1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/super.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Big-endian to little-endian byte-swapping/bitmaps by 17 * David S. Miller (davem@caip.rutgers.edu), 1995 18 */ 19 20 #include <linux/module.h> 21 #include <linux/string.h> 22 #include <linux/fs.h> 23 #include <linux/time.h> 24 #include <linux/vmalloc.h> 25 #include <linux/slab.h> 26 #include <linux/init.h> 27 #include <linux/blkdev.h> 28 #include <linux/backing-dev.h> 29 #include <linux/parser.h> 30 #include <linux/buffer_head.h> 31 #include <linux/exportfs.h> 32 #include <linux/vfs.h> 33 #include <linux/random.h> 34 #include <linux/mount.h> 35 #include <linux/namei.h> 36 #include <linux/quotaops.h> 37 #include <linux/seq_file.h> 38 #include <linux/ctype.h> 39 #include <linux/log2.h> 40 #include <linux/crc16.h> 41 #include <linux/dax.h> 42 #include <linux/cleancache.h> 43 #include <linux/uaccess.h> 44 #include <linux/iversion.h> 45 #include <linux/unicode.h> 46 #include <linux/part_stat.h> 47 #include <linux/kthread.h> 48 #include <linux/freezer.h> 49 50 #include "ext4.h" 51 #include "ext4_extents.h" /* Needed for trace points definition */ 52 #include "ext4_jbd2.h" 53 #include "xattr.h" 54 #include "acl.h" 55 #include "mballoc.h" 56 #include "fsmap.h" 57 58 #define CREATE_TRACE_POINTS 59 #include <trace/events/ext4.h> 60 61 static struct ext4_lazy_init *ext4_li_info; 62 static DEFINE_MUTEX(ext4_li_mtx); 63 static struct ratelimit_state ext4_mount_msg_ratelimit; 64 65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 66 unsigned long journal_devnum); 67 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 68 static void ext4_update_super(struct super_block *sb); 69 static int ext4_commit_super(struct super_block *sb); 70 static int ext4_mark_recovery_complete(struct super_block *sb, 71 struct ext4_super_block *es); 72 static int ext4_clear_journal_err(struct super_block *sb, 73 struct ext4_super_block *es); 74 static int ext4_sync_fs(struct super_block *sb, int wait); 75 static int ext4_remount(struct super_block *sb, int *flags, char *data); 76 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 77 static int ext4_unfreeze(struct super_block *sb); 78 static int ext4_freeze(struct super_block *sb); 79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 80 const char *dev_name, void *data); 81 static inline int ext2_feature_set_ok(struct super_block *sb); 82 static inline int ext3_feature_set_ok(struct super_block *sb); 83 static int ext4_feature_set_ok(struct super_block *sb, int readonly); 84 static void ext4_destroy_lazyinit_thread(void); 85 static void ext4_unregister_li_request(struct super_block *sb); 86 static void ext4_clear_request_list(void); 87 static struct inode *ext4_get_journal_inode(struct super_block *sb, 88 unsigned int journal_inum); 89 90 /* 91 * Lock ordering 92 * 93 * page fault path: 94 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start 95 * -> page lock -> i_data_sem (rw) 96 * 97 * buffered write path: 98 * sb_start_write -> i_mutex -> mmap_lock 99 * sb_start_write -> i_mutex -> transaction start -> page lock -> 100 * i_data_sem (rw) 101 * 102 * truncate: 103 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) -> 104 * page lock 105 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start -> 106 * i_data_sem (rw) 107 * 108 * direct IO: 109 * sb_start_write -> i_mutex -> mmap_lock 110 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 111 * 112 * writepages: 113 * transaction start -> page lock(s) -> i_data_sem (rw) 114 */ 115 116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 117 static struct file_system_type ext2_fs_type = { 118 .owner = THIS_MODULE, 119 .name = "ext2", 120 .mount = ext4_mount, 121 .kill_sb = kill_block_super, 122 .fs_flags = FS_REQUIRES_DEV, 123 }; 124 MODULE_ALIAS_FS("ext2"); 125 MODULE_ALIAS("ext2"); 126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type) 127 #else 128 #define IS_EXT2_SB(sb) (0) 129 #endif 130 131 132 static struct file_system_type ext3_fs_type = { 133 .owner = THIS_MODULE, 134 .name = "ext3", 135 .mount = ext4_mount, 136 .kill_sb = kill_block_super, 137 .fs_flags = FS_REQUIRES_DEV, 138 }; 139 MODULE_ALIAS_FS("ext3"); 140 MODULE_ALIAS("ext3"); 141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type) 142 143 144 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags, 145 bh_end_io_t *end_io) 146 { 147 /* 148 * buffer's verified bit is no longer valid after reading from 149 * disk again due to write out error, clear it to make sure we 150 * recheck the buffer contents. 151 */ 152 clear_buffer_verified(bh); 153 154 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 155 get_bh(bh); 156 submit_bh(REQ_OP_READ, op_flags, bh); 157 } 158 159 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags, 160 bh_end_io_t *end_io) 161 { 162 BUG_ON(!buffer_locked(bh)); 163 164 if (ext4_buffer_uptodate(bh)) { 165 unlock_buffer(bh); 166 return; 167 } 168 __ext4_read_bh(bh, op_flags, end_io); 169 } 170 171 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io) 172 { 173 BUG_ON(!buffer_locked(bh)); 174 175 if (ext4_buffer_uptodate(bh)) { 176 unlock_buffer(bh); 177 return 0; 178 } 179 180 __ext4_read_bh(bh, op_flags, end_io); 181 182 wait_on_buffer(bh); 183 if (buffer_uptodate(bh)) 184 return 0; 185 return -EIO; 186 } 187 188 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait) 189 { 190 if (trylock_buffer(bh)) { 191 if (wait) 192 return ext4_read_bh(bh, op_flags, NULL); 193 ext4_read_bh_nowait(bh, op_flags, NULL); 194 return 0; 195 } 196 if (wait) { 197 wait_on_buffer(bh); 198 if (buffer_uptodate(bh)) 199 return 0; 200 return -EIO; 201 } 202 return 0; 203 } 204 205 /* 206 * This works like __bread_gfp() except it uses ERR_PTR for error 207 * returns. Currently with sb_bread it's impossible to distinguish 208 * between ENOMEM and EIO situations (since both result in a NULL 209 * return. 210 */ 211 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 212 sector_t block, int op_flags, 213 gfp_t gfp) 214 { 215 struct buffer_head *bh; 216 int ret; 217 218 bh = sb_getblk_gfp(sb, block, gfp); 219 if (bh == NULL) 220 return ERR_PTR(-ENOMEM); 221 if (ext4_buffer_uptodate(bh)) 222 return bh; 223 224 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 225 if (ret) { 226 put_bh(bh); 227 return ERR_PTR(ret); 228 } 229 return bh; 230 } 231 232 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 233 int op_flags) 234 { 235 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 236 } 237 238 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 239 sector_t block) 240 { 241 return __ext4_sb_bread_gfp(sb, block, 0, 0); 242 } 243 244 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 245 { 246 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 247 248 if (likely(bh)) { 249 ext4_read_bh_lock(bh, REQ_RAHEAD, false); 250 brelse(bh); 251 } 252 } 253 254 static int ext4_verify_csum_type(struct super_block *sb, 255 struct ext4_super_block *es) 256 { 257 if (!ext4_has_feature_metadata_csum(sb)) 258 return 1; 259 260 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 261 } 262 263 static __le32 ext4_superblock_csum(struct super_block *sb, 264 struct ext4_super_block *es) 265 { 266 struct ext4_sb_info *sbi = EXT4_SB(sb); 267 int offset = offsetof(struct ext4_super_block, s_checksum); 268 __u32 csum; 269 270 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 271 272 return cpu_to_le32(csum); 273 } 274 275 static int ext4_superblock_csum_verify(struct super_block *sb, 276 struct ext4_super_block *es) 277 { 278 if (!ext4_has_metadata_csum(sb)) 279 return 1; 280 281 return es->s_checksum == ext4_superblock_csum(sb, es); 282 } 283 284 void ext4_superblock_csum_set(struct super_block *sb) 285 { 286 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 287 288 if (!ext4_has_metadata_csum(sb)) 289 return; 290 291 es->s_checksum = ext4_superblock_csum(sb, es); 292 } 293 294 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 295 struct ext4_group_desc *bg) 296 { 297 return le32_to_cpu(bg->bg_block_bitmap_lo) | 298 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 299 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 300 } 301 302 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 303 struct ext4_group_desc *bg) 304 { 305 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 306 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 307 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 308 } 309 310 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 311 struct ext4_group_desc *bg) 312 { 313 return le32_to_cpu(bg->bg_inode_table_lo) | 314 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 315 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 316 } 317 318 __u32 ext4_free_group_clusters(struct super_block *sb, 319 struct ext4_group_desc *bg) 320 { 321 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 322 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 323 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 324 } 325 326 __u32 ext4_free_inodes_count(struct super_block *sb, 327 struct ext4_group_desc *bg) 328 { 329 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 330 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 331 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 332 } 333 334 __u32 ext4_used_dirs_count(struct super_block *sb, 335 struct ext4_group_desc *bg) 336 { 337 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 338 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 339 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 340 } 341 342 __u32 ext4_itable_unused_count(struct super_block *sb, 343 struct ext4_group_desc *bg) 344 { 345 return le16_to_cpu(bg->bg_itable_unused_lo) | 346 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 347 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 348 } 349 350 void ext4_block_bitmap_set(struct super_block *sb, 351 struct ext4_group_desc *bg, ext4_fsblk_t blk) 352 { 353 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 354 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 355 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 356 } 357 358 void ext4_inode_bitmap_set(struct super_block *sb, 359 struct ext4_group_desc *bg, ext4_fsblk_t blk) 360 { 361 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 362 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 363 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 364 } 365 366 void ext4_inode_table_set(struct super_block *sb, 367 struct ext4_group_desc *bg, ext4_fsblk_t blk) 368 { 369 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 370 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 371 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 372 } 373 374 void ext4_free_group_clusters_set(struct super_block *sb, 375 struct ext4_group_desc *bg, __u32 count) 376 { 377 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 378 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 379 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 380 } 381 382 void ext4_free_inodes_set(struct super_block *sb, 383 struct ext4_group_desc *bg, __u32 count) 384 { 385 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 386 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 387 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 388 } 389 390 void ext4_used_dirs_set(struct super_block *sb, 391 struct ext4_group_desc *bg, __u32 count) 392 { 393 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 394 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 395 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 396 } 397 398 void ext4_itable_unused_set(struct super_block *sb, 399 struct ext4_group_desc *bg, __u32 count) 400 { 401 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 402 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 403 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 404 } 405 406 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 407 { 408 now = clamp_val(now, 0, (1ull << 40) - 1); 409 410 *lo = cpu_to_le32(lower_32_bits(now)); 411 *hi = upper_32_bits(now); 412 } 413 414 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 415 { 416 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 417 } 418 #define ext4_update_tstamp(es, tstamp) \ 419 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 420 ktime_get_real_seconds()) 421 #define ext4_get_tstamp(es, tstamp) \ 422 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 423 424 /* 425 * The del_gendisk() function uninitializes the disk-specific data 426 * structures, including the bdi structure, without telling anyone 427 * else. Once this happens, any attempt to call mark_buffer_dirty() 428 * (for example, by ext4_commit_super), will cause a kernel OOPS. 429 * This is a kludge to prevent these oops until we can put in a proper 430 * hook in del_gendisk() to inform the VFS and file system layers. 431 */ 432 static int block_device_ejected(struct super_block *sb) 433 { 434 struct inode *bd_inode = sb->s_bdev->bd_inode; 435 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 436 437 return bdi->dev == NULL; 438 } 439 440 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 441 { 442 struct super_block *sb = journal->j_private; 443 struct ext4_sb_info *sbi = EXT4_SB(sb); 444 int error = is_journal_aborted(journal); 445 struct ext4_journal_cb_entry *jce; 446 447 BUG_ON(txn->t_state == T_FINISHED); 448 449 ext4_process_freed_data(sb, txn->t_tid); 450 451 spin_lock(&sbi->s_md_lock); 452 while (!list_empty(&txn->t_private_list)) { 453 jce = list_entry(txn->t_private_list.next, 454 struct ext4_journal_cb_entry, jce_list); 455 list_del_init(&jce->jce_list); 456 spin_unlock(&sbi->s_md_lock); 457 jce->jce_func(sb, jce, error); 458 spin_lock(&sbi->s_md_lock); 459 } 460 spin_unlock(&sbi->s_md_lock); 461 } 462 463 /* 464 * This writepage callback for write_cache_pages() 465 * takes care of a few cases after page cleaning. 466 * 467 * write_cache_pages() already checks for dirty pages 468 * and calls clear_page_dirty_for_io(), which we want, 469 * to write protect the pages. 470 * 471 * However, we may have to redirty a page (see below.) 472 */ 473 static int ext4_journalled_writepage_callback(struct page *page, 474 struct writeback_control *wbc, 475 void *data) 476 { 477 transaction_t *transaction = (transaction_t *) data; 478 struct buffer_head *bh, *head; 479 struct journal_head *jh; 480 481 bh = head = page_buffers(page); 482 do { 483 /* 484 * We have to redirty a page in these cases: 485 * 1) If buffer is dirty, it means the page was dirty because it 486 * contains a buffer that needs checkpointing. So the dirty bit 487 * needs to be preserved so that checkpointing writes the buffer 488 * properly. 489 * 2) If buffer is not part of the committing transaction 490 * (we may have just accidentally come across this buffer because 491 * inode range tracking is not exact) or if the currently running 492 * transaction already contains this buffer as well, dirty bit 493 * needs to be preserved so that the buffer gets writeprotected 494 * properly on running transaction's commit. 495 */ 496 jh = bh2jh(bh); 497 if (buffer_dirty(bh) || 498 (jh && (jh->b_transaction != transaction || 499 jh->b_next_transaction))) { 500 redirty_page_for_writepage(wbc, page); 501 goto out; 502 } 503 } while ((bh = bh->b_this_page) != head); 504 505 out: 506 return AOP_WRITEPAGE_ACTIVATE; 507 } 508 509 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 510 { 511 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 512 struct writeback_control wbc = { 513 .sync_mode = WB_SYNC_ALL, 514 .nr_to_write = LONG_MAX, 515 .range_start = jinode->i_dirty_start, 516 .range_end = jinode->i_dirty_end, 517 }; 518 519 return write_cache_pages(mapping, &wbc, 520 ext4_journalled_writepage_callback, 521 jinode->i_transaction); 522 } 523 524 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 525 { 526 int ret; 527 528 if (ext4_should_journal_data(jinode->i_vfs_inode)) 529 ret = ext4_journalled_submit_inode_data_buffers(jinode); 530 else 531 ret = jbd2_journal_submit_inode_data_buffers(jinode); 532 533 return ret; 534 } 535 536 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 537 { 538 int ret = 0; 539 540 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 541 ret = jbd2_journal_finish_inode_data_buffers(jinode); 542 543 return ret; 544 } 545 546 static bool system_going_down(void) 547 { 548 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 549 || system_state == SYSTEM_RESTART; 550 } 551 552 struct ext4_err_translation { 553 int code; 554 int errno; 555 }; 556 557 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 558 559 static struct ext4_err_translation err_translation[] = { 560 EXT4_ERR_TRANSLATE(EIO), 561 EXT4_ERR_TRANSLATE(ENOMEM), 562 EXT4_ERR_TRANSLATE(EFSBADCRC), 563 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 564 EXT4_ERR_TRANSLATE(ENOSPC), 565 EXT4_ERR_TRANSLATE(ENOKEY), 566 EXT4_ERR_TRANSLATE(EROFS), 567 EXT4_ERR_TRANSLATE(EFBIG), 568 EXT4_ERR_TRANSLATE(EEXIST), 569 EXT4_ERR_TRANSLATE(ERANGE), 570 EXT4_ERR_TRANSLATE(EOVERFLOW), 571 EXT4_ERR_TRANSLATE(EBUSY), 572 EXT4_ERR_TRANSLATE(ENOTDIR), 573 EXT4_ERR_TRANSLATE(ENOTEMPTY), 574 EXT4_ERR_TRANSLATE(ESHUTDOWN), 575 EXT4_ERR_TRANSLATE(EFAULT), 576 }; 577 578 static int ext4_errno_to_code(int errno) 579 { 580 int i; 581 582 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 583 if (err_translation[i].errno == errno) 584 return err_translation[i].code; 585 return EXT4_ERR_UNKNOWN; 586 } 587 588 static void save_error_info(struct super_block *sb, int error, 589 __u32 ino, __u64 block, 590 const char *func, unsigned int line) 591 { 592 struct ext4_sb_info *sbi = EXT4_SB(sb); 593 594 /* We default to EFSCORRUPTED error... */ 595 if (error == 0) 596 error = EFSCORRUPTED; 597 598 spin_lock(&sbi->s_error_lock); 599 sbi->s_add_error_count++; 600 sbi->s_last_error_code = error; 601 sbi->s_last_error_line = line; 602 sbi->s_last_error_ino = ino; 603 sbi->s_last_error_block = block; 604 sbi->s_last_error_func = func; 605 sbi->s_last_error_time = ktime_get_real_seconds(); 606 if (!sbi->s_first_error_time) { 607 sbi->s_first_error_code = error; 608 sbi->s_first_error_line = line; 609 sbi->s_first_error_ino = ino; 610 sbi->s_first_error_block = block; 611 sbi->s_first_error_func = func; 612 sbi->s_first_error_time = sbi->s_last_error_time; 613 } 614 spin_unlock(&sbi->s_error_lock); 615 } 616 617 /* Deal with the reporting of failure conditions on a filesystem such as 618 * inconsistencies detected or read IO failures. 619 * 620 * On ext2, we can store the error state of the filesystem in the 621 * superblock. That is not possible on ext4, because we may have other 622 * write ordering constraints on the superblock which prevent us from 623 * writing it out straight away; and given that the journal is about to 624 * be aborted, we can't rely on the current, or future, transactions to 625 * write out the superblock safely. 626 * 627 * We'll just use the jbd2_journal_abort() error code to record an error in 628 * the journal instead. On recovery, the journal will complain about 629 * that error until we've noted it down and cleared it. 630 * 631 * If force_ro is set, we unconditionally force the filesystem into an 632 * ABORT|READONLY state, unless the error response on the fs has been set to 633 * panic in which case we take the easy way out and panic immediately. This is 634 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 635 * at a critical moment in log management. 636 */ 637 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 638 __u32 ino, __u64 block, 639 const char *func, unsigned int line) 640 { 641 journal_t *journal = EXT4_SB(sb)->s_journal; 642 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 643 644 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 645 if (test_opt(sb, WARN_ON_ERROR)) 646 WARN_ON_ONCE(1); 647 648 if (!continue_fs && !sb_rdonly(sb)) { 649 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 650 if (journal) 651 jbd2_journal_abort(journal, -EIO); 652 } 653 654 if (!bdev_read_only(sb->s_bdev)) { 655 save_error_info(sb, error, ino, block, func, line); 656 /* 657 * In case the fs should keep running, we need to writeout 658 * superblock through the journal. Due to lock ordering 659 * constraints, it may not be safe to do it right here so we 660 * defer superblock flushing to a workqueue. 661 */ 662 if (continue_fs) 663 schedule_work(&EXT4_SB(sb)->s_error_work); 664 else 665 ext4_commit_super(sb); 666 } 667 668 /* 669 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 670 * could panic during 'reboot -f' as the underlying device got already 671 * disabled. 672 */ 673 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 674 panic("EXT4-fs (device %s): panic forced after error\n", 675 sb->s_id); 676 } 677 678 if (sb_rdonly(sb) || continue_fs) 679 return; 680 681 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 682 /* 683 * Make sure updated value of ->s_mount_flags will be visible before 684 * ->s_flags update 685 */ 686 smp_wmb(); 687 sb->s_flags |= SB_RDONLY; 688 } 689 690 static void flush_stashed_error_work(struct work_struct *work) 691 { 692 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 693 s_error_work); 694 journal_t *journal = sbi->s_journal; 695 handle_t *handle; 696 697 /* 698 * If the journal is still running, we have to write out superblock 699 * through the journal to avoid collisions of other journalled sb 700 * updates. 701 * 702 * We use directly jbd2 functions here to avoid recursing back into 703 * ext4 error handling code during handling of previous errors. 704 */ 705 if (!sb_rdonly(sbi->s_sb) && journal) { 706 struct buffer_head *sbh = sbi->s_sbh; 707 handle = jbd2_journal_start(journal, 1); 708 if (IS_ERR(handle)) 709 goto write_directly; 710 if (jbd2_journal_get_write_access(handle, sbh)) { 711 jbd2_journal_stop(handle); 712 goto write_directly; 713 } 714 ext4_update_super(sbi->s_sb); 715 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 716 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 717 "superblock detected"); 718 clear_buffer_write_io_error(sbh); 719 set_buffer_uptodate(sbh); 720 } 721 722 if (jbd2_journal_dirty_metadata(handle, sbh)) { 723 jbd2_journal_stop(handle); 724 goto write_directly; 725 } 726 jbd2_journal_stop(handle); 727 ext4_notify_error_sysfs(sbi); 728 return; 729 } 730 write_directly: 731 /* 732 * Write through journal failed. Write sb directly to get error info 733 * out and hope for the best. 734 */ 735 ext4_commit_super(sbi->s_sb); 736 ext4_notify_error_sysfs(sbi); 737 } 738 739 #define ext4_error_ratelimit(sb) \ 740 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 741 "EXT4-fs error") 742 743 void __ext4_error(struct super_block *sb, const char *function, 744 unsigned int line, bool force_ro, int error, __u64 block, 745 const char *fmt, ...) 746 { 747 struct va_format vaf; 748 va_list args; 749 750 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 751 return; 752 753 trace_ext4_error(sb, function, line); 754 if (ext4_error_ratelimit(sb)) { 755 va_start(args, fmt); 756 vaf.fmt = fmt; 757 vaf.va = &args; 758 printk(KERN_CRIT 759 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 760 sb->s_id, function, line, current->comm, &vaf); 761 va_end(args); 762 } 763 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 764 } 765 766 void __ext4_error_inode(struct inode *inode, const char *function, 767 unsigned int line, ext4_fsblk_t block, int error, 768 const char *fmt, ...) 769 { 770 va_list args; 771 struct va_format vaf; 772 773 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 774 return; 775 776 trace_ext4_error(inode->i_sb, function, line); 777 if (ext4_error_ratelimit(inode->i_sb)) { 778 va_start(args, fmt); 779 vaf.fmt = fmt; 780 vaf.va = &args; 781 if (block) 782 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 783 "inode #%lu: block %llu: comm %s: %pV\n", 784 inode->i_sb->s_id, function, line, inode->i_ino, 785 block, current->comm, &vaf); 786 else 787 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 788 "inode #%lu: comm %s: %pV\n", 789 inode->i_sb->s_id, function, line, inode->i_ino, 790 current->comm, &vaf); 791 va_end(args); 792 } 793 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 794 function, line); 795 } 796 797 void __ext4_error_file(struct file *file, const char *function, 798 unsigned int line, ext4_fsblk_t block, 799 const char *fmt, ...) 800 { 801 va_list args; 802 struct va_format vaf; 803 struct inode *inode = file_inode(file); 804 char pathname[80], *path; 805 806 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 807 return; 808 809 trace_ext4_error(inode->i_sb, function, line); 810 if (ext4_error_ratelimit(inode->i_sb)) { 811 path = file_path(file, pathname, sizeof(pathname)); 812 if (IS_ERR(path)) 813 path = "(unknown)"; 814 va_start(args, fmt); 815 vaf.fmt = fmt; 816 vaf.va = &args; 817 if (block) 818 printk(KERN_CRIT 819 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 820 "block %llu: comm %s: path %s: %pV\n", 821 inode->i_sb->s_id, function, line, inode->i_ino, 822 block, current->comm, path, &vaf); 823 else 824 printk(KERN_CRIT 825 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 826 "comm %s: path %s: %pV\n", 827 inode->i_sb->s_id, function, line, inode->i_ino, 828 current->comm, path, &vaf); 829 va_end(args); 830 } 831 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 832 function, line); 833 } 834 835 const char *ext4_decode_error(struct super_block *sb, int errno, 836 char nbuf[16]) 837 { 838 char *errstr = NULL; 839 840 switch (errno) { 841 case -EFSCORRUPTED: 842 errstr = "Corrupt filesystem"; 843 break; 844 case -EFSBADCRC: 845 errstr = "Filesystem failed CRC"; 846 break; 847 case -EIO: 848 errstr = "IO failure"; 849 break; 850 case -ENOMEM: 851 errstr = "Out of memory"; 852 break; 853 case -EROFS: 854 if (!sb || (EXT4_SB(sb)->s_journal && 855 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 856 errstr = "Journal has aborted"; 857 else 858 errstr = "Readonly filesystem"; 859 break; 860 default: 861 /* If the caller passed in an extra buffer for unknown 862 * errors, textualise them now. Else we just return 863 * NULL. */ 864 if (nbuf) { 865 /* Check for truncated error codes... */ 866 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 867 errstr = nbuf; 868 } 869 break; 870 } 871 872 return errstr; 873 } 874 875 /* __ext4_std_error decodes expected errors from journaling functions 876 * automatically and invokes the appropriate error response. */ 877 878 void __ext4_std_error(struct super_block *sb, const char *function, 879 unsigned int line, int errno) 880 { 881 char nbuf[16]; 882 const char *errstr; 883 884 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 885 return; 886 887 /* Special case: if the error is EROFS, and we're not already 888 * inside a transaction, then there's really no point in logging 889 * an error. */ 890 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 891 return; 892 893 if (ext4_error_ratelimit(sb)) { 894 errstr = ext4_decode_error(sb, errno, nbuf); 895 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 896 sb->s_id, function, line, errstr); 897 } 898 899 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 900 } 901 902 void __ext4_msg(struct super_block *sb, 903 const char *prefix, const char *fmt, ...) 