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 ext4_show_quota_options(seq, sb); 2642 return 0; 2643 } 2644 2645 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2646 { 2647 return _ext4_show_options(seq, root->d_sb, 0); 2648 } 2649 2650 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2651 { 2652 struct super_block *sb = seq->private; 2653 int rc; 2654 2655 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2656 rc = _ext4_show_options(seq, sb, 1); 2657 seq_puts(seq, "\n"); 2658 return rc; 2659 } 2660 2661 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2662 int read_only) 2663 { 2664 struct ext4_sb_info *sbi = EXT4_SB(sb); 2665 int err = 0; 2666 2667 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2668 ext4_msg(sb, KERN_ERR, "revision level too high, " 2669 "forcing read-only mode"); 2670 err = -EROFS; 2671 goto done; 2672 } 2673 if (read_only) 2674 goto done; 2675 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2676 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2677 "running e2fsck is recommended"); 2678 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2679 ext4_msg(sb, KERN_WARNING, 2680 "warning: mounting fs with errors, " 2681 "running e2fsck is recommended"); 2682 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2683 le16_to_cpu(es->s_mnt_count) >= 2684 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2685 ext4_msg(sb, KERN_WARNING, 2686 "warning: maximal mount count reached, " 2687 "running e2fsck is recommended"); 2688 else if (le32_to_cpu(es->s_checkinterval) && 2689 (ext4_get_tstamp(es, s_lastcheck) + 2690 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2691 ext4_msg(sb, KERN_WARNING, 2692 "warning: checktime reached, " 2693 "running e2fsck is recommended"); 2694 if (!sbi->s_journal) 2695 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2696 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2697 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2698 le16_add_cpu(&es->s_mnt_count, 1); 2699 ext4_update_tstamp(es, s_mtime); 2700 if (sbi->s_journal) 2701 ext4_set_feature_journal_needs_recovery(sb); 2702 2703 err = ext4_commit_super(sb, 1); 2704 done: 2705 if (test_opt(sb, DEBUG)) 2706 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2707 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2708 sb->s_blocksize, 2709 sbi->s_groups_count, 2710 EXT4_BLOCKS_PER_GROUP(sb), 2711 EXT4_INODES_PER_GROUP(sb), 2712 sbi->s_mount_opt, sbi->s_mount_opt2); 2713 2714 cleancache_init_fs(sb); 2715 return err; 2716 } 2717 2718 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2719 { 2720 struct ext4_sb_info *sbi = EXT4_SB(sb); 2721 struct flex_groups **old_groups, **new_groups; 2722 int size, i, j; 2723 2724 if (!sbi->s_log_groups_per_flex) 2725 return 0; 2726 2727 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2728 if (size <= sbi->s_flex_groups_allocated) 2729 return 0; 2730 2731 new_groups = kvzalloc(roundup_pow_of_two(size * 2732 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 2733 if (!new_groups) { 2734 ext4_msg(sb, KERN_ERR, 2735 "not enough memory for %d flex group pointers", size); 2736 return -ENOMEM; 2737 } 2738 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 2739 new_groups[i] = kvzalloc(roundup_pow_of_two( 2740 sizeof(struct flex_groups)), 2741 GFP_KERNEL); 2742 if (!new_groups[i]) { 2743 for (j = sbi->s_flex_groups_allocated; j < i; j++) 2744 kvfree(new_groups[j]); 2745 kvfree(new_groups); 2746 ext4_msg(sb, KERN_ERR, 2747 "not enough memory for %d flex groups", size); 2748 return -ENOMEM; 2749 } 2750 } 2751 rcu_read_lock(); 2752 old_groups = rcu_dereference(sbi->s_flex_groups); 2753 if (old_groups) 2754 memcpy(new_groups, old_groups, 2755 (sbi->s_flex_groups_allocated * 2756 sizeof(struct flex_groups *))); 2757 rcu_read_unlock(); 2758 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 2759 sbi->s_flex_groups_allocated = size; 2760 if (old_groups) 2761 ext4_kvfree_array_rcu(old_groups); 2762 return 0; 2763 } 2764 2765 static int ext4_fill_flex_info(struct super_block *sb) 2766 { 2767 struct ext4_sb_info *sbi = EXT4_SB(sb); 2768 struct ext4_group_desc *gdp = NULL; 2769 struct flex_groups *fg; 2770 ext4_group_t flex_group; 2771 int i, err; 2772 2773 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2774 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2775 sbi->s_log_groups_per_flex = 0; 2776 return 1; 2777 } 2778 2779 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2780 if (err) 2781 goto failed; 2782 2783 for (i = 0; i < sbi->s_groups_count; i++) { 2784 gdp = ext4_get_group_desc(sb, i, NULL); 2785 2786 flex_group = ext4_flex_group(sbi, i); 2787 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 2788 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 2789 atomic64_add(ext4_free_group_clusters(sb, gdp), 2790 &fg->free_clusters); 2791 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 2792 } 2793 2794 return 1; 2795 failed: 2796 return 0; 2797 } 2798 2799 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2800 struct ext4_group_desc *gdp) 2801 { 2802 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2803 __u16 crc = 0; 2804 __le32 le_group = cpu_to_le32(block_group); 2805 struct ext4_sb_info *sbi = EXT4_SB(sb); 2806 2807 if (ext4_has_metadata_csum(sbi->s_sb)) { 2808 /* Use new metadata_csum algorithm */ 2809 __u32 csum32; 2810 __u16 dummy_csum = 0; 2811 2812 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2813 sizeof(le_group)); 2814 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2815 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2816 sizeof(dummy_csum)); 2817 offset += sizeof(dummy_csum); 2818 if (offset < sbi->s_desc_size) 2819 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2820 sbi->s_desc_size - offset); 2821 2822 crc = csum32 & 0xFFFF; 2823 goto out; 2824 } 2825 2826 /* old crc16 code */ 2827 if (!ext4_has_feature_gdt_csum(sb)) 2828 return 0; 2829 2830 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2831 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2832 crc = crc16(crc, (__u8 *)gdp, offset); 2833 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2834 /* for checksum of struct ext4_group_desc do the rest...*/ 2835 if (ext4_has_feature_64bit(sb) && 2836 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2837 crc = crc16(crc, (__u8 *)gdp + offset, 2838 le16_to_cpu(sbi->s_es->s_desc_size) - 2839 offset); 2840 2841 out: 2842 return cpu_to_le16(crc); 2843 } 2844 2845 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2846 struct ext4_group_desc *gdp) 2847 { 2848 if (ext4_has_group_desc_csum(sb) && 2849 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2850 return 0; 2851 2852 return 1; 2853 } 2854 2855 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2856 struct ext4_group_desc *gdp) 2857 { 2858 if (!ext4_has_group_desc_csum(sb)) 2859 return; 2860 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2861 } 2862 2863 /* Called at mount-time, super-block is locked */ 2864 static int ext4_check_descriptors(struct super_block *sb, 2865 ext4_fsblk_t sb_block, 2866 ext4_group_t *first_not_zeroed) 2867 { 2868 struct ext4_sb_info *sbi = EXT4_SB(sb); 2869 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2870 ext4_fsblk_t last_block; 2871 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2872 ext4_fsblk_t block_bitmap; 2873 ext4_fsblk_t inode_bitmap; 2874 ext4_fsblk_t inode_table; 2875 int flexbg_flag = 0; 2876 ext4_group_t i, grp = sbi->s_groups_count; 2877 2878 if (ext4_has_feature_flex_bg(sb)) 2879 flexbg_flag = 1; 2880 2881 ext4_debug("Checking group descriptors"); 2882 2883 for (i = 0; i < sbi->s_groups_count; i++) { 2884 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2885 2886 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2887 last_block = ext4_blocks_count(sbi->s_es) - 1; 2888 else 2889 last_block = first_block + 2890 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2891 2892 if ((grp == sbi->s_groups_count) && 2893 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2894 grp = i; 2895 2896 block_bitmap = ext4_block_bitmap(sb, gdp); 2897 if (block_bitmap == sb_block) { 2898 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2899 "Block bitmap for group %u overlaps " 2900 "superblock", i); 2901 if (!sb_rdonly(sb)) 2902 return 0; 2903 } 2904 if (block_bitmap >= sb_block + 1 && 2905 block_bitmap <= last_bg_block) { 2906 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2907 "Block bitmap for group %u overlaps " 2908 "block group descriptors", i); 2909 if (!sb_rdonly(sb)) 2910 return 0; 2911 } 2912 if (block_bitmap < first_block || block_bitmap > last_block) { 2913 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2914 "Block bitmap for group %u not in group " 2915 "(block %llu)!", i, block_bitmap); 2916 return 0; 2917 } 2918 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2919 if (inode_bitmap == sb_block) { 2920 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2921 "Inode bitmap for group %u overlaps " 2922 "superblock", i); 2923 if (!sb_rdonly(sb)) 2924 return 0; 2925 } 2926 if (inode_bitmap >= sb_block + 1 && 2927 inode_bitmap <= last_bg_block) { 2928 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2929 "Inode bitmap for group %u overlaps " 2930 "block group descriptors", i); 2931 if (!sb_rdonly(sb)) 2932 return 0; 2933 } 2934 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2935 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2936 "Inode bitmap for group %u not in group " 2937 "(block %llu)!", i, inode_bitmap); 2938 return 0; 2939 } 2940 inode_table = ext4_inode_table(sb, gdp); 2941 if (inode_table == sb_block) { 2942 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2943 "Inode table for group %u overlaps " 2944 "superblock", i); 2945 if (!sb_rdonly(sb)) 2946 return 0; 2947 } 2948 if (inode_table >= sb_block + 1 && 2949 inode_table <= last_bg_block) { 2950 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2951 "Inode table for group %u overlaps " 2952 "block group descriptors", i); 2953 if (!sb_rdonly(sb)) 2954 return 0; 2955 } 2956 if (inode_table < first_block || 2957 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2958 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2959 "Inode table for group %u not in group " 2960 "(block %llu)!", i, inode_table); 2961 return 0; 2962 } 2963 ext4_lock_group(sb, i); 2964 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2965 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2966 "Checksum for group %u failed (%u!=%u)", 2967 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2968 gdp)), le16_to_cpu(gdp->bg_checksum)); 2969 if (!sb_rdonly(sb)) { 2970 ext4_unlock_group(sb, i); 2971 return 0; 2972 } 2973 } 2974 ext4_unlock_group(sb, i); 2975 if (!flexbg_flag) 2976 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2977 } 2978 if (NULL != first_not_zeroed) 2979 *first_not_zeroed = grp; 2980 return 1; 2981 } 2982 2983 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2984 * the superblock) which were deleted from all directories, but held open by 2985 * a process at the time of a crash. We walk the list and try to delete these 2986 * inodes at recovery time (only with a read-write filesystem). 2987 * 2988 * In order to keep the orphan inode chain consistent during traversal (in 2989 * case of crash during recovery), we link each inode into the superblock 2990 * orphan list_head and handle it the same way as an inode deletion during 2991 * normal operation (which journals the operations for us). 2992 * 2993 * We only do an iget() and an iput() on each inode, which is very safe if we 2994 * accidentally point at an in-use or already deleted inode. The worst that 2995 * can happen in this case is that we get a "bit already cleared" message from 2996 * ext4_free_inode(). The only reason we would point at a wrong inode is if 2997 * e2fsck was run on this filesystem, and it must have already done the orphan 2998 * inode cleanup for us, so we can safely abort without any further action. 2999 */ 3000 static void ext4_orphan_cleanup(struct super_block *sb, 3001 struct ext4_super_block *es) 3002 { 3003 unsigned int s_flags = sb->s_flags; 3004 int ret, nr_orphans = 0, nr_truncates = 0; 3005 #ifdef CONFIG_QUOTA 3006 int quota_update = 0; 3007 int i; 3008 #endif 3009 if (!es->s_last_orphan) { 3010 jbd_debug(4, "no orphan inodes to clean up\n"); 3011 return; 3012 } 3013 3014 if (bdev_read_only(sb->s_bdev)) { 3015 ext4_msg(sb, KERN_ERR, "write access " 3016 "unavailable, skipping orphan cleanup"); 3017 return; 3018 } 3019 3020 /* Check if feature set would not allow a r/w mount */ 3021 if (!ext4_feature_set_ok(sb, 0)) { 3022 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 3023 "unknown ROCOMPAT features"); 3024 return; 3025 } 3026 3027 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3028 /* don't clear list on RO mount w/ errors */ 3029 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 3030 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 3031 "clearing orphan list.\n"); 3032 es->s_last_orphan = 0; 3033 } 3034 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3035 return; 3036 } 3037 3038 if (s_flags & SB_RDONLY) { 3039 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 3040 sb->s_flags &= ~SB_RDONLY; 3041 } 3042 #ifdef CONFIG_QUOTA 3043 /* Needed for iput() to work correctly and not trash data */ 3044 sb->s_flags |= SB_ACTIVE; 3045 3046 /* 3047 * Turn on quotas which were not enabled for read-only mounts if 3048 * filesystem has quota feature, so that they are updated correctly. 3049 */ 3050 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 3051 int ret = ext4_enable_quotas(sb); 3052 3053 if (!ret) 3054 quota_update = 1; 3055 else 3056 ext4_msg(sb, KERN_ERR, 3057 "Cannot turn on quotas: error %d", ret); 3058 } 3059 3060 /* Turn on journaled quotas used for old sytle */ 3061 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3062 if (EXT4_SB(sb)->s_qf_names[i]) { 3063 int ret = ext4_quota_on_mount(sb, i); 3064 3065 if (!ret) 3066 quota_update = 1; 3067 else 3068 ext4_msg(sb, KERN_ERR, 3069 "Cannot turn on journaled " 3070 "quota: type %d: error %d", i, ret); 3071 } 3072 } 3073 #endif 3074 3075 while (es->s_last_orphan) { 3076 struct inode *inode; 3077 3078 /* 3079 * We may have encountered an error during cleanup; if 3080 * so, skip the rest. 3081 */ 3082 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3083 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3084 es->s_last_orphan = 0; 3085 break; 3086 } 3087 3088 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 3089 if (IS_ERR(inode)) { 3090 es->s_last_orphan = 0; 3091 break; 3092 } 3093 3094 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 3095 dquot_initialize(inode); 3096 if (inode->i_nlink) { 3097 if (test_opt(sb, DEBUG)) 3098 ext4_msg(sb, KERN_DEBUG, 3099 "%s: truncating inode %lu to %lld bytes", 3100 __func__, inode->i_ino, inode->i_size); 3101 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 3102 inode->i_ino, inode->i_size); 3103 inode_lock(inode); 3104 truncate_inode_pages(inode->i_mapping, inode->i_size); 3105 ret = ext4_truncate(inode); 3106 if (ret) 3107 ext4_std_error(inode->i_sb, ret); 3108 inode_unlock(inode); 3109 nr_truncates++; 3110 } else { 3111 if (test_opt(sb, DEBUG)) 3112 ext4_msg(sb, KERN_DEBUG, 3113 "%s: deleting unreferenced inode %lu", 3114 __func__, inode->i_ino); 3115 jbd_debug(2, "deleting unreferenced inode %lu\n", 3116 inode->i_ino); 3117 nr_orphans++; 3118 } 3119 iput(inode); /* The delete magic happens here! */ 3120 } 3121 3122 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 3123 3124 if (nr_orphans) 3125 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 3126 PLURAL(nr_orphans)); 3127 if (nr_truncates) 3128 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 3129 PLURAL(nr_truncates)); 3130 #ifdef CONFIG_QUOTA 3131 /* Turn off quotas if they were enabled for orphan cleanup */ 3132 if (quota_update) { 3133 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3134 if (sb_dqopt(sb)->files[i]) 3135 dquot_quota_off(sb, i); 3136 } 3137 } 3138 #endif 3139 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 3140 } 3141 3142 /* 3143 * Maximal extent format file size. 3144 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3145 * extent format containers, within a sector_t, and within i_blocks 3146 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3147 * so that won't be a limiting factor. 