 4402 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4403 "using the ext4 subsystem"); 4404 else { 4405 /* 4406 * If we're probing be silent, if this looks like 4407 * it's actually an ext4 filesystem. 4408 */ 4409 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4410 goto failed_mount; 4411 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4412 "to feature incompatibilities"); 4413 goto failed_mount; 4414 } 4415 } 4416 4417 /* 4418 * Check feature flags regardless of the revision level, since we 4419 * previously didn't change the revision level when setting the flags, 4420 * so there is a chance incompat flags are set on a rev 0 filesystem. 4421 */ 4422 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4423 goto failed_mount; 4424 4425 if (le32_to_cpu(es->s_log_block_size) > 4426 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4427 ext4_msg(sb, KERN_ERR, 4428 "Invalid log block size: %u", 4429 le32_to_cpu(es->s_log_block_size)); 4430 goto failed_mount; 4431 } 4432 if (le32_to_cpu(es->s_log_cluster_size) > 4433 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4434 ext4_msg(sb, KERN_ERR, 4435 "Invalid log cluster size: %u", 4436 le32_to_cpu(es->s_log_cluster_size)); 4437 goto failed_mount; 4438 } 4439 4440 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 4441 ext4_msg(sb, KERN_ERR, 4442 "Number of reserved GDT blocks insanely large: %d", 4443 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4444 goto failed_mount; 4445 } 4446 4447 if (bdev_dax_supported(sb->s_bdev, blocksize)) 4448 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4449 4450 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4451 if (ext4_has_feature_inline_data(sb)) { 4452 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4453 " that may contain inline data"); 4454 goto failed_mount; 4455 } 4456 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4457 ext4_msg(sb, KERN_ERR, 4458 "DAX unsupported by block device."); 4459 goto failed_mount; 4460 } 4461 } 4462 4463 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4464 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4465 es->s_encryption_level); 4466 goto failed_mount; 4467 } 4468 4469 if (sb->s_blocksize != blocksize) { 4470 /* Validate the filesystem blocksize */ 4471 if (!sb_set_blocksize(sb, blocksize)) { 4472 ext4_msg(sb, KERN_ERR, "bad block size %d", 4473 blocksize); 4474 goto failed_mount; 4475 } 4476 4477 brelse(bh); 4478 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4479 offset = do_div(logical_sb_block, blocksize); 4480 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4481 if (IS_ERR(bh)) { 4482 ext4_msg(sb, KERN_ERR, 4483 "Can't read superblock on 2nd try"); 4484 ret = PTR_ERR(bh); 4485 bh = NULL; 4486 goto failed_mount; 4487 } 4488 es = (struct ext4_super_block *)(bh->b_data + offset); 4489 sbi->s_es = es; 4490 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4491 ext4_msg(sb, KERN_ERR, 4492 "Magic mismatch, very weird!"); 4493 goto failed_mount; 4494 } 4495 } 4496 4497 has_huge_files = ext4_has_feature_huge_file(sb); 4498 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4499 has_huge_files); 4500 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4501 4502 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4503 if (ext4_has_feature_64bit(sb)) { 4504 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4505 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4506 !is_power_of_2(sbi->s_desc_size)) { 4507 ext4_msg(sb, KERN_ERR, 4508 "unsupported descriptor size %lu", 4509 sbi->s_desc_size); 4510 goto failed_mount; 4511 } 4512 } else 4513 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4514 4515 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4516 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4517 4518 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4519 if (sbi->s_inodes_per_block == 0) 4520 goto cantfind_ext4; 4521 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4522 sbi->s_inodes_per_group > blocksize * 8) { 4523 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4524 sbi->s_inodes_per_group); 4525 goto failed_mount; 4526 } 4527 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4528 sbi->s_inodes_per_block; 4529 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4530 sbi->s_sbh = bh; 4531 sbi->s_mount_state = le16_to_cpu(es->s_state); 4532 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4533 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4534 4535 for (i = 0; i < 4; i++) 4536 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4537 sbi->s_def_hash_version = es->s_def_hash_version; 4538 if (ext4_has_feature_dir_index(sb)) { 4539 i = le32_to_cpu(es->s_flags); 4540 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4541 sbi->s_hash_unsigned = 3; 4542 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4543 #ifdef __CHAR_UNSIGNED__ 4544 if (!sb_rdonly(sb)) 4545 es->s_flags |= 4546 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4547 sbi->s_hash_unsigned = 3; 4548 #else 4549 if (!sb_rdonly(sb)) 4550 es->s_flags |= 4551 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4552 #endif 4553 } 4554 } 4555 4556 /* Handle clustersize */ 4557 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4558 if (ext4_has_feature_bigalloc(sb)) { 4559 if (clustersize < blocksize) { 4560 ext4_msg(sb, KERN_ERR, 4561 "cluster size (%d) smaller than " 4562 "block size (%d)", clustersize, blocksize); 4563 goto failed_mount; 4564 } 4565 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4566 le32_to_cpu(es->s_log_block_size); 4567 sbi->s_clusters_per_group = 4568 le32_to_cpu(es->s_clusters_per_group); 4569 if (sbi->s_clusters_per_group > blocksize * 8) { 4570 ext4_msg(sb, KERN_ERR, 4571 "#clusters per group too big: %lu", 4572 sbi->s_clusters_per_group); 4573 goto failed_mount; 4574 } 4575 if (sbi->s_blocks_per_group != 4576 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4577 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4578 "clusters per group (%lu) inconsistent", 4579 sbi->s_blocks_per_group, 4580 sbi->s_clusters_per_group); 4581 goto failed_mount; 4582 } 4583 } else { 4584 if (clustersize != blocksize) { 4585 ext4_msg(sb, KERN_ERR, 4586 "fragment/cluster size (%d) != " 4587 "block size (%d)", clustersize, blocksize); 4588 goto failed_mount; 4589 } 4590 if (sbi->s_blocks_per_group > blocksize * 8) { 4591 ext4_msg(sb, KERN_ERR, 4592 "#blocks per group too big: %lu", 4593 sbi->s_blocks_per_group); 4594 goto failed_mount; 4595 } 4596 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4597 sbi->s_cluster_bits = 0; 4598 } 4599 sbi->s_cluster_ratio = clustersize / blocksize; 4600 4601 /* Do we have standard group size of clustersize * 8 blocks ? */ 4602 if (sbi->s_blocks_per_group == clustersize << 3) 4603 set_opt2(sb, STD_GROUP_SIZE); 4604 4605 /* 4606 * Test whether we have more sectors than will fit in sector_t, 4607 * and whether the max offset is addressable by the page cache. 4608 */ 4609 err = generic_check_addressable(sb->s_blocksize_bits, 4610 ext4_blocks_count(es)); 4611 if (err) { 4612 ext4_msg(sb, KERN_ERR, "filesystem" 4613 " too large to mount safely on this system"); 4614 goto failed_mount; 4615 } 4616 4617 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4618 goto cantfind_ext4; 4619 4620 /* check blocks count against device size */ 4621 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4622 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4623 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4624 "exceeds size of device (%llu blocks)", 4625 ext4_blocks_count(es), blocks_count); 4626 goto failed_mount; 4627 } 4628 4629 /* 4630 * It makes no sense for the first data block to be beyond the end 4631 * of the filesystem. 