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/uaccess.h> 43 #include <linux/iversion.h> 44 #include <linux/unicode.h> 45 #include <linux/part_stat.h> 46 #include <linux/kthread.h> 47 #include <linux/freezer.h> 48 #include <linux/fsnotify.h> 49 #include <linux/fs_context.h> 50 #include <linux/fs_parser.h> 51 52 #include "ext4.h" 53 #include "ext4_extents.h" /* Needed for trace points definition */ 54 #include "ext4_jbd2.h" 55 #include "xattr.h" 56 #include "acl.h" 57 #include "mballoc.h" 58 #include "fsmap.h" 59 60 #define CREATE_TRACE_POINTS 61 #include <trace/events/ext4.h> 62 63 static struct ext4_lazy_init *ext4_li_info; 64 static DEFINE_MUTEX(ext4_li_mtx); 65 static struct ratelimit_state ext4_mount_msg_ratelimit; 66 67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 68 unsigned long journal_devnum); 69 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 70 static void ext4_update_super(struct super_block *sb); 71 static int ext4_commit_super(struct super_block *sb); 72 static int ext4_mark_recovery_complete(struct super_block *sb, 73 struct ext4_super_block *es); 74 static int ext4_clear_journal_err(struct super_block *sb, 75 struct ext4_super_block *es); 76 static int ext4_sync_fs(struct super_block *sb, int wait); 77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 78 static int ext4_unfreeze(struct super_block *sb); 79 static int ext4_freeze(struct super_block *sb); 80 static inline int ext2_feature_set_ok(struct super_block *sb); 81 static inline int ext3_feature_set_ok(struct super_block *sb); 82 static void ext4_destroy_lazyinit_thread(void); 83 static void ext4_unregister_li_request(struct super_block *sb); 84 static void ext4_clear_request_list(void); 85 static struct inode *ext4_get_journal_inode(struct super_block *sb, 86 unsigned int journal_inum); 87 static int ext4_validate_options(struct fs_context *fc); 88 static int ext4_check_opt_consistency(struct fs_context *fc, 89 struct super_block *sb); 90 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb); 91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param); 92 static int ext4_get_tree(struct fs_context *fc); 93 static int ext4_reconfigure(struct fs_context *fc); 94 static void ext4_fc_free(struct fs_context *fc); 95 static int ext4_init_fs_context(struct fs_context *fc); 96 static const struct fs_parameter_spec ext4_param_specs[]; 97 98 /* 99 * Lock ordering 100 * 101 * page fault path: 102 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start 103 * -> page lock -> i_data_sem (rw) 104 * 105 * buffered write path: 106 * sb_start_write -> i_mutex -> mmap_lock 107 * sb_start_write -> i_mutex -> transaction start -> page lock -> 108 * i_data_sem (rw) 109 * 110 * truncate: 111 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) -> 112 * page lock 113 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start -> 114 * i_data_sem (rw) 115 * 116 * direct IO: 117 * sb_start_write -> i_mutex -> mmap_lock 118 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 119 * 120 * writepages: 121 * transaction start -> page lock(s) -> i_data_sem (rw) 122 */ 123 124 static const struct fs_context_operations ext4_context_ops = { 125 .parse_param = ext4_parse_param, 126 .get_tree = ext4_get_tree, 127 .reconfigure = ext4_reconfigure, 128 .free = ext4_fc_free, 129 }; 130 131 132 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 133 static struct file_system_type ext2_fs_type = { 134 .owner = THIS_MODULE, 135 .name = "ext2", 136 .init_fs_context = ext4_init_fs_context, 137 .parameters = ext4_param_specs, 138 .kill_sb = kill_block_super, 139 .fs_flags = FS_REQUIRES_DEV, 140 }; 141 MODULE_ALIAS_FS("ext2"); 142 MODULE_ALIAS("ext2"); 143 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type) 144 #else 145 #define IS_EXT2_SB(sb) (0) 146 #endif 147 148 149 static struct file_system_type ext3_fs_type = { 150 .owner = THIS_MODULE, 151 .name = "ext3", 152 .init_fs_context = ext4_init_fs_context, 153 .parameters = ext4_param_specs, 154 .kill_sb = kill_block_super, 155 .fs_flags = FS_REQUIRES_DEV, 156 }; 157 MODULE_ALIAS_FS("ext3"); 158 MODULE_ALIAS("ext3"); 159 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type) 160 161 162 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, 163 bh_end_io_t *end_io) 164 { 165 /* 166 * buffer's verified bit is no longer valid after reading from 167 * disk again due to write out error, clear it to make sure we 168 * recheck the buffer contents. 