3148 * 3149 * However there is other limiting factor. We do store extents in the form 3150 * of starting block and length, hence the resulting length of the extent 3151 * covering maximum file size must fit into on-disk format containers as 3152 * well. Given that length is always by 1 unit bigger than max unit (because 3153 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3154 * 3155 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3156 */ 3157 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3158 { 3159 loff_t res; 3160 loff_t upper_limit = MAX_LFS_FILESIZE; 3161 3162 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3163 3164 if (!has_huge_files) { 3165 upper_limit = (1LL << 32) - 1; 3166 3167 /* total blocks in file system block size */ 3168 upper_limit >>= (blkbits - 9); 3169 upper_limit <<= blkbits; 3170 } 3171 3172 /* 3173 * 32-bit extent-start container, ee_block. We lower the maxbytes 3174 * by one fs block, so ee_len can cover the extent of maximum file 3175 * size 3176 */ 3177 res = (1LL << 32) - 1; 3178 res <<= blkbits; 3179 3180 /* Sanity check against vm- & vfs- imposed limits */ 3181 if (res > upper_limit) 3182 res = upper_limit; 3183 3184 return res; 3185 } 3186 3187 /* 3188 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3189 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3190 * We need to be 1 filesystem block less than the 2^48 sector limit. 3191 */ 3192 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3193 { 3194 loff_t res = EXT4_NDIR_BLOCKS; 3195 int meta_blocks; 3196 loff_t upper_limit; 3197 /* This is calculated to be the largest file size for a dense, block 3198 * mapped file such that the file's total number of 512-byte sectors, 3199 * including data and all indirect blocks, does not exceed (2^48 - 1). 3200 * 3201 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3202 * number of 512-byte sectors of the file. 3203 */ 3204 3205 if (!has_huge_files) { 3206 /* 3207 * !has_huge_files or implies that the inode i_block field 3208 * represents total file blocks in 2^32 512-byte sectors == 3209 * size of vfs inode i_blocks * 8 3210 */ 3211 upper_limit = (1LL << 32) - 1; 3212 3213 /* total blocks in file system block size */ 3214 upper_limit >>= (bits - 9); 3215 3216 } else { 3217 /* 3218 * We use 48 bit ext4_inode i_blocks 3219 * With EXT4_HUGE_FILE_FL set the i_blocks 3220 * represent total number of blocks in 3221 * file system block size 3222 */ 3223 upper_limit = (1LL << 48) - 1; 3224 3225 } 3226 3227 /* indirect blocks */ 3228 meta_blocks = 1; 3229 /* double indirect blocks */ 3230 meta_blocks += 1 + (1LL << (bits-2)); 3231 /* tripple indirect blocks */ 3232 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 3233 3234 upper_limit -= meta_blocks; 3235 upper_limit <<= bits; 3236 3237 res += 1LL << (bits-2); 3238 res += 1LL << (2*(bits-2)); 3239 res += 1LL << (3*(bits-2)); 3240 res <<= bits; 3241 if (res > upper_limit) 3242 res = upper_limit; 3243 3244 if (res > MAX_LFS_FILESIZE) 3245 res = MAX_LFS_FILESIZE; 3246 3247 return res; 3248 } 3249 3250 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3251 ext4_fsblk_t logical_sb_block, int nr) 3252 { 3253 struct ext4_sb_info *sbi = EXT4_SB(sb); 3254 ext4_group_t bg, first_meta_bg; 3255 int has_super = 0; 3256 3257 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3258 3259 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3260 return logical_sb_block + nr + 1; 3261 bg = sbi->s_desc_per_block * nr; 3262 if (ext4_bg_has_super(sb, bg)) 3263 has_super = 1; 3264 3265 /* 3266 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3267 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3268 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3269 * compensate. 3270 */ 3271 if (sb->s_blocksize == 1024 && nr == 0 && 3272 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3273 has_super++; 3274 3275 return (has_super + ext4_group_first_block_no(sb, bg)); 3276 } 3277 3278 /** 3279 * ext4_get_stripe_size: Get the stripe size. 3280 * @sbi: In memory super block info 3281 * 3282 * If we have specified it via mount option, then 3283 * use the mount option value. If the value specified at mount time is 3284 * greater than the blocks per group use the super block value. 3285 * If the super block value is greater than blocks per group return 0. 3286 * Allocator needs it be less than blocks per group. 3287 * 3288 */ 3289 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3290 { 3291 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3292 unsigned long stripe_width = 3293 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3294 int ret; 3295 3296 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3297 ret = sbi->s_stripe; 3298 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3299 ret = stripe_width; 3300 else if (stride && stride <= sbi->s_blocks_per_group) 3301 ret = stride; 3302 else 3303 ret = 0; 3304 3305 /* 3306 * If the stripe width is 1, this makes no sense and 3307 * we set it to 0 to turn off stripe handling code. 3308 */ 3309 if (ret <= 1) 3310 ret = 0; 3311 3312 return ret; 3313 } 3314 3315 /* 3316 * Check whether this filesystem can be mounted based on 3317 * the features present and the RDONLY/RDWR mount requested. 3318 * Returns 1 if this filesystem can be mounted as requested, 3319 * 0 if it cannot be. 3320 */ 3321 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 3322 { 3323 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3324 ext4_msg(sb, KERN_ERR, 3325 "Couldn't mount because of " 3326 "unsupported optional features (%x)", 3327 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3328 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3329 return 0; 3330 } 3331 3332 #ifndef CONFIG_UNICODE 3333 if (ext4_has_feature_casefold(sb)) { 3334 ext4_msg(sb, KERN_ERR, 3335 "Filesystem with casefold feature cannot be " 3336 "mounted without CONFIG_UNICODE"); 3337 return 0; 3338 } 3339 #endif 3340 3341 if (readonly) 3342 return 1; 3343 3344 if (ext4_has_feature_readonly(sb)) { 3345 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3346 sb->s_flags |= SB_RDONLY; 3347 return 1; 3348 } 3349 3350 /* Check that feature set is OK for a read-write mount */ 3351 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3352 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3353 "unsupported optional features (%x)", 3354 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3355 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3356 return 0; 3357 } 3358 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3359 ext4_msg(sb, KERN_ERR, 3360 "Can't support bigalloc feature without " 3361 "extents feature\n"); 3362 return 0; 3363 } 3364 3365 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3366 if (!readonly && (ext4_has_feature_quota(sb) || 3367 ext4_has_feature_project(sb))) { 3368 ext4_msg(sb, KERN_ERR, 3369 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3370 return 0; 3371 } 3372 #endif /* CONFIG_QUOTA */ 3373 return 1; 3374 } 3375 3376 /* 3377 * This function is called once a day if we have errors logged 3378 * on the file system 3379 */ 3380 static void print_daily_error_info(struct timer_list *t) 3381 { 3382 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3383 struct super_block *sb = sbi->s_sb; 3384 struct ext4_super_block *es = sbi->s_es; 3385 3386 if (es->s_error_count) 3387 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3388 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3389 le32_to_cpu(es->s_error_count)); 3390 if (es->s_first_error_time) { 3391 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3392 sb->s_id, 3393 ext4_get_tstamp(es, s_first_error_time), 3394 (int) sizeof(es->s_first_error_func), 3395 es->s_first_error_func, 3396 le32_to_cpu(es->s_first_error_line)); 3397 if (es->s_first_error_ino) 3398 printk(KERN_CONT ": inode %u", 3399 le32_to_cpu(es->s_first_error_ino)); 3400 if (es->s_first_error_block) 3401 printk(KERN_CONT ": block %llu", (unsigned long long) 3402 le64_to_cpu(es->s_first_error_block)); 3403 printk(KERN_CONT "\n"); 3404 } 3405 if (es->s_last_error_time) { 3406 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3407 sb->s_id, 3408 ext4_get_tstamp(es, s_last_error_time), 3409 (int) sizeof(es->s_last_error_func), 3410 es->s_last_error_func, 3411 le32_to_cpu(es->s_last_error_line)); 3412 if (es->s_last_error_ino) 3413 printk(KERN_CONT ": inode %u", 3414 le32_to_cpu(es->s_last_error_ino)); 3415 if (es->s_last_error_block) 3416 printk(KERN_CONT ": block %llu", (unsigned long long) 3417 le64_to_cpu(es->s_last_error_block)); 3418 printk(KERN_CONT "\n"); 3419 } 3420 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3421 } 3422 3423 /* Find next suitable group and run ext4_init_inode_table */ 3424 static int ext4_run_li_request(struct ext4_li_request *elr) 3425 { 3426 struct ext4_group_desc *gdp = NULL; 3427 struct super_block *sb = elr->lr_super; 3428 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3429 ext4_group_t group = elr->lr_next_group; 3430 unsigned long timeout = 0; 3431 unsigned int prefetch_ios = 0; 3432 int ret = 0; 3433 3434 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3435 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3436 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3437 if (prefetch_ios) 3438 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3439 prefetch_ios); 3440 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3441 prefetch_ios); 3442 if (group >= elr->lr_next_group) { 3443 ret = 1; 3444 if (elr->lr_first_not_zeroed != ngroups && 3445 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3446 elr->lr_next_group = elr->lr_first_not_zeroed; 3447 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3448 ret = 0; 3449 } 3450 } 3451 return ret; 3452 } 3453 3454 for (; group < ngroups; group++) { 3455 gdp = ext4_get_group_desc(sb, group, NULL); 3456 if (!gdp) { 3457 ret = 1; 3458 break; 3459 } 3460 3461 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3462 break; 3463 } 3464 3465 if (group >= ngroups) 3466 ret = 1; 3467 3468 if (!ret) { 3469 timeout = jiffies; 3470 ret = ext4_init_inode_table(sb, group, 3471 elr->lr_timeout ? 0 : 1); 3472 trace_ext4_lazy_itable_init(sb, group); 3473 if (elr->lr_timeout == 0) { 3474 timeout = (jiffies - timeout) * 3475 EXT4_SB(elr->lr_super)->s_li_wait_mult; 3476 elr->lr_timeout = timeout; 3477 } 3478 elr->lr_next_sched = jiffies + elr->lr_timeout; 3479 elr->lr_next_group = group + 1; 3480 } 3481 return ret; 3482 } 3483 3484 /* 3485 * Remove lr_request from the list_request and free the 3486 * request structure. Should be called with li_list_mtx held 3487 */ 3488 static void ext4_remove_li_request(struct ext4_li_request *elr) 3489 { 3490 if (!elr) 3491 return; 3492 3493 list_del(&elr->lr_request); 3494 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3495 kfree(elr); 3496 } 3497 3498 static void ext4_unregister_li_request(struct super_block *sb) 3499 { 3500 mutex_lock(&ext4_li_mtx); 3501 if (!ext4_li_info) { 3502 mutex_unlock(&ext4_li_mtx); 3503 return; 3504 } 3505 3506 mutex_lock(&ext4_li_info->li_list_mtx); 3507 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3508 mutex_unlock(&ext4_li_info->li_list_mtx); 3509 mutex_unlock(&ext4_li_mtx); 3510 } 3511 3512 static struct task_struct *ext4_lazyinit_task; 3513 3514 /* 3515 * This is the function where ext4lazyinit thread lives. It walks 3516 * through the request list searching for next scheduled filesystem. 3517 * When such a fs is found, run the lazy initialization request 3518 * (ext4_rn_li_request) and keep track of the time spend in this 3519 * function. Based on that time we compute next schedule time of 3520 * the request. When walking through the list is complete, compute 3521 * next waking time and put itself into sleep. 3522 */ 3523 static int ext4_lazyinit_thread(void *arg) 3524 { 3525 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3526 struct list_head *pos, *n; 3527 struct ext4_li_request *elr; 3528 unsigned long next_wakeup, cur; 3529 3530 BUG_ON(NULL == eli); 3531 3532 cont_thread: 3533 while (true) { 3534 next_wakeup = MAX_JIFFY_OFFSET; 3535 3536 mutex_lock(&eli->li_list_mtx); 3537 if (list_empty(&eli->li_request_list)) { 3538 mutex_unlock(&eli->li_list_mtx); 3539 goto exit_thread; 3540 } 3541 list_for_each_safe(pos, n, &eli->li_request_list) { 3542 int err = 0; 3543 int progress = 0; 3544 elr = list_entry(pos, struct ext4_li_request, 3545 lr_request); 3546 3547 if (time_before(jiffies, elr->lr_next_sched)) { 3548 if (time_before(elr->lr_next_sched, next_wakeup)) 3549 next_wakeup = elr->lr_next_sched; 3550 continue; 3551 } 3552 if (down_read_trylock(&elr->lr_super->s_umount)) { 3553 if (sb_start_write_trylock(elr->lr_super)) { 3554 progress = 1; 3555 /* 3556 * We hold sb->s_umount, sb can not 3557 * be removed from the list, it is 3558 * now safe to drop li_list_mtx 3559 */ 3560 mutex_unlock(&eli->li_list_mtx); 3561 err = ext4_run_li_request(elr); 3562 sb_end_write(elr->lr_super); 3563 mutex_lock(&eli->li_list_mtx); 3564 n = pos->next; 3565 } 3566 up_read((&elr->lr_super->s_umount)); 3567 } 3568 /* error, remove the lazy_init job */ 3569 if (err) { 3570 ext4_remove_li_request(elr); 3571 continue; 3572 } 3573 if (!progress) { 3574 elr->lr_next_sched = jiffies + 3575 (prandom_u32() 3576 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3577 } 3578 if (time_before(elr->lr_next_sched, next_wakeup)) 3579 next_wakeup = elr->lr_next_sched; 3580 } 3581 mutex_unlock(&eli->li_list_mtx); 3582 3583 try_to_freeze(); 3584 3585 cur = jiffies; 3586 if ((time_after_eq(cur, next_wakeup)) || 3587 (MAX_JIFFY_OFFSET == next_wakeup)) { 3588 cond_resched(); 3589 continue; 3590 } 3591 3592 schedule_timeout_interruptible(next_wakeup - cur); 3593 3594 if (kthread_should_stop()) { 3595 ext4_clear_request_list(); 3596 goto exit_thread; 3597 } 3598 } 3599 3600 exit_thread: 3601 /* 3602 * It looks like the request list is empty, but we need 3603 * to check it under the li_list_mtx lock, to prevent any 3604 * additions into it, and of course we should lock ext4_li_mtx 3605 * to atomically free the list and ext4_li_info, because at 3606 * this point another ext4 filesystem could be registering 3607 * new one. 3608 */ 3609 mutex_lock(&ext4_li_mtx); 3610 mutex_lock(&eli->li_list_mtx); 3611 if (!list_empty(&eli->li_request_list)) { 3612 mutex_unlock(&eli->li_list_mtx); 3613 mutex_unlock(&ext4_li_mtx); 3614 goto cont_thread; 3615 } 3616 mutex_unlock(&eli->li_list_mtx); 3617 kfree(ext4_li_info); 3618 ext4_li_info = NULL; 3619 mutex_unlock(&ext4_li_mtx); 3620 3621 return 0; 3622 } 3623 3624 static void ext4_clear_request_list(void) 3625 { 3626 struct list_head *pos, *n; 3627 struct ext4_li_request *elr; 3628 3629 mutex_lock(&ext4_li_info->li_list_mtx); 3630 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3631 elr = list_entry(pos, struct ext4_li_request, 3632 lr_request); 3633 ext4_remove_li_request(elr); 3634 } 3635 mutex_unlock(&ext4_li_info->li_list_mtx); 3636 } 3637 3638 static int ext4_run_lazyinit_thread(void) 3639 { 3640 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3641 ext4_li_info, "ext4lazyinit"); 3642 if (IS_ERR(ext4_lazyinit_task)) { 3643 int err = PTR_ERR(ext4_lazyinit_task); 3644 ext4_clear_request_list(); 3645 kfree(ext4_li_info); 3646 ext4_li_info = NULL; 3647 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3648 "initialization thread\n", 3649 err); 3650 return err; 3651 } 3652 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3653 return 0; 3654 } 3655 3656 /* 3657 * Check whether it make sense to run itable init. thread or not. 3658 * If there is at least one uninitialized inode table, return 3659 * corresponding group number, else the loop goes through all 3660 * groups and return total number of groups. 