904 { 905 struct va_format vaf; 906 va_list args; 907 908 atomic_inc(&EXT4_SB(sb)->s_msg_count); 909 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs")) 910 return; 911 912 va_start(args, fmt); 913 vaf.fmt = fmt; 914 vaf.va = &args; 915 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 916 va_end(args); 917 } 918 919 static int ext4_warning_ratelimit(struct super_block *sb) 920 { 921 atomic_inc(&EXT4_SB(sb)->s_warning_count); 922 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 923 "EXT4-fs warning"); 924 } 925 926 void __ext4_warning(struct super_block *sb, const char *function, 927 unsigned int line, const char *fmt, ...) 928 { 929 struct va_format vaf; 930 va_list args; 931 932 if (!ext4_warning_ratelimit(sb)) 933 return; 934 935 va_start(args, fmt); 936 vaf.fmt = fmt; 937 vaf.va = &args; 938 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 939 sb->s_id, function, line, &vaf); 940 va_end(args); 941 } 942 943 void __ext4_warning_inode(const struct inode *inode, const char *function, 944 unsigned int line, const char *fmt, ...) 945 { 946 struct va_format vaf; 947 va_list args; 948 949 if (!ext4_warning_ratelimit(inode->i_sb)) 950 return; 951 952 va_start(args, fmt); 953 vaf.fmt = fmt; 954 vaf.va = &args; 955 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 956 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 957 function, line, inode->i_ino, current->comm, &vaf); 958 va_end(args); 959 } 960 961 void __ext4_grp_locked_error(const char *function, unsigned int line, 962 struct super_block *sb, ext4_group_t grp, 963 unsigned long ino, ext4_fsblk_t block, 964 const char *fmt, ...) 965 __releases(bitlock) 966 __acquires(bitlock) 967 { 968 struct va_format vaf; 969 va_list args; 970 971 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 972 return; 973 974 trace_ext4_error(sb, function, line); 975 if (ext4_error_ratelimit(sb)) { 976 va_start(args, fmt); 977 vaf.fmt = fmt; 978 vaf.va = &args; 979 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 980 sb->s_id, function, line, grp); 981 if (ino) 982 printk(KERN_CONT "inode %lu: ", ino); 983 if (block) 984 printk(KERN_CONT "block %llu:", 985 (unsigned long long) block); 986 printk(KERN_CONT "%pV\n", &vaf); 987 va_end(args); 988 } 989 990 if (test_opt(sb, ERRORS_CONT)) { 991 if (test_opt(sb, WARN_ON_ERROR)) 992 WARN_ON_ONCE(1); 993 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 994 if (!bdev_read_only(sb->s_bdev)) { 995 save_error_info(sb, EFSCORRUPTED, ino, block, function, 996 line); 997 schedule_work(&EXT4_SB(sb)->s_error_work); 998 } 999 return; 1000 } 1001 ext4_unlock_group(sb, grp); 1002 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1003 /* 1004 * We only get here in the ERRORS_RO case; relocking the group 1005 * may be dangerous, but nothing bad will happen since the 1006 * filesystem will have already been marked read/only and the 1007 * journal has been aborted. We return 1 as a hint to callers 1008 * who might what to use the return value from 1009 * ext4_grp_locked_error() to distinguish between the 1010 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1011 * aggressively from the ext4 function in question, with a 1012 * more appropriate error code. 1013 */ 1014 ext4_lock_group(sb, grp); 1015 return; 1016 } 1017 1018 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1019 ext4_group_t group, 1020 unsigned int flags) 1021 { 1022 struct ext4_sb_info *sbi = EXT4_SB(sb); 1023 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1024 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1025 int ret; 1026 1027 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1028 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1029 &grp->bb_state); 1030 if (!ret) 1031 percpu_counter_sub(&sbi->s_freeclusters_counter, 1032 grp->bb_free); 1033 } 1034 1035 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1036 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1037 &grp->bb_state); 1038 if (!ret && gdp) { 1039 int count; 1040 1041 count = ext4_free_inodes_count(sb, gdp); 1042 percpu_counter_sub(&sbi->s_freeinodes_counter, 1043 count); 1044 } 1045 } 1046 } 1047 1048 void ext4_update_dynamic_rev(struct super_block *sb) 1049 { 1050 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1051 1052 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1053 return; 1054 1055 ext4_warning(sb, 1056 "updating to rev %d because of new feature flag, " 1057 "running e2fsck is recommended", 1058 EXT4_DYNAMIC_REV); 1059 1060 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1061 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1062 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1063 /* leave es->s_feature_*compat flags alone */ 1064 /* es->s_uuid will be set by e2fsck if empty */ 1065 1066 /* 1067 * The rest of the superblock fields should be zero, and if not it 1068 * means they are likely already in use, so leave them alone. We 1069 * can leave it up to e2fsck to clean up any inconsistencies there. 1070 */ 1071 } 1072 1073 /* 1074 * Open the external journal device 1075 */ 1076 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 1077 { 1078 struct block_device *bdev; 1079 1080 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 1081 if (IS_ERR(bdev)) 1082 goto fail; 1083 return bdev; 1084 1085 fail: 1086 ext4_msg(sb, KERN_ERR, 1087 "failed to open journal device unknown-block(%u,%u) %ld", 1088 MAJOR(dev), MINOR(dev), PTR_ERR(bdev)); 1089 return NULL; 1090 } 1091 1092 /* 1093 * Release the journal device 1094 */ 1095 static void ext4_blkdev_put(struct block_device *bdev) 1096 { 1097 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1098 } 1099 1100 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 1101 { 1102 struct block_device *bdev; 1103 bdev = sbi->s_journal_bdev; 1104 if (bdev) { 1105 ext4_blkdev_put(bdev); 1106 sbi->s_journal_bdev = NULL; 1107 } 1108 } 1109 1110 static inline struct inode *orphan_list_entry(struct list_head *l) 1111 { 1112 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1113 } 1114 1115 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1116 { 1117 struct list_head *l; 1118 1119 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1120 le32_to_cpu(sbi->s_es->s_last_orphan)); 1121 1122 printk(KERN_ERR "sb_info orphan list:\n"); 1123 list_for_each(l, &sbi->s_orphan) { 1124 struct inode *inode = orphan_list_entry(l); 1125 printk(KERN_ERR " " 1126 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1127 inode->i_sb->s_id, inode->i_ino, inode, 1128 inode->i_mode, inode->i_nlink, 1129 NEXT_ORPHAN(inode)); 1130 } 1131 } 1132 1133 #ifdef CONFIG_QUOTA 1134 static int ext4_quota_off(struct super_block *sb, int type); 1135 1136 static inline void ext4_quota_off_umount(struct super_block *sb) 1137 { 1138 int type; 1139 1140 /* Use our quota_off function to clear inode flags etc. */ 1141 for (type = 0; type < EXT4_MAXQUOTAS; type++) 1142 ext4_quota_off(sb, type); 1143 } 1144 1145 /* 1146 * This is a helper function which is used in the mount/remount 1147 * codepaths (which holds s_umount) to fetch the quota file name. 1148 */ 1149 static inline char *get_qf_name(struct super_block *sb, 1150 struct ext4_sb_info *sbi, 1151 int type) 1152 { 1153 return rcu_dereference_protected(sbi->s_qf_names[type], 1154 lockdep_is_held(&sb->s_umount)); 1155 } 1156 #else 1157 static inline void ext4_quota_off_umount(struct super_block *sb) 1158 { 1159 } 1160 #endif 1161 1162 static void ext4_put_super(struct super_block *sb) 1163 { 1164 struct ext4_sb_info *sbi = EXT4_SB(sb); 1165 struct ext4_super_block *es = sbi->s_es; 1166 struct buffer_head **group_desc; 1167 struct flex_groups **flex_groups; 1168 int aborted = 0; 1169 int i, err; 1170 1171 ext4_unregister_li_request(sb); 1172 ext4_quota_off_umount(sb); 1173 1174 flush_work(&sbi->s_error_work); 1175 destroy_workqueue(sbi->rsv_conversion_wq); 1176 1177 /* 1178 * Unregister sysfs before destroying jbd2 journal. 1179 * Since we could still access attr_journal_task attribute via sysfs 1180 * path which could have sbi->s_journal->j_task as NULL 1181 */ 1182 ext4_unregister_sysfs(sb); 1183 1184 if (sbi->s_journal) { 1185 aborted = is_journal_aborted(sbi->s_journal); 1186 err = jbd2_journal_destroy(sbi->s_journal); 1187 sbi->s_journal = NULL; 1188 if ((err < 0) && !aborted) { 1189 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1190 } 1191 } 1192 1193 ext4_es_unregister_shrinker(sbi); 1194 del_timer_sync(&sbi->s_err_report); 1195 ext4_release_system_zone(sb); 1196 ext4_mb_release(sb); 1197 ext4_ext_release(sb); 1198 1199 if (!sb_rdonly(sb) && !aborted) { 1200 ext4_clear_feature_journal_needs_recovery(sb); 1201 es->s_state = cpu_to_le16(sbi->s_mount_state); 1202 } 1203 if (!sb_rdonly(sb)) 1204 ext4_commit_super(sb); 1205 1206 rcu_read_lock(); 1207 group_desc = rcu_dereference(sbi->s_group_desc); 1208 for (i = 0; i < sbi->s_gdb_count; i++) 1209 brelse(group_desc[i]); 1210 kvfree(group_desc); 1211 flex_groups = rcu_dereference(sbi->s_flex_groups); 1212 if (flex_groups) { 1213 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1214 kvfree(flex_groups[i]); 1215 kvfree(flex_groups); 1216 } 1217 rcu_read_unlock(); 1218 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1219 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1220 percpu_counter_destroy(&sbi->s_dirs_counter); 1221 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1222 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1223 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1224 #ifdef CONFIG_QUOTA 1225 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1226 kfree(get_qf_name(sb, sbi, i)); 1227 #endif 1228 1229 /* Debugging code just in case the in-memory inode orphan list 1230 * isn't empty. The on-disk one can be non-empty if we've 1231 * detected an error and taken the fs readonly, but the 1232 * in-memory list had better be clean by this point. */ 1233 if (!list_empty(&sbi->s_orphan)) 1234 dump_orphan_list(sb, sbi); 1235 ASSERT(list_empty(&sbi->s_orphan)); 1236 1237 sync_blockdev(sb->s_bdev); 1238 invalidate_bdev(sb->s_bdev); 1239 if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) { 1240 /* 1241 * Invalidate the journal device's buffers. We don't want them 1242 * floating about in memory - the physical journal device may 1243 * hotswapped, and it breaks the `ro-after' testing code. 1244 */ 1245 sync_blockdev(sbi->s_journal_bdev); 1246 invalidate_bdev(sbi->s_journal_bdev); 1247 ext4_blkdev_remove(sbi); 1248 } 1249 1250 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1251 sbi->s_ea_inode_cache = NULL; 1252 1253 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1254 sbi->s_ea_block_cache = NULL; 1255 1256 ext4_stop_mmpd(sbi); 1257 1258 brelse(sbi->s_sbh); 1259 sb->s_fs_info = NULL; 1260 /* 1261 * Now that we are completely done shutting down the 1262 * superblock, we need to actually destroy the kobject. 1263 */ 1264 kobject_put(&sbi->s_kobj); 1265 wait_for_completion(&sbi->s_kobj_unregister); 1266 if (sbi->s_chksum_driver) 1267 crypto_free_shash(sbi->s_chksum_driver); 1268 kfree(sbi->s_blockgroup_lock); 1269 fs_put_dax(sbi->s_daxdev); 1270 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1271 #ifdef CONFIG_UNICODE 1272 utf8_unload(sb->s_encoding); 1273 #endif 1274 kfree(sbi); 1275 } 1276 1277 static struct kmem_cache *ext4_inode_cachep; 1278 1279 /* 1280 * Called inside transaction, so use GFP_NOFS 1281 */ 1282 static struct inode *ext4_alloc_inode(struct super_block *sb) 1283 { 1284 struct ext4_inode_info *ei; 1285 1286 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS); 1287 if (!ei) 1288 return NULL; 1289 1290 inode_set_iversion(&ei->vfs_inode, 1); 1291 spin_lock_init(&ei->i_raw_lock); 1292 INIT_LIST_HEAD(&ei->i_prealloc_list); 1293 atomic_set(&ei->i_prealloc_active, 0); 1294 spin_lock_init(&ei->i_prealloc_lock); 1295 ext4_es_init_tree(&ei->i_es_tree); 1296 rwlock_init(&ei->i_es_lock); 1297 INIT_LIST_HEAD(&ei->i_es_list); 1298 ei->i_es_all_nr = 0; 1299 ei->i_es_shk_nr = 0; 1300 ei->i_es_shrink_lblk = 0; 1301 ei->i_reserved_data_blocks = 0; 1302 spin_lock_init(&(ei->i_block_reservation_lock)); 1303 ext4_init_pending_tree(&ei->i_pending_tree); 1304 #ifdef CONFIG_QUOTA 1305 ei->i_reserved_quota = 0; 1306 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1307 #endif 1308 ei->jinode = NULL; 1309 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1310 spin_lock_init(&ei->i_completed_io_lock); 1311 ei->i_sync_tid = 0; 1312 ei->i_datasync_tid = 0; 1313 atomic_set(&ei->i_unwritten, 0); 1314 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1315 ext4_fc_init_inode(&ei->vfs_inode); 1316 mutex_init(&ei->i_fc_lock); 1317 return &ei->vfs_inode; 1318 } 1319 1320 static int ext4_drop_inode(struct inode *inode) 1321 { 1322 int drop = generic_drop_inode(inode); 1323 1324 if (!drop) 1325 drop = fscrypt_drop_inode(inode); 1326 1327 trace_ext4_drop_inode(inode, drop); 1328 return drop; 1329 } 1330 1331 static void ext4_free_in_core_inode(struct inode *inode) 1332 { 1333 fscrypt_free_inode(inode); 1334 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1335 pr_warn("%s: inode %ld still in fc list", 1336 __func__, inode->i_ino); 1337 } 1338 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1339 } 1340 1341 static void ext4_destroy_inode(struct inode *inode) 1342 { 1343 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1344 ext4_msg(inode->i_sb, KERN_ERR, 1345 "Inode %lu (%p): orphan list check failed!", 1346 inode->i_ino, EXT4_I(inode)); 1347 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1348 EXT4_I(inode), sizeof(struct ext4_inode_info), 1349 true); 1350 dump_stack(); 1351 } 1352 } 1353 1354 static void init_once(void *foo) 1355 { 1356 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo; 1357 1358 INIT_LIST_HEAD(&ei->i_orphan); 1359 init_rwsem(&ei->xattr_sem); 1360 init_rwsem(&ei->i_data_sem); 1361 inode_init_once(&ei->vfs_inode); 1362 ext4_fc_init_inode(&ei->vfs_inode); 1363 } 1364 1365 static int __init init_inodecache(void) 1366 { 1367 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1368 sizeof(struct ext4_inode_info), 0, 1369 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1370 SLAB_ACCOUNT), 1371 offsetof(struct ext4_inode_info, i_data), 1372 sizeof_field(struct ext4_inode_info, i_data), 1373 init_once); 1374 if (ext4_inode_cachep == NULL) 1375 return -ENOMEM; 1376 return 0; 1377 } 1378 1379 static void destroy_inodecache(void) 1380 { 1381 /* 1382 * Make sure all delayed rcu free inodes are flushed before we 1383 * destroy cache. 1384 */ 1385 rcu_barrier(); 1386 kmem_cache_destroy(ext4_inode_cachep); 1387 } 1388 1389 void ext4_clear_inode(struct inode *inode) 1390 { 1391 ext4_fc_del(inode); 1392 invalidate_inode_buffers(inode); 1393 clear_inode(inode); 1394 ext4_discard_preallocations(inode, 0); 1395 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1396 dquot_drop(inode); 1397 if (EXT4_I(inode)->jinode) { 1398 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1399 EXT4_I(inode)->jinode); 1400 jbd2_free_inode(EXT4_I(inode)->jinode); 1401 EXT4_I(inode)->jinode = NULL; 1402 } 1403 fscrypt_put_encryption_info(inode); 1404 fsverity_cleanup_inode(inode); 1405 } 1406 1407 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1408 u64 ino, u32 generation) 1409 { 1410 struct inode *inode; 1411 1412 /* 1413 * Currently we don't know the generation for parent directory, so 1414 * a generation of 0 means "accept any" 1415 */ 1416 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1417 if (IS_ERR(inode)) 1418 return ERR_CAST(inode); 1419 if (generation && inode->i_generation != generation) { 1420 iput(inode); 1421 return ERR_PTR(-ESTALE); 1422 } 1423 1424 return inode; 1425 } 1426 1427 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1428 int fh_len, int fh_type) 1429 { 1430 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1431 ext4_nfs_get_inode); 1432 } 1433 1434 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1435 int fh_len, int fh_type) 1436 { 1437 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1438 ext4_nfs_get_inode); 1439 } 1440 1441 static int ext4_nfs_commit_metadata(struct inode *inode) 1442 { 1443 struct writeback_control wbc = { 1444 .sync_mode = WB_SYNC_ALL 1445 }; 1446 1447 trace_ext4_nfs_commit_metadata(inode); 1448 return ext4_write_inode(inode, &wbc); 1449 } 1450 1451 #ifdef CONFIG_FS_ENCRYPTION 1452 static int ext4_get_context(struct inode *inode, void *ctx, size_t len) 1453 { 1454 return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, 1455 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len); 1456 } 1457 1458 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len, 1459 void *fs_data) 1460 { 1461 handle_t *handle = fs_data; 1462 int res, res2, credits, retries = 0; 1463 1464 /* 1465 * Encrypting the root directory is not allowed because e2fsck expects 1466 * lost+found to exist and be unencrypted, and encrypting the root 1467 * directory would imply encrypting the lost+found directory as well as 1468 * the filename "lost+found" itself. 1469 */ 1470 if (inode->i_ino == EXT4_ROOT_INO) 1471 return -EPERM; 1472 1473 if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode))) 1474 return -EINVAL; 1475 1476 if (ext4_test_inode_flag(inode, EXT4_INODE_DAX)) 1477 return -EOPNOTSUPP; 1478 1479 res = ext4_convert_inline_data(inode); 1480 if (res) 1481 return res; 1482 1483 /* 1484 * If a journal handle was specified, then the encryption context is 1485 * being set on a new inode via inheritance and is part of a larger 1486 * transaction to create the inode. Otherwise the encryption context is 1487 * being set on an existing inode in its own transaction. Only in the 1488 * latter case should the "retry on ENOSPC" logic be used. 1489 */ 1490 1491 if (handle) { 1492 res = ext4_xattr_set_handle(handle, inode, 1493 EXT4_XATTR_INDEX_ENCRYPTION, 1494 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1495 ctx, len, 0); 1496 if (!res) { 1497 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1498 ext4_clear_inode_state(inode, 1499 EXT4_STATE_MAY_INLINE_DATA); 1500 /* 1501 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1502 * S_DAX may be disabled 1503 */ 1504 ext4_set_inode_flags(inode, false); 1505 } 1506 return res; 1507 } 1508 1509 res = dquot_initialize(inode); 1510 if (res) 1511 return res; 1512 retry: 1513 res = ext4_xattr_set_credits(inode, len, false /* is_create */, 1514 &credits); 1515 if (res) 1516 return res; 1517 1518 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); 1519 if (IS_ERR(handle)) 1520 return PTR_ERR(handle); 1521 1522 res = ext4_xattr_set_handle(handle, inode, 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 /* 1528 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1529 * S_DAX may be disabled 1530 */ 1531 ext4_set_inode_flags(inode, false); 1532 res = ext4_mark_inode_dirty(handle, inode); 1533 if (res) 1534 EXT4_ERROR_INODE(inode, "Failed to mark inode dirty"); 1535 } 1536 res2 = ext4_journal_stop(handle); 1537 1538 if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 1539 goto retry; 1540 if (!res) 1541 res = res2; 1542 return res; 1543 } 1544 1545 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb) 1546 { 1547 return EXT4_SB(sb)->s_dummy_enc_policy.policy; 1548 } 1549 1550 static bool ext4_has_stable_inodes(struct super_block *sb) 1551 { 1552 return ext4_has_feature_stable_inodes(sb); 1553 } 1554 1555 static void ext4_get_ino_and_lblk_bits(struct super_block *sb, 1556 int *ino_bits_ret, int *lblk_bits_ret) 1557 { 1558 *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count); 1559 *lblk_bits_ret = 8 * sizeof(ext4_lblk_t); 1560 } 1561 1562 static const struct fscrypt_operations ext4_cryptops = { 1563 .key_prefix = "ext4:", 1564 .get_context = ext4_get_context, 1565 .set_context = ext4_set_context, 1566 .get_dummy_policy = ext4_get_dummy_policy, 1567 .empty_dir = ext4_empty_dir, 1568 .max_namelen = EXT4_NAME_LEN, 1569 .has_stable_inodes = ext4_has_stable_inodes, 1570 .get_ino_and_lblk_bits = ext4_get_ino_and_lblk_bits, 1571 }; 1572 #endif 1573 1574 #ifdef CONFIG_QUOTA 1575 static const char * const quotatypes[] = INITQFNAMES; 1576 #define QTYPE2NAME(t) (quotatypes[t]) 1577 1578 static int ext4_write_dquot(struct dquot *dquot); 1579 static int ext4_acquire_dquot(struct dquot *dquot); 1580 static int ext4_release_dquot(struct dquot *dquot); 1581 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1582 static int ext4_write_info(struct super_block *sb, int type); 1583 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1584 const struct path *path); 1585 static int ext4_quota_on_mount(struct super_block *sb, int type); 1586 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1587 size_t len, loff_t off); 1588 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1589 const char *data, size_t len, loff_t off); 1590 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1591 unsigned int flags); 1592 static int ext4_enable_quotas(struct super_block *sb); 1593 1594 static struct dquot **ext4_get_dquots(struct inode *inode) 1595 { 1596 return EXT4_I(inode)->i_dquot; 1597 } 1598 1599 static const struct dquot_operations ext4_quota_operations = { 1600 .get_reserved_space = ext4_get_reserved_space, 1601 .write_dquot = ext4_write_dquot, 1602 .acquire_dquot = ext4_acquire_dquot, 1603 .release_dquot = ext4_release_dquot, 1604 .mark_dirty = ext4_mark_dquot_dirty, 1605 .write_info = ext4_write_info, 1606 .alloc_dquot = dquot_alloc, 1607 .destroy_dquot = dquot_destroy, 1608 .get_projid = ext4_get_projid, 1609 .get_inode_usage = ext4_get_inode_usage, 1610 .get_next_id = dquot_get_next_id, 1611 }; 1612 1613 static const struct quotactl_ops ext4_qctl_operations = { 1614 .quota_on = ext4_quota_on, 1615 .quota_off = ext4_quota_off, 1616 .quota_sync = dquot_quota_sync, 1617 .get_state = dquot_get_state, 1618 .set_info = dquot_set_dqinfo, 1619 .get_dqblk = dquot_get_dqblk, 1620 .set_dqblk = dquot_set_dqblk, 1621 .get_nextdqblk = dquot_get_next_dqblk, 1622 }; 1623 #endif 1624 1625 static const struct super_operations ext4_sops = { 1626 .alloc_inode = ext4_alloc_inode, 1627 .free_inode = ext4_free_in_core_inode, 1628 .destroy_inode = ext4_destroy_inode, 1629 .write_inode = ext4_write_inode, 1630 .dirty_inode = ext4_dirty_inode, 1631 .drop_inode = ext4_drop_inode, 1632 .evict_inode = ext4_evict_inode, 1633 .put_super = ext4_put_super, 1634 .sync_fs = ext4_sync_fs, 1635 .freeze_fs = ext4_freeze, 1636 .unfreeze_fs = ext4_unfreeze, 1637 .statfs = ext4_statfs, 1638 .remount_fs = ext4_remount, 1639 .show_options = ext4_show_options, 1640 #ifdef CONFIG_QUOTA 1641 .quota_read = ext4_quota_read, 1642 .quota_write = ext4_quota_write, 1643 .get_dquots = ext4_get_dquots, 1644 #endif 1645 }; 1646 1647 static const struct export_operations ext4_export_ops = { 1648 .fh_to_dentry = ext4_fh_to_dentry, 1649 .fh_to_parent = ext4_fh_to_parent, 1650 .get_parent = ext4_get_parent, 1651 .commit_metadata = ext4_nfs_commit_metadata, 1652 }; 1653 1654 enum { 1655 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1656 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro, 1657 Opt_nouid32, Opt_debug, Opt_removed, 1658 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, 1659 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1660 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1661 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1662 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1663 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1664 Opt_inlinecrypt, 1665 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota, 1666 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota, 1667 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1668 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, 1669 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1670 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1671 Opt_nowarn_on_error, Opt_mblk_io_submit, 1672 Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize, 1673 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1674 Opt_inode_readahead_blks, Opt_journal_ioprio, 1675 Opt_dioread_nolock, Opt_dioread_lock, 1676 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1677 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1678 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1679 #ifdef CONFIG_EXT4_DEBUG 1680 Opt_fc_debug_max_replay, Opt_fc_debug_force 1681 #endif 1682 }; 1683 1684 static const match_table_t tokens = { 1685 {Opt_bsd_df, "bsddf"}, 1686 {Opt_minix_df, "minixdf"}, 1687 {Opt_grpid, "grpid"}, 1688 {Opt_grpid, "bsdgroups"}, 1689 {Opt_nogrpid, "nogrpid"}, 1690 {Opt_nogrpid, "sysvgroups"}, 1691 {Opt_resgid, "resgid=%u"}, 1692 {Opt_resuid, "resuid=%u"}, 1693 {Opt_sb, "sb=%u"}, 1694 {Opt_err_cont, "errors=continue"}, 1695 {Opt_err_panic, "errors=panic"}, 1696 {Opt_err_ro, "errors=remount-ro"}, 1697 {Opt_nouid32, "nouid32"}, 1698 {Opt_debug, "debug"}, 1699 {Opt_removed, "oldalloc"}, 1700 {Opt_removed, "orlov"}, 1701 {Opt_user_xattr, "user_xattr"}, 1702 {Opt_nouser_xattr, "nouser_xattr"}, 1703 {Opt_acl, "acl"}, 1704 {Opt_noacl, "noacl"}, 1705 {Opt_noload, "norecovery"}, 1706 {Opt_noload, "noload"}, 1707 {Opt_removed, "nobh"}, 1708 {Opt_removed, "bh"}, 1709 {Opt_commit, "commit=%u"}, 1710 {Opt_min_batch_time, "min_batch_time=%u"}, 1711 {Opt_max_batch_time, "max_batch_time=%u"}, 1712 {Opt_journal_dev, "journal_dev=%u"}, 1713 {Opt_journal_path, "journal_path=%s"}, 1714 {Opt_journal_checksum, "journal_checksum"}, 1715 {Opt_nojournal_checksum, "nojournal_checksum"}, 1716 {Opt_journal_async_commit, "journal_async_commit"}, 1717 {Opt_abort, "abort"}, 1718 {Opt_data_journal, "data=journal"}, 1719 {Opt_data_ordered, "data=ordered"}, 1720 {Opt_data_writeback, "data=writeback"}, 1721 {Opt_data_err_abort, "data_err=abort"}, 1722 {Opt_data_err_ignore, "data_err=ignore"}, 1723 {Opt_offusrjquota, "usrjquota="}, 1724 {Opt_usrjquota, "usrjquota=%s"}, 1725 {Opt_offgrpjquota, "grpjquota="}, 1726 {Opt_grpjquota, "grpjquota=%s"}, 1727 {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, 1728 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, 1729 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, 1730 {Opt_grpquota, "grpquota"}, 1731 {Opt_noquota, "noquota"}, 1732 {Opt_quota, "quota"}, 1733 {Opt_usrquota, "usrquota"}, 1734 {Opt_prjquota, "prjquota"}, 1735 {Opt_barrier, "barrier=%u"}, 1736 {Opt_barrier, "barrier"}, 1737 {Opt_nobarrier, "nobarrier"}, 1738 {Opt_i_version, "i_version"}, 1739 {Opt_dax, "dax"}, 