4632 */ 4633 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4634 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4635 "block %u is beyond end of filesystem (%llu)", 4636 le32_to_cpu(es->s_first_data_block), 4637 ext4_blocks_count(es)); 4638 goto failed_mount; 4639 } 4640 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4641 (sbi->s_cluster_ratio == 1)) { 4642 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4643 "block is 0 with a 1k block and cluster size"); 4644 goto failed_mount; 4645 } 4646 4647 blocks_count = (ext4_blocks_count(es) - 4648 le32_to_cpu(es->s_first_data_block) + 4649 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4650 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4651 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4652 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4653 "(block count %llu, first data block %u, " 4654 "blocks per group %lu)", blocks_count, 4655 ext4_blocks_count(es), 4656 le32_to_cpu(es->s_first_data_block), 4657 EXT4_BLOCKS_PER_GROUP(sb)); 4658 goto failed_mount; 4659 } 4660 sbi->s_groups_count = blocks_count; 4661 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4662 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4663 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4664 le32_to_cpu(es->s_inodes_count)) { 4665 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4666 le32_to_cpu(es->s_inodes_count), 4667 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4668 ret = -EINVAL; 4669 goto failed_mount; 4670 } 4671 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4672 EXT4_DESC_PER_BLOCK(sb); 4673 if (ext4_has_feature_meta_bg(sb)) { 4674 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4675 ext4_msg(sb, KERN_WARNING, 4676 "first meta block group too large: %u " 4677 "(group descriptor block count %u)", 4678 le32_to_cpu(es->s_first_meta_bg), db_count); 4679 goto failed_mount; 4680 } 4681 } 4682 rcu_assign_pointer(sbi->s_group_desc, 4683 kvmalloc_array(db_count, 4684 sizeof(struct buffer_head *), 4685 GFP_KERNEL)); 4686 if (sbi->s_group_desc == NULL) { 4687 ext4_msg(sb, KERN_ERR, "not enough memory"); 4688 ret = -ENOMEM; 4689 goto failed_mount; 4690 } 4691 4692 bgl_lock_init(sbi->s_blockgroup_lock); 4693 4694 /* Pre-read the descriptors into the buffer cache */ 4695 for (i = 0; i < db_count; i++) { 4696 block = descriptor_loc(sb, logical_sb_block, i); 4697 ext4_sb_breadahead_unmovable(sb, block); 4698 } 4699 4700 for (i = 0; i < db_count; i++) { 4701 struct buffer_head *bh; 4702 4703 block = descriptor_loc(sb, logical_sb_block, i); 4704 bh = ext4_sb_bread_unmovable(sb, block); 4705 if (IS_ERR(bh)) { 4706 ext4_msg(sb, KERN_ERR, 4707 "can't read group descriptor %d", i); 4708 db_count = i; 4709 ret = PTR_ERR(bh); 4710 bh = NULL; 4711 goto failed_mount2; 4712 } 4713 rcu_read_lock(); 4714 rcu_dereference(sbi->s_group_desc)[i] = bh; 4715 rcu_read_unlock(); 4716 } 4717 sbi->s_gdb_count = db_count; 4718 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4719 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4720 ret = -EFSCORRUPTED; 4721 goto failed_mount2; 4722 } 4723 4724 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4725 4726 /* Register extent status tree shrinker */ 4727 if (ext4_es_register_shrinker(sbi)) 4728 goto failed_mount3; 4729 4730 sbi->s_stripe = ext4_get_stripe_size(sbi); 4731 sbi->s_extent_max_zeroout_kb = 32; 4732 4733 /* 4734 * set up enough so that it can read an inode 4735 */ 4736 sb->s_op = &ext4_sops; 4737 sb->s_export_op = &ext4_export_ops; 4738 sb->s_xattr = ext4_xattr_handlers; 4739 #ifdef CONFIG_FS_ENCRYPTION 4740 sb->s_cop = &ext4_cryptops; 4741 #endif 4742 #ifdef CONFIG_FS_VERITY 4743 sb->s_vop = &ext4_verityops; 4744 #endif 4745 #ifdef CONFIG_QUOTA 4746 sb->dq_op = &ext4_quota_operations; 4747 if (ext4_has_feature_quota(sb)) 4748 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4749 else 4750 sb->s_qcop = &ext4_qctl_operations; 4751 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4752 #endif 4753 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4754 4755 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4756 mutex_init(&sbi->s_orphan_lock); 4757 4758 /* Initialize fast commit stuff */ 4759 atomic_set(&sbi->s_fc_subtid, 0); 4760 atomic_set(&sbi->s_fc_ineligible_updates, 0); 4761 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4762 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4763 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4764 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4765 sbi->s_fc_bytes = 0; 4766 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4767 ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING); 4768 spin_lock_init(&sbi->s_fc_lock); 4769 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4770 sbi->s_fc_replay_state.fc_regions = NULL; 4771 sbi->s_fc_replay_state.fc_regions_size = 0; 4772 sbi->s_fc_replay_state.fc_regions_used = 0; 4773 sbi->s_fc_replay_state.fc_regions_valid = 0; 4774 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4775 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4776 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4777 4778 sb->s_root = NULL; 4779 4780 needs_recovery = (es->s_last_orphan != 0 || 4781 ext4_has_feature_journal_needs_recovery(sb)); 4782 4783 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4784 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4785 goto failed_mount3a; 4786 4787 /* 4788 * The first inode we look at is the journal inode. Don't try 4789 * root first: it may be modified in the journal! 4790 */ 4791 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4792 err = ext4_load_journal(sb, es, journal_devnum); 4793 if (err) 4794 goto failed_mount3a; 4795 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4796 ext4_has_feature_journal_needs_recovery(sb)) { 4797 ext4_msg(sb, KERN_ERR, "required journal recovery " 4798 "suppressed and not mounted read-only"); 4799 goto failed_mount_wq; 4800 } else { 4801 /* Nojournal mode, all journal mount options are illegal */ 4802 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4803 ext4_msg(sb, KERN_ERR, "can't mount with " 4804 "journal_checksum, fs mounted w/o journal"); 4805 goto failed_mount_wq; 4806 } 4807 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4808 ext4_msg(sb, KERN_ERR, "can't mount with " 4809 "journal_async_commit, fs mounted w/o journal"); 4810 goto failed_mount_wq; 4811 } 4812 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4813 ext4_msg(sb, KERN_ERR, "can't mount with " 4814 "commit=%lu, fs mounted w/o journal", 4815 sbi->s_commit_interval / HZ); 4816 goto failed_mount_wq; 4817 } 4818 if (EXT4_MOUNT_DATA_FLAGS & 4819 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4820 ext4_msg(sb, KERN_ERR, "can't mount with " 4821 "data=, fs mounted w/o journal"); 4822 goto failed_mount_wq; 4823 } 4824 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4825 clear_opt(sb, JOURNAL_CHECKSUM); 4826 clear_opt(sb, DATA_FLAGS); 4827 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4828 sbi->s_journal = NULL; 4829 needs_recovery = 0; 4830 goto no_journal; 4831 } 4832 4833 if (ext4_has_feature_64bit(sb) && 4834 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4835 JBD2_FEATURE_INCOMPAT_64BIT)) { 4836 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4837 goto failed_mount_wq; 4838 } 4839 4840 if (!set_journal_csum_feature_set(sb)) { 4841 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4842 "feature set"); 4843 goto failed_mount_wq; 4844 } 4845 4846 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4847 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4848 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4849 ext4_msg(sb, KERN_ERR, 4850 "Failed to set fast commit journal feature"); 4851 goto failed_mount_wq; 4852 } 4853 4854 /* We have now updated the journal if required, so we can 4855 * validate the data journaling mode. */ 4856 switch (test_opt(sb, DATA_FLAGS)) { 4857 case 0: 4858 /* No mode set, assume a default based on the journal 4859 * capabilities: ORDERED_DATA if the journal can 4860 * cope, else JOURNAL_DATA 4861 */ 4862 if (jbd2_journal_check_available_features 4863 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4864 set_opt(sb, ORDERED_DATA); 4865 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4866 } else { 4867 set_opt(sb, JOURNAL_DATA); 4868 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4869 } 4870 break; 4871 4872 case EXT4_MOUNT_ORDERED_DATA: 4873 case EXT4_MOUNT_WRITEBACK_DATA: 4874 if (!jbd2_journal_check_available_features 4875 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4876 ext4_msg(sb, KERN_ERR, "Journal does not support " 4877 "requested data journaling mode"); 4878 goto failed_mount_wq; 4879 } 4880 default: 4881 break; 4882 } 4883 4884 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4885 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4886 ext4_msg(sb, KERN_ERR, "can't mount with " 4887 "journal_async_commit in data=ordered mode"); 4888 goto failed_mount_wq; 4889 } 4890 4891 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 4892 4893 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback; 4894 sbi->s_journal->j_submit_inode_data_buffers = 4895 ext4_journal_submit_inode_data_buffers; 4896 sbi->s_journal->j_finish_inode_data_buffers = 4897 ext4_journal_finish_inode_data_buffers; 4898 4899 no_journal: 4900 if (!test_opt(sb, NO_MBCACHE)) { 4901 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4902 if (!sbi->s_ea_block_cache) { 4903 ext4_msg(sb, KERN_ERR, 4904 "Failed to create ea_block_cache"); 4905 goto failed_mount_wq; 4906 } 4907 4908 if (ext4_has_feature_ea_inode(sb)) { 4909 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4910 if (!sbi->s_ea_inode_cache) { 4911 ext4_msg(sb, KERN_ERR, 4912 "Failed to create ea_inode_cache"); 4913 goto failed_mount_wq; 4914 } 4915 } 4916 } 4917 4918 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4919 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4920 goto failed_mount_wq; 4921 } 4922 4923 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4924 !ext4_has_feature_encrypt(sb)) { 4925 ext4_set_feature_encrypt(sb); 4926 ext4_commit_super(sb, 1); 4927 } 4928 4929 /* 4930 * Get the # of file system overhead blocks from the 4931 * superblock if present. 