3661 */ 3662 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3663 { 3664 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3665 struct ext4_group_desc *gdp = NULL; 3666 3667 if (!ext4_has_group_desc_csum(sb)) 3668 return ngroups; 3669 3670 for (group = 0; group < ngroups; group++) { 3671 gdp = ext4_get_group_desc(sb, group, NULL); 3672 if (!gdp) 3673 continue; 3674 3675 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3676 break; 3677 } 3678 3679 return group; 3680 } 3681 3682 static int ext4_li_info_new(void) 3683 { 3684 struct ext4_lazy_init *eli = NULL; 3685 3686 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3687 if (!eli) 3688 return -ENOMEM; 3689 3690 INIT_LIST_HEAD(&eli->li_request_list); 3691 mutex_init(&eli->li_list_mtx); 3692 3693 eli->li_state |= EXT4_LAZYINIT_QUIT; 3694 3695 ext4_li_info = eli; 3696 3697 return 0; 3698 } 3699 3700 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3701 ext4_group_t start) 3702 { 3703 struct ext4_li_request *elr; 3704 3705 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3706 if (!elr) 3707 return NULL; 3708 3709 elr->lr_super = sb; 3710 elr->lr_first_not_zeroed = start; 3711 if (test_opt(sb, PREFETCH_BLOCK_BITMAPS)) 3712 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3713 else { 3714 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3715 elr->lr_next_group = start; 3716 } 3717 3718 /* 3719 * Randomize first schedule time of the request to 3720 * spread the inode table initialization requests 3721 * better. 3722 */ 3723 elr->lr_next_sched = jiffies + (prandom_u32() % 3724 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3725 return elr; 3726 } 3727 3728 int ext4_register_li_request(struct super_block *sb, 3729 ext4_group_t first_not_zeroed) 3730 { 3731 struct ext4_sb_info *sbi = EXT4_SB(sb); 3732 struct ext4_li_request *elr = NULL; 3733 ext4_group_t ngroups = sbi->s_groups_count; 3734 int ret = 0; 3735 3736 mutex_lock(&ext4_li_mtx); 3737 if (sbi->s_li_request != NULL) { 3738 /* 3739 * Reset timeout so it can be computed again, because 3740 * s_li_wait_mult might have changed. 3741 */ 3742 sbi->s_li_request->lr_timeout = 0; 3743 goto out; 3744 } 3745 3746 if (!test_opt(sb, PREFETCH_BLOCK_BITMAPS) && 3747 (first_not_zeroed == ngroups || sb_rdonly(sb) || 3748 !test_opt(sb, INIT_INODE_TABLE))) 3749 goto out; 3750 3751 elr = ext4_li_request_new(sb, first_not_zeroed); 3752 if (!elr) { 3753 ret = -ENOMEM; 3754 goto out; 3755 } 3756 3757 if (NULL == ext4_li_info) { 3758 ret = ext4_li_info_new(); 3759 if (ret) 3760 goto out; 3761 } 3762 3763 mutex_lock(&ext4_li_info->li_list_mtx); 3764 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3765 mutex_unlock(&ext4_li_info->li_list_mtx); 3766 3767 sbi->s_li_request = elr; 3768 /* 3769 * set elr to NULL here since it has been inserted to 3770 * the request_list and the removal and free of it is 3771 * handled by ext4_clear_request_list from now on. 3772 */ 3773 elr = NULL; 3774 3775 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3776 ret = ext4_run_lazyinit_thread(); 3777 if (ret) 3778 goto out; 3779 } 3780 out: 3781 mutex_unlock(&ext4_li_mtx); 3782 if (ret) 3783 kfree(elr); 3784 return ret; 3785 } 3786 3787 /* 3788 * We do not need to lock anything since this is called on 3789 * module unload. 3790 */ 3791 static void ext4_destroy_lazyinit_thread(void) 3792 { 3793 /* 3794 * If thread exited earlier 3795 * there's nothing to be done. 3796 */ 3797 if (!ext4_li_info || !ext4_lazyinit_task) 3798 return; 3799 3800 kthread_stop(ext4_lazyinit_task); 3801 } 3802 3803 static int set_journal_csum_feature_set(struct super_block *sb) 3804 { 3805 int ret = 1; 3806 int compat, incompat; 3807 struct ext4_sb_info *sbi = EXT4_SB(sb); 3808 3809 if (ext4_has_metadata_csum(sb)) { 3810 /* journal checksum v3 */ 3811 compat = 0; 3812 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3813 } else { 3814 /* journal checksum v1 */ 3815 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3816 incompat = 0; 3817 } 3818 3819 jbd2_journal_clear_features(sbi->s_journal, 3820 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3821 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3822 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3823 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3824 ret = jbd2_journal_set_features(sbi->s_journal, 3825 compat, 0, 3826 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3827 incompat); 3828 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3829 ret = jbd2_journal_set_features(sbi->s_journal, 3830 compat, 0, 3831 incompat); 3832 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3833 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3834 } else { 3835 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3836 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3837 } 3838 3839 return ret; 3840 } 3841 3842 /* 3843 * Note: calculating the overhead so we can be compatible with 3844 * historical BSD practice is quite difficult in the face of 3845 * clusters/bigalloc. This is because multiple metadata blocks from 3846 * different block group can end up in the same allocation cluster. 3847 * Calculating the exact overhead in the face of clustered allocation 3848 * requires either O(all block bitmaps) in memory or O(number of block 3849 * groups**2) in time. We will still calculate the superblock for 3850 * older file systems --- and if we come across with a bigalloc file 3851 * system with zero in s_overhead_clusters the estimate will be close to 3852 * correct especially for very large cluster sizes --- but for newer 3853 * file systems, it's better to calculate this figure once at mkfs 3854 * time, and store it in the superblock. If the superblock value is 3855 * present (even for non-bigalloc file systems), we will use it. 3856 */ 3857 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3858 char *buf) 3859 { 3860 struct ext4_sb_info *sbi = EXT4_SB(sb); 3861 struct ext4_group_desc *gdp; 3862 ext4_fsblk_t first_block, last_block, b; 3863 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3864 int s, j, count = 0; 3865 3866 if (!ext4_has_feature_bigalloc(sb)) 3867 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3868 sbi->s_itb_per_group + 2); 3869 3870 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3871 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3872 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3873 for (i = 0; i < ngroups; i++) { 3874 gdp = ext4_get_group_desc(sb, i, NULL); 3875 b = ext4_block_bitmap(sb, gdp); 3876 if (b >= first_block && b <= last_block) { 3877 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3878 count++; 3879 } 3880 b = ext4_inode_bitmap(sb, gdp); 3881 if (b >= first_block && b <= last_block) { 3882 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3883 count++; 3884 } 3885 b = ext4_inode_table(sb, gdp); 3886 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3887 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3888 int c = EXT4_B2C(sbi, b - first_block); 3889 ext4_set_bit(c, buf); 3890 count++; 3891 } 3892 if (i != grp) 3893 continue; 3894 s = 0; 3895 if (ext4_bg_has_super(sb, grp)) { 3896 ext4_set_bit(s++, buf); 3897 count++; 3898 } 3899 j = ext4_bg_num_gdb(sb, grp); 3900 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3901 ext4_error(sb, "Invalid number of block group " 3902 "descriptor blocks: %d", j); 3903 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3904 } 3905 count += j; 3906 for (; j > 0; j--) 3907 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3908 } 3909 if (!count) 3910 return 0; 3911 return EXT4_CLUSTERS_PER_GROUP(sb) - 3912 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3913 } 3914 3915 /* 3916 * Compute the overhead and stash it in sbi->s_overhead 3917 */ 3918 int ext4_calculate_overhead(struct super_block *sb) 3919 { 3920 struct ext4_sb_info *sbi = EXT4_SB(sb); 3921 struct ext4_super_block *es = sbi->s_es; 3922 struct inode *j_inode; 3923 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3924 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3925 ext4_fsblk_t overhead = 0; 3926 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3927 3928 if (!buf) 3929 return -ENOMEM; 3930 3931 /* 3932 * Compute the overhead (FS structures). This is constant 3933 * for a given filesystem unless the number of block groups 3934 * changes so we cache the previous value until it does. 3935 */ 3936 3937 /* 3938 * All of the blocks before first_data_block are overhead 3939 */ 3940 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3941 3942 /* 3943 * Add the overhead found in each block group 3944 */ 3945 for (i = 0; i < ngroups; i++) { 3946 int blks; 3947 3948 blks = count_overhead(sb, i, buf); 3949 overhead += blks; 3950 if (blks) 3951 memset(buf, 0, PAGE_SIZE); 3952 cond_resched(); 3953 } 3954 3955 /* 3956 * Add the internal journal blocks whether the journal has been 3957 * loaded or not 3958 */ 3959 if (sbi->s_journal && !sbi->s_journal_bdev) 3960 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 3961 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 3962 /* j_inum for internal journal is non-zero */ 3963 j_inode = ext4_get_journal_inode(sb, j_inum); 3964 if (j_inode) { 3965 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3966 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3967 iput(j_inode); 3968 } else { 3969 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3970 } 3971 } 3972 sbi->s_overhead = overhead; 3973 smp_wmb(); 3974 free_page((unsigned long) buf); 3975 return 0; 3976 } 3977 3978 static void ext4_set_resv_clusters(struct super_block *sb) 3979 { 3980 ext4_fsblk_t resv_clusters; 3981 struct ext4_sb_info *sbi = EXT4_SB(sb); 3982 3983 /* 3984 * There's no need to reserve anything when we aren't using extents. 3985 * The space estimates are exact, there are no unwritten extents, 3986 * hole punching doesn't need new metadata... This is needed especially 3987 * to keep ext2/3 backward compatibility. 3988 */ 3989 if (!ext4_has_feature_extents(sb)) 3990 return; 3991 /* 3992 * By default we reserve 2% or 4096 clusters, whichever is smaller. 3993 * This should cover the situations where we can not afford to run 3994 * out of space like for example punch hole, or converting 3995 * unwritten extents in delalloc path. In most cases such 3996 * allocation would require 1, or 2 blocks, higher numbers are 3997 * very rare. 3998 */ 3999 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4000 sbi->s_cluster_bits); 4001 4002 do_div(resv_clusters, 50); 4003 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4004 4005 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4006 } 4007 4008 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 4009 { 4010 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 4011 char *orig_data = kstrdup(data, GFP_KERNEL); 4012 struct buffer_head *bh, **group_desc; 4013 struct ext4_super_block *es = NULL; 4014 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4015 struct flex_groups **flex_groups; 4016 ext4_fsblk_t block; 4017 ext4_fsblk_t sb_block = get_sb_block(&data); 4018 ext4_fsblk_t logical_sb_block; 4019 unsigned long offset = 0; 4020 unsigned long journal_devnum = 0; 4021 unsigned long def_mount_opts; 4022 struct inode *root; 4023 const char *descr; 4024 int ret = -ENOMEM; 4025 int blocksize, clustersize; 4026 unsigned int db_count; 4027 unsigned int i; 4028 int needs_recovery, has_huge_files; 4029 __u64 blocks_count; 4030 int err = 0; 4031 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 4032 ext4_group_t first_not_zeroed; 4033 4034 if ((data && !orig_data) || !sbi) 4035 goto out_free_base; 4036 4037 sbi->s_daxdev = dax_dev; 4038 sbi->s_blockgroup_lock = 4039 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4040 if (!sbi->s_blockgroup_lock) 4041 goto out_free_base; 4042 4043 sb->s_fs_info = sbi; 4044 sbi->s_sb = sb; 4045 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 4046 sbi->s_sb_block = sb_block; 4047 if (sb->s_bdev->bd_part) 4048 sbi->s_sectors_written_start = 4049 part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]); 4050 4051 /* Cleanup superblock name */ 4052 strreplace(sb->s_id, '/', '!'); 4053 4054 /* -EINVAL is default */ 4055 ret = -EINVAL; 4056 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4057 if (!blocksize) { 4058 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4059 goto out_fail; 4060 } 4061 4062 /* 4063 * The ext4 superblock will not be buffer aligned for other than 1kB 4064 * block sizes. We need to calculate the offset from buffer start. 4065 */ 4066 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4067 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4068 offset = do_div(logical_sb_block, blocksize); 4069 } else { 4070 logical_sb_block = sb_block; 4071 } 4072 4073 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4074 if (IS_ERR(bh)) { 4075 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4076 ret = PTR_ERR(bh); 4077 bh = NULL; 4078 goto out_fail; 4079 } 4080 /* 4081 * Note: s_es must be initialized as soon as possible because 4082 * some ext4 macro-instructions depend on its value 4083 */ 4084 es = (struct ext4_super_block *) (bh->b_data + offset); 4085 sbi->s_es = es; 4086 sb->s_magic = le16_to_cpu(es->s_magic); 4087 if (sb->s_magic != EXT4_SUPER_MAGIC) 4088 goto cantfind_ext4; 4089 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 4090 4091 /* Warn if metadata_csum and gdt_csum are both set. */ 4092 if (ext4_has_feature_metadata_csum(sb) && 4093 ext4_has_feature_gdt_csum(sb)) 4094 ext4_warning(sb, "metadata_csum and uninit_bg are " 4095 "redundant flags; please run fsck."); 4096 4097 /* Check for a known checksum algorithm */ 4098 if (!ext4_verify_csum_type(sb, es)) { 4099 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4100 "unknown checksum algorithm."); 4101 silent = 1; 4102 goto cantfind_ext4; 4103 } 4104 4105 /* Load the checksum driver */ 4106 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4107 if (IS_ERR(sbi->s_chksum_driver)) { 4108 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4109 ret = PTR_ERR(sbi->s_chksum_driver); 4110 sbi->s_chksum_driver = NULL; 4111 goto failed_mount; 4112 } 4113 4114 /* Check superblock checksum */ 4115 if (!ext4_superblock_csum_verify(sb, es)) { 4116 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4117 "invalid superblock checksum. Run e2fsck?"); 4118 silent = 1; 4119 ret = -EFSBADCRC; 4120 goto cantfind_ext4; 4121 } 4122 4123 /* Precompute checksum seed for all metadata */ 4124 if (ext4_has_feature_csum_seed(sb)) 4125 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4126 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4127 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4128 sizeof(es->s_uuid)); 4129 4130 /* Set defaults before we parse the mount options */ 4131 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4132 set_opt(sb, INIT_INODE_TABLE); 4133 if (def_mount_opts & EXT4_DEFM_DEBUG) 4134 set_opt(sb, DEBUG); 4135 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4136 set_opt(sb, GRPID); 4137 if (def_mount_opts & EXT4_DEFM_UID16) 4138 set_opt(sb, NO_UID32); 4139 /* xattr user namespace & acls are now defaulted on */ 4140 set_opt(sb, XATTR_USER); 4141 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4142 set_opt(sb, POSIX_ACL); 4143 #endif 4144 if (ext4_has_feature_fast_commit(sb)) 4145 set_opt2(sb, JOURNAL_FAST_COMMIT); 4146 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4147 if (ext4_has_metadata_csum(sb)) 4148 set_opt(sb, JOURNAL_CHECKSUM); 4149 4150 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4151 set_opt(sb, JOURNAL_DATA); 4152 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4153 set_opt(sb, ORDERED_DATA); 4154 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4155 set_opt(sb, WRITEBACK_DATA); 4156 4157 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 4158 set_opt(sb, ERRORS_PANIC); 4159 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 4160 set_opt(sb, ERRORS_CONT); 4161 else 4162 set_opt(sb, ERRORS_RO); 4163 /* block_validity enabled by default; disable with noblock_validity */ 4164 set_opt(sb, BLOCK_VALIDITY); 4165 if (def_mount_opts & EXT4_DEFM_DISCARD) 4166 set_opt(sb, DISCARD); 4167 4168 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 4169 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 4170 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 4171 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 4172 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 4173 4174 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4175 set_opt(sb, BARRIER); 4176 4177 /* 4178 * enable delayed allocation by default 4179 * Use -o nodelalloc to turn it off 4180 */ 4181 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4182 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4183 set_opt(sb, DELALLOC); 4184 4185 /* 4186 * set default s_li_wait_mult for lazyinit, for the case there is 4187 * no mount option specified. 