1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/super.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Big-endian to little-endian byte-swapping/bitmaps by 17 * David S. Miller (davem@caip.rutgers.edu), 1995 18 */ 19 20 #include <linux/module.h> 21 #include <linux/string.h> 22 #include <linux/fs.h> 23 #include <linux/time.h> 24 #include <linux/vmalloc.h> 25 #include <linux/slab.h> 26 #include <linux/init.h> 27 #include <linux/blkdev.h> 28 #include <linux/backing-dev.h> 29 #include <linux/parser.h> 30 #include <linux/buffer_head.h> 31 #include <linux/exportfs.h> 32 #include <linux/vfs.h> 33 #include <linux/random.h> 34 #include <linux/mount.h> 35 #include <linux/namei.h> 36 #include <linux/quotaops.h> 37 #include <linux/seq_file.h> 38 #include <linux/ctype.h> 39 #include <linux/log2.h> 40 #include <linux/crc16.h> 41 #include <linux/dax.h> 42 #include <linux/cleancache.h> 43 #include <linux/uaccess.h> 44 #include <linux/iversion.h> 45 #include <linux/unicode.h> 46 #include <linux/part_stat.h> 47 #include <linux/kthread.h> 48 #include <linux/freezer.h> 49 50 #include "ext4.h" 51 #include "ext4_extents.h" /* Needed for trace points definition */ 52 #include "ext4_jbd2.h" 53 #include "xattr.h" 54 #include "acl.h" 55 #include "mballoc.h" 56 #include "fsmap.h" 57 58 #define CREATE_TRACE_POINTS 59 #include <trace/events/ext4.h> 60 61 static struct ext4_lazy_init *ext4_li_info; 62 static DEFINE_MUTEX(ext4_li_mtx); 63 static struct ratelimit_state ext4_mount_msg_ratelimit; 64 65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 66 unsigned long journal_devnum); 67 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 68 static void ext4_update_super(struct super_block *sb); 69 static int ext4_commit_super(struct super_block *sb); 70 static int ext4_mark_recovery_complete(struct super_block *sb, 71 struct ext4_super_block *es); 72 static int ext4_clear_journal_err(struct super_block *sb, 73 struct ext4_super_block *es); 74 static int ext4_sync_fs(struct super_block *sb, int wait); 75 static int ext4_remount(struct super_block *sb, int *flags, char *data); 76 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 77 static int ext4_unfreeze(struct super_block *sb); 78 static int ext4_freeze(struct super_block *sb); 79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 80 const char *dev_name, void *data); 81 static inline int ext2_feature_set_ok(struct super_block *sb); 82 static inline int ext3_feature_set_ok(struct super_block *sb); 83 static int ext4_feature_set_ok(struct super_block *sb, int readonly); 84 static void ext4_destroy_lazyinit_thread(void); 85 static void ext4_unregister_li_request(struct super_block *sb); 86 static void ext4_clear_request_list(void); 87 static struct inode *ext4_get_journal_inode(struct super_block *sb, 88 unsigned int journal_inum); 89 90 /* 91 * Lock ordering 92 * 93 * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and 94 * i_mmap_rwsem (inode->i_mmap_rwsem)! 95 * 96 * page fault path: 97 * mmap_lock -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start -> 98 * page lock -> i_data_sem (rw) 99 * 100 * buffered write path: 101 * sb_start_write -> i_mutex -> mmap_lock 102 * sb_start_write -> i_mutex -> transaction start -> page lock -> 103 * i_data_sem (rw) 104 * 105 * truncate: 106 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock 107 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start -> 108 * i_data_sem (rw) 109 * 110 * direct IO: 111 * sb_start_write -> i_mutex -> mmap_lock 112 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 113 * 114 * writepages: 115 * transaction start -> page lock(s) -> i_data_sem (rw) 116 */ 117 118 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 119 static struct file_system_type ext2_fs_type = { 120 .owner = THIS_MODULE, 121 .name = "ext2", 122 .mount = ext4_mount, 123 .kill_sb = kill_block_super, 124 .fs_flags = FS_REQUIRES_DEV, 125 }; 126 MODULE_ALIAS_FS("ext2"); 127 MODULE_ALIAS("ext2"); 128 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type) 129 #else 130 #define IS_EXT2_SB(sb) (0) 131 #endif 132 133 134 static struct file_system_type ext3_fs_type = { 135 .owner = THIS_MODULE, 136 .name = "ext3", 137 .mount = ext4_mount, 138 .kill_sb = kill_block_super, 139 .fs_flags = FS_REQUIRES_DEV, 140 }; 141 MODULE_ALIAS_FS("ext3"); 142 MODULE_ALIAS("ext3"); 143 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type) 144 145 146 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags, 147 bh_end_io_t *end_io) 148 { 149 /* 150 * buffer's verified bit is no longer valid after reading from 151 * disk again due to write out error, clear it to make sure we 152 * recheck the buffer contents. 153 */ 154 clear_buffer_verified(bh); 155 156 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 157 get_bh(bh); 158 submit_bh(REQ_OP_READ, op_flags, bh); 159 } 160 161 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags, 162 bh_end_io_t *end_io) 163 { 164 BUG_ON(!buffer_locked(bh)); 165 166 if (ext4_buffer_uptodate(bh)) { 167 unlock_buffer(bh); 168 return; 169 } 170 __ext4_read_bh(bh, op_flags, end_io); 171 } 172 173 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io) 174 { 175 BUG_ON(!buffer_locked(bh)); 176 177 if (ext4_buffer_uptodate(bh)) { 178 unlock_buffer(bh); 179 return 0; 180 } 181 182 __ext4_read_bh(bh, op_flags, end_io); 183 184 wait_on_buffer(bh); 185 if (buffer_uptodate(bh)) 186 return 0; 187 return -EIO; 188 } 189 190 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait) 191 { 192 if (trylock_buffer(bh)) { 193 if (wait) 194 return ext4_read_bh(bh, op_flags, NULL); 195 ext4_read_bh_nowait(bh, op_flags, NULL); 196 return 0; 197 } 198 if (wait) { 199 wait_on_buffer(bh); 200 if (buffer_uptodate(bh)) 201 return 0; 202 return -EIO; 203 } 204 return 0; 205 } 206 207 /* 208 * This works like __bread_gfp() except it uses ERR_PTR for error 209 * returns. Currently with sb_bread it's impossible to distinguish 210 * between ENOMEM and EIO situations (since both result in a NULL 211 * return. 212 */ 213 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 214 sector_t block, int op_flags, 215 gfp_t gfp) 216 { 217 struct buffer_head *bh; 218 int ret; 219 220 bh = sb_getblk_gfp(sb, block, gfp); 221 if (bh == NULL) 222 return ERR_PTR(-ENOMEM); 223 if (ext4_buffer_uptodate(bh)) 224 return bh; 225 226 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 227 if (ret) { 228 put_bh(bh); 229 return ERR_PTR(ret); 230 } 231 return bh; 232 } 233 234 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 235 int op_flags) 236 { 237 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 238 } 239 240 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 241 sector_t block) 242 { 243 return __ext4_sb_bread_gfp(sb, block, 0, 0); 244 } 245 246 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 247 { 248 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 249 250 if (likely(bh)) { 251 ext4_read_bh_lock(bh, REQ_RAHEAD, false); 252 brelse(bh); 253 } 254 } 255 256 static int ext4_verify_csum_type(struct super_block *sb, 257 struct ext4_super_block *es) 258 { 259 if (!ext4_has_feature_metadata_csum(sb)) 260 return 1; 261 262 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 263 } 264 265 static __le32 ext4_superblock_csum(struct super_block *sb, 266 struct ext4_super_block *es) 267 { 268 struct ext4_sb_info *sbi = EXT4_SB(sb); 269 int offset = offsetof(struct ext4_super_block, s_checksum); 270 __u32 csum; 271 272 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 273 274 return cpu_to_le32(csum); 275 } 276 277 static int ext4_superblock_csum_verify(struct super_block *sb, 278 struct ext4_super_block *es) 279 { 280 if (!ext4_has_metadata_csum(sb)) 281 return 1; 282 283 return es->s_checksum == ext4_superblock_csum(sb, es); 284 } 285 286 void ext4_superblock_csum_set(struct super_block *sb) 287 { 288 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 289 290 if (!ext4_has_metadata_csum(sb)) 291 return; 292 293 es->s_checksum = ext4_superblock_csum(sb, es); 294 } 295 296 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 297 struct ext4_group_desc *bg) 298 { 299 return le32_to_cpu(bg->bg_block_bitmap_lo) | 300 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 301 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 302 } 303 304 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 305 struct ext4_group_desc *bg) 306 { 307 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 308 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 309 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 310 } 311 312 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 313 struct ext4_group_desc *bg) 314 { 315 return le32_to_cpu(bg->bg_inode_table_lo) | 316 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 317 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 318 } 319 320 __u32 ext4_free_group_clusters(struct super_block *sb, 321 struct ext4_group_desc *bg) 322 { 323 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 324 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 325 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 326 } 327 328 __u32 ext4_free_inodes_count(struct super_block *sb, 329 struct ext4_group_desc *bg) 330 { 331 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 332 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 333 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 334 } 335 336 __u32 ext4_used_dirs_count(struct super_block *sb, 337 struct ext4_group_desc *bg) 338 { 339 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 340 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 341 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 342 } 343 344 __u32 ext4_itable_unused_count(struct super_block *sb, 345 struct ext4_group_desc *bg) 346 { 347 return le16_to_cpu(bg->bg_itable_unused_lo) | 348 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 349 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 350 } 351 352 void ext4_block_bitmap_set(struct super_block *sb, 353 struct ext4_group_desc *bg, ext4_fsblk_t blk) 354 { 355 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 356 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 357 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 358 } 359 360 void ext4_inode_bitmap_set(struct super_block *sb, 361 struct ext4_group_desc *bg, ext4_fsblk_t blk) 362 { 363 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 364 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 365 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 366 } 367 368 void ext4_inode_table_set(struct super_block *sb, 369 struct ext4_group_desc *bg, ext4_fsblk_t blk) 370 { 371 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 372 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 373 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 374 } 375 376 void ext4_free_group_clusters_set(struct super_block *sb, 377 struct ext4_group_desc *bg, __u32 count) 378 { 379 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 380 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 381 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 382 } 383 384 void ext4_free_inodes_set(struct super_block *sb, 385 struct ext4_group_desc *bg, __u32 count) 386 { 387 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 388 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 389 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 390 } 391 392 void ext4_used_dirs_set(struct super_block *sb, 393 struct ext4_group_desc *bg, __u32 count) 394 { 395 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 396 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 397 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 398 } 399 400 void ext4_itable_unused_set(struct super_block *sb, 401 struct ext4_group_desc *bg, __u32 count) 402 { 403 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 404 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 405 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 406 } 407 408 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 409 { 410 now = clamp_val(now, 0, (1ull << 40) - 1); 411 412 *lo = cpu_to_le32(lower_32_bits(now)); 413 *hi = upper_32_bits(now); 414 } 415 416 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 417 { 418 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 419 } 420 #define ext4_update_tstamp(es, tstamp) \ 421 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 422 ktime_get_real_seconds()) 423 #define ext4_get_tstamp(es, tstamp) \ 424 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 425 426 /* 427 * The del_gendisk() function uninitializes the disk-specific data 428 * structures, including the bdi structure, without telling anyone 429 * else. Once this happens, any attempt to call mark_buffer_dirty() 430 * (for example, by ext4_commit_super), will cause a kernel OOPS. 431 * This is a kludge to prevent these oops until we can put in a proper 432 * hook in del_gendisk() to inform the VFS and file system layers. 433 */ 434 static int block_device_ejected(struct super_block *sb) 435 { 436 struct inode *bd_inode = sb->s_bdev->bd_inode; 437 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 438 439 return bdi->dev == NULL; 440 } 441 442 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 443 { 444 struct super_block *sb = journal->j_private; 445 struct ext4_sb_info *sbi = EXT4_SB(sb); 446 int error = is_journal_aborted(journal); 447 struct ext4_journal_cb_entry *jce; 448 449 BUG_ON(txn->t_state == T_FINISHED); 450 451 ext4_process_freed_data(sb, txn->t_tid); 452 453 spin_lock(&sbi->s_md_lock); 454 while (!list_empty(&txn->t_private_list)) { 455 jce = list_entry(txn->t_private_list.next, 456 struct ext4_journal_cb_entry, jce_list); 457 list_del_init(&jce->jce_list); 458 spin_unlock(&sbi->s_md_lock); 459 jce->jce_func(sb, jce, error); 460 spin_lock(&sbi->s_md_lock); 461 } 462 spin_unlock(&sbi->s_md_lock); 463 } 464 465 /* 466 * This writepage callback for write_cache_pages() 467 * takes care of a few cases after page cleaning. 468 * 469 * write_cache_pages() already checks for dirty pages 470 * and calls clear_page_dirty_for_io(), which we want, 471 * to write protect the pages. 472 * 473 * However, we may have to redirty a page (see below.) 474 */ 475 static int ext4_journalled_writepage_callback(struct page *page, 476 struct writeback_control *wbc, 477 void *data) 478 { 479 transaction_t *transaction = (transaction_t *) data; 480 struct buffer_head *bh, *head; 481 struct journal_head *jh; 482 483 bh = head = page_buffers(page); 484 do { 485 /* 486 * We have to redirty a page in these cases: 487 * 1) If buffer is dirty, it means the page was dirty because it 488 * contains a buffer that needs checkpointing. So the dirty bit 489 * needs to be preserved so that checkpointing writes the buffer 490 * properly. 491 * 2) If buffer is not part of the committing transaction 492 * (we may have just accidentally come across this buffer because 493 * inode range tracking is not exact) or if the currently running 494 * transaction already contains this buffer as well, dirty bit 495 * needs to be preserved so that the buffer gets writeprotected 496 * properly on running transaction's commit. 497 */ 498 jh = bh2jh(bh); 499 if (buffer_dirty(bh) || 500 (jh && (jh->b_transaction != transaction || 501 jh->b_next_transaction))) { 502 redirty_page_for_writepage(wbc, page); 503 goto out; 504 } 505 } while ((bh = bh->b_this_page) != head); 506 507 out: 508 return AOP_WRITEPAGE_ACTIVATE; 509 } 510 511 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 512 { 513 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 514 struct writeback_control wbc = { 515 .sync_mode = WB_SYNC_ALL, 516 .nr_to_write = LONG_MAX, 517 .range_start = jinode->i_dirty_start, 518 .range_end = jinode->i_dirty_end, 519 }; 520 521 return write_cache_pages(mapping, &wbc, 522 ext4_journalled_writepage_callback, 523 jinode->i_transaction); 524 } 525 526 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 527 { 528 int ret; 529 530 if (ext4_should_journal_data(jinode->i_vfs_inode)) 531 ret = ext4_journalled_submit_inode_data_buffers(jinode); 532 else 533 ret = jbd2_journal_submit_inode_data_buffers(jinode); 534 535 return ret; 536 } 537 538 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 539 { 540 int ret = 0; 541 542 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 543 ret = jbd2_journal_finish_inode_data_buffers(jinode); 544 545 return ret; 546 } 547 548 static bool system_going_down(void) 549 { 550 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 551 || system_state == SYSTEM_RESTART; 552 } 553 554 struct ext4_err_translation { 555 int code; 556 int errno; 557 }; 558 559 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 560 561 static struct ext4_err_translation err_translation[] = { 562 EXT4_ERR_TRANSLATE(EIO), 563 EXT4_ERR_TRANSLATE(ENOMEM), 564 EXT4_ERR_TRANSLATE(EFSBADCRC), 565 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 566 EXT4_ERR_TRANSLATE(ENOSPC), 567 EXT4_ERR_TRANSLATE(ENOKEY), 568 EXT4_ERR_TRANSLATE(EROFS), 569 EXT4_ERR_TRANSLATE(EFBIG), 570 EXT4_ERR_TRANSLATE(EEXIST), 571 EXT4_ERR_TRANSLATE(ERANGE), 572 EXT4_ERR_TRANSLATE(EOVERFLOW), 573 EXT4_ERR_TRANSLATE(EBUSY), 574 EXT4_ERR_TRANSLATE(ENOTDIR), 575 EXT4_ERR_TRANSLATE(ENOTEMPTY), 576 EXT4_ERR_TRANSLATE(ESHUTDOWN), 577 EXT4_ERR_TRANSLATE(EFAULT), 578 }; 579 580 static int ext4_errno_to_code(int errno) 581 { 582 int i; 583 584 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 585 if (err_translation[i].errno == errno) 586 return err_translation[i].code; 587 return EXT4_ERR_UNKNOWN; 588 } 589 590 static void save_error_info(struct super_block *sb, int error, 591 __u32 ino, __u64 block, 592 const char *func, unsigned int line) 593 { 594 struct ext4_sb_info *sbi = EXT4_SB(sb); 595 596 /* We default to EFSCORRUPTED error... */ 597 if (error == 0) 598 error = EFSCORRUPTED; 599 600 spin_lock(&sbi->s_error_lock); 601 sbi->s_add_error_count++; 602 sbi->s_last_error_code = error; 603 sbi->s_last_error_line = line; 604 sbi->s_last_error_ino = ino; 605 sbi->s_last_error_block = block; 606 sbi->s_last_error_func = func; 607 sbi->s_last_error_time = ktime_get_real_seconds(); 608 if (!sbi->s_first_error_time) { 609 sbi->s_first_error_code = error; 610 sbi->s_first_error_line = line; 611 sbi->s_first_error_ino = ino; 612 sbi->s_first_error_block = block; 613 sbi->s_first_error_func = func; 614 sbi->s_first_error_time = sbi->s_last_error_time; 615 } 616 spin_unlock(&sbi->s_error_lock); 617 } 618 619 /* Deal with the reporting of failure conditions on a filesystem such as 620 * inconsistencies detected or read IO failures. 621 * 622 * On ext2, we can store the error state of the filesystem in the 623 * superblock. That is not possible on ext4, because we may have other 624 * write ordering constraints on the superblock which prevent us from 625 * writing it out straight away; and given that the journal is about to 626 * be aborted, we can't rely on the current, or future, transactions to 627 * write out the superblock safely. 628 * 629 * We'll just use the jbd2_journal_abort() error code to record an error in 630 * the journal instead. On recovery, the journal will complain about 631 * that error until we've noted it down and cleared it. 632 * 633 * If force_ro is set, we unconditionally force the filesystem into an 634 * ABORT|READONLY state, unless the error response on the fs has been set to 635 * panic in which case we take the easy way out and panic immediately. This is 636 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 637 * at a critical moment in log management. 638 */ 639 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 640 __u32 ino, __u64 block, 641 const char *func, unsigned int line) 642 { 643 journal_t *journal = EXT4_SB(sb)->s_journal; 644 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 645 646 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 647 if (test_opt(sb, WARN_ON_ERROR)) 648 WARN_ON_ONCE(1); 649 650 if (!continue_fs && !sb_rdonly(sb)) { 651 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 652 if (journal) 653 jbd2_journal_abort(journal, -EIO); 654 } 655 656 if (!bdev_read_only(sb->s_bdev)) { 657 save_error_info(sb, error, ino, block, func, line); 658 /* 659 * In case the fs should keep running, we need to writeout 660 * superblock through the journal. Due to lock ordering 661 * constraints, it may not be safe to do it right here so we 662 * defer superblock flushing to a workqueue. 663 */ 664 if (continue_fs) 665 schedule_work(&EXT4_SB(sb)->s_error_work); 666 else 667 ext4_commit_super(sb); 668 } 669 670 /* 671 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 672 * could panic during 'reboot -f' as the underlying device got already 673 * disabled. 674 */ 675 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 676 panic("EXT4-fs (device %s): panic forced after error\n", 677 sb->s_id); 678 } 679 680 if (sb_rdonly(sb) || continue_fs) 681 return; 682 683 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 684 /* 685 * Make sure updated value of ->s_mount_flags will be visible before 686 * ->s_flags update 687 */ 688 smp_wmb(); 689 sb->s_flags |= SB_RDONLY; 690 } 691 692 static void flush_stashed_error_work(struct work_struct *work) 693 { 694 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 695 s_error_work); 696 journal_t *journal = sbi->s_journal; 697 handle_t *handle; 698 699 /* 700 * If the journal is still running, we have to write out superblock 701 * through the journal to avoid collisions of other journalled sb 702 * updates. 703 * 704 * We use directly jbd2 functions here to avoid recursing back into 705 * ext4 error handling code during handling of previous errors. 706 */ 707 if (!sb_rdonly(sbi->s_sb) && journal) { 708 handle = jbd2_journal_start(journal, 1); 709 if (IS_ERR(handle)) 710 goto write_directly; 711 if (jbd2_journal_get_write_access(handle, sbi->s_sbh)) { 712 jbd2_journal_stop(handle); 713 goto write_directly; 714 } 715 ext4_update_super(sbi->s_sb); 716 if (jbd2_journal_dirty_metadata(handle, sbi->s_sbh)) { 717 jbd2_journal_stop(handle); 718 goto write_directly; 719 } 720 jbd2_journal_stop(handle); 721 return; 722 } 723 write_directly: 724 /* 725 * Write through journal failed. Write sb directly to get error info 726 * out and hope for the best. 727 */ 728 ext4_commit_super(sbi->s_sb); 729 } 730 731 #define ext4_error_ratelimit(sb) \ 732 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 733 "EXT4-fs error") 734 735 void __ext4_error(struct super_block *sb, const char *function, 736 unsigned int line, bool force_ro, int error, __u64 block, 737 const char *fmt, ...) 738 { 739 struct va_format vaf; 740 va_list args; 741 742 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 743 return; 744 745 trace_ext4_error(sb, function, line); 746 if (ext4_error_ratelimit(sb)) { 747 va_start(args, fmt); 748 vaf.fmt = fmt; 749 vaf.va = &args; 750 printk(KERN_CRIT 751 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 752 sb->s_id, function, line, current->comm, &vaf); 753 va_end(args); 754 } 755 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 756 } 757 758 void __ext4_error_inode(struct inode *inode, const char *function, 759 unsigned int line, ext4_fsblk_t block, int error, 760 const char *fmt, ...) 