169 */ 170 clear_buffer_verified(bh); 171 172 bh->b_end_io = end_io ? end_io : end_buffer_read_sync; 173 get_bh(bh); 174 submit_bh(REQ_OP_READ | op_flags, bh); 175 } 176 177 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags, 178 bh_end_io_t *end_io) 179 { 180 BUG_ON(!buffer_locked(bh)); 181 182 if (ext4_buffer_uptodate(bh)) { 183 unlock_buffer(bh); 184 return; 185 } 186 __ext4_read_bh(bh, op_flags, end_io); 187 } 188 189 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io) 190 { 191 BUG_ON(!buffer_locked(bh)); 192 193 if (ext4_buffer_uptodate(bh)) { 194 unlock_buffer(bh); 195 return 0; 196 } 197 198 __ext4_read_bh(bh, op_flags, end_io); 199 200 wait_on_buffer(bh); 201 if (buffer_uptodate(bh)) 202 return 0; 203 return -EIO; 204 } 205 206 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 207 { 208 lock_buffer(bh); 209 if (!wait) { 210 ext4_read_bh_nowait(bh, op_flags, NULL); 211 return 0; 212 } 213 return ext4_read_bh(bh, op_flags, NULL); 214 } 215 216 /* 217 * This works like __bread_gfp() except it uses ERR_PTR for error 218 * returns. Currently with sb_bread it's impossible to distinguish 219 * between ENOMEM and EIO situations (since both result in a NULL 220 * return. 221 */ 222 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb, 223 sector_t block, 224 blk_opf_t op_flags, gfp_t gfp) 225 { 226 struct buffer_head *bh; 227 int ret; 228 229 bh = sb_getblk_gfp(sb, block, gfp); 230 if (bh == NULL) 231 return ERR_PTR(-ENOMEM); 232 if (ext4_buffer_uptodate(bh)) 233 return bh; 234 235 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true); 236 if (ret) { 237 put_bh(bh); 238 return ERR_PTR(ret); 239 } 240 return bh; 241 } 242 243 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block, 244 blk_opf_t op_flags) 245 { 246 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE); 247 } 248 249 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb, 250 sector_t block) 251 { 252 return __ext4_sb_bread_gfp(sb, block, 0, 0); 253 } 254 255 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block) 256 { 257 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0); 258 259 if (likely(bh)) { 260 if (trylock_buffer(bh)) 261 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL); 262 brelse(bh); 263 } 264 } 265 266 static int ext4_verify_csum_type(struct super_block *sb, 267 struct ext4_super_block *es) 268 { 269 if (!ext4_has_feature_metadata_csum(sb)) 270 return 1; 271 272 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 273 } 274 275 __le32 ext4_superblock_csum(struct super_block *sb, 276 struct ext4_super_block *es) 277 { 278 struct ext4_sb_info *sbi = EXT4_SB(sb); 279 int offset = offsetof(struct ext4_super_block, s_checksum); 280 __u32 csum; 281 282 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 283 284 return cpu_to_le32(csum); 285 } 286 287 static int ext4_superblock_csum_verify(struct super_block *sb, 288 struct ext4_super_block *es) 289 { 290 if (!ext4_has_metadata_csum(sb)) 291 return 1; 292 293 return es->s_checksum == ext4_superblock_csum(sb, es); 294 } 295 296 void ext4_superblock_csum_set(struct super_block *sb) 297 { 298 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 299 300 if (!ext4_has_metadata_csum(sb)) 301 return; 302 303 es->s_checksum = ext4_superblock_csum(sb, es); 304 } 305 306 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 307 struct ext4_group_desc *bg) 308 { 309 return le32_to_cpu(bg->bg_block_bitmap_lo) | 310 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 311 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 312 } 313 314 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 315 struct ext4_group_desc *bg) 316 { 317 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 318 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 319 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 320 } 321 322 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 323 struct ext4_group_desc *bg) 324 { 325 return le32_to_cpu(bg->bg_inode_table_lo) | 326 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 327 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 328 } 329 330 __u32 ext4_free_group_clusters(struct super_block *sb, 331 struct ext4_group_desc *bg) 332 { 333 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 334 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 335 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 336 } 337 338 __u32 