4188 */ 4189 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 4190 4191 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4192 4193 if (blocksize == PAGE_SIZE) 4194 set_opt(sb, DIOREAD_NOLOCK); 4195 4196 if (blocksize < EXT4_MIN_BLOCK_SIZE || 4197 blocksize > EXT4_MAX_BLOCK_SIZE) { 4198 ext4_msg(sb, KERN_ERR, 4199 "Unsupported filesystem blocksize %d (%d log_block_size)", 4200 blocksize, le32_to_cpu(es->s_log_block_size)); 4201 goto failed_mount; 4202 } 4203 4204 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4205 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4206 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4207 } else { 4208 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4209 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4210 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4211 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4212 sbi->s_first_ino); 4213 goto failed_mount; 4214 } 4215 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4216 (!is_power_of_2(sbi->s_inode_size)) || 4217 (sbi->s_inode_size > blocksize)) { 4218 ext4_msg(sb, KERN_ERR, 4219 "unsupported inode size: %d", 4220 sbi->s_inode_size); 4221 ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize); 4222 goto failed_mount; 4223 } 4224 /* 4225 * i_atime_extra is the last extra field available for 4226 * [acm]times in struct ext4_inode. Checking for that 4227 * field should suffice to ensure we have extra space 4228 * for all three. 4229 */ 4230 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4231 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4232 sb->s_time_gran = 1; 4233 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4234 } else { 4235 sb->s_time_gran = NSEC_PER_SEC; 4236 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4237 } 4238 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4239 } 4240 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4241 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4242 EXT4_GOOD_OLD_INODE_SIZE; 4243 if (ext4_has_feature_extra_isize(sb)) { 4244 unsigned v, max = (sbi->s_inode_size - 4245 EXT4_GOOD_OLD_INODE_SIZE); 4246 4247 v = le16_to_cpu(es->s_want_extra_isize); 4248 if (v > max) { 4249 ext4_msg(sb, KERN_ERR, 4250 "bad s_want_extra_isize: %d", v); 4251 goto failed_mount; 4252 } 4253 if (sbi->s_want_extra_isize < v) 4254 sbi->s_want_extra_isize = v; 4255 4256 v = le16_to_cpu(es->s_min_extra_isize); 4257 if (v > max) { 4258 ext4_msg(sb, KERN_ERR, 4259 "bad s_min_extra_isize: %d", v); 4260 goto failed_mount; 4261 } 4262 if (sbi->s_want_extra_isize < v) 4263 sbi->s_want_extra_isize = v; 4264 } 4265 } 4266 4267 if (sbi->s_es->s_mount_opts[0]) { 4268 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 4269 sizeof(sbi->s_es->s_mount_opts), 4270 GFP_KERNEL); 4271 if (!s_mount_opts) 4272 goto failed_mount; 4273 if (!parse_options(s_mount_opts, sb, &journal_devnum, 4274 &journal_ioprio, 0)) { 4275 ext4_msg(sb, KERN_WARNING, 4276 "failed to parse options in superblock: %s", 4277 s_mount_opts); 4278 } 4279 kfree(s_mount_opts); 4280 } 4281 sbi->s_def_mount_opt = sbi->s_mount_opt; 4282 if (!parse_options((char *) data, sb, &journal_devnum, 4283 &journal_ioprio, 0)) 4284 goto failed_mount; 4285 4286 #ifdef CONFIG_UNICODE 4287 if (ext4_has_feature_casefold(sb) && !sb->s_encoding) { 4288 const struct ext4_sb_encodings *encoding_info; 4289 struct unicode_map *encoding; 4290 __u16 encoding_flags; 4291 4292 if (ext4_has_feature_encrypt(sb)) { 4293 ext4_msg(sb, KERN_ERR, 4294 "Can't mount with encoding and encryption"); 4295 goto failed_mount; 4296 } 4297 4298 if (ext4_sb_read_encoding(es, &encoding_info, 4299 &encoding_flags)) { 4300 ext4_msg(sb, KERN_ERR, 4301 "Encoding requested by superblock is unknown"); 4302 goto failed_mount; 4303 } 4304 4305 encoding = utf8_load(encoding_info->version); 4306 if (IS_ERR(encoding)) { 4307 ext4_msg(sb, KERN_ERR, 4308 "can't mount with superblock charset: %s-%s " 4309 "not supported by the kernel. flags: 0x%x.", 4310 encoding_info->name, encoding_info->version, 4311 encoding_flags); 4312 goto failed_mount; 4313 } 4314 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4315 "%s-%s with flags 0x%hx", encoding_info->name, 4316 encoding_info->version?:"\b", encoding_flags); 4317 4318 sb->s_encoding = encoding; 4319 sb->s_encoding_flags = encoding_flags; 4320 } 4321 #endif 4322 4323 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4324 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n"); 4325 /* can't mount with both data=journal and dioread_nolock. */ 4326 clear_opt(sb, DIOREAD_NOLOCK); 4327 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4328 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4329 ext4_msg(sb, KERN_ERR, "can't mount with " 4330 "both data=journal and delalloc"); 4331 goto failed_mount; 4332 } 4333 if (test_opt(sb, DAX_ALWAYS)) { 4334 ext4_msg(sb, KERN_ERR, "can't mount with " 4335 "both data=journal and dax"); 4336 goto failed_mount; 4337 } 4338 if (ext4_has_feature_encrypt(sb)) { 4339 ext4_msg(sb, KERN_WARNING, 4340 "encrypted files will use data=ordered " 4341 "instead of data journaling mode"); 4342 } 4343 if (test_opt(sb, DELALLOC)) 4344 clear_opt(sb, DELALLOC); 4345 } else { 4346 sb->s_iflags |= SB_I_CGROUPWB; 4347 } 4348 4349 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 4350 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 4351 4352 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4353 (ext4_has_compat_features(sb) || 4354 ext4_has_ro_compat_features(sb) || 4355 ext4_has_incompat_features(sb))) 4356 ext4_msg(sb, KERN_WARNING, 4357 "feature flags set on rev 0 fs, " 4358 "running e2fsck is recommended"); 4359 4360 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4361 set_opt2(sb, HURD_COMPAT); 4362 if (ext4_has_feature_64bit(sb)) { 4363 ext4_msg(sb, KERN_ERR, 4364 "The Hurd can't support 64-bit file systems"); 4365 goto failed_mount; 4366 } 4367 4368 /* 4369 * ea_inode feature uses l_i_version field which is not 4370 * available in HURD_COMPAT mode. 4371 */ 4372 if (ext4_has_feature_ea_inode(sb)) { 4373 ext4_msg(sb, KERN_ERR, 4374 "ea_inode feature is not supported for Hurd"); 4375 goto failed_mount; 4376 } 4377 } 4378 4379 if (IS_EXT2_SB(sb)) { 4380 if (ext2_feature_set_ok(sb)) 4381 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4382 "using the ext4 subsystem"); 4383 else { 4384 /* 4385 * If we're probing be silent, if this looks like 4386 * it's actually an ext[34] filesystem. 4387 */ 4388 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4389 goto failed_mount; 4390 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4391 "to feature incompatibilities"); 4392 goto failed_mount; 4393 } 4394 } 4395 4396 if (IS_EXT3_SB(sb)) { 4397 if (ext3_feature_set_ok(sb)) 4398 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4399 "using the ext4 subsystem"); 4400 else { 4401 /* 4402 * If we're probing be silent, if this looks like 4403 * it's actually an ext4 filesystem. 4404 */ 4405 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4406 goto failed_mount; 4407 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4408 "to feature incompatibilities"); 4409 goto failed_mount; 4410 } 4411 } 4412 4413 /* 4414 * Check feature flags regardless of the revision level, since we 4415 * previously didn't change the revision level when setting the flags, 4416 * so there is a chance incompat flags are set on a rev 0 filesystem. 4417 */ 4418 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4419 goto failed_mount; 4420 4421 if (le32_to_cpu(es->s_log_block_size) > 4422 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4423 ext4_msg(sb, KERN_ERR, 4424 "Invalid log block size: %u", 4425 le32_to_cpu(es->s_log_block_size)); 4426 goto failed_mount; 4427 } 4428 if (le32_to_cpu(es->s_log_cluster_size) > 4429 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4430 ext4_msg(sb, KERN_ERR, 4431 "Invalid log cluster size: %u", 4432 le32_to_cpu(es->s_log_cluster_size)); 4433 goto failed_mount; 4434 } 4435 4436 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 4437 ext4_msg(sb, KERN_ERR, 4438 "Number of reserved GDT blocks insanely large: %d", 4439 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4440 goto failed_mount; 4441 } 4442 4443 if (bdev_dax_supported(sb->s_bdev, blocksize)) 4444 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4445 4446 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4447 if (ext4_has_feature_inline_data(sb)) { 4448 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4449 " that may contain inline data"); 4450 goto failed_mount; 4451 } 4452 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4453 ext4_msg(sb, KERN_ERR, 4454 "DAX unsupported by block device."); 4455 goto failed_mount; 4456 } 4457 } 4458 4459 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4460 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4461 es->s_encryption_level); 4462 goto failed_mount; 4463 } 4464 4465 if (sb->s_blocksize != blocksize) { 4466 /* Validate the filesystem blocksize */ 4467 if (!sb_set_blocksize(sb, blocksize)) { 4468 ext4_msg(sb, KERN_ERR, "bad block size %d", 4469 blocksize); 4470 goto failed_mount; 4471 } 4472 4473 brelse(bh); 4474 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4475 offset = do_div(logical_sb_block, blocksize); 4476 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4477 if (IS_ERR(bh)) { 4478 ext4_msg(sb, KERN_ERR, 4479 "Can't read superblock on 2nd try"); 4480 ret = PTR_ERR(bh); 4481 bh = NULL; 4482 goto failed_mount; 4483 } 4484 es = (struct ext4_super_block *)(bh->b_data + offset); 4485 sbi->s_es = es; 4486 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4487 ext4_msg(sb, KERN_ERR, 4488 "Magic mismatch, very weird!"); 4489 goto failed_mount; 4490 } 4491 } 4492 4493 has_huge_files = ext4_has_feature_huge_file(sb); 4494 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4495 has_huge_files); 4496 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4497 4498 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4499 if (ext4_has_feature_64bit(sb)) { 4500 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4501 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4502 !is_power_of_2(sbi->s_desc_size)) { 4503 ext4_msg(sb, KERN_ERR, 4504 "unsupported descriptor size %lu", 4505 sbi->s_desc_size); 4506 goto failed_mount; 4507 } 4508 } else 4509 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4510 4511 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4512 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4513 4514 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4515 if (sbi->s_inodes_per_block == 0) 4516 goto cantfind_ext4; 4517 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4518 sbi->s_inodes_per_group > blocksize * 8) { 4519 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4520 sbi->s_inodes_per_group); 4521 goto failed_mount; 4522 } 4523 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4524 sbi->s_inodes_per_block; 4525 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4526 sbi->s_sbh = bh; 4527 sbi->s_mount_state = le16_to_cpu(es->s_state); 4528 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4529 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4530 4531 for (i = 0; i < 4; i++) 4532 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4533 sbi->s_def_hash_version = es->s_def_hash_version; 4534 if (ext4_has_feature_dir_index(sb)) { 4535 i = le32_to_cpu(es->s_flags); 4536 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4537 sbi->s_hash_unsigned = 3; 4538 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4539 #ifdef __CHAR_UNSIGNED__ 4540 if (!sb_rdonly(sb)) 4541 es->s_flags |= 4542 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4543 sbi->s_hash_unsigned = 3; 4544 #else 4545 if (!sb_rdonly(sb)) 4546 es->s_flags |= 4547 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4548 #endif 4549 } 4550 } 4551 4552 /* Handle clustersize */ 4553 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4554 if (ext4_has_feature_bigalloc(sb)) { 4555 if (clustersize < blocksize) { 4556 ext4_msg(sb, KERN_ERR, 4557 "cluster size (%d) smaller than " 4558 "block size (%d)", clustersize, blocksize); 4559 goto failed_mount; 4560 } 4561 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4562 le32_to_cpu(es->s_log_block_size); 4563 sbi->s_clusters_per_group = 4564 le32_to_cpu(es->s_clusters_per_group); 4565 if (sbi->s_clusters_per_group > blocksize * 8) { 4566 ext4_msg(sb, KERN_ERR, 4567 "#clusters per group too big: %lu", 4568 sbi->s_clusters_per_group); 4569 goto failed_mount; 4570 } 4571 if (sbi->s_blocks_per_group != 4572 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4573 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4574 "clusters per group (%lu) inconsistent", 4575 sbi->s_blocks_per_group, 4576 sbi->s_clusters_per_group); 4577 goto failed_mount; 4578 } 4579 } else { 4580 if (clustersize != blocksize) { 4581 ext4_msg(sb, KERN_ERR, 4582 "fragment/cluster size (%d) != " 4583 "block size (%d)", clustersize, blocksize); 4584 goto failed_mount; 4585 } 4586 if (sbi->s_blocks_per_group > blocksize * 8) { 4587 ext4_msg(sb, KERN_ERR, 4588 "#blocks per group too big: %lu", 4589 sbi->s_blocks_per_group); 4590 goto failed_mount; 4591 } 4592 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4593 sbi->s_cluster_bits = 0; 4594 } 4595 sbi->s_cluster_ratio = clustersize / blocksize; 4596 4597 /* Do we have standard group size of clustersize * 8 blocks ? */ 4598 if (sbi->s_blocks_per_group == clustersize << 3) 4599 set_opt2(sb, STD_GROUP_SIZE); 4600 4601 /* 4602 * Test whether we have more sectors than will fit in sector_t, 4603 * and whether the max offset is addressable by the page cache. 4604 */ 4605 err = generic_check_addressable(sb->s_blocksize_bits, 4606 ext4_blocks_count(es)); 4607 if (err) { 4608 ext4_msg(sb, KERN_ERR, "filesystem" 4609 " too large to mount safely on this system"); 4610 goto failed_mount; 4611 } 4612 4613 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4614 goto cantfind_ext4; 4615 4616 /* check blocks count against device size */ 4617 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4618 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4619 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4620 "exceeds size of device (%llu blocks)", 4621 ext4_blocks_count(es), blocks_count); 4622 goto failed_mount; 4623 } 4624 4625 /* 4626 * It makes no sense for the first data block to be beyond the end 4627 * of the filesystem. 