761 { 762 va_list args; 763 struct va_format vaf; 764 765 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 766 return; 767 768 trace_ext4_error(inode->i_sb, function, line); 769 if (ext4_error_ratelimit(inode->i_sb)) { 770 va_start(args, fmt); 771 vaf.fmt = fmt; 772 vaf.va = &args; 773 if (block) 774 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 775 "inode #%lu: block %llu: comm %s: %pV\n", 776 inode->i_sb->s_id, function, line, inode->i_ino, 777 block, current->comm, &vaf); 778 else 779 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 780 "inode #%lu: comm %s: %pV\n", 781 inode->i_sb->s_id, function, line, inode->i_ino, 782 current->comm, &vaf); 783 va_end(args); 784 } 785 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 786 function, line); 787 } 788 789 void __ext4_error_file(struct file *file, const char *function, 790 unsigned int line, ext4_fsblk_t block, 791 const char *fmt, ...) 792 { 793 va_list args; 794 struct va_format vaf; 795 struct inode *inode = file_inode(file); 796 char pathname[80], *path; 797 798 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 799 return; 800 801 trace_ext4_error(inode->i_sb, function, line); 802 if (ext4_error_ratelimit(inode->i_sb)) { 803 path = file_path(file, pathname, sizeof(pathname)); 804 if (IS_ERR(path)) 805 path = "(unknown)"; 806 va_start(args, fmt); 807 vaf.fmt = fmt; 808 vaf.va = &args; 809 if (block) 810 printk(KERN_CRIT 811 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 812 "block %llu: comm %s: path %s: %pV\n", 813 inode->i_sb->s_id, function, line, inode->i_ino, 814 block, current->comm, path, &vaf); 815 else 816 printk(KERN_CRIT 817 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 818 "comm %s: path %s: %pV\n", 819 inode->i_sb->s_id, function, line, inode->i_ino, 820 current->comm, path, &vaf); 821 va_end(args); 822 } 823 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 824 function, line); 825 } 826 827 const char *ext4_decode_error(struct super_block *sb, int errno, 828 char nbuf[16]) 829 { 830 char *errstr = NULL; 831 832 switch (errno) { 833 case -EFSCORRUPTED: 834 errstr = "Corrupt filesystem"; 835 break; 836 case -EFSBADCRC: 837 errstr = "Filesystem failed CRC"; 838 break; 839 case -EIO: 840 errstr = "IO failure"; 841 break; 842 case -ENOMEM: 843 errstr = "Out of memory"; 844 break; 845 case -EROFS: 846 if (!sb || (EXT4_SB(sb)->s_journal && 847 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 848 errstr = "Journal has aborted"; 849 else 850 errstr = "Readonly filesystem"; 851 break; 852 default: 853 /* If the caller passed in an extra buffer for unknown 854 * errors, textualise them now. Else we just return 855 * NULL. */ 856 if (nbuf) { 857 /* Check for truncated error codes... */ 858 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 859 errstr = nbuf; 860 } 861 break; 862 } 863 864 return errstr; 865 } 866 867 /* __ext4_std_error decodes expected errors from journaling functions 868 * automatically and invokes the appropriate error response. */ 869 870 void __ext4_std_error(struct super_block *sb, const char *function, 871 unsigned int line, int errno) 872 { 873 char nbuf[16]; 874 const char *errstr; 875 876 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 877 return; 878 879 /* Special case: if the error is EROFS, and we're not already 880 * inside a transaction, then there's really no point in logging 881 * an error. */ 882 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 883 return; 884 885 if (ext4_error_ratelimit(sb)) { 886 errstr = ext4_decode_error(sb, errno, nbuf); 887 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 888 sb->s_id, function, line, errstr); 889 } 890 891 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 892 } 893 894 void __ext4_msg(struct super_block *sb, 895 const char *prefix, const char *fmt, ...) 896 { 897 struct va_format vaf; 898 va_list args; 899 900 atomic_inc(&EXT4_SB(sb)->s_msg_count); 901 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs")) 902 return; 903 904 va_start(args, fmt); 905 vaf.fmt = fmt; 906 vaf.va = &args; 907 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 908 va_end(args); 909 } 910 911 static int ext4_warning_ratelimit(struct super_block *sb) 912 { 913 atomic_inc(&EXT4_SB(sb)->s_warning_count); 914 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 915 "EXT4-fs warning"); 916 } 917 918 void __ext4_warning(struct super_block *sb, const char *function, 919 unsigned int line, const char *fmt, ...) 920 { 921 struct va_format vaf; 922 va_list args; 923 924 if (!ext4_warning_ratelimit(sb)) 925 return; 926 927 va_start(args, fmt); 928 vaf.fmt = fmt; 929 vaf.va = &args; 930 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 931 sb->s_id, function, line, &vaf); 932 va_end(args); 933 } 934 935 void __ext4_warning_inode(const struct inode *inode, const char *function, 936 unsigned int line, const char *fmt, ...) 937 { 938 struct va_format vaf; 939 va_list args; 940 941 if (!ext4_warning_ratelimit(inode->i_sb)) 942 return; 943 944 va_start(args, fmt); 945 vaf.fmt = fmt; 946 vaf.va = &args; 947 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 948 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 949 function, line, inode->i_ino, current->comm, &vaf); 950 va_end(args); 951 } 952 953 void __ext4_grp_locked_error(const char *function, unsigned int line, 954 struct super_block *sb, ext4_group_t grp, 955 unsigned long ino, ext4_fsblk_t block, 956 const char *fmt, ...) 957 __releases(bitlock) 958 __acquires(bitlock) 959 { 960 struct va_format vaf; 961 va_list args; 962 963 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 964 return; 965 966 trace_ext4_error(sb, function, line); 967 if (ext4_error_ratelimit(sb)) { 968 va_start(args, fmt); 969 vaf.fmt = fmt; 970 vaf.va = &args; 971 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 972 sb->s_id, function, line, grp); 973 if (ino) 974 printk(KERN_CONT "inode %lu: ", ino); 975 if (block) 976 printk(KERN_CONT "block %llu:", 977 (unsigned long long) block); 978 printk(KERN_CONT "%pV\n", &vaf); 979 va_end(args); 980 } 981 982 if (test_opt(sb, ERRORS_CONT)) { 983 if (test_opt(sb, WARN_ON_ERROR)) 984 WARN_ON_ONCE(1); 985 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 986 if (!bdev_read_only(sb->s_bdev)) { 987 save_error_info(sb, EFSCORRUPTED, ino, block, function, 988 line); 989 schedule_work(&EXT4_SB(sb)->s_error_work); 990 } 991 return; 992 } 993 ext4_unlock_group(sb, grp); 994 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 995 /* 996 * We only get here in the ERRORS_RO case; relocking the group 997 * may be dangerous, but nothing bad will happen since the 998 * filesystem will have already been marked read/only and the 999 * journal has been aborted. We return 1 as a hint to callers 1000 * who might what to use the return value from 1001 * ext4_grp_locked_error() to distinguish between the 1002 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1003 * aggressively from the ext4 function in question, with a 1004 * more appropriate error code. 1005 */ 1006 ext4_lock_group(sb, grp); 1007 return; 1008 } 1009 1010 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1011 ext4_group_t group, 1012 unsigned int flags) 1013 { 1014 struct ext4_sb_info *sbi = EXT4_SB(sb); 1015 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1016 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1017 int ret; 1018 1019 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1020 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1021 &grp->bb_state); 1022 if (!ret) 1023 percpu_counter_sub(&sbi->s_freeclusters_counter, 1024 grp->bb_free); 1025 } 1026 1027 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1028 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1029 &grp->bb_state); 1030 if (!ret && gdp) { 1031 int count; 1032 1033 count = ext4_free_inodes_count(sb, gdp); 1034 percpu_counter_sub(&sbi->s_freeinodes_counter, 1035 count); 1036 } 1037 } 1038 } 1039 1040 void ext4_update_dynamic_rev(struct super_block *sb) 1041 { 1042 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1043 1044 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1045 return; 1046 1047 ext4_warning(sb, 1048 "updating to rev %d because of new feature flag, " 1049 "running e2fsck is recommended", 1050 EXT4_DYNAMIC_REV); 1051 1052 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1053 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1054 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1055 /* leave es->s_feature_*compat flags alone */ 1056 /* es->s_uuid will be set by e2fsck if empty */ 1057 1058 /* 1059 * The rest of the superblock fields should be zero, and if not it 1060 * means they are likely already in use, so leave them alone. We 1061 * can leave it up to e2fsck to clean up any inconsistencies there. 1062 */ 1063 } 1064 1065 /* 1066 * Open the external journal device 1067 */ 1068 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 1069 { 1070 struct block_device *bdev; 1071 1072 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 1073 if (IS_ERR(bdev)) 1074 goto fail; 1075 return bdev; 1076 1077 fail: 1078 ext4_msg(sb, KERN_ERR, 1079 "failed to open journal device unknown-block(%u,%u) %ld", 1080 MAJOR(dev), MINOR(dev), PTR_ERR(bdev)); 1081 return NULL; 1082 } 1083 1084 /* 1085 * Release the journal device 1086 */ 1087 static void ext4_blkdev_put(struct block_device *bdev) 1088 { 1089 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1090 } 1091 1092 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 1093 { 1094 struct block_device *bdev; 1095 bdev = sbi->s_journal_bdev; 1096 if (bdev) { 1097 ext4_blkdev_put(bdev); 1098 sbi->s_journal_bdev = NULL; 1099 } 1100 } 1101 1102 static inline struct inode *orphan_list_entry(struct list_head *l) 1103 { 1104 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1105 } 1106 1107 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1108 { 1109 struct list_head *l; 1110 1111 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1112 le32_to_cpu(sbi->s_es->s_last_orphan)); 1113 1114 printk(KERN_ERR "sb_info orphan list:\n"); 1115 list_for_each(l, &sbi->s_orphan) { 1116 struct inode *inode = orphan_list_entry(l); 1117 printk(KERN_ERR " " 1118 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1119 inode->i_sb->s_id, inode->i_ino, inode, 1120 inode->i_mode, inode->i_nlink, 1121 NEXT_ORPHAN(inode)); 1122 } 1123 } 1124 1125 #ifdef CONFIG_QUOTA 1126 static int ext4_quota_off(struct super_block *sb, int type); 1127 1128 static inline void ext4_quota_off_umount(struct super_block *sb) 1129 { 1130 int type; 1131 1132 /* Use our quota_off function to clear inode flags etc. */ 1133 for (type = 0; type < EXT4_MAXQUOTAS; type++) 1134 ext4_quota_off(sb, type); 1135 } 1136 1137 /* 1138 * This is a helper function which is used in the mount/remount 1139 * codepaths (which holds s_umount) to fetch the quota file name. 1140 */ 1141 static inline char *get_qf_name(struct super_block *sb, 1142 struct ext4_sb_info *sbi, 1143 int type) 1144 { 1145 return rcu_dereference_protected(sbi->s_qf_names[type], 1146 lockdep_is_held(&sb->s_umount)); 1147 } 1148 #else 1149 static inline void ext4_quota_off_umount(struct super_block *sb) 1150 { 1151 } 1152 #endif 1153 1154 static void ext4_put_super(struct super_block *sb) 1155 { 1156 struct ext4_sb_info *sbi = EXT4_SB(sb); 1157 struct ext4_super_block *es = sbi->s_es; 1158 struct buffer_head **group_desc; 1159 struct flex_groups **flex_groups; 1160 int aborted = 0; 1161 int i, err; 1162 1163 ext4_unregister_li_request(sb); 1164 ext4_quota_off_umount(sb); 1165 1166 flush_work(&sbi->s_error_work); 1167 destroy_workqueue(sbi->rsv_conversion_wq); 1168 1169 /* 1170 * Unregister sysfs before destroying jbd2 journal. 1171 * Since we could still access attr_journal_task attribute via sysfs 1172 * path which could have sbi->s_journal->j_task as NULL 1173 */ 1174 ext4_unregister_sysfs(sb); 1175 1176 if (sbi->s_journal) { 1177 aborted = is_journal_aborted(sbi->s_journal); 1178 err = jbd2_journal_destroy(sbi->s_journal); 1179 sbi->s_journal = NULL; 1180 if ((err < 0) && !aborted) { 1181 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1182 } 1183 } 1184 1185 ext4_es_unregister_shrinker(sbi); 1186 del_timer_sync(&sbi->s_err_report); 1187 ext4_release_system_zone(sb); 1188 ext4_mb_release(sb); 1189 ext4_ext_release(sb); 1190 1191 if (!sb_rdonly(sb) && !aborted) { 1192 ext4_clear_feature_journal_needs_recovery(sb); 1193 es->s_state = cpu_to_le16(sbi->s_mount_state); 1194 } 1195 if (!sb_rdonly(sb)) 1196 ext4_commit_super(sb); 1197 1198 rcu_read_lock(); 1199 group_desc = rcu_dereference(sbi->s_group_desc); 1200 for (i = 0; i < sbi->s_gdb_count; i++) 1201 brelse(group_desc[i]); 1202 kvfree(group_desc); 1203 flex_groups = rcu_dereference(sbi->s_flex_groups); 1204 if (flex_groups) { 1205 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1206 kvfree(flex_groups[i]); 1207 kvfree(flex_groups); 1208 } 1209 rcu_read_unlock(); 1210 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1211 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1212 percpu_counter_destroy(&sbi->s_dirs_counter); 1213 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1214 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1215 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1216 #ifdef CONFIG_QUOTA 1217 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1218 kfree(get_qf_name(sb, sbi, i)); 1219 #endif 1220 1221 /* Debugging code just in case the in-memory inode orphan list 1222 * isn't empty. The on-disk one can be non-empty if we've 1223 * detected an error and taken the fs readonly, but the 1224 * in-memory list had better be clean by this point. */ 1225 if (!list_empty(&sbi->s_orphan)) 1226 dump_orphan_list(sb, sbi); 1227 ASSERT(list_empty(&sbi->s_orphan)); 1228 1229 sync_blockdev(sb->s_bdev); 1230 invalidate_bdev(sb->s_bdev); 1231 if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) { 1232 /* 1233 * Invalidate the journal device's buffers. We don't want them 1234 * floating about in memory - the physical journal device may 1235 * hotswapped, and it breaks the `ro-after' testing code. 1236 */ 1237 sync_blockdev(sbi->s_journal_bdev); 1238 invalidate_bdev(sbi->s_journal_bdev); 1239 ext4_blkdev_remove(sbi); 1240 } 1241 1242 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1243 sbi->s_ea_inode_cache = NULL; 1244 1245 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1246 sbi->s_ea_block_cache = NULL; 1247 1248 if (sbi->s_mmp_tsk) 1249 kthread_stop(sbi->s_mmp_tsk); 1250 brelse(sbi->s_sbh); 1251 sb->s_fs_info = NULL; 1252 /* 1253 * Now that we are completely done shutting down the 1254 * superblock, we need to actually destroy the kobject. 1255 */ 1256 kobject_put(&sbi->s_kobj); 1257 wait_for_completion(&sbi->s_kobj_unregister); 1258 if (sbi->s_chksum_driver) 1259 crypto_free_shash(sbi->s_chksum_driver); 1260 kfree(sbi->s_blockgroup_lock); 1261 fs_put_dax(sbi->s_daxdev); 1262 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1263 #ifdef CONFIG_UNICODE 1264 utf8_unload(sb->s_encoding); 1265 #endif 1266 kfree(sbi); 1267 } 1268 1269 static struct kmem_cache *ext4_inode_cachep; 1270 1271 /* 1272 * Called inside transaction, so use GFP_NOFS 1273 */ 1274 static struct inode *ext4_alloc_inode(struct super_block *sb) 1275 { 1276 struct ext4_inode_info *ei; 1277 1278 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS); 1279 if (!ei) 1280 return NULL; 1281 1282 inode_set_iversion(&ei->vfs_inode, 1); 1283 spin_lock_init(&ei->i_raw_lock); 1284 INIT_LIST_HEAD(&ei->i_prealloc_list); 1285 atomic_set(&ei->i_prealloc_active, 0); 1286 spin_lock_init(&ei->i_prealloc_lock); 1287 ext4_es_init_tree(&ei->i_es_tree); 1288 rwlock_init(&ei->i_es_lock); 1289 INIT_LIST_HEAD(&ei->i_es_list); 1290 ei->i_es_all_nr = 0; 1291 ei->i_es_shk_nr = 0; 1292 ei->i_es_shrink_lblk = 0; 1293 ei->i_reserved_data_blocks = 0; 1294 spin_lock_init(&(ei->i_block_reservation_lock)); 1295 ext4_init_pending_tree(&ei->i_pending_tree); 1296 #ifdef CONFIG_QUOTA 1297 ei->i_reserved_quota = 0; 1298 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1299 #endif 1300 ei->jinode = NULL; 1301 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1302 spin_lock_init(&ei->i_completed_io_lock); 1303 ei->i_sync_tid = 0; 1304 ei->i_datasync_tid = 0; 1305 atomic_set(&ei->i_unwritten, 0); 1306 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1307 ext4_fc_init_inode(&ei->vfs_inode); 1308 mutex_init(&ei->i_fc_lock); 1309 return &ei->vfs_inode; 1310 } 1311 1312 static int ext4_drop_inode(struct inode *inode) 1313 { 1314 int drop = generic_drop_inode(inode); 1315 1316 if (!drop) 1317 drop = fscrypt_drop_inode(inode); 1318 1319 trace_ext4_drop_inode(inode, drop); 1320 return drop; 1321 } 1322 1323 static void ext4_free_in_core_inode(struct inode *inode) 1324 { 1325 fscrypt_free_inode(inode); 1326 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1327 pr_warn("%s: inode %ld still in fc list", 1328 __func__, inode->i_ino); 1329 } 1330 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1331 } 1332 1333 static void ext4_destroy_inode(struct inode *inode) 1334 { 1335 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1336 ext4_msg(inode->i_sb, KERN_ERR, 1337 "Inode %lu (%p): orphan list check failed!", 1338 inode->i_ino, EXT4_I(inode)); 1339 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1340 EXT4_I(inode), sizeof(struct ext4_inode_info), 1341 true); 1342 dump_stack(); 1343 } 1344 } 1345 1346 static void init_once(void *foo) 1347 { 1348 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo; 1349 1350 INIT_LIST_HEAD(&ei->i_orphan); 1351 init_rwsem(&ei->xattr_sem); 1352 init_rwsem(&ei->i_data_sem); 1353 init_rwsem(&ei->i_mmap_sem); 1354 inode_init_once(&ei->vfs_inode); 1355 ext4_fc_init_inode(&ei->vfs_inode); 1356 } 1357 1358 static int __init init_inodecache(void) 1359 { 1360 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1361 sizeof(struct ext4_inode_info), 0, 1362 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1363 SLAB_ACCOUNT), 1364 offsetof(struct ext4_inode_info, i_data), 1365 sizeof_field(struct ext4_inode_info, i_data), 1366 init_once); 1367 if (ext4_inode_cachep == NULL) 1368 return -ENOMEM; 1369 return 0; 1370 } 1371 1372 static void destroy_inodecache(void) 1373 { 1374 /* 1375 * Make sure all delayed rcu free inodes are flushed before we 1376 * destroy cache. 1377 */ 1378 rcu_barrier(); 1379 kmem_cache_destroy(ext4_inode_cachep); 1380 } 1381 1382 void ext4_clear_inode(struct inode *inode) 1383 { 1384 ext4_fc_del(inode); 1385 invalidate_inode_buffers(inode); 1386 clear_inode(inode); 1387 ext4_discard_preallocations(inode, 0); 1388 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1389 dquot_drop(inode); 1390 if (EXT4_I(inode)->jinode) { 1391 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1392 EXT4_I(inode)->jinode); 1393 jbd2_free_inode(EXT4_I(inode)->jinode); 1394 EXT4_I(inode)->jinode = NULL; 1395 } 1396 fscrypt_put_encryption_info(inode); 1397 fsverity_cleanup_inode(inode); 1398 } 1399 1400 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1401 u64 ino, u32 generation) 1402 { 1403 struct inode *inode; 1404 1405 /* 1406 * Currently we don't know the generation for parent directory, so 1407 * a generation of 0 means "accept any" 1408 */ 1409 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1410 if (IS_ERR(inode)) 1411 return ERR_CAST(inode); 1412 if (generation && inode->i_generation != generation) { 1413 iput(inode); 1414 return ERR_PTR(-ESTALE); 1415 } 1416 1417 return inode; 1418 } 1419 1420 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1421 int fh_len, int fh_type) 1422 { 1423 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1424 ext4_nfs_get_inode); 1425 } 1426 1427 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1428 int fh_len, int fh_type) 1429 { 1430 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1431 ext4_nfs_get_inode); 1432 } 1433 1434 static int ext4_nfs_commit_metadata(struct inode *inode) 1435 { 1436 struct writeback_control wbc = { 1437 .sync_mode = WB_SYNC_ALL 1438 }; 1439 1440 trace_ext4_nfs_commit_metadata(inode); 1441 return ext4_write_inode(inode, &wbc); 1442 } 1443 1444 /* 1445 * Try to release metadata pages (indirect blocks, directories) which are 1446 * mapped via the block device. Since these pages could have journal heads 1447 * which would prevent try_to_free_buffers() from freeing them, we must use 1448 * jbd2 layer's try_to_free_buffers() function to release them. 1449 */ 1450 static int bdev_try_to_free_page(struct super_block *sb, struct page *page, 1451 gfp_t wait) 1452 { 1453 journal_t *journal = EXT4_SB(sb)->s_journal; 1454 1455 WARN_ON(PageChecked(page)); 1456 if (!page_has_buffers(page)) 1457 return 0; 1458 if (journal) 1459 return jbd2_journal_try_to_free_buffers(journal, page); 1460 1461 return try_to_free_buffers(page); 1462 } 1463 1464 #ifdef CONFIG_FS_ENCRYPTION 1465 static int ext4_get_context(struct inode *inode, void *ctx, size_t len) 1466 { 1467 return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, 1468 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len); 1469 } 1470 1471 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len, 1472 void *fs_data) 1473 { 1474 handle_t *handle = fs_data; 1475 int res, res2, credits, retries = 0; 1476 1477 /* 1478 * Encrypting the root directory is not allowed because e2fsck expects 1479 * lost+found to exist and be unencrypted, and encrypting the root 1480 * directory would imply encrypting the lost+found directory as well as 1481 * the filename "lost+found" itself. 