1740 {Opt_dax_always, "dax=always"}, 1741 {Opt_dax_inode, "dax=inode"}, 1742 {Opt_dax_never, "dax=never"}, 1743 {Opt_stripe, "stripe=%u"}, 1744 {Opt_delalloc, "delalloc"}, 1745 {Opt_warn_on_error, "warn_on_error"}, 1746 {Opt_nowarn_on_error, "nowarn_on_error"}, 1747 {Opt_lazytime, "lazytime"}, 1748 {Opt_nolazytime, "nolazytime"}, 1749 {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"}, 1750 {Opt_nodelalloc, "nodelalloc"}, 1751 {Opt_removed, "mblk_io_submit"}, 1752 {Opt_removed, "nomblk_io_submit"}, 1753 {Opt_block_validity, "block_validity"}, 1754 {Opt_noblock_validity, "noblock_validity"}, 1755 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, 1756 {Opt_journal_ioprio, "journal_ioprio=%u"}, 1757 {Opt_auto_da_alloc, "auto_da_alloc=%u"}, 1758 {Opt_auto_da_alloc, "auto_da_alloc"}, 1759 {Opt_noauto_da_alloc, "noauto_da_alloc"}, 1760 {Opt_dioread_nolock, "dioread_nolock"}, 1761 {Opt_dioread_lock, "nodioread_nolock"}, 1762 {Opt_dioread_lock, "dioread_lock"}, 1763 {Opt_discard, "discard"}, 1764 {Opt_nodiscard, "nodiscard"}, 1765 {Opt_init_itable, "init_itable=%u"}, 1766 {Opt_init_itable, "init_itable"}, 1767 {Opt_noinit_itable, "noinit_itable"}, 1768 #ifdef CONFIG_EXT4_DEBUG 1769 {Opt_fc_debug_force, "fc_debug_force"}, 1770 {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"}, 1771 #endif 1772 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, 1773 {Opt_test_dummy_encryption, "test_dummy_encryption=%s"}, 1774 {Opt_test_dummy_encryption, "test_dummy_encryption"}, 1775 {Opt_inlinecrypt, "inlinecrypt"}, 1776 {Opt_nombcache, "nombcache"}, 1777 {Opt_nombcache, "no_mbcache"}, /* for backward compatibility */ 1778 {Opt_removed, "prefetch_block_bitmaps"}, 1779 {Opt_no_prefetch_block_bitmaps, "no_prefetch_block_bitmaps"}, 1780 {Opt_mb_optimize_scan, "mb_optimize_scan=%d"}, 1781 {Opt_removed, "check=none"}, /* mount option from ext2/3 */ 1782 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */ 1783 {Opt_removed, "reservation"}, /* mount option from ext2/3 */ 1784 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */ 1785 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */ 1786 {Opt_err, NULL}, 1787 }; 1788 1789 static ext4_fsblk_t get_sb_block(void **data) 1790 { 1791 ext4_fsblk_t sb_block; 1792 char *options = (char *) *data; 1793 1794 if (!options || strncmp(options, "sb=", 3) != 0) 1795 return 1; /* Default location */ 1796 1797 options += 3; 1798 /* TODO: use simple_strtoll with >32bit ext4 */ 1799 sb_block = simple_strtoul(options, &options, 0); 1800 if (*options && *options != ',') { 1801 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n", 1802 (char *) *data); 1803 return 1; 1804 } 1805 if (*options == ',') 1806 options++; 1807 *data = (void *) options; 1808 1809 return sb_block; 1810 } 1811 1812 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1813 #define DEFAULT_MB_OPTIMIZE_SCAN (-1) 1814 1815 static const char deprecated_msg[] = 1816 "Mount option \"%s\" will be removed by %s\n" 1817 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1818 1819 #ifdef CONFIG_QUOTA 1820 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args) 1821 { 1822 struct ext4_sb_info *sbi = EXT4_SB(sb); 1823 char *qname, *old_qname = get_qf_name(sb, sbi, qtype); 1824 int ret = -1; 1825 1826 if (sb_any_quota_loaded(sb) && !old_qname) { 1827 ext4_msg(sb, KERN_ERR, 1828 "Cannot change journaled " 1829 "quota options when quota turned on"); 1830 return -1; 1831 } 1832 if (ext4_has_feature_quota(sb)) { 1833 ext4_msg(sb, KERN_INFO, "Journaled quota options " 1834 "ignored when QUOTA feature is enabled"); 1835 return 1; 1836 } 1837 qname = match_strdup(args); 1838 if (!qname) { 1839 ext4_msg(sb, KERN_ERR, 1840 "Not enough memory for storing quotafile name"); 1841 return -1; 1842 } 1843 if (old_qname) { 1844 if (strcmp(old_qname, qname) == 0) 1845 ret = 1; 1846 else 1847 ext4_msg(sb, KERN_ERR, 1848 "%s quota file already specified", 1849 QTYPE2NAME(qtype)); 1850 goto errout; 1851 } 1852 if (strchr(qname, '/')) { 1853 ext4_msg(sb, KERN_ERR, 1854 "quotafile must be on filesystem root"); 1855 goto errout; 1856 } 1857 rcu_assign_pointer(sbi->s_qf_names[qtype], qname); 1858 set_opt(sb, QUOTA); 1859 return 1; 1860 errout: 1861 kfree(qname); 1862 return ret; 1863 } 1864 1865 static int clear_qf_name(struct super_block *sb, int qtype) 1866 { 1867 1868 struct ext4_sb_info *sbi = EXT4_SB(sb); 1869 char *old_qname = get_qf_name(sb, sbi, qtype); 1870 1871 if (sb_any_quota_loaded(sb) && old_qname) { 1872 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options" 1873 " when quota turned on"); 1874 return -1; 1875 } 1876 rcu_assign_pointer(sbi->s_qf_names[qtype], NULL); 1877 synchronize_rcu(); 1878 kfree(old_qname); 1879 return 1; 1880 } 1881 #endif 1882 1883 #define MOPT_SET 0x0001 1884 #define MOPT_CLEAR 0x0002 1885 #define MOPT_NOSUPPORT 0x0004 1886 #define MOPT_EXPLICIT 0x0008 1887 #define MOPT_CLEAR_ERR 0x0010 1888 #define MOPT_GTE0 0x0020 1889 #ifdef CONFIG_QUOTA 1890 #define MOPT_Q 0 1891 #define MOPT_QFMT 0x0040 1892 #else 1893 #define MOPT_Q MOPT_NOSUPPORT 1894 #define MOPT_QFMT MOPT_NOSUPPORT 1895 #endif 1896 #define MOPT_DATAJ 0x0080 1897 #define MOPT_NO_EXT2 0x0100 1898 #define MOPT_NO_EXT3 0x0200 1899 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1900 #define MOPT_STRING 0x0400 1901 #define MOPT_SKIP 0x0800 1902 #define MOPT_2 0x1000 1903 1904 static const struct mount_opts { 1905 int token; 1906 int mount_opt; 1907 int flags; 1908 } ext4_mount_opts[] = { 1909 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1910 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1911 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1912 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1913 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1914 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1915 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1916 MOPT_EXT4_ONLY | MOPT_SET}, 1917 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1918 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1919 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1920 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1921 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1922 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1923 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1924 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1925 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1926 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1927 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1928 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1929 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1930 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1931 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1932 EXT4_MOUNT_JOURNAL_CHECKSUM), 1933 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1934 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1935 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR}, 1936 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR}, 1937 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR}, 1938 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, 1939 MOPT_NO_EXT2}, 1940 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, 1941 MOPT_NO_EXT2}, 1942 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1943 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1944 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1945 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1946 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1947 {Opt_commit, 0, MOPT_GTE0}, 1948 {Opt_max_batch_time, 0, MOPT_GTE0}, 1949 {Opt_min_batch_time, 0, MOPT_GTE0}, 1950 {Opt_inode_readahead_blks, 0, MOPT_GTE0}, 1951 {Opt_init_itable, 0, MOPT_GTE0}, 1952 {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP}, 1953 {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS, 1954 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1955 {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE, 1956 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1957 {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER, 1958 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1959 {Opt_stripe, 0, MOPT_GTE0}, 1960 {Opt_resuid, 0, MOPT_GTE0}, 1961 {Opt_resgid, 0, MOPT_GTE0}, 1962 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1963 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING}, 1964 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1965 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1966 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1967 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, 1968 MOPT_NO_EXT2 | MOPT_DATAJ}, 1969 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1970 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR}, 1971 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1972 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1973 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR}, 1974 #else 1975 {Opt_acl, 0, MOPT_NOSUPPORT}, 1976 {Opt_noacl, 0, MOPT_NOSUPPORT}, 1977 #endif 1978 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1979 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1980 {Opt_debug_want_extra_isize, 0, MOPT_GTE0}, 1981 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1982 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1983 MOPT_SET | MOPT_Q}, 1984 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1985 MOPT_SET | MOPT_Q}, 1986 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1987 MOPT_SET | MOPT_Q}, 1988 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1989 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1990 MOPT_CLEAR | MOPT_Q}, 1991 {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING}, 1992 {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING}, 1993 {Opt_offusrjquota, 0, MOPT_Q}, 1994 {Opt_offgrpjquota, 0, MOPT_Q}, 1995 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT}, 1996 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT}, 1997 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT}, 1998 {Opt_max_dir_size_kb, 0, MOPT_GTE0}, 1999 {Opt_test_dummy_encryption, 0, MOPT_STRING}, 2000 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 2001 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 2002 MOPT_SET}, 2003 {Opt_mb_optimize_scan, EXT4_MOUNT2_MB_OPTIMIZE_SCAN, MOPT_GTE0}, 2004 #ifdef CONFIG_EXT4_DEBUG 2005 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 2006 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 2007 {Opt_fc_debug_max_replay, 0, MOPT_GTE0}, 2008 #endif 2009 {Opt_err, 0, 0} 2010 }; 2011 2012 #ifdef CONFIG_UNICODE 2013 static const struct ext4_sb_encodings { 2014 __u16 magic; 2015 char *name; 2016 char *version; 2017 } ext4_sb_encoding_map[] = { 2018 {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"}, 2019 }; 2020 2021 static int ext4_sb_read_encoding(const struct ext4_super_block *es, 2022 const struct ext4_sb_encodings **encoding, 2023 __u16 *flags) 2024 { 2025 __u16 magic = le16_to_cpu(es->s_encoding); 2026 int i; 2027 2028 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 2029 if (magic == ext4_sb_encoding_map[i].magic) 2030 break; 2031 2032 if (i >= ARRAY_SIZE(ext4_sb_encoding_map)) 2033 return -EINVAL; 2034 2035 *encoding = &ext4_sb_encoding_map[i]; 2036 *flags = le16_to_cpu(es->s_encoding_flags); 2037 2038 return 0; 2039 } 2040 #endif 2041 2042 static int ext4_set_test_dummy_encryption(struct super_block *sb, 2043 const char *opt, 2044 const substring_t *arg, 2045 bool is_remount) 2046 { 2047 #ifdef CONFIG_FS_ENCRYPTION 2048 struct ext4_sb_info *sbi = EXT4_SB(sb); 2049 int err; 2050 2051 /* 2052 * This mount option is just for testing, and it's not worthwhile to 2053 * implement the extra complexity (e.g. RCU protection) that would be 2054 * needed to allow it to be set or changed during remount. We do allow 2055 * it to be specified during remount, but only if there is no change. 2056 */ 2057 if (is_remount && !sbi->s_dummy_enc_policy.policy) { 2058 ext4_msg(sb, KERN_WARNING, 2059 "Can't set test_dummy_encryption on remount"); 2060 return -1; 2061 } 2062 err = fscrypt_set_test_dummy_encryption(sb, arg->from, 2063 &sbi->s_dummy_enc_policy); 2064 if (err) { 2065 if (err == -EEXIST) 2066 ext4_msg(sb, KERN_WARNING, 2067 "Can't change test_dummy_encryption on remount"); 2068 else if (err == -EINVAL) 2069 ext4_msg(sb, KERN_WARNING, 2070 "Value of option \"%s\" is unrecognized", opt); 2071 else 2072 ext4_msg(sb, KERN_WARNING, 2073 "Error processing option \"%s\" [%d]", 2074 opt, err); 2075 return -1; 2076 } 2077 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2078 #else 2079 ext4_msg(sb, KERN_WARNING, 2080 "Test dummy encryption mount option ignored"); 2081 #endif 2082 return 1; 2083 } 2084 2085 struct ext4_parsed_options { 2086 unsigned long journal_devnum; 2087 unsigned int journal_ioprio; 2088 int mb_optimize_scan; 2089 }; 2090 2091 static int handle_mount_opt(struct super_block *sb, char *opt, int token, 2092 substring_t *args, struct ext4_parsed_options *parsed_opts, 2093 int is_remount) 2094 { 2095 struct ext4_sb_info *sbi = EXT4_SB(sb); 2096 const struct mount_opts *m; 2097 kuid_t uid; 2098 kgid_t gid; 2099 int arg = 0; 2100 2101 #ifdef CONFIG_QUOTA 2102 if (token == Opt_usrjquota) 2103 return set_qf_name(sb, USRQUOTA, &args[0]); 2104 else if (token == Opt_grpjquota) 2105 return set_qf_name(sb, GRPQUOTA, &args[0]); 2106 else if (token == Opt_offusrjquota) 2107 return clear_qf_name(sb, USRQUOTA); 2108 else if (token == Opt_offgrpjquota) 2109 return clear_qf_name(sb, GRPQUOTA); 2110 #endif 2111 switch (token) { 2112 case Opt_noacl: 2113 case Opt_nouser_xattr: 2114 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5"); 2115 break; 2116 case Opt_sb: 2117 return 1; /* handled by get_sb_block() */ 2118 case Opt_removed: 2119 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt); 2120 return 1; 2121 case Opt_abort: 2122 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 2123 return 1; 2124 case Opt_i_version: 2125 sb->s_flags |= SB_I_VERSION; 2126 return 1; 2127 case Opt_lazytime: 2128 sb->s_flags |= SB_LAZYTIME; 2129 return 1; 2130 case Opt_nolazytime: 2131 sb->s_flags &= ~SB_LAZYTIME; 2132 return 1; 2133 case Opt_inlinecrypt: 2134 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2135 sb->s_flags |= SB_INLINECRYPT; 2136 #else 2137 ext4_msg(sb, KERN_ERR, "inline encryption not supported"); 2138 #endif 2139 return 1; 2140 } 2141 2142 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2143 if (token == m->token) 2144 break; 2145 2146 if (m->token == Opt_err) { 2147 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" " 2148 "or missing value", opt); 2149 return -1; 2150 } 2151 2152 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2153 ext4_msg(sb, KERN_ERR, 2154 "Mount option \"%s\" incompatible with ext2", opt); 2155 return -1; 2156 } 2157 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2158 ext4_msg(sb, KERN_ERR, 2159 "Mount option \"%s\" incompatible with ext3", opt); 2160 return -1; 2161 } 2162 2163 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg)) 2164 return -1; 2165 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0)) 2166 return -1; 2167 if (m->flags & MOPT_EXPLICIT) { 2168 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2169 set_opt2(sb, EXPLICIT_DELALLOC); 2170 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2171 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM); 2172 } else 2173 return -1; 2174 } 2175 if (m->flags & MOPT_CLEAR_ERR) 2176 clear_opt(sb, ERRORS_MASK); 2177 if (token == Opt_noquota && sb_any_quota_loaded(sb)) { 2178 ext4_msg(sb, KERN_ERR, "Cannot change quota " 2179 "options when quota turned on"); 2180 return -1; 2181 } 2182 2183 if (m->flags & MOPT_NOSUPPORT) { 2184 ext4_msg(sb, KERN_ERR, "%s option not supported", opt); 2185 } else if (token == Opt_commit) { 2186 if (arg == 0) 2187 arg = JBD2_DEFAULT_MAX_COMMIT_AGE; 2188 else if (arg > INT_MAX / HZ) { 2189 ext4_msg(sb, KERN_ERR, 2190 "Invalid commit interval %d, " 2191 "must be smaller than %d", 2192 arg, INT_MAX / HZ); 2193 return -1; 2194 } 2195 sbi->s_commit_interval = HZ * arg; 2196 } else if (token == Opt_debug_want_extra_isize) { 2197 if ((arg & 1) || 2198 (arg < 4) || 2199 (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) { 2200 ext4_msg(sb, KERN_ERR, 2201 "Invalid want_extra_isize %d", arg); 2202 return -1; 2203 } 2204 sbi->s_want_extra_isize = arg; 2205 } else if (token == Opt_max_batch_time) { 2206 sbi->s_max_batch_time = arg; 2207 } else if (token == Opt_min_batch_time) { 2208 sbi->s_min_batch_time = arg; 2209 } else if (token == Opt_inode_readahead_blks) { 2210 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) { 2211 ext4_msg(sb, KERN_ERR, 2212 "EXT4-fs: inode_readahead_blks must be " 2213 "0 or a power of 2 smaller than 2^31"); 2214 return -1; 2215 } 2216 sbi->s_inode_readahead_blks = arg; 2217 } else if (token == Opt_init_itable) { 2218 set_opt(sb, INIT_INODE_TABLE); 2219 if (!args->from) 2220 arg = EXT4_DEF_LI_WAIT_MULT; 2221 sbi->s_li_wait_mult = arg; 2222 } else if (token == Opt_max_dir_size_kb) { 2223 sbi->s_max_dir_size_kb = arg; 2224 #ifdef CONFIG_EXT4_DEBUG 2225 } else if (token == Opt_fc_debug_max_replay) { 2226 sbi->s_fc_debug_max_replay = arg; 2227 #endif 2228 } else if (token == Opt_stripe) { 2229 sbi->s_stripe = arg; 2230 } else if (token == Opt_resuid) { 2231 uid = make_kuid(current_user_ns(), arg); 2232 if (!uid_valid(uid)) { 2233 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg); 2234 return -1; 2235 } 2236 sbi->s_resuid = uid; 2237 } else if (token == Opt_resgid) { 2238 gid = make_kgid(current_user_ns(), arg); 2239 if (!gid_valid(gid)) { 2240 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg); 2241 return -1; 2242 } 2243 sbi->s_resgid = gid; 2244 } else if (token == Opt_journal_dev) { 2245 if (is_remount) { 2246 ext4_msg(sb, KERN_ERR, 2247 "Cannot specify journal on remount"); 2248 return -1; 2249 } 2250 parsed_opts->journal_devnum = arg; 2251 } else if (token == Opt_journal_path) { 2252 char *journal_path; 2253 struct inode *journal_inode; 2254 struct path path; 2255 int error; 2256 2257 if (is_remount) { 2258 ext4_msg(sb, KERN_ERR, 2259 "Cannot specify journal on remount"); 2260 return -1; 2261 } 2262 journal_path = match_strdup(&args[0]); 2263 if (!journal_path) { 2264 ext4_msg(sb, KERN_ERR, "error: could not dup " 2265 "journal device string"); 2266 return -1; 2267 } 2268 2269 error = kern_path(journal_path, LOOKUP_FOLLOW, &path); 2270 if (error) { 2271 ext4_msg(sb, KERN_ERR, "error: could not find " 2272 "journal device path: error %d", error); 2273 kfree(journal_path); 2274 return -1; 2275 } 2276 2277 journal_inode = d_inode(path.dentry); 2278 if (!S_ISBLK(journal_inode->i_mode)) { 2279 ext4_msg(sb, KERN_ERR, "error: journal path %s " 2280 "is not a block device", journal_path); 2281 path_put(&path); 2282 kfree(journal_path); 2283 return -1; 2284 } 2285 2286 parsed_opts->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2287 path_put(&path); 2288 kfree(journal_path); 2289 } else if (token == Opt_journal_ioprio) { 2290 if (arg > 7) { 2291 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority" 2292 " (must be 0-7)"); 2293 return -1; 2294 } 2295 parsed_opts->journal_ioprio = 2296 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg); 2297 } else if (token == Opt_test_dummy_encryption) { 2298 return ext4_set_test_dummy_encryption(sb, opt, &args[0], 2299 is_remount); 2300 } else if (m->flags & MOPT_DATAJ) { 2301 if (is_remount) { 2302 if (!sbi->s_journal) 2303 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option"); 2304 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) { 2305 ext4_msg(sb, KERN_ERR, 2306 "Cannot change data mode on remount"); 2307 return -1; 2308 } 2309 } else { 2310 clear_opt(sb, DATA_FLAGS); 2311 sbi->s_mount_opt |= m->mount_opt; 2312 } 2313 #ifdef CONFIG_QUOTA 2314 } else if (m->flags & MOPT_QFMT) { 2315 if (sb_any_quota_loaded(sb) && 2316 sbi->s_jquota_fmt != m->mount_opt) { 2317 ext4_msg(sb, KERN_ERR, "Cannot change journaled " 2318 "quota options when quota turned on"); 2319 return -1; 2320 } 2321 if (ext4_has_feature_quota(sb)) { 2322 ext4_msg(sb, KERN_INFO, 2323 "Quota format mount options ignored " 2324 "when QUOTA feature is enabled"); 2325 return 1; 2326 } 2327 sbi->s_jquota_fmt = m->mount_opt; 2328 #endif 2329 } else if (token == Opt_dax || token == Opt_dax_always || 2330 token == Opt_dax_inode || token == Opt_dax_never) { 2331 #ifdef CONFIG_FS_DAX 2332 switch (token) { 2333 case Opt_dax: 2334 case Opt_dax_always: 2335 if (is_remount && 2336 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2337 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2338 fail_dax_change_remount: 2339 ext4_msg(sb, KERN_ERR, "can't change " 2340 "dax mount option while remounting"); 2341 return -1; 2342 } 2343 if (is_remount && 2344 (test_opt(sb, DATA_FLAGS) == 2345 EXT4_MOUNT_JOURNAL_DATA)) { 2346 ext4_msg(sb, KERN_ERR, "can't mount with " 2347 "both data=journal and dax"); 2348 return -1; 2349 } 2350 ext4_msg(sb, KERN_WARNING, 2351 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 2352 sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS; 2353 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2354 break; 2355 case Opt_dax_never: 2356 if (is_remount && 2357 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2358 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) 2359 goto fail_dax_change_remount; 2360 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2361 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2362 break; 2363 case Opt_dax_inode: 2364 if (is_remount && 2365 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2366 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2367 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) 2368 goto fail_dax_change_remount; 2369 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2370 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2371 /* Strictly for printing options */ 2372 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE; 2373 break; 2374 } 2375 #else 2376 ext4_msg(sb, KERN_INFO, "dax option not supported"); 2377 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2378 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2379 return -1; 2380 #endif 2381 } else if (token == Opt_data_err_abort) { 2382 sbi->s_mount_opt |= m->mount_opt; 2383 } else if (token == Opt_data_err_ignore) { 2384 sbi->s_mount_opt &= ~m->mount_opt; 2385 } else if (token == Opt_mb_optimize_scan) { 2386 if (arg != 0 && arg != 1) { 2387 ext4_msg(sb, KERN_WARNING, 2388 "mb_optimize_scan should be set to 0 or 1."); 2389 return -1; 2390 } 2391 parsed_opts->mb_optimize_scan = arg; 2392 } else { 2393 if (!args->from) 2394 arg = 1; 2395 if (m->flags & MOPT_CLEAR) 2396 arg = !arg; 2397 else if (unlikely(!(m->flags & MOPT_SET))) { 2398 ext4_msg(sb, KERN_WARNING, 2399 "buggy handling of option %s", opt); 2400 WARN_ON(1); 2401 return -1; 2402 } 2403 if (m->flags & MOPT_2) { 2404 if (arg != 0) 2405 sbi->s_mount_opt2 |= m->mount_opt; 2406 else 2407 sbi->s_mount_opt2 &= ~m->mount_opt; 2408 } else { 2409 if (arg != 0) 2410 sbi->s_mount_opt |= m->mount_opt; 2411 else 2412 sbi->s_mount_opt &= ~m->mount_opt; 2413 } 2414 } 2415 return 1; 2416 } 2417 2418 static int parse_options(char *options, struct super_block *sb, 2419 struct ext4_parsed_options *ret_opts, 2420 int is_remount) 2421 { 2422 struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb); 2423 char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name; 2424 substring_t args[MAX_OPT_ARGS]; 2425 int token; 2426 2427 if (!options) 2428 return 1; 2429 2430 while ((p = strsep(&options, ",")) != NULL) { 2431 if (!*p) 2432 continue; 2433 /* 2434 * Initialize args struct so we know whether arg was 2435 * found; some options take optional arguments. 2436 */ 2437 args[0].to = args[0].from = NULL; 2438 token = match_token(p, tokens, args); 2439 if (handle_mount_opt(sb, p, token, args, ret_opts, 2440 is_remount) < 0) 2441 return 0; 2442 } 2443 #ifdef CONFIG_QUOTA 2444 /* 2445 * We do the test below only for project quotas. 'usrquota' and 2446 * 'grpquota' mount options are allowed even without quota feature 2447 * to support legacy quotas in quota files. 