4628 */ 4629 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4630 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4631 "block %u is beyond end of filesystem (%llu)", 4632 le32_to_cpu(es->s_first_data_block), 4633 ext4_blocks_count(es)); 4634 goto failed_mount; 4635 } 4636 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4637 (sbi->s_cluster_ratio == 1)) { 4638 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4639 "block is 0 with a 1k block and cluster size"); 4640 goto failed_mount; 4641 } 4642 4643 blocks_count = (ext4_blocks_count(es) - 4644 le32_to_cpu(es->s_first_data_block) + 4645 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4646 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4647 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4648 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4649 "(block count %llu, first data block %u, " 4650 "blocks per group %lu)", blocks_count, 4651 ext4_blocks_count(es), 4652 le32_to_cpu(es->s_first_data_block), 4653 EXT4_BLOCKS_PER_GROUP(sb)); 4654 goto failed_mount; 4655 } 4656 sbi->s_groups_count = blocks_count; 4657 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4658 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4659 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4660 le32_to_cpu(es->s_inodes_count)) { 4661 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4662 le32_to_cpu(es->s_inodes_count), 4663 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4664 ret = -EINVAL; 4665 goto failed_mount; 4666 } 4667 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4668 EXT4_DESC_PER_BLOCK(sb); 4669 if (ext4_has_feature_meta_bg(sb)) { 4670 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4671 ext4_msg(sb, KERN_WARNING, 4672 "first meta block group too large: %u " 4673 "(group descriptor block count %u)", 4674 le32_to_cpu(es->s_first_meta_bg), db_count); 4675 goto failed_mount; 4676 } 4677 } 4678 rcu_assign_pointer(sbi->s_group_desc, 4679 kvmalloc_array(db_count, 4680 sizeof(struct buffer_head *), 4681 GFP_KERNEL)); 4682 if (sbi->s_group_desc == NULL) { 4683 ext4_msg(sb, KERN_ERR, "not enough memory"); 4684 ret = -ENOMEM; 4685 goto failed_mount; 4686 } 4687 4688 bgl_lock_init(sbi->s_blockgroup_lock); 4689 4690 /* Pre-read the descriptors into the buffer cache */ 4691 for (i = 0; i < db_count; i++) { 4692 block = descriptor_loc(sb, logical_sb_block, i); 4693 ext4_sb_breadahead_unmovable(sb, block); 4694 } 4695 4696 for (i = 0; i < db_count; i++) { 4697 struct buffer_head *bh; 4698 4699 block = descriptor_loc(sb, logical_sb_block, i); 4700 bh = ext4_sb_bread_unmovable(sb, block); 4701 if (IS_ERR(bh)) { 4702 ext4_msg(sb, KERN_ERR, 4703 "can't read group descriptor %d", i); 4704 db_count = i; 4705 ret = PTR_ERR(bh); 4706 bh = NULL; 4707 goto failed_mount2; 4708 } 4709 rcu_read_lock(); 4710 rcu_dereference(sbi->s_group_desc)[i] = bh; 4711 rcu_read_unlock(); 4712 } 4713 sbi->s_gdb_count = db_count; 4714 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4715 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4716 ret = -EFSCORRUPTED; 4717 goto failed_mount2; 4718 } 4719 4720 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4721 4722 /* Register extent status tree shrinker */ 4723 if (ext4_es_register_shrinker(sbi)) 4724 goto failed_mount3; 4725 4726 sbi->s_stripe = ext4_get_stripe_size(sbi); 4727 sbi->s_extent_max_zeroout_kb = 32; 4728 4729 /* 4730 * set up enough so that it can read an inode 4731 */ 4732 sb->s_op = &ext4_sops; 4733 sb->s_export_op = &ext4_export_ops; 4734 sb->s_xattr = ext4_xattr_handlers; 4735 #ifdef CONFIG_FS_ENCRYPTION 4736 sb->s_cop = &ext4_cryptops; 4737 #endif 4738 #ifdef CONFIG_FS_VERITY 4739 sb->s_vop = &ext4_verityops; 4740 #endif 4741 #ifdef CONFIG_QUOTA 4742 sb->dq_op = &ext4_quota_operations; 4743 if (ext4_has_feature_quota(sb)) 4744 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4745 else 4746 sb->s_qcop = &ext4_qctl_operations; 4747 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4748 #endif 4749 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4750 4751 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4752 mutex_init(&sbi->s_orphan_lock); 4753 4754 /* Initialize fast commit stuff */ 4755 atomic_set(&sbi->s_fc_subtid, 0); 4756 atomic_set(&sbi->s_fc_ineligible_updates, 0); 4757 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4758 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4759 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4760 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4761 sbi->s_fc_bytes = 0; 4762 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4763 ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING); 4764 spin_lock_init(&sbi->s_fc_lock); 4765 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4766 sbi->s_fc_replay_state.fc_regions = NULL; 4767 sbi->s_fc_replay_state.fc_regions_size = 0; 4768 sbi->s_fc_replay_state.fc_regions_used = 0; 4769 sbi->s_fc_replay_state.fc_regions_valid = 0; 4770 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4771 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4772 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4773 4774 sb->s_root = NULL; 4775 4776 needs_recovery = (es->s_last_orphan != 0 || 4777 ext4_has_feature_journal_needs_recovery(sb)); 4778 4779 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4780 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4781 goto failed_mount3a; 4782 4783 /* 4784 * The first inode we look at is the journal inode. Don't try 4785 * root first: it may be modified in the journal! 4786 */ 4787 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4788 err = ext4_load_journal(sb, es, journal_devnum); 4789 if (err) 4790 goto failed_mount3a; 4791 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4792 ext4_has_feature_journal_needs_recovery(sb)) { 4793 ext4_msg(sb, KERN_ERR, "required journal recovery " 4794 "suppressed and not mounted read-only"); 4795 goto failed_mount_wq; 4796 } else { 4797 /* Nojournal mode, all journal mount options are illegal */ 4798 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4799 ext4_msg(sb, KERN_ERR, "can't mount with " 4800 "journal_checksum, fs mounted w/o journal"); 4801 goto failed_mount_wq; 4802 } 4803 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4804 ext4_msg(sb, KERN_ERR, "can't mount with " 4805 "journal_async_commit, fs mounted w/o journal"); 4806 goto failed_mount_wq; 4807 } 4808 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4809 ext4_msg(sb, KERN_ERR, "can't mount with " 4810 "commit=%lu, fs mounted w/o journal", 4811 sbi->s_commit_interval / HZ); 4812 goto failed_mount_wq; 4813 } 4814 if (EXT4_MOUNT_DATA_FLAGS & 4815 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4816 ext4_msg(sb, KERN_ERR, "can't mount with " 4817 "data=, fs mounted w/o journal"); 4818 goto failed_mount_wq; 4819 } 4820 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4821 clear_opt(sb, JOURNAL_CHECKSUM); 4822 clear_opt(sb, DATA_FLAGS); 4823 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4824 sbi->s_journal = NULL; 4825 needs_recovery = 0; 4826 goto no_journal; 4827 } 4828 4829 if (ext4_has_feature_64bit(sb) && 4830 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4831 JBD2_FEATURE_INCOMPAT_64BIT)) { 4832 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4833 goto failed_mount_wq; 4834 } 4835 4836 if (!set_journal_csum_feature_set(sb)) { 4837 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4838 "feature set"); 4839 goto failed_mount_wq; 4840 } 4841 4842 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4843 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4844 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4845 ext4_msg(sb, KERN_ERR, 4846 "Failed to set fast commit journal feature"); 4847 goto failed_mount_wq; 4848 } 4849 4850 /* We have now updated the journal if required, so we can 4851 * validate the data journaling mode. */ 4852 switch (test_opt(sb, DATA_FLAGS)) { 4853 case 0: 4854 /* No mode set, assume a default based on the journal 4855 * capabilities: ORDERED_DATA if the journal can 4856 * cope, else JOURNAL_DATA 4857 */ 4858 if (jbd2_journal_check_available_features 4859 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4860 set_opt(sb, ORDERED_DATA); 4861 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4862 } else { 4863 set_opt(sb, JOURNAL_DATA); 4864 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4865 } 4866 break; 4867 4868 case EXT4_MOUNT_ORDERED_DATA: 4869 case EXT4_MOUNT_WRITEBACK_DATA: 4870 if (!jbd2_journal_check_available_features 4871 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4872 ext4_msg(sb, KERN_ERR, "Journal does not support " 4873 "requested data journaling mode"); 4874 goto failed_mount_wq; 4875 } 4876 default: 4877 break; 4878 } 4879 4880 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4881 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4882 ext4_msg(sb, KERN_ERR, "can't mount with " 4883 "journal_async_commit in data=ordered mode"); 4884 goto failed_mount_wq; 4885 } 4886 4887 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 4888 4889 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback; 4890 sbi->s_journal->j_submit_inode_data_buffers = 4891 ext4_journal_submit_inode_data_buffers; 4892 sbi->s_journal->j_finish_inode_data_buffers = 4893 ext4_journal_finish_inode_data_buffers; 4894 4895 no_journal: 4896 if (!test_opt(sb, NO_MBCACHE)) { 4897 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4898 if (!sbi->s_ea_block_cache) { 4899 ext4_msg(sb, KERN_ERR, 4900 "Failed to create ea_block_cache"); 4901 goto failed_mount_wq; 4902 } 4903 4904 if (ext4_has_feature_ea_inode(sb)) { 4905 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4906 if (!sbi->s_ea_inode_cache) { 4907 ext4_msg(sb, KERN_ERR, 4908 "Failed to create ea_inode_cache"); 4909 goto failed_mount_wq; 4910 } 4911 } 4912 } 4913 4914 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4915 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4916 goto failed_mount_wq; 4917 } 4918 4919 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4920 !ext4_has_feature_encrypt(sb)) { 4921 ext4_set_feature_encrypt(sb); 4922 ext4_commit_super(sb, 1); 4923 } 4924 4925 /* 4926 * Get the # of file system overhead blocks from the 4927 * superblock if present. 4928 */ 4929 if (es->s_overhead_clusters) 4930 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4931 else { 4932 err = ext4_calculate_overhead(sb); 4933 if (err) 4934 goto failed_mount_wq; 4935 } 4936 4937 /* 4938 * The maximum number of concurrent works can be high and 4939 * concurrency isn't really necessary. Limit it to 1. 4940 */ 4941 EXT4_SB(sb)->rsv_conversion_wq = 4942 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4943 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4944 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4945 ret = -ENOMEM; 4946 goto failed_mount4; 4947 } 4948 4949 /* 4950 * The jbd2_journal_load will have done any necessary log recovery, 4951 * so we can safely mount the rest of the filesystem now. 4952 */ 4953 4954 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4955 if (IS_ERR(root)) { 4956 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4957 ret = PTR_ERR(root); 4958 root = NULL; 4959 goto failed_mount4; 4960 } 4961 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4962 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4963 iput(root); 4964 goto failed_mount4; 4965 } 4966 4967 #ifdef CONFIG_UNICODE 4968 if (sb->s_encoding) 4969 sb->s_d_op = &ext4_dentry_ops; 4970 #endif 4971 4972 sb->s_root = d_make_root(root); 4973 if (!sb->s_root) { 4974 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4975 ret = -ENOMEM; 4976 goto failed_mount4; 4977 } 4978 4979 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4980 if (ret == -EROFS) { 4981 sb->s_flags |= SB_RDONLY; 4982 ret = 0; 4983 } else if (ret) 4984 goto failed_mount4a; 4985 4986 ext4_set_resv_clusters(sb); 4987 4988 if (test_opt(sb, BLOCK_VALIDITY)) { 4989 err = ext4_setup_system_zone(sb); 4990 if (err) { 4991 ext4_msg(sb, KERN_ERR, "failed to initialize system " 4992 "zone (%d)", err); 4993 goto failed_mount4a; 4994 } 4995 } 4996 ext4_fc_replay_cleanup(sb); 4997 4998 ext4_ext_init(sb); 4999 err = ext4_mb_init(sb); 5000 if (err) { 5001 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5002 err); 5003 goto failed_mount5; 5004 } 5005 5006 block = ext4_count_free_clusters(sb); 5007 ext4_free_blocks_count_set(sbi->s_es, 5008 EXT4_C2B(sbi, block)); 5009 ext4_superblock_csum_set(sb); 5010 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5011 GFP_KERNEL); 5012 if (!err) { 5013 unsigned long freei = ext4_count_free_inodes(sb); 5014 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5015 ext4_superblock_csum_set(sb); 5016 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5017 GFP_KERNEL); 5018 } 5019 if (!err) 5020 err = percpu_counter_init(&sbi->s_dirs_counter, 5021 ext4_count_dirs(sb), GFP_KERNEL); 5022 if (!err) 5023 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5024 GFP_KERNEL); 5025 if (!err) 5026 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5027 5028 if (err) { 5029 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5030 goto failed_mount6; 5031 } 5032 5033 if (ext4_has_feature_flex_bg(sb)) 5034 if (!ext4_fill_flex_info(sb)) { 5035 ext4_msg(sb, KERN_ERR, 5036 "unable to initialize " 5037 "flex_bg meta info!"); 5038 goto failed_mount6; 5039 } 5040 5041 err = ext4_register_li_request(sb, first_not_zeroed); 5042 if (err) 5043 goto failed_mount6; 5044 5045 err = ext4_register_sysfs(sb); 5046 if (err) 5047 goto failed_mount7; 5048 5049 #ifdef CONFIG_QUOTA 5050 /* Enable quota usage during mount. */ 5051 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5052 err = ext4_enable_quotas(sb); 5053 if (err) 5054 goto failed_mount8; 5055 } 5056 #endif /* CONFIG_QUOTA */ 5057 5058 /* 5059 * Save the original bdev mapping's wb_err value which could be 5060 * used to detect the metadata async write error. 5061 */ 5062 spin_lock_init(&sbi->s_bdev_wb_lock); 5063 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5064 &sbi->s_bdev_wb_err); 5065 sb->s_bdev->bd_super = sb; 5066 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5067 ext4_orphan_cleanup(sb, es); 5068 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5069 if (needs_recovery) { 5070 ext4_msg(sb, KERN_INFO, "recovery complete"); 5071 err = ext4_mark_recovery_complete(sb, es); 5072 if (err) 5073 goto failed_mount8; 5074 } 5075 if (EXT4_SB(sb)->s_journal) { 5076 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5077 descr = " journalled data mode"; 5078 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5079 descr = " ordered data mode"; 5080 else 5081 descr = " writeback data mode"; 5082 } else 5083 descr = "out journal"; 5084 5085 if (test_opt(sb, DISCARD)) { 5086 struct request_queue *q = bdev_get_queue(sb->s_bdev); 5087 if (!blk_queue_discard(q)) 5088 ext4_msg(sb, KERN_WARNING, 5089 "mounting with \"discard\" option, but " 5090 "the device does not support discard"); 5091 } 5092 5093 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5094 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 5095 "Opts: %.