1482 */ 1483 if (inode->i_ino == EXT4_ROOT_INO) 1484 return -EPERM; 1485 1486 if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode))) 1487 return -EINVAL; 1488 1489 if (ext4_test_inode_flag(inode, EXT4_INODE_DAX)) 1490 return -EOPNOTSUPP; 1491 1492 res = ext4_convert_inline_data(inode); 1493 if (res) 1494 return res; 1495 1496 /* 1497 * If a journal handle was specified, then the encryption context is 1498 * being set on a new inode via inheritance and is part of a larger 1499 * transaction to create the inode. Otherwise the encryption context is 1500 * being set on an existing inode in its own transaction. Only in the 1501 * latter case should the "retry on ENOSPC" logic be used. 1502 */ 1503 1504 if (handle) { 1505 res = ext4_xattr_set_handle(handle, inode, 1506 EXT4_XATTR_INDEX_ENCRYPTION, 1507 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1508 ctx, len, 0); 1509 if (!res) { 1510 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1511 ext4_clear_inode_state(inode, 1512 EXT4_STATE_MAY_INLINE_DATA); 1513 /* 1514 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1515 * S_DAX may be disabled 1516 */ 1517 ext4_set_inode_flags(inode, false); 1518 } 1519 return res; 1520 } 1521 1522 res = dquot_initialize(inode); 1523 if (res) 1524 return res; 1525 retry: 1526 res = ext4_xattr_set_credits(inode, len, false /* is_create */, 1527 &credits); 1528 if (res) 1529 return res; 1530 1531 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); 1532 if (IS_ERR(handle)) 1533 return PTR_ERR(handle); 1534 1535 res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION, 1536 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1537 ctx, len, 0); 1538 if (!res) { 1539 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1540 /* 1541 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1542 * S_DAX may be disabled 1543 */ 1544 ext4_set_inode_flags(inode, false); 1545 res = ext4_mark_inode_dirty(handle, inode); 1546 if (res) 1547 EXT4_ERROR_INODE(inode, "Failed to mark inode dirty"); 1548 } 1549 res2 = ext4_journal_stop(handle); 1550 1551 if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 1552 goto retry; 1553 if (!res) 1554 res = res2; 1555 return res; 1556 } 1557 1558 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb) 1559 { 1560 return EXT4_SB(sb)->s_dummy_enc_policy.policy; 1561 } 1562 1563 static bool ext4_has_stable_inodes(struct super_block *sb) 1564 { 1565 return ext4_has_feature_stable_inodes(sb); 1566 } 1567 1568 static void ext4_get_ino_and_lblk_bits(struct super_block *sb, 1569 int *ino_bits_ret, int *lblk_bits_ret) 1570 { 1571 *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count); 1572 *lblk_bits_ret = 8 * sizeof(ext4_lblk_t); 1573 } 1574 1575 static const struct fscrypt_operations ext4_cryptops = { 1576 .key_prefix = "ext4:", 1577 .get_context = ext4_get_context, 1578 .set_context = ext4_set_context, 1579 .get_dummy_policy = ext4_get_dummy_policy, 1580 .empty_dir = ext4_empty_dir, 1581 .max_namelen = EXT4_NAME_LEN, 1582 .has_stable_inodes = ext4_has_stable_inodes, 1583 .get_ino_and_lblk_bits = ext4_get_ino_and_lblk_bits, 1584 }; 1585 #endif 1586 1587 #ifdef CONFIG_QUOTA 1588 static const char * const quotatypes[] = INITQFNAMES; 1589 #define QTYPE2NAME(t) (quotatypes[t]) 1590 1591 static int ext4_write_dquot(struct dquot *dquot); 1592 static int ext4_acquire_dquot(struct dquot *dquot); 1593 static int ext4_release_dquot(struct dquot *dquot); 1594 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1595 static int ext4_write_info(struct super_block *sb, int type); 1596 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1597 const struct path *path); 1598 static int ext4_quota_on_mount(struct super_block *sb, int type); 1599 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1600 size_t len, loff_t off); 1601 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1602 const char *data, size_t len, loff_t off); 1603 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1604 unsigned int flags); 1605 static int ext4_enable_quotas(struct super_block *sb); 1606 1607 static struct dquot **ext4_get_dquots(struct inode *inode) 1608 { 1609 return EXT4_I(inode)->i_dquot; 1610 } 1611 1612 static const struct dquot_operations ext4_quota_operations = { 1613 .get_reserved_space = ext4_get_reserved_space, 1614 .write_dquot = ext4_write_dquot, 1615 .acquire_dquot = ext4_acquire_dquot, 1616 .release_dquot = ext4_release_dquot, 1617 .mark_dirty = ext4_mark_dquot_dirty, 1618 .write_info = ext4_write_info, 1619 .alloc_dquot = dquot_alloc, 1620 .destroy_dquot = dquot_destroy, 1621 .get_projid = ext4_get_projid, 1622 .get_inode_usage = ext4_get_inode_usage, 1623 .get_next_id = dquot_get_next_id, 1624 }; 1625 1626 static const struct quotactl_ops ext4_qctl_operations = { 1627 .quota_on = ext4_quota_on, 1628 .quota_off = ext4_quota_off, 1629 .quota_sync = dquot_quota_sync, 1630 .get_state = dquot_get_state, 1631 .set_info = dquot_set_dqinfo, 1632 .get_dqblk = dquot_get_dqblk, 1633 .set_dqblk = dquot_set_dqblk, 1634 .get_nextdqblk = dquot_get_next_dqblk, 1635 }; 1636 #endif 1637 1638 static const struct super_operations ext4_sops = { 1639 .alloc_inode = ext4_alloc_inode, 1640 .free_inode = ext4_free_in_core_inode, 1641 .destroy_inode = ext4_destroy_inode, 1642 .write_inode = ext4_write_inode, 1643 .dirty_inode = ext4_dirty_inode, 1644 .drop_inode = ext4_drop_inode, 1645 .evict_inode = ext4_evict_inode, 1646 .put_super = ext4_put_super, 1647 .sync_fs = ext4_sync_fs, 1648 .freeze_fs = ext4_freeze, 1649 .unfreeze_fs = ext4_unfreeze, 1650 .statfs = ext4_statfs, 1651 .remount_fs = ext4_remount, 1652 .show_options = ext4_show_options, 1653 #ifdef CONFIG_QUOTA 1654 .quota_read = ext4_quota_read, 1655 .quota_write = ext4_quota_write, 1656 .get_dquots = ext4_get_dquots, 1657 #endif 1658 .bdev_try_to_free_page = bdev_try_to_free_page, 1659 }; 1660 1661 static const struct export_operations ext4_export_ops = { 1662 .fh_to_dentry = ext4_fh_to_dentry, 1663 .fh_to_parent = ext4_fh_to_parent, 1664 .get_parent = ext4_get_parent, 1665 .commit_metadata = ext4_nfs_commit_metadata, 1666 }; 1667 1668 enum { 1669 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1670 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro, 1671 Opt_nouid32, Opt_debug, Opt_removed, 1672 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, 1673 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1674 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1675 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1676 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1677 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1678 Opt_inlinecrypt, 1679 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota, 1680 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota, 1681 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1682 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, 1683 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1684 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1685 Opt_nowarn_on_error, Opt_mblk_io_submit, 1686 Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize, 1687 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1688 Opt_inode_readahead_blks, Opt_journal_ioprio, 1689 Opt_dioread_nolock, Opt_dioread_lock, 1690 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1691 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1692 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1693 #ifdef CONFIG_EXT4_DEBUG 1694 Opt_fc_debug_max_replay, Opt_fc_debug_force 1695 #endif 1696 }; 1697 1698 static const match_table_t tokens = { 1699 {Opt_bsd_df, "bsddf"}, 1700 {Opt_minix_df, "minixdf"}, 1701 {Opt_grpid, "grpid"}, 1702 {Opt_grpid, "bsdgroups"}, 1703 {Opt_nogrpid, "nogrpid"}, 1704 {Opt_nogrpid, "sysvgroups"}, 1705 {Opt_resgid, "resgid=%u"}, 1706 {Opt_resuid, "resuid=%u"}, 1707 {Opt_sb, "sb=%u"}, 1708 {Opt_err_cont, "errors=continue"}, 1709 {Opt_err_panic, "errors=panic"}, 1710 {Opt_err_ro, "errors=remount-ro"}, 1711 {Opt_nouid32, "nouid32"}, 1712 {Opt_debug, "debug"}, 1713 {Opt_removed, "oldalloc"}, 1714 {Opt_removed, "orlov"}, 1715 {Opt_user_xattr, "user_xattr"}, 1716 {Opt_nouser_xattr, "nouser_xattr"}, 1717 {Opt_acl, "acl"}, 1718 {Opt_noacl, "noacl"}, 1719 {Opt_noload, "norecovery"}, 1720 {Opt_noload, "noload"}, 1721 {Opt_removed, "nobh"}, 1722 {Opt_removed, "bh"}, 1723 {Opt_commit, "commit=%u"}, 1724 {Opt_min_batch_time, "min_batch_time=%u"}, 1725 {Opt_max_batch_time, "max_batch_time=%u"}, 1726 {Opt_journal_dev, "journal_dev=%u"}, 1727 {Opt_journal_path, "journal_path=%s"}, 1728 {Opt_journal_checksum, "journal_checksum"}, 1729 {Opt_nojournal_checksum, "nojournal_checksum"}, 1730 {Opt_journal_async_commit, "journal_async_commit"}, 1731 {Opt_abort, "abort"}, 1732 {Opt_data_journal, "data=journal"}, 1733 {Opt_data_ordered, "data=ordered"}, 1734 {Opt_data_writeback, "data=writeback"}, 1735 {Opt_data_err_abort, "data_err=abort"}, 1736 {Opt_data_err_ignore, "data_err=ignore"}, 1737 {Opt_offusrjquota, "usrjquota="}, 1738 {Opt_usrjquota, "usrjquota=%s"}, 1739 {Opt_offgrpjquota, "grpjquota="}, 1740 {Opt_grpjquota, "grpjquota=%s"}, 1741 {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, 1742 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, 1743 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, 1744 {Opt_grpquota, "grpquota"}, 1745 {Opt_noquota, "noquota"}, 1746 {Opt_quota, "quota"}, 1747 {Opt_usrquota, "usrquota"}, 1748 {Opt_prjquota, "prjquota"}, 1749 {Opt_barrier, "barrier=%u"}, 1750 {Opt_barrier, "barrier"}, 1751 {Opt_nobarrier, "nobarrier"}, 1752 {Opt_i_version, "i_version"}, 1753 {Opt_dax, "dax"}, 1754 {Opt_dax_always, "dax=always"}, 1755 {Opt_dax_inode, "dax=inode"}, 1756 {Opt_dax_never, "dax=never"}, 1757 {Opt_stripe, "stripe=%u"}, 1758 {Opt_delalloc, "delalloc"}, 1759 {Opt_warn_on_error, "warn_on_error"}, 1760 {Opt_nowarn_on_error, "nowarn_on_error"}, 1761 {Opt_lazytime, "lazytime"}, 1762 {Opt_nolazytime, "nolazytime"}, 1763 {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"}, 1764 {Opt_nodelalloc, "nodelalloc"}, 1765 {Opt_removed, "mblk_io_submit"}, 1766 {Opt_removed, "nomblk_io_submit"}, 1767 {Opt_block_validity, "block_validity"}, 1768 {Opt_noblock_validity, "noblock_validity"}, 1769 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, 1770 {Opt_journal_ioprio, "journal_ioprio=%u"}, 1771 {Opt_auto_da_alloc, "auto_da_alloc=%u"}, 1772 {Opt_auto_da_alloc, "auto_da_alloc"}, 1773 {Opt_noauto_da_alloc, "noauto_da_alloc"}, 1774 {Opt_dioread_nolock, "dioread_nolock"}, 1775 {Opt_dioread_lock, "nodioread_nolock"}, 1776 {Opt_dioread_lock, "dioread_lock"}, 1777 {Opt_discard, "discard"}, 1778 {Opt_nodiscard, "nodiscard"}, 1779 {Opt_init_itable, "init_itable=%u"}, 1780 {Opt_init_itable, "init_itable"}, 1781 {Opt_noinit_itable, "noinit_itable"}, 1782 #ifdef CONFIG_EXT4_DEBUG 1783 {Opt_fc_debug_force, "fc_debug_force"}, 1784 {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"}, 1785 #endif 1786 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, 1787 {Opt_test_dummy_encryption, "test_dummy_encryption=%s"}, 1788 {Opt_test_dummy_encryption, "test_dummy_encryption"}, 1789 {Opt_inlinecrypt, "inlinecrypt"}, 1790 {Opt_nombcache, "nombcache"}, 1791 {Opt_nombcache, "no_mbcache"}, /* for backward compatibility */ 1792 {Opt_removed, "prefetch_block_bitmaps"}, 1793 {Opt_no_prefetch_block_bitmaps, "no_prefetch_block_bitmaps"}, 1794 {Opt_mb_optimize_scan, "mb_optimize_scan=%d"}, 1795 {Opt_removed, "check=none"}, /* mount option from ext2/3 */ 1796 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */ 1797 {Opt_removed, "reservation"}, /* mount option from ext2/3 */ 1798 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */ 1799 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */ 1800 {Opt_err, NULL}, 1801 }; 1802 1803 static ext4_fsblk_t get_sb_block(void **data) 1804 { 1805 ext4_fsblk_t sb_block; 1806 char *options = (char *) *data; 1807 1808 if (!options || strncmp(options, "sb=", 3) != 0) 1809 return 1; /* Default location */ 1810 1811 options += 3; 1812 /* TODO: use simple_strtoll with >32bit ext4 */ 1813 sb_block = simple_strtoul(options, &options, 0); 1814 if (*options && *options != ',') { 1815 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n", 1816 (char *) *data); 1817 return 1; 1818 } 1819 if (*options == ',') 1820 options++; 1821 *data = (void *) options; 1822 1823 return sb_block; 1824 } 1825 1826 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1827 #define DEFAULT_MB_OPTIMIZE_SCAN (-1) 1828 1829 static const char deprecated_msg[] = 1830 "Mount option \"%s\" will be removed by %s\n" 1831 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1832 1833 #ifdef CONFIG_QUOTA 1834 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args) 1835 { 1836 struct ext4_sb_info *sbi = EXT4_SB(sb); 1837 char *qname, *old_qname = get_qf_name(sb, sbi, qtype); 1838 int ret = -1; 1839 1840 if (sb_any_quota_loaded(sb) && !old_qname) { 1841 ext4_msg(sb, KERN_ERR, 1842 "Cannot change journaled " 1843 "quota options when quota turned on"); 1844 return -1; 1845 } 1846 if (ext4_has_feature_quota(sb)) { 1847 ext4_msg(sb, KERN_INFO, "Journaled quota options " 1848 "ignored when QUOTA feature is enabled"); 1849 return 1; 1850 } 1851 qname = match_strdup(args); 1852 if (!qname) { 1853 ext4_msg(sb, KERN_ERR, 1854 "Not enough memory for storing quotafile name"); 1855 return -1; 1856 } 1857 if (old_qname) { 1858 if (strcmp(old_qname, qname) == 0) 1859 ret = 1; 1860 else 1861 ext4_msg(sb, KERN_ERR, 1862 "%s quota file already specified", 1863 QTYPE2NAME(qtype)); 1864 goto errout; 1865 } 1866 if (strchr(qname, '/')) { 1867 ext4_msg(sb, KERN_ERR, 1868 "quotafile must be on filesystem root"); 1869 goto errout; 1870 } 1871 rcu_assign_pointer(sbi->s_qf_names[qtype], qname); 1872 set_opt(sb, QUOTA); 1873 return 1; 1874 errout: 1875 kfree(qname); 1876 return ret; 1877 } 1878 1879 static int clear_qf_name(struct super_block *sb, int qtype) 1880 { 1881 1882 struct ext4_sb_info *sbi = EXT4_SB(sb); 1883 char *old_qname = get_qf_name(sb, sbi, qtype); 1884 1885 if (sb_any_quota_loaded(sb) && old_qname) { 1886 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options" 1887 " when quota turned on"); 1888 return -1; 1889 } 1890 rcu_assign_pointer(sbi->s_qf_names[qtype], NULL); 1891 synchronize_rcu(); 1892 kfree(old_qname); 1893 return 1; 1894 } 1895 #endif 1896 1897 #define MOPT_SET 0x0001 1898 #define MOPT_CLEAR 0x0002 1899 #define MOPT_NOSUPPORT 0x0004 1900 #define MOPT_EXPLICIT 0x0008 1901 #define MOPT_CLEAR_ERR 0x0010 1902 #define MOPT_GTE0 0x0020 1903 #ifdef CONFIG_QUOTA 1904 #define MOPT_Q 0 1905 #define MOPT_QFMT 0x0040 1906 #else 1907 #define MOPT_Q MOPT_NOSUPPORT 1908 #define MOPT_QFMT MOPT_NOSUPPORT 1909 #endif 1910 #define MOPT_DATAJ 0x0080 1911 #define MOPT_NO_EXT2 0x0100 1912 #define MOPT_NO_EXT3 0x0200 1913 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1914 #define MOPT_STRING 0x0400 1915 #define MOPT_SKIP 0x0800 1916 #define MOPT_2 0x1000 1917 1918 static const struct mount_opts { 1919 int token; 1920 int mount_opt; 1921 int flags; 1922 } ext4_mount_opts[] = { 1923 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1924 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1925 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1926 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1927 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1928 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1929 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1930 MOPT_EXT4_ONLY | MOPT_SET}, 1931 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1932 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1933 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1934 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1935 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1936 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1937 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1938 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1939 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1940 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1941 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1942 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1943 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1944 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1945 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1946 EXT4_MOUNT_JOURNAL_CHECKSUM), 1947 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1948 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1949 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR}, 1950 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR}, 1951 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR}, 1952 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, 1953 MOPT_NO_EXT2}, 1954 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, 1955 MOPT_NO_EXT2}, 1956 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1957 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1958 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1959 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1960 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1961 {Opt_commit, 0, MOPT_GTE0}, 1962 {Opt_max_batch_time, 0, MOPT_GTE0}, 1963 {Opt_min_batch_time, 0, MOPT_GTE0}, 1964 {Opt_inode_readahead_blks, 0, MOPT_GTE0}, 1965 {Opt_init_itable, 0, MOPT_GTE0}, 1966 {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP}, 1967 {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS, 1968 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1969 {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE, 1970 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1971 {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER, 1972 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP}, 1973 {Opt_stripe, 0, MOPT_GTE0}, 1974 {Opt_resuid, 0, MOPT_GTE0}, 1975 {Opt_resgid, 0, MOPT_GTE0}, 1976 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1977 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING}, 1978 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1979 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1980 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1981 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, 1982 MOPT_NO_EXT2 | MOPT_DATAJ}, 1983 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1984 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR}, 1985 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1986 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1987 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR}, 1988 #else 1989 {Opt_acl, 0, MOPT_NOSUPPORT}, 1990 {Opt_noacl, 0, MOPT_NOSUPPORT}, 1991 #endif 1992 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1993 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1994 {Opt_debug_want_extra_isize, 0, MOPT_GTE0}, 1995 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1996 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1997 MOPT_SET | MOPT_Q}, 1998 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1999 MOPT_SET | MOPT_Q}, 2000 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 2001 MOPT_SET | MOPT_Q}, 2002 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2003 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 2004 MOPT_CLEAR | MOPT_Q}, 2005 {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING}, 2006 {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING}, 2007 {Opt_offusrjquota, 0, MOPT_Q}, 2008 {Opt_offgrpjquota, 0, MOPT_Q}, 2009 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT}, 2010 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT}, 2011 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT}, 2012 {Opt_max_dir_size_kb, 0, MOPT_GTE0}, 2013 {Opt_test_dummy_encryption, 0, MOPT_STRING}, 2014 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 2015 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 2016 MOPT_SET}, 2017 {Opt_mb_optimize_scan, EXT4_MOUNT2_MB_OPTIMIZE_SCAN, MOPT_GTE0}, 2018 #ifdef CONFIG_EXT4_DEBUG 2019 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 2020 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 2021 {Opt_fc_debug_max_replay, 0, MOPT_GTE0}, 2022 #endif 2023 {Opt_err, 0, 0} 2024 }; 2025 2026 #ifdef CONFIG_UNICODE 2027 static const struct ext4_sb_encodings { 2028 __u16 magic; 2029 char *name; 2030 char *version; 2031 } ext4_sb_encoding_map[] = { 2032 {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"}, 2033 }; 2034 2035 static int ext4_sb_read_encoding(const struct ext4_super_block *es, 2036 const struct ext4_sb_encodings **encoding, 2037 __u16 *flags) 2038 { 2039 __u16 magic = le16_to_cpu(es->s_encoding); 2040 int i; 2041 2042 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 2043 if (magic == ext4_sb_encoding_map[i].magic) 2044 break; 2045 2046 if (i >= ARRAY_SIZE(ext4_sb_encoding_map)) 2047 return -EINVAL; 2048 2049 *encoding = &ext4_sb_encoding_map[i]; 2050 *flags = le16_to_cpu(es->s_encoding_flags); 2051 2052 return 0; 2053 } 2054 #endif 2055 2056 static int ext4_set_test_dummy_encryption(struct super_block *sb, 2057 const char *opt, 2058 const substring_t *arg, 2059 bool is_remount) 2060 { 2061 #ifdef CONFIG_FS_ENCRYPTION 2062 struct ext4_sb_info *sbi = EXT4_SB(sb); 2063 int err; 2064 2065 /* 2066 * This mount option is just for testing, and it's not worthwhile to 2067 * implement the extra complexity (e.g. RCU protection) that would be 2068 * needed to allow it to be set or changed during remount. We do allow 2069 * it to be specified during remount, but only if there is no change. 2070 */ 2071 if (is_remount && !sbi->s_dummy_enc_policy.policy) { 2072 ext4_msg(sb, KERN_WARNING, 2073 "Can't set test_dummy_encryption on remount"); 2074 return -1; 2075 } 2076 err = fscrypt_set_test_dummy_encryption(sb, arg->from, 2077 &sbi->s_dummy_enc_policy); 2078 if (err) { 2079 if (err == -EEXIST) 2080 ext4_msg(sb, KERN_WARNING, 2081 "Can't change test_dummy_encryption on remount"); 2082 else if (err == -EINVAL) 2083 ext4_msg(sb, KERN_WARNING, 2084 "Value of option \"%s\" is unrecognized", opt); 2085 else 2086 ext4_msg(sb, KERN_WARNING, 2087 "Error processing option \"%s\" [%d]", 2088 opt, err); 2089 return -1; 2090 } 2091 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2092 #else 2093 ext4_msg(sb, KERN_WARNING, 2094 "Test dummy encryption mount option ignored"); 2095 #endif 2096 return 1; 2097 } 2098 2099 struct ext4_parsed_options { 2100 unsigned long journal_devnum; 2101 unsigned int journal_ioprio; 2102 int mb_optimize_scan; 2103 }; 2104 2105 static int handle_mount_opt(struct super_block *sb, char *opt, int token, 2106 substring_t *args, struct ext4_parsed_options *parsed_opts, 2107 int is_remount) 2108 { 2109 struct ext4_sb_info *sbi = EXT4_SB(sb); 2110 const struct mount_opts *m; 2111 kuid_t uid; 2112 kgid_t gid; 2113 int arg = 0; 2114 2115 #ifdef CONFIG_QUOTA 2116 if (token == Opt_usrjquota) 2117 return set_qf_name(sb, USRQUOTA, &args[0]); 2118 else if (token == Opt_grpjquota) 2119 return set_qf_name(sb, GRPQUOTA, &args[0]); 2120 else if (token == Opt_offusrjquota) 2121 return clear_qf_name(sb, USRQUOTA); 2122 else if (token == Opt_offgrpjquota) 2123 return clear_qf_name(sb, GRPQUOTA); 2124 #endif 2125 switch (token) { 2126 case Opt_noacl: 2127 case Opt_nouser_xattr: 2128 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5"); 2129 break; 2130 case Opt_sb: 2131 return 1; /* handled by get_sb_block() */ 2132 case Opt_removed: 2133 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt); 2134 return 1; 2135 case Opt_abort: 2136 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 2137 return 1; 2138 case Opt_i_version: 2139 sb->s_flags |= SB_I_VERSION; 2140 return 1; 2141 case Opt_lazytime: 2142 sb->s_flags |= SB_LAZYTIME; 2143 return 1; 2144 case Opt_nolazytime: 2145 sb->s_flags &= ~SB_LAZYTIME; 2146 return 1; 2147 case Opt_inlinecrypt: 2148 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2149 sb->s_flags |= SB_INLINECRYPT; 2150 #else 2151 ext4_msg(sb, KERN_ERR, "inline encryption not supported"); 2152 #endif 2153 return 1; 2154 } 2155 2156 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2157 if (token == m->token) 2158 break; 2159 2160 if (m->token == Opt_err) { 2161 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" " 2162 "or missing value", opt); 2163 return -1; 2164 } 2165 2166 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2167 ext4_msg(sb, KERN_ERR, 2168 "Mount option \"%s\" incompatible with ext2", opt); 2169 return -1; 2170 } 2171 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2172 ext4_msg(sb, KERN_ERR, 2173 "Mount option \"%s\" incompatible with ext3", opt); 2174 return -1; 2175 } 2176 2177 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg)) 2178 return -1; 2179 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0)) 2180 return -1; 2181 if (m->flags & MOPT_EXPLICIT) { 2182 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2183 set_opt2(sb, EXPLICIT_DELALLOC); 2184 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2185 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM); 2186 } else 2187 return -1; 2188 } 2189 if (m->flags & MOPT_CLEAR_ERR) 2190 clear_opt(sb, ERRORS_MASK); 2191 if (token == Opt_noquota && sb_any_quota_loaded(sb)) { 2192 ext4_msg(sb, KERN_ERR, "Cannot change quota " 2193 "options when quota turned on"); 2194 return -1; 2195 } 2196 2197 if (m->flags & MOPT_NOSUPPORT) { 2198 ext4_msg(sb, KERN_ERR, "%s option not