2448 */ 2449 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) { 2450 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. " 2451 "Cannot enable project quota enforcement."); 2452 return 0; 2453 } 2454 usr_qf_name = get_qf_name(sb, sbi, USRQUOTA); 2455 grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA); 2456 if (usr_qf_name || grp_qf_name) { 2457 if (test_opt(sb, USRQUOTA) && usr_qf_name) 2458 clear_opt(sb, USRQUOTA); 2459 2460 if (test_opt(sb, GRPQUOTA) && grp_qf_name) 2461 clear_opt(sb, GRPQUOTA); 2462 2463 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) { 2464 ext4_msg(sb, KERN_ERR, "old and new quota " 2465 "format mixing"); 2466 return 0; 2467 } 2468 2469 if (!sbi->s_jquota_fmt) { 2470 ext4_msg(sb, KERN_ERR, "journaled quota format " 2471 "not specified"); 2472 return 0; 2473 } 2474 } 2475 #endif 2476 if (test_opt(sb, DIOREAD_NOLOCK)) { 2477 int blocksize = 2478 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2479 if (blocksize < PAGE_SIZE) 2480 ext4_msg(sb, KERN_WARNING, "Warning: mounting with an " 2481 "experimental mount option 'dioread_nolock' " 2482 "for blocksize < PAGE_SIZE"); 2483 } 2484 return 1; 2485 } 2486 2487 static inline void ext4_show_quota_options(struct seq_file *seq, 2488 struct super_block *sb) 2489 { 2490 #if defined(CONFIG_QUOTA) 2491 struct ext4_sb_info *sbi = EXT4_SB(sb); 2492 char *usr_qf_name, *grp_qf_name; 2493 2494 if (sbi->s_jquota_fmt) { 2495 char *fmtname = ""; 2496 2497 switch (sbi->s_jquota_fmt) { 2498 case QFMT_VFS_OLD: 2499 fmtname = "vfsold"; 2500 break; 2501 case QFMT_VFS_V0: 2502 fmtname = "vfsv0"; 2503 break; 2504 case QFMT_VFS_V1: 2505 fmtname = "vfsv1"; 2506 break; 2507 } 2508 seq_printf(seq, ",jqfmt=%s", fmtname); 2509 } 2510 2511 rcu_read_lock(); 2512 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2513 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2514 if (usr_qf_name) 2515 seq_show_option(seq, "usrjquota", usr_qf_name); 2516 if (grp_qf_name) 2517 seq_show_option(seq, "grpjquota", grp_qf_name); 2518 rcu_read_unlock(); 2519 #endif 2520 } 2521 2522 static const char *token2str(int token) 2523 { 2524 const struct match_token *t; 2525 2526 for (t = tokens; t->token != Opt_err; t++) 2527 if (t->token == token && !strchr(t->pattern, '=')) 2528 break; 2529 return t->pattern; 2530 } 2531 2532 /* 2533 * Show an option if 2534 * - it's set to a non-default value OR 2535 * - if the per-sb default is different from the global default 2536 */ 2537 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2538 int nodefs) 2539 { 2540 struct ext4_sb_info *sbi = EXT4_SB(sb); 2541 struct ext4_super_block *es = sbi->s_es; 2542 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2543 const struct mount_opts *m; 2544 char sep = nodefs ? '\n' : ','; 2545 2546 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2547 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2548 2549 if (sbi->s_sb_block != 1) 2550 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2551 2552 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2553 int want_set = m->flags & MOPT_SET; 2554 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2555 (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP) 2556 continue; 2557 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2558 continue; /* skip if same as the default */ 2559 if ((want_set && 2560 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2561 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2562 continue; /* select Opt_noFoo vs Opt_Foo */ 2563 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2564 } 2565 2566 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2567 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2568 SEQ_OPTS_PRINT("resuid=%u", 2569 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2570 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2571 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2572 SEQ_OPTS_PRINT("resgid=%u", 2573 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2574 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2575 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2576 SEQ_OPTS_PUTS("errors=remount-ro"); 2577 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2578 SEQ_OPTS_PUTS("errors=continue"); 2579 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2580 SEQ_OPTS_PUTS("errors=panic"); 2581 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2582 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2583 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2584 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2585 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2586 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2587 if (sb->s_flags & SB_I_VERSION) 2588 SEQ_OPTS_PUTS("i_version"); 2589 if (nodefs || sbi->s_stripe) 2590 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2591 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2592 (sbi->s_mount_opt ^ def_mount_opt)) { 2593 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2594 SEQ_OPTS_PUTS("data=journal"); 2595 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2596 SEQ_OPTS_PUTS("data=ordered"); 2597 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2598 SEQ_OPTS_PUTS("data=writeback"); 2599 } 2600 if (nodefs || 2601 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2602 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2603 sbi->s_inode_readahead_blks); 2604 2605 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2606 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2607 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2608 if (nodefs || sbi->s_max_dir_size_kb) 2609 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2610 if (test_opt(sb, DATA_ERR_ABORT)) 2611 SEQ_OPTS_PUTS("data_err=abort"); 2612 2613 fscrypt_show_test_dummy_encryption(seq, sep, sb); 2614 2615 if (sb->s_flags & SB_INLINECRYPT) 2616 SEQ_OPTS_PUTS("inlinecrypt"); 2617 2618 if (test_opt(sb, DAX_ALWAYS)) { 2619 if (IS_EXT2_SB(sb)) 2620 SEQ_OPTS_PUTS("dax"); 2621 else 2622 SEQ_OPTS_PUTS("dax=always"); 2623 } else if (test_opt2(sb, DAX_NEVER)) { 2624 SEQ_OPTS_PUTS("dax=never"); 2625 } else if (test_opt2(sb, DAX_INODE)) { 2626 SEQ_OPTS_PUTS("dax=inode"); 2627 } 2628 ext4_show_quota_options(seq, sb); 2629 return 0; 2630 } 2631 2632 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2633 { 2634 return _ext4_show_options(seq, root->d_sb, 0); 2635 } 2636 2637 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2638 { 2639 struct super_block *sb = seq->private; 2640 int rc; 2641 2642 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2643 rc = _ext4_show_options(seq, sb, 1); 2644 seq_puts(seq, "\n"); 2645 return rc; 2646 } 2647 2648 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2649 int read_only) 2650 { 2651 struct ext4_sb_info *sbi = EXT4_SB(sb); 2652 int err = 0; 2653 2654 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2655 ext4_msg(sb, KERN_ERR, "revision level too high, " 2656 "forcing read-only mode"); 2657 err = -EROFS; 2658 goto done; 2659 } 2660 if (read_only) 2661 goto done; 2662 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2663 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2664 "running e2fsck is recommended"); 2665 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2666 ext4_msg(sb, KERN_WARNING, 2667 "warning: mounting fs with errors, " 2668 "running e2fsck is recommended"); 2669 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2670 le16_to_cpu(es->s_mnt_count) >= 2671 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2672 ext4_msg(sb, KERN_WARNING, 2673 "warning: maximal mount count reached, " 2674 "running e2fsck is recommended"); 2675 else if (le32_to_cpu(es->s_checkinterval) && 2676 (ext4_get_tstamp(es, s_lastcheck) + 2677 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2678 ext4_msg(sb, KERN_WARNING, 2679 "warning: checktime reached, " 2680 "running e2fsck is recommended"); 2681 if (!sbi->s_journal) 2682 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2683 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2684 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2685 le16_add_cpu(&es->s_mnt_count, 1); 2686 ext4_update_tstamp(es, s_mtime); 2687 if (sbi->s_journal) 2688 ext4_set_feature_journal_needs_recovery(sb); 2689 2690 err = ext4_commit_super(sb); 2691 done: 2692 if (test_opt(sb, DEBUG)) 2693 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2694 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2695 sb->s_blocksize, 2696 sbi->s_groups_count, 2697 EXT4_BLOCKS_PER_GROUP(sb), 2698 EXT4_INODES_PER_GROUP(sb), 2699 sbi->s_mount_opt, sbi->s_mount_opt2); 2700 2701 cleancache_init_fs(sb); 2702 return err; 2703 } 2704 2705 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2706 { 2707 struct ext4_sb_info *sbi = EXT4_SB(sb); 2708 struct flex_groups **old_groups, **new_groups; 2709 int size, i, j; 2710 2711 if (!sbi->s_log_groups_per_flex) 2712 return 0; 2713 2714 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2715 if (size <= sbi->s_flex_groups_allocated) 2716 return 0; 2717 2718 new_groups = kvzalloc(roundup_pow_of_two(size * 2719 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 2720 if (!new_groups) { 2721 ext4_msg(sb, KERN_ERR, 2722 "not enough memory for %d flex group pointers", size); 2723 return -ENOMEM; 2724 } 2725 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 2726 new_groups[i] = kvzalloc(roundup_pow_of_two( 2727 sizeof(struct flex_groups)), 2728 GFP_KERNEL); 2729 if (!new_groups[i]) { 2730 for (j = sbi->s_flex_groups_allocated; j < i; j++) 2731 kvfree(new_groups[j]); 2732 kvfree(new_groups); 2733 ext4_msg(sb, KERN_ERR, 2734 "not enough memory for %d flex groups", size); 2735 return -ENOMEM; 2736 } 2737 } 2738 rcu_read_lock(); 2739 old_groups = rcu_dereference(sbi->s_flex_groups); 2740 if (old_groups) 2741 memcpy(new_groups, old_groups, 2742 (sbi->s_flex_groups_allocated * 2743 sizeof(struct flex_groups *))); 2744 rcu_read_unlock(); 2745 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 2746 sbi->s_flex_groups_allocated = size; 2747 if (old_groups) 2748 ext4_kvfree_array_rcu(old_groups); 2749 return 0; 2750 } 2751 2752 static int ext4_fill_flex_info(struct super_block *sb) 2753 { 2754 struct ext4_sb_info *sbi = EXT4_SB(sb); 2755 struct ext4_group_desc *gdp = NULL; 2756 struct flex_groups *fg; 2757 ext4_group_t flex_group; 2758 int i, err; 2759 2760 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2761 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2762 sbi->s_log_groups_per_flex = 0; 2763 return 1; 2764 } 2765 2766 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2767 if (err) 2768 goto failed; 2769 2770 for (i = 0; i < sbi->s_groups_count; i++) { 2771 gdp = ext4_get_group_desc(sb, i, NULL); 2772 2773 flex_group = ext4_flex_group(sbi, i); 2774 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 2775 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 2776 atomic64_add(ext4_free_group_clusters(sb, gdp), 2777 &fg->free_clusters); 2778 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 2779 } 2780 2781 return 1; 2782 failed: 2783 return 0; 2784 } 2785 2786 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2787 struct ext4_group_desc *gdp) 2788 { 2789 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2790 __u16 crc = 0; 2791 __le32 le_group = cpu_to_le32(block_group); 2792 struct ext4_sb_info *sbi = EXT4_SB(sb); 2793 2794 if (ext4_has_metadata_csum(sbi->s_sb)) { 2795 /* Use new metadata_csum algorithm */ 2796 __u32 csum32; 2797 __u16 dummy_csum = 0; 2798 2799 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2800 sizeof(le_group)); 2801 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2802 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2803 sizeof(dummy_csum)); 2804 offset += sizeof(dummy_csum); 2805 if (offset < sbi->s_desc_size) 2806 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2807 sbi->s_desc_size - offset); 2808 2809 crc = csum32 & 0xFFFF; 2810 goto out; 2811 } 2812 2813 /* old crc16 code */ 2814 if (!ext4_has_feature_gdt_csum(sb)) 2815 return 0; 2816 2817 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2818 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2819 crc = crc16(crc, (__u8 *)gdp, offset); 2820 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2821 /* for checksum of struct ext4_group_desc do the rest...*/ 2822 if (ext4_has_feature_64bit(sb) && 2823 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2824 crc = crc16(crc, (__u8 *)gdp + offset, 2825 le16_to_cpu(sbi->s_es->s_desc_size) - 2826 offset); 2827 2828 out: 2829 return cpu_to_le16(crc); 2830 } 2831 2832 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2833 struct ext4_group_desc *gdp) 2834 { 2835 if (ext4_has_group_desc_csum(sb) && 2836 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2837 return 0; 2838 2839 return 1; 2840 } 2841 2842 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2843 struct ext4_group_desc *gdp) 2844 { 2845 if (!ext4_has_group_desc_csum(sb)) 2846 return; 2847 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2848 } 2849 2850 /* Called at mount-time, super-block is locked */ 2851 static int ext4_check_descriptors(struct super_block *sb, 2852 ext4_fsblk_t sb_block, 2853 ext4_group_t *first_not_zeroed) 2854 { 2855 struct ext4_sb_info *sbi = EXT4_SB(sb); 2856 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2857 ext4_fsblk_t last_block; 2858 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2859 ext4_fsblk_t block_bitmap; 2860 ext4_fsblk_t inode_bitmap; 2861 ext4_fsblk_t inode_table; 2862 int flexbg_flag = 0; 2863 ext4_group_t i, grp = sbi->s_groups_count; 2864 2865 if (ext4_has_feature_flex_bg(sb)) 2866 flexbg_flag = 1; 2867 2868 ext4_debug("Checking group descriptors"); 2869 2870 for (i = 0; i < sbi->s_groups_count; i++) { 2871 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2872 2873 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2874 last_block = ext4_blocks_count(sbi->s_es) - 1; 2875 else 2876 last_block = first_block + 2877 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2878 2879 if ((grp == sbi->s_groups_count) && 2880 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2881 grp = i; 2882 2883 block_bitmap = ext4_block_bitmap(sb, gdp); 2884 if (block_bitmap == sb_block) { 2885 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2886 "Block bitmap for group %u overlaps " 2887 "superblock", i); 2888 if (!sb_rdonly(sb)) 2889 return 0; 2890 } 2891 if (block_bitmap >= sb_block + 1 && 2892 block_bitmap <= last_bg_block) { 2893 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2894 "Block bitmap for group %u overlaps " 2895 "block group descriptors", i); 2896 if (!sb_rdonly(sb)) 2897 return 0; 2898 } 2899 if (block_bitmap < first_block || block_bitmap > last_block) { 2900 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2901 "Block bitmap for group %u not in group " 2902 "(block %llu)!", i, block_bitmap); 2903 return 0; 2904 } 2905 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2906 if (inode_bitmap == sb_block) { 2907 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2908 "Inode bitmap for group %u overlaps " 2909 "superblock", i); 2910 if (!sb_rdonly(sb)) 2911 return 0; 2912 } 2913 if (inode_bitmap >= sb_block + 1 && 2914 inode_bitmap <= last_bg_block) { 2915 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2916 "Inode bitmap for group %u overlaps " 2917 "block group descriptors", i); 2918 if (!sb_rdonly(sb)) 2919 return 0; 2920 } 2921 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2922 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2923 "Inode bitmap for group %u not in group " 2924 "(block %llu)!", i, inode_bitmap); 2925 return 0; 2926 } 2927 inode_table = ext4_inode_table(sb, gdp); 2928 if (inode_table == sb_block) { 2929 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2930 "Inode table for group %u overlaps " 2931 "superblock", i); 2932 if (!sb_rdonly(sb)) 2933 return 0; 2934 } 2935 if (inode_table >= sb_block + 1 && 2936 inode_table <= last_bg_block) { 2937 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2938 "Inode table for group %u overlaps " 2939 "block group descriptors", i); 2940 if (!sb_rdonly(sb)) 2941 return 0; 2942 } 2943 if (inode_table < first_block || 2944 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2945 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2946 "Inode table for group %u not in group " 2947 "(block %llu)!", i, inode_table); 2948 return 0; 2949 } 2950 ext4_lock_group(sb, i); 2951 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2952 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2953 "Checksum for group %u failed (%u!=%u)", 2954 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2955 gdp)), le16_to_cpu(gdp->bg_checksum)); 2956 if (!sb_rdonly(sb)) { 2957 ext4_unlock_group(sb, i); 2958 return 0; 2959 } 2960 } 2961 ext4_unlock_group(sb, i); 2962 if (!flexbg_flag) 2963 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2964 } 2965 if (NULL != first_not_zeroed) 2966 *first_not_zeroed = grp; 2967 return 1; 2968 } 2969 2970 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2971 * the superblock) which were deleted from all directories, but held open by 2972 * a process at the time of a crash. We walk the list and try to delete these 2973 * inodes at recovery time (only with a read-write filesystem). 2974 * 2975 * In order to keep the orphan inode chain consistent during traversal (in 2976 * case of crash during recovery), we link each inode into the superblock 2977 * orphan list_head and handle it the same way as an inode deletion during 2978 * normal operation (which journals the operations for us). 2979 * 2980 * We only do an iget() and an iput() on each inode, which is very safe if we 2981 * accidentally point at an in-use or already deleted inode. The worst that 2982 * can happen in this case is that we get a "bit already cleared" message from 2983 * ext4_free_inode(). The only reason we would point at a wrong inode is if 2984 * e2fsck was run on this filesystem, and it must have already done the orphan 2985 * inode cleanup for us, so we can safely abort without any further action. 2986 */ 2987 static void ext4_orphan_cleanup(struct super_block *sb, 2988 struct ext4_super_block *es) 2989 { 2990 unsigned int s_flags = sb->s_flags; 2991 int ret, nr_orphans = 0, nr_truncates = 0; 2992 #ifdef CONFIG_QUOTA 2993 int quota_update = 0; 2994 int i; 2995 #endif 2996 if (!es->s_last_orphan) { 2997 jbd_debug(4, "no orphan inodes to clean up\n"); 2998 return; 2999 } 3000 3001 if (bdev_read_only(sb->s_bdev)) { 3002 ext4_msg(sb, KERN_ERR, "write access " 3003 "unavailable, skipping orphan cleanup"); 3004 return; 3005 } 3006 3007 /* Check if feature set would not allow a r/w mount */ 3008 if (!ext4_feature_set_ok(sb, 0)) { 3009 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 3010 "unknown ROCOMPAT features"); 3011 return; 3012 } 3013 3014 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3015 /* don't clear list on RO mount w/ errors */ 3016 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 3017 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 3018 "clearing orphan list.\n"); 3019 es->s_last_orphan = 0; 3020 } 3021 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3022 return; 3023 } 3024 3025 if (s_flags & SB_RDONLY) { 3026 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 3027 sb->s_flags &= ~SB_RDONLY; 3028 } 3029 #ifdef CONFIG_QUOTA 3030 /* 3031 * Turn on quotas which were not enabled for read-only mounts if 3032 * filesystem has quota feature, so that they are updated correctly. 3033 */ 3034 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 3035 int ret = ext4_enable_quotas(sb); 3036 3037 if (!ret) 3038 quota_update = 1; 3039 else 3040 ext4_msg(sb, KERN_ERR, 3041 "Cannot turn on quotas: error %d", ret); 3042 } 3043 3044 /* Turn on journaled quotas used for old sytle */ 3045 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3046 if (EXT4_SB(sb)->s_qf_names[i]) { 3047 int ret = ext4_quota_on_mount(sb, i); 3048 3049 if (!ret) 3050 quota_update = 1; 3051 else 3052 ext4_msg(sb, KERN_ERR, 3053 "Cannot turn on journaled " 3054 "quota: type %d: error %d", i, ret); 3055 } 3056 } 3057 #endif 3058 3059 while (es->s_last_orphan) { 3060 struct inode *inode; 3061 3062 /* 3063 * We may have encountered an error during cleanup; if 3064 * so, skip the rest. 3065 */ 3066 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3067 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3068 es->s_last_orphan = 0; 3069 break; 3070 } 3071 3072 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 3073 if (IS_ERR(inode)) { 3074 es->s_last_orphan = 0; 3075 break; 3076 } 3077 3078 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 3079 dquot_initialize(inode); 3080 if (inode->i_nlink) { 3081 if (test_opt(sb, DEBUG)) 3082 ext4_msg(sb, KERN_DEBUG, 3083 "%s: truncating inode %lu to %lld bytes", 3084 __func__, inode->i_ino, inode->i_size); 3085 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 3086 inode->i_ino, inode->i_size); 3087 inode_lock(inode); 3088 truncate_inode_pages(inode->i_mapping, inode->i_size); 3089 ret = ext4_truncate(inode); 3090 if (ret) { 3091 /* 3092 * We need to clean up the in-core orphan list 3093 * manually if ext4_truncate() failed to get a 3094 * transaction handle. 3095 */ 3096 ext4_orphan_del(NULL, inode); 3097 ext4_std_error(inode->i_sb, ret); 3098 } 3099 inode_unlock(inode); 3100 nr_truncates++; 3101 } else { 3102 if (test_opt(sb, DEBUG)) 3103 ext4_msg(sb, KERN_DEBUG, 3104 "%s: deleting unreferenced inode %lu", 3105 __func__, inode->i_ino); 3106 jbd_debug(2, "deleting unreferenced inode %lu\n", 3107 inode->i_ino); 3108 nr_orphans++; 3109 } 3110 iput(inode); /* The delete magic happens here! */ 3111 } 3112 3113 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 3114 3115 if (nr_orphans) 3116 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 3117 PLURAL(nr_orphans)); 3118 if (nr_truncates) 3119 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 3120 PLURAL(nr_truncates)); 3121 #ifdef CONFIG_QUOTA 3122 /* Turn off quotas if they were enabled for orphan cleanup */ 3123 if (quota_update) { 3124 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3125 if (sb_dqopt(sb)->files[i]) 3126 dquot_quota_off(sb, i); 3127 } 3128 } 3129 #endif 3130 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 3131 } 3132 3133 /* 3134 * Maximal extent format file size. 3135 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3136 * extent format containers, within a sector_t, and within i_blocks 3137 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3138 * so that won't be a limiting factor. 3139 * 3140 * However there is other limiting factor. We do store extents in the form 3141 * of starting block and length, hence the resulting length of the extent 3142 * covering maximum file size must fit into on-disk format containers as 3143 * well. Given that length is always by 1 unit bigger than max unit (because 3144 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3145 * 3146 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3147 */ 3148 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3149 { 3150 loff_t res; 3151 loff_t upper_limit = MAX_LFS_FILESIZE; 3152 3153 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3154 3155 if (!has_huge_files) { 3156 upper_limit = (1LL << 32) - 1; 3157 3158 /* total blocks in file system block size */ 3159 upper_limit >>= (blkbits - 9); 3160 upper_limit <<= blkbits; 3161 } 3162 3163 /* 3164 * 32-bit extent-start container, ee_block. We lower the maxbytes 3165 * by one fs block, so ee_len can cover the extent of maximum file 3166 * size 3167 */ 3168 res = (1LL << 32) - 1; 3169 res <<= blkbits; 3170 3171 /* Sanity check against vm- & vfs- imposed limits */ 3172 if (res > upper_limit) 3173 res = upper_limit; 3174 3175 return res; 3176 } 3177 3178 /* 3179 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3180 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3181 * We need to be 1 filesystem block less than the 2^48 sector limit. 3182 */ 3183 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3184 { 3185 loff_t res = EXT4_NDIR_BLOCKS; 3186 int meta_blocks; 3187 loff_t upper_limit; 3188 /* This is calculated to be the largest file size for a dense, block 3189 * mapped file such that the file's total number of 512-byte sectors, 3190 * including data and all indirect blocks, does not exceed (2^48 - 1). 3191 * 3192 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3193 * number of 512-byte sectors of the file. 3194 */ 3195 3196 if (!has_huge_files) { 3197 /* 3198 * !has_huge_files or implies that the inode i_block field 3199 * represents total file blocks in 2^32 512-byte sectors == 3200 * size of vfs inode i_blocks * 8 3201 */ 3202 upper_limit = (1LL << 32) - 1; 3203 3204 /* total blocks in file system block size */ 3205 upper_limit >>= (bits - 9); 3206 3207 } else { 3208 /* 3209 * We use 48 bit ext4_inode i_blocks 3210 * With EXT4_HUGE_FILE_FL set the i_blocks 3211 * represent total number of blocks in 3212 * file system block size 3213 */ 3214 upper_limit = (1LL << 48) - 1; 3215 3216 } 3217 3218 /* indirect blocks */ 3219 meta_blocks = 1; 3220 /* double indirect blocks */ 3221 meta_blocks += 1 + (1LL << (bits-2)); 3222 /* tripple indirect blocks */ 3223 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 3224 3225 upper_limit -= meta_blocks; 3226 upper_limit <<= bits; 3227 3228 res += 1LL << (bits-2); 3229 res += 1LL << (2*(bits-2)); 3230 res += 1LL << (3*(bits-2)); 3231 res <<= bits; 3232 if (res > upper_limit) 3233 res = upper_limit; 3234 3235 if (res > MAX_LFS_FILESIZE) 3236 res = MAX_LFS_FILESIZE; 3237 3238 return res; 3239 } 3240 3241 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3242 ext4_fsblk_t logical_sb_block, int nr) 3243 { 3244 struct ext4_sb_info *sbi = EXT4_SB(sb); 3245 ext4_group_t bg, first_meta_bg; 3246 int has_super = 0; 3247 3248 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3249 3250 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3251 return logical_sb_block + nr + 1; 3252 bg = sbi->s_desc_per_block * nr; 3253 if (ext4_bg_has_super(sb, bg)) 3254 has_super = 1; 3255 3256 /* 3257 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3258 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3259 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3260 * compensate. 3261 */ 3262 if (sb->s_blocksize == 1024 && nr == 0 && 3263 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3264 has_super++; 3265 3266 return (has_super + ext4_group_first_block_no(sb, bg)); 3267 } 3268 3269 /** 3270 * ext4_get_stripe_size: Get the stripe size. 3271 * @sbi: In memory super block info 3272 * 3273 * If we have specified it via mount option, then 3274 * use the mount option value. If the value specified at mount time is 3275 * greater than the blocks per group use the super block value. 3276 * If the super block value is greater than blocks per group return 0. 3277 * Allocator needs it be less than blocks per group. 3278 * 3279 */ 3280 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3281 { 3282 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3283 unsigned long stripe_width = 3284 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3285 int ret; 3286 3287 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3288 ret = sbi->s_stripe; 3289 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3290 ret = stripe_width; 3291 else if (stride && stride <= sbi->s_blocks_per_group) 3292 ret = stride; 3293 else 3294 ret = 0; 3295 3296 /* 3297 * If the stripe width is 1, this makes no sense and 3298 * we set it to 0 to turn off stripe handling code. 3299 */ 3300 if (ret <= 1) 3301 ret = 0; 3302 3303 return ret; 3304 } 3305 3306 /* 3307 * Check whether this filesystem can be mounted based on 3308 * the features present and the RDONLY/RDWR mount requested. 