4932 */ 4933 if (es->s_overhead_clusters) 4934 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4935 else { 4936 err = ext4_calculate_overhead(sb); 4937 if (err) 4938 goto failed_mount_wq; 4939 } 4940 4941 /* 4942 * The maximum number of concurrent works can be high and 4943 * concurrency isn't really necessary. Limit it to 1. 4944 */ 4945 EXT4_SB(sb)->rsv_conversion_wq = 4946 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4947 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4948 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4949 ret = -ENOMEM; 4950 goto failed_mount4; 4951 } 4952 4953 /* 4954 * The jbd2_journal_load will have done any necessary log recovery, 4955 * so we can safely mount the rest of the filesystem now. 4956 */ 4957 4958 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4959 if (IS_ERR(root)) { 4960 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4961 ret = PTR_ERR(root); 4962 root = NULL; 4963 goto failed_mount4; 4964 } 4965 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4966 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4967 iput(root); 4968 goto failed_mount4; 4969 } 4970 4971 #ifdef CONFIG_UNICODE 4972 if (sb->s_encoding) 4973 sb->s_d_op = &ext4_dentry_ops; 4974 #endif 4975 4976 sb->s_root = d_make_root(root); 4977 if (!sb->s_root) { 4978 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4979 ret = -ENOMEM; 4980 goto failed_mount4; 4981 } 4982 4983 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4984 if (ret == -EROFS) { 4985 sb->s_flags |= SB_RDONLY; 4986 ret = 0; 4987 } else if (ret) 4988 goto failed_mount4a; 4989 4990 ext4_set_resv_clusters(sb); 4991 4992 if (test_opt(sb, BLOCK_VALIDITY)) { 4993 err = ext4_setup_system_zone(sb); 4994 if (err) { 4995 ext4_msg(sb, KERN_ERR, "failed to initialize system " 4996 "zone (%d)", err); 4997 goto failed_mount4a; 4998 } 4999 } 5000 ext4_fc_replay_cleanup(sb); 5001 5002 ext4_ext_init(sb); 5003 err = ext4_mb_init(sb); 5004 if (err) { 5005 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5006 err); 5007 goto failed_mount5; 5008 } 5009 5010 block = ext4_count_free_clusters(sb); 5011 ext4_free_blocks_count_set(sbi->s_es, 5012 EXT4_C2B(sbi, block)); 5013 ext4_superblock_csum_set(sb); 5014 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5015 GFP_KERNEL); 5016 if (!err) { 5017 unsigned long freei = ext4_count_free_inodes(sb); 5018 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5019 ext4_superblock_csum_set(sb); 5020 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5021 GFP_KERNEL); 5022 } 5023 if (!err) 5024 err = percpu_counter_init(&sbi->s_dirs_counter, 5025 ext4_count_dirs(sb), GFP_KERNEL); 5026 if (!err) 5027 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5028 GFP_KERNEL); 5029 if (!err) 5030 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5031 5032 if (err) { 5033 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5034 goto failed_mount6; 5035 } 5036 5037 if (ext4_has_feature_flex_bg(sb)) 5038 if (!ext4_fill_flex_info(sb)) { 5039 ext4_msg(sb, KERN_ERR, 5040 "unable to initialize " 5041 "flex_bg meta info!"); 5042 goto failed_mount6; 5043 } 5044 5045 err = ext4_register_li_request(sb, first_not_zeroed); 5046 if (err) 5047 goto failed_mount6; 5048 5049 err = ext4_register_sysfs(sb); 5050 if (err) 5051 goto failed_mount7; 5052 5053 #ifdef CONFIG_QUOTA 5054 /* Enable quota usage during mount. */ 5055 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5056 err = ext4_enable_quotas(sb); 5057 if (err) 5058 goto failed_mount8; 5059 } 5060 #endif /* CONFIG_QUOTA */ 5061 5062 /* 5063 * Save the original bdev mapping's wb_err value which could be 5064 * used to detect the metadata async write error. 5065 */ 5066 spin_lock_init(&sbi->s_bdev_wb_lock); 5067 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5068 &sbi->s_bdev_wb_err); 5069 sb->s_bdev->bd_super = sb; 5070 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5071 ext4_orphan_cleanup(sb, es); 5072 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5073 if (needs_recovery) { 5074 ext4_msg(sb, KERN_INFO, "recovery complete"); 5075 err = ext4_mark_recovery_complete(sb, es); 5076 if (err) 5077 goto failed_mount8; 5078 } 5079 if (EXT4_SB(sb)->s_journal) { 5080 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5081 descr = " journalled data mode"; 5082 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5083 descr = " ordered data mode"; 5084 else 5085 descr = " writeback data mode"; 5086 } else 5087 descr = "out journal"; 5088 5089 if (test_opt(sb, DISCARD)) { 5090 struct request_queue *q = bdev_get_queue(sb->s_bdev); 5091 if (!blk_queue_discard(q)) 5092 ext4_msg(sb, KERN_WARNING, 5093 "mounting with \"discard\" option, but " 5094 "the device does not support discard"); 5095 } 5096 5097 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5098 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 5099 "Opts: %.*s%s%s", descr, 5100 (int) sizeof(sbi->s_es->s_mount_opts), 5101 sbi->s_es->s_mount_opts, 5102 *sbi->s_es->s_mount_opts ? "; " : "", orig_data); 5103 5104 if (es->s_error_count) 5105 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5106 5107 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5108 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5109 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5110 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5111 atomic_set(&sbi->s_warning_count, 0); 5112 atomic_set(&sbi->s_msg_count, 0); 5113 5114 kfree(orig_data); 5115 return 0; 5116 5117 cantfind_ext4: 5118 if (!silent) 5119 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5120 goto failed_mount; 5121 5122 failed_mount8: 5123 ext4_unregister_sysfs(sb); 5124 kobject_put(&sbi->s_kobj); 5125 failed_mount7: 5126 ext4_unregister_li_request(sb); 5127 failed_mount6: 5128 ext4_mb_release(sb); 5129 rcu_read_lock(); 5130 flex_groups = rcu_dereference(sbi->s_flex_groups); 5131 if (flex_groups) { 5132 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 5133 kvfree(flex_groups[i]); 5134 kvfree(flex_groups); 5135 } 5136 rcu_read_unlock(); 5137 percpu_counter_destroy(&sbi->s_freeclusters_counter); 5138 percpu_counter_destroy(&sbi->s_freeinodes_counter); 5139 percpu_counter_destroy(&sbi->s_dirs_counter); 5140 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 5141 percpu_free_rwsem(&sbi->s_writepages_rwsem); 5142 failed_mount5: 5143 ext4_ext_release(sb); 5144 ext4_release_system_zone(sb); 5145 failed_mount4a: 5146 dput(sb->s_root); 5147 sb->s_root = NULL; 5148 failed_mount4: 5149 ext4_msg(sb, KERN_ERR, "mount failed"); 5150 if (EXT4_SB(sb)->rsv_conversion_wq) 5151 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5152 failed_mount_wq: 5153 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5154 sbi->s_ea_inode_cache = NULL; 5155 5156 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5157 sbi->s_ea_block_cache = NULL; 5158 5159 if (sbi->s_journal) { 5160 jbd2_journal_destroy(sbi->s_journal); 5161 sbi->s_journal = NULL; 5162 } 5163 failed_mount3a: 5164 ext4_es_unregister_shrinker(sbi); 5165 failed_mount3: 5166 del_timer_sync(&sbi->s_err_report); 5167 if (sbi->s_mmp_tsk) 5168 kthread_stop(sbi->s_mmp_tsk); 5169 failed_mount2: 5170 rcu_read_lock(); 5171 group_desc = rcu_dereference(sbi->s_group_desc); 5172 for (i = 0; i < db_count; i++) 5173 brelse(group_desc[i]); 5174 kvfree(group_desc); 5175 rcu_read_unlock(); 5176 failed_mount: 5177 if (sbi->s_chksum_driver) 5178 crypto_free_shash(sbi->s_chksum_driver); 5179 5180 #ifdef CONFIG_UNICODE 5181 utf8_unload(sb->s_encoding); 5182 #endif 5183 5184 #ifdef CONFIG_QUOTA 5185 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5186 kfree(get_qf_name(sb, sbi, i)); 5187 #endif 5188 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5189 ext4_blkdev_remove(sbi); 5190 brelse(bh); 5191 out_fail: 5192 sb->s_fs_info = NULL; 5193 kfree(sbi->s_blockgroup_lock); 5194 out_free_base: 5195 kfree(sbi); 5196 kfree(orig_data); 5197 fs_put_dax(dax_dev); 5198 return err ? err : ret; 5199 } 5200 5201 /* 5202 * Setup any per-fs journal parameters now. We'll do this both on 5203 * initial mount, once the journal has been initialised but before we've 5204 * done any recovery; and again on any subsequent remount. 