supported", opt); 2199 } else if (token == Opt_commit) { 2200 if (arg == 0) 2201 arg = JBD2_DEFAULT_MAX_COMMIT_AGE; 2202 else if (arg > INT_MAX / HZ) { 2203 ext4_msg(sb, KERN_ERR, 2204 "Invalid commit interval %d, " 2205 "must be smaller than %d", 2206 arg, INT_MAX / HZ); 2207 return -1; 2208 } 2209 sbi->s_commit_interval = HZ * arg; 2210 } else if (token == Opt_debug_want_extra_isize) { 2211 if ((arg & 1) || 2212 (arg < 4) || 2213 (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) { 2214 ext4_msg(sb, KERN_ERR, 2215 "Invalid want_extra_isize %d", arg); 2216 return -1; 2217 } 2218 sbi->s_want_extra_isize = arg; 2219 } else if (token == Opt_max_batch_time) { 2220 sbi->s_max_batch_time = arg; 2221 } else if (token == Opt_min_batch_time) { 2222 sbi->s_min_batch_time = arg; 2223 } else if (token == Opt_inode_readahead_blks) { 2224 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) { 2225 ext4_msg(sb, KERN_ERR, 2226 "EXT4-fs: inode_readahead_blks must be " 2227 "0 or a power of 2 smaller than 2^31"); 2228 return -1; 2229 } 2230 sbi->s_inode_readahead_blks = arg; 2231 } else if (token == Opt_init_itable) { 2232 set_opt(sb, INIT_INODE_TABLE); 2233 if (!args->from) 2234 arg = EXT4_DEF_LI_WAIT_MULT; 2235 sbi->s_li_wait_mult = arg; 2236 } else if (token == Opt_max_dir_size_kb) { 2237 sbi->s_max_dir_size_kb = arg; 2238 #ifdef CONFIG_EXT4_DEBUG 2239 } else if (token == Opt_fc_debug_max_replay) { 2240 sbi->s_fc_debug_max_replay = arg; 2241 #endif 2242 } else if (token == Opt_stripe) { 2243 sbi->s_stripe = arg; 2244 } else if (token == Opt_resuid) { 2245 uid = make_kuid(current_user_ns(), arg); 2246 if (!uid_valid(uid)) { 2247 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg); 2248 return -1; 2249 } 2250 sbi->s_resuid = uid; 2251 } else if (token == Opt_resgid) { 2252 gid = make_kgid(current_user_ns(), arg); 2253 if (!gid_valid(gid)) { 2254 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg); 2255 return -1; 2256 } 2257 sbi->s_resgid = gid; 2258 } else if (token == Opt_journal_dev) { 2259 if (is_remount) { 2260 ext4_msg(sb, KERN_ERR, 2261 "Cannot specify journal on remount"); 2262 return -1; 2263 } 2264 parsed_opts->journal_devnum = arg; 2265 } else if (token == Opt_journal_path) { 2266 char *journal_path; 2267 struct inode *journal_inode; 2268 struct path path; 2269 int error; 2270 2271 if (is_remount) { 2272 ext4_msg(sb, KERN_ERR, 2273 "Cannot specify journal on remount"); 2274 return -1; 2275 } 2276 journal_path = match_strdup(&args[0]); 2277 if (!journal_path) { 2278 ext4_msg(sb, KERN_ERR, "error: could not dup " 2279 "journal device string"); 2280 return -1; 2281 } 2282 2283 error = kern_path(journal_path, LOOKUP_FOLLOW, &path); 2284 if (error) { 2285 ext4_msg(sb, KERN_ERR, "error: could not find " 2286 "journal device path: error %d", error); 2287 kfree(journal_path); 2288 return -1; 2289 } 2290 2291 journal_inode = d_inode(path.dentry); 2292 if (!S_ISBLK(journal_inode->i_mode)) { 2293 ext4_msg(sb, KERN_ERR, "error: journal path %s " 2294 "is not a block device", journal_path); 2295 path_put(&path); 2296 kfree(journal_path); 2297 return -1; 2298 } 2299 2300 parsed_opts->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2301 path_put(&path); 2302 kfree(journal_path); 2303 } else if (token == Opt_journal_ioprio) { 2304 if (arg > 7) { 2305 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority" 2306 " (must be 0-7)"); 2307 return -1; 2308 } 2309 parsed_opts->journal_ioprio = 2310 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg); 2311 } else if (token == Opt_test_dummy_encryption) { 2312 return ext4_set_test_dummy_encryption(sb, opt, &args[0], 2313 is_remount); 2314 } else if (m->flags & MOPT_DATAJ) { 2315 if (is_remount) { 2316 if (!sbi->s_journal) 2317 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option"); 2318 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) { 2319 ext4_msg(sb, KERN_ERR, 2320 "Cannot change data mode on remount"); 2321 return -1; 2322 } 2323 } else { 2324 clear_opt(sb, DATA_FLAGS); 2325 sbi->s_mount_opt |= m->mount_opt; 2326 } 2327 #ifdef CONFIG_QUOTA 2328 } else if (m->flags & MOPT_QFMT) { 2329 if (sb_any_quota_loaded(sb) && 2330 sbi->s_jquota_fmt != m->mount_opt) { 2331 ext4_msg(sb, KERN_ERR, "Cannot change journaled " 2332 "quota options when quota turned on"); 2333 return -1; 2334 } 2335 if (ext4_has_feature_quota(sb)) { 2336 ext4_msg(sb, KERN_INFO, 2337 "Quota format mount options ignored " 2338 "when QUOTA feature is enabled"); 2339 return 1; 2340 } 2341 sbi->s_jquota_fmt = m->mount_opt; 2342 #endif 2343 } else if (token == Opt_dax || token == Opt_dax_always || 2344 token == Opt_dax_inode || token == Opt_dax_never) { 2345 #ifdef CONFIG_FS_DAX 2346 switch (token) { 2347 case Opt_dax: 2348 case Opt_dax_always: 2349 if (is_remount && 2350 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2351 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2352 fail_dax_change_remount: 2353 ext4_msg(sb, KERN_ERR, "can't change " 2354 "dax mount option while remounting"); 2355 return -1; 2356 } 2357 if (is_remount && 2358 (test_opt(sb, DATA_FLAGS) == 2359 EXT4_MOUNT_JOURNAL_DATA)) { 2360 ext4_msg(sb, KERN_ERR, "can't mount with " 2361 "both data=journal and dax"); 2362 return -1; 2363 } 2364 ext4_msg(sb, KERN_WARNING, 2365 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 2366 sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS; 2367 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2368 break; 2369 case Opt_dax_never: 2370 if (is_remount && 2371 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2372 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) 2373 goto fail_dax_change_remount; 2374 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2375 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2376 break; 2377 case Opt_dax_inode: 2378 if (is_remount && 2379 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2380 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2381 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) 2382 goto fail_dax_change_remount; 2383 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2384 sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER; 2385 /* Strictly for printing options */ 2386 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE; 2387 break; 2388 } 2389 #else 2390 ext4_msg(sb, KERN_INFO, "dax option not supported"); 2391 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER; 2392 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS; 2393 return -1; 2394 #endif 2395 } else if (token == Opt_data_err_abort) { 2396 sbi->s_mount_opt |= m->mount_opt; 2397 } else if (token == Opt_data_err_ignore) { 2398 sbi->s_mount_opt &= ~m->mount_opt; 2399 } else if (token == Opt_mb_optimize_scan) { 2400 if (arg != 0 && arg != 1) { 2401 ext4_msg(sb, KERN_WARNING, 2402 "mb_optimize_scan should be set to 0 or 1."); 2403 return -1; 2404 } 2405 parsed_opts->mb_optimize_scan = arg; 2406 } else { 2407 if (!args->from) 2408 arg = 1; 2409 if (m->flags & MOPT_CLEAR) 2410 arg = !arg; 2411 else if (unlikely(!(m->flags & MOPT_SET))) { 2412 ext4_msg(sb, KERN_WARNING, 2413 "buggy handling of option %s", opt); 2414 WARN_ON(1); 2415 return -1; 2416 } 2417 if (m->flags & MOPT_2) { 2418 if (arg != 0) 2419 sbi->s_mount_opt2 |= m->mount_opt; 2420 else 2421 sbi->s_mount_opt2 &= ~m->mount_opt; 2422 } else { 2423 if (arg != 0) 2424 sbi->s_mount_opt |= m->mount_opt; 2425 else 2426 sbi->s_mount_opt &= ~m->mount_opt; 2427 } 2428 } 2429 return 1; 2430 } 2431 2432 static int parse_options(char *options, struct super_block *sb, 2433 struct ext4_parsed_options *ret_opts, 2434 int is_remount) 2435 { 2436 struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb); 2437 char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name; 2438 substring_t args[MAX_OPT_ARGS]; 2439 int token; 2440 2441 if (!options) 2442 return 1; 2443 2444 while ((p = strsep(&options, ",")) != NULL) { 2445 if (!*p) 2446 continue; 2447 /* 2448 * Initialize args struct so we know whether arg was 2449 * found; some options take optional arguments. 2450 */ 2451 args[0].to = args[0].from = NULL; 2452 token = match_token(p, tokens, args); 2453 if (handle_mount_opt(sb, p, token, args, ret_opts, 2454 is_remount) < 0) 2455 return 0; 2456 } 2457 #ifdef CONFIG_QUOTA 2458 /* 2459 * We do the test below only for project quotas. 'usrquota' and 2460 * 'grpquota' mount options are allowed even without quota feature 2461 * to support legacy quotas in quota files. 2462 */ 2463 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) { 2464 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. " 2465 "Cannot enable project quota enforcement."); 2466 return 0; 2467 } 2468 usr_qf_name = get_qf_name(sb, sbi, USRQUOTA); 2469 grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA); 2470 if (usr_qf_name || grp_qf_name) { 2471 if (test_opt(sb, USRQUOTA) && usr_qf_name) 2472 clear_opt(sb, USRQUOTA); 2473 2474 if (test_opt(sb, GRPQUOTA) && grp_qf_name) 2475 clear_opt(sb, GRPQUOTA); 2476 2477 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) { 2478 ext4_msg(sb, KERN_ERR, "old and new quota " 2479 "format mixing"); 2480 return 0; 2481 } 2482 2483 if (!sbi->s_jquota_fmt) { 2484 ext4_msg(sb, KERN_ERR, "journaled quota format " 2485 "not specified"); 2486 return 0; 2487 } 2488 } 2489 #endif 2490 if (test_opt(sb, DIOREAD_NOLOCK)) { 2491 int blocksize = 2492 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2493 if (blocksize < PAGE_SIZE) 2494 ext4_msg(sb, KERN_WARNING, "Warning: mounting with an " 2495 "experimental mount option 'dioread_nolock' " 2496 "for blocksize < PAGE_SIZE"); 2497 } 2498 return 1; 2499 } 2500 2501 static inline void ext4_show_quota_options(struct seq_file *seq, 2502 struct super_block *sb) 2503 { 2504 #if defined(CONFIG_QUOTA) 2505 struct ext4_sb_info *sbi = EXT4_SB(sb); 2506 char *usr_qf_name, *grp_qf_name; 2507 2508 if (sbi->s_jquota_fmt) { 2509 char *fmtname = ""; 2510 2511 switch (sbi->s_jquota_fmt) { 2512 case QFMT_VFS_OLD: 2513 fmtname = "vfsold"; 2514 break; 2515 case QFMT_VFS_V0: 2516 fmtname = "vfsv0"; 2517 break; 2518 case QFMT_VFS_V1: 2519 fmtname = "vfsv1"; 2520 break; 2521 } 2522 seq_printf(seq, ",jqfmt=%s", fmtname); 2523 } 2524 2525 rcu_read_lock(); 2526 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2527 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2528 if (usr_qf_name) 2529 seq_show_option(seq, "usrjquota", usr_qf_name); 2530 if (grp_qf_name) 2531 seq_show_option(seq, "grpjquota", grp_qf_name); 2532 rcu_read_unlock(); 2533 #endif 2534 } 2535 2536 static const char *token2str(int token) 2537 { 2538 const struct match_token *t; 2539 2540 for (t = tokens; t->token != Opt_err; t++) 2541 if (t->token == token && !strchr(t->pattern, '=')) 2542 break; 2543 return t->pattern; 2544 } 2545 2546 /* 2547 * Show an option if 2548 * - it's set to a non-default value OR 2549 * - if the per-sb default is different from the global default 2550 */ 2551 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2552 int nodefs) 2553 { 2554 struct ext4_sb_info *sbi = EXT4_SB(sb); 2555 struct ext4_super_block *es = sbi->s_es; 2556 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2557 const struct mount_opts *m; 2558 char sep = nodefs ? '\n' : ','; 2559 2560 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2561 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2562 2563 if (sbi->s_sb_block != 1) 2564 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2565 2566 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2567 int want_set = m->flags & MOPT_SET; 2568 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2569 (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP) 2570 continue; 2571 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2572 continue; /* skip if same as the default */ 2573 if ((want_set && 2574 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2575 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2576 continue; /* select Opt_noFoo vs Opt_Foo */ 2577 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2578 } 2579 2580 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2581 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2582 SEQ_OPTS_PRINT("resuid=%u", 2583 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2584 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2585 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2586 SEQ_OPTS_PRINT("resgid=%u", 2587 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2588 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2589 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2590 SEQ_OPTS_PUTS("errors=remount-ro"); 2591 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2592 SEQ_OPTS_PUTS("errors=continue"); 2593 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2594 SEQ_OPTS_PUTS("errors=panic"); 2595 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2596 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2597 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2598 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2599 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2600 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2601 if (sb->s_flags & SB_I_VERSION) 2602 SEQ_OPTS_PUTS("i_version"); 2603 if (nodefs || sbi->s_stripe) 2604 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2605 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2606 (sbi->s_mount_opt ^ def_mount_opt)) { 2607 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2608 SEQ_OPTS_PUTS("data=journal"); 2609 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2610 SEQ_OPTS_PUTS("data=ordered"); 2611 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2612 SEQ_OPTS_PUTS("data=writeback"); 2613 } 2614 if (nodefs || 2615 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2616 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2617 sbi->s_inode_readahead_blks); 2618 2619 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2620 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2621 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2622 if (nodefs || sbi->s_max_dir_size_kb) 2623 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2624 if (test_opt(sb, DATA_ERR_ABORT)) 2625 SEQ_OPTS_PUTS("data_err=abort"); 2626 2627 fscrypt_show_test_dummy_encryption(seq, sep, sb); 2628 2629 if (sb->s_flags & SB_INLINECRYPT) 2630 SEQ_OPTS_PUTS("inlinecrypt"); 2631 2632 if (test_opt(sb, DAX_ALWAYS)) { 2633 if (IS_EXT2_SB(sb)) 2634 SEQ_OPTS_PUTS("dax"); 2635 else 2636 SEQ_OPTS_PUTS("dax=always"); 2637 } else if (test_opt2(sb, DAX_NEVER)) { 2638 SEQ_OPTS_PUTS("dax=never"); 2639 } else if (test_opt2(sb, DAX_INODE)) { 2640 SEQ_OPTS_PUTS("dax=inode"); 2641 } 2642 ext4_show_quota_options(seq, sb); 2643 return 0; 2644 } 2645 2646 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2647 { 2648 return _ext4_show_options(seq, root->d_sb, 0); 2649 } 2650 2651 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2652 { 2653 struct super_block *sb = seq->private; 2654 int rc; 2655 2656 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2657 rc = _ext4_show_options(seq, sb, 1); 2658 seq_puts(seq, "\n"); 2659 return rc; 2660 } 2661 2662 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2663 int read_only) 2664 { 2665 struct ext4_sb_info *sbi = EXT4_SB(sb); 2666 int err = 0; 2667 2668 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2669 ext4_msg(sb, KERN_ERR, "revision level too high, " 2670 "forcing read-only mode"); 2671 err = -EROFS; 2672 goto done; 2673 } 2674 if (read_only) 2675 goto done; 2676 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2677 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2678 "running e2fsck is recommended"); 2679 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2680 ext4_msg(sb, KERN_WARNING, 2681 "warning: mounting fs with errors, " 2682 "running e2fsck is recommended"); 2683 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2684 le16_to_cpu(es->s_mnt_count) >= 2685 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2686 ext4_msg(sb, KERN_WARNING, 2687 "warning: maximal mount count reached, " 2688 "running e2fsck is recommended"); 2689 else if (le32_to_cpu(es->s_checkinterval) && 2690 (ext4_get_tstamp(es, s_lastcheck) + 2691 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2692 ext4_msg(sb, KERN_WARNING, 2693 "warning: checktime reached, " 2694 "running e2fsck is recommended"); 2695 if (!sbi->s_journal) 2696 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2697 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2698 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2699 le16_add_cpu(&es->s_mnt_count, 1); 2700 ext4_update_tstamp(es, s_mtime); 2701 if (sbi->s_journal) 2702 ext4_set_feature_journal_needs_recovery(sb); 2703 2704 err = ext4_commit_super(sb); 2705 done: 2706 if (test_opt(sb, DEBUG)) 2707 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2708 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2709 sb->s_blocksize, 2710 sbi->s_groups_count, 2711 EXT4_BLOCKS_PER_GROUP(sb), 2712 EXT4_INODES_PER_GROUP(sb), 2713 sbi->s_mount_opt, sbi->s_mount_opt2); 2714 2715 cleancache_init_fs(sb); 2716 return err; 2717 } 2718 2719 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2720 { 2721 struct ext4_sb_info *sbi = EXT4_SB(sb); 2722 struct flex_groups **old_groups, **new_groups; 2723 int size, i, j; 2724 2725 if (!sbi->s_log_groups_per_flex) 2726 return 0; 2727 2728 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2729 if (size <= sbi->s_flex_groups_allocated) 2730 return 0; 2731 2732 new_groups = kvzalloc(roundup_pow_of_two(size * 2733 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 2734 if (!new_groups) { 2735 ext4_msg(sb, KERN_ERR, 2736 "not enough memory for %d flex group pointers", size); 2737 return -ENOMEM; 2738 } 2739 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 2740 new_groups[i] = kvzalloc(roundup_pow_of_two( 2741 sizeof(struct flex_groups)), 2742 GFP_KERNEL); 2743 if (!new_groups[i]) { 2744 for (j = sbi->s_flex_groups_allocated; j < i; j++) 2745 kvfree(new_groups[j]); 2746 kvfree(new_groups); 2747 ext4_msg(sb, KERN_ERR, 2748 "not enough memory for %d flex groups", size); 2749 return -ENOMEM; 2750 } 2751 } 2752 rcu_read_lock(); 2753 old_groups = rcu_dereference(sbi->s_flex_groups); 2754 if (old_groups) 2755 memcpy(new_groups, old_groups, 2756 (sbi->s_flex_groups_allocated * 2757 sizeof(struct flex_groups *))); 2758 rcu_read_unlock(); 2759 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 2760 sbi->s_flex_groups_allocated = size; 2761 if (old_groups) 2762 ext4_kvfree_array_rcu(old_groups); 2763 return 0; 2764 } 2765 2766 static int ext4_fill_flex_info(struct super_block *sb) 2767 { 2768 struct ext4_sb_info *sbi = EXT4_SB(sb); 2769 struct ext4_group_desc *gdp = NULL; 2770 struct flex_groups *fg; 2771 ext4_group_t flex_group; 2772 int i, err; 2773 2774 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2775 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2776 sbi->s_log_groups_per_flex = 0; 2777 return 1; 2778 } 2779 2780 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2781 if (err) 2782 goto failed; 2783 2784 for (i = 0; i < sbi->s_groups_count; i++) { 2785 gdp = ext4_get_group_desc(sb, i, NULL); 2786 2787 flex_group = ext4_flex_group(sbi, i); 2788 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 2789 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 2790 atomic64_add(ext4_free_group_clusters(sb, gdp), 2791 &fg->free_clusters); 2792 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 2793 } 2794 2795 return 1; 2796 failed: 2797 return 0; 2798 } 2799 2800 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2801 struct ext4_group_desc *gdp) 2802 { 2803 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2804 __u16 crc = 0; 2805 __le32 le_group = cpu_to_le32(block_group); 2806 struct ext4_sb_info *sbi = EXT4_SB(sb); 2807 2808 if (ext4_has_metadata_csum(sbi->s_sb)) { 2809 /* Use new metadata_csum algorithm */ 2810 __u32 csum32; 2811 __u16 dummy_csum = 0; 2812 2813 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2814 sizeof(le_group)); 2815 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2816 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2817 sizeof(dummy_csum)); 2818 offset += sizeof(dummy_csum); 2819 if (offset < sbi->s_desc_size) 2820 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2821 sbi->s_desc_size - offset); 2822 2823 crc = csum32 & 0xFFFF; 2824 goto out; 2825 } 2826 2827 /* old crc16 code */ 2828 if (!ext4_has_feature_gdt_csum(sb)) 2829 return 0; 2830 2831 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2832 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2833 crc = crc16(crc, (__u8 *)gdp, offset); 2834 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2835 /* for checksum of struct ext4_group_desc do the rest...*/ 2836 if (ext4_has_feature_64bit(sb) && 2837 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2838 crc = crc16(crc, (__u8 *)gdp + offset, 2839 le16_to_cpu(sbi->s_es->s_desc_size) - 2840 offset); 2841 2842 out: 2843 return cpu_to_le16(crc); 2844 } 2845 2846 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2847 struct ext4_group_desc *gdp) 2848 { 2849 if (ext4_has_group_desc_csum(sb) && 2850 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2851 return 0; 2852 2853 return 1; 2854 } 2855 2856 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2857 struct ext4_group_desc *gdp) 2858 { 2859 if (!ext4_has_group_desc_csum(sb)) 2860 return; 2861 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2862 } 2863 2864 /* Called at mount-time, super-block is locked */ 2865 static int ext4_check_descriptors(struct super_block *sb, 2866 ext4_fsblk_t sb_block, 2867 ext4_group_t *first_not_zeroed) 2868 { 2869 struct ext4_sb_info *sbi = EXT4_SB(sb); 2870 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2871 ext4_fsblk_t last_block; 2872 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2873 ext4_fsblk_t block_bitmap; 2874 ext4_fsblk_t inode_bitmap; 2875 ext4_fsblk_t inode_table; 2876 int flexbg_flag = 0; 2877 ext4_group_t i, grp = sbi->s_groups_count; 2878 2879 if (ext4_has_feature_flex_bg(sb)) 2880 flexbg_flag = 1; 2881 2882 ext4_debug("Checking group descriptors"); 2883 2884 for (i = 0; i < sbi->s_groups_count; i++) { 2885 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2886 2887 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2888 last_block = ext4_blocks_count(sbi->s_es) - 1; 2889 else 2890 last_block = first_block + 2891 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2892 2893 if ((grp == sbi->s_groups_count) && 2894 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2895 grp = i; 2896 2897 block_bitmap = ext4_block_bitmap(sb, gdp); 2898 if (block_bitmap == sb_block) { 2899 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2900 "Block bitmap for group %u overlaps " 2901 "superblock", i); 2902 if (!sb_rdonly(sb)) 2903 return 0; 2904 } 2905 if (block_bitmap >= sb_block + 1 && 2906 block_bitmap <= last_bg_block) { 2907 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2908 "Block bitmap for group %u overlaps " 2909 "block group descriptors", i); 2910 if (!sb_rdonly(sb)) 2911 return 0; 2912 } 2913 if (block_bitmap < first_block || block_bitmap > last_block) { 2914 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2915 "Block bitmap for group %u not in group " 2916 "(block %llu)!", i, block_bitmap); 2917 return 0; 2918 } 2919 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2920 if (inode_bitmap == sb_block) { 2921 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2922 "Inode bitmap for group %u overlaps " 2923 "superblock", i); 2924 if (!sb_rdonly(sb)) 2925 return 0; 2926 } 2927 if (inode_bitmap >= sb_block + 1 && 2928 inode_bitmap <= last_bg_block) { 2929 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2930 "Inode bitmap for group %u overlaps " 2931 "block group descriptors", i); 2932 if (!sb_rdonly(sb)) 2933 return 0; 2934 } 2935 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2936 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2937 "Inode bitmap for group %u not in group " 2938 "(block %llu)!", i, inode_bitmap); 2939 return 0; 2940 } 2941 inode_table = ext4_inode_table(sb, gdp); 2942 if (inode_table == sb_block) { 2943 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2944 "Inode table for group %u overlaps " 2945 "superblock", i); 2946 if (!sb_rdonly(sb)) 2947 return 0; 2948 } 2949 if (inode_table >= sb_block + 1 && 2950 inode_table <= last_bg_block) { 2951 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2952 "Inode table for group %u overlaps " 2953 "block group descriptors", i); 2954 if (!sb_rdonly(sb)) 2955 return 0; 2956 } 2957 if (inode_table < first_block || 2958 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2959 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2960 "Inode table for group %u not in group " 2961 "(block %llu)!", i, inode_table); 2962 return 0; 2963 } 2964 ext4_lock_group(sb, i); 2965 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2966 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2967 "Checksum for group %u failed (%u!