3309 * Returns 1 if this filesystem can be mounted as requested, 3310 * 0 if it cannot be. 3311 */ 3312 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 3313 { 3314 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3315 ext4_msg(sb, KERN_ERR, 3316 "Couldn't mount because of " 3317 "unsupported optional features (%x)", 3318 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3319 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3320 return 0; 3321 } 3322 3323 #ifndef CONFIG_UNICODE 3324 if (ext4_has_feature_casefold(sb)) { 3325 ext4_msg(sb, KERN_ERR, 3326 "Filesystem with casefold feature cannot be " 3327 "mounted without CONFIG_UNICODE"); 3328 return 0; 3329 } 3330 #endif 3331 3332 if (readonly) 3333 return 1; 3334 3335 if (ext4_has_feature_readonly(sb)) { 3336 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3337 sb->s_flags |= SB_RDONLY; 3338 return 1; 3339 } 3340 3341 /* Check that feature set is OK for a read-write mount */ 3342 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3343 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3344 "unsupported optional features (%x)", 3345 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3346 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3347 return 0; 3348 } 3349 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3350 ext4_msg(sb, KERN_ERR, 3351 "Can't support bigalloc feature without " 3352 "extents feature\n"); 3353 return 0; 3354 } 3355 3356 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3357 if (!readonly && (ext4_has_feature_quota(sb) || 3358 ext4_has_feature_project(sb))) { 3359 ext4_msg(sb, KERN_ERR, 3360 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3361 return 0; 3362 } 3363 #endif /* CONFIG_QUOTA */ 3364 return 1; 3365 } 3366 3367 /* 3368 * This function is called once a day if we have errors logged 3369 * on the file system 3370 */ 3371 static void print_daily_error_info(struct timer_list *t) 3372 { 3373 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3374 struct super_block *sb = sbi->s_sb; 3375 struct ext4_super_block *es = sbi->s_es; 3376 3377 if (es->s_error_count) 3378 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3379 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3380 le32_to_cpu(es->s_error_count)); 3381 if (es->s_first_error_time) { 3382 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3383 sb->s_id, 3384 ext4_get_tstamp(es, s_first_error_time), 3385 (int) sizeof(es->s_first_error_func), 3386 es->s_first_error_func, 3387 le32_to_cpu(es->s_first_error_line)); 3388 if (es->s_first_error_ino) 3389 printk(KERN_CONT ": inode %u", 3390 le32_to_cpu(es->s_first_error_ino)); 3391 if (es->s_first_error_block) 3392 printk(KERN_CONT ": block %llu", (unsigned long long) 3393 le64_to_cpu(es->s_first_error_block)); 3394 printk(KERN_CONT "\n"); 3395 } 3396 if (es->s_last_error_time) { 3397 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3398 sb->s_id, 3399 ext4_get_tstamp(es, s_last_error_time), 3400 (int) sizeof(es->s_last_error_func), 3401 es->s_last_error_func, 3402 le32_to_cpu(es->s_last_error_line)); 3403 if (es->s_last_error_ino) 3404 printk(KERN_CONT ": inode %u", 3405 le32_to_cpu(es->s_last_error_ino)); 3406 if (es->s_last_error_block) 3407 printk(KERN_CONT ": block %llu", (unsigned long long) 3408 le64_to_cpu(es->s_last_error_block)); 3409 printk(KERN_CONT "\n"); 3410 } 3411 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3412 } 3413 3414 /* Find next suitable group and run ext4_init_inode_table */ 3415 static int ext4_run_li_request(struct ext4_li_request *elr) 3416 { 3417 struct ext4_group_desc *gdp = NULL; 3418 struct super_block *sb = elr->lr_super; 3419 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3420 ext4_group_t group = elr->lr_next_group; 3421 unsigned long timeout = 0; 3422 unsigned int prefetch_ios = 0; 3423 int ret = 0; 3424 3425 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3426 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3427 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3428 if (prefetch_ios) 3429 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3430 prefetch_ios); 3431 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3432 prefetch_ios); 3433 if (group >= elr->lr_next_group) { 3434 ret = 1; 3435 if (elr->lr_first_not_zeroed != ngroups && 3436 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3437 elr->lr_next_group = elr->lr_first_not_zeroed; 3438 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3439 ret = 0; 3440 } 3441 } 3442 return ret; 3443 } 3444 3445 for (; group < ngroups; group++) { 3446 gdp = ext4_get_group_desc(sb, group, NULL); 3447 if (!gdp) { 3448 ret = 1; 3449 break; 3450 } 3451 3452 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3453 break; 3454 } 3455 3456 if (group >= ngroups) 3457 ret = 1; 3458 3459 if (!ret) { 3460 timeout = jiffies; 3461 ret = ext4_init_inode_table(sb, group, 3462 elr->lr_timeout ? 0 : 1); 3463 trace_ext4_lazy_itable_init(sb, group); 3464 if (elr->lr_timeout == 0) { 3465 timeout = (jiffies - timeout) * 3466 EXT4_SB(elr->lr_super)->s_li_wait_mult; 3467 elr->lr_timeout = timeout; 3468 } 3469 elr->lr_next_sched = jiffies + elr->lr_timeout; 3470 elr->lr_next_group = group + 1; 3471 } 3472 return ret; 3473 } 3474 3475 /* 3476 * Remove lr_request from the list_request and free the 3477 * request structure. Should be called with li_list_mtx held 3478 */ 3479 static void ext4_remove_li_request(struct ext4_li_request *elr) 3480 { 3481 if (!elr) 3482 return; 3483 3484 list_del(&elr->lr_request); 3485 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3486 kfree(elr); 3487 } 3488 3489 static void ext4_unregister_li_request(struct super_block *sb) 3490 { 3491 mutex_lock(&ext4_li_mtx); 3492 if (!ext4_li_info) { 3493 mutex_unlock(&ext4_li_mtx); 3494 return; 3495 } 3496 3497 mutex_lock(&ext4_li_info->li_list_mtx); 3498 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3499 mutex_unlock(&ext4_li_info->li_list_mtx); 3500 mutex_unlock(&ext4_li_mtx); 3501 } 3502 3503 static struct task_struct *ext4_lazyinit_task; 3504 3505 /* 3506 * This is the function where ext4lazyinit thread lives. It walks 3507 * through the request list searching for next scheduled filesystem. 3508 * When such a fs is found, run the lazy initialization request 3509 * (ext4_rn_li_request) and keep track of the time spend in this 3510 * function. Based on that time we compute next schedule time of 3511 * the request. When walking through the list is complete, compute 3512 * next waking time and put itself into sleep. 3513 */ 3514 static int ext4_lazyinit_thread(void *arg) 3515 { 3516 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3517 struct list_head *pos, *n; 3518 struct ext4_li_request *elr; 3519 unsigned long next_wakeup, cur; 3520 3521 BUG_ON(NULL == eli); 3522 3523 cont_thread: 3524 while (true) { 3525 next_wakeup = MAX_JIFFY_OFFSET; 3526 3527 mutex_lock(&eli->li_list_mtx); 3528 if (list_empty(&eli->li_request_list)) { 3529 mutex_unlock(&eli->li_list_mtx); 3530 goto exit_thread; 3531 } 3532 list_for_each_safe(pos, n, &eli->li_request_list) { 3533 int err = 0; 3534 int progress = 0; 3535 elr = list_entry(pos, struct ext4_li_request, 3536 lr_request); 3537 3538 if (time_before(jiffies, elr->lr_next_sched)) { 3539 if (time_before(elr->lr_next_sched, next_wakeup)) 3540 next_wakeup = elr->lr_next_sched; 3541 continue; 3542 } 3543 if (down_read_trylock(&elr->lr_super->s_umount)) { 3544 if (sb_start_write_trylock(elr->lr_super)) { 3545 progress = 1; 3546 /* 3547 * We hold sb->s_umount, sb can not 3548 * be removed from the list, it is 3549 * now safe to drop li_list_mtx 3550 */ 3551 mutex_unlock(&eli->li_list_mtx); 3552 err = ext4_run_li_request(elr); 3553 sb_end_write(elr->lr_super); 3554 mutex_lock(&eli->li_list_mtx); 3555 n = pos->next; 3556 } 3557 up_read((&elr->lr_super->s_umount)); 3558 } 3559 /* error, remove the lazy_init job */ 3560 if (err) { 3561 ext4_remove_li_request(elr); 3562 continue; 3563 } 3564 if (!progress) { 3565 elr->lr_next_sched = jiffies + 3566 (prandom_u32() 3567 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3568 } 3569 if (time_before(elr->lr_next_sched, next_wakeup)) 3570 next_wakeup = elr->lr_next_sched; 3571 } 3572 mutex_unlock(&eli->li_list_mtx); 3573 3574 try_to_freeze(); 3575 3576 cur = jiffies; 3577 if ((time_after_eq(cur, next_wakeup)) || 3578 (MAX_JIFFY_OFFSET == next_wakeup)) { 3579 cond_resched(); 3580 continue; 3581 } 3582 3583 schedule_timeout_interruptible(next_wakeup - cur); 3584 3585 if (kthread_should_stop()) { 3586 ext4_clear_request_list(); 3587 goto exit_thread; 3588 } 3589 } 3590 3591 exit_thread: 3592 /* 3593 * It looks like the request list is empty, but we need 3594 * to check it under the li_list_mtx lock, to prevent any 3595 * additions into it, and of course we should lock ext4_li_mtx 3596 * to atomically free the list and ext4_li_info, because at 3597 * this point another ext4 filesystem could be registering 3598 * new one. 3599 */ 3600 mutex_lock(&ext4_li_mtx); 3601 mutex_lock(&eli->li_list_mtx); 3602 if (!list_empty(&eli->li_request_list)) { 3603 mutex_unlock(&eli->li_list_mtx); 3604 mutex_unlock(&ext4_li_mtx); 3605 goto cont_thread; 3606 } 3607 mutex_unlock(&eli->li_list_mtx); 3608 kfree(ext4_li_info); 3609 ext4_li_info = NULL; 3610 mutex_unlock(&ext4_li_mtx); 3611 3612 return 0; 3613 } 3614 3615 static void ext4_clear_request_list(void) 3616 { 3617 struct list_head *pos, *n; 3618 struct ext4_li_request *elr; 3619 3620 mutex_lock(&ext4_li_info->li_list_mtx); 3621 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3622 elr = list_entry(pos, struct ext4_li_request, 3623 lr_request); 3624 ext4_remove_li_request(elr); 3625 } 3626 mutex_unlock(&ext4_li_info->li_list_mtx); 3627 } 3628 3629 static int ext4_run_lazyinit_thread(void) 3630 { 3631 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3632 ext4_li_info, "ext4lazyinit"); 3633 if (IS_ERR(ext4_lazyinit_task)) { 3634 int err = PTR_ERR(ext4_lazyinit_task); 3635 ext4_clear_request_list(); 3636 kfree(ext4_li_info); 3637 ext4_li_info = NULL; 3638 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3639 "initialization thread\n", 3640 err); 3641 return err; 3642 } 3643 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3644 return 0; 3645 } 3646 3647 /* 3648 * Check whether it make sense to run itable init. thread or not. 3649 * If there is at least one uninitialized inode table, return 3650 * corresponding group number, else the loop goes through all 3651 * groups and return total number of groups. 3652 */ 3653 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3654 { 3655 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3656 struct ext4_group_desc *gdp = NULL; 3657 3658 if (!ext4_has_group_desc_csum(sb)) 3659 return ngroups; 3660 3661 for (group = 0; group < ngroups; group++) { 3662 gdp = ext4_get_group_desc(sb, group, NULL); 3663 if (!gdp) 3664 continue; 3665 3666 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3667 break; 3668 } 3669 3670 return group; 3671 } 3672 3673 static int ext4_li_info_new(void) 3674 { 3675 struct ext4_lazy_init *eli = NULL; 3676 3677 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3678 if (!eli) 3679 return -ENOMEM; 3680 3681 INIT_LIST_HEAD(&eli->li_request_list); 3682 mutex_init(&eli->li_list_mtx); 3683 3684 eli->li_state |= EXT4_LAZYINIT_QUIT; 3685 3686 ext4_li_info = eli; 3687 3688 return 0; 3689 } 3690 3691 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3692 ext4_group_t start) 3693 { 3694 struct ext4_li_request *elr; 3695 3696 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3697 if (!elr) 3698 return NULL; 3699 3700 elr->lr_super = sb; 3701 elr->lr_first_not_zeroed = start; 3702 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3703 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3704 elr->lr_next_group = start; 3705 } else { 3706 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3707 } 3708 3709 /* 3710 * Randomize first schedule time of the request to 3711 * spread the inode table initialization requests 3712 * better. 3713 */ 3714 elr->lr_next_sched = jiffies + (prandom_u32() % 3715 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3716 return elr; 3717 } 3718 3719 int ext4_register_li_request(struct super_block *sb, 3720 ext4_group_t first_not_zeroed) 3721 { 3722 struct ext4_sb_info *sbi = EXT4_SB(sb); 3723 struct ext4_li_request *elr = NULL; 3724 ext4_group_t ngroups = sbi->s_groups_count; 3725 int ret = 0; 3726 3727 mutex_lock(&ext4_li_mtx); 3728 if (sbi->s_li_request != NULL) { 3729 /* 3730 * Reset timeout so it can be computed again, because 3731 * s_li_wait_mult might have changed. 3732 */ 3733 sbi->s_li_request->lr_timeout = 0; 3734 goto out; 3735 } 3736 3737 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 3738 (first_not_zeroed == ngroups || sb_rdonly(sb) || 3739 !test_opt(sb, INIT_INODE_TABLE))) 3740 goto out; 3741 3742 elr = ext4_li_request_new(sb, first_not_zeroed); 3743 if (!elr) { 3744 ret = -ENOMEM; 3745 goto out; 3746 } 3747 3748 if (NULL == ext4_li_info) { 3749 ret = ext4_li_info_new(); 3750 if (ret) 3751 goto out; 3752 } 3753 3754 mutex_lock(&ext4_li_info->li_list_mtx); 3755 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3756 mutex_unlock(&ext4_li_info->li_list_mtx); 3757 3758 sbi->s_li_request = elr; 3759 /* 3760 * set elr to NULL here since it has been inserted to 3761 * the request_list and the removal and free of it is 3762 * handled by ext4_clear_request_list from now on. 3763 */ 3764 elr = NULL; 3765 3766 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3767 ret = ext4_run_lazyinit_thread(); 3768 if (ret) 3769 goto out; 3770 } 3771 out: 3772 mutex_unlock(&ext4_li_mtx); 3773 if (ret) 3774 kfree(elr); 3775 return ret; 3776 } 3777 3778 /* 3779 * We do not need to lock anything since this is called on 3780 * module unload. 3781 */ 3782 static void ext4_destroy_lazyinit_thread(void) 3783 { 3784 /* 3785 * If thread exited earlier 3786 * there's nothing to be done. 3787 */ 3788 if (!ext4_li_info || !ext4_lazyinit_task) 3789 return; 3790 3791 kthread_stop(ext4_lazyinit_task); 3792 } 3793 3794 static int set_journal_csum_feature_set(struct super_block *sb) 3795 { 3796 int ret = 1; 3797 int compat, incompat; 3798 struct ext4_sb_info *sbi = EXT4_SB(sb); 3799 3800 if (ext4_has_metadata_csum(sb)) { 3801 /* journal checksum v3 */ 3802 compat = 0; 3803 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3804 } else { 3805 /* journal checksum v1 */ 3806 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3807 incompat = 0; 3808 } 3809 3810 jbd2_journal_clear_features(sbi->s_journal, 3811 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3812 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3813 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3814 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3815 ret = jbd2_journal_set_features(sbi->s_journal, 3816 compat, 0, 3817 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3818 incompat); 3819 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3820 ret = jbd2_journal_set_features(sbi->s_journal, 3821 compat, 0, 3822 incompat); 3823 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3824 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3825 } else { 3826 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3827 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3828 } 3829 3830 return ret; 3831 } 3832 3833 /* 3834 * Note: calculating the overhead so we can be compatible with 3835 * historical BSD practice is quite difficult in the face of 3836 * clusters/bigalloc. This is because multiple metadata blocks from 3837 * different block group can end up in the same allocation cluster. 3838 * Calculating the exact overhead in the face of clustered allocation 3839 * requires either O(all block bitmaps) in memory or O(number of block 3840 * groups**2) in time. We will still calculate the superblock for 3841 * older file systems --- and if we come across with a bigalloc file 3842 * system with zero in s_overhead_clusters the estimate will be close to 3843 * correct especially for very large cluster sizes --- but for newer 3844 * file systems, it's better to calculate this figure once at mkfs 3845 * time, and store it in the superblock. If the superblock value is 3846 * present (even for non-bigalloc file systems), we will use it. 3847 */ 3848 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3849 char *buf) 3850 { 3851 struct ext4_sb_info *sbi = EXT4_SB(sb); 3852 struct ext4_group_desc *gdp; 3853 ext4_fsblk_t first_block, last_block, b; 3854 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3855 int s, j, count = 0; 3856 3857 if (!ext4_has_feature_bigalloc(sb)) 3858 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3859 sbi->s_itb_per_group + 2); 3860 3861 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3862 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3863 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3864 for (i = 0; i < ngroups; i++) { 3865 gdp = ext4_get_group_desc(sb, i, NULL); 3866 b = ext4_block_bitmap(sb, gdp); 3867 if (b >= first_block && b <= last_block) { 3868 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3869 count++; 3870 } 3871 b = ext4_inode_bitmap(sb, gdp); 3872 if (b >= first_block && b <= last_block) { 3873 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3874 count++; 3875 } 3876 b = ext4_inode_table(sb, gdp); 3877 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3878 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3879 int c = EXT4_B2C(sbi, b - first_block); 3880 ext4_set_bit(c, buf); 3881 count++; 3882 } 3883 if (i != grp) 3884 continue; 3885 s = 0; 3886 if (ext4_bg_has_super(sb, grp)) { 3887 ext4_set_bit(s++, buf); 3888 count++; 3889 } 3890 j = ext4_bg_num_gdb(sb, grp); 3891 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3892 ext4_error(sb, "Invalid number of block group " 3893 "descriptor blocks: %d", j); 3894 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3895 } 3896 count += j; 3897 for (; j > 0; j--) 3898 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3899 } 3900 if (!count) 3901 return 0; 3902 return EXT4_CLUSTERS_PER_GROUP(sb) - 3903 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3904 } 3905 3906 /* 3907 * Compute the overhead and stash it in sbi->s_overhead 3908 */ 3909 int ext4_calculate_overhead(struct super_block *sb) 3910 { 3911 struct ext4_sb_info *sbi = EXT4_SB(sb); 3912 struct ext4_super_block *es = sbi->s_es; 3913 struct inode *j_inode; 3914 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3915 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3916 ext4_fsblk_t overhead = 0; 3917 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3918 3919 if (!buf) 3920 return -ENOMEM; 3921 3922 /* 3923 * Compute the overhead (FS structures). This is constant 3924 * for a given filesystem unless the number of block groups 3925 * changes so we cache the previous value until it does. 3926 */ 3927 3928 /* 3929 * All of the blocks before first_data_block are overhead 3930 */ 3931 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3932 3933 /* 3934 * Add the overhead found in each block group 3935 */ 3936 for (i = 0; i < ngroups; i++) { 3937 int blks; 3938 3939 blks = count_overhead(sb, i, buf); 3940 overhead += blks; 3941 if (blks) 3942 memset(buf, 0, PAGE_SIZE); 3943 cond_resched(); 3944 } 3945 3946 /* 3947 * Add the internal journal blocks whether the journal has been 3948 * loaded or not 3949 */ 3950 if (sbi->s_journal && !sbi->s_journal_bdev) 3951 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 3952 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 3953 /* j_inum for internal journal is non-zero */ 3954 j_inode = ext4_get_journal_inode(sb, j_inum); 3955 if (j_inode) { 3956 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3957 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3958 iput(j_inode); 3959 } else { 3960 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3961 } 3962 } 3963 sbi->s_overhead = overhead; 3964 smp_wmb(); 3965 free_page((unsigned long) buf); 3966 return 0; 3967 } 3968 3969 static void ext4_set_resv_clusters(struct super_block *sb) 3970 { 3971 ext4_fsblk_t resv_clusters; 3972 struct ext4_sb_info *sbi = EXT4_SB(sb); 3973 3974 /* 3975 * There's no need to reserve anything when we aren't using extents. 3976 * The space estimates are exact, there are no unwritten extents, 3977 * hole punching doesn't need new metadata... This is needed especially 3978 * to keep ext2/3 backward compatibility. 3979 */ 3980 if (!ext4_has_feature_extents(sb)) 3981 return; 3982 /* 3983 * By default we reserve 2% or 4096 clusters, whichever is smaller. 3984 * This should cover the situations where we can not afford to run 3985 * out of space like for example punch hole, or converting 3986 * unwritten extents in delalloc path. In most cases such 3987 * allocation would require 1, or 2 blocks, higher numbers are 3988 * very rare. 3989 */ 3990 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 3991 sbi->s_cluster_bits); 3992 3993 do_div(resv_clusters, 50); 3994 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 3995 3996 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 3997 } 3998 3999 static const char *ext4_quota_mode(struct super_block *sb) 4000 { 4001 #ifdef CONFIG_QUOTA 4002 if (!ext4_quota_capable(sb)) 4003 return "none"; 4004 4005 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4006 return "journalled"; 4007 else 4008 return "writeback"; 4009 #else 4010 return "disabled"; 4011 #endif 4012 } 4013 4014 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 4015 { 4016 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 4017 char *orig_data = kstrdup(data, GFP_KERNEL); 4018 struct buffer_head *bh, **group_desc; 4019 struct ext4_super_block *es = NULL; 4020 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4021 struct flex_groups **flex_groups; 4022 ext4_fsblk_t block; 4023 ext4_fsblk_t sb_block = get_sb_block(&data); 4024 ext4_fsblk_t logical_sb_block; 4025 unsigned long offset = 0; 4026 unsigned long def_mount_opts; 4027 struct inode *root; 4028 const char *descr; 4029 int ret = -ENOMEM; 4030 int blocksize, clustersize; 4031 unsigned int db_count; 4032 unsigned int i; 4033 int needs_recovery, has_huge_files; 4034 __u64 blocks_count; 4035 int err = 0; 4036 ext4_group_t first_not_zeroed; 4037 struct ext4_parsed_options parsed_opts; 4038 4039 /* Set defaults for the variables that will be set during parsing */ 4040 parsed_opts.journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 4041 parsed_opts.journal_devnum = 0; 4042 parsed_opts.mb_optimize_scan = DEFAULT_MB_OPTIMIZE_SCAN; 4043 4044 if ((data && !orig_data) || !sbi) 4045 goto out_free_base; 4046 4047 sbi->s_daxdev = dax_dev; 4048 sbi->s_blockgroup_lock = 4049 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4050 if (!sbi->s_blockgroup_lock) 4051 goto out_free_base; 4052 4053 sb->s_fs_info = sbi; 4054 sbi->s_sb = sb; 4055 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 4056 sbi->s_sb_block = sb_block; 4057 sbi->s_sectors_written_start = 4058 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 4059 4060 /* Cleanup superblock name */ 4061 strreplace(sb->s_id, '/', '!'); 4062 4063 /* -EINVAL is default */ 4064 ret = -EINVAL; 4065 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4066 if (!blocksize) { 4067 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4068 goto out_fail; 4069 } 4070 4071 /* 4072 * The ext4 superblock will not be buffer aligned for other than 1kB 4073 * block sizes. We need to calculate the offset from buffer start. 4074 */ 4075 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4076 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4077 offset = do_div(logical_sb_block, blocksize); 4078 } else { 4079 logical_sb_block = sb_block; 4080 } 4081 4082 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4083 if (IS_ERR(bh)) { 4084 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4085 ret = PTR_ERR(bh); 4086 goto out_fail; 4087 } 4088 /* 4089 * Note: s_es must be initialized as soon as possible because 4090 * some ext4 macro-instructions depend on its value 4091 */ 4092 es = (struct ext4_super_block *) (bh->b_data + offset); 4093 sbi->s_es = es; 4094 sb->s_magic = le16_to_cpu(es->s_magic); 4095 if (sb->s_magic != EXT4_SUPER_MAGIC) 4096 goto cantfind_ext4; 4097 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 4098 4099 /* Warn if metadata_csum and gdt_csum are both set. */ 4100 if (ext4_has_feature_metadata_csum(sb) && 4101 ext4_has_feature_gdt_csum(sb)) 4102 ext4_warning(sb, "metadata_csum and uninit_bg are " 4103 "redundant flags; please run fsck."); 4104 4105 /* Check for a known checksum algorithm */ 4106 if (!ext4_verify_csum_type(sb, es)) { 4107 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4108 "unknown checksum algorithm."); 4109 silent = 1; 4110 goto cantfind_ext4; 4111 } 4112 4113 /* Load the checksum driver */ 4114 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4115 if (IS_ERR(sbi->s_chksum_driver)) { 4116 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4117 ret = PTR_ERR(sbi->s_chksum_driver); 4118 sbi->s_chksum_driver = NULL; 4119 goto failed_mount; 4120 } 4121 4122 /* Check superblock checksum */ 4123 if (!ext4_superblock_csum_verify(sb, es)) { 4124 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4125 "invalid superblock checksum. Run e2fsck?"); 4126 silent = 1; 4127 ret = -EFSBADCRC; 4128 goto cantfind_ext4; 4129 } 4130 4131 /* Precompute checksum seed for all metadata */ 4132 if (ext4_has_feature_csum_seed(sb)) 4133 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4134 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4135 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4136 sizeof(es->s_uuid)); 4137 4138 /* Set defaults before we parse the mount options */ 4139 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4140 set_opt(sb, INIT_INODE_TABLE); 4141 if (def_mount_opts & EXT4_DEFM_DEBUG) 4142 set_opt(sb, DEBUG); 4143 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4144 set_opt(sb, GRPID); 4145 if (def_mount_opts & EXT4_DEFM_UID16) 4146 set_opt(sb, NO_UID32); 4147 /* xattr user namespace & acls are now defaulted on */ 4148 set_opt(sb, XATTR_USER); 4149 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4150 set_opt(sb, POSIX_ACL); 4151 #endif 4152 if (ext4_has_feature_fast_commit(sb)) 4153 set_opt2(sb, JOURNAL_FAST_COMMIT); 4154 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4155 if (ext4_has_metadata_csum(sb)) 4156 set_opt(sb, JOURNAL_CHECKSUM); 4157 4158 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4159 set_opt(sb, JOURNAL_DATA); 4160 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4161 set_opt(sb, ORDERED_DATA); 4162 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4163 set_opt(sb, WRITEBACK_DATA); 4164 4165 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 4166 set_opt(sb, ERRORS_PANIC); 4167 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 4168 set_opt(sb, ERRORS_CONT); 4169 else 4170 set_opt(sb, ERRORS_RO); 4171 /* block_validity enabled by default; disable with noblock_validity */ 4172 set_opt(sb, BLOCK_VALIDITY); 4173 if (def_mount_opts & EXT4_DEFM_DISCARD) 4174 set_opt(sb, DISCARD); 4175 4176 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 4177 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 4178 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 4179 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 4180 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 4181 4182 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4183 set_opt(sb, BARRIER); 4184 4185 /* 4186 * enable delayed allocation by default 4187 * Use -o nodelalloc to turn it off 4188 */ 4189 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4190 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4191 set_opt(sb, DELALLOC); 4192 4193 /* 4194 * set default s_li_wait_mult for lazyinit, for the case there is 4195 * no mount option specified. 