ext4_free_inodes_count(struct super_block *sb, 339 struct ext4_group_desc *bg) 340 { 341 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 342 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 343 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 344 } 345 346 __u32 ext4_used_dirs_count(struct super_block *sb, 347 struct ext4_group_desc *bg) 348 { 349 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 350 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 351 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 352 } 353 354 __u32 ext4_itable_unused_count(struct super_block *sb, 355 struct ext4_group_desc *bg) 356 { 357 return le16_to_cpu(bg->bg_itable_unused_lo) | 358 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 359 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 360 } 361 362 void ext4_block_bitmap_set(struct super_block *sb, 363 struct ext4_group_desc *bg, ext4_fsblk_t blk) 364 { 365 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 366 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 367 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 368 } 369 370 void ext4_inode_bitmap_set(struct super_block *sb, 371 struct ext4_group_desc *bg, ext4_fsblk_t blk) 372 { 373 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 374 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 375 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 376 } 377 378 void ext4_inode_table_set(struct super_block *sb, 379 struct ext4_group_desc *bg, ext4_fsblk_t blk) 380 { 381 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 382 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 383 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 384 } 385 386 void ext4_free_group_clusters_set(struct super_block *sb, 387 struct ext4_group_desc *bg, __u32 count) 388 { 389 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 390 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 391 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 392 } 393 394 void ext4_free_inodes_set(struct super_block *sb, 395 struct ext4_group_desc *bg, __u32 count) 396 { 397 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 398 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 399 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 400 } 401 402 void ext4_used_dirs_set(struct super_block *sb, 403 struct ext4_group_desc *bg, __u32 count) 404 { 405 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 406 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 407 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 408 } 409 410 void ext4_itable_unused_set(struct super_block *sb, 411 struct ext4_group_desc *bg, __u32 count) 412 { 413 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 414 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 415 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 416 } 417 418 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now) 419 { 420 now = clamp_val(now, 0, (1ull << 40) - 1); 421 422 *lo = cpu_to_le32(lower_32_bits(now)); 423 *hi = upper_32_bits(now); 424 } 425 426 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 427 { 428 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 429 } 430 #define ext4_update_tstamp(es, tstamp) \ 431 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \ 432 ktime_get_real_seconds()) 433 #define ext4_get_tstamp(es, tstamp) \ 434 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 435 436 /* 437 * The del_gendisk() function uninitializes the disk-specific data 438 * structures, including the bdi structure, without telling anyone 439 * else. Once this happens, any attempt to call mark_buffer_dirty() 440 * (for example, by ext4_commit_super), will cause a kernel OOPS. 441 * This is a kludge to prevent these oops until we can put in a proper 442 * hook in del_gendisk() to inform the VFS and file system layers. 443 */ 444 static int block_device_ejected(struct super_block *sb) 445 { 446 struct inode *bd_inode = sb->s_bdev->bd_inode; 447 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 448 449 return bdi->dev == NULL; 450 } 451 452 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 453 { 454 struct super_block *sb = journal->j_private; 455 struct ext4_sb_info *sbi = EXT4_SB(sb); 456 int error = is_journal_aborted(journal); 457 struct ext4_journal_cb_entry *jce; 458 459 BUG_ON(txn->t_state == T_FINISHED); 460 461 ext4_process_freed_data(sb, txn->t_tid); 462 463 spin_lock(&sbi->s_md_lock); 464 while (!list_empty(&txn->t_private_list)) { 465 jce = list_entry(txn->t_private_list.next, 466 struct ext4_journal_cb_entry, jce_list); 467 list_del_init(&jce->jce_list); 468 spin_unlock(&sbi->s_md_lock); 469 jce->jce_func(sb, jce, error); 470 spin_lock(&sbi->s_md_lock); 471 } 472 spin_unlock(&sbi->s_md_lock); 473 } 474 475 /* 476 * This writepage callback for write_cache_pages() 477 * takes care of a few cases after page cleaning. 