=%u)", 2968 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2969 gdp)), le16_to_cpu(gdp->bg_checksum)); 2970 if (!sb_rdonly(sb)) { 2971 ext4_unlock_group(sb, i); 2972 return 0; 2973 } 2974 } 2975 ext4_unlock_group(sb, i); 2976 if (!flexbg_flag) 2977 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2978 } 2979 if (NULL != first_not_zeroed) 2980 *first_not_zeroed = grp; 2981 return 1; 2982 } 2983 2984 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2985 * the superblock) which were deleted from all directories, but held open by 2986 * a process at the time of a crash. We walk the list and try to delete these 2987 * inodes at recovery time (only with a read-write filesystem). 2988 * 2989 * In order to keep the orphan inode chain consistent during traversal (in 2990 * case of crash during recovery), we link each inode into the superblock 2991 * orphan list_head and handle it the same way as an inode deletion during 2992 * normal operation (which journals the operations for us). 2993 * 2994 * We only do an iget() and an iput() on each inode, which is very safe if we 2995 * accidentally point at an in-use or already deleted inode. The worst that 2996 * can happen in this case is that we get a "bit already cleared" message from 2997 * ext4_free_inode(). The only reason we would point at a wrong inode is if 2998 * e2fsck was run on this filesystem, and it must have already done the orphan 2999 * inode cleanup for us, so we can safely abort without any further action. 3000 */ 3001 static void ext4_orphan_cleanup(struct super_block *sb, 3002 struct ext4_super_block *es) 3003 { 3004 unsigned int s_flags = sb->s_flags; 3005 int ret, nr_orphans = 0, nr_truncates = 0; 3006 #ifdef CONFIG_QUOTA 3007 int quota_update = 0; 3008 int i; 3009 #endif 3010 if (!es->s_last_orphan) { 3011 jbd_debug(4, "no orphan inodes to clean up\n"); 3012 return; 3013 } 3014 3015 if (bdev_read_only(sb->s_bdev)) { 3016 ext4_msg(sb, KERN_ERR, "write access " 3017 "unavailable, skipping orphan cleanup"); 3018 return; 3019 } 3020 3021 /* Check if feature set would not allow a r/w mount */ 3022 if (!ext4_feature_set_ok(sb, 0)) { 3023 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 3024 "unknown ROCOMPAT features"); 3025 return; 3026 } 3027 3028 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3029 /* don't clear list on RO mount w/ errors */ 3030 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 3031 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 3032 "clearing orphan list.\n"); 3033 es->s_last_orphan = 0; 3034 } 3035 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3036 return; 3037 } 3038 3039 if (s_flags & SB_RDONLY) { 3040 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 3041 sb->s_flags &= ~SB_RDONLY; 3042 } 3043 #ifdef CONFIG_QUOTA 3044 /* 3045 * Turn on quotas which were not enabled for read-only mounts if 3046 * filesystem has quota feature, so that they are updated correctly. 3047 */ 3048 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 3049 int ret = ext4_enable_quotas(sb); 3050 3051 if (!ret) 3052 quota_update = 1; 3053 else 3054 ext4_msg(sb, KERN_ERR, 3055 "Cannot turn on quotas: error %d", ret); 3056 } 3057 3058 /* Turn on journaled quotas used for old sytle */ 3059 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3060 if (EXT4_SB(sb)->s_qf_names[i]) { 3061 int ret = ext4_quota_on_mount(sb, i); 3062 3063 if (!ret) 3064 quota_update = 1; 3065 else 3066 ext4_msg(sb, KERN_ERR, 3067 "Cannot turn on journaled " 3068 "quota: type %d: error %d", i, ret); 3069 } 3070 } 3071 #endif 3072 3073 while (es->s_last_orphan) { 3074 struct inode *inode; 3075 3076 /* 3077 * We may have encountered an error during cleanup; if 3078 * so, skip the rest. 3079 */ 3080 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 3081 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 3082 es->s_last_orphan = 0; 3083 break; 3084 } 3085 3086 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 3087 if (IS_ERR(inode)) { 3088 es->s_last_orphan = 0; 3089 break; 3090 } 3091 3092 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 3093 dquot_initialize(inode); 3094 if (inode->i_nlink) { 3095 if (test_opt(sb, DEBUG)) 3096 ext4_msg(sb, KERN_DEBUG, 3097 "%s: truncating inode %lu to %lld bytes", 3098 __func__, inode->i_ino, inode->i_size); 3099 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 3100 inode->i_ino, inode->i_size); 3101 inode_lock(inode); 3102 truncate_inode_pages(inode->i_mapping, inode->i_size); 3103 ret = ext4_truncate(inode); 3104 if (ret) 3105 ext4_std_error(inode->i_sb, ret); 3106 inode_unlock(inode); 3107 nr_truncates++; 3108 } else { 3109 if (test_opt(sb, DEBUG)) 3110 ext4_msg(sb, KERN_DEBUG, 3111 "%s: deleting unreferenced inode %lu", 3112 __func__, inode->i_ino); 3113 jbd_debug(2, "deleting unreferenced inode %lu\n", 3114 inode->i_ino); 3115 nr_orphans++; 3116 } 3117 iput(inode); /* The delete magic happens here! */ 3118 } 3119 3120 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 3121 3122 if (nr_orphans) 3123 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 3124 PLURAL(nr_orphans)); 3125 if (nr_truncates) 3126 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 3127 PLURAL(nr_truncates)); 3128 #ifdef CONFIG_QUOTA 3129 /* Turn off quotas if they were enabled for orphan cleanup */ 3130 if (quota_update) { 3131 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 3132 if (sb_dqopt(sb)->files[i]) 3133 dquot_quota_off(sb, i); 3134 } 3135 } 3136 #endif 3137 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 3138 } 3139 3140 /* 3141 * Maximal extent format file size. 3142 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3143 * extent format containers, within a sector_t, and within i_blocks 3144 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3145 * so that won't be a limiting factor. 3146 * 3147 * However there is other limiting factor. We do store extents in the form 3148 * of starting block and length, hence the resulting length of the extent 3149 * covering maximum file size must fit into on-disk format containers as 3150 * well. Given that length is always by 1 unit bigger than max unit (because 3151 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3152 * 3153 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3154 */ 3155 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3156 { 3157 loff_t res; 3158 loff_t upper_limit = MAX_LFS_FILESIZE; 3159 3160 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3161 3162 if (!has_huge_files) { 3163 upper_limit = (1LL << 32) - 1; 3164 3165 /* total blocks in file system block size */ 3166 upper_limit >>= (blkbits - 9); 3167 upper_limit <<= blkbits; 3168 } 3169 3170 /* 3171 * 32-bit extent-start container, ee_block. We lower the maxbytes 3172 * by one fs block, so ee_len can cover the extent of maximum file 3173 * size 3174 */ 3175 res = (1LL << 32) - 1; 3176 res <<= blkbits; 3177 3178 /* Sanity check against vm- & vfs- imposed limits */ 3179 if (res > upper_limit) 3180 res = upper_limit; 3181 3182 return res; 3183 } 3184 3185 /* 3186 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3187 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3188 * We need to be 1 filesystem block less than the 2^48 sector limit. 3189 */ 3190 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3191 { 3192 loff_t res = EXT4_NDIR_BLOCKS; 3193 int meta_blocks; 3194 loff_t upper_limit; 3195 /* This is calculated to be the largest file size for a dense, block 3196 * mapped file such that the file's total number of 512-byte sectors, 3197 * including data and all indirect blocks, does not exceed (2^48 - 1). 3198 * 3199 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3200 * number of 512-byte sectors of the file. 3201 */ 3202 3203 if (!has_huge_files) { 3204 /* 3205 * !has_huge_files or implies that the inode i_block field 3206 * represents total file blocks in 2^32 512-byte sectors == 3207 * size of vfs inode i_blocks * 8 3208 */ 3209 upper_limit = (1LL << 32) - 1; 3210 3211 /* total blocks in file system block size */ 3212 upper_limit >>= (bits - 9); 3213 3214 } else { 3215 /* 3216 * We use 48 bit ext4_inode i_blocks 3217 * With EXT4_HUGE_FILE_FL set the i_blocks 3218 * represent total number of blocks in 3219 * file system block size 3220 */ 3221 upper_limit = (1LL << 48) - 1; 3222 3223 } 3224 3225 /* indirect blocks */ 3226 meta_blocks = 1; 3227 /* double indirect blocks */ 3228 meta_blocks += 1 + (1LL << (bits-2)); 3229 /* tripple indirect blocks */ 3230 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 3231 3232 upper_limit -= meta_blocks; 3233 upper_limit <<= bits; 3234 3235 res += 1LL << (bits-2); 3236 res += 1LL << (2*(bits-2)); 3237 res += 1LL << (3*(bits-2)); 3238 res <<= bits; 3239 if (res > upper_limit) 3240 res = upper_limit; 3241 3242 if (res > MAX_LFS_FILESIZE) 3243 res = MAX_LFS_FILESIZE; 3244 3245 return res; 3246 } 3247 3248 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3249 ext4_fsblk_t logical_sb_block, int nr) 3250 { 3251 struct ext4_sb_info *sbi = EXT4_SB(sb); 3252 ext4_group_t bg, first_meta_bg; 3253 int has_super = 0; 3254 3255 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3256 3257 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3258 return logical_sb_block + nr + 1; 3259 bg = sbi->s_desc_per_block * nr; 3260 if (ext4_bg_has_super(sb, bg)) 3261 has_super = 1; 3262 3263 /* 3264 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3265 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3266 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3267 * compensate. 3268 */ 3269 if (sb->s_blocksize == 1024 && nr == 0 && 3270 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3271 has_super++; 3272 3273 return (has_super + ext4_group_first_block_no(sb, bg)); 3274 } 3275 3276 /** 3277 * ext4_get_stripe_size: Get the stripe size. 3278 * @sbi: In memory super block info 3279 * 3280 * If we have specified it via mount option, then 3281 * use the mount option value. If the value specified at mount time is 3282 * greater than the blocks per group use the super block value. 3283 * If the super block value is greater than blocks per group return 0. 3284 * Allocator needs it be less than blocks per group. 3285 * 3286 */ 3287 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3288 { 3289 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3290 unsigned long stripe_width = 3291 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3292 int ret; 3293 3294 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3295 ret = sbi->s_stripe; 3296 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3297 ret = stripe_width; 3298 else if (stride && stride <= sbi->s_blocks_per_group) 3299 ret = stride; 3300 else 3301 ret = 0; 3302 3303 /* 3304 * If the stripe width is 1, this makes no sense and 3305 * we set it to 0 to turn off stripe handling code. 3306 */ 3307 if (ret <= 1) 3308 ret = 0; 3309 3310 return ret; 3311 } 3312 3313 /* 3314 * Check whether this filesystem can be mounted based on 3315 * the features present and the RDONLY/RDWR mount requested. 3316 * Returns 1 if this filesystem can be mounted as requested, 3317 * 0 if it cannot be. 3318 */ 3319 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 3320 { 3321 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3322 ext4_msg(sb, KERN_ERR, 3323 "Couldn't mount because of " 3324 "unsupported optional features (%x)", 3325 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3326 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3327 return 0; 3328 } 3329 3330 #ifndef CONFIG_UNICODE 3331 if (ext4_has_feature_casefold(sb)) { 3332 ext4_msg(sb, KERN_ERR, 3333 "Filesystem with casefold feature cannot be " 3334 "mounted without CONFIG_UNICODE"); 3335 return 0; 3336 } 3337 #endif 3338 3339 if (readonly) 3340 return 1; 3341 3342 if (ext4_has_feature_readonly(sb)) { 3343 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3344 sb->s_flags |= SB_RDONLY; 3345 return 1; 3346 } 3347 3348 /* Check that feature set is OK for a read-write mount */ 3349 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3350 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3351 "unsupported optional features (%x)", 3352 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3353 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3354 return 0; 3355 } 3356 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3357 ext4_msg(sb, KERN_ERR, 3358 "Can't support bigalloc feature without " 3359 "extents feature\n"); 3360 return 0; 3361 } 3362 3363 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3364 if (!readonly && (ext4_has_feature_quota(sb) || 3365 ext4_has_feature_project(sb))) { 3366 ext4_msg(sb, KERN_ERR, 3367 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3368 return 0; 3369 } 3370 #endif /* CONFIG_QUOTA */ 3371 return 1; 3372 } 3373 3374 /* 3375 * This function is called once a day if we have errors logged 3376 * on the file system 3377 */ 3378 static void print_daily_error_info(struct timer_list *t) 3379 { 3380 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3381 struct super_block *sb = sbi->s_sb; 3382 struct ext4_super_block *es = sbi->s_es; 3383 3384 if (es->s_error_count) 3385 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3386 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3387 le32_to_cpu(es->s_error_count)); 3388 if (es->s_first_error_time) { 3389 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3390 sb->s_id, 3391 ext4_get_tstamp(es, s_first_error_time), 3392 (int) sizeof(es->s_first_error_func), 3393 es->s_first_error_func, 3394 le32_to_cpu(es->s_first_error_line)); 3395 if (es->s_first_error_ino) 3396 printk(KERN_CONT ": inode %u", 3397 le32_to_cpu(es->s_first_error_ino)); 3398 if (es->s_first_error_block) 3399 printk(KERN_CONT ": block %llu", (unsigned long long) 3400 le64_to_cpu(es->s_first_error_block)); 3401 printk(KERN_CONT "\n"); 3402 } 3403 if (es->s_last_error_time) { 3404 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3405 sb->s_id, 3406 ext4_get_tstamp(es, s_last_error_time), 3407 (int) sizeof(es->s_last_error_func), 3408 es->s_last_error_func, 3409 le32_to_cpu(es->s_last_error_line)); 3410 if (es->s_last_error_ino) 3411 printk(KERN_CONT ": inode %u", 3412 le32_to_cpu(es->s_last_error_ino)); 3413 if (es->s_last_error_block) 3414 printk(KERN_CONT ": block %llu", (unsigned long long) 3415 le64_to_cpu(es->s_last_error_block)); 3416 printk(KERN_CONT "\n"); 3417 } 3418 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3419 } 3420 3421 /* Find next suitable group and run ext4_init_inode_table */ 3422 static int ext4_run_li_request(struct ext4_li_request *elr) 3423 { 3424 struct ext4_group_desc *gdp = NULL; 3425 struct super_block *sb = elr->lr_super; 3426 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3427 ext4_group_t group = elr->lr_next_group; 3428 unsigned long timeout = 0; 3429 unsigned int prefetch_ios = 0; 3430 int ret = 0; 3431 3432 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3433 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3434 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3435 if (prefetch_ios) 3436 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3437 prefetch_ios); 3438 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3439 prefetch_ios); 3440 if (group >= elr->lr_next_group) { 3441 ret = 1; 3442 if (elr->lr_first_not_zeroed != ngroups && 3443 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3444 elr->lr_next_group = elr->lr_first_not_zeroed; 3445 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3446 ret = 0; 3447 } 3448 } 3449 return ret; 3450 } 3451 3452 for (; group < ngroups; group++) { 3453 gdp = ext4_get_group_desc(sb, group, NULL); 3454 if (!gdp) { 3455 ret = 1; 3456 break; 3457 } 3458 3459 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3460 break; 3461 } 3462 3463 if (group >= ngroups) 3464 ret = 1; 3465 3466 if (!ret) { 3467 timeout = jiffies; 3468 ret = ext4_init_inode_table(sb, group, 3469 elr->lr_timeout ? 0 : 1); 3470 trace_ext4_lazy_itable_init(sb, group); 3471 if (elr->lr_timeout == 0) { 3472 timeout = (jiffies - timeout) * 3473 EXT4_SB(elr->lr_super)->s_li_wait_mult; 3474 elr->lr_timeout = timeout; 3475 } 3476 elr->lr_next_sched = jiffies + elr->lr_timeout; 3477 elr->lr_next_group = group + 1; 3478 } 3479 return ret; 3480 } 3481 3482 /* 3483 * Remove lr_request from the list_request and free the 3484 * request structure. Should be called with li_list_mtx held 3485 */ 3486 static void ext4_remove_li_request(struct ext4_li_request *elr) 3487 { 3488 if (!elr) 3489 return; 3490 3491 list_del(&elr->lr_request); 3492 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3493 kfree(elr); 3494 } 3495 3496 static void ext4_unregister_li_request(struct super_block *sb) 3497 { 3498 mutex_lock(&ext4_li_mtx); 3499 if (!ext4_li_info) { 3500 mutex_unlock(&ext4_li_mtx); 3501 return; 3502 } 3503 3504 mutex_lock(&ext4_li_info->li_list_mtx); 3505 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3506 mutex_unlock(&ext4_li_info->li_list_mtx); 3507 mutex_unlock(&ext4_li_mtx); 3508 } 3509 3510 static struct task_struct *ext4_lazyinit_task; 3511 3512 /* 3513 * This is the function where ext4lazyinit thread lives. It walks 3514 * through the request list searching for next scheduled filesystem. 3515 * When such a fs is found, run the lazy initialization request 3516 * (ext4_rn_li_request) and keep track of the time spend in this 3517 * function. Based on that time we compute next schedule time of 3518 * the request. When walking through the list is complete, compute 3519 * next waking time and put itself into sleep. 3520 */ 3521 static int ext4_lazyinit_thread(void *arg) 3522 { 3523 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3524 struct list_head *pos, *n; 3525 struct ext4_li_request *elr; 3526 unsigned long next_wakeup, cur; 3527 3528 BUG_ON(NULL == eli); 3529 3530 cont_thread: 3531 while (true) { 3532 next_wakeup = MAX_JIFFY_OFFSET; 3533 3534 mutex_lock(&eli->li_list_mtx); 3535 if (list_empty(&eli->li_request_list)) { 3536 mutex_unlock(&eli->li_list_mtx); 3537 goto exit_thread; 3538 } 3539 list_for_each_safe(pos, n, &eli->li_request_list) { 3540 int err = 0; 3541 int progress = 0; 3542 elr = list_entry(pos, struct ext4_li_request, 3543 lr_request); 3544 3545 if (time_before(jiffies, elr->lr_next_sched)) { 3546 if (time_before(elr->lr_next_sched, next_wakeup)) 3547 next_wakeup = elr->lr_next_sched; 3548 continue; 3549 } 3550 if (down_read_trylock(&elr->lr_super->s_umount)) { 3551 if (sb_start_write_trylock(elr->lr_super)) { 3552 progress = 1; 3553 /* 3554 * We hold sb->s_umount, sb can not 3555 * be removed from the list, it is 3556 * now safe to drop li_list_mtx 3557 */ 3558 mutex_unlock(&eli->li_list_mtx); 3559 err = ext4_run_li_request(elr); 3560 sb_end_write(elr->lr_super); 3561 mutex_lock(&eli->li_list_mtx); 3562 n = pos->next; 3563 } 3564 up_read((&elr->lr_super->s_umount)); 3565 } 3566 /* error, remove the lazy_init job */ 3567 if (err) { 3568 ext4_remove_li_request(elr); 3569 continue; 3570 } 3571 if (!progress) { 3572 elr->lr_next_sched = jiffies + 3573 (prandom_u32() 3574 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3575 } 3576 if (time_before(elr->lr_next_sched, next_wakeup)) 3577 next_wakeup = elr->lr_next_sched; 3578 } 3579 mutex_unlock(&eli->li_list_mtx); 3580 3581 try_to_freeze(); 3582 3583 cur = jiffies; 3584 if ((time_after_eq(cur, next_wakeup)) || 3585 (MAX_JIFFY_OFFSET == next_wakeup)) { 3586 cond_resched(); 3587 continue; 3588 } 3589 3590 schedule_timeout_interruptible(next_wakeup - cur); 3591 3592 if (kthread_should_stop()) { 3593 ext4_clear_request_list(); 3594 goto exit_thread; 3595 } 3596 } 3597 3598 exit_thread: 3599 /* 3600 * It looks like the request list is empty, but we need 3601 * to check it under the li_list_mtx lock, to prevent any 3602 * additions into it, and of course we should lock ext4_li_mtx 3603 * to atomically free the list and ext4_li_info, because at 3604 * this point another ext4 filesystem could be registering 3605 * new one. 3606 */ 3607 mutex_lock(&ext4_li_mtx); 3608 mutex_lock(&eli->li_list_mtx); 3609 if (!list_empty(&eli->li_request_list)) { 3610 mutex_unlock(&eli->li_list_mtx); 3611 mutex_unlock(&ext4_li_mtx); 3612 goto cont_thread; 3613 } 3614 mutex_unlock(&eli->li_list_mtx); 3615 kfree(ext4_li_info); 3616 ext4_li_info = NULL; 3617 mutex_unlock(&ext4_li_mtx); 3618 3619 return 0; 3620 } 3621 3622 static void ext4_clear_request_list(void) 3623 { 3624 struct list_head *pos, *n; 3625 struct ext4_li_request *elr; 3626 3627 mutex_lock(&ext4_li_info->li_list_mtx); 3628 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3629 elr = list_entry(pos, struct ext4_li_request, 3630 lr_request); 3631 ext4_remove_li_request(elr); 3632 } 3633 mutex_unlock(&ext4_li_info->li_list_mtx); 3634 } 3635 3636 static int ext4_run_lazyinit_thread(void) 3637 { 3638 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3639 ext4_li_info, "ext4lazyinit"); 3640 if (IS_ERR(ext4_lazyinit_task)) { 3641 int err = PTR_ERR(ext4_lazyinit_task); 3642 ext4_clear_request_list(); 3643 kfree(ext4_li_info); 3644 ext4_li_info = NULL; 3645 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3646 "initialization thread\n", 3647 err); 3648 return err; 3649 } 3650 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3651 return 0; 3652 } 3653 3654 /* 3655 * Check whether it make sense to run itable init. thread or not. 3656 * If there is at least one uninitialized inode table, return 3657 * corresponding group number, else the loop goes through all 3658 * groups and return total number of groups. 