*s%s%s", descr, 5096 (int) sizeof(sbi->s_es->s_mount_opts), 5097 sbi->s_es->s_mount_opts, 5098 *sbi->s_es->s_mount_opts ? "; " : "", orig_data); 5099 5100 if (es->s_error_count) 5101 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5102 5103 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5104 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5105 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5106 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5107 atomic_set(&sbi->s_warning_count, 0); 5108 atomic_set(&sbi->s_msg_count, 0); 5109 5110 kfree(orig_data); 5111 return 0; 5112 5113 cantfind_ext4: 5114 if (!silent) 5115 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5116 goto failed_mount; 5117 5118 failed_mount8: 5119 ext4_unregister_sysfs(sb); 5120 kobject_put(&sbi->s_kobj); 5121 failed_mount7: 5122 ext4_unregister_li_request(sb); 5123 failed_mount6: 5124 ext4_mb_release(sb); 5125 rcu_read_lock(); 5126 flex_groups = rcu_dereference(sbi->s_flex_groups); 5127 if (flex_groups) { 5128 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 5129 kvfree(flex_groups[i]); 5130 kvfree(flex_groups); 5131 } 5132 rcu_read_unlock(); 5133 percpu_counter_destroy(&sbi->s_freeclusters_counter); 5134 percpu_counter_destroy(&sbi->s_freeinodes_counter); 5135 percpu_counter_destroy(&sbi->s_dirs_counter); 5136 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 5137 percpu_free_rwsem(&sbi->s_writepages_rwsem); 5138 failed_mount5: 5139 ext4_ext_release(sb); 5140 ext4_release_system_zone(sb); 5141 failed_mount4a: 5142 dput(sb->s_root); 5143 sb->s_root = NULL; 5144 failed_mount4: 5145 ext4_msg(sb, KERN_ERR, "mount failed"); 5146 if (EXT4_SB(sb)->rsv_conversion_wq) 5147 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5148 failed_mount_wq: 5149 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5150 sbi->s_ea_inode_cache = NULL; 5151 5152 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5153 sbi->s_ea_block_cache = NULL; 5154 5155 if (sbi->s_journal) { 5156 jbd2_journal_destroy(sbi->s_journal); 5157 sbi->s_journal = NULL; 5158 } 5159 failed_mount3a: 5160 ext4_es_unregister_shrinker(sbi); 5161 failed_mount3: 5162 del_timer_sync(&sbi->s_err_report); 5163 if (sbi->s_mmp_tsk) 5164 kthread_stop(sbi->s_mmp_tsk); 5165 failed_mount2: 5166 rcu_read_lock(); 5167 group_desc = rcu_dereference(sbi->s_group_desc); 5168 for (i = 0; i < db_count; i++) 5169 brelse(group_desc[i]); 5170 kvfree(group_desc); 5171 rcu_read_unlock(); 5172 failed_mount: 5173 if (sbi->s_chksum_driver) 5174 crypto_free_shash(sbi->s_chksum_driver); 5175 5176 #ifdef CONFIG_UNICODE 5177 utf8_unload(sb->s_encoding); 5178 #endif 5179 5180 #ifdef CONFIG_QUOTA 5181 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5182 kfree(get_qf_name(sb, sbi, i)); 5183 #endif 5184 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5185 ext4_blkdev_remove(sbi); 5186 brelse(bh); 5187 out_fail: 5188 sb->s_fs_info = NULL; 5189 kfree(sbi->s_blockgroup_lock); 5190 out_free_base: 5191 kfree(sbi); 5192 kfree(orig_data); 5193 fs_put_dax(dax_dev); 5194 return err ? err : ret; 5195 } 5196 5197 /* 5198 * Setup any per-fs journal parameters now. We'll do this both on 5199 * initial mount, once the journal has been initialised but before we've 5200 * done any recovery; and again on any subsequent remount. 5201 */ 5202 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5203 { 5204 struct ext4_sb_info *sbi = EXT4_SB(sb); 5205 5206 journal->j_commit_interval = sbi->s_commit_interval; 5207 journal->j_min_batch_time = sbi->s_min_batch_time; 5208 journal->j_max_batch_time = sbi->s_max_batch_time; 5209 ext4_fc_init(sb, journal); 5210 5211 write_lock(&journal->j_state_lock); 5212 if (test_opt(sb, BARRIER)) 5213 journal->j_flags |= JBD2_BARRIER; 5214 else 5215 journal->j_flags &= ~JBD2_BARRIER; 5216 if (test_opt(sb, DATA_ERR_ABORT)) 5217 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5218 else 5219 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5220 write_unlock(&journal->j_state_lock); 5221 } 5222 5223 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5224 unsigned int journal_inum) 5225 { 5226 struct inode *journal_inode; 5227 5228 /* 5229 * Test for the existence of a valid inode on disk. Bad things 5230 * happen if we iget() an unused inode, as the subsequent iput() 5231 * will try to delete it. 5232 */ 5233 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5234 if (IS_ERR(journal_inode)) { 5235 ext4_msg(sb, KERN_ERR, "no journal found"); 5236 return NULL; 5237 } 5238 if (!journal_inode->i_nlink) { 5239 make_bad_inode(journal_inode); 5240 iput(journal_inode); 5241 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5242 return NULL; 5243 } 5244 5245 jbd_debug(2, "Journal inode found at %p: %lld bytes\n", 5246 journal_inode, journal_inode->i_size); 5247 if (!S_ISREG(journal_inode->i_mode)) { 5248 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5249 iput(journal_inode); 5250 return NULL; 5251 } 5252 return journal_inode; 5253 } 5254 5255 static journal_t *ext4_get_journal(struct super_block *sb, 5256 unsigned int journal_inum) 5257 { 5258 struct inode *journal_inode; 5259 journal_t *journal; 5260 5261 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5262 return NULL; 5263 5264 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5265 if (!journal_inode) 5266 return NULL; 5267 5268 journal = jbd2_journal_init_inode(journal_inode); 5269 if (!journal) { 5270 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5271 iput(journal_inode); 5272 return NULL; 5273 } 5274 journal->j_private = sb; 5275 ext4_init_journal_params(sb, journal); 5276 return journal; 5277 } 5278 5279 static journal_t *ext4_get_dev_journal(struct super_block *sb, 5280 dev_t j_dev) 5281 { 5282 struct buffer_head *bh; 5283 journal_t *journal; 5284 ext4_fsblk_t start; 5285 ext4_fsblk_t len; 5286 int hblock, blocksize; 5287 ext4_fsblk_t sb_block; 5288 unsigned long offset; 5289 struct ext4_super_block *es; 5290 struct block_device *bdev; 5291 5292 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5293 return NULL; 5294 5295 bdev = ext4_blkdev_get(j_dev, sb); 5296 if (bdev == NULL) 5297 return NULL; 5298 5299 blocksize = sb->s_blocksize; 5300 hblock = bdev_logical_block_size(bdev); 5301 if (blocksize < hblock) { 5302 ext4_msg(sb, KERN_ERR, 5303 "blocksize too small for journal device"); 5304 goto out_bdev; 5305 } 5306 5307 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5308 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5309 set_blocksize(bdev, blocksize); 5310 if (!(bh = __bread(bdev, sb_block, blocksize))) { 5311 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5312 "external journal"); 5313 goto out_bdev; 5314 } 5315 5316 es = (struct ext4_super_block *) (bh->b_data + offset); 5317 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5318 !(le32_to_cpu(es->s_feature_incompat) & 5319 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5320 ext4_msg(sb, KERN_ERR, "external journal has " 5321 "bad superblock"); 5322 brelse(bh); 5323 goto out_bdev; 5324 } 5325 5326 if ((le32_to_cpu(es->s_feature_ro_compat) & 5327 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5328 es->s_checksum != ext4_superblock_csum(sb, es)) { 5329 ext4_msg(sb, KERN_ERR, "external journal has " 5330 "corrupt superblock"); 5331 brelse(bh); 5332 goto out_bdev; 5333 } 5334 5335 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5336 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5337 brelse(bh); 5338 goto out_bdev; 5339 } 5340 5341 len = ext4_blocks_count(es); 5342 start = sb_block + 1; 5343 brelse(bh); /* we're done with the superblock */ 5344 5345 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 5346 start, len, blocksize); 5347 if (!journal) { 5348 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5349 goto out_bdev; 5350 } 5351 journal->j_private = sb; 5352 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) { 5353 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 5354 goto out_journal; 5355 } 5356 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5357 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5358 "user (unsupported) - %d", 5359 be32_to_cpu(journal->j_superblock->s_nr_users)); 5360 goto out_journal; 5361 } 5362 EXT4_SB(sb)->s_journal_bdev = bdev; 5363 ext4_init_journal_params(sb, journal); 5364 return journal; 5365 5366 out_journal: 5367 jbd2_journal_destroy(journal); 5368 out_bdev: 5369 ext4_blkdev_put(bdev); 5370 return NULL; 5371 } 5372 5373 static int ext4_load_journal(struct super_block *sb, 5374 struct ext4_super_block *es, 5375 unsigned long journal_devnum) 5376 { 5377 journal_t *journal; 5378 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5379 dev_t journal_dev; 5380 int err = 0; 5381 int really_read_only; 5382 int journal_dev_ro; 5383 5384 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5385 return -EFSCORRUPTED; 5386 5387 if (journal_devnum && 5388 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5389 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5390 "numbers have changed"); 5391 journal_dev = new_decode_dev(journal_devnum); 5392 } else 5393 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5394 5395 if (journal_inum && journal_dev) { 5396 ext4_msg(sb, KERN_ERR, 5397 "filesystem has both journal inode and journal device!"); 5398 return -EINVAL; 5399 } 5400 5401 if (journal_inum) { 5402 journal = ext4_get_journal(sb, journal_inum); 5403 if (!journal) 5404 return -EINVAL; 5405 } else { 5406 journal = ext4_get_dev_journal(sb, journal_dev); 5407 if (!journal) 5408 return -EINVAL; 5409 } 5410 5411 journal_dev_ro = bdev_read_only(journal->j_dev); 5412 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5413 5414 if (journal_dev_ro && !sb_rdonly(sb)) { 5415 ext4_msg(sb, KERN_ERR, 5416 "journal device read-only, try mounting with '-o ro'"); 5417 err = -EROFS; 5418 goto err_out; 5419 } 5420 5421 /* 5422 * Are we loading a blank journal or performing recovery after a 5423 * crash? For recovery, we need to check in advance whether we 5424 * can get read-write access to the device. 5425 */ 5426 if (ext4_has_feature_journal_needs_recovery(sb)) { 5427 if (sb_rdonly(sb)) { 5428 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5429 "required on readonly filesystem"); 5430 if (really_read_only) { 5431 ext4_msg(sb, KERN_ERR, "write access " 5432 "unavailable, cannot proceed " 5433 "(try mounting with noload)"); 5434 err = -EROFS; 5435 goto err_out; 5436 } 5437 ext4_msg(sb, KERN_INFO, "write access will " 5438 "be enabled during recovery"); 5439 } 5440 } 5441 5442 if (!(journal->j_flags & JBD2_BARRIER)) 5443 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5444 5445 if (!ext4_has_feature_journal_needs_recovery(sb)) 5446 err = jbd2_journal_wipe(journal, !really_read_only); 5447 if (!err) { 5448 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5449 if (save) 5450 memcpy(save, ((char *) es) + 5451 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 5452 err = jbd2_journal_load(journal); 5453 if (save) 5454 memcpy(((char *) es) + EXT4_S_ERR_START, 5455 save, EXT4_S_ERR_LEN); 5456 kfree(save); 5457 } 5458 5459 if (err) { 5460 ext4_msg(sb, KERN_ERR, "error loading journal"); 5461 goto err_out; 5462 } 5463 5464 EXT4_SB(sb)->s_journal = journal; 5465 err = ext4_clear_journal_err(sb, es); 5466 if (err) { 5467 EXT4_SB(sb)->s_journal = NULL; 5468 jbd2_journal_destroy(journal); 5469 return err; 5470 } 5471 5472 if (!really_read_only && journal_devnum && 5473 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5474 es->s_journal_dev = cpu_to_le32(journal_devnum); 5475 5476 /* Make sure we flush the recovery flag to disk. */ 5477 ext4_commit_super(sb, 1); 5478 } 5479 5480 return 0; 5481 5482 err_out: 5483 jbd2_journal_destroy(journal); 5484 return err; 5485 } 5486 5487 static int ext4_commit_super(struct super_block *sb, int sync) 5488 { 5489 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 5490 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 5491 int error = 0; 5492 5493 if (!sbh || block_device_ejected(sb)) 5494 return error; 5495 5496 /* 5497 * If the file system is mounted read-only, don't update the 5498 * superblock write time. This avoids updating the superblock 5499 * write time when we are mounting the root file system 5500 * read/only but we need to replay the journal; at that point, 5501 * for people who are east of GMT and who make their clock 5502 * tick in localtime for Windows bug-for-bug compatibility, 5503 * the clock is set in the future, and this will cause e2fsck 5504 * to complain and force a full file system check. 5505 */ 5506 if (!(sb->s_flags & SB_RDONLY)) 5507 ext4_update_tstamp(es, s_wtime); 5508 if (sb->s_bdev->bd_part) 5509 es->s_kbytes_written = 5510 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 5511 ((part_stat_read(sb->s_bdev->bd_part, 5512 sectors[STAT_WRITE]) - 5513 EXT4_SB(sb)->s_sectors_written_start) >> 1)); 5514 else 5515 es->s_kbytes_written = 5516 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written); 5517 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter)) 5518 ext4_free_blocks_count_set(es, 5519 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive( 5520 &EXT4_SB(sb)->s_freeclusters_counter))); 5521 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter)) 5522 es->s_free_inodes_count = 5523 cpu_to_le32(percpu_counter_sum_positive( 5524 &EXT4_SB(sb)->s_freeinodes_counter)); 5525 BUFFER_TRACE(sbh, "marking dirty"); 5526 ext4_superblock_csum_set(sb); 5527 if (sync) 5528 lock_buffer(sbh); 5529 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 5530 /* 5531 * Oh, dear. A previous attempt to write the 5532 * superblock failed. This could happen because the 5533 * USB device was yanked out. Or it could happen to 5534 * be a transient write error and maybe the block will 5535 * be remapped. Nothing we can do but to retry the 5536 * write and hope for the best. 5537 */ 5538 ext4_msg(sb, KERN_ERR, "previous I/O error to " 5539 "superblock detected"); 5540 clear_buffer_write_io_error(sbh); 5541 set_buffer_uptodate(sbh); 5542 } 5543 mark_buffer_dirty(sbh); 5544 if (sync) { 5545 unlock_buffer(sbh); 5546 error = __sync_dirty_buffer(sbh, 5547 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0)); 5548 if (buffer_write_io_error(sbh)) { 5549 ext4_msg(sb, KERN_ERR, "I/O error while writing " 5550 "superblock"); 5551 clear_buffer_write_io_error(sbh); 5552 set_buffer_uptodate(sbh); 5553 } 5554 } 5555 return error; 5556 } 5557 5558 /* 5559 * Have we just finished recovery? If so, and if we are mounting (or 5560 * remounting) the filesystem readonly, then we will end up with a 5561 * consistent fs on disk. Record that fact. 5562 */ 5563 static int ext4_mark_recovery_complete(struct super_block *sb, 5564 struct ext4_super_block *es) 5565 { 5566 int err; 5567 journal_t *journal = EXT4_SB(sb)->s_journal; 5568 5569 if (!ext4_has_feature_journal(sb)) { 5570 if (journal != NULL) { 5571 ext4_error(sb, "Journal got removed while the fs was " 5572 "mounted!"); 5573 return -EFSCORRUPTED; 5574 } 5575 return 0; 5576 } 5577 jbd2_journal_lock_updates(journal); 5578 err = jbd2_journal_flush(journal); 5579 if (err < 0) 5580 goto out; 5581 5582 if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) { 5583 ext4_clear_feature_journal_needs_recovery(sb); 5584 ext4_commit_super(sb, 1); 5585 } 5586 out: 5587 jbd2_journal_unlock_updates(journal); 5588 return err; 5589 } 5590 5591 /* 5592 * If we are mounting (or read-write remounting) a filesystem whose journal 5593 * has recorded an error from a previous lifetime, move that error to the 5594 * main filesystem now. 5595 */ 5596 static int ext4_clear_journal_err(struct super_block *sb, 5597 struct ext4_super_block *es) 5598 { 5599 journal_t *journal; 5600 int j_errno; 5601 const char *errstr; 5602 5603 if (!ext4_has_feature_journal(sb)) { 5604 ext4_error(sb, "Journal got removed while the fs was mounted!"); 5605 return -EFSCORRUPTED; 5606 } 5607 5608 journal = EXT4_SB(sb)->s_journal; 5609 5610 /* 5611 * Now check for any error status which may have been recorded in the 5612 * journal by a prior ext4_error() or ext4_abort() 5613 */ 5614 5615 j_errno = jbd2_journal_errno(journal); 5616 if (j_errno) { 5617 char nbuf[16]; 5618 5619 errstr = ext4_decode_error(sb, j_errno, nbuf); 5620 ext4_warning(sb, "Filesystem error recorded " 5621 "from previous mount: %s", errstr); 5622 ext4_warning(sb, "Marking fs in need of filesystem check."); 5623 5624 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 5625 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5626 ext4_commit_super(sb, 1); 5627 5628 jbd2_journal_clear_err(journal); 5629 jbd2_journal_update_sb_errno(journal); 5630 } 5631 return 0; 5632 } 5633 5634 /* 5635 * Force the running and committing transactions to commit, 5636 * and wait on the commit. 