5205 */ 5206 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5207 { 5208 struct ext4_sb_info *sbi = EXT4_SB(sb); 5209 5210 journal->j_commit_interval = sbi->s_commit_interval; 5211 journal->j_min_batch_time = sbi->s_min_batch_time; 5212 journal->j_max_batch_time = sbi->s_max_batch_time; 5213 ext4_fc_init(sb, journal); 5214 5215 write_lock(&journal->j_state_lock); 5216 if (test_opt(sb, BARRIER)) 5217 journal->j_flags |= JBD2_BARRIER; 5218 else 5219 journal->j_flags &= ~JBD2_BARRIER; 5220 if (test_opt(sb, DATA_ERR_ABORT)) 5221 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5222 else 5223 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5224 write_unlock(&journal->j_state_lock); 5225 } 5226 5227 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5228 unsigned int journal_inum) 5229 { 5230 struct inode *journal_inode; 5231 5232 /* 5233 * Test for the existence of a valid inode on disk. Bad things 5234 * happen if we iget() an unused inode, as the subsequent iput() 5235 * will try to delete it. 5236 */ 5237 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5238 if (IS_ERR(journal_inode)) { 5239 ext4_msg(sb, KERN_ERR, "no journal found"); 5240 return NULL; 5241 } 5242 if (!journal_inode->i_nlink) { 5243 make_bad_inode(journal_inode); 5244 iput(journal_inode); 5245 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5246 return NULL; 5247 } 5248 5249 jbd_debug(2, "Journal inode found at %p: %lld bytes\n", 5250 journal_inode, journal_inode->i_size); 5251 if (!S_ISREG(journal_inode->i_mode)) { 5252 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5253 iput(journal_inode); 5254 return NULL; 5255 } 5256 return journal_inode; 5257 } 5258 5259 static journal_t *ext4_get_journal(struct super_block *sb, 5260 unsigned int journal_inum) 5261 { 5262 struct inode *journal_inode; 5263 journal_t *journal; 5264 5265 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5266 return NULL; 5267 5268 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5269 if (!journal_inode) 5270 return NULL; 5271 5272 journal = jbd2_journal_init_inode(journal_inode); 5273 if (!journal) { 5274 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5275 iput(journal_inode); 5276 return NULL; 5277 } 5278 journal->j_private = sb; 5279 ext4_init_journal_params(sb, journal); 5280 return journal; 5281 } 5282 5283 static journal_t *ext4_get_dev_journal(struct super_block *sb, 5284 dev_t j_dev) 5285 { 5286 struct buffer_head *bh; 5287 journal_t *journal; 5288 ext4_fsblk_t start; 5289 ext4_fsblk_t len; 5290 int hblock, blocksize; 5291 ext4_fsblk_t sb_block; 5292 unsigned long offset; 5293 struct ext4_super_block *es; 5294 struct block_device *bdev; 5295 5296 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5297 return NULL; 5298 5299 bdev = ext4_blkdev_get(j_dev, sb); 5300 if (bdev == NULL) 5301 return NULL; 5302 5303 blocksize = sb->s_blocksize; 5304 hblock = bdev_logical_block_size(bdev); 5305 if (blocksize < hblock) { 5306 ext4_msg(sb, KERN_ERR, 5307 "blocksize too small for journal device"); 5308 goto out_bdev; 5309 } 5310 5311 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5312 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5313 set_blocksize(bdev, blocksize); 5314 if (!(bh = __bread(bdev, sb_block, blocksize))) { 5315 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5316 "external journal"); 5317 goto out_bdev; 5318 } 5319 5320 es = (struct ext4_super_block *) (bh->b_data + offset); 5321 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5322 !(le32_to_cpu(es->s_feature_incompat) & 5323 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5324 ext4_msg(sb, KERN_ERR, "external journal has " 5325 "bad superblock"); 5326 brelse(bh); 5327 goto out_bdev; 5328 } 5329 5330 if ((le32_to_cpu(es->s_feature_ro_compat) & 5331 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5332 es->s_checksum != ext4_superblock_csum(sb, es)) { 5333 ext4_msg(sb, KERN_ERR, "external journal has " 5334 "corrupt superblock"); 5335 brelse(bh); 5336 goto out_bdev; 5337 } 5338 5339 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5340 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5341 brelse(bh); 5342 goto out_bdev; 5343 } 5344 5345 len = ext4_blocks_count(es); 5346 start = sb_block + 1; 5347 brelse(bh); /* we're done with the superblock */ 5348 5349 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 5350 start, len, blocksize); 5351 if (!journal) { 5352 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5353 goto out_bdev; 5354 } 5355 journal->j_private = sb; 5356 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) { 5357 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 5358 goto out_journal; 5359 } 5360 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5361 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5362 "user (unsupported) - %d", 5363 be32_to_cpu(journal->j_superblock->s_nr_users)); 5364 goto out_journal; 5365 } 5366 EXT4_SB(sb)->s_journal_bdev = bdev; 5367 ext4_init_journal_params(sb, journal); 5368 return journal; 5369 5370 out_journal: 5371 jbd2_journal_destroy(journal); 5372 out_bdev: 5373 ext4_blkdev_put(bdev); 5374 return NULL; 5375 } 5376 5377 static int ext4_load_journal(struct super_block *sb, 5378 struct ext4_super_block *es, 5379 unsigned long journal_devnum) 5380 { 5381 journal_t *journal; 5382 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5383 dev_t journal_dev; 5384 int err = 0; 5385 int really_read_only; 5386 int journal_dev_ro; 5387 5388 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5389 return -EFSCORRUPTED; 5390 5391 if (journal_devnum && 5392 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5393 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5394 "numbers have changed"); 5395 journal_dev = new_decode_dev(journal_devnum); 5396 } else 5397 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5398 5399 if (journal_inum && journal_dev) { 5400 ext4_msg(sb, KERN_ERR, 5401 "filesystem has both journal inode and journal device!"); 5402 return -EINVAL; 5403 } 5404 5405 if (journal_inum) { 5406 journal = ext4_get_journal(sb, journal_inum); 5407 if (!journal) 5408 return -EINVAL; 5409 } else { 5410 journal = ext4_get_dev_journal(sb, journal_dev); 5411 if (!journal) 5412 return -EINVAL; 5413 } 5414 5415 journal_dev_ro = bdev_read_only(journal->j_dev); 5416 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5417 5418 if (journal_dev_ro && !sb_rdonly(sb)) { 5419 ext4_msg(sb, KERN_ERR, 5420 "journal device read-only, try mounting with '-o ro'"); 5421 err = -EROFS; 5422 goto err_out; 5423 } 5424 5425 /* 5426 * Are we loading a blank journal or performing recovery after a 5427 * crash? For recovery, we need to check in advance whether we 5428 * can get read-write access to the device. 5429 */ 5430 if (ext4_has_feature_journal_needs_recovery(sb)) { 5431 if (sb_rdonly(sb)) { 5432 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5433 "required on readonly filesystem"); 5434 if (really_read_only) { 5435 ext4_msg(sb, KERN_ERR, "write access " 5436 "unavailable, cannot proceed " 5437 "(try mounting with noload)"); 5438 err = -EROFS; 5439 goto err_out; 5440 } 5441 ext4_msg(sb, KERN_INFO, "write access will " 5442 "be enabled during recovery"); 5443 } 5444 } 5445 5446 if (!(journal->j_flags & JBD2_BARRIER)) 5447 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5448 5449 if (!ext4_has_feature_journal_needs_recovery(sb)) 5450 err = jbd2_journal_wipe(journal, !really_read_only); 5451 if (!err) { 5452 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5453 if (save) 5454 memcpy(save, ((char *) es) + 5455 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 5456 err = jbd2_journal_load(journal); 5457 if (save) 5458 memcpy(((char *) es) + EXT4_S_ERR_START, 5459 save, EXT4_S_ERR_LEN); 5460 kfree(save); 5461 } 5462 5463 if (err) { 5464 ext4_msg(sb, KERN_ERR, "error loading journal"); 5465 goto err_out; 5466 } 5467 5468 EXT4_SB(sb)->s_journal = journal; 5469 err = ext4_clear_journal_err(sb, es); 5470 if (err) { 5471 EXT4_SB(sb)->s_journal = NULL; 5472 jbd2_journal_destroy(journal); 5473 return err; 5474 } 5475 5476 if (!really_read_only && journal_devnum && 5477 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5478 es->s_journal_dev = cpu_to_le32(journal_devnum); 5479 5480 /* Make sure we flush the recovery flag to disk. */ 5481 ext4_commit_super(sb, 1); 5482 } 5483 5484 return 0; 5485 5486 err_out: 5487 jbd2_journal_destroy(journal); 5488 return err; 5489 } 5490 5491 static int ext4_commit_super(struct super_block *sb, int sync) 5492 { 5493 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 5494 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 5495 int error = 0; 5496 5497 if (!sbh || block_device_ejected(sb)) 5498 return error; 5499 5500 /* 5501 * If the file system is mounted read-only, don't update the 5502 * superblock write time. This avoids updating the superblock 5503 * write time when we are mounting the root file system 5504 * read/only but we need to replay the journal; at that point, 5505 * for people who are east of GMT and who make their clock 5506 * tick in localtime for Windows bug-for-bug compatibility, 5507 * the clock is set in the future, and this will cause e2fsck 5508 * to complain and force a full file system check. 5509 */ 5510 if (!