3659 */ 3660 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3661 { 3662 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3663 struct ext4_group_desc *gdp = NULL; 3664 3665 if (!ext4_has_group_desc_csum(sb)) 3666 return ngroups; 3667 3668 for (group = 0; group < ngroups; group++) { 3669 gdp = ext4_get_group_desc(sb, group, NULL); 3670 if (!gdp) 3671 continue; 3672 3673 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3674 break; 3675 } 3676 3677 return group; 3678 } 3679 3680 static int ext4_li_info_new(void) 3681 { 3682 struct ext4_lazy_init *eli = NULL; 3683 3684 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3685 if (!eli) 3686 return -ENOMEM; 3687 3688 INIT_LIST_HEAD(&eli->li_request_list); 3689 mutex_init(&eli->li_list_mtx); 3690 3691 eli->li_state |= EXT4_LAZYINIT_QUIT; 3692 3693 ext4_li_info = eli; 3694 3695 return 0; 3696 } 3697 3698 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3699 ext4_group_t start) 3700 { 3701 struct ext4_li_request *elr; 3702 3703 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3704 if (!elr) 3705 return NULL; 3706 3707 elr->lr_super = sb; 3708 elr->lr_first_not_zeroed = start; 3709 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3710 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3711 elr->lr_next_group = start; 3712 } else { 3713 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3714 } 3715 3716 /* 3717 * Randomize first schedule time of the request to 3718 * spread the inode table initialization requests 3719 * better. 3720 */ 3721 elr->lr_next_sched = jiffies + (prandom_u32() % 3722 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3723 return elr; 3724 } 3725 3726 int ext4_register_li_request(struct super_block *sb, 3727 ext4_group_t first_not_zeroed) 3728 { 3729 struct ext4_sb_info *sbi = EXT4_SB(sb); 3730 struct ext4_li_request *elr = NULL; 3731 ext4_group_t ngroups = sbi->s_groups_count; 3732 int ret = 0; 3733 3734 mutex_lock(&ext4_li_mtx); 3735 if (sbi->s_li_request != NULL) { 3736 /* 3737 * Reset timeout so it can be computed again, because 3738 * s_li_wait_mult might have changed. 3739 */ 3740 sbi->s_li_request->lr_timeout = 0; 3741 goto out; 3742 } 3743 3744 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 3745 (first_not_zeroed == ngroups || sb_rdonly(sb) || 3746 !test_opt(sb, INIT_INODE_TABLE))) 3747 goto out; 3748 3749 elr = ext4_li_request_new(sb, first_not_zeroed); 3750 if (!elr) { 3751 ret = -ENOMEM; 3752 goto out; 3753 } 3754 3755 if (NULL == ext4_li_info) { 3756 ret = ext4_li_info_new(); 3757 if (ret) 3758 goto out; 3759 } 3760 3761 mutex_lock(&ext4_li_info->li_list_mtx); 3762 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3763 mutex_unlock(&ext4_li_info->li_list_mtx); 3764 3765 sbi->s_li_request = elr; 3766 /* 3767 * set elr to NULL here since it has been inserted to 3768 * the request_list and the removal and free of it is 3769 * handled by ext4_clear_request_list from now on. 3770 */ 3771 elr = NULL; 3772 3773 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3774 ret = ext4_run_lazyinit_thread(); 3775 if (ret) 3776 goto out; 3777 } 3778 out: 3779 mutex_unlock(&ext4_li_mtx); 3780 if (ret) 3781 kfree(elr); 3782 return ret; 3783 } 3784 3785 /* 3786 * We do not need to lock anything since this is called on 3787 * module unload. 3788 */ 3789 static void ext4_destroy_lazyinit_thread(void) 3790 { 3791 /* 3792 * If thread exited earlier 3793 * there's nothing to be done. 3794 */ 3795 if (!ext4_li_info || !ext4_lazyinit_task) 3796 return; 3797 3798 kthread_stop(ext4_lazyinit_task); 3799 } 3800 3801 static int set_journal_csum_feature_set(struct super_block *sb) 3802 { 3803 int ret = 1; 3804 int compat, incompat; 3805 struct ext4_sb_info *sbi = EXT4_SB(sb); 3806 3807 if (ext4_has_metadata_csum(sb)) { 3808 /* journal checksum v3 */ 3809 compat = 0; 3810 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3811 } else { 3812 /* journal checksum v1 */ 3813 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3814 incompat = 0; 3815 } 3816 3817 jbd2_journal_clear_features(sbi->s_journal, 3818 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3819 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3820 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3821 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3822 ret = jbd2_journal_set_features(sbi->s_journal, 3823 compat, 0, 3824 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3825 incompat); 3826 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3827 ret = jbd2_journal_set_features(sbi->s_journal, 3828 compat, 0, 3829 incompat); 3830 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3831 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3832 } else { 3833 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3834 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3835 } 3836 3837 return ret; 3838 } 3839 3840 /* 3841 * Note: calculating the overhead so we can be compatible with 3842 * historical BSD practice is quite difficult in the face of 3843 * clusters/bigalloc. This is because multiple metadata blocks from 3844 * different block group can end up in the same allocation cluster. 3845 * Calculating the exact overhead in the face of clustered allocation 3846 * requires either O(all block bitmaps) in memory or O(number of block 3847 * groups**2) in time. We will still calculate the superblock for 3848 * older file systems --- and if we come across with a bigalloc file 3849 * system with zero in s_overhead_clusters the estimate will be close to 3850 * correct especially for very large cluster sizes --- but for newer 3851 * file systems, it's better to calculate this figure once at mkfs 3852 * time, and store it in the superblock. If the superblock value is 3853 * present (even for non-bigalloc file systems), we will use it. 3854 */ 3855 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3856 char *buf) 3857 { 3858 struct ext4_sb_info *sbi = EXT4_SB(sb); 3859 struct ext4_group_desc *gdp; 3860 ext4_fsblk_t first_block, last_block, b; 3861 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3862 int s, j, count = 0; 3863 3864 if (!ext4_has_feature_bigalloc(sb)) 3865 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3866 sbi->s_itb_per_group + 2); 3867 3868 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3869 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3870 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3871 for (i = 0; i < ngroups; i++) { 3872 gdp = ext4_get_group_desc(sb, i, NULL); 3873 b = ext4_block_bitmap(sb, gdp); 3874 if (b >= first_block && b <= last_block) { 3875 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3876 count++; 3877 } 3878 b = ext4_inode_bitmap(sb, gdp); 3879 if (b >= first_block && b <= last_block) { 3880 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3881 count++; 3882 } 3883 b = ext4_inode_table(sb, gdp); 3884 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3885 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3886 int c = EXT4_B2C(sbi, b - first_block); 3887 ext4_set_bit(c, buf); 3888 count++; 3889 } 3890 if (i != grp) 3891 continue; 3892 s = 0; 3893 if (ext4_bg_has_super(sb, grp)) { 3894 ext4_set_bit(s++, buf); 3895 count++; 3896 } 3897 j = ext4_bg_num_gdb(sb, grp); 3898 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3899 ext4_error(sb, "Invalid number of block group " 3900 "descriptor blocks: %d", j); 3901 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3902 } 3903 count += j; 3904 for (; j > 0; j--) 3905 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3906 } 3907 if (!count) 3908 return 0; 3909 return EXT4_CLUSTERS_PER_GROUP(sb) - 3910 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3911 } 3912 3913 /* 3914 * Compute the overhead and stash it in sbi->s_overhead 3915 */ 3916 int ext4_calculate_overhead(struct super_block *sb) 3917 { 3918 struct ext4_sb_info *sbi = EXT4_SB(sb); 3919 struct ext4_super_block *es = sbi->s_es; 3920 struct inode *j_inode; 3921 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3922 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3923 ext4_fsblk_t overhead = 0; 3924 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3925 3926 if (!buf) 3927 return -ENOMEM; 3928 3929 /* 3930 * Compute the overhead (FS structures). This is constant 3931 * for a given filesystem unless the number of block groups 3932 * changes so we cache the previous value until it does. 3933 */ 3934 3935 /* 3936 * All of the blocks before first_data_block are overhead 3937 */ 3938 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3939 3940 /* 3941 * Add the overhead found in each block group 3942 */ 3943 for (i = 0; i < ngroups; i++) { 3944 int blks; 3945 3946 blks = count_overhead(sb, i, buf); 3947 overhead += blks; 3948 if (blks) 3949 memset(buf, 0, PAGE_SIZE); 3950 cond_resched(); 3951 } 3952 3953 /* 3954 * Add the internal journal blocks whether the journal has been 3955 * loaded or not 3956 */ 3957 if (sbi->s_journal && !sbi->s_journal_bdev) 3958 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 3959 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 3960 /* j_inum for internal journal is non-zero */ 3961 j_inode = ext4_get_journal_inode(sb, j_inum); 3962 if (j_inode) { 3963 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3964 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3965 iput(j_inode); 3966 } else { 3967 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3968 } 3969 } 3970 sbi->s_overhead = overhead; 3971 smp_wmb(); 3972 free_page((unsigned long) buf); 3973 return 0; 3974 } 3975 3976 static void ext4_set_resv_clusters(struct super_block *sb) 3977 { 3978 ext4_fsblk_t resv_clusters; 3979 struct ext4_sb_info *sbi = EXT4_SB(sb); 3980 3981 /* 3982 * There's no need to reserve anything when we aren't using extents. 3983 * The space estimates are exact, there are no unwritten extents, 3984 * hole punching doesn't need new metadata... This is needed especially 3985 * to keep ext2/3 backward compatibility. 3986 */ 3987 if (!ext4_has_feature_extents(sb)) 3988 return; 3989 /* 3990 * By default we reserve 2% or 4096 clusters, whichever is smaller. 3991 * This should cover the situations where we can not afford to run 3992 * out of space like for example punch hole, or converting 3993 * unwritten extents in delalloc path. In most cases such 3994 * allocation would require 1, or 2 blocks, higher numbers are 3995 * very rare. 3996 */ 3997 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 3998 sbi->s_cluster_bits); 3999 4000 do_div(resv_clusters, 50); 4001 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4002 4003 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4004 } 4005 4006 static const char *ext4_quota_mode(struct super_block *sb) 4007 { 4008 #ifdef CONFIG_QUOTA 4009 if (!ext4_quota_capable(sb)) 4010 return "none"; 4011 4012 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4013 return "journalled"; 4014 else 4015 return "writeback"; 4016 #else 4017 return "disabled"; 4018 #endif 4019 } 4020 4021 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 4022 { 4023 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 4024 char *orig_data = kstrdup(data, GFP_KERNEL); 4025 struct buffer_head *bh, **group_desc; 4026 struct ext4_super_block *es = NULL; 4027 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4028 struct flex_groups **flex_groups; 4029 ext4_fsblk_t block; 4030 ext4_fsblk_t sb_block = get_sb_block(&data); 4031 ext4_fsblk_t logical_sb_block; 4032 unsigned long offset = 0; 4033 unsigned long def_mount_opts; 4034 struct inode *root; 4035 const char *descr; 4036 int ret = -ENOMEM; 4037 int blocksize, clustersize; 4038 unsigned int db_count; 4039 unsigned int i; 4040 int needs_recovery, has_huge_files; 4041 __u64 blocks_count; 4042 int err = 0; 4043 ext4_group_t first_not_zeroed; 4044 struct ext4_parsed_options parsed_opts; 4045 4046 /* Set defaults for the variables that will be set during parsing */ 4047 parsed_opts.journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 4048 parsed_opts.journal_devnum = 0; 4049 parsed_opts.mb_optimize_scan = DEFAULT_MB_OPTIMIZE_SCAN; 4050 4051 if ((data && !orig_data) || !sbi) 4052 goto out_free_base; 4053 4054 sbi->s_daxdev = dax_dev; 4055 sbi->s_blockgroup_lock = 4056 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4057 if (!sbi->s_blockgroup_lock) 4058 goto out_free_base; 4059 4060 sb->s_fs_info = sbi; 4061 sbi->s_sb = sb; 4062 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 4063 sbi->s_sb_block = sb_block; 4064 sbi->s_sectors_written_start = 4065 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 4066 4067 /* Cleanup superblock name */ 4068 strreplace(sb->s_id, '/', '!'); 4069 4070 /* -EINVAL is default */ 4071 ret = -EINVAL; 4072 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4073 if (!blocksize) { 4074 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4075 goto out_fail; 4076 } 4077 4078 /* 4079 * The ext4 superblock will not be buffer aligned for other than 1kB 4080 * block sizes. We need to calculate the offset from buffer start. 4081 */ 4082 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4083 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4084 offset = do_div(logical_sb_block, blocksize); 4085 } else { 4086 logical_sb_block = sb_block; 4087 } 4088 4089 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4090 if (IS_ERR(bh)) { 4091 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4092 ret = PTR_ERR(bh); 4093 goto out_fail; 4094 } 4095 /* 4096 * Note: s_es must be initialized as soon as possible because 4097 * some ext4 macro-instructions depend on its value 4098 */ 4099 es = (struct ext4_super_block *) (bh->b_data + offset); 4100 sbi->s_es = es; 4101 sb->s_magic = le16_to_cpu(es->s_magic); 4102 if (sb->s_magic != EXT4_SUPER_MAGIC) 4103 goto cantfind_ext4; 4104 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 4105 4106 /* Warn if metadata_csum and gdt_csum are both set. */ 4107 if (ext4_has_feature_metadata_csum(sb) && 4108 ext4_has_feature_gdt_csum(sb)) 4109 ext4_warning(sb, "metadata_csum and uninit_bg are " 4110 "redundant flags; please run fsck."); 4111 4112 /* Check for a known checksum algorithm */ 4113 if (!ext4_verify_csum_type(sb, es)) { 4114 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4115 "unknown checksum algorithm."); 4116 silent = 1; 4117 goto cantfind_ext4; 4118 } 4119 4120 /* Load the checksum driver */ 4121 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4122 if (IS_ERR(sbi->s_chksum_driver)) { 4123 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4124 ret = PTR_ERR(sbi->s_chksum_driver); 4125 sbi->s_chksum_driver = NULL; 4126 goto failed_mount; 4127 } 4128 4129 /* Check superblock checksum */ 4130 if (!ext4_superblock_csum_verify(sb, es)) { 4131 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4132 "invalid superblock checksum. Run e2fsck?"); 4133 silent = 1; 4134 ret = -EFSBADCRC; 4135 goto cantfind_ext4; 4136 } 4137 4138 /* Precompute checksum seed for all metadata */ 4139 if (ext4_has_feature_csum_seed(sb)) 4140 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4141 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4142 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4143 sizeof(es->s_uuid)); 4144 4145 /* Set defaults before we parse the mount options */ 4146 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4147 set_opt(sb, INIT_INODE_TABLE); 4148 if (def_mount_opts & EXT4_DEFM_DEBUG) 4149 set_opt(sb, DEBUG); 4150 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4151 set_opt(sb, GRPID); 4152 if (def_mount_opts & EXT4_DEFM_UID16) 4153 set_opt(sb, NO_UID32); 4154 /* xattr user namespace & acls are now defaulted on */ 4155 set_opt(sb, XATTR_USER); 4156 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4157 set_opt(sb, POSIX_ACL); 4158 #endif 4159 if (ext4_has_feature_fast_commit(sb)) 4160 set_opt2(sb, JOURNAL_FAST_COMMIT); 4161 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4162 if (ext4_has_metadata_csum(sb)) 4163 set_opt(sb, JOURNAL_CHECKSUM); 4164 4165 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4166 set_opt(sb, JOURNAL_DATA); 4167 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4168 set_opt(sb, ORDERED_DATA); 4169 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4170 set_opt(sb, WRITEBACK_DATA); 4171 4172 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 4173 set_opt(sb, ERRORS_PANIC); 4174 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 4175 set_opt(sb, ERRORS_CONT); 4176 else 4177 set_opt(sb, ERRORS_RO); 4178 /* block_validity enabled by default; disable with noblock_validity */ 4179 set_opt(sb, BLOCK_VALIDITY); 4180 if (def_mount_opts & EXT4_DEFM_DISCARD) 4181 set_opt(sb, DISCARD); 4182 4183 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 4184 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 4185 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 4186 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 4187 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 4188 4189 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4190 set_opt(sb, BARRIER); 4191 4192 /* 4193 * enable delayed allocation by default 4194 * Use -o nodelalloc to turn it off 4195 */ 4196 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4197 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4198 set_opt(sb, DELALLOC); 4199 4200 /* 4201 * set default s_li_wait_mult for lazyinit, for the case there is 4202 * no mount option specified. 4203 */ 4204 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 4205 4206 if (le32_to_cpu(es->s_log_block_size) > 4207 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4208 ext4_msg(sb, KERN_ERR, 4209 "Invalid log block size: %u", 4210 le32_to_cpu(es->s_log_block_size)); 4211 goto failed_mount; 4212 } 4213 if (le32_to_cpu(es->s_log_cluster_size) > 4214 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4215 ext4_msg(sb, KERN_ERR, 4216 "Invalid log cluster size: %u", 4217 le32_to_cpu(es->s_log_cluster_size)); 4218 goto failed_mount; 4219 } 4220 4221 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4222 4223 if (blocksize == PAGE_SIZE) 4224 set_opt(sb, DIOREAD_NOLOCK); 4225 4226 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4227 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4228 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4229 } else { 4230 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4231 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4232 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4233 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4234 sbi->s_first_ino); 4235 goto failed_mount; 4236 } 4237 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4238 (!is_power_of_2(sbi->s_inode_size)) || 4239 (sbi->s_inode_size > blocksize)) { 4240 ext4_msg(sb, KERN_ERR, 4241 "unsupported inode size: %d", 4242 sbi->s_inode_size); 4243 ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize); 4244 goto failed_mount; 4245 } 4246 /* 4247 * i_atime_extra is the last extra field available for 4248 * [acm]times in struct ext4_inode. Checking for that 4249 * field should suffice to ensure we have extra space 4250 * for all three. 4251 */ 4252 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4253 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4254 sb->s_time_gran = 1; 4255 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4256 } else { 4257 sb->s_time_gran = NSEC_PER_SEC; 4258 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4259 } 4260 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4261 } 4262 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4263 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4264 EXT4_GOOD_OLD_INODE_SIZE; 4265 if (ext4_has_feature_extra_isize(sb)) { 4266 unsigned v, max = (sbi->s_inode_size - 4267 EXT4_GOOD_OLD_INODE_SIZE); 4268 4269 v = le16_to_cpu(es->s_want_extra_isize); 4270 if (v > max) { 4271 ext4_msg(sb, KERN_ERR, 4272 "bad s_want_extra_isize: %d", v); 4273 goto failed_mount; 4274 } 4275 if (sbi->s_want_extra_isize < v) 4276 sbi->s_want_extra_isize = v; 4277 4278 v = le16_to_cpu(es->s_min_extra_isize); 4279 if (v > max) { 4280 ext4_msg(sb, KERN_ERR, 4281 "bad s_min_extra_isize: %d", v); 4282 goto failed_mount; 4283 } 4284 if (sbi->s_want_extra_isize < v) 4285 sbi->s_want_extra_isize = v; 4286 } 4287 } 4288 4289 if (sbi->s_es->s_mount_opts[0]) { 4290 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 4291 sizeof(sbi->s_es->s_mount_opts), 4292 GFP_KERNEL); 4293 if (!s_mount_opts) 4294 goto failed_mount; 4295 if (!parse_options(s_mount_opts, sb, &parsed_opts, 0)) { 4296 ext4_msg(sb, KERN_WARNING, 4297 "failed to parse options in superblock: %s", 4298 s_mount_opts); 4299 } 4300 kfree(s_mount_opts); 4301 } 4302 sbi->s_def_mount_opt = sbi->s_mount_opt; 4303 if (!parse_options((char *) data, sb, &parsed_opts, 0)) 4304 goto failed_mount; 4305 4306 #ifdef CONFIG_UNICODE 4307 if (ext4_has_feature_casefold(sb) && !sb->s_encoding) { 4308 const struct ext4_sb_encodings *encoding_info; 4309 struct unicode_map *encoding; 4310 __u16 encoding_flags; 4311 4312 if (ext4_sb_read_encoding(es, &encoding_info, 4313 &encoding_flags)) { 4314 ext4_msg(sb, KERN_ERR, 4315 "Encoding requested by superblock is unknown"); 4316 goto failed_mount; 4317 } 4318 4319 encoding = utf8_load(encoding_info->version); 4320 if (IS_ERR(encoding)) { 4321 ext4_msg(sb, KERN_ERR, 4322 "can't mount with superblock charset: %s-%s " 4323 "not supported by the kernel. flags: 0x%x.", 4324 encoding_info->name, encoding_info->version, 4325 encoding_flags); 4326 goto failed_mount; 4327 } 4328 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4329 "%s-%s with flags 0x%hx", encoding_info->name, 4330 encoding_info->version?:"\b", encoding_flags); 4331 4332 sb->s_encoding = encoding; 4333 sb->s_encoding_flags = encoding_flags; 4334 } 4335 #endif 4336 4337 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4338 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n"); 4339 /* can't mount with both data=journal and dioread_nolock. */ 4340 clear_opt(sb, DIOREAD_NOLOCK); 4341 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4342 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4343 ext4_msg(sb, KERN_ERR, "can't mount with " 4344 "both data=journal and delalloc"); 4345 goto failed_mount; 4346 } 4347 if (test_opt(sb, DAX_ALWAYS)) { 4348 ext4_msg(sb, KERN_ERR, "can't mount with " 4349 "both data=journal and dax"); 4350 goto failed_mount; 4351 } 4352 if (ext4_has_feature_encrypt(sb)) { 4353 ext4_msg(sb, KERN_WARNING, 4354 "encrypted files will use data=ordered " 4355 "instead of data journaling mode"); 4356 } 4357 if (test_opt(sb, DELALLOC)) 4358 clear_opt(sb, DELALLOC); 4359 } else { 4360 sb->s_iflags |= SB_I_CGROUPWB; 4361 } 4362 4363 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 4364 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 4365 4366 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4367 (ext4_has_compat_features(sb) || 4368 ext4_has_ro_compat_features(sb) || 4369 ext4_has_incompat_features(sb))) 4370 ext4_msg(sb, KERN_WARNING, 4371 "feature flags set on rev 0 fs, " 4372 "running e2fsck is recommended"); 4373 4374 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4375 set_opt2(sb, HURD_COMPAT); 4376 if (ext4_has_feature_64bit(sb)) { 4377 ext4_msg(sb, KERN_ERR, 4378 "The Hurd can't support 64-bit file systems"); 4379 goto failed_mount; 4380 } 4381 4382 /* 4383 * ea_inode feature uses l_i_version field which is not 4384 * available in HURD_COMPAT mode. 4385 */ 4386 if (ext4_has_feature_ea_inode(sb)) { 4387 ext4_msg(sb, KERN_ERR, 4388 "ea_inode feature is not supported for Hurd"); 4389 goto failed_mount; 4390 } 4391 } 4392 4393 if (IS_EXT2_SB(sb)) { 4394 if (ext2_feature_set_ok(sb)) 4395 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4396 "using the ext4 subsystem"); 4397 else { 4398 /* 4399 * If we're probing be silent, if this looks like 4400 * it's actually an ext[34] filesystem. 