5637 */ 5638 int ext4_force_commit(struct super_block *sb) 5639 { 5640 journal_t *journal; 5641 5642 if (sb_rdonly(sb)) 5643 return 0; 5644 5645 journal = EXT4_SB(sb)->s_journal; 5646 return ext4_journal_force_commit(journal); 5647 } 5648 5649 static int ext4_sync_fs(struct super_block *sb, int wait) 5650 { 5651 int ret = 0; 5652 tid_t target; 5653 bool needs_barrier = false; 5654 struct ext4_sb_info *sbi = EXT4_SB(sb); 5655 5656 if (unlikely(ext4_forced_shutdown(sbi))) 5657 return 0; 5658 5659 trace_ext4_sync_fs(sb, wait); 5660 flush_workqueue(sbi->rsv_conversion_wq); 5661 /* 5662 * Writeback quota in non-journalled quota case - journalled quota has 5663 * no dirty dquots 5664 */ 5665 dquot_writeback_dquots(sb, -1); 5666 /* 5667 * Data writeback is possible w/o journal transaction, so barrier must 5668 * being sent at the end of the function. But we can skip it if 5669 * transaction_commit will do it for us. 5670 */ 5671 if (sbi->s_journal) { 5672 target = jbd2_get_latest_transaction(sbi->s_journal); 5673 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 5674 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 5675 needs_barrier = true; 5676 5677 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 5678 if (wait) 5679 ret = jbd2_log_wait_commit(sbi->s_journal, 5680 target); 5681 } 5682 } else if (wait && test_opt(sb, BARRIER)) 5683 needs_barrier = true; 5684 if (needs_barrier) { 5685 int err; 5686 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL); 5687 if (!ret) 5688 ret = err; 5689 } 5690 5691 return ret; 5692 } 5693 5694 /* 5695 * LVM calls this function before a (read-only) snapshot is created. This 5696 * gives us a chance to flush the journal completely and mark the fs clean. 5697 * 5698 * Note that only this function cannot bring a filesystem to be in a clean 5699 * state independently. It relies on upper layer to stop all data & metadata 5700 * modifications. 5701 */ 5702 static int ext4_freeze(struct super_block *sb) 5703 { 5704 int error = 0; 5705 journal_t *journal; 5706 5707 if (sb_rdonly(sb)) 5708 return 0; 5709 5710 journal = EXT4_SB(sb)->s_journal; 5711 5712 if (journal) { 5713 /* Now we set up the journal barrier. */ 5714 jbd2_journal_lock_updates(journal); 5715 5716 /* 5717 * Don't clear the needs_recovery flag if we failed to 5718 * flush the journal. 5719 */ 5720 error = jbd2_journal_flush(journal); 5721 if (error < 0) 5722 goto out; 5723 5724 /* Journal blocked and flushed, clear needs_recovery flag. */ 5725 ext4_clear_feature_journal_needs_recovery(sb); 5726 } 5727 5728 error = ext4_commit_super(sb, 1); 5729 out: 5730 if (journal) 5731 /* we rely on upper layer to stop further updates */ 5732 jbd2_journal_unlock_updates(journal); 5733 return error; 5734 } 5735 5736 /* 5737 * Called by LVM after the snapshot is done. We need to reset the RECOVER 5738 * flag here, even though the filesystem is not technically dirty yet. 5739 */ 5740 static int ext4_unfreeze(struct super_block *sb) 5741 { 5742 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 5743 return 0; 5744 5745 if (EXT4_SB(sb)->s_journal) { 5746 /* Reset the needs_recovery flag before the fs is unlocked. */ 5747 ext4_set_feature_journal_needs_recovery(sb); 5748 } 5749 5750 ext4_commit_super(sb, 1); 5751 return 0; 5752 } 5753 5754 /* 5755 * Structure to save mount options for ext4_remount's benefit 5756 */ 5757 struct ext4_mount_options { 5758 unsigned long s_mount_opt; 5759 unsigned long s_mount_opt2; 5760 kuid_t s_resuid; 5761 kgid_t s_resgid; 5762 unsigned long s_commit_interval; 5763 u32 s_min_batch_time, s_max_batch_time; 5764 #ifdef CONFIG_QUOTA 5765 int s_jquota_fmt; 5766 char *s_qf_names[EXT4_MAXQUOTAS]; 5767 #endif 5768 }; 5769 5770 static int ext4_remount(struct super_block *sb, int *flags, char *data) 5771 { 5772 struct ext4_super_block *es; 5773 struct ext4_sb_info *sbi = EXT4_SB(sb); 5774 unsigned long old_sb_flags, vfs_flags; 5775 struct ext4_mount_options old_opts; 5776 int enable_quota = 0; 5777 ext4_group_t g; 5778 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5779 int err = 0; 5780 #ifdef CONFIG_QUOTA 5781 int i, j; 5782 char *to_free[EXT4_MAXQUOTAS]; 5783 #endif 5784 char *orig_data = kstrdup(data, GFP_KERNEL); 5785 5786 if (data && !orig_data) 5787 return -ENOMEM; 5788 5789 /* Store the original options */ 5790 old_sb_flags = sb->s_flags; 5791 old_opts.s_mount_opt = sbi->s_mount_opt; 5792 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 5793 old_opts.s_resuid = sbi->s_resuid; 5794 old_opts.s_resgid = sbi->s_resgid; 5795 old_opts.s_commit_interval = sbi->s_commit_interval; 5796 old_opts.s_min_batch_time = sbi->s_min_batch_time; 5797 old_opts.s_max_batch_time = sbi->s_max_batch_time; 5798 #ifdef CONFIG_QUOTA 5799 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 5800 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5801 if (sbi->s_qf_names[i]) { 5802 char *qf_name = get_qf_name(sb, sbi, i); 5803 5804 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 5805 if (!old_opts.s_qf_names[i]) { 5806 for (j = 0; j < i; j++) 5807 kfree(old_opts.s_qf_names[j]); 5808 kfree(orig_data); 5809 return -ENOMEM; 5810 } 5811 } else 5812 old_opts.s_qf_names[i] = NULL; 5813 #endif 5814 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 5815 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio; 5816 5817 /* 5818 * Some options can be enabled by ext4 and/or by VFS mount flag 5819 * either way we need to make sure it matches in both *flags and 5820 * s_flags. Copy those selected flags from *flags to s_flags 5821 */ 5822 vfs_flags = SB_LAZYTIME | SB_I_VERSION; 5823 sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags); 5824 5825 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) { 5826 err = -EINVAL; 5827 goto restore_opts; 5828 } 5829 5830 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 5831 test_opt(sb, JOURNAL_CHECKSUM)) { 5832 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 5833 "during remount not supported; ignoring"); 5834 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 5835 } 5836 5837 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5838 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5839 ext4_msg(sb, KERN_ERR, "can't mount with " 5840 "both data=journal and delalloc"); 5841 err = -EINVAL; 5842 goto restore_opts; 5843 } 5844 if (test_opt(sb, DIOREAD_NOLOCK)) { 5845 ext4_msg(sb, KERN_ERR, "can't mount with " 5846 "both data=journal and dioread_nolock"); 5847 err = -EINVAL; 5848 goto restore_opts; 5849 } 5850 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 5851 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5852 ext4_msg(sb, KERN_ERR, "can't mount with " 5853 "journal_async_commit in data=ordered mode"); 5854 err = -EINVAL; 5855 goto restore_opts; 5856 } 5857 } 5858 5859 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 5860 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 5861 err = -EINVAL; 5862 goto restore_opts; 5863 } 5864 5865 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) 5866 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user"); 5867 5868 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5869 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5870 5871 es = sbi->s_es; 5872 5873 if (sbi->s_journal) { 5874 ext4_init_journal_params(sb, sbi->s_journal); 5875 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 5876 } 5877 5878 if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) { 5879 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) { 5880 err = -EROFS; 5881 goto restore_opts; 5882 } 5883 5884 if (*flags & SB_RDONLY) { 5885 err = sync_filesystem(sb); 5886 if (err < 0) 5887 goto restore_opts; 5888 err = dquot_suspend(sb, -1); 5889 if (err < 0) 5890 goto restore_opts; 5891 5892 /* 5893 * First of all, the unconditional stuff we have to do 5894 * to disable replay of the journal when we next remount 5895 */ 5896 sb->s_flags |= SB_RDONLY; 5897 5898 /* 5899 * OK, test if we are remounting a valid rw partition 5900 * readonly, and if so set the rdonly flag and then 5901 * mark the partition as valid again. 5902 */ 5903 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 5904 (sbi->s_mount_state & EXT4_VALID_FS)) 5905 es->s_state = cpu_to_le16(sbi->s_mount_state); 5906 5907 if (sbi->s_journal) { 5908 /* 5909 * We let remount-ro finish even if marking fs 5910 * as clean failed... 5911 */ 5912 ext4_mark_recovery_complete(sb, es); 5913 } 5914 if (sbi->s_mmp_tsk) 5915 kthread_stop(sbi->s_mmp_tsk); 5916 } else { 5917 /* Make sure we can mount this feature set readwrite */ 5918 if (ext4_has_feature_readonly(sb) || 5919 !ext4_feature_set_ok(sb, 0)) { 5920 err = -EROFS; 5921 goto restore_opts; 5922 } 5923 /* 5924 * Make sure the group descriptor checksums 5925 * are sane. If they aren't, refuse to remount r/w. 5926 */ 5927 for (g = 0; g < sbi->s_groups_count; g++) { 5928 struct ext4_group_desc *gdp = 5929 ext4_get_group_desc(sb, g, NULL); 5930 5931 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 5932 ext4_msg(sb, KERN_ERR, 5933 "ext4_remount: Checksum for group %u failed (%u!=%u)", 5934 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 5935 le16_to_cpu(gdp->bg_checksum)); 5936 err = -EFSBADCRC; 5937 goto restore_opts; 5938 } 5939 } 5940 5941 /* 5942 * If we have an unprocessed orphan list hanging 5943 * around from a previously readonly bdev mount, 5944 * require a full umount/remount for now. 5945 */ 5946 if (es->s_last_orphan) { 5947 ext4_msg(sb, KERN_WARNING, "Couldn't " 5948 "remount RDWR because of unprocessed " 5949 "orphan inode list. Please " 5950 "umount/remount instead"); 5951 err = -EINVAL; 5952 goto restore_opts; 5953 } 5954 5955 /* 5956 * Mounting a RDONLY partition read-write, so reread 5957 * and store the current valid flag. (It may have 5958 * been changed by e2fsck since we originally mounted 5959 * the partition.) 5960 */ 5961 if (sbi->s_journal) { 5962 err = ext4_clear_journal_err(sb, es); 5963 if (err) 5964 goto restore_opts; 5965 } 5966 sbi->s_mount_state = le16_to_cpu(es->s_state); 5967 5968 err = ext4_setup_super(sb, es, 0); 5969 if (err) 5970 goto restore_opts; 5971 5972 sb->s_flags &= ~SB_RDONLY; 5973 if (ext4_has_feature_mmp(sb)) 5974 if (ext4_multi_mount_protect(sb, 5975 le64_to_cpu(es->s_mmp_block))) { 5976 err = -EROFS; 5977 goto restore_opts; 5978 } 5979 enable_quota = 1; 5980 } 5981 } 5982 5983 /* 5984 * Reinitialize lazy itable initialization thread based on 5985 * current settings 5986 */ 5987 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 5988 ext4_unregister_li_request(sb); 5989 else { 5990 ext4_group_t first_not_zeroed; 5991 first_not_zeroed = ext4_has_uninit_itable(sb); 5992 ext4_register_li_request(sb, first_not_zeroed); 5993 } 5994 5995 /* 5996 * Handle creation of system zone data early because it can fail. 5997 * Releasing of existing data is done when we are sure remount will 5998 * succeed. 5999 */ 6000 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6001 err = ext4_setup_system_zone(sb); 6002 if (err) 6003 goto restore_opts; 6004 } 6005 6006 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6007 err = ext4_commit_super(sb, 1); 6008 if (err) 6009 goto restore_opts; 6010 } 6011 6012 #ifdef CONFIG_QUOTA 6013 /* Release old quota file names */ 6014 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6015 kfree(old_opts.s_qf_names[i]); 6016 if (enable_quota) { 6017 if (sb_any_quota_suspended(sb)) 6018 dquot_resume(sb, -1); 6019 else if (ext4_has_feature_quota(sb)) { 6020 err = ext4_enable_quotas(sb); 6021 if (err) 6022 goto restore_opts; 6023 } 6024 } 6025 #endif 6026 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6027 ext4_release_system_zone(sb); 6028 6029 /* 6030 * Some options can be enabled by ext4 and/or by VFS mount flag 6031 * either way we need to make sure it matches in both *flags and 6032 * s_flags. Copy those selected flags from s_flags to *flags 6033 */ 6034 *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags); 6035 6036 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data); 6037 kfree(orig_data); 6038 return 0; 6039 6040 restore_opts: 6041 sb->s_flags = old_sb_flags; 6042 sbi->s_mount_opt = old_opts.s_mount_opt; 6043 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6044 sbi->s_resuid = old_opts.s_resuid; 6045 sbi->s_resgid = old_opts.s_resgid; 6046 sbi->s_commit_interval = old_opts.s_commit_interval; 6047 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6048 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6049 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6050 ext4_release_system_zone(sb); 6051 #ifdef CONFIG_QUOTA 6052 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6053 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6054 to_free[i] = get_qf_name(sb, sbi, i); 6055 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6056 } 6057 synchronize_rcu(); 6058 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6059 kfree(to_free[i]); 6060 #endif 6061 kfree(orig_data); 6062 return err; 6063 } 6064 6065 #ifdef CONFIG_QUOTA 6066 static int ext4_statfs_project(struct super_block *sb, 6067 kprojid_t projid, struct kstatfs *buf) 6068 { 6069 struct kqid qid; 6070 struct dquot *dquot; 6071 u64 limit; 6072 u64 curblock; 6073 6074 qid = make_kqid_projid(projid); 6075 dquot = dqget(sb, qid); 6076 if (IS_ERR(dquot)) 6077 return PTR_ERR(dquot); 6078 spin_lock(&dquot->dq_dqb_lock); 6079 6080 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6081 dquot->dq_dqb.dqb_bhardlimit); 6082 limit >>= sb->s_blocksize_bits; 6083 6084 if (limit && buf->f_blocks > limit) { 6085 curblock = (dquot->dq_dqb.dqb_curspace + 6086 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6087 buf->f_blocks = limit; 6088 buf->f_bfree = buf->f_bavail = 6089 (buf->f_blocks > curblock) ? 6090 (buf->f_blocks - curblock) : 0; 6091 } 6092 6093 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6094 dquot->dq_dqb.dqb_ihardlimit); 6095 if (limit && buf->f_files > limit) { 6096 buf->f_files = limit; 6097 buf->f_ffree = 6098 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6099 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6100 } 6101 6102 spin_unlock(&dquot->dq_dqb_lock); 6103 dqput(dquot); 6104 return 0; 6105 } 6106 #endif 6107 6108 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6109 { 6110 struct super_block *sb = dentry->d_sb; 6111 struct ext4_sb_info *sbi = EXT4_SB(sb); 6112 struct ext4_super_block *es = sbi->s_es; 6113 ext4_fsblk_t overhead = 0, resv_blocks; 6114 u64 fsid; 6115 s64 bfree; 6116 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6117 6118 if (!test_opt(sb, MINIX_DF)) 6119 overhead = sbi->s_overhead; 6120 6121 buf->f_type = EXT4_SUPER_MAGIC; 6122 buf->f_bsize = sb->s_blocksize; 6123 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6124 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6125 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6126 /* prevent underflow in case that few free space is available */ 6127 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6128 buf->f_bavail = buf->f_bfree - 6129 (ext4_r_blocks_count(es) + resv_blocks); 6130 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6131 buf->f_bavail = 0; 6132 buf->f_files = le32_to_cpu(es->s_inodes_count); 6133 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6134 buf->f_namelen = EXT4_NAME_LEN; 6135 fsid = le64_to_cpup((void *)es->s_uuid) ^ 6136 le64_to_cpup((void *)es->s_uuid + sizeof(u64)); 6137 buf->f_fsid = u64_to_fsid(fsid); 6138 6139 #ifdef CONFIG_QUOTA 6140 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6141 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6142 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6143 #endif 6144 return 0; 6145 } 6146 6147 6148 #ifdef CONFIG_QUOTA 6149 6150 /* 6151 * Helper functions so that transaction is started before we acquire dqio_sem 6152 * to keep correct lock ordering of transaction > dqio_sem 6153 */ 6154 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6155 { 6156 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6157 } 6158 6159 static int ext4_write_dquot(struct dquot *dquot) 6160 { 6161 int ret, err; 6162 handle_t *handle; 6163 struct inode *inode; 6164 6165 inode = dquot_to_inode(dquot); 6166 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6167 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6168 if (IS_ERR(handle)) 6169 return PTR_ERR(handle); 6170 ret = dquot_commit(dquot); 6171 err = ext4_journal_stop(handle); 6172 if (!ret) 6173 ret = err; 6174 return ret; 6175 } 6176 6177 static int ext4_acquire_dquot(struct dquot *dquot) 6178 { 6179 