478 * 479 * write_cache_pages() already checks for dirty pages 480 * and calls clear_page_dirty_for_io(), which we want, 481 * to write protect the pages. 482 * 483 * However, we may have to redirty a page (see below.) 484 */ 485 static int ext4_journalled_writepage_callback(struct page *page, 486 struct writeback_control *wbc, 487 void *data) 488 { 489 transaction_t *transaction = (transaction_t *) data; 490 struct buffer_head *bh, *head; 491 struct journal_head *jh; 492 493 bh = head = page_buffers(page); 494 do { 495 /* 496 * We have to redirty a page in these cases: 497 * 1) If buffer is dirty, it means the page was dirty because it 498 * contains a buffer that needs checkpointing. So the dirty bit 499 * needs to be preserved so that checkpointing writes the buffer 500 * properly. 501 * 2) If buffer is not part of the committing transaction 502 * (we may have just accidentally come across this buffer because 503 * inode range tracking is not exact) or if the currently running 504 * transaction already contains this buffer as well, dirty bit 505 * needs to be preserved so that the buffer gets writeprotected 506 * properly on running transaction's commit. 507 */ 508 jh = bh2jh(bh); 509 if (buffer_dirty(bh) || 510 (jh && (jh->b_transaction != transaction || 511 jh->b_next_transaction))) { 512 redirty_page_for_writepage(wbc, page); 513 goto out; 514 } 515 } while ((bh = bh->b_this_page) != head); 516 517 out: 518 return AOP_WRITEPAGE_ACTIVATE; 519 } 520 521 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode) 522 { 523 struct address_space *mapping = jinode->i_vfs_inode->i_mapping; 524 struct writeback_control wbc = { 525 .sync_mode = WB_SYNC_ALL, 526 .nr_to_write = LONG_MAX, 527 .range_start = jinode->i_dirty_start, 528 .range_end = jinode->i_dirty_end, 529 }; 530 531 return write_cache_pages(mapping, &wbc, 532 ext4_journalled_writepage_callback, 533 jinode->i_transaction); 534 } 535 536 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode) 537 { 538 int ret; 539 540 if (ext4_should_journal_data(jinode->i_vfs_inode)) 541 ret = ext4_journalled_submit_inode_data_buffers(jinode); 542 else 543 ret = jbd2_journal_submit_inode_data_buffers(jinode); 544 545 return ret; 546 } 547 548 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode) 549 { 550 int ret = 0; 551 552 if (!ext4_should_journal_data(jinode->i_vfs_inode)) 553 ret = jbd2_journal_finish_inode_data_buffers(jinode); 554 555 return ret; 556 } 557 558 static bool system_going_down(void) 559 { 560 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 561 || system_state == SYSTEM_RESTART; 562 } 563 564 struct ext4_err_translation { 565 int code; 566 int errno; 567 }; 568 569 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err } 570 571 static struct ext4_err_translation err_translation[] = { 572 EXT4_ERR_TRANSLATE(EIO), 573 EXT4_ERR_TRANSLATE(ENOMEM), 574 EXT4_ERR_TRANSLATE(EFSBADCRC), 575 EXT4_ERR_TRANSLATE(EFSCORRUPTED), 576 EXT4_ERR_TRANSLATE(ENOSPC), 577 EXT4_ERR_TRANSLATE(ENOKEY), 578 EXT4_ERR_TRANSLATE(EROFS), 579 EXT4_ERR_TRANSLATE(EFBIG), 580 EXT4_ERR_TRANSLATE(EEXIST), 581 EXT4_ERR_TRANSLATE(ERANGE), 582 EXT4_ERR_TRANSLATE(EOVERFLOW), 583 EXT4_ERR_TRANSLATE(EBUSY), 584 EXT4_ERR_TRANSLATE(ENOTDIR), 585 EXT4_ERR_TRANSLATE(ENOTEMPTY), 586 EXT4_ERR_TRANSLATE(ESHUTDOWN), 587 EXT4_ERR_TRANSLATE(EFAULT), 588 }; 589 590 static int ext4_errno_to_code(int errno) 591 { 592 int i; 593 594 for (i = 0; i < ARRAY_SIZE(err_translation); i++) 595 if (err_translation[i].errno == errno) 596 return err_translation[i].code; 597 return EXT4_ERR_UNKNOWN; 598 } 599 600 static void save_error_info(struct super_block *sb, int error, 601 __u32 ino, __u64 block, 602 const char *func, unsigned int line) 603 { 604 struct ext4_sb_info *sbi = EXT4_SB(sb); 605 606 /* We default to EFSCORRUPTED error... */ 607 if (error == 0) 608 error = EFSCORRUPTED; 609 610 spin_lock(&sbi->s_error_lock); 611 sbi->s_add_error_count++; 612 