4401 */ 4402 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4403 goto failed_mount; 4404 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4405 "to feature incompatibilities"); 4406 goto failed_mount; 4407 } 4408 } 4409 4410 if (IS_EXT3_SB(sb)) { 4411 if (ext3_feature_set_ok(sb)) 4412 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4413 "using the ext4 subsystem"); 4414 else { 4415 /* 4416 * If we're probing be silent, if this looks like 4417 * it's actually an ext4 filesystem. 4418 */ 4419 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4420 goto failed_mount; 4421 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4422 "to feature incompatibilities"); 4423 goto failed_mount; 4424 } 4425 } 4426 4427 /* 4428 * Check feature flags regardless of the revision level, since we 4429 * previously didn't change the revision level when setting the flags, 4430 * so there is a chance incompat flags are set on a rev 0 filesystem. 4431 */ 4432 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4433 goto failed_mount; 4434 4435 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 4436 ext4_msg(sb, KERN_ERR, 4437 "Number of reserved GDT blocks insanely large: %d", 4438 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 4439 goto failed_mount; 4440 } 4441 4442 if (bdev_dax_supported(sb->s_bdev, blocksize)) 4443 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 4444 4445 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 4446 if (ext4_has_feature_inline_data(sb)) { 4447 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 4448 " that may contain inline data"); 4449 goto failed_mount; 4450 } 4451 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 4452 ext4_msg(sb, KERN_ERR, 4453 "DAX unsupported by block device."); 4454 goto failed_mount; 4455 } 4456 } 4457 4458 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4459 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4460 es->s_encryption_level); 4461 goto failed_mount; 4462 } 4463 4464 if (sb->s_blocksize != blocksize) { 4465 /* Validate the filesystem blocksize */ 4466 if (!sb_set_blocksize(sb, blocksize)) { 4467 ext4_msg(sb, KERN_ERR, "bad block size %d", 4468 blocksize); 4469 goto failed_mount; 4470 } 4471 4472 brelse(bh); 4473 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4474 offset = do_div(logical_sb_block, blocksize); 4475 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4476 if (IS_ERR(bh)) { 4477 ext4_msg(sb, KERN_ERR, 4478 "Can't read superblock on 2nd try"); 4479 ret = PTR_ERR(bh); 4480 bh = NULL; 4481 goto failed_mount; 4482 } 4483 es = (struct ext4_super_block *)(bh->b_data + offset); 4484 sbi->s_es = es; 4485 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4486 ext4_msg(sb, KERN_ERR, 4487 "Magic mismatch, very weird!"); 4488 goto failed_mount; 4489 } 4490 } 4491 4492 has_huge_files = ext4_has_feature_huge_file(sb); 4493 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4494 has_huge_files); 4495 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4496 4497 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4498 if (ext4_has_feature_64bit(sb)) { 4499 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4500 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4501 !is_power_of_2(sbi->s_desc_size)) { 4502 ext4_msg(sb, KERN_ERR, 4503 "unsupported descriptor size %lu", 4504 sbi->s_desc_size); 4505 goto failed_mount; 4506 } 4507 } else 4508 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4509 4510 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4511 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4512 4513 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4514 if (sbi->s_inodes_per_block == 0) 4515 goto cantfind_ext4; 4516 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4517 sbi->s_inodes_per_group > blocksize * 8) { 4518 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4519 sbi->s_inodes_per_group); 4520 goto failed_mount; 4521 } 4522 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4523 sbi->s_inodes_per_block; 4524 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4525 sbi->s_sbh = bh; 4526 sbi->s_mount_state = le16_to_cpu(es->s_state); 4527 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4528 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4529 4530 for (i = 0; i < 4; i++) 4531 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4532 sbi->s_def_hash_version = es->s_def_hash_version; 4533 if (ext4_has_feature_dir_index(sb)) { 4534 i = le32_to_cpu(es->s_flags); 4535 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4536 sbi->s_hash_unsigned = 3; 4537 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4538 #ifdef __CHAR_UNSIGNED__ 4539 if (!sb_rdonly(sb)) 4540 es->s_flags |= 4541 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4542 sbi->s_hash_unsigned = 3; 4543 #else 4544 if (!sb_rdonly(sb)) 4545 es->s_flags |= 4546 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4547 #endif 4548 } 4549 } 4550 4551 /* Handle clustersize */ 4552 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4553 if (ext4_has_feature_bigalloc(sb)) { 4554 if (clustersize < blocksize) { 4555 ext4_msg(sb, KERN_ERR, 4556 "cluster size (%d) smaller than " 4557 "block size (%d)", clustersize, blocksize); 4558 goto failed_mount; 4559 } 4560 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4561 le32_to_cpu(es->s_log_block_size); 4562 sbi->s_clusters_per_group = 4563 le32_to_cpu(es->s_clusters_per_group); 4564 if (sbi->s_clusters_per_group > blocksize * 8) { 4565 ext4_msg(sb, KERN_ERR, 4566 "#clusters per group too big: %lu", 4567 sbi->s_clusters_per_group); 4568 goto failed_mount; 4569 } 4570 if (sbi->s_blocks_per_group != 4571 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4572 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4573 "clusters per group (%lu) inconsistent", 4574 sbi->s_blocks_per_group, 4575 sbi->s_clusters_per_group); 4576 goto failed_mount; 4577 } 4578 } else { 4579 if (clustersize != blocksize) { 4580 ext4_msg(sb, KERN_ERR, 4581 "fragment/cluster size (%d) != " 4582 "block size (%d)", clustersize, blocksize); 4583 goto failed_mount; 4584 } 4585 if (sbi->s_blocks_per_group > blocksize * 8) { 4586 ext4_msg(sb, KERN_ERR, 4587 "#blocks per group too big: %lu", 4588 sbi->s_blocks_per_group); 4589 goto failed_mount; 4590 } 4591 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4592 sbi->s_cluster_bits = 0; 4593 } 4594 sbi->s_cluster_ratio = clustersize / blocksize; 4595 4596 /* Do we have standard group size of clustersize * 8 blocks ? */ 4597 if (sbi->s_blocks_per_group == clustersize << 3) 4598 set_opt2(sb, STD_GROUP_SIZE); 4599 4600 /* 4601 * Test whether we have more sectors than will fit in sector_t, 4602 * and whether the max offset is addressable by the page cache. 4603 */ 4604 err = generic_check_addressable(sb->s_blocksize_bits, 4605 ext4_blocks_count(es)); 4606 if (err) { 4607 ext4_msg(sb, KERN_ERR, "filesystem" 4608 " too large to mount safely on this system"); 4609 goto failed_mount; 4610 } 4611 4612 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4613 goto cantfind_ext4; 4614 4615 /* check blocks count against device size */ 4616 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4617 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4618 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4619 "exceeds size of device (%llu blocks)", 4620 ext4_blocks_count(es), blocks_count); 4621 goto failed_mount; 4622 } 4623 4624 /* 4625 * It makes no sense for the first data block to be beyond the end 4626 * of the filesystem. 4627 */ 4628 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4629 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4630 "block %u is beyond end of filesystem (%llu)", 4631 le32_to_cpu(es->s_first_data_block), 4632 ext4_blocks_count(es)); 4633 goto failed_mount; 4634 } 4635 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4636 (sbi->s_cluster_ratio == 1)) { 4637 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4638 "block is 0 with a 1k block and cluster size"); 4639 goto failed_mount; 4640 } 4641 4642 blocks_count = (ext4_blocks_count(es) - 4643 le32_to_cpu(es->s_first_data_block) + 4644 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4645 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4646 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4647 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4648 "(block count %llu, first data block %u, " 4649 "blocks per group %lu)", blocks_count, 4650 ext4_blocks_count(es), 4651 le32_to_cpu(es->s_first_data_block), 4652 EXT4_BLOCKS_PER_GROUP(sb)); 4653 goto failed_mount; 4654 } 4655 sbi->s_groups_count = blocks_count; 4656 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4657 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4658 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4659 le32_to_cpu(es->s_inodes_count)) { 4660 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4661 le32_to_cpu(es->s_inodes_count), 4662 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4663 ret = -EINVAL; 4664 goto failed_mount; 4665 } 4666 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4667 EXT4_DESC_PER_BLOCK(sb); 4668 if (ext4_has_feature_meta_bg(sb)) { 4669 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4670 ext4_msg(sb, KERN_WARNING, 4671 "first meta block group too large: %u " 4672 "(group descriptor block count %u)", 4673 le32_to_cpu(es->s_first_meta_bg), db_count); 4674 goto failed_mount; 4675 } 4676 } 4677 rcu_assign_pointer(sbi->s_group_desc, 4678 kvmalloc_array(db_count, 4679 sizeof(struct buffer_head *), 4680 GFP_KERNEL)); 4681 if (sbi->s_group_desc == NULL) { 4682 ext4_msg(sb, KERN_ERR, "not enough memory"); 4683 ret = -ENOMEM; 4684 goto failed_mount; 4685 } 4686 4687 bgl_lock_init(sbi->s_blockgroup_lock); 4688 4689 /* Pre-read the descriptors into the buffer cache */ 4690 for (i = 0; i < db_count; i++) { 4691 block = descriptor_loc(sb, logical_sb_block, i); 4692 ext4_sb_breadahead_unmovable(sb, block); 4693 } 4694 4695 for (i = 0; i < db_count; i++) { 4696 struct buffer_head *bh; 4697 4698 block = descriptor_loc(sb, logical_sb_block, i); 4699 bh = ext4_sb_bread_unmovable(sb, block); 4700 if (IS_ERR(bh)) { 4701 ext4_msg(sb, KERN_ERR, 4702 "can't read group descriptor %d", i); 4703 db_count = i; 4704 ret = PTR_ERR(bh); 4705 goto failed_mount2; 4706 } 4707 rcu_read_lock(); 4708 rcu_dereference(sbi->s_group_desc)[i] = bh; 4709 rcu_read_unlock(); 4710 } 4711 sbi->s_gdb_count = db_count; 4712 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4713 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4714 ret = -EFSCORRUPTED; 4715 goto failed_mount2; 4716 } 4717 4718 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4719 spin_lock_init(&sbi->s_error_lock); 4720 INIT_WORK(&sbi->s_error_work, flush_stashed_error_work); 4721 4722 /* Register extent status tree shrinker */ 4723 if (ext4_es_register_shrinker(sbi)) 4724 goto failed_mount3; 4725 4726 sbi->s_stripe = ext4_get_stripe_size(sbi); 4727 sbi->s_extent_max_zeroout_kb = 32; 4728 4729 /* 4730 * set up enough so that it can read an inode 4731 */ 4732 sb->s_op = &ext4_sops; 4733 sb->s_export_op = &ext4_export_ops; 4734 sb->s_xattr = ext4_xattr_handlers; 4735 #ifdef CONFIG_FS_ENCRYPTION 4736 sb->s_cop = &ext4_cryptops; 4737 #endif 4738 #ifdef CONFIG_FS_VERITY 4739 sb->s_vop = &ext4_verityops; 4740 #endif 4741 #ifdef CONFIG_QUOTA 4742 sb->dq_op = &ext4_quota_operations; 4743 if (ext4_has_feature_quota(sb)) 4744 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4745 else 4746 sb->s_qcop = &ext4_qctl_operations; 4747 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4748 #endif 4749 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4750 4751 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4752 mutex_init(&sbi->s_orphan_lock); 4753 4754 /* Initialize fast commit stuff */ 4755 atomic_set(&sbi->s_fc_subtid, 0); 4756 atomic_set(&sbi->s_fc_ineligible_updates, 0); 4757 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4758 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4759 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4760 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4761 sbi->s_fc_bytes = 0; 4762 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4763 ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING); 4764 spin_lock_init(&sbi->s_fc_lock); 4765 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4766 sbi->s_fc_replay_state.fc_regions = NULL; 4767 sbi->s_fc_replay_state.fc_regions_size = 0; 4768 sbi->s_fc_replay_state.fc_regions_used = 0; 4769 sbi->s_fc_replay_state.fc_regions_valid = 0; 4770 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4771 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4772 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4773 4774 sb->s_root = NULL; 4775 4776 needs_recovery = (es->s_last_orphan != 0 || 4777 ext4_has_feature_journal_needs_recovery(sb)); 4778 4779 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4780 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4781 goto failed_mount3a; 4782 4783 /* 4784 * The first inode we look at is the journal inode. Don't try 4785 * root first: it may be modified in the journal! 4786 */ 4787 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4788 err = ext4_load_journal(sb, es, parsed_opts.journal_devnum); 4789 if (err) 4790 goto failed_mount3a; 4791 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4792 ext4_has_feature_journal_needs_recovery(sb)) { 4793 ext4_msg(sb, KERN_ERR, "required journal recovery " 4794 "suppressed and not mounted read-only"); 4795 goto failed_mount_wq; 4796 } else { 4797 /* Nojournal mode, all journal mount options are illegal */ 4798 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4799 ext4_msg(sb, KERN_ERR, "can't mount with " 4800 "journal_checksum, fs mounted w/o journal"); 4801 goto failed_mount_wq; 4802 } 4803 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4804 ext4_msg(sb, KERN_ERR, "can't mount with " 4805 "journal_async_commit, fs mounted w/o journal"); 4806 goto failed_mount_wq; 4807 } 4808 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4809 ext4_msg(sb, KERN_ERR, "can't mount with " 4810 "commit=%lu, fs mounted w/o journal", 4811 sbi->s_commit_interval / HZ); 4812 goto failed_mount_wq; 4813 } 4814 if (EXT4_MOUNT_DATA_FLAGS & 4815 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4816 ext4_msg(sb, KERN_ERR, "can't mount with " 4817 "data=, fs mounted w/o journal"); 4818 goto failed_mount_wq; 4819 } 4820 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4821 clear_opt(sb, JOURNAL_CHECKSUM); 4822 clear_opt(sb, DATA_FLAGS); 4823 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4824 sbi->s_journal = NULL; 4825 needs_recovery = 0; 4826 goto no_journal; 4827 } 4828 4829 if (ext4_has_feature_64bit(sb) && 4830 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4831 JBD2_FEATURE_INCOMPAT_64BIT)) { 4832 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4833 goto failed_mount_wq; 4834 } 4835 4836 if (!set_journal_csum_feature_set(sb)) { 4837 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4838 "feature set"); 4839 goto failed_mount_wq; 4840 } 4841 4842 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4843 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4844 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4845 ext4_msg(sb, KERN_ERR, 4846 "Failed to set fast commit journal feature"); 4847 goto failed_mount_wq; 4848 } 4849 4850 /* We have now updated the journal if required, so we can 4851 * validate the data journaling mode. */ 4852 switch (test_opt(sb, DATA_FLAGS)) { 4853 case 0: 4854 /* No mode set, assume a default based on the journal 4855 * capabilities: ORDERED_DATA if the journal can 4856 * cope, else JOURNAL_DATA 4857 */ 4858 if (jbd2_journal_check_available_features 4859 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4860 set_opt(sb, ORDERED_DATA); 4861 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4862 } else { 4863 set_opt(sb, JOURNAL_DATA); 4864 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4865 } 4866 break; 4867 4868 case EXT4_MOUNT_ORDERED_DATA: 4869 case EXT4_MOUNT_WRITEBACK_DATA: 4870 if (!jbd2_journal_check_available_features 4871 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4872 ext4_msg(sb, KERN_ERR, "Journal does not support " 4873 "requested data journaling mode"); 4874 goto failed_mount_wq; 4875 } 4876 break; 4877 default: 4878 break; 4879 } 4880 4881 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4882 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4883 ext4_msg(sb, KERN_ERR, "can't mount with " 4884 "journal_async_commit in data=ordered mode"); 4885 goto failed_mount_wq; 4886 } 4887 4888 set_task_ioprio(sbi->s_journal->j_task, parsed_opts.journal_ioprio); 4889 4890 sbi->s_journal->j_submit_inode_data_buffers = 4891 ext4_journal_submit_inode_data_buffers; 4892 sbi->s_journal->j_finish_inode_data_buffers = 4893 ext4_journal_finish_inode_data_buffers; 4894 4895 no_journal: 4896 if (!test_opt(sb, NO_MBCACHE)) { 4897 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4898 if (!sbi->s_ea_block_cache) { 4899 ext4_msg(sb, KERN_ERR, 4900 "Failed to create ea_block_cache"); 4901 goto failed_mount_wq; 4902 } 4903 4904 if (ext4_has_feature_ea_inode(sb)) { 4905 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4906 if (!sbi->s_ea_inode_cache) { 4907 ext4_msg(sb, KERN_ERR, 4908 "Failed to create ea_inode_cache"); 4909 goto failed_mount_wq; 4910 } 4911 } 4912 } 4913 4914 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4915 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4916 goto failed_mount_wq; 4917 } 4918 4919 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4920 !ext4_has_feature_encrypt(sb)) { 4921 ext4_set_feature_encrypt(sb); 4922 ext4_commit_super(sb); 4923 } 4924 4925 /* 4926 * Get the # of file system overhead blocks from the 4927 * superblock if present. 4928 */ 4929 if (es->s_overhead_clusters) 4930 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4931 else { 4932 err = ext4_calculate_overhead(sb); 4933 if (err) 4934 goto failed_mount_wq; 4935 } 4936 4937 /* 4938 * The maximum number of concurrent works can be high and 4939 * concurrency isn't really necessary. Limit it to 1. 4940 */ 4941 EXT4_SB(sb)->rsv_conversion_wq = 4942 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4943 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4944 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4945 ret = -ENOMEM; 4946 goto failed_mount4; 4947 } 4948 4949 /* 4950 * The jbd2_journal_load will have done any necessary log recovery, 4951 * so we can safely mount the rest of the filesystem now. 4952 */ 4953 4954 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4955 if (IS_ERR(root)) { 4956 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4957 ret = PTR_ERR(root); 4958 root = NULL; 4959 goto failed_mount4; 4960 } 4961 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4962 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4963 iput(root); 4964 goto failed_mount4; 4965 } 4966 4967 sb->s_root = d_make_root(root); 4968 if (!sb->s_root) { 4969 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4970 ret = -ENOMEM; 4971 goto failed_mount4; 4972 } 4973 4974 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4975 if (ret == -EROFS) { 4976 sb->s_flags |= SB_RDONLY; 4977 ret = 0; 4978 } else if (ret) 4979 goto failed_mount4a; 4980 4981 ext4_set_resv_clusters(sb); 4982 4983 if (test_opt(sb, BLOCK_VALIDITY)) { 4984 err = ext4_setup_system_zone(sb); 4985 if (err) { 4986 ext4_msg(sb, KERN_ERR, "failed to initialize system " 4987 "zone (%d)", err); 4988 goto failed_mount4a; 4989 } 4990 } 4991 ext4_fc_replay_cleanup(sb); 4992 4993 ext4_ext_init(sb); 4994 4995 /* 4996 * Enable optimize_scan if number of groups is > threshold. This can be 4997 * turned off by passing "mb_optimize_scan=0". This can also be 4998 * turned on forcefully by passing "mb_optimize_scan=1". 4999 */ 5000 if (parsed_opts.mb_optimize_scan == 1) 5001 set_opt2(sb, MB_OPTIMIZE_SCAN); 5002 else if (parsed_opts.mb_optimize_scan == 0) 5003 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5004 else if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5005 set_opt2(sb, MB_OPTIMIZE_SCAN); 5006 5007 err = ext4_mb_init(sb); 5008 if (err) { 5009 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5010 err); 5011 goto failed_mount5; 5012 } 5013 5014 /* 5015 * We can only set up the journal commit callback once 5016 * mballoc is initialized 5017 */ 5018 if (sbi->s_journal) 5019 sbi->s_journal->j_commit_callback = 5020 ext4_journal_commit_callback; 5021 5022 block = ext4_count_free_clusters(sb); 5023 ext4_free_blocks_count_set(sbi->s_es, 5024 EXT4_C2B(sbi, block)); 5025 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5026 GFP_KERNEL); 5027 if (!err) { 5028 unsigned long freei = ext4_count_free_inodes(sb); 5029 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5030 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5031 GFP_KERNEL); 5032 } 5033 if (!err) 5034 err = percpu_counter_init(&sbi->s_dirs_counter, 5035 ext4_count_dirs(sb), GFP_KERNEL); 5036 if (!err) 5037 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5038 GFP_KERNEL); 5039 if (!err) 5040 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 5041 GFP_KERNEL); 5042 if (!err) 5043 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5044 5045 if (err) { 5046 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5047 goto failed_mount6; 5048 } 5049 5050 if (ext4_has_feature_flex_bg(sb)) 5051 if (!ext4_fill_flex_info(sb)) { 5052 ext4_msg(sb, KERN_ERR, 5053 "unable to initialize " 5054 "flex_bg meta info!"); 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 if (sbi->s_mmp_tsk) 5184 kthread_stop(sbi->s_mmp_tsk); 5185 failed_mount2: 5186 rcu_read_lock(); 5187 group_desc = rcu_dereference(sbi->s_group_desc); 5188 for (i = 0; i < db_count; i++) 5189 brelse(group_desc[i]); 5190 kvfree(group_desc); 5191 rcu_read_unlock(); 5192 failed_mount: 5193 if (sbi->s_chksum_driver) 5194 crypto_free_shash(sbi->s_chksum_driver); 5195 5196 #ifdef CONFIG_UNICODE 5197 utf8_unload(sb->s_encoding); 5198 #endif 5199 5200 #ifdef CONFIG_QUOTA 5201 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5202 kfree(get_qf_name(sb, sbi, i)); 5203 #endif 5204 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5205 ext4_blkdev_remove(sbi); 5206 brelse(bh); 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); 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); 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 if (sbi->s_mmp_tsk) 5986 kthread_stop(sbi->s_mmp_tsk); 5987 } else { 5988 /* Make sure we can mount this feature set readwrite */ 5989 if (ext4_has_feature_readonly(sb) || 5990 !ext4_feature_set_ok(sb, 0)) { 5991 err = -EROFS; 5992 goto restore_opts; 5993 } 5994 /* 5995 * Make sure the group descriptor checksums 5996 * are sane. If they aren't, refuse to remount r/w. 5997 */ 5998 for (g = 0; g < sbi->s_groups_count; g++) { 5999 struct ext4_group_desc *gdp = 6000 ext4_get_group_desc(sb, g, NULL); 6001 6002 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6003 ext4_msg(sb, KERN_ERR, 6004 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6005 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6006 le16_to_cpu(gdp->bg_checksum)); 6007 err = -EFSBADCRC; 6008 goto restore_opts; 6009 } 6010 } 6011 6012 /* 6013 * If we have an unprocessed orphan list hanging 6014 * around from a previously readonly bdev mount, 6015 * require a full umount/remount for now. 6016 */ 6017 if (es->s_last_orphan) { 6018 ext4_msg(sb, KERN_WARNING, "Couldn't " 6019 "remount RDWR because of unprocessed " 6020 "orphan inode list. Please " 6021 "umount/remount instead"); 6022 err = -EINVAL; 6023 goto restore_opts; 6024 } 6025 6026 /* 6027 * Mounting a RDONLY partition read-write, so reread 6028 * and store the current valid flag. (It may have 6029 * been changed by e2fsck since we originally mounted 6030 * the partition.) 