4196 */ 4197 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 4198 4199 if (le32_to_cpu(es->s_log_block_size) > 4200 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4201 ext4_msg(sb, KERN_ERR, 4202 "Invalid log block size: %u", 4203 le32_to_cpu(es->s_log_block_size)); 4204 goto failed_mount; 4205 } 4206 if (le32_to_cpu(es->s_log_cluster_size) > 4207 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4208 ext4_msg(sb, KERN_ERR, 4209 "Invalid log cluster size: %u", 4210 le32_to_cpu(es->s_log_cluster_size)); 4211 goto failed_mount; 4212 } 4213 4214 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4215 4216 if (blocksize == PAGE_SIZE) 4217 set_opt(sb, DIOREAD_NOLOCK); 4218 4219 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4220 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4221 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4222 } else { 4223 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4224 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4225 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4226 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4227 sbi->s_first_ino); 4228 goto failed_mount; 4229 } 4230 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4231 (!is_power_of_2(sbi->s_inode_size)) || 4232 (sbi->s_inode_size > blocksize)) { 4233 ext4_msg(sb, KERN_ERR, 4234 "unsupported inode size: %d", 4235 sbi->s_inode_size); 4236 ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize); 4237 goto failed_mount; 4238 } 4239 /* 4240 * i_atime_extra is the last extra field available for 4241 * [acm]times in struct ext4_inode. Checking for that 4242 * field should suffice to ensure we have extra space 4243 * for all three. 4244 */ 4245 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4246 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4247 sb->s_time_gran = 1; 4248 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4249 } else { 4250 sb->s_time_gran = NSEC_PER_SEC; 4251 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4252 } 4253 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4254 } 4255 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4256 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4257 EXT4_GOOD_OLD_INODE_SIZE; 4258 if (ext4_has_feature_extra_isize(sb)) { 4259 unsigned v, max = (sbi->s_inode_size - 4260 EXT4_GOOD_OLD_INODE_SIZE); 4261 4262 v = le16_to_cpu(es->s_want_extra_isize); 4263 if (v > max) { 4264 ext4_msg(sb, KERN_ERR, 4265 "bad s_want_extra_isize: %d", v); 4266 goto failed_mount; 4267 } 4268 if (sbi->s_want_extra_isize < v) 4269 sbi->s_want_extra_isize = v; 4270 4271 v = le16_to_cpu(es->s_min_extra_isize); 4272 if (v > max) { 4273 ext4_msg(sb, KERN_ERR, 4274 "bad s_min_extra_isize: %d", v); 4275 goto failed_mount; 4276 } 4277 if (sbi->s_want_extra_isize < v) 4278 sbi->s_want_extra_isize = v; 4279 } 4280 } 4281 4282 if (sbi->s_es->s_mount_opts[0]) { 4283 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 4284 sizeof(sbi->s_es->s_mount_opts), 4285 GFP_KERNEL); 4286 if (!s_mount_opts) 4287 goto failed_mount; 4288 if (!parse_options(s_mount_opts, sb, &parsed_opts, 0)) { 4289 ext4_msg(sb, KERN_WARNING, 4290 "failed to parse options in superblock: %s", 4291 s_mount_opts); 4292 } 4293 kfree(s_mount_opts); 4294 } 4295 sbi->s_def_mount_opt = sbi->s_mount_opt; 4296 if (!parse_options((char *) data, sb, &parsed_opts, 0)) 4297 goto failed_mount; 4298 4299 #ifdef CONFIG_UNICODE 4300 if (ext4_has_feature_casefold(sb) && !sb->s_encoding) { 4301 const struct ext4_sb_encodings *encoding_info; 4302 struct unicode_map *encoding; 4303 __u16 encoding_flags; 4304 4305 if (ext4_sb_read_encoding(es, &encoding_info, 4306 &encoding_flags)) { 4307 ext4_msg(sb, KERN_ERR, 4308 "Encoding requested by superblock is unknown"); 4309 goto failed_mount; 4310 } 4311 4312 encoding = utf8_load(encoding_info->version); 4313 if (IS_ERR(encoding)) { 4314 ext4_msg(sb, KERN_ERR, 4315 "can't mount with superblock charset: %s-%s " 4316 "not supported by the kernel. flags: 0x%x.", 4317 encoding_info->name, encoding_info->version, 4318 encoding_flags); 4319 goto failed_mount; 4320 } 4321 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4322 "%s-%s with flags 0x%hx", encoding_info->name, 4323 encoding_info->version?:"\b", encoding_flags); 4324 4325 sb->s_encoding = encoding; 4326 sb->s_encoding_flags = encoding_flags; 4327 } 4328 #endif 4329 4330 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4331 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n"); 4332 /* can't mount with both data=journal and dioread_nolock. */ 4333 clear_opt(sb, DIOREAD_NOLOCK); 4334 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4335 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4336 ext4_msg(sb, KERN_ERR, "can't mount with " 4337 "both data=journal and delalloc"); 4338 goto failed_mount; 4339 } 4340 if (test_opt(sb, DAX_ALWAYS)) { 4341 ext4_msg(sb, KERN_ERR, "can't mount with " 4342 "both data=journal and dax"); 4343 goto failed_mount; 4344 } 4345 if (ext4_has_feature_encrypt(sb)) { 4346 ext4_msg(sb, KERN_WARNING, 4347 "encrypted files will use data=ordered " 4348 "instead of data journaling mode"); 4349 } 4350 if (test_opt(sb, DELALLOC)) 4351 clear_opt(sb, DELALLOC); 4352 } else { 4353 sb->s_iflags |= SB_I_CGROUPWB; 4354 } 4355 4356 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 4357 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 4358 4359 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4360 (ext4_has_compat_features(sb) || 4361 ext4_has_ro_compat_features(sb) || 4362 ext4_has_incompat_features(sb))) 4363 ext4_msg(sb, KERN_WARNING, 4364 "feature flags set on rev 0 fs, " 4365 "running e2fsck is recommended"); 4366 4367 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4368 set_opt2(sb, HURD_COMPAT); 4369 if (ext4_has_feature_64bit(sb)) { 4370 ext4_msg(sb, KERN_ERR, 4371 "The Hurd can't support 64-bit file systems"); 4372 goto failed_mount; 4373 } 4374 4375 /* 4376 * ea_inode feature uses l_i_version field which is not 4377 * available in HURD_COMPAT mode. 4378 */ 4379 if (ext4_has_feature_ea_inode(sb)) { 4380 ext4_msg(sb, KERN_ERR, 4381 "ea_inode feature is not supported for Hurd"); 4382 goto failed_mount; 4383 } 4384 } 4385 4386 if (IS_EXT2_SB(sb)) { 4387 if (ext2_feature_set_ok(sb)) 4388 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4389 "using the ext4 subsystem"); 4390 else { 4391 /* 4392 * If we're probing be silent, if this looks like 4393 * it's actually an ext[34] filesystem. 4394 */ 4395 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4396 goto failed_mount; 4397 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4398 "to feature incompatibilities"); 4399 goto failed_mount; 4400 } 4401 } 4402 4403 if (IS_EXT3_SB(sb)) { 4404 if (ext3_feature_set_ok(sb)) 4405 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4406 "using the ext4 subsystem"); 4407 else { 4408 /* 4409 * If we're probing be silent, if this looks like 4410 * it's actually an ext4 filesystem. 4411 */ 4412 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4413 goto failed_mount; 4414 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4415 "to feature incompatibilities"); 4416 goto failed_mount; 4417 } 4418 } 4419 4420 /* 4421 * Check feature flags regardless of the revision level, since we 4422 * previously didn't change the revision level when setting the flags, 4423 * so there is a chance incompat flags are set on a rev 0 filesystem. 4424 */ 4425 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4426 goto failed_mount; 4427 4428 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 4429 ext4_msg(sb, KERN_ERR, 4430 "Number of reserved GDT blocks insanely large: %d", 4431 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4432 goto failed_mount; 4433 } 4434 4435 if (bdev_dax_supported(sb->s_bdev, blocksize)) 4436 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4437 4438 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4439 if (ext4_has_feature_inline_data(sb)) { 4440 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4441 " that may contain inline data"); 4442 goto failed_mount; 4443 } 4444 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4445 ext4_msg(sb, KERN_ERR, 4446 "DAX unsupported by block device."); 4447 goto failed_mount; 4448 } 4449 } 4450 4451 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4452 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4453 es->s_encryption_level); 4454 goto failed_mount; 4455 } 4456 4457 if (sb->s_blocksize != blocksize) { 4458 /* 4459 * bh must be released before kill_bdev(), otherwise 4460 * it won't be freed and its page also. kill_bdev() 4461 * is called by sb_set_blocksize(). 4462 */ 4463 brelse(bh); 4464 /* Validate the filesystem blocksize */ 4465 if (!sb_set_blocksize(sb, blocksize)) { 4466 ext4_msg(sb, KERN_ERR, "bad block size %d", 4467 blocksize); 4468 bh = NULL; 4469 goto failed_mount; 4470 } 4471 4472 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4473 offset = do_div(logical_sb_block, blocksize); 4474 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4475 if (IS_ERR(bh)) { 4476 ext4_msg(sb, KERN_ERR, 4477 "Can't read superblock on 2nd try"); 4478 ret = PTR_ERR(bh); 4479 bh = NULL; 4480 goto failed_mount; 4481 } 4482 es = (struct ext4_super_block *)(bh->b_data + offset); 4483 sbi->s_es = es; 4484 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4485 ext4_msg(sb, KERN_ERR, 4486 "Magic mismatch, very weird!"); 4487 goto failed_mount; 4488 } 4489 } 4490 4491 has_huge_files = ext4_has_feature_huge_file(sb); 4492 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4493 has_huge_files); 4494 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4495 4496 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4497 if (ext4_has_feature_64bit(sb)) { 4498 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4499 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4500 !is_power_of_2(sbi->s_desc_size)) { 4501 ext4_msg(sb, KERN_ERR, 4502 "unsupported descriptor size %lu", 4503 sbi->s_desc_size); 4504 goto failed_mount; 4505 } 4506 } else 4507 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4508 4509 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4510 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4511 4512 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4513 if (sbi->s_inodes_per_block == 0) 4514 goto cantfind_ext4; 4515 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4516 sbi->s_inodes_per_group > blocksize * 8) { 4517 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4518 sbi->s_inodes_per_group); 4519 goto failed_mount; 4520 } 4521 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4522 sbi->s_inodes_per_block; 4523 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4524 sbi->s_sbh = bh; 4525 sbi->s_mount_state = le16_to_cpu(es->s_state); 4526 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4527 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4528 4529 for (i = 0; i < 4; i++) 4530 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4531 sbi->s_def_hash_version = es->s_def_hash_version; 4532 if (ext4_has_feature_dir_index(sb)) { 4533 i = le32_to_cpu(es->s_flags); 4534 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4535 sbi->s_hash_unsigned = 3; 4536 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4537 #ifdef __CHAR_UNSIGNED__ 4538 if (!sb_rdonly(sb)) 4539 es->s_flags |= 4540 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4541 sbi->s_hash_unsigned = 3; 4542 #else 4543 if (!sb_rdonly(sb)) 4544 es->s_flags |= 4545 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4546 #endif 4547 } 4548 } 4549 4550 /* Handle clustersize */ 4551 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4552 if (ext4_has_feature_bigalloc(sb)) { 4553 if (clustersize < blocksize) { 4554 ext4_msg(sb, KERN_ERR, 4555 "cluster size (%d) smaller than " 4556 "block size (%d)", clustersize, blocksize); 4557 goto failed_mount; 4558 } 4559 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4560 le32_to_cpu(es->s_log_block_size); 4561 sbi->s_clusters_per_group = 4562 le32_to_cpu(es->s_clusters_per_group); 4563 if (sbi->s_clusters_per_group > blocksize * 8) { 4564 ext4_msg(sb, KERN_ERR, 4565 "#clusters per group too big: %lu", 4566 sbi->s_clusters_per_group); 4567 goto failed_mount; 4568 } 4569 if (sbi->s_blocks_per_group != 4570 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4571 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4572 "clusters per group (%lu) inconsistent", 4573 sbi->s_blocks_per_group, 4574 sbi->s_clusters_per_group); 4575 goto failed_mount; 4576 } 4577 } else { 4578 if (clustersize != blocksize) { 4579 ext4_msg(sb, KERN_ERR, 4580 "fragment/cluster size (%d) != " 4581 "block size (%d)", clustersize, blocksize); 4582 goto failed_mount; 4583 } 4584 if (sbi->s_blocks_per_group > blocksize * 8) { 4585 ext4_msg(sb, KERN_ERR, 4586 "#blocks per group too big: %lu", 4587 sbi->s_blocks_per_group); 4588 goto failed_mount; 4589 } 4590 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4591 sbi->s_cluster_bits = 0; 4592 } 4593 sbi->s_cluster_ratio = clustersize / blocksize; 4594 4595 /* Do we have standard group size of clustersize * 8 blocks ? */ 4596 if (sbi->s_blocks_per_group == clustersize << 3) 4597 set_opt2(sb, STD_GROUP_SIZE); 4598 4599 /* 4600 * Test whether we have more sectors than will fit in sector_t, 4601 * and whether the max offset is addressable by the page cache. 4602 */ 4603 err = generic_check_addressable(sb->s_blocksize_bits, 4604 ext4_blocks_count(es)); 4605 if (err) { 4606 ext4_msg(sb, KERN_ERR, "filesystem" 4607 " too large to mount safely on this system"); 4608 goto failed_mount; 4609 } 4610 4611 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4612 goto cantfind_ext4; 4613 4614 /* check blocks count against device size */ 4615 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4616 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4617 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4618 "exceeds size of device (%llu blocks)", 4619 ext4_blocks_count(es), blocks_count); 4620 goto failed_mount; 4621 } 4622 4623 /* 4624 * It makes no sense for the first data block to be beyond the end 4625 * of the filesystem. 4626 */ 4627 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4628 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4629 "block %u is beyond end of filesystem (%llu)", 4630 le32_to_cpu(es->s_first_data_block), 4631 ext4_blocks_count(es)); 4632 goto failed_mount; 4633 } 4634 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4635 (sbi->s_cluster_ratio == 1)) { 4636 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4637 "block is 0 with a 1k block and cluster size"); 4638 goto failed_mount; 4639 } 4640 4641 blocks_count = (ext4_blocks_count(es) - 4642 le32_to_cpu(es->s_first_data_block) + 4643 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4644 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4645 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4646 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4647 "(block count %llu, first data block %u, " 4648 "blocks per group %lu)", blocks_count, 4649 ext4_blocks_count(es), 4650 le32_to_cpu(es->s_first_data_block), 4651 EXT4_BLOCKS_PER_GROUP(sb)); 4652 goto failed_mount; 4653 } 4654 sbi->s_groups_count = blocks_count; 4655 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4656 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4657 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4658 le32_to_cpu(es->s_inodes_count)) { 4659 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4660 le32_to_cpu(es->s_inodes_count), 4661 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4662 ret = -EINVAL; 4663 goto failed_mount; 4664 } 4665 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4666 EXT4_DESC_PER_BLOCK(sb); 4667 if (ext4_has_feature_meta_bg(sb)) { 4668 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4669 ext4_msg(sb, KERN_WARNING, 4670 "first meta block group too large: %u " 4671 "(group descriptor block count %u)", 4672 le32_to_cpu(es->s_first_meta_bg), db_count); 4673 goto failed_mount; 4674 } 4675 } 4676 rcu_assign_pointer(sbi->s_group_desc, 4677 kvmalloc_array(db_count, 4678 sizeof(struct buffer_head *), 4679 GFP_KERNEL)); 4680 if (sbi->s_group_desc == NULL) { 4681 ext4_msg(sb, KERN_ERR, "not enough memory"); 4682 ret = -ENOMEM; 4683 goto failed_mount; 4684 } 4685 4686 bgl_lock_init(sbi->s_blockgroup_lock); 4687 4688 /* Pre-read the descriptors into the buffer cache */ 4689 for (i = 0; i < db_count; i++) { 4690 block = descriptor_loc(sb, logical_sb_block, i); 4691 ext4_sb_breadahead_unmovable(sb, block); 4692 } 4693 4694 for (i = 0; i < db_count; i++) { 4695 struct buffer_head *bh; 4696 4697 block = descriptor_loc(sb, logical_sb_block, i); 4698 bh = ext4_sb_bread_unmovable(sb, block); 4699 if (IS_ERR(bh)) { 4700 ext4_msg(sb, KERN_ERR, 4701 "can't read group descriptor %d", i); 4702 db_count = i; 4703 ret = PTR_ERR(bh); 4704 goto failed_mount2; 4705 } 4706 rcu_read_lock(); 4707 rcu_dereference(sbi->s_group_desc)[i] = bh; 4708 rcu_read_unlock(); 4709 } 4710 sbi->s_gdb_count = db_count; 4711 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4712 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4713 ret = -EFSCORRUPTED; 4714 goto failed_mount2; 4715 } 4716 4717 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4718 spin_lock_init(&sbi->s_error_lock); 4719 INIT_WORK(&sbi->s_error_work, flush_stashed_error_work); 4720 4721 /* Register extent status tree shrinker */ 4722 if (ext4_es_register_shrinker(sbi)) 4723 goto failed_mount3; 4724 4725 sbi->s_stripe = ext4_get_stripe_size(sbi); 4726 sbi->s_extent_max_zeroout_kb = 32; 4727 4728 /* 4729 * set up enough so that it can read an inode 4730 */ 4731 sb->s_op = &ext4_sops; 4732 sb->s_export_op = &ext4_export_ops; 4733 sb->s_xattr = ext4_xattr_handlers; 4734 #ifdef CONFIG_FS_ENCRYPTION 4735 sb->s_cop = &ext4_cryptops; 4736 #endif 4737 #ifdef CONFIG_FS_VERITY 4738 sb->s_vop = &ext4_verityops; 4739 #endif 4740 #ifdef CONFIG_QUOTA 4741 sb->dq_op = &ext4_quota_operations; 4742 if (ext4_has_feature_quota(sb)) 4743 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4744 else 4745 sb->s_qcop = &ext4_qctl_operations; 4746 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4747 #endif 4748 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4749 4750 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4751 mutex_init(&sbi->s_orphan_lock); 4752 4753 /* Initialize fast commit stuff */ 4754 atomic_set(&sbi->s_fc_subtid, 0); 4755 atomic_set(&sbi->s_fc_ineligible_updates, 0); 4756 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4757 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4758 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4759 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4760 sbi->s_fc_bytes = 0; 4761 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4762 ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING); 4763 spin_lock_init(&sbi->s_fc_lock); 4764 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4765 sbi->s_fc_replay_state.fc_regions = NULL; 4766 sbi->s_fc_replay_state.fc_regions_size = 0; 4767 sbi->s_fc_replay_state.fc_regions_used = 0; 4768 sbi->s_fc_replay_state.fc_regions_valid = 0; 4769 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4770 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4771 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4772 4773 sb->s_root = NULL; 4774 4775 needs_recovery = (es->s_last_orphan != 0 || 4776 ext4_has_feature_journal_needs_recovery(sb)); 4777 4778 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4779 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4780 goto failed_mount3a; 4781 4782 /* 4783 * The first inode we look at is the journal inode. Don't try 4784 * root first: it may be modified in the journal! 4785 */ 4786 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4787 err = ext4_load_journal(sb, es, parsed_opts.journal_devnum); 4788 if (err) 4789 goto failed_mount3a; 4790 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4791 ext4_has_feature_journal_needs_recovery(sb)) { 4792 ext4_msg(sb, KERN_ERR, "required journal recovery " 4793 "suppressed and not mounted read-only"); 4794 goto failed_mount_wq; 4795 } else { 4796 /* Nojournal mode, all journal mount options are illegal */ 4797 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4798 ext4_msg(sb, KERN_ERR, "can't mount with " 4799 "journal_checksum, fs mounted w/o journal"); 4800 goto failed_mount_wq; 4801 } 4802 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4803 ext4_msg(sb, KERN_ERR, "can't mount with " 4804 "journal_async_commit, fs mounted w/o journal"); 4805 goto failed_mount_wq; 4806 } 4807 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4808 ext4_msg(sb, KERN_ERR, "can't mount with " 4809 "commit=%lu, fs mounted w/o journal", 4810 sbi->s_commit_interval / HZ); 4811 goto failed_mount_wq; 4812 } 4813 if (EXT4_MOUNT_DATA_FLAGS & 4814 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4815 ext4_msg(sb, KERN_ERR, "can't mount with " 4816 "data=, fs mounted w/o journal"); 4817 goto failed_mount_wq; 4818 } 4819 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4820 clear_opt(sb, JOURNAL_CHECKSUM); 4821 clear_opt(sb, DATA_FLAGS); 4822 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4823 sbi->s_journal = NULL; 4824 needs_recovery = 0; 4825 goto no_journal; 4826 } 4827 4828 if (ext4_has_feature_64bit(sb) && 4829 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4830 JBD2_FEATURE_INCOMPAT_64BIT)) { 4831 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4832 goto failed_mount_wq; 4833 } 4834 4835 if (!set_journal_csum_feature_set(sb)) { 4836 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4837 "feature set"); 4838 goto failed_mount_wq; 4839 } 4840 4841 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4842 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4843 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4844 ext4_msg(sb, KERN_ERR, 4845 "Failed to set fast commit journal feature"); 4846 goto failed_mount_wq; 4847 } 4848 4849 /* We have now updated the journal if required, so we can 4850 * validate the data journaling mode. */ 4851 switch (test_opt(sb, DATA_FLAGS)) { 4852 case 0: 4853 /* No mode set, assume a default based on the journal 4854 * capabilities: ORDERED_DATA if the journal can 4855 * cope, else JOURNAL_DATA 4856 */ 4857 if (jbd2_journal_check_available_features 4858 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4859 set_opt(sb, ORDERED_DATA); 4860 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4861 } else { 4862 set_opt(sb, JOURNAL_DATA); 4863 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4864 } 4865 break; 4866 4867 case EXT4_MOUNT_ORDERED_DATA: 4868 case EXT4_MOUNT_WRITEBACK_DATA: 4869 if (!jbd2_journal_check_available_features 4870 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4871 ext4_msg(sb, KERN_ERR, "Journal does not support " 4872 "requested data journaling mode"); 4873 goto failed_mount_wq; 4874 } 4875 break; 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, parsed_opts.journal_ioprio); 4888 4889 sbi->s_journal->j_submit_inode_data_buffers = 4890 ext4_journal_submit_inode_data_buffers; 4891 sbi->s_journal->j_finish_inode_data_buffers = 4892 ext4_journal_finish_inode_data_buffers; 4893 4894 no_journal: 4895 if (!test_opt(sb, NO_MBCACHE)) { 4896 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4897 if (!sbi->s_ea_block_cache) { 4898 ext4_msg(sb, KERN_ERR, 4899 "Failed to create ea_block_cache"); 4900 goto failed_mount_wq; 4901 } 4902 4903 if (ext4_has_feature_ea_inode(sb)) { 4904 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4905 if (!sbi->s_ea_inode_cache) { 4906 ext4_msg(sb, KERN_ERR, 4907 "Failed to create ea_inode_cache"); 4908 goto failed_mount_wq; 4909 } 4910 } 4911 } 4912 4913 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4914 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4915 goto failed_mount_wq; 4916 } 4917 4918 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4919 !ext4_has_feature_encrypt(sb)) { 4920 ext4_set_feature_encrypt(sb); 4921 ext4_commit_super(sb); 4922 } 4923 4924 /* 4925 * Get the # of file system overhead blocks from the 4926 * superblock if present. 4927 */ 4928 if (es->s_overhead_clusters) 4929 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4930 else { 4931 err = ext4_calculate_overhead(sb); 4932 if (err) 4933 goto failed_mount_wq; 4934 } 4935 4936 /* 4937 * The maximum number of concurrent works can be high and 4938 * concurrency isn't really necessary. Limit it to 1. 4939 */ 4940 EXT4_SB(sb)->rsv_conversion_wq = 4941 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4942 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4943 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4944 ret = -ENOMEM; 4945 goto failed_mount4; 4946 } 4947 4948 /* 4949 * The jbd2_journal_load will have done any necessary log recovery, 4950 * so we can safely mount the rest of the filesystem now. 4951 */ 4952 4953 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4954 if (IS_ERR(root)) { 4955 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4956 ret = PTR_ERR(root); 4957 root = NULL; 4958 goto failed_mount4; 4959 } 4960 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4961 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4962 iput(root); 4963 goto failed_mount4; 4964 } 4965 4966 sb->s_root = d_make_root(root); 4967 if (!sb->s_root) { 4968 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4969 ret = -ENOMEM; 4970 goto failed_mount4; 4971 } 4972 4973 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4974 if (ret == -EROFS) { 4975 sb->s_flags |= SB_RDONLY; 4976 ret = 0; 4977 } else if (ret) 4978 goto failed_mount4a; 4979 4980 ext4_set_resv_clusters(sb); 4981 4982 if (test_opt(sb, BLOCK_VALIDITY)) { 4983 err = ext4_setup_system_zone(sb); 4984 if (err) { 4985 ext4_msg(sb, KERN_ERR, "failed to initialize system " 4986 "zone (%d)", err); 4987 goto failed_mount4a; 4988 } 4989 } 4990 ext4_fc_replay_cleanup(sb); 4991 4992 ext4_ext_init(sb); 4993 4994 /* 4995 * Enable optimize_scan if number of groups is > threshold. This can be 4996 * turned off by passing "mb_optimize_scan=0". This can also be 4997 * turned on forcefully by passing "mb_optimize_scan=1". 