(sb->s_flags & SB_RDONLY)) 5511 ext4_update_tstamp(es, s_wtime); 5512 if (sb->s_bdev->bd_part) 5513 es->s_kbytes_written = 5514 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 5515 ((part_stat_read(sb->s_bdev->bd_part, 5516 sectors[STAT_WRITE]) - 5517 EXT4_SB(sb)->s_sectors_written_start) >> 1)); 5518 else 5519 es->s_kbytes_written = 5520 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written); 5521 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter)) 5522 ext4_free_blocks_count_set(es, 5523 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive( 5524 &EXT4_SB(sb)->s_freeclusters_counter))); 5525 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter)) 5526 es->s_free_inodes_count = 5527 cpu_to_le32(percpu_counter_sum_positive( 5528 &EXT4_SB(sb)->s_freeinodes_counter)); 5529 BUFFER_TRACE(sbh, "marking dirty"); 5530 ext4_superblock_csum_set(sb); 5531 if (sync) 5532 lock_buffer(sbh); 5533 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 5534 /* 5535 * Oh, dear. A previous attempt to write the 5536 * superblock failed. This could happen because the 5537 * USB device was yanked out. Or it could happen to 5538 * be a transient write error and maybe the block will 5539 * be remapped. Nothing we can do but to retry the 5540 * write and hope for the best. 5541 */ 5542 ext4_msg(sb, KERN_ERR, "previous I/O error to " 5543 "superblock detected"); 5544 clear_buffer_write_io_error(sbh); 5545 set_buffer_uptodate(sbh); 5546 } 5547 mark_buffer_dirty(sbh); 5548 if (sync) { 5549 unlock_buffer(sbh); 5550 error = __sync_dirty_buffer(sbh, 5551 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0)); 5552 if (buffer_write_io_error(sbh)) { 5553 ext4_msg(sb, KERN_ERR, "I/O error while writing " 5554 "superblock"); 5555 clear_buffer_write_io_error(sbh); 5556 set_buffer_uptodate(sbh); 5557 } 5558 } 5559 return error; 5560 } 5561 5562 /* 5563 * Have we just finished recovery? If so, and if we are mounting (or 5564 * remounting) the filesystem readonly, then we will end up with a 5565 * consistent fs on disk. Record that fact. 5566 */ 5567 static int ext4_mark_recovery_complete(struct super_block *sb, 5568 struct ext4_super_block *es) 5569 { 5570 int err; 5571 journal_t *journal = EXT4_SB(sb)->s_journal; 5572 5573 if (!ext4_has_feature_journal(sb)) { 5574 if (journal != NULL) { 5575 ext4_error(sb, "Journal got removed while the fs was " 5576 "mounted!"); 5577 return -EFSCORRUPTED; 5578 } 5579 return 0; 5580 } 5581 jbd2_journal_lock_updates(journal); 5582 err = jbd2_journal_flush(journal); 5583 if (err < 0) 5584 goto out; 5585 5586 if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) { 5587 ext4_clear_feature_journal_needs_recovery(sb); 5588 ext4_commit_super(sb, 1); 5589 } 5590 out: 5591 jbd2_journal_unlock_updates(journal); 5592 return err; 5593 } 5594 5595 /* 5596 * If we are mounting (or read-write remounting) a filesystem whose journal 5597 * has recorded an error from a previous lifetime, move that error to the 5598 * main filesystem now. 5599 */ 5600 static int ext4_clear_journal_err(struct super_block *sb, 5601 struct ext4_super_block *es) 5602 { 5603 journal_t *journal; 5604 int j_errno; 5605 const char *errstr; 5606 5607 if (!ext4_has_feature_journal(sb)) { 5608 ext4_error(sb, "Journal got removed while the fs was mounted!"); 5609 return -EFSCORRUPTED; 5610 } 5611 5612 journal = EXT4_SB(sb)->s_journal; 5613 5614 /* 5615 * Now check for any error status which may have been recorded in the 5616 * journal by a prior ext4_error() or ext4_abort() 5617 */ 5618 5619 j_errno = jbd2_journal_errno(journal); 5620 if (j_errno) { 5621 char nbuf[16]; 5622 5623 errstr = ext4_decode_error(sb, j_errno, nbuf); 5624 ext4_warning(sb, "Filesystem error recorded " 5625 "from previous mount: %s", errstr); 5626 ext4_warning(sb, "Marking fs in need of filesystem check."); 5627 5628 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 5629 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5630 ext4_commit_super(sb, 1); 5631 5632 jbd2_journal_clear_err(journal); 5633 jbd2_journal_update_sb_errno(journal); 5634 } 5635 return 0; 5636 } 5637 5638 /* 5639 * Force the running and committing transactions to commit, 5640 * and wait on the commit. 5641 */ 5642 int ext4_force_commit(struct super_block *sb) 5643 { 5644 journal_t *journal; 5645 5646 if (sb_rdonly(sb)) 5647 return 0; 5648 5649 journal = EXT4_SB(sb)->s_journal; 5650 return ext4_journal_force_commit(journal); 5651 } 5652 5653 static int ext4_sync_fs(struct super_block *sb, int wait) 5654 { 5655 int ret = 0; 5656 tid_t target; 5657 bool needs_barrier = false; 5658 struct ext4_sb_info *sbi = EXT4_SB(sb); 5659 5660 if (unlikely(ext4_forced_shutdown(sbi))) 5661 return 0; 5662 5663 trace_ext4_sync_fs(sb, wait); 5664 flush_workqueue(sbi->rsv_conversion_wq); 5665 /* 5666 * Writeback quota in non-journalled quota case - journalled quota has 5667 * no dirty dquots 5668 */ 5669 dquot_writeback_dquots(sb, -1); 5670 /* 5671 * Data writeback is possible w/o journal transaction, so barrier must 5672 * being sent at the end of the function. But we can skip it if 5673 * transaction_commit will do it for us. 5674 */ 5675 if (sbi->s_journal) { 5676 target = jbd2_get_latest_transaction(sbi->s_journal); 5677 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 5678 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 5679 needs_barrier = true; 5680 5681 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 5682 if (wait) 5683 ret = jbd2_log_wait_commit(sbi->s_journal, 5684 target); 5685 } 5686 } else if (wait && test_opt(sb, BARRIER)) 5687 needs_barrier = true; 5688 if (needs_barrier) { 5689 int err; 5690 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL); 5691 if (!ret) 5692 ret = err; 5693 } 5694 5695 return ret; 5696 } 5697 5698 /* 5699 * LVM calls this function before a (read-only) snapshot is created. This 5700 * gives us a chance to flush the journal completely and mark the fs clean. 5701 * 5702 * Note that only this function cannot bring a filesystem to be in a clean 5703 * state independently. It relies on upper layer to stop all data & metadata 5704 * modifications. 5705 */ 5706 static int ext4_freeze(struct super_block *sb) 5707 { 5708 int error = 0; 5709 journal_t *journal; 5710 5711 if (sb_rdonly(sb)) 5712 return 0; 5713 5714 journal = EXT4_SB(sb)->s_journal; 5715 5716 if (journal) { 5717 /* Now we set up the journal barrier. */ 5718 jbd2_journal_lock_updates(journal); 5719 5720 /* 5721 * Don't clear the needs_recovery flag if we failed to 5722 * flush the journal. 5723 */ 5724 error = jbd2_journal_flush(journal); 5725 if (error < 0) 5726 goto out; 5727 5728 /* Journal blocked and flushed, clear needs_recovery flag. */ 5729 ext4_clear_feature_journal_needs_recovery(sb); 5730 } 5731 5732 error = ext4_commit_super(sb, 1); 5733 out: 5734 if (journal) 5735 /* we rely on upper layer to stop further updates */ 5736 jbd2_journal_unlock_updates(journal); 5737 return error; 5738 } 5739 5740 /* 5741 * Called by LVM after the snapshot is done. We need to reset the RECOVER 5742 * flag here, even though the filesystem is not technically dirty yet. 5743 */ 5744 static int ext4_unfreeze(struct super_block *sb) 5745 { 5746 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 5747 return 0; 5748 5749 if (EXT4_SB(sb)->s_journal) { 5750 /* Reset the needs_recovery flag before the fs is unlocked. */ 5751 ext4_set_feature_journal_needs_recovery(sb); 5752 } 5753 5754 ext4_commit_super(sb, 1); 5755 return 0; 5756 } 5757 5758 /* 5759 * Structure to save mount options for ext4_remount's benefit 5760 */ 5761 struct ext4_mount_options { 5762 unsigned long s_mount_opt; 5763 unsigned long s_mount_opt2; 5764 kuid_t s_resuid; 5765 kgid_t s_resgid; 5766 unsigned long s_commit_interval; 5767 u32 s_min_batch_time, s_max_batch_time; 5768 #ifdef CONFIG_QUOTA 5769 int s_jquota_fmt; 5770 char *s_qf_names[EXT4_MAXQUOTAS]; 5771 #endif 5772 }; 5773 5774 static int ext4_remount(struct super_block *sb, int *flags, char *data) 5775 { 5776 struct ext4_super_block *es; 5777 struct ext4_sb_info *sbi = EXT4_SB(sb); 5778 unsigned long old_sb_flags, vfs_flags; 5779 struct ext4_mount_options old_opts; 5780 int enable_quota = 0; 5781 ext4_group_t g; 5782 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5783 int err = 0; 5784 #ifdef CONFIG_QUOTA 5785 int i, j; 5786 char *to_free[EXT4_MAXQUOTAS]; 5787 #endif 5788 char *orig_data = kstrdup(data, GFP_KERNEL); 5789 5790 if (data && !orig_data) 5791 return -ENOMEM; 5792 5793 /* Store the original options */ 5794 old_sb_flags = sb->s_flags; 5795 old_opts.s_mount_opt = sbi->s_mount_opt; 5796 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 5797 old_opts.s_resuid = sbi->s_resuid; 5798 old_opts.s_resgid = sbi->s_resgid; 5799 old_opts.s_commit_interval = sbi->s_commit_interval; 5800 old_opts.s_min_batch_time = sbi->s_min_batch_time; 5801 old_opts.s_max_batch_time = sbi->s_max_batch_time; 5802 #ifdef CONFIG_QUOTA 5803 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 5804 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5805 if (sbi->s_qf_names[i]) { 5806 char *qf_name = get_qf_name(sb, sbi, i); 5807 5808 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 5809 if (!old_opts.s_qf_names[i]) { 5810 for (j = 0; j < i; j++) 5811 kfree(old_opts.s_qf_names[j]); 5812 kfree(orig_data); 5813 return -ENOMEM; 5814 } 5815 } else 5816 old_opts.s_qf_names[i] = NULL; 5817 #endif 5818 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 5819 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio; 5820 5821 /* 5822 * Some options can be enabled by ext4 and/or by VFS mount flag 5823 * either way we need to make sure it matches in both *flags and 5824 * s_flags. Copy those selected flags from *flags to s_flags 5825 */ 5826 vfs_flags = SB_LAZYTIME | SB_I_VERSION; 5827 sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags); 5828 5829 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) { 5830 err = -EINVAL; 5831 goto restore_opts; 5832 } 5833 5834 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 5835 test_opt(sb, JOURNAL_CHECKSUM)) { 5836 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 5837 "during remount not supported; ignoring"); 5838 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 5839 } 5840 5841 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5842 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5843 ext4_msg(sb, KERN_ERR, "can't mount with " 5844 "both data=journal and delalloc"); 5845 err = -EINVAL; 5846 goto restore_opts; 5847 } 5848 if (test_opt(sb, DIOREAD_NOLOCK)) { 5849 ext4_msg(sb, KERN_ERR, "can't mount with " 5850 "both data=journal and dioread_nolock"); 5851 err = -EINVAL; 5852 goto restore_opts; 5853 } 5854 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 5855 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5856 ext4_msg(sb, KERN_ERR, "can't mount with " 5857 "journal_async_commit in data=ordered mode"); 5858 err = -EINVAL; 5859 goto restore_opts; 5860 } 5861 } 5862 5863 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 5864 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 5865 err = -EINVAL; 5866 goto restore_opts; 5867 } 5868 5869 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) 5870 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user"); 5871 5872 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5873 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5874 5875 es = sbi->s_es; 5876 5877 if (sbi->s_journal) { 5878 ext4_init_journal_params(sb, sbi->s_journal); 5879 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 5880 } 5881 5882 if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) { 5883 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) { 5884 err = -EROFS; 5885 goto restore_opts; 5886 } 5887 5888 if (*flags & SB_RDONLY) { 5889 err = sync_filesystem(sb); 5890 if (err < 0) 5891 goto restore_opts; 5892 err = dquot_suspend(sb, -1); 5893 if (err < 0) 5894 goto restore_opts; 5895 5896 /* 5897 * First of all, the unconditional stuff we have to do 5898 * to disable replay of the journal when we next remount 5899 */ 5900 sb->s_flags |= SB_RDONLY; 5901 5902 /* 5903 * OK, test if we are remounting a valid rw partition 5904 * readonly, and if so set the rdonly flag and then 5905 * mark the partition as valid again. 5906 */ 5907 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 5908 (sbi->s_mount_state & EXT4_VALID_FS)) 5909 es->s_state = cpu_to_le16(sbi->s_mount_state); 5910 5911 if (sbi->s_journal) { 5912 /* 5913 * We let remount-ro finish even if marking fs 5914 * as clean failed... 5915 */ 5916 ext4_mark_recovery_complete(sb, es); 5917 } 5918 if (sbi->s_mmp_tsk) 5919 kthread_stop(sbi->s_mmp_tsk); 5920 } else { 5921 /* Make sure we can mount this feature set readwrite */ 5922 if (ext4_has_feature_readonly(sb) || 5923 !ext4_feature_set_ok(sb, 0)) { 5924 err = -EROFS; 5925 goto restore_opts; 5926 } 5927 /* 5928 * Make sure the group descriptor checksums 5929 * are sane. If they aren't, refuse to remount r/w. 5930 */ 5931 for (g = 0; g < sbi->s_groups_count; g++) { 5932 struct ext4_group_desc *gdp = 5933 ext4_get_group_desc(sb, g, NULL); 5934 5935 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 5936 ext4_msg(sb, KERN_ERR, 5937 "ext4_remount: Checksum for group %u failed (%u!=%u)", 5938 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 5939 le16_to_cpu(gdp->bg_checksum)); 5940 err = -EFSBADCRC; 5941 goto restore_opts; 5942 } 5943 } 5944 5945 /* 5946 * If we have an unprocessed orphan list hanging 5947 * around from a previously readonly bdev mount, 5948 * require a full umount/remount for now. 5949 */ 5950 if (es->s_last_orphan) { 5951 ext4_msg(sb, KERN_WARNING, "Couldn't " 5952 "remount RDWR because of unprocessed " 5953 "orphan inode list. Please " 5954 "umount/remount instead"); 5955 err = -EINVAL; 5956 goto restore_opts; 5957 } 5958 5959 /* 5960 * Mounting a RDONLY partition read-write, so reread 5961 * and store the current valid flag. (It may have 5962 * been changed by e2fsck since we originally mounted 5963 * the partition.) 5964 */ 5965 if (sbi->s_journal) { 5966 err = ext4_clear_journal_err(sb, es); 5967 if (err) 5968 goto restore_opts; 5969 } 5970 sbi->s_mount_state = le16_to_cpu(es->s_state); 5971 5972 err = ext4_setup_super(sb, es, 0); 5973 if (err) 5974 goto restore_opts; 5975 5976 sb->s_flags &= ~SB_RDONLY; 5977 if (ext4_has_feature_mmp(sb)) 5978 if (ext4_multi_mount_protect(sb, 5979 le64_to_cpu(es->s_mmp_block))) { 5980 err = -EROFS; 5981 goto restore_opts; 5982 } 5983 enable_quota = 1; 5984 } 5985 } 5986 5987 /* 5988 * Reinitialize lazy itable initialization thread based on 5989 * current settings 5990 */ 5991 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 5992 ext4_unregister_li_request(sb); 5993 else { 5994 ext4_group_t first_not_zeroed; 5995 first_not_zeroed = ext4_has_uninit_itable(sb); 5996 ext4_register_li_request(sb, first_not_zeroed); 5997 } 5998 5999 /* 6000 * Handle creation of system zone data early because it can fail. 6001 * Releasing of existing data is done when we are sure remount will 6002 * succeed. 6003 */ 6004 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6005 err = ext4_setup_system_zone(sb); 6006 if (err) 6007 goto restore_opts; 6008 } 6009 6010 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6011 err = ext4_commit_super(sb, 1); 6012 if (err) 6013 goto restore_opts; 6014 } 6015 6016 #ifdef CONFIG_QUOTA 6017 /* Release old quota file names */ 6018 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6019 kfree(old_opts.s_qf_names[i]); 6020 if (enable_quota) { 6021 if (sb_any_quota_suspended(sb)) 6022 dquot_resume(sb, -1); 6023 else if (ext4_has_feature_quota(sb)) { 6024 err = ext4_enable_quotas(sb); 6025 if (err) 6026 goto restore_opts; 6027 } 6028 } 6029 #endif 6030 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6031 ext4_release_system_zone(sb); 6032 6033 /* 6034 * Some options can be enabled by ext4 and/or by VFS mount flag 6035 * either way we need to make sure it matches in both *flags and 6036 * s_flags. Copy those selected flags from s_flags to *flags 6037 */ 6038 *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags); 6039 6040 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data); 6041 kfree(orig_data); 6042 return 0; 6043 6044 restore_opts: 6045 sb->s_flags = old_sb_flags; 6046 sbi->s_mount_opt = old_opts.s_mount_opt; 6047 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6048 sbi->s_resuid = old_opts.s_resuid; 6049 sbi->s_resgid = old_opts.s_resgid; 6050 sbi->s_commit_interval = old_opts.s_commit_interval; 6051 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6052 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6053 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6054 ext4_release_system_zone(sb); 6055 #ifdef CONFIG_QUOTA 6056 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6057 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6058 to_free[i] = get_qf_name(sb, sbi, i); 6059 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6060 } 6061 synchronize_rcu(); 6062 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6063 kfree(to_free[i]); 6064 #endif 6065 kfree(orig_data); 6066 return err; 6067 } 6068 6069 #ifdef CONFIG_QUOTA 6070 static int ext4_statfs_project(struct super_block *sb, 6071 kprojid_t projid, struct kstatfs *buf) 6072 { 6073 struct kqid qid; 6074 struct dquot *dquot; 6075 u64 limit; 6076 u64 curblock; 6077 6078 qid = make_kqid_projid(projid); 6079 dquot = dqget(sb, qid); 6080 if (IS_ERR(dquot)) 6081 return PTR_ERR(dquot); 6082 spin_lock(&dquot->dq_dqb_lock); 6083 6084 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6085 dquot->dq_dqb.dqb_bhardlimit); 6086 limit >>= sb->s_blocksize_bits; 6087 6088 if (limit && buf->f_blocks > limit) { 6089 curblock = (dquot->dq_dqb.dqb_curspace + 6090 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6091 buf->f_blocks = limit; 6092 buf->f_bfree = buf->f_bavail = 6093 (buf->f_blocks > curblock) ? 