int ret, err; 6180 handle_t *handle; 6181 6182 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6183 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6184 if (IS_ERR(handle)) 6185 return PTR_ERR(handle); 6186 ret = dquot_acquire(dquot); 6187 err = ext4_journal_stop(handle); 6188 if (!ret) 6189 ret = err; 6190 return ret; 6191 } 6192 6193 static int ext4_release_dquot(struct dquot *dquot) 6194 { 6195 int ret, err; 6196 handle_t *handle; 6197 6198 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6199 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6200 if (IS_ERR(handle)) { 6201 /* Release dquot anyway to avoid endless cycle in dqput() */ 6202 dquot_release(dquot); 6203 return PTR_ERR(handle); 6204 } 6205 ret = dquot_release(dquot); 6206 err = ext4_journal_stop(handle); 6207 if (!ret) 6208 ret = err; 6209 return ret; 6210 } 6211 6212 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6213 { 6214 struct super_block *sb = dquot->dq_sb; 6215 struct ext4_sb_info *sbi = EXT4_SB(sb); 6216 6217 /* Are we journaling quotas? */ 6218 if (ext4_has_feature_quota(sb) || 6219 sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) { 6220 dquot_mark_dquot_dirty(dquot); 6221 return ext4_write_dquot(dquot); 6222 } else { 6223 return dquot_mark_dquot_dirty(dquot); 6224 } 6225 } 6226 6227 static int ext4_write_info(struct super_block *sb, int type) 6228 { 6229 int ret, err; 6230 handle_t *handle; 6231 6232 /* Data block + inode block */ 6233 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 6234 if (IS_ERR(handle)) 6235 return PTR_ERR(handle); 6236 ret = dquot_commit_info(sb, type); 6237 err = ext4_journal_stop(handle); 6238 if (!ret) 6239 ret = err; 6240 return ret; 6241 } 6242 6243 /* 6244 * Turn on quotas during mount time - we need to find 6245 * the quota file and such... 6246 */ 6247 static int ext4_quota_on_mount(struct super_block *sb, int type) 6248 { 6249 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 6250 EXT4_SB(sb)->s_jquota_fmt, type); 6251 } 6252 6253 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6254 { 6255 struct ext4_inode_info *ei = EXT4_I(inode); 6256 6257 /* The first argument of lockdep_set_subclass has to be 6258 * *exactly* the same as the argument to init_rwsem() --- in 6259 * this case, in init_once() --- or lockdep gets unhappy 6260 * because the name of the lock is set using the 6261 * stringification of the argument to init_rwsem(). 6262 */ 6263 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6264 lockdep_set_subclass(&ei->i_data_sem, subclass); 6265 } 6266 6267 /* 6268 * Standard function to be called on quota_on 6269 */ 6270 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6271 const struct path *path) 6272 { 6273 int err; 6274 6275 if (!test_opt(sb, QUOTA)) 6276 return -EINVAL; 6277 6278 /* Quotafile not on the same filesystem? */ 6279 if (path->dentry->d_sb != sb) 6280 return -EXDEV; 6281 6282 /* Quota already enabled for this file? */ 6283 if (IS_NOQUOTA(d_inode(path->dentry))) 6284 return -EBUSY; 6285 6286 /* Journaling quota? */ 6287 if (EXT4_SB(sb)->s_qf_names[type]) { 6288 /* Quotafile not in fs root? */ 6289 if (path->dentry->d_parent != sb->s_root) 6290 ext4_msg(sb, KERN_WARNING, 6291 "Quota file not on filesystem root. " 6292 "Journaled quota will not work"); 6293 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6294 } else { 6295 /* 6296 * Clear the flag just in case mount options changed since 6297 * last time. 6298 */ 6299 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6300 } 6301 6302 /* 6303 * When we journal data on quota file, we have to flush journal to see 6304 * all updates to the file when we bypass pagecache... 6305 */ 6306 if (EXT4_SB(sb)->s_journal && 6307 ext4_should_journal_data(d_inode(path->dentry))) { 6308 /* 6309 * We don't need to lock updates but journal_flush() could 6310 * otherwise be livelocked... 6311 */ 6312 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6313 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 6314 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6315 if (err) 6316 return err; 6317 } 6318 6319 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6320 err = dquot_quota_on(sb, type, format_id, path); 6321 if (err) { 6322 lockdep_set_quota_inode(path->dentry->d_inode, 6323 I_DATA_SEM_NORMAL); 6324 } else { 6325 struct inode *inode = d_inode(path->dentry); 6326 handle_t *handle; 6327 6328 /* 6329 * Set inode flags to prevent userspace from messing with quota 6330 * files. If this fails, we return success anyway since quotas 6331 * are already enabled and this is not a hard failure. 6332 */ 6333 inode_lock(inode); 6334 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6335 if (IS_ERR(handle)) 6336 goto unlock_inode; 6337 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6338 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6339 S_NOATIME | S_IMMUTABLE); 6340 err = ext4_mark_inode_dirty(handle, inode); 6341 ext4_journal_stop(handle); 6342 unlock_inode: 6343 inode_unlock(inode); 6344 } 6345 return err; 6346 } 6347 6348 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6349 unsigned int flags) 6350 { 6351 int err; 6352 struct inode *qf_inode; 6353 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6354 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6355 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6356 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6357 }; 6358 6359 BUG_ON(!ext4_has_feature_quota(sb)); 6360 6361 if (!qf_inums[type]) 6362 return -EPERM; 6363 6364 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6365 if (IS_ERR(qf_inode)) { 6366 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6367 return PTR_ERR(qf_inode); 6368 } 6369 6370 /* Don't account quota for quota files to avoid recursion */ 6371 qf_inode->i_flags |= S_NOQUOTA; 6372 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6373 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6374 if (err) 6375 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6376 iput(qf_inode); 6377 6378 return err; 6379 } 6380 6381 /* Enable usage tracking for all quota types. */ 6382 static int ext4_enable_quotas(struct super_block *sb) 6383 { 6384 int type, err = 0; 6385 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6386 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6387 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6388 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6389 }; 6390 bool quota_mopt[EXT4_MAXQUOTAS] = { 6391 test_opt(sb, USRQUOTA), 6392 test_opt(sb, GRPQUOTA), 6393 test_opt(sb, PRJQUOTA), 6394 }; 6395 6396 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6397 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6398 if (qf_inums[type]) { 6399 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6400 DQUOT_USAGE_ENABLED | 6401 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6402 if (err) { 6403 ext4_warning(sb, 6404 "Failed to enable quota tracking " 6405 "(type=%d, err=%d). Please run " 6406 "e2fsck to fix.", type, err); 6407 for (type--; type >= 0; type--) 6408 dquot_quota_off(sb, type); 6409 6410 return err; 6411 } 6412 } 6413 } 6414 return 0; 6415 } 6416 6417 static int ext4_quota_off(struct super_block *sb, int type) 6418 { 6419 struct inode *inode = sb_dqopt(sb)->files[type]; 6420 handle_t *handle; 6421 int err; 6422 6423 /* Force all delayed allocation blocks to be allocated. 6424 * Caller already holds s_umount sem */ 6425 if (test_opt(sb, DELALLOC)) 6426 sync_filesystem(sb); 6427 6428 if (!inode || !igrab(inode)) 6429 goto out; 6430 6431 err = dquot_quota_off(sb, type); 6432 if (err || ext4_has_feature_quota(sb)) 6433 goto out_put; 6434 6435 inode_lock(inode); 6436 /* 6437 * Update modification times of quota files when userspace can 6438 * start looking at them. If we fail, we return success anyway since 6439 * this is not a hard failure and quotas are already disabled. 6440 */ 6441 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6442 if (IS_ERR(handle)) { 6443 err = PTR_ERR(handle); 6444 goto out_unlock; 6445 } 6446 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 6447 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 6448 inode->i_mtime = inode->i_ctime = current_time(inode); 6449 err = ext4_mark_inode_dirty(handle, inode); 6450 ext4_journal_stop(handle); 6451 out_unlock: 6452 inode_unlock(inode); 6453 out_put: 6454 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 6455 iput(inode); 6456 return err; 6457 out: 6458 return dquot_quota_off(sb, type); 6459 } 6460 6461 /* Read data from quotafile - avoid pagecache and such because we cannot afford 6462 * acquiring the locks... As quota files are never truncated and quota code 6463 * itself serializes the operations (and no one else should touch the files) 6464 * we don't have to be afraid of races */ 6465 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 6466 size_t len, loff_t off) 6467 { 6468 struct inode *inode = sb_dqopt(sb)->files[type]; 6469 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6470 int offset = off & (sb->s_blocksize - 1); 6471 int tocopy; 6472 size_t toread; 6473 struct buffer_head *bh; 6474 loff_t i_size = i_size_read(inode); 6475 6476 if (off > i_size) 6477 return 0; 6478 if (off+len > i_size) 6479 len = i_size-off; 6480 toread = len; 6481 while (toread > 0) { 6482 tocopy = sb->s_blocksize - offset < toread ? 6483 sb->s_blocksize - offset : toread; 6484 bh = ext4_bread(NULL, inode, blk, 0); 6485 if (IS_ERR(bh)) 6486 return PTR_ERR(bh); 6487 if (!bh) /* A hole? */ 6488 memset(data, 0, tocopy); 6489 else 6490 memcpy(data, bh->b_data+offset, tocopy); 6491 brelse(bh); 6492 offset = 0; 6493 toread -= tocopy; 6494 data += tocopy; 6495 blk++; 6496 } 6497 return len; 6498 } 6499 6500 /* Write to quotafile (we know the transaction is already started and has 6501 * enough credits) */ 6502 static ssize_t ext4_quota_write(struct super_block *sb, int type, 6503 const char *data, size_t len, loff_t off) 6504 { 6505 struct inode *inode = sb_dqopt(sb)->files[type]; 6506 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6507 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 6508 int retries = 0; 6509 struct buffer_head *bh; 6510 handle_t *handle = journal_current_handle(); 6511 6512 if (EXT4_SB(sb)->s_journal && !handle) { 6513 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6514 " cancelled because transaction is not started", 6515 (unsigned long long)off, (unsigned long long)len); 6516 return -EIO; 6517 } 6518 /* 6519 * Since we account only one data block in transaction credits, 6520 * then it is impossible to cross a block boundary. 6521 */ 6522 if (sb->s_blocksize - offset < len) { 6523 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6524 " cancelled because not block aligned", 6525 (unsigned long long)off, (unsigned long long)len); 6526 return -EIO; 6527 } 6528 6529 do { 6530 bh = ext4_bread(handle, inode, blk, 6531 EXT4_GET_BLOCKS_CREATE | 6532 EXT4_GET_BLOCKS_METADATA_NOFAIL); 6533 } while (PTR_ERR(bh) == -ENOSPC && 6534 ext4_should_retry_alloc(inode->i_sb, &retries)); 6535 if (IS_ERR(bh)) 6536 return PTR_ERR(bh); 6537 if (!bh) 6538 goto out; 6539 BUFFER_TRACE(bh, "get write access"); 6540 err = ext4_journal_get_write_access(handle, bh); 6541 if (err) { 6542 brelse(bh); 6543 return err; 6544 } 6545 lock_buffer(bh); 6546 memcpy(bh->b_data+offset, data, len); 6547 flush_dcache_page(bh->b_page); 6548 unlock_buffer(bh); 6549 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6550 brelse(bh); 6551 out: 6552 if (inode->i_size < off + len) { 6553 i_size_write(inode, off + len); 6554 EXT4_I(inode)->i_disksize = inode->i_size; 6555 err2 = ext4_mark_inode_dirty(handle, inode); 6556 if (unlikely(err2 && !err)) 6557 err = err2; 6558 } 6559 return err ? err : len; 6560 } 6561 #endif 6562 6563 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6564 const char *dev_name, void *data) 6565 { 6566 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6567 } 6568 6569 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6570 static inline void register_as_ext2(void) 6571 { 6572 int err = register_filesystem(&ext2_fs_type); 6573 if (err) 6574 printk(KERN_WARNING 6575 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6576 } 6577 6578 static inline void unregister_as_ext2(void) 6579 { 6580 unregister_filesystem(&ext2_fs_type); 6581 } 6582 6583 static inline int ext2_feature_set_ok(struct super_block *sb) 6584 { 6585 if (ext4_has_unknown_ext2_incompat_features(sb)) 6586 return 0; 6587 if (sb_rdonly(sb)) 6588 return 1; 6589 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6590 return 0; 6591 return 1; 6592 } 6593 #else 6594 static inline void register_as_ext2(void) { } 6595 static inline void unregister_as_ext2(void) { } 6596 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6597 #endif 6598 6599 static inline void register_as_ext3(void) 6600 { 6601 int err = register_filesystem(&ext3_fs_type); 6602 if (err) 6603 printk(KERN_WARNING 6604 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6605 } 6606 6607 static inline void unregister_as_ext3(void) 6608 { 6609 unregister_filesystem(&ext3_fs_type); 6610 } 6611 6612 static inline int ext3_feature_set_ok(struct super_block *sb) 6613 { 6614 if (ext4_has_unknown_ext3_incompat_features(sb)) 6615 return 0; 6616 if (!ext4_has_feature_journal(sb)) 6617 return 0; 6618 if (sb_rdonly(sb)) 6619 return 1; 6620 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6621 return 0; 6622 return 1; 6623 } 6624 6625 static struct file_system_type ext4_fs_type = { 6626 .owner = THIS_MODULE, 6627 .name = "ext4", 6628 .mount = ext4_mount, 6629 .kill_sb = kill_block_super, 6630 .fs_flags = FS_REQUIRES_DEV, 6631 }; 6632 MODULE_ALIAS_FS("ext4"); 6633 6634 /* Shared across all ext4 file systems */ 6635 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6636 6637 static int __init ext4_init_fs(void) 6638 { 6639 int i, err; 6640 6641 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6642 ext4_li_info = NULL; 6643 mutex_init(&ext4_li_mtx); 6644 6645 /* Build-time check for flags consistency */ 6646 ext4_check_flag_values(); 6647 6648 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6649 init_waitqueue_head(&ext4__ioend_wq[i]); 6650 6651 err = ext4_init_es(); 6652 if (err) 6653 return err; 6654 6655 err = ext4_init_pending(); 6656 if (err) 6657 goto out7; 6658 6659 err = ext4_init_post_read_processing(); 6660 if (err) 6661 goto out6; 6662 6663 err = ext4_init_pageio(); 6664 if (err) 6665 goto out5; 6666 6667 err = ext4_init_system_zone(); 6668 if (err) 6669 goto out4; 6670 6671 err = ext4_init_sysfs(); 6672 if (err) 6673 goto out3; 6674 6675 err = ext4_init_mballoc(); 6676 if (err) 6677 goto out2; 6678 err = init_inodecache(); 6679 if (err) 6680 goto out1; 6681 6682 err = ext4_fc_init_dentry_cache(); 6683 if (err) 6684 goto out05; 6685 6686 register_as_ext3(); 6687 register_as_ext2(); 6688 err = register_filesystem(&ext4_fs_type); 6689 if (err) 6690 goto out; 6691 6692 return 0; 6693 out: 6694 unregister_as_ext2(); 6695 unregister_as_ext3(); 6696 out05: 6697 destroy_inodecache(); 6698 out1: 6699 ext4_exit_mballoc(); 6700 out2: 6701 ext4_exit_sysfs(); 6702 out3: 6703 ext4_exit_system_zone(); 6704 out4: 6705 ext4_exit_pageio(); 6706 out5: 6707 ext4_exit_post_read_processing(); 6708 out6: 6709 ext4_exit_pending(); 6710 out7: 6711 ext4_exit_es(); 6712 6713 return err; 6714 } 6715 6716 static void __exit ext4_exit_fs(void) 6717 { 6718 ext4_destroy_lazyinit_thread(); 6719 unregister_as_ext2(); 6720 unregister_as_ext3(); 6721 unregister_filesystem(&ext4_fs_type); 6722 destroy_inodecache(); 6723 ext4_exit_mballoc(); 6724 ext4_exit_sysfs(); 6725 ext4_exit_system_zone(); 6726 ext4_exit_pageio(); 6727 ext4_exit_post_read_processing(); 6728 ext4_exit_es(); 6729 ext4_exit_pending(); 6730 } 6731 6732 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6733 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6734 MODULE_LICENSE("GPL"); 6735 MODULE_SOFTDEP("pre: crc32c"); 6736 module_init(ext4_init_fs) 6737 module_exit(ext4_exit_fs) 6738