4998 */ 4999 if (parsed_opts.mb_optimize_scan == 1) 5000 set_opt2(sb, MB_OPTIMIZE_SCAN); 5001 else if (parsed_opts.mb_optimize_scan == 0) 5002 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5003 else if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5004 set_opt2(sb, MB_OPTIMIZE_SCAN); 5005 5006 err = ext4_mb_init(sb); 5007 if (err) { 5008 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5009 err); 5010 goto failed_mount5; 5011 } 5012 5013 /* 5014 * We can only set up the journal commit callback once 5015 * mballoc is initialized 5016 */ 5017 if (sbi->s_journal) 5018 sbi->s_journal->j_commit_callback = 5019 ext4_journal_commit_callback; 5020 5021 block = ext4_count_free_clusters(sb); 5022 ext4_free_blocks_count_set(sbi->s_es, 5023 EXT4_C2B(sbi, block)); 5024 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5025 GFP_KERNEL); 5026 if (!err) { 5027 unsigned long freei = ext4_count_free_inodes(sb); 5028 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5029 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5030 GFP_KERNEL); 5031 } 5032 if (!err) 5033 err = percpu_counter_init(&sbi->s_dirs_counter, 5034 ext4_count_dirs(sb), GFP_KERNEL); 5035 if (!err) 5036 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5037 GFP_KERNEL); 5038 if (!err) 5039 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 5040 GFP_KERNEL); 5041 if (!err) 5042 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5043 5044 if (err) { 5045 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5046 goto failed_mount6; 5047 } 5048 5049 if (ext4_has_feature_flex_bg(sb)) 5050 if (!ext4_fill_flex_info(sb)) { 5051 ext4_msg(sb, KERN_ERR, 5052 "unable to initialize " 5053 "flex_bg meta info!"); 5054 ret = -ENOMEM; 5055 goto failed_mount6; 5056 } 5057 5058 err = ext4_register_li_request(sb, first_not_zeroed); 5059 if (err) 5060 goto failed_mount6; 5061 5062 err = ext4_register_sysfs(sb); 5063 if (err) 5064 goto failed_mount7; 5065 5066 #ifdef CONFIG_QUOTA 5067 /* Enable quota usage during mount. */ 5068 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5069 err = ext4_enable_quotas(sb); 5070 if (err) 5071 goto failed_mount8; 5072 } 5073 #endif /* CONFIG_QUOTA */ 5074 5075 /* 5076 * Save the original bdev mapping's wb_err value which could be 5077 * used to detect the metadata async write error. 5078 */ 5079 spin_lock_init(&sbi->s_bdev_wb_lock); 5080 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5081 &sbi->s_bdev_wb_err); 5082 sb->s_bdev->bd_super = sb; 5083 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5084 ext4_orphan_cleanup(sb, es); 5085 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5086 if (needs_recovery) { 5087 ext4_msg(sb, KERN_INFO, "recovery complete"); 5088 err = ext4_mark_recovery_complete(sb, es); 5089 if (err) 5090 goto failed_mount8; 5091 } 5092 if (EXT4_SB(sb)->s_journal) { 5093 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5094 descr = " journalled data mode"; 5095 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5096 descr = " ordered data mode"; 5097 else 5098 descr = " writeback data mode"; 5099 } else 5100 descr = "out journal"; 5101 5102 if (test_opt(sb, DISCARD)) { 5103 struct request_queue *q = bdev_get_queue(sb->s_bdev); 5104 if (!blk_queue_discard(q)) 5105 ext4_msg(sb, KERN_WARNING, 5106 "mounting with \"discard\" option, but " 5107 "the device does not support discard"); 5108 } 5109 5110 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5111 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 5112 "Opts: %.*s%s%s. Quota mode: %s.", descr, 5113 (int) sizeof(sbi->s_es->s_mount_opts), 5114 sbi->s_es->s_mount_opts, 5115 *sbi->s_es->s_mount_opts ? "; " : "", orig_data, 5116 ext4_quota_mode(sb)); 5117 5118 if (es->s_error_count) 5119 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5120 5121 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5122 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5123 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5124 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5125 atomic_set(&sbi->s_warning_count, 0); 5126 atomic_set(&sbi->s_msg_count, 0); 5127 5128 kfree(orig_data); 5129 return 0; 5130 5131 cantfind_ext4: 5132 if (!silent) 5133 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5134 goto failed_mount; 5135 5136 failed_mount8: 5137 ext4_unregister_sysfs(sb); 5138 kobject_put(&sbi->s_kobj); 5139 failed_mount7: 5140 ext4_unregister_li_request(sb); 5141 failed_mount6: 5142 ext4_mb_release(sb); 5143 rcu_read_lock(); 5144 flex_groups = rcu_dereference(sbi->s_flex_groups); 5145 if (flex_groups) { 5146 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 5147 kvfree(flex_groups[i]); 5148 kvfree(flex_groups); 5149 } 5150 rcu_read_unlock(); 5151 percpu_counter_destroy(&sbi->s_freeclusters_counter); 5152 percpu_counter_destroy(&sbi->s_freeinodes_counter); 5153 percpu_counter_destroy(&sbi->s_dirs_counter); 5154 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 5155 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 5156 percpu_free_rwsem(&sbi->s_writepages_rwsem); 5157 failed_mount5: 5158 ext4_ext_release(sb); 5159 ext4_release_system_zone(sb); 5160 failed_mount4a: 5161 dput(sb->s_root); 5162 sb->s_root = NULL; 5163 failed_mount4: 5164 ext4_msg(sb, KERN_ERR, "mount failed"); 5165 if (EXT4_SB(sb)->rsv_conversion_wq) 5166 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5167 failed_mount_wq: 5168 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5169 sbi->s_ea_inode_cache = NULL; 5170 5171 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5172 sbi->s_ea_block_cache = NULL; 5173 5174 if (sbi->s_journal) { 5175 jbd2_journal_destroy(sbi->s_journal); 5176 sbi->s_journal = NULL; 5177 } 5178 failed_mount3a: 5179 ext4_es_unregister_shrinker(sbi); 5180 failed_mount3: 5181 flush_work(&sbi->s_error_work); 5182 del_timer_sync(&sbi->s_err_report); 5183 ext4_stop_mmpd(sbi); 5184 failed_mount2: 5185 rcu_read_lock(); 5186 group_desc = rcu_dereference(sbi->s_group_desc); 5187 for (i = 0; i < db_count; i++) 5188 brelse(group_desc[i]); 5189 kvfree(group_desc); 5190 rcu_read_unlock(); 5191 failed_mount: 5192 if (sbi->s_chksum_driver) 5193 crypto_free_shash(sbi->s_chksum_driver); 5194 5195 #ifdef CONFIG_UNICODE 5196 utf8_unload(sb->s_encoding); 5197 #endif 5198 5199 #ifdef CONFIG_QUOTA 5200 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5201 kfree(get_qf_name(sb, sbi, i)); 5202 #endif 5203 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5204 /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */ 5205 brelse(bh); 5206 ext4_blkdev_remove(sbi); 5207 out_fail: 5208 sb->s_fs_info = NULL; 5209 kfree(sbi->s_blockgroup_lock); 5210 out_free_base: 5211 kfree(sbi); 5212 kfree(orig_data); 5213 fs_put_dax(dax_dev); 5214 return err ? err : ret; 5215 } 5216 5217 /* 5218 * Setup any per-fs journal parameters now. We'll do this both on 5219 * initial mount, once the journal has been initialised but before we've 5220 * done any recovery; and again on any subsequent remount. 5221 */ 5222 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5223 { 5224 struct ext4_sb_info *sbi = EXT4_SB(sb); 5225 5226 journal->j_commit_interval = sbi->s_commit_interval; 5227 journal->j_min_batch_time = sbi->s_min_batch_time; 5228 journal->j_max_batch_time = sbi->s_max_batch_time; 5229 ext4_fc_init(sb, journal); 5230 5231 write_lock(&journal->j_state_lock); 5232 if (test_opt(sb, BARRIER)) 5233 journal->j_flags |= JBD2_BARRIER; 5234 else 5235 journal->j_flags &= ~JBD2_BARRIER; 5236 if (test_opt(sb, DATA_ERR_ABORT)) 5237 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5238 else 5239 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5240 write_unlock(&journal->j_state_lock); 5241 } 5242 5243 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5244 unsigned int journal_inum) 5245 { 5246 struct inode *journal_inode; 5247 5248 /* 5249 * Test for the existence of a valid inode on disk. Bad things 5250 * happen if we iget() an unused inode, as the subsequent iput() 5251 * will try to delete it. 5252 */ 5253 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5254 if (IS_ERR(journal_inode)) { 5255 ext4_msg(sb, KERN_ERR, "no journal found"); 5256 return NULL; 5257 } 5258 if (!journal_inode->i_nlink) { 5259 make_bad_inode(journal_inode); 5260 iput(journal_inode); 5261 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5262 return NULL; 5263 } 5264 5265 jbd_debug(2, "Journal inode found at %p: %lld bytes\n", 5266 journal_inode, journal_inode->i_size); 5267 if (!S_ISREG(journal_inode->i_mode)) { 5268 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5269 iput(journal_inode); 5270 return NULL; 5271 } 5272 return journal_inode; 5273 } 5274 5275 static journal_t *ext4_get_journal(struct super_block *sb, 5276 unsigned int journal_inum) 5277 { 5278 struct inode *journal_inode; 5279 journal_t *journal; 5280 5281 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5282 return NULL; 5283 5284 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5285 if (!journal_inode) 5286 return NULL; 5287 5288 journal = jbd2_journal_init_inode(journal_inode); 5289 if (!journal) { 5290 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5291 iput(journal_inode); 5292 return NULL; 5293 } 5294 journal->j_private = sb; 5295 ext4_init_journal_params(sb, journal); 5296 return journal; 5297 } 5298 5299 static journal_t *ext4_get_dev_journal(struct super_block *sb, 5300 dev_t j_dev) 5301 { 5302 struct buffer_head *bh; 5303 journal_t *journal; 5304 ext4_fsblk_t start; 5305 ext4_fsblk_t len; 5306 int hblock, blocksize; 5307 ext4_fsblk_t sb_block; 5308 unsigned long offset; 5309 struct ext4_super_block *es; 5310 struct block_device *bdev; 5311 5312 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5313 return NULL; 5314 5315 bdev = ext4_blkdev_get(j_dev, sb); 5316 if (bdev == NULL) 5317 return NULL; 5318 5319 blocksize = sb->s_blocksize; 5320 hblock = bdev_logical_block_size(bdev); 5321 if (blocksize < hblock) { 5322 ext4_msg(sb, KERN_ERR, 5323 "blocksize too small for journal device"); 5324 goto out_bdev; 5325 } 5326 5327 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5328 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5329 set_blocksize(bdev, blocksize); 5330 if (!(bh = __bread(bdev, sb_block, blocksize))) { 5331 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5332 "external journal"); 5333 goto out_bdev; 5334 } 5335 5336 es = (struct ext4_super_block *) (bh->b_data + offset); 5337 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5338 !(le32_to_cpu(es->s_feature_incompat) & 5339 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5340 ext4_msg(sb, KERN_ERR, "external journal has " 5341 "bad superblock"); 5342 brelse(bh); 5343 goto out_bdev; 5344 } 5345 5346 if ((le32_to_cpu(es->s_feature_ro_compat) & 5347 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5348 es->s_checksum != ext4_superblock_csum(sb, es)) { 5349 ext4_msg(sb, KERN_ERR, "external journal has " 5350 "corrupt superblock"); 5351 brelse(bh); 5352 goto out_bdev; 5353 } 5354 5355 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5356 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5357 brelse(bh); 5358 goto out_bdev; 5359 } 5360 5361 len = ext4_blocks_count(es); 5362 start = sb_block + 1; 5363 brelse(bh); /* we're done with the superblock */ 5364 5365 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 5366 start, len, blocksize); 5367 if (!journal) { 5368 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5369 goto out_bdev; 5370 } 5371 journal->j_private = sb; 5372 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) { 5373 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 5374 goto out_journal; 5375 } 5376 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5377 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5378 "user (unsupported) - %d", 5379 be32_to_cpu(journal->j_superblock->s_nr_users)); 5380 goto out_journal; 5381 } 5382 EXT4_SB(sb)->s_journal_bdev = bdev; 5383 ext4_init_journal_params(sb, journal); 5384 return journal; 5385 5386 out_journal: 5387 jbd2_journal_destroy(journal); 5388 out_bdev: 5389 ext4_blkdev_put(bdev); 5390 return NULL; 5391 } 5392 5393 static int ext4_load_journal(struct super_block *sb, 5394 struct ext4_super_block *es, 5395 unsigned long journal_devnum) 5396 { 5397 journal_t *journal; 5398 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5399 dev_t journal_dev; 5400 int err = 0; 5401 int really_read_only; 5402 int journal_dev_ro; 5403 5404 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5405 return -EFSCORRUPTED; 5406 5407 if (journal_devnum && 5408 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5409 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5410 "numbers have changed"); 5411 journal_dev = new_decode_dev(journal_devnum); 5412 } else 5413 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5414 5415 if (journal_inum && journal_dev) { 5416 ext4_msg(sb, KERN_ERR, 5417 "filesystem has both journal inode and journal device!"); 5418 return -EINVAL; 5419 } 5420 5421 if (journal_inum) { 5422 journal = ext4_get_journal(sb, journal_inum); 5423 if (!journal) 5424 return -EINVAL; 5425 } else { 5426 journal = ext4_get_dev_journal(sb, journal_dev); 5427 if (!journal) 5428 return -EINVAL; 5429 } 5430 5431 journal_dev_ro = bdev_read_only(journal->j_dev); 5432 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5433 5434 if (journal_dev_ro && !sb_rdonly(sb)) { 5435 ext4_msg(sb, KERN_ERR, 5436 "journal device read-only, try mounting with '-o ro'"); 5437 err = -EROFS; 5438 goto err_out; 5439 } 5440 5441 /* 5442 * Are we loading a blank journal or performing recovery after a 5443 * crash? For recovery, we need to check in advance whether we 5444 * can get read-write access to the device. 5445 */ 5446 if (ext4_has_feature_journal_needs_recovery(sb)) { 5447 if (sb_rdonly(sb)) { 5448 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5449 "required on readonly filesystem"); 5450 if (really_read_only) { 5451 ext4_msg(sb, KERN_ERR, "write access " 5452 "unavailable, cannot proceed " 5453 "(try mounting with noload)"); 5454 err = -EROFS; 5455 goto err_out; 5456 } 5457 ext4_msg(sb, KERN_INFO, "write access will " 5458 "be enabled during recovery"); 5459 } 5460 } 5461 5462 if (!(journal->j_flags & JBD2_BARRIER)) 5463 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5464 5465 if (!ext4_has_feature_journal_needs_recovery(sb)) 5466 err = jbd2_journal_wipe(journal, !really_read_only); 5467 if (!err) { 5468 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5469 if (save) 5470 memcpy(save, ((char *) es) + 5471 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 5472 err = jbd2_journal_load(journal); 5473 if (save) 5474 memcpy(((char *) es) + EXT4_S_ERR_START, 5475 save, EXT4_S_ERR_LEN); 5476 kfree(save); 5477 } 5478 5479 if (err) { 5480 ext4_msg(sb, KERN_ERR, "error loading journal"); 5481 goto err_out; 5482 } 5483 5484 EXT4_SB(sb)->s_journal = journal; 5485 err = ext4_clear_journal_err(sb, es); 5486 if (err) { 5487 EXT4_SB(sb)->s_journal = NULL; 5488 jbd2_journal_destroy(journal); 5489 return err; 5490 } 5491 5492 if (!really_read_only && journal_devnum && 5493 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5494 es->s_journal_dev = cpu_to_le32(journal_devnum); 5495 5496 /* Make sure we flush the recovery flag to disk. */ 5497 ext4_commit_super(sb); 5498 } 5499 5500 return 0; 5501 5502 err_out: 5503 jbd2_journal_destroy(journal); 5504 return err; 5505 } 5506 5507 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 5508 static void ext4_update_super(struct super_block *sb) 5509 { 5510 struct ext4_sb_info *sbi = EXT4_SB(sb); 5511 struct ext4_super_block *es = sbi->s_es; 5512 struct buffer_head *sbh = sbi->s_sbh; 5513 5514 lock_buffer(sbh); 5515 /* 5516 * If the file system is mounted read-only, don't update the 5517 * superblock write time. This avoids updating the superblock 5518 * write time when we are mounting the root file system 5519 * read/only but we need to replay the journal; at that point, 5520 * for people who are east of GMT and who make their clock 5521 * tick in localtime for Windows bug-for-bug compatibility, 5522 * the clock is set in the future, and this will cause e2fsck 5523 * to complain and force a full file system check. 5524 */ 5525 if (!(sb->s_flags & SB_RDONLY)) 5526 ext4_update_tstamp(es, s_wtime); 5527 es->s_kbytes_written = 5528 cpu_to_le64(sbi->s_kbytes_written + 5529 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 5530 sbi->s_sectors_written_start) >> 1)); 5531 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 5532 ext4_free_blocks_count_set(es, 5533 EXT4_C2B(sbi, percpu_counter_sum_positive( 5534 &sbi->s_freeclusters_counter))); 5535 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 5536 es->s_free_inodes_count = 5537 cpu_to_le32(percpu_counter_sum_positive( 5538 &sbi->s_freeinodes_counter)); 5539 /* Copy error information to the on-disk superblock */ 5540 spin_lock(&sbi->s_error_lock); 5541 if (sbi->s_add_error_count > 0) { 5542 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5543 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 5544 __ext4_update_tstamp(&es->s_first_error_time, 5545 &es->s_first_error_time_hi, 5546 sbi->s_first_error_time); 5547 strncpy(es->s_first_error_func, sbi->s_first_error_func, 5548 sizeof(es->s_first_error_func)); 5549 es->s_first_error_line = 5550 cpu_to_le32(sbi->s_first_error_line); 5551 es->s_first_error_ino = 5552 cpu_to_le32(sbi->s_first_error_ino); 5553 es->s_first_error_block = 5554 cpu_to_le64(sbi->s_first_error_block); 5555 es->s_first_error_errcode = 5556 ext4_errno_to_code(sbi->s_first_error_code); 5557 } 5558 __ext4_update_tstamp(&es->s_last_error_time, 5559 &es->s_last_error_time_hi, 5560 sbi->s_last_error_time); 5561 strncpy(es->s_last_error_func, sbi->s_last_error_func, 5562 sizeof(es->s_last_error_func)); 5563 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 5564 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 5565 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 5566 es->s_last_error_errcode = 5567 ext4_errno_to_code(sbi->s_last_error_code); 5568 /* 5569 * Start the daily error reporting function if it hasn't been 5570 * started already 5571 */ 5572 if (!es->s_error_count) 5573 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 5574 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 5575 sbi->s_add_error_count = 0; 5576 } 5577 spin_unlock(&sbi->s_error_lock); 5578 5579 ext4_superblock_csum_set(sb); 5580 unlock_buffer(sbh); 5581 } 5582 5583 static int ext4_commit_super(struct super_block *sb) 5584 { 5585 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 5586 int error = 0; 5587 5588 if (!sbh) 5589 return -EINVAL; 5590 if (block_device_ejected(sb)) 5591 return -ENODEV; 5592 5593 ext4_update_super(sb); 5594 5595 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 5596 /* 5597 * Oh, dear. A previous attempt to write the 5598 * superblock failed. This could happen because the 5599 * USB device was yanked out. Or it could happen to 5600 * be a transient write error and maybe the block will 5601 * be remapped. Nothing we can do but to retry the 5602 * write and hope for the best. 5603 */ 5604 ext4_msg(sb, KERN_ERR, "previous I/O error to " 5605 "superblock detected"); 5606 clear_buffer_write_io_error(sbh); 5607 set_buffer_uptodate(sbh); 5608 } 5609 BUFFER_TRACE(sbh, "marking dirty"); 5610 mark_buffer_dirty(sbh); 5611 error = __sync_dirty_buffer(sbh, 5612 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0)); 5613 if (buffer_write_io_error(sbh)) { 5614 ext4_msg(sb, KERN_ERR, "I/O error while writing " 5615 "superblock"); 5616 clear_buffer_write_io_error(sbh); 5617 set_buffer_uptodate(sbh); 5618 } 5619 return error; 5620 } 5621 5622 /* 5623 * Have we just finished recovery? If so, and if we are mounting (or 5624 * remounting) the filesystem readonly, then we will end up with a 5625 * consistent fs on disk. Record that fact. 5626 */ 5627 static int ext4_mark_recovery_complete(struct super_block *sb, 5628 struct ext4_super_block *es) 5629 { 5630 int err; 5631 journal_t *journal = EXT4_SB(sb)->s_journal; 5632 5633 if (!ext4_has_feature_journal(sb)) { 5634 if (journal != NULL) { 5635 ext4_error(sb, "Journal got removed while the fs was " 5636 "mounted!"); 5637 return -EFSCORRUPTED; 5638 } 5639 return 0; 5640 } 5641 jbd2_journal_lock_updates(journal); 5642 err = jbd2_journal_flush(journal, 0); 5643 if (err < 0) 5644 goto out; 5645 5646 if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) { 5647 ext4_clear_feature_journal_needs_recovery(sb); 5648 ext4_commit_super(sb); 5649 } 5650 out: 5651 jbd2_journal_unlock_updates(journal); 5652 return err; 5653 } 5654 5655 /* 5656 * If we are mounting (or read-write remounting) a filesystem whose journal 5657 * has recorded an error from a previous lifetime, move that error to the 5658 * main filesystem now. 5659 */ 5660 static int ext4_clear_journal_err(struct super_block *sb, 5661 struct ext4_super_block *es) 5662 { 5663 journal_t *journal; 5664 int j_errno; 5665 const char *errstr; 5666 5667 if (!ext4_has_feature_journal(sb)) { 5668 ext4_error(sb, "Journal got removed while the fs was mounted!"); 5669 return -EFSCORRUPTED; 5670 } 5671 5672 journal = EXT4_SB(sb)->s_journal; 5673 5674 /* 5675 * Now check for any error status which may have been recorded in the 5676 * journal by a prior ext4_error() or ext4_abort() 5677 */ 5678 5679 j_errno = jbd2_journal_errno(journal); 5680 if (j_errno) { 5681 char nbuf[16]; 5682 5683 errstr = ext4_decode_error(sb, j_errno, nbuf); 5684 ext4_warning(sb, "Filesystem error recorded " 5685 "from previous mount: %s", errstr); 5686 ext4_warning(sb, "Marking fs in need of filesystem check."); 5687 5688 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 5689 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5690 ext4_commit_super(sb); 5691 5692 jbd2_journal_clear_err(journal); 5693 jbd2_journal_update_sb_errno(journal); 5694 } 5695 return 0; 5696 } 5697 5698 /* 5699 * Force the running and committing transactions to commit, 5700 * and wait on the commit. 5701 */ 5702 int ext4_force_commit(struct super_block *sb) 5703 { 5704 journal_t *journal; 5705 5706 if (sb_rdonly(sb)) 5707 return 0; 5708 5709 journal = EXT4_SB(sb)->s_journal; 5710 return ext4_journal_force_commit(journal); 5711 } 5712 5713 static int ext4_sync_fs(struct super_block *sb, int wait) 5714 { 5715 int ret = 0; 5716 tid_t target; 5717 bool needs_barrier = false; 5718 struct ext4_sb_info *sbi = EXT4_SB(sb); 5719 5720 if (unlikely(ext4_forced_shutdown(sbi))) 5721 return 0; 5722 5723 trace_ext4_sync_fs(sb, wait); 5724 flush_workqueue(sbi->rsv_conversion_wq); 5725 /* 5726 * Writeback quota in non-journalled quota case - journalled quota has 5727 * no dirty dquots 5728 */ 5729 dquot_writeback_dquots(sb, -1); 5730 /* 5731 * Data writeback is possible w/o journal transaction, so barrier must 5732 * being sent at the end of the function. But we can skip it if 5733 * transaction_commit will do it for us. 5734 */ 5735 if (sbi->s_journal) { 5736 target = jbd2_get_latest_transaction(sbi->s_journal); 5737 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 5738 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 5739 needs_barrier = true; 5740 5741 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 5742 if (wait) 5743 ret = jbd2_log_wait_commit(sbi->s_journal, 5744 target); 5745 } 5746 } else if (wait && test_opt(sb, BARRIER)) 5747 needs_barrier = true; 5748 if (needs_barrier) { 5749 int err; 5750 err = blkdev_issue_flush(sb->s_bdev); 5751 if (!ret) 5752 ret = err; 5753 } 5754 5755 return ret; 5756 } 5757 5758 /* 5759 * LVM calls this function before a (read-only) snapshot is created. This 5760 * gives us a chance to flush the journal completely and mark the fs clean. 5761 * 5762 * Note that only this function cannot bring a filesystem to be in a clean 5763 * state independently. It relies on upper layer to stop all data & metadata 5764 * modifications. 5765 */ 5766 static int ext4_freeze(struct super_block *sb) 5767 { 5768 int error = 0; 5769 journal_t *journal; 5770 5771 if (sb_rdonly(sb)) 5772 return 0; 5773 5774 journal = EXT4_SB(sb)->s_journal; 5775 5776 if (journal) { 5777 /* Now we set up the journal barrier. */ 5778 jbd2_journal_lock_updates(journal); 5779 5780 /* 5781 * Don't clear the needs_recovery flag if we failed to 5782 * flush the journal. 5783 */ 5784 error = jbd2_journal_flush(journal, 0); 5785 if (error < 0) 5786 goto out; 5787 5788 /* Journal blocked and flushed, clear needs_recovery flag. */ 5789 ext4_clear_feature_journal_needs_recovery(sb); 5790 } 5791 5792 error = ext4_commit_super(sb); 5793 out: 5794 if (journal) 5795 /* we rely on upper layer to stop further updates */ 5796 jbd2_journal_unlock_updates(journal); 5797 return error; 5798 } 5799 5800 /* 5801 * Called by LVM after the snapshot is done. We need to reset the RECOVER 5802 * flag here, even though the filesystem is not technically dirty yet. 5803 */ 5804 static int ext4_unfreeze(struct super_block *sb) 5805 { 5806 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 5807 return 0; 5808 5809 if (EXT4_SB(sb)->s_journal) { 5810 /* Reset the needs_recovery flag before the fs is unlocked. */ 5811 ext4_set_feature_journal_needs_recovery(sb); 5812 } 5813 5814 ext4_commit_super(sb); 5815 return 0; 5816 } 5817 5818 /* 5819 * Structure to save mount options for ext4_remount's benefit 5820 */ 5821 struct ext4_mount_options { 5822 unsigned long s_mount_opt; 5823 unsigned long s_mount_opt2; 5824 kuid_t s_resuid; 5825 kgid_t s_resgid; 5826 unsigned long s_commit_interval; 5827 u32 s_min_batch_time, s_max_batch_time; 5828 #ifdef CONFIG_QUOTA 5829 int s_jquota_fmt; 5830 char *s_qf_names[EXT4_MAXQUOTAS]; 5831 #endif 5832 }; 5833 5834 static int ext4_remount(struct super_block *sb, int *flags, char *data) 5835 { 5836 struct ext4_super_block *es; 5837 struct ext4_sb_info *sbi = EXT4_SB(sb); 5838 unsigned long old_sb_flags, vfs_flags; 5839 struct ext4_mount_options old_opts; 5840 int enable_quota = 0; 5841 ext4_group_t g; 5842 int err = 0; 5843 #ifdef CONFIG_QUOTA 5844 int i, j; 5845 char *to_free[EXT4_MAXQUOTAS]; 5846 #endif 5847 char *orig_data = kstrdup(data, GFP_KERNEL); 5848 struct ext4_parsed_options parsed_opts; 5849 5850 parsed_opts.journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5851 parsed_opts.journal_devnum = 0; 5852 5853 if (data && !orig_data) 5854 return -ENOMEM; 5855 5856 /* Store the original options */ 5857 old_sb_flags = sb->s_flags; 5858 old_opts.s_mount_opt = sbi->s_mount_opt; 5859 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 5860 old_opts.s_resuid = sbi->s_resuid; 5861 old_opts.s_resgid = sbi->s_resgid; 5862 old_opts.s_commit_interval = sbi->s_commit_interval; 5863 old_opts.s_min_batch_time = sbi->s_min_batch_time; 5864 old_opts.s_max_batch_time = sbi->s_max_batch_time; 5865 #ifdef CONFIG_QUOTA 5866 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 5867 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5868 if (sbi->s_qf_names[i]) { 5869 char *qf_name = get_qf_name(sb, sbi, i); 5870 5871 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 5872 if (!old_opts.s_qf_names[i]) { 5873 for (j = 0; j < i; j++) 5874 kfree(old_opts.s_qf_names[j]); 5875 kfree(orig_data); 5876 return -ENOMEM; 5877 } 5878 } else 5879 old_opts.s_qf_names[i] = NULL; 5880 #endif 5881 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 5882 parsed_opts.journal_ioprio = 5883 sbi->s_journal->j_task->io_context->ioprio; 5884 5885 /* 5886 * Some options can be enabled by ext4 and/or by VFS mount flag 5887 * either way we need to make sure it matches in both *flags and 5888 * s_flags. Copy those selected flags from *flags to s_flags 5889 */ 5890 vfs_flags = SB_LAZYTIME | SB_I_VERSION; 5891 sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags); 5892 5893 if (!parse_options(data, sb, &parsed_opts, 1)) { 5894 err = -EINVAL; 5895 goto restore_opts; 5896 } 5897 5898 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 5899 test_opt(sb, JOURNAL_CHECKSUM)) { 5900 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 5901 "during remount not supported; ignoring"); 5902 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 5903 } 5904 5905 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5906 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5907 ext4_msg(sb, KERN_ERR, "can't mount with " 5908 "both data=journal and delalloc"); 5909 err = -EINVAL; 5910 goto restore_opts; 5911 } 5912 if (test_opt(sb, DIOREAD_NOLOCK)) { 5913 ext4_msg(sb, KERN_ERR, "can't mount with " 5914 "both data=journal and dioread_nolock"); 5915 err = -EINVAL; 5916 goto restore_opts; 5917 } 5918 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 5919 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5920 ext4_msg(sb, KERN_ERR, "can't mount with " 5921 "journal_async_commit in data=ordered mode"); 5922 err = -EINVAL; 5923 goto restore_opts; 5924 } 5925 } 5926 5927 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 5928 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 5929 err = -EINVAL; 5930 goto restore_opts; 5931 } 5932 5933 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) 5934 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user"); 5935 5936 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5937 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5938 5939 es = sbi->s_es; 5940 5941 if (sbi->s_journal) { 5942 ext4_init_journal_params(sb, sbi->s_journal); 5943 set_task_ioprio(sbi->s_journal->j_task, parsed_opts.journal_ioprio); 5944 } 5945 5946 /* Flush outstanding errors before changing fs state */ 5947 flush_work(&sbi->s_error_work); 5948 5949 if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) { 5950 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) { 5951 err = -EROFS; 5952 goto restore_opts; 5953 } 5954 5955 if (*flags & SB_RDONLY) { 5956 err = sync_filesystem(sb); 5957 if (err < 0) 5958 goto restore_opts; 5959 err = dquot_suspend(sb, -1); 5960 if (err < 0) 5961 goto restore_opts; 5962 5963 /* 5964 * First of all, the unconditional stuff we have to do 5965 * to disable replay of the journal when we next remount 5966 */ 5967 sb->s_flags |= SB_RDONLY; 5968 5969 /* 5970 * OK, test if we are remounting a valid rw partition 5971 * readonly, and if so set the rdonly flag and then 5972 * mark the partition as valid again. 