6094 (buf->f_blocks - curblock) : 0; 6095 } 6096 6097 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6098 dquot->dq_dqb.dqb_ihardlimit); 6099 if (limit && buf->f_files > limit) { 6100 buf->f_files = limit; 6101 buf->f_ffree = 6102 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6103 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6104 } 6105 6106 spin_unlock(&dquot->dq_dqb_lock); 6107 dqput(dquot); 6108 return 0; 6109 } 6110 #endif 6111 6112 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6113 { 6114 struct super_block *sb = dentry->d_sb; 6115 struct ext4_sb_info *sbi = EXT4_SB(sb); 6116 struct ext4_super_block *es = sbi->s_es; 6117 ext4_fsblk_t overhead = 0, resv_blocks; 6118 u64 fsid; 6119 s64 bfree; 6120 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6121 6122 if (!test_opt(sb, MINIX_DF)) 6123 overhead = sbi->s_overhead; 6124 6125 buf->f_type = EXT4_SUPER_MAGIC; 6126 buf->f_bsize = sb->s_blocksize; 6127 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6128 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6129 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6130 /* prevent underflow in case that few free space is available */ 6131 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6132 buf->f_bavail = buf->f_bfree - 6133 (ext4_r_blocks_count(es) + resv_blocks); 6134 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6135 buf->f_bavail = 0; 6136 buf->f_files = le32_to_cpu(es->s_inodes_count); 6137 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6138 buf->f_namelen = EXT4_NAME_LEN; 6139 fsid = le64_to_cpup((void *)es->s_uuid) ^ 6140 le64_to_cpup((void *)es->s_uuid + sizeof(u64)); 6141 buf->f_fsid = u64_to_fsid(fsid); 6142 6143 #ifdef CONFIG_QUOTA 6144 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6145 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6146 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6147 #endif 6148 return 0; 6149 } 6150 6151 6152 #ifdef CONFIG_QUOTA 6153 6154 /* 6155 * Helper functions so that transaction is started before we acquire dqio_sem 6156 * to keep correct lock ordering of transaction > dqio_sem 6157 */ 6158 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6159 { 6160 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6161 } 6162 6163 static int ext4_write_dquot(struct dquot *dquot) 6164 { 6165 int ret, err; 6166 handle_t *handle; 6167 struct inode *inode; 6168 6169 inode = dquot_to_inode(dquot); 6170 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6171 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6172 if (IS_ERR(handle)) 6173 return PTR_ERR(handle); 6174 ret = dquot_commit(dquot); 6175 err = ext4_journal_stop(handle); 6176 if (!ret) 6177 ret = err; 6178 return ret; 6179 } 6180 6181 static int ext4_acquire_dquot(struct dquot *dquot) 6182 { 6183 int ret, err; 6184 handle_t *handle; 6185 6186 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6187 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6188 if (IS_ERR(handle)) 6189 return PTR_ERR(handle); 6190 ret = dquot_acquire(dquot); 6191 err = ext4_journal_stop(handle); 6192 if (!ret) 6193 ret = err; 6194 return ret; 6195 } 6196 6197 static int ext4_release_dquot(struct dquot *dquot) 6198 { 6199 int ret, err; 6200 handle_t *handle; 6201 6202 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6203 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6204 if (IS_ERR(handle)) { 6205 /* Release dquot anyway to avoid endless cycle in dqput() */ 6206 dquot_release(dquot); 6207 return PTR_ERR(handle); 6208 } 6209 ret = dquot_release(dquot); 6210 err = ext4_journal_stop(handle); 6211 if (!ret) 6212 ret = err; 6213 return ret; 6214 } 6215 6216 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6217 { 6218 struct super_block *sb = dquot->dq_sb; 6219 struct ext4_sb_info *sbi = EXT4_SB(sb); 6220 6221 /* Are we journaling quotas? */ 6222 if (ext4_has_feature_quota(sb) || 6223 sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) { 6224 dquot_mark_dquot_dirty(dquot); 6225 return ext4_write_dquot(dquot); 6226 } else { 6227 return dquot_mark_dquot_dirty(dquot); 6228 } 6229 } 6230 6231 static int ext4_write_info(struct super_block *sb, int type) 6232 { 6233 int ret, err; 6234 handle_t *handle; 6235 6236 /* Data block + inode block */ 6237 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 6238 if (IS_ERR(handle)) 6239 return PTR_ERR(handle); 6240 ret = dquot_commit_info(sb, type); 6241 err = ext4_journal_stop(handle); 6242 if (!ret) 6243 ret = err; 6244 return ret; 6245 } 6246 6247 /* 6248 * Turn on quotas during mount time - we need to find 6249 * the quota file and such... 6250 */ 6251 static int ext4_quota_on_mount(struct super_block *sb, int type) 6252 { 6253 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 6254 EXT4_SB(sb)->s_jquota_fmt, type); 6255 } 6256 6257 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6258 { 6259 struct ext4_inode_info *ei = EXT4_I(inode); 6260 6261 /* The first argument of lockdep_set_subclass has to be 6262 * *exactly* the same as the argument to init_rwsem() --- in 6263 * this case, in init_once() --- or lockdep gets unhappy 6264 * because the name of the lock is set using the 6265 * stringification of the argument to init_rwsem(). 6266 */ 6267 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6268 lockdep_set_subclass(&ei->i_data_sem, subclass); 6269 } 6270 6271 /* 6272 * Standard function to be called on quota_on 6273 */ 6274 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6275 const struct path *path) 6276 { 6277 int err; 6278 6279 if (!test_opt(sb, QUOTA)) 6280 return -EINVAL; 6281 6282 /* Quotafile not on the same filesystem? */ 6283 if (path->dentry->d_sb != sb) 6284 return -EXDEV; 6285 6286 /* Quota already enabled for this file? */ 6287 if (IS_NOQUOTA(d_inode(path->dentry))) 6288 return -EBUSY; 6289 6290 /* Journaling quota? */ 6291 if (EXT4_SB(sb)->s_qf_names[type]) { 6292 /* Quotafile not in fs root? */ 6293 if (path->dentry->d_parent != sb->s_root) 6294 ext4_msg(sb, KERN_WARNING, 6295 "Quota file not on filesystem root. " 6296 "Journaled quota will not work"); 6297 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6298 } else { 6299 /* 6300 * Clear the flag just in case mount options changed since 6301 * last time. 6302 */ 6303 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6304 } 6305 6306 /* 6307 * When we journal data on quota file, we have to flush journal to see 6308 * all updates to the file when we bypass pagecache... 6309 */ 6310 if (EXT4_SB(sb)->s_journal && 6311 ext4_should_journal_data(d_inode(path->dentry))) { 6312 /* 6313 * We don't need to lock updates but journal_flush() could 6314 * otherwise be livelocked... 6315 */ 6316 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6317 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 6318 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6319 if (err) 6320 return err; 6321 } 6322 6323 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6324 err = dquot_quota_on(sb, type, format_id, path); 6325 if (err) { 6326 lockdep_set_quota_inode(path->dentry->d_inode, 6327 I_DATA_SEM_NORMAL); 6328 } else { 6329 struct inode *inode = d_inode(path->dentry); 6330 handle_t *handle; 6331 6332 /* 6333 * Set inode flags to prevent userspace from messing with quota 6334 * files. If this fails, we return success anyway since quotas 6335 * are already enabled and this is not a hard failure. 6336 */ 6337 inode_lock(inode); 6338 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6339 if (IS_ERR(handle)) 6340 goto unlock_inode; 6341 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6342 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6343 S_NOATIME | S_IMMUTABLE); 6344 err = ext4_mark_inode_dirty(handle, inode); 6345 ext4_journal_stop(handle); 6346 unlock_inode: 6347 inode_unlock(inode); 6348 } 6349 return err; 6350 } 6351 6352 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6353 unsigned int flags) 6354 { 6355 int err; 6356 struct inode *qf_inode; 6357 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6358 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6359 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6360 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6361 }; 6362 6363 BUG_ON(!ext4_has_feature_quota(sb)); 6364 6365 if (!qf_inums[type]) 6366 return -EPERM; 6367 6368 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6369 if (IS_ERR(qf_inode)) { 6370 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6371 return PTR_ERR(qf_inode); 6372 } 6373 6374 /* Don't account quota for quota files to avoid recursion */ 6375 qf_inode->i_flags |= S_NOQUOTA; 6376 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6377 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6378 if (err) 6379 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6380 iput(qf_inode); 6381 6382 return err; 6383 } 6384 6385 /* Enable usage tracking for all quota types. */ 6386 static int ext4_enable_quotas(struct super_block *sb) 6387 { 6388 int type, err = 0; 6389 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6390 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6391 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6392 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6393 }; 6394 bool quota_mopt[EXT4_MAXQUOTAS] = { 6395 test_opt(sb, USRQUOTA), 6396 test_opt(sb, GRPQUOTA), 6397 test_opt(sb, PRJQUOTA), 6398 }; 6399 6400 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6401 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6402 if (qf_inums[type]) { 6403 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6404 DQUOT_USAGE_ENABLED | 6405 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6406 if (err) { 6407 ext4_warning(sb, 6408 "Failed to enable quota tracking " 6409 "(type=%d, err=%d). Please run " 6410 "e2fsck to fix.", type, err); 6411 for (type--; type >= 0; type--) 6412 dquot_quota_off(sb, type); 6413 6414 return err; 6415 } 6416 } 6417 } 6418 return 0; 6419 } 6420 6421 static int ext4_quota_off(struct super_block *sb, int type) 6422 { 6423 struct inode *inode = sb_dqopt(sb)->files[type]; 6424 handle_t *handle; 6425 int err; 6426 6427 /* Force all delayed allocation blocks to be allocated. 