sbi->s_last_error_code = error; 613 sbi->s_last_error_line = line; 614 sbi->s_last_error_ino = ino; 615 sbi->s_last_error_block = block; 616 sbi->s_last_error_func = func; 617 sbi->s_last_error_time = ktime_get_real_seconds(); 618 if (!sbi->s_first_error_time) { 619 sbi->s_first_error_code = error; 620 sbi->s_first_error_line = line; 621 sbi->s_first_error_ino = ino; 622 sbi->s_first_error_block = block; 623 sbi->s_first_error_func = func; 624 sbi->s_first_error_time = sbi->s_last_error_time; 625 } 626 spin_unlock(&sbi->s_error_lock); 627 } 628 629 /* Deal with the reporting of failure conditions on a filesystem such as 630 * inconsistencies detected or read IO failures. 631 * 632 * On ext2, we can store the error state of the filesystem in the 633 * superblock. That is not possible on ext4, because we may have other 634 * write ordering constraints on the superblock which prevent us from 635 * writing it out straight away; and given that the journal is about to 636 * be aborted, we can't rely on the current, or future, transactions to 637 * write out the superblock safely. 638 * 639 * We'll just use the jbd2_journal_abort() error code to record an error in 640 * the journal instead. On recovery, the journal will complain about 641 * that error until we've noted it down and cleared it. 642 * 643 * If force_ro is set, we unconditionally force the filesystem into an 644 * ABORT|READONLY state, unless the error response on the fs has been set to 645 * panic in which case we take the easy way out and panic immediately. This is 646 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM 647 * at a critical moment in log management. 648 */ 649 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error, 650 __u32 ino, __u64 block, 651 const char *func, unsigned int line) 652 { 653 journal_t *journal = EXT4_SB(sb)->s_journal; 654 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT); 655 656 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 657 if (test_opt(sb, WARN_ON_ERROR)) 658 WARN_ON_ONCE(1); 659 660 if (!continue_fs && !sb_rdonly(sb)) { 661 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED); 662 if (journal) 663 jbd2_journal_abort(journal, -EIO); 664 } 665 666 if (!bdev_read_only(sb->s_bdev)) { 667 save_error_info(sb, error, ino, block, func, line); 668 /* 669 * In case the fs should keep running, we need to writeout 670 * superblock through the journal. Due to lock ordering 671 * constraints, it may not be safe to do it right here so we 672 * defer superblock flushing to a workqueue. 673 */ 674 if (continue_fs && journal) 675 schedule_work(&EXT4_SB(sb)->s_error_work); 676 else 677 ext4_commit_super(sb); 678 } 679 680 /* 681 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 682 * could panic during 'reboot -f' as the underlying device got already 683 * disabled. 684 */ 685 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 686 panic("EXT4-fs (device %s): panic forced after error\n", 687 sb->s_id); 688 } 689 690 if (sb_rdonly(sb) || continue_fs) 691 return; 692 693 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 694 /* 695 * Make sure updated value of ->s_mount_flags will be visible before 696 * ->s_flags update 697 */ 698 smp_wmb(); 699 sb->s_flags |= SB_RDONLY; 700 } 701 702 static void flush_stashed_error_work(struct work_struct *work) 703 { 704 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info, 705 s_error_work); 706 journal_t *journal = sbi->s_journal; 707 handle_t *handle; 708 709 /* 710 * If the journal is still running, we have to write out superblock 711 * through the journal to avoid collisions of other journalled sb 712 * updates. 713 * 714 * We use directly jbd2 functions here to avoid recursing back into 715 * ext4 error handling code during handling of previous errors. 716 */ 717 if (!sb_rdonly(sbi->s_sb) && journal) { 718 struct buffer_head *sbh = sbi->s_sbh; 719 handle = jbd2_journal_start(journal, 1); 720 if (IS_ERR(handle)) 721 goto write_directly; 722 if (jbd2_journal_get_write_access(handle, sbh)) { 723 jbd2_journal_stop(handle); 724 goto write_directly; 725 } 726 ext4_update_super(sbi->s_sb); 727 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 728 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to " 729 "superblock detected"); 730 clear_buffer_write_io_error(sbh); 731 set_buffer_uptodate(sbh); 732 } 733 734 if (jbd2_journal_dirty_metadata(handle, sbh)) { 735 jbd2_journal_stop(handle); 736 goto write_directly; 737 } 738 jbd2_journal_stop(handle); 739 ext4_notify_error_sysfs(sbi); 740 return; 741 } 742 write_directly: 743 /* 744 * Write through journal failed. Write sb directly to get error info 745 * out and hope for the best. 746 */ 747 ext4_commit_super(sbi->s_sb); 748 ext4_notify_error_sysfs(sbi); 749 } 750 751 #define ext4_error_ratelimit(sb) \ 752 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 753 "EXT4-fs error") 754 755 void __ext4_error(struct super_block *sb, const char *function, 756 unsigned int line, bool force_ro, int error, __u64 block, 757 const char *fmt, ...) 