6031 */ 6032 if (sbi->s_journal) { 6033 err = ext4_clear_journal_err(sb, es); 6034 if (err) 6035 goto restore_opts; 6036 } 6037 sbi->s_mount_state = le16_to_cpu(es->s_state); 6038 6039 err = ext4_setup_super(sb, es, 0); 6040 if (err) 6041 goto restore_opts; 6042 6043 sb->s_flags &= ~SB_RDONLY; 6044 if (ext4_has_feature_mmp(sb)) 6045 if (ext4_multi_mount_protect(sb, 6046 le64_to_cpu(es->s_mmp_block))) { 6047 err = -EROFS; 6048 goto restore_opts; 6049 } 6050 enable_quota = 1; 6051 } 6052 } 6053 6054 /* 6055 * Reinitialize lazy itable initialization thread based on 6056 * current settings 6057 */ 6058 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6059 ext4_unregister_li_request(sb); 6060 else { 6061 ext4_group_t first_not_zeroed; 6062 first_not_zeroed = ext4_has_uninit_itable(sb); 6063 ext4_register_li_request(sb, first_not_zeroed); 6064 } 6065 6066 /* 6067 * Handle creation of system zone data early because it can fail. 6068 * Releasing of existing data is done when we are sure remount will 6069 * succeed. 6070 */ 6071 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6072 err = ext4_setup_system_zone(sb); 6073 if (err) 6074 goto restore_opts; 6075 } 6076 6077 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6078 err = ext4_commit_super(sb); 6079 if (err) 6080 goto restore_opts; 6081 } 6082 6083 #ifdef CONFIG_QUOTA 6084 /* Release old quota file names */ 6085 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6086 kfree(old_opts.s_qf_names[i]); 6087 if (enable_quota) { 6088 if (sb_any_quota_suspended(sb)) 6089 dquot_resume(sb, -1); 6090 else if (ext4_has_feature_quota(sb)) { 6091 err = ext4_enable_quotas(sb); 6092 if (err) 6093 goto restore_opts; 6094 } 6095 } 6096 #endif 6097 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6098 ext4_release_system_zone(sb); 6099 6100 /* 6101 * Some options can be enabled by ext4 and/or by VFS mount flag 6102 * either way we need to make sure it matches in both *flags and 6103 * s_flags. Copy those selected flags from s_flags to *flags 6104 */ 6105 *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags); 6106 6107 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s. Quota mode: %s.", 6108 orig_data, ext4_quota_mode(sb)); 6109 kfree(orig_data); 6110 return 0; 6111 6112 restore_opts: 6113 sb->s_flags = old_sb_flags; 6114 sbi->s_mount_opt = old_opts.s_mount_opt; 6115 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6116 sbi->s_resuid = old_opts.s_resuid; 6117 sbi->s_resgid = old_opts.s_resgid; 6118 sbi->s_commit_interval = old_opts.s_commit_interval; 6119 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6120 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6121 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6122 ext4_release_system_zone(sb); 6123 #ifdef CONFIG_QUOTA 6124 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6125 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6126 to_free[i] = get_qf_name(sb, sbi, i); 6127 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6128 } 6129 synchronize_rcu(); 6130 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6131 kfree(to_free[i]); 6132 #endif 6133 kfree(orig_data); 6134 return err; 6135 } 6136 6137 #ifdef CONFIG_QUOTA 6138 static int ext4_statfs_project(struct super_block *sb, 6139 kprojid_t projid, struct kstatfs *buf) 6140 { 6141 struct kqid qid; 6142 struct dquot *dquot; 6143 u64 limit; 6144 u64 curblock; 6145 6146 qid = make_kqid_projid(projid); 6147 dquot = dqget(sb, qid); 6148 if (IS_ERR(dquot)) 6149 return PTR_ERR(dquot); 6150 spin_lock(&dquot->dq_dqb_lock); 6151 6152 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6153 dquot->dq_dqb.dqb_bhardlimit); 6154 limit >>= sb->s_blocksize_bits; 6155 6156 if (limit && buf->f_blocks > limit) { 6157 curblock = (dquot->dq_dqb.dqb_curspace + 6158 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6159 buf->f_blocks = limit; 6160 buf->f_bfree = buf->f_bavail = 6161 (buf->f_blocks > curblock) ? 6162 (buf->f_blocks - curblock) : 0; 6163 } 6164 6165 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6166 dquot->dq_dqb.dqb_ihardlimit); 6167 if (limit && buf->f_files > limit) { 6168 buf->f_files = limit; 6169 buf->f_ffree = 6170 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6171 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6172 } 6173 6174 spin_unlock(&dquot->dq_dqb_lock); 6175 dqput(dquot); 6176 return 0; 6177 } 6178 #endif 6179 6180 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6181 { 6182 struct super_block *sb = dentry->d_sb; 6183 struct ext4_sb_info *sbi = EXT4_SB(sb); 6184 struct ext4_super_block *es = sbi->s_es; 6185 ext4_fsblk_t overhead = 0, resv_blocks; 6186 s64 bfree; 6187 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6188 6189 if (!test_opt(sb, MINIX_DF)) 6190 overhead = sbi->s_overhead; 6191 6192 buf->f_type = EXT4_SUPER_MAGIC; 6193 buf->f_bsize = sb->s_blocksize; 6194 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6195 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6196 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6197 /* prevent underflow in case that few free space is available */ 6198 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6199 buf->f_bavail = buf->f_bfree - 6200 (ext4_r_blocks_count(es) + resv_blocks); 6201 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6202 buf->f_bavail = 0; 6203 buf->f_files = le32_to_cpu(es->s_inodes_count); 6204 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6205 buf->f_namelen = EXT4_NAME_LEN; 6206 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6207 6208 #ifdef CONFIG_QUOTA 6209 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6210 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6211 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6212 #endif 6213 return 0; 6214 } 6215 6216 6217 #ifdef CONFIG_QUOTA 6218 6219 /* 6220 * Helper functions so that transaction is started before we acquire dqio_sem 6221 * to keep correct lock ordering of transaction > dqio_sem 6222 */ 6223 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6224 { 6225 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6226 } 6227 6228 static int ext4_write_dquot(struct dquot *dquot) 6229 { 6230 int ret, err; 6231 handle_t *handle; 6232 struct inode *inode; 6233 6234 inode = dquot_to_inode(dquot); 6235 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6236 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6237 if (IS_ERR(handle)) 6238 return PTR_ERR(handle); 6239 ret = dquot_commit(dquot); 6240 err = ext4_journal_stop(handle); 6241 if (!ret) 6242 ret = err; 6243 return ret; 6244 } 6245 6246 static int ext4_acquire_dquot(struct dquot *dquot) 6247 { 6248 int ret, err; 6249 handle_t *handle; 6250 6251 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6252 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6253 if (IS_ERR(handle)) 6254 return PTR_ERR(handle); 6255 ret = dquot_acquire(dquot); 6256 err = ext4_journal_stop(handle); 6257 if (!ret) 6258 ret = err; 6259 return ret; 6260 } 6261 6262 static int ext4_release_dquot(struct dquot *dquot) 6263 { 6264 int ret, err; 6265 handle_t *handle; 6266 6267 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6268 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6269 if (IS_ERR(handle)) { 6270 /* Release dquot anyway to avoid endless cycle in dqput() */ 6271 dquot_release(dquot); 6272 return PTR_ERR(handle); 6273 } 6274 ret = dquot_release(dquot); 6275 err = ext4_journal_stop(handle); 6276 if (!ret) 6277 ret = err; 6278 return ret; 6279 } 6280 6281 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6282 { 6283 struct super_block *sb = dquot->dq_sb; 6284 6285 if (ext4_is_quota_journalled(sb)) { 6286 dquot_mark_dquot_dirty(dquot); 6287 return ext4_write_dquot(dquot); 6288 } else { 6289 return dquot_mark_dquot_dirty(dquot); 6290 } 6291 } 6292 6293 static int ext4_write_info(struct super_block *sb, int type) 6294 { 6295 int ret, err; 6296 handle_t *handle; 6297 6298 /* Data block + inode block */ 6299 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 6300 if (IS_ERR(handle)) 6301 return PTR_ERR(handle); 6302 ret = dquot_commit_info(sb, type); 6303 err = ext4_journal_stop(handle); 6304 if (!ret) 6305 ret = err; 6306 return ret; 6307 } 6308 6309 /* 6310 * Turn on quotas during mount time - we need to find 6311 * the quota file and such... 6312 */ 6313 static int ext4_quota_on_mount(struct super_block *sb, int type) 6314 { 6315 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 6316 EXT4_SB(sb)->s_jquota_fmt, type); 6317 } 6318 6319 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6320 { 6321 struct ext4_inode_info *ei = EXT4_I(inode); 6322 6323 /* The first argument of lockdep_set_subclass has to be 6324 * *exactly* the same as the argument to init_rwsem() --- in 6325 * this case, in init_once() --- or lockdep gets unhappy 6326 * because the name of the lock is set using the 6327 * stringification of the argument to init_rwsem(). 6328 */ 6329 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6330 lockdep_set_subclass(&ei->i_data_sem, subclass); 6331 } 6332 6333 /* 6334 * Standard function to be called on quota_on 6335 */ 6336 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6337 const struct path *path) 6338 { 6339 int err; 6340 6341 if (!test_opt(sb, QUOTA)) 6342 return -EINVAL; 6343 6344 /* Quotafile not on the same filesystem? */ 6345 if (path->dentry->d_sb != sb) 6346 return -EXDEV; 6347 6348 /* Quota already enabled for this file? */ 6349 if (IS_NOQUOTA(d_inode(path->dentry))) 6350 return -EBUSY; 6351 6352 /* Journaling quota? */ 6353 if (EXT4_SB(sb)->s_qf_names[type]) { 6354 /* Quotafile not in fs root? */ 6355 if (path->dentry->d_parent != sb->s_root) 6356 ext4_msg(sb, KERN_WARNING, 6357 "Quota file not on filesystem root. " 6358 "Journaled quota will not work"); 6359 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6360 } else { 6361 /* 6362 * Clear the flag just in case mount options changed since 6363 * last time. 6364 */ 6365 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6366 } 6367 6368 /* 6369 * When we journal data on quota file, we have to flush journal to see 6370 * all updates to the file when we bypass pagecache... 6371 */ 6372 if (EXT4_SB(sb)->s_journal && 6373 ext4_should_journal_data(d_inode(path->dentry))) { 6374 /* 6375 * We don't need to lock updates but journal_flush() could 6376 * otherwise be livelocked... 6377 */ 6378 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6379 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 6380 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6381 if (err) 6382 return err; 6383 } 6384 6385 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6386 err = dquot_quota_on(sb, type, format_id, path); 6387 if (err) { 6388 lockdep_set_quota_inode(path->dentry->d_inode, 6389 I_DATA_SEM_NORMAL); 6390 } else { 6391 struct inode *inode = d_inode(path->dentry); 6392 handle_t *handle; 6393 6394 /* 6395 * Set inode flags to prevent userspace from messing with quota 6396 * files. If this fails, we return success anyway since quotas 6397 * are already enabled and this is not a hard failure. 6398 */ 6399 inode_lock(inode); 6400 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6401 if (IS_ERR(handle)) 6402 goto unlock_inode; 6403 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6404 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6405 S_NOATIME | S_IMMUTABLE); 6406 err = ext4_mark_inode_dirty(handle, inode); 6407 ext4_journal_stop(handle); 6408 unlock_inode: 6409 inode_unlock(inode); 6410 } 6411 return err; 6412 } 6413 6414 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6415 unsigned int flags) 6416 { 6417 int err; 6418 struct inode *qf_inode; 6419 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6420 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6421 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6422 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6423 }; 6424 6425 BUG_ON(!ext4_has_feature_quota(sb)); 6426 6427 if (!qf_inums[type]) 6428 return -EPERM; 6429 6430 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6431 if (IS_ERR(qf_inode)) { 6432 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6433 return PTR_ERR(qf_inode); 6434 } 6435 6436 /* Don't account quota for quota files to avoid recursion */ 6437 qf_inode->i_flags |= S_NOQUOTA; 6438 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6439 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6440 if (err) 6441 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6442 iput(qf_inode); 6443 6444 return err; 6445 } 6446 6447 /* Enable usage tracking for all quota types. */ 6448 static int ext4_enable_quotas(struct super_block *sb) 6449 { 6450 int type, err = 0; 6451 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6452 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6453 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6454 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6455 }; 6456 bool quota_mopt[EXT4_MAXQUOTAS] = { 6457 test_opt(sb, USRQUOTA), 6458 test_opt(sb, GRPQUOTA), 6459 test_opt(sb, PRJQUOTA), 6460 }; 6461 6462 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6463 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6464 if (qf_inums[type]) { 6465 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6466 DQUOT_USAGE_ENABLED | 6467 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6468 if (err) { 6469 ext4_warning(sb, 6470 "Failed to enable quota tracking " 6471 "(type=%d, err=%d). Please run " 6472 "e2fsck to fix.", type, err); 6473 for (type--; type >= 0; type--) 6474 dquot_quota_off(sb, type); 6475 6476 return err; 6477 } 6478 } 6479 } 6480 return 0; 6481 } 6482 6483 static int ext4_quota_off(struct super_block *sb, int type) 6484 { 6485 struct inode *inode = sb_dqopt(sb)->files[type]; 6486 handle_t *handle; 6487 int err; 6488 6489 /* Force all delayed allocation blocks to be allocated. 6490 * Caller already holds s_umount sem */ 6491 if (test_opt(sb, DELALLOC)) 6492 sync_filesystem(sb); 6493 6494 if (!inode || !igrab(inode)) 6495 goto out; 6496 6497 err = dquot_quota_off(sb, type); 6498 if (err || ext4_has_feature_quota(sb)) 6499 goto out_put; 6500 6501 inode_lock(inode); 6502 /* 6503 * Update modification times of quota files when userspace can 6504 * start looking at them. If we fail, we return success anyway since 6505 * this is not a hard failure and quotas are already disabled. 6506 */ 6507 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6508 if (IS_ERR(handle)) { 6509 err = PTR_ERR(handle); 6510 goto out_unlock; 6511 } 6512 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 6513 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 6514 inode->i_mtime = inode->i_ctime = current_time(inode); 6515 err = ext4_mark_inode_dirty(handle, inode); 6516 ext4_journal_stop(handle); 6517 out_unlock: 6518 inode_unlock(inode); 6519 out_put: 6520 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 6521 iput(inode); 6522 return err; 6523 out: 6524 return dquot_quota_off(sb, type); 6525 } 6526 6527 /* Read data from quotafile - avoid pagecache and such because we cannot afford 6528 * acquiring the locks... As quota files are never truncated and quota code 6529 * itself serializes the operations (and no one else should touch the files) 6530 * we don't have to be afraid of races */ 6531 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 6532 size_t len, loff_t off) 6533 { 6534 struct inode *inode = sb_dqopt(sb)->files[type]; 6535 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6536 int offset = off & (sb->s_blocksize - 1); 6537 int tocopy; 6538 size_t toread; 6539 struct buffer_head *bh; 6540 loff_t i_size = i_size_read(inode); 6541 6542 if (off > i_size) 6543 return 0; 6544 if (off+len > i_size) 6545 len = i_size-off; 6546 toread = len; 6547 while (toread > 0) { 6548 tocopy = sb->s_blocksize - offset < toread ? 6549 sb->s_blocksize - offset : toread; 6550 bh = ext4_bread(NULL, inode, blk, 0); 6551 if (IS_ERR(bh)) 6552 return PTR_ERR(bh); 6553 if (!bh) /* A hole? */ 6554 memset(data, 0, tocopy); 6555 else 6556 memcpy(data, bh->b_data+offset, tocopy); 6557 brelse(bh); 6558 offset = 0; 6559 toread -= tocopy; 6560 data += tocopy; 6561 blk++; 6562 } 6563 return len; 6564 } 6565 6566 /* Write to quotafile (we know the transaction is already started and has 6567 * enough credits) */ 6568 static ssize_t ext4_quota_write(struct super_block *sb, int type, 6569 const char *data, size_t len, loff_t off) 6570 { 6571 struct inode *inode = sb_dqopt(sb)->files[type]; 6572 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 6573 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 6574 int retries = 0; 6575 struct buffer_head *bh; 6576 handle_t *handle = journal_current_handle(); 6577 6578 if (EXT4_SB(sb)->s_journal && !handle) { 6579 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6580 " cancelled because transaction is not started", 6581 (unsigned long long)off, (unsigned long long)len); 6582 return -EIO; 6583 } 6584 /* 6585 * Since we account only one data block in transaction credits, 6586 * then it is impossible to cross a block boundary. 6587 */ 6588 if (sb->s_blocksize - offset < len) { 6589 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 6590 " cancelled because not block aligned", 6591 (unsigned long long)off, (unsigned long long)len); 6592 return -EIO; 6593 } 6594 6595 do { 6596 bh = ext4_bread(handle, inode, blk, 6597 EXT4_GET_BLOCKS_CREATE | 6598 EXT4_GET_BLOCKS_METADATA_NOFAIL); 6599 } while (PTR_ERR(bh) == -ENOSPC && 6600 ext4_should_retry_alloc(inode->i_sb, &retries)); 6601 if (IS_ERR(bh)) 6602 return PTR_ERR(bh); 6603 if (!bh) 6604 goto out; 6605 BUFFER_TRACE(bh, "get write access"); 6606 err = ext4_journal_get_write_access(handle, bh); 6607 if (err) { 6608 brelse(bh); 6609 return err; 6610 } 6611 lock_buffer(bh); 6612 memcpy(bh->b_data+offset, data, len); 6613 flush_dcache_page(bh->b_page); 6614 unlock_buffer(bh); 6615 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6616 brelse(bh); 6617 out: 6618 if (inode->i_size < off + len) { 6619 i_size_write(inode, off + len); 6620 EXT4_I(inode)->i_disksize = inode->i_size; 6621 err2 = ext4_mark_inode_dirty(handle, inode); 6622 if (unlikely(err2 && !err)) 6623 err = err2; 6624 } 6625 return err ? err : len; 6626 } 6627 #endif 6628 6629 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6630 const char *dev_name, void *data) 6631 { 6632 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6633 } 6634 6635 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6636 static inline void register_as_ext2(void) 6637 { 6638 int err = register_filesystem(&ext2_fs_type); 6639 if (err) 6640 printk(KERN_WARNING 6641 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6642 } 6643 6644 static inline void unregister_as_ext2(void) 6645 { 6646 unregister_filesystem(&ext2_fs_type); 6647 } 6648 6649 static inline int ext2_feature_set_ok(struct super_block *sb) 6650 { 6651 if (ext4_has_unknown_ext2_incompat_features(sb)) 6652 return 0; 6653 if (sb_rdonly(sb)) 6654 return 1; 6655 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6656 return 0; 6657 return 1; 6658 } 6659 #else 6660 static inline void register_as_ext2(void) { } 6661 static inline void unregister_as_ext2(void) { } 6662 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6663 #endif 6664 6665 static inline void register_as_ext3(void) 6666 { 6667 int err = register_filesystem(&ext3_fs_type); 6668 if (err) 6669 printk(KERN_WARNING 6670 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6671 } 6672 6673 static inline void unregister_as_ext3(void) 6674 { 6675 unregister_filesystem(&ext3_fs_type); 6676 } 6677 6678 static inline int ext3_feature_set_ok(struct super_block *sb) 6679 { 6680 if (ext4_has_unknown_ext3_incompat_features(sb)) 6681 return 0; 6682 if (!ext4_has_feature_journal(sb)) 6683 return 0; 6684 if (sb_rdonly(sb)) 6685 return 1; 6686 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6687 return 0; 6688 return 1; 6689 } 6690 6691 static struct file_system_type ext4_fs_type = { 6692 .owner = THIS_MODULE, 6693 .name = "ext4", 6694 .mount = ext4_mount, 6695 .kill_sb = kill_block_super, 6696 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 6697 }; 6698 MODULE_ALIAS_FS("ext4"); 6699 6700 /* Shared across all ext4 file systems */ 6701 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6702 6703 static int __init ext4_init_fs(void) 6704 { 6705 int i, err; 6706 6707 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6708 ext4_li_info = NULL; 6709 6710 /* Build-time check for flags consistency */ 6711 ext4_check_flag_values(); 6712 6713 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6714 init_waitqueue_head(&ext4__ioend_wq[i]); 6715 6716 err = ext4_init_es(); 6717 if (err) 6718 return err; 6719 6720 err = ext4_init_pending(); 6721 if (err) 6722 goto out7; 6723 6724 err = ext4_init_post_read_processing(); 6725 if (err) 6726 goto out6; 6727 6728 err = ext4_init_pageio(); 6729 if (err) 6730 goto out5; 6731 6732 err = ext4_init_system_zone(); 6733 if (err) 6734 goto out4; 6735 6736 err = ext4_init_sysfs(); 6737 if (err) 6738 goto out3; 6739 6740 err = ext4_init_mballoc(); 6741 if (err) 6742 goto out2; 6743 err = init_inodecache(); 6744 if (err) 6745 goto out1; 6746 6747 err = ext4_fc_init_dentry_cache(); 6748 if (err) 6749 goto out05; 6750 6751 register_as_ext3(); 6752 register_as_ext2(); 6753 err = register_filesystem(&ext4_fs_type); 6754 if (err) 6755 goto out; 6756 6757 return 0; 6758 out: 6759 unregister_as_ext2(); 6760 unregister_as_ext3(); 6761 out05: 6762 destroy_inodecache(); 6763 out1: 6764 ext4_exit_mballoc(); 6765 out2: 6766 ext4_exit_sysfs(); 6767 out3: 6768 ext4_exit_system_zone(); 6769 out4: 6770 ext4_exit_pageio(); 6771 out5: 6772 ext4_exit_post_read_processing(); 6773 out6: 6774 ext4_exit_pending(); 6775 out7: 6776 ext4_exit_es(); 6777 6778 return err; 6779 } 6780 6781 static void __exit ext4_exit_fs(void) 6782 { 6783 ext4_destroy_lazyinit_thread(); 6784 unregister_as_ext2(); 6785 unregister_as_ext3(); 6786 unregister_filesystem(&ext4_fs_type); 6787 destroy_inodecache(); 6788 ext4_exit_mballoc(); 6789 ext4_exit_sysfs(); 6790 ext4_exit_system_zone(); 6791 ext4_exit_pageio(); 6792 ext4_exit_post_read_processing(); 6793 ext4_exit_es(); 6794 ext4_exit_pending(); 6795 } 6796 6797 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6798 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6799 MODULE_LICENSE("GPL"); 6800 MODULE_SOFTDEP("pre: crc32c"); 6801 module_init(ext4_init_fs) 6802 module_exit(ext4_exit_fs) 6803