6428 * Caller already holds s_umount sem */ 6429 if (test_opt(sb, DELALLOC)) 6430 sync_filesystem(sb); 6431 6432 if (!inode || !igrab(inode)) 6433 goto out; 6434 6435 err = dquot_quota_off(sb, type); 6436 if (err || ext4_has_feature_quota(sb)) 6437 goto out_put; 6438 6439 inode_lock(inode); 6440 /* 6441 * Update modification times of quota files when userspace can 6442 * start looking at them. If we fail, we return success anyway since 6443 * this is not a hard failure and quotas are already disabled. 6444 */ 6445 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6446 if (IS_ERR(handle)) { 6447 err = PTR_ERR(handle); 6448 goto out_unlock; 6449 } 6450 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 6451 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 6452 inode->i_mtime = inode->i_ctime = current_time(inode); 6453 err = ext4_mark_inode_dirty(handle, inode); 6454 ext4_journal_stop(handle); 6455 out_unlock: 6456 inode_unlock(inode); 6457 out_put: 6458 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 6459 iput(inode); 6460 return err; 6461 out: 6462 return dquot_quota_off(sb, type); 6463 } 6464 6465 /* Read data from quotafile - avoid pagecache and such because we cannot afford 6466 * acquiring the locks... As quota files are never truncated and quota code 6467 * itself serializes the operations (and no one else should touch the files) 6468 * we don't have to be afraid of races */ 6469 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 6470 size_t len, loff_t off) 6471 { 6472 struct inode *inode = sb_dqopt(sb)->files[type]; 6473 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6474 int offset = off & (sb->s_blocksize - 1); 6475 int tocopy; 6476 size_t toread; 6477 struct buffer_head *bh; 6478 loff_t i_size = i_size_read(inode); 6479 6480 if (off > i_size) 6481 return 0; 6482 if (off+len > i_size) 6483 len = i_size-off; 6484 toread = len; 6485 while (toread > 0) { 6486 tocopy = sb->s_blocksize - offset < toread ? 6487 sb->s_blocksize - offset : toread; 6488 bh = ext4_bread(NULL, inode, blk, 0); 6489 if (IS_ERR(bh)) 6490 return PTR_ERR(bh); 6491 if (!bh) /* A hole? */ 6492 memset(data, 0, tocopy); 6493 else 6494 memcpy(data, bh->b_data+offset, tocopy); 6495 brelse(bh); 6496 offset = 0; 6497 toread -= tocopy; 6498 data += tocopy; 6499 blk++; 6500 } 6501 return len; 6502 } 6503 6504 /* Write to quotafile (we know the transaction is already started and has 6505 * enough credits) */ 6506 static ssize_t ext4_quota_write(struct super_block *sb, int type, 6507 const char *data, size_t len, loff_t off) 6508 { 6509 struct inode *inode = sb_dqopt(sb)->files[type]; 6510 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6511 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 6512 int retries = 0; 6513 struct buffer_head *bh; 6514 handle_t *handle = journal_current_handle(); 6515 6516 if (EXT4_SB(sb)->s_journal && !handle) { 6517 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6518 " cancelled because transaction is not started", 6519 (unsigned long long)off, (unsigned long long)len); 6520 return -EIO; 6521 } 6522 /* 6523 * Since we account only one data block in transaction credits, 6524 * then it is impossible to cross a block boundary. 6525 */ 6526 if (sb->s_blocksize - offset < len) { 6527 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6528 " cancelled because not block aligned", 6529 (unsigned long long)off, (unsigned long long)len); 6530 return -EIO; 6531 } 6532 6533 do { 6534 bh = ext4_bread(handle, inode, blk, 6535 EXT4_GET_BLOCKS_CREATE | 6536 EXT4_GET_BLOCKS_METADATA_NOFAIL); 6537 } while (PTR_ERR(bh) == -ENOSPC && 6538 ext4_should_retry_alloc(inode->i_sb, &retries)); 6539 if (IS_ERR(bh)) 6540 return PTR_ERR(bh); 6541 if (!bh) 6542 goto out; 6543 BUFFER_TRACE(bh, "get write access"); 6544 err = ext4_journal_get_write_access(handle, bh); 6545 if (err) { 6546 brelse(bh); 6547 return err; 6548 } 6549 lock_buffer(bh); 6550 memcpy(bh->b_data+offset, data, len); 6551 flush_dcache_page(bh->b_page); 6552 unlock_buffer(bh); 6553 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6554 brelse(bh); 6555 out: 6556 if (inode->i_size < off + len) { 6557 i_size_write(inode, off + len); 6558 EXT4_I(inode)->i_disksize = inode->i_size; 6559 err2 = ext4_mark_inode_dirty(handle, inode); 6560 if (unlikely(err2 && !err)) 6561 err = err2; 6562 } 6563 return err ? err : len; 6564 } 6565 #endif 6566 6567 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6568 const char *dev_name, void *data) 6569 { 6570 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6571 } 6572 6573 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6574 static inline void register_as_ext2(void) 6575 { 6576 int err = register_filesystem(&ext2_fs_type); 6577 if (err) 6578 printk(KERN_WARNING 6579 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6580 } 6581 6582 static inline void unregister_as_ext2(void) 6583 { 6584 unregister_filesystem(&ext2_fs_type); 6585 } 6586 6587 static inline int ext2_feature_set_ok(struct super_block *sb) 6588 { 6589 if (ext4_has_unknown_ext2_incompat_features(sb)) 6590 return 0; 6591 if (sb_rdonly(sb)) 6592 return 1; 6593 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6594 return 0; 6595 return 1; 6596 } 6597 #else 6598 static inline void register_as_ext2(void) { } 6599 static inline void unregister_as_ext2(void) { } 6600 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6601 #endif 6602 6603 static inline void register_as_ext3(void) 6604 { 6605 int err = register_filesystem(&ext3_fs_type); 6606 if (err) 6607 printk(KERN_WARNING 6608 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6609 } 6610 6611 static inline void unregister_as_ext3(void) 6612 { 6613 unregister_filesystem(&ext3_fs_type); 6614 } 6615 6616 static inline int ext3_feature_set_ok(struct super_block *sb) 6617 { 6618 if (ext4_has_unknown_ext3_incompat_features(sb)) 6619 return 0; 6620 if (!ext4_has_feature_journal(sb)) 6621 return 0; 6622 if (sb_rdonly(sb)) 6623 return 1; 6624 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6625 return 0; 6626 return 1; 6627 } 6628 6629 static struct file_system_type ext4_fs_type = { 6630 .owner = THIS_MODULE, 6631 .name = "ext4", 6632 .mount = ext4_mount, 6633 .kill_sb = kill_block_super, 6634 .fs_flags = FS_REQUIRES_DEV, 6635 }; 6636 MODULE_ALIAS_FS("ext4"); 6637 6638 /* Shared across all ext4 file systems */ 6639 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6640 6641 static int __init ext4_init_fs(void) 6642 { 6643 int i, err; 6644 6645 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6646 ext4_li_info = NULL; 6647 mutex_init(&ext4_li_mtx); 6648 6649 /* Build-time check for flags consistency */ 6650 ext4_check_flag_values(); 6651 6652 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6653 init_waitqueue_head(&ext4__ioend_wq[i]); 6654 6655 err = ext4_init_es(); 6656 if (err) 6657 return err; 6658 6659 err = ext4_init_pending(); 6660 if (err) 6661 goto out7; 6662 6663 err = ext4_init_post_read_processing(); 6664 if (err) 6665 goto out6; 6666 6667 err = ext4_init_pageio(); 6668 if (err) 6669 goto out5; 6670 6671 err = ext4_init_system_zone(); 6672 if (err) 6673 goto out4; 6674 6675 err = ext4_init_sysfs(); 6676 if (err) 6677 goto out3; 6678 6679 err = ext4_init_mballoc(); 6680 if (err) 6681 goto out2; 6682 err = init_inodecache(); 6683 if (err) 6684 goto out1; 6685 6686 err = ext4_fc_init_dentry_cache(); 6687 if (err) 6688 goto out05; 6689 6690 register_as_ext3(); 6691 register_as_ext2(); 6692 err = register_filesystem(&ext4_fs_type); 6693 if (err) 6694 goto out; 6695 6696 return 0; 6697 out: 6698 unregister_as_ext2(); 6699 unregister_as_ext3(); 6700 out05: 6701 destroy_inodecache(); 6702 out1: 6703 ext4_exit_mballoc(); 6704 out2: 6705 ext4_exit_sysfs(); 6706 out3: 6707 ext4_exit_system_zone(); 6708 out4: 6709 ext4_exit_pageio(); 6710 out5: 6711 ext4_exit_post_read_processing(); 6712 out6: 6713 ext4_exit_pending(); 6714 out7: 6715 ext4_exit_es(); 6716 6717 return err; 6718 } 6719 6720 static void __exit ext4_exit_fs(void) 6721 { 6722 ext4_destroy_lazyinit_thread(); 6723 unregister_as_ext2(); 6724 unregister_as_ext3(); 6725 unregister_filesystem(&ext4_fs_type); 6726 destroy_inodecache(); 6727 ext4_exit_mballoc(); 6728 ext4_exit_sysfs(); 6729 ext4_exit_system_zone(); 6730 ext4_exit_pageio(); 6731 ext4_exit_post_read_processing(); 6732 ext4_exit_es(); 6733 ext4_exit_pending(); 6734 } 6735 6736 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6737 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6738 MODULE_LICENSE("GPL"); 6739 MODULE_SOFTDEP("pre: crc32c"); 6740 module_init(ext4_init_fs) 6741 module_exit(ext4_exit_fs) 6742