758 { 759 struct va_format vaf; 760 va_list args; 761 762 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 763 return; 764 765 trace_ext4_error(sb, function, line); 766 if (ext4_error_ratelimit(sb)) { 767 va_start(args, fmt); 768 vaf.fmt = fmt; 769 vaf.va = &args; 770 printk(KERN_CRIT 771 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 772 sb->s_id, function, line, current->comm, &vaf); 773 va_end(args); 774 } 775 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED); 776 777 ext4_handle_error(sb, force_ro, error, 0, block, function, line); 778 } 779 780 void __ext4_error_inode(struct inode *inode, const char *function, 781 unsigned int line, ext4_fsblk_t block, int error, 782 const char *fmt, ...) 783 { 784 va_list args; 785 struct va_format vaf; 786 787 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 788 return; 789 790 trace_ext4_error(inode->i_sb, function, line); 791 if (ext4_error_ratelimit(inode->i_sb)) { 792 va_start(args, fmt); 793 vaf.fmt = fmt; 794 vaf.va = &args; 795 if (block) 796 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 797 "inode #%lu: block %llu: comm %s: %pV\n", 798 inode->i_sb->s_id, function, line, inode->i_ino, 799 block, current->comm, &vaf); 800 else 801 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 802 "inode #%lu: comm %s: %pV\n", 803 inode->i_sb->s_id, function, line, inode->i_ino, 804 current->comm, &vaf); 805 va_end(args); 806 } 807 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED); 808 809 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block, 810 function, line); 811 } 812 813 void __ext4_error_file(struct file *file, const char *function, 814 unsigned int line, ext4_fsblk_t block, 815 const char *fmt, ...) 816 { 817 va_list args; 818 struct va_format vaf; 819 struct inode *inode = file_inode(file); 820 char pathname[80], *path; 821 822 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 823 return; 824 825 trace_ext4_error(inode->i_sb, function, line); 826 if (ext4_error_ratelimit(inode->i_sb)) { 827 path = file_path(file, pathname, sizeof(pathname)); 828 if (IS_ERR(path)) 829 path = "(unknown)"; 830 va_start(args, fmt); 831 vaf.fmt = fmt; 832 vaf.va = &args; 833 if (block) 834 printk(KERN_CRIT 835 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 836 "block %llu: comm %s: path %s: %pV\n", 837 inode->i_sb->s_id, function, line, inode->i_ino, 838 block, current->comm, path, &vaf); 839 else 840 printk(KERN_CRIT 841 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 842 "comm %s: path %s: %pV\n", 843 inode->i_sb->s_id, function, line, inode->i_ino, 844 current->comm, path, &vaf); 845 va_end(args); 846 } 847 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED); 848 849 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block, 850 function, line); 851 } 852 853 const char *ext4_decode_error(struct super_block *sb, int errno, 854 char nbuf[16]) 855 { 856 char *errstr = NULL; 857 858 switch (errno) { 859 case -EFSCORRUPTED: 860 errstr = "Corrupt filesystem"; 861 break; 862 case -EFSBADCRC: 863 errstr = "Filesystem failed CRC"; 864 break; 865 case -EIO: 866 errstr = "IO failure"; 867 break; 868 case -ENOMEM: 869 errstr = "Out of memory"; 870 break; 871 case -EROFS: 872 if (!sb || (EXT4_SB(sb)->s_journal && 873 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 874 errstr = "Journal has aborted"; 875 else 876 errstr = "Readonly filesystem"; 877 break; 878 default: 879 /* If the caller passed in an extra buffer for unknown 880 * errors, textualise them now. Else we just return 881 * NULL. */ 882 if (nbuf) { 883 /* Check for truncated error codes... */ 884 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 885 errstr = nbuf; 886 } 887 break; 888 } 889 890 return errstr; 891 } 892 893 /* __ext4_std_error decodes expected errors from journaling functions 894 * automatically and invokes the appropriate error response. */ 895 896 void __ext4_std_error(struct super_block *sb, const char *function, 897 unsigned int line, int errno) 898 { 899 char nbuf[16]; 900 const char *errstr; 901 902 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 903 return; 904 905 /* Special case: if the error is EROFS, and we're not already 906 * inside a transaction, then there's really no point in logging 907 * an error. */ 908 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 909 return; 910 911 if (ext4_error_ratelimit(sb)) { 912 errstr = ext4_decode_error(sb, errno, nbuf); 913 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 914 sb->s_id, function, line, errstr); 915 } 916 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED); 917 918 ext4_handle_error(sb, false, -errno, 0, 0, function, line); 919 } 920 921 void __ext4_msg(struct super_block *sb, 922 const char *prefix, const char *fmt, ...) 923 { 924 struct va_format vaf; 925 va_list args; 926 927 if (sb) { 928 atomic_inc(&EXT4_SB(sb)->s_msg_count); 929 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), 930 "EXT4-fs")) 931 return; 932 } 933 934 va_start(args, fmt); 935 vaf.fmt = fmt; 936 vaf.va = &args; 937 if (sb) 938 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 939 else 940 printk("%sEXT4-fs: %pV\n", prefix, &vaf); 941 va_end(args); 942 } 943 944 static int ext4_warning_ratelimit(struct super_block *sb) 945 { 946 atomic_inc(&EXT4_SB(sb)->s_warning_count); 947 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), 948 "EXT4-fs warning"); 949 } 950 951 void __ext4_warning(struct super_block *sb, const char *function, 952 unsigned int line, const char *fmt, ...) 953 { 954 struct va_format vaf; 955 va_list args; 956 957 if (!ext4_warning_ratelimit(sb)) 958 return; 959 960 va_start(args, fmt); 961 vaf.fmt = fmt; 962 vaf.va = &args; 963 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 964 sb->s_id, function, line, &vaf); 965 va_end(args); 966 } 967 968 void __ext4_warning_inode(const struct inode *inode, const char *function, 969 unsigned int line, const char *fmt, ...) 970 { 971 struct va_format vaf; 972 va_list args; 973 974 if (!ext4_warning_ratelimit(inode->i_sb)) 975 return; 976 977 va_start(args, fmt); 978 vaf.fmt = fmt; 979 vaf.va = &args; 980 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 981 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 982 function, line, inode->i_ino, current->comm, &vaf); 983 va_end(args); 984 } 985 986 void __ext4_grp_locked_error(const char *function, unsigned int line, 987 struct super_block *sb, ext4_group_t grp, 988 unsigned long ino, ext4_fsblk_t block, 989 const char *fmt, ...) 990 __releases(bitlock) 991 __acquires(bitlock) 992 { 993 struct va_format vaf; 994 va_list args; 995 996 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 997 return; 998 999 trace_ext4_error(sb, function, line); 1000 if (ext4_error_ratelimit(sb)) { 1001 va_start(args, fmt); 1002 vaf.fmt = fmt; 1003 vaf.va = &args; 1004 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 1005 sb->s_id, function, line, grp); 1006 if (ino) 1007 printk(KERN_CONT "inode %lu: ", ino); 1008 if (block) 1009 printk(KERN_CONT "block %llu:", 1010 (unsigned long long) block); 1011 printk(KERN_CONT "%pV\n", &vaf); 1012 va_end(args); 1013 } 1014 1015 if (test_opt(sb, ERRORS_CONT)) { 1016 if (test_opt(sb, WARN_ON_ERROR)) 1017 WARN_ON_ONCE(1); 1018 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 1019 if (!bdev_read_only(sb->s_bdev)) { 1020 save_error_info(sb, EFSCORRUPTED, ino, block, function, 1021 line); 1022 schedule_work(&EXT4_SB(sb)->s_error_work); 1023 } 1024 return; 1025 } 1026 ext4_unlock_group(sb, grp); 1027 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line); 1028 /* 1029 * We only get here in the ERRORS_RO case; relocking the group 1030 * may be dangerous, but nothing bad will happen since the 1031 * filesystem will have already been marked read/only and the 1032 * journal has been aborted. We return 1 as a hint to callers 1033 * who might what to use the return value from 1034 * ext4_grp_locked_error() to distinguish between the 1035 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 1036 * aggressively from the ext4 function in question, with a 1037 * more appropriate error code. 1038 */ 1039 ext4_lock_group(sb, grp); 1040 return; 1041 } 1042 1043 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 1044 