5973 */ 5974 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 5975 (sbi->s_mount_state & EXT4_VALID_FS)) 5976 es->s_state = cpu_to_le16(sbi->s_mount_state); 5977 5978 if (sbi->s_journal) { 5979 /* 5980 * We let remount-ro finish even if marking fs 5981 * as clean failed... 5982 */ 5983 ext4_mark_recovery_complete(sb, es); 5984 } 5985 } else { 5986 /* Make sure we can mount this feature set readwrite */ 5987 if (ext4_has_feature_readonly(sb) || 5988 !ext4_feature_set_ok(sb, 0)) { 5989 err = -EROFS; 5990 goto restore_opts; 5991 } 5992 /* 5993 * Make sure the group descriptor checksums 5994 * are sane. If they aren't, refuse to remount r/w. 5995 */ 5996 for (g = 0; g < sbi->s_groups_count; g++) { 5997 struct ext4_group_desc *gdp = 5998 ext4_get_group_desc(sb, g, NULL); 5999 6000 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6001 ext4_msg(sb, KERN_ERR, 6002 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6003 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6004 le16_to_cpu(gdp->bg_checksum)); 6005 err = -EFSBADCRC; 6006 goto restore_opts; 6007 } 6008 } 6009 6010 /* 6011 * If we have an unprocessed orphan list hanging 6012 * around from a previously readonly bdev mount, 6013 * require a full umount/remount for now. 6014 */ 6015 if (es->s_last_orphan) { 6016 ext4_msg(sb, KERN_WARNING, "Couldn't " 6017 "remount RDWR because of unprocessed " 6018 "orphan inode list. Please " 6019 "umount/remount instead"); 6020 err = -EINVAL; 6021 goto restore_opts; 6022 } 6023 6024 /* 6025 * Mounting a RDONLY partition read-write, so reread 6026 * and store the current valid flag. (It may have 6027 * been changed by e2fsck since we originally mounted 6028 * the partition.) 6029 */ 6030 if (sbi->s_journal) { 6031 err = ext4_clear_journal_err(sb, es); 6032 if (err) 6033 goto restore_opts; 6034 } 6035 sbi->s_mount_state = le16_to_cpu(es->s_state); 6036 6037 err = ext4_setup_super(sb, es, 0); 6038 if (err) 6039 goto restore_opts; 6040 6041 sb->s_flags &= ~SB_RDONLY; 6042 if (ext4_has_feature_mmp(sb)) 6043 if (ext4_multi_mount_protect(sb, 6044 le64_to_cpu(es->s_mmp_block))) { 6045 err = -EROFS; 6046 goto restore_opts; 6047 } 6048 enable_quota = 1; 6049 } 6050 } 6051 6052 /* 6053 * Reinitialize lazy itable initialization thread based on 6054 * current settings 6055 */ 6056 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6057 ext4_unregister_li_request(sb); 6058 else { 6059 ext4_group_t first_not_zeroed; 6060 first_not_zeroed = ext4_has_uninit_itable(sb); 6061 ext4_register_li_request(sb, first_not_zeroed); 6062 } 6063 6064 /* 6065 * Handle creation of system zone data early because it can fail. 6066 * Releasing of existing data is done when we are sure remount will 6067 * succeed. 6068 */ 6069 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6070 err = ext4_setup_system_zone(sb); 6071 if (err) 6072 goto restore_opts; 6073 } 6074 6075 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6076 err = ext4_commit_super(sb); 6077 if (err) 6078 goto restore_opts; 6079 } 6080 6081 #ifdef CONFIG_QUOTA 6082 /* Release old quota file names */ 6083 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6084 kfree(old_opts.s_qf_names[i]); 6085 if (enable_quota) { 6086 if (sb_any_quota_suspended(sb)) 6087 dquot_resume(sb, -1); 6088 else if (ext4_has_feature_quota(sb)) { 6089 err = ext4_enable_quotas(sb); 6090 if (err) 6091 goto restore_opts; 6092 } 6093 } 6094 #endif 6095 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6096 ext4_release_system_zone(sb); 6097 6098 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6099 ext4_stop_mmpd(sbi); 6100 6101 /* 6102 * Some options can be enabled by ext4 and/or by VFS mount flag 6103 * either way we need to make sure it matches in both *flags and 6104 * s_flags. Copy those selected flags from s_flags to *flags 6105 */ 6106 *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags); 6107 6108 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s. Quota mode: %s.", 6109 orig_data, ext4_quota_mode(sb)); 6110 kfree(orig_data); 6111 return 0; 6112 6113 restore_opts: 6114 sb->s_flags = old_sb_flags; 6115 sbi->s_mount_opt = old_opts.s_mount_opt; 6116 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6117 sbi->s_resuid = old_opts.s_resuid; 6118 sbi->s_resgid = old_opts.s_resgid; 6119 sbi->s_commit_interval = old_opts.s_commit_interval; 6120 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6121 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6122 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6123 ext4_release_system_zone(sb); 6124 #ifdef CONFIG_QUOTA 6125 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6126 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6127 to_free[i] = get_qf_name(sb, sbi, i); 6128 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6129 } 6130 synchronize_rcu(); 6131 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6132 kfree(to_free[i]); 6133 #endif 6134 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6135 ext4_stop_mmpd(sbi); 6136 kfree(orig_data); 6137 return err; 6138 } 6139 6140 #ifdef CONFIG_QUOTA 6141 static int ext4_statfs_project(struct super_block *sb, 6142 kprojid_t projid, struct kstatfs *buf) 6143 { 6144 struct kqid qid; 6145 struct dquot *dquot; 6146 u64 limit; 6147 u64 curblock; 6148 6149 qid = make_kqid_projid(projid); 6150 dquot = dqget(sb, qid); 6151 if (IS_ERR(dquot)) 6152 return PTR_ERR(dquot); 6153 spin_lock(&dquot->dq_dqb_lock); 6154 6155 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6156 dquot->dq_dqb.dqb_bhardlimit); 6157 limit >>= sb->s_blocksize_bits; 6158 6159 if (limit && buf->f_blocks > limit) { 6160 curblock = (dquot->dq_dqb.dqb_curspace + 6161 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6162 buf->f_blocks = limit; 6163 buf->f_bfree = buf->f_bavail = 6164 (buf->f_blocks > curblock) ? 6165 (buf->f_blocks - curblock) : 0; 6166 } 6167 6168 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6169 dquot->dq_dqb.dqb_ihardlimit); 6170 if (limit && buf->f_files > limit) { 6171 buf->f_files = limit; 6172 buf->f_ffree = 6173 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6174 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6175 } 6176 6177 spin_unlock(&dquot->dq_dqb_lock); 6178 dqput(dquot); 6179 return 0; 6180 } 6181 #endif 6182 6183 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6184 { 6185 struct super_block *sb = dentry->d_sb; 6186 struct ext4_sb_info *sbi = EXT4_SB(sb); 6187 struct ext4_super_block *es = sbi->s_es; 6188 ext4_fsblk_t overhead = 0, resv_blocks; 6189 s64 bfree; 6190 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6191 6192 if (!test_opt(sb, MINIX_DF)) 6193 overhead = sbi->s_overhead; 6194 6195 buf->f_type = EXT4_SUPER_MAGIC; 6196 buf->f_bsize = sb->s_blocksize; 6197 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6198 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6199 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6200 /* prevent underflow in case that few free space is available */ 6201 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6202 buf->f_bavail = buf->f_bfree - 6203 (ext4_r_blocks_count(es) + resv_blocks); 6204 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6205 buf->f_bavail = 0; 6206 buf->f_files = le32_to_cpu(es->s_inodes_count); 6207 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6208 buf->f_namelen = EXT4_NAME_LEN; 6209 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6210 6211 #ifdef CONFIG_QUOTA 6212 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6213 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6214 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6215 #endif 6216 return 0; 6217 } 6218 6219 6220 #ifdef CONFIG_QUOTA 6221 6222 /* 6223 * Helper functions so that transaction is started before we acquire dqio_sem 6224 * to keep correct lock ordering of transaction > dqio_sem 6225 */ 6226 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6227 { 6228 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6229 } 6230 6231 static int ext4_write_dquot(struct dquot *dquot) 6232 { 6233 int ret, err; 6234 handle_t *handle; 6235 struct inode *inode; 6236 6237 inode = dquot_to_inode(dquot); 6238 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6239 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6240 if (IS_ERR(handle)) 6241 return PTR_ERR(handle); 6242 ret = dquot_commit(dquot); 6243 err = ext4_journal_stop(handle); 6244 if (!ret) 6245 ret = err; 6246 return ret; 6247 } 6248 6249 static int ext4_acquire_dquot(struct dquot *dquot) 6250 { 6251 int ret, err; 6252 handle_t *handle; 6253 6254 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6255 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6256 if (IS_ERR(handle)) 6257 return PTR_ERR(handle); 6258 ret = dquot_acquire(dquot); 6259 err = ext4_journal_stop(handle); 6260 if (!ret) 6261 ret = err; 6262 return ret; 6263 } 6264 6265 static int ext4_release_dquot(struct dquot *dquot) 6266 { 6267 int ret, err; 6268 handle_t *handle; 6269 6270 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6271 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6272 if (IS_ERR(handle)) { 6273 /* Release dquot anyway to avoid endless cycle in dqput() */ 6274 dquot_release(dquot); 6275 return PTR_ERR(handle); 6276 } 6277 ret = dquot_release(dquot); 6278 err = ext4_journal_stop(handle); 6279 if (!ret) 6280 ret = err; 6281 return ret; 6282 } 6283 6284 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6285 { 6286 struct super_block *sb = dquot->dq_sb; 6287 6288 if (ext4_is_quota_journalled(sb)) { 6289 dquot_mark_dquot_dirty(dquot); 6290 return ext4_write_dquot(dquot); 6291 } else { 6292 return dquot_mark_dquot_dirty(dquot); 6293 } 6294 } 6295 6296 static int ext4_write_info(struct super_block *sb, int type) 6297 { 6298 int ret, err; 6299 handle_t *handle; 6300 6301 /* Data block + inode block */ 6302 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 6303 if (IS_ERR(handle)) 6304 return PTR_ERR(handle); 6305 ret = dquot_commit_info(sb, type); 6306 err = ext4_journal_stop(handle); 6307 if (!ret) 6308 ret = err; 6309 return ret; 6310 } 6311 6312 /* 6313 * Turn on quotas during mount time - we need to find 6314 * the quota file and such... 6315 */ 6316 static int ext4_quota_on_mount(struct super_block *sb, int type) 6317 { 6318 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 6319 EXT4_SB(sb)->s_jquota_fmt, type); 6320 } 6321 6322 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6323 { 6324 struct ext4_inode_info *ei = EXT4_I(inode); 6325 6326 /* The first argument of lockdep_set_subclass has to be 6327 * *exactly* the same as the argument to init_rwsem() --- in 6328 * this case, in init_once() --- or lockdep gets unhappy 6329 * because the name of the lock is set using the 6330 * stringification of the argument to init_rwsem(). 6331 */ 6332 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6333 lockdep_set_subclass(&ei->i_data_sem, subclass); 6334 } 6335 6336 /* 6337 * Standard function to be called on quota_on 6338 */ 6339 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6340 const struct path *path) 6341 { 6342 int err; 6343 6344 if (!test_opt(sb, QUOTA)) 6345 return -EINVAL; 6346 6347 /* Quotafile not on the same filesystem? */ 6348 if (path->dentry->d_sb != sb) 6349 return -EXDEV; 6350 6351 /* Quota already enabled for this file? */ 6352 if (IS_NOQUOTA(d_inode(path->dentry))) 6353 return -EBUSY; 6354 6355 /* Journaling quota? */ 6356 if (EXT4_SB(sb)->s_qf_names[type]) { 6357 /* Quotafile not in fs root? */ 6358 if (path->dentry->d_parent != sb->s_root) 6359 ext4_msg(sb, KERN_WARNING, 6360 "Quota file not on filesystem root. " 6361 "Journaled quota will not work"); 6362 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6363 } else { 6364 /* 6365 * Clear the flag just in case mount options changed since 6366 * last time. 6367 */ 6368 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6369 } 6370 6371 /* 6372 * When we journal data on quota file, we have to flush journal to see 6373 * all updates to the file when we bypass pagecache... 6374 */ 6375 if (EXT4_SB(sb)->s_journal && 6376 ext4_should_journal_data(d_inode(path->dentry))) { 6377 /* 6378 * We don't need to lock updates but journal_flush() could 6379 * otherwise be livelocked... 6380 */ 6381 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6382 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0); 6383 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6384 if (err) 6385 return err; 6386 } 6387 6388 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6389 err = dquot_quota_on(sb, type, format_id, path); 6390 if (err) { 6391 lockdep_set_quota_inode(path->dentry->d_inode, 6392 I_DATA_SEM_NORMAL); 6393 } else { 6394 struct inode *inode = d_inode(path->dentry); 6395 handle_t *handle; 6396 6397 /* 6398 * Set inode flags to prevent userspace from messing with quota 6399 * files. If this fails, we return success anyway since quotas 6400 * are already enabled and this is not a hard failure. 6401 */ 6402 inode_lock(inode); 6403 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6404 if (IS_ERR(handle)) 6405 goto unlock_inode; 6406 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6407 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6408 S_NOATIME | S_IMMUTABLE); 6409 err = ext4_mark_inode_dirty(handle, inode); 6410 ext4_journal_stop(handle); 6411 unlock_inode: 6412 inode_unlock(inode); 6413 } 6414 return err; 6415 } 6416 6417 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6418 unsigned int flags) 6419 { 6420 int err; 6421 struct inode *qf_inode; 6422 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6423 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6424 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6425 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6426 }; 6427 6428 BUG_ON(!ext4_has_feature_quota(sb)); 6429 6430 if (!qf_inums[type]) 6431 return -EPERM; 6432 6433 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6434 if (IS_ERR(qf_inode)) { 6435 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6436 return PTR_ERR(qf_inode); 6437 } 6438 6439 /* Don't account quota for quota files to avoid recursion */ 6440 qf_inode->i_flags |= S_NOQUOTA; 6441 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6442 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6443 if (err) 6444 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6445 iput(qf_inode); 6446 6447 return err; 6448 } 6449 6450 /* Enable usage tracking for all quota types. */ 6451 static int ext4_enable_quotas(struct super_block *sb) 6452 { 6453 int type, err = 0; 6454 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6455 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6456 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6457 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6458 }; 6459 bool quota_mopt[EXT4_MAXQUOTAS] = { 6460 test_opt(sb, USRQUOTA), 6461 test_opt(sb, GRPQUOTA), 6462 test_opt(sb, PRJQUOTA), 6463 }; 6464 6465 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6466 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6467 if (qf_inums[type]) { 6468 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6469 DQUOT_USAGE_ENABLED | 6470 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6471 if (err) { 6472 ext4_warning(sb, 6473 "Failed to enable quota tracking " 6474 "(type=%d, err=%d). Please run " 6475 "e2fsck to fix.", type, err); 6476 for (type--; type >= 0; type--) 6477 dquot_quota_off(sb, type); 6478 6479 return err; 6480 } 6481 } 6482 } 6483 return 0; 6484 } 6485 6486 static int ext4_quota_off(struct super_block *sb, int type) 6487 { 6488 struct inode *inode = sb_dqopt(sb)->files[type]; 6489 handle_t *handle; 6490 int err; 6491 6492 /* Force all delayed allocation blocks to be allocated. 6493 * Caller already holds s_umount sem */ 6494 if (test_opt(sb, DELALLOC)) 6495 sync_filesystem(sb); 6496 6497 if (!inode || !igrab(inode)) 6498 goto out; 6499 6500 err = dquot_quota_off(sb, type); 6501 if (err || ext4_has_feature_quota(sb)) 6502 goto out_put; 6503 6504 inode_lock(inode); 6505 /* 6506 * Update modification times of quota files when userspace can 6507 * start looking at them. If we fail, we return success anyway since 6508 * this is not a hard failure and quotas are already disabled. 6509 */ 6510 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6511 if (IS_ERR(handle)) { 6512 err = PTR_ERR(handle); 6513 goto out_unlock; 6514 } 6515 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 6516 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 6517 inode->i_mtime = inode->i_ctime = current_time(inode); 6518 err = ext4_mark_inode_dirty(handle, inode); 6519 ext4_journal_stop(handle); 6520 out_unlock: 6521 inode_unlock(inode); 6522 out_put: 6523 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 6524 iput(inode); 6525 return err; 6526 out: 6527 return dquot_quota_off(sb, type); 6528 } 6529 6530 /* Read data from quotafile - avoid pagecache and such because we cannot afford 6531 * acquiring the locks... As quota files are never truncated and quota code 6532 * itself serializes the operations (and no one else should touch the files) 6533 * we don't have to be afraid of races */ 6534 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 6535 size_t len, loff_t off) 6536 { 6537 struct inode *inode = sb_dqopt(sb)->files[type]; 6538 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6539 int offset = off & (sb->s_blocksize - 1); 6540 int tocopy; 6541 size_t toread; 6542 struct buffer_head *bh; 6543 loff_t i_size = i_size_read(inode); 6544 6545 if (off > i_size) 6546 return 0; 6547 if (off+len > i_size) 6548 len = i_size-off; 6549 toread = len; 6550 while (toread > 0) { 6551 tocopy = sb->s_blocksize - offset < toread ? 6552 sb->s_blocksize - offset : toread; 6553 bh = ext4_bread(NULL, inode, blk, 0); 6554 if (IS_ERR(bh)) 6555 return PTR_ERR(bh); 6556 if (!bh) /* A hole? */ 6557 memset(data, 0, tocopy); 6558 else 6559 memcpy(data, bh->b_data+offset, tocopy); 6560 brelse(bh); 6561 offset = 0; 6562 toread -= tocopy; 6563 data += tocopy; 6564 blk++; 6565 } 6566 return len; 6567 } 6568 6569 /* Write to quotafile (we know the transaction is already started and has 6570 * enough credits) */ 6571 static ssize_t ext4_quota_write(struct super_block *sb, int type, 6572 const char *data, size_t len, loff_t off) 6573 { 6574 struct inode *inode = sb_dqopt(sb)->files[type]; 6575 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6576 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 6577 int retries = 0; 6578 struct buffer_head *bh; 6579 handle_t *handle = journal_current_handle(); 6580 6581 if (EXT4_SB(sb)->s_journal && !handle) { 6582 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6583 " cancelled because transaction is not started", 6584 (unsigned long long)off, (unsigned long long)len); 6585 return -EIO; 6586 } 6587 /* 6588 * Since we account only one data block in transaction credits, 6589 * then it is impossible to cross a block boundary. 6590 */ 6591 if (sb->s_blocksize - offset < len) { 6592 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6593 " cancelled because not block aligned", 6594 (unsigned long long)off, (unsigned long long)len); 6595 return -EIO; 6596 } 6597 6598 do { 6599 bh = ext4_bread(handle, inode, blk, 6600 EXT4_GET_BLOCKS_CREATE | 6601 EXT4_GET_BLOCKS_METADATA_NOFAIL); 6602 } while (PTR_ERR(bh) == -ENOSPC && 6603 ext4_should_retry_alloc(inode->i_sb, &retries)); 6604 if (IS_ERR(bh)) 6605 return PTR_ERR(bh); 6606 if (!bh) 6607 goto out; 6608 BUFFER_TRACE(bh, "get write access"); 6609 err = ext4_journal_get_write_access(handle, bh); 6610 if (err) { 6611 brelse(bh); 6612 return err; 6613 } 6614 lock_buffer(bh); 6615 memcpy(bh->b_data+offset, data, len); 6616 flush_dcache_page(bh->b_page); 6617 unlock_buffer(bh); 6618 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6619 brelse(bh); 6620 out: 6621 if (inode->i_size < off + len) { 6622 i_size_write(inode, off + len); 6623 EXT4_I(inode)->i_disksize = inode->i_size; 6624 err2 = ext4_mark_inode_dirty(handle, inode); 6625 if (unlikely(err2 && !err)) 6626 err = err2; 6627 } 6628 return err ? err : len; 6629 } 6630 #endif 6631 6632 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6633 const char *dev_name, void *data) 6634 { 6635 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6636 } 6637 6638 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6639 static inline void register_as_ext2(void) 6640 { 6641 int err = register_filesystem(&ext2_fs_type); 6642 if (err) 6643 printk(KERN_WARNING 6644 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6645 } 6646 6647 static inline void unregister_as_ext2(void) 6648 { 6649 unregister_filesystem(&ext2_fs_type); 6650 } 6651 6652 static inline int ext2_feature_set_ok(struct super_block *sb) 6653 { 6654 if (ext4_has_unknown_ext2_incompat_features(sb)) 6655 return 0; 6656 if (sb_rdonly(sb)) 6657 return 1; 6658 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6659 return 0; 6660 return 1; 6661 } 6662 #else 6663 static inline void register_as_ext2(void) { } 6664 static inline void unregister_as_ext2(void) { } 6665 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6666 #endif 6667 6668 static inline void register_as_ext3(void) 6669 { 6670 int err = register_filesystem(&ext3_fs_type); 6671 if (err) 6672 printk(KERN_WARNING 6673 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6674 } 6675 6676 static inline void unregister_as_ext3(void) 6677 { 6678 unregister_filesystem(&ext3_fs_type); 6679 } 6680 6681 static inline int ext3_feature_set_ok(struct super_block *sb) 6682 { 6683 if (ext4_has_unknown_ext3_incompat_features(sb)) 6684 return 0; 6685 if (!ext4_has_feature_journal(sb)) 6686 return 0; 6687 if (sb_rdonly(sb)) 6688 return 1; 6689 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6690 return 0; 6691 return 1; 6692 } 6693 6694 static struct file_system_type ext4_fs_type = { 6695 .owner = THIS_MODULE, 6696 .name = "ext4", 6697 .mount = ext4_mount, 6698 .kill_sb = kill_block_super, 6699 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 6700 }; 6701 MODULE_ALIAS_FS("ext4"); 6702 6703 /* Shared across all ext4 file systems */ 6704 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6705 6706 static int __init ext4_init_fs(void) 6707 { 6708 int i, err; 6709 6710 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6711 ext4_li_info = NULL; 6712 6713 /* Build-time check for flags consistency */ 6714 ext4_check_flag_values(); 6715 6716 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6717 init_waitqueue_head(&ext4__ioend_wq[i]); 6718 6719 err = ext4_init_es(); 6720 if (err) 6721 return err; 6722 6723 err = ext4_init_pending(); 6724 if (err) 6725 goto out7; 6726 6727 err = ext4_init_post_read_processing(); 6728 if (err) 6729 goto out6; 6730 6731 err = ext4_init_pageio(); 6732 if (err) 6733 goto out5; 6734 6735 err = ext4_init_system_zone(); 6736 if (err) 6737 goto out4; 6738 6739 err = ext4_init_sysfs(); 6740 if (err) 6741 goto out3; 6742 6743 err = ext4_init_mballoc(); 6744 if (err) 6745 goto out2; 6746 err = init_inodecache(); 6747 if (err) 6748 goto out1; 6749 6750 err = ext4_fc_init_dentry_cache(); 6751 if (err) 6752 goto out05; 6753 6754 register_as_ext3(); 6755 register_as_ext2(); 6756 err = register_filesystem(&ext4_fs_type); 6757 if (err) 6758 goto out; 6759 6760 return 0; 6761 out: 6762 unregister_as_ext2(); 6763 unregister_as_ext3(); 6764 out05: 6765 destroy_inodecache(); 6766 out1: 6767 ext4_exit_mballoc(); 6768 out2: 6769 ext4_exit_sysfs(); 6770 out3: 6771 ext4_exit_system_zone(); 6772 out4: 6773 ext4_exit_pageio(); 6774 out5: 6775 ext4_exit_post_read_processing(); 6776 out6: 6777 ext4_exit_pending(); 6778 out7: 6779 ext4_exit_es(); 6780 6781 return err; 6782 } 6783 6784 static void __exit ext4_exit_fs(void) 6785 { 6786 ext4_destroy_lazyinit_thread(); 6787 unregister_as_ext2(); 6788 unregister_as_ext3(); 6789 unregister_filesystem(&ext4_fs_type); 6790 destroy_inodecache(); 6791 ext4_exit_mballoc(); 6792 ext4_exit_sysfs(); 6793 ext4_exit_system_zone(); 6794 ext4_exit_pageio(); 6795 ext4_exit_post_read_processing(); 6796 ext4_exit_es(); 6797 ext4_exit_pending(); 6798 } 6799 6800 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6801 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6802 MODULE_LICENSE("GPL"); 6803 MODULE_SOFTDEP("pre: crc32c"); 6804 module_init(ext4_init_fs) 6805 module_exit(ext4_exit_fs) 6806