ext4_group_t group, 1045 unsigned int flags) 1046 { 1047 struct ext4_sb_info *sbi = EXT4_SB(sb); 1048 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1049 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 1050 int ret; 1051 1052 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 1053 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 1054 &grp->bb_state); 1055 if (!ret) 1056 percpu_counter_sub(&sbi->s_freeclusters_counter, 1057 grp->bb_free); 1058 } 1059 1060 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 1061 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 1062 &grp->bb_state); 1063 if (!ret && gdp) { 1064 int count; 1065 1066 count = ext4_free_inodes_count(sb, gdp); 1067 percpu_counter_sub(&sbi->s_freeinodes_counter, 1068 count); 1069 } 1070 } 1071 } 1072 1073 void ext4_update_dynamic_rev(struct super_block *sb) 1074 { 1075 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1076 1077 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 1078 return; 1079 1080 ext4_warning(sb, 1081 "updating to rev %d because of new feature flag, " 1082 "running e2fsck is recommended", 1083 EXT4_DYNAMIC_REV); 1084 1085 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 1086 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 1087 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 1088 /* leave es->s_feature_*compat flags alone */ 1089 /* es->s_uuid will be set by e2fsck if empty */ 1090 1091 /* 1092 * The rest of the superblock fields should be zero, and if not it 1093 * means they are likely already in use, so leave them alone. We 1094 * can leave it up to e2fsck to clean up any inconsistencies there. 1095 */ 1096 } 1097 1098 /* 1099 * Open the external journal device 1100 */ 1101 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 1102 { 1103 struct block_device *bdev; 1104 1105 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 1106 if (IS_ERR(bdev)) 1107 goto fail; 1108 return bdev; 1109 1110 fail: 1111 ext4_msg(sb, KERN_ERR, 1112 "failed to open journal device unknown-block(%u,%u) %ld", 1113 MAJOR(dev), MINOR(dev), PTR_ERR(bdev)); 1114 return NULL; 1115 } 1116 1117 /* 1118 * Release the journal device 1119 */ 1120 static void ext4_blkdev_put(struct block_device *bdev) 1121 { 1122 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 1123 } 1124 1125 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 1126 { 1127 struct block_device *bdev; 1128 bdev = sbi->s_journal_bdev; 1129 if (bdev) { 1130 ext4_blkdev_put(bdev); 1131 sbi->s_journal_bdev = NULL; 1132 } 1133 } 1134 1135 static inline struct inode *orphan_list_entry(struct list_head *l) 1136 { 1137 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 1138 } 1139 1140 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 1141 { 1142 struct list_head *l; 1143 1144 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 1145 le32_to_cpu(sbi->s_es->s_last_orphan)); 1146 1147 printk(KERN_ERR "sb_info orphan list:\n"); 1148 list_for_each(l, &sbi->s_orphan) { 1149 struct inode *inode = orphan_list_entry(l); 1150 printk(KERN_ERR " " 1151 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 1152 inode->i_sb->s_id, inode->i_ino, inode, 1153 inode->i_mode, inode->i_nlink, 1154 NEXT_ORPHAN(inode)); 1155 } 1156 } 1157 1158 #ifdef CONFIG_QUOTA 1159 static int ext4_quota_off(struct super_block *sb, int type); 1160 1161 static inline void ext4_quota_off_umount(struct super_block *sb) 1162 { 1163 int type; 1164 1165 /* Use our quota_off function to clear inode flags etc. */ 1166 for (type = 0; type < EXT4_MAXQUOTAS; type++) 1167 ext4_quota_off(sb, type); 1168 } 1169 1170 /* 1171 * This is a helper function which is used in the mount/remount 1172 * codepaths (which holds s_umount) to fetch the quota file name. 1173 */ 1174 static inline char *get_qf_name(struct super_block *sb, 1175 struct ext4_sb_info *sbi, 1176 int type) 1177 { 1178 return rcu_dereference_protected(sbi->s_qf_names[type], 1179 lockdep_is_held(&sb->s_umount)); 1180 } 1181 #else 1182 static inline void ext4_quota_off_umount(struct super_block *sb) 1183 { 1184 } 1185 #endif 1186 1187 static void ext4_put_super(struct super_block *sb) 1188 { 1189 struct ext4_sb_info *sbi = EXT4_SB(sb); 1190 struct ext4_super_block *es = sbi->s_es; 1191 struct buffer_head **group_desc; 1192 struct flex_groups **flex_groups; 1193 int aborted = 0; 1194 int i, err; 1195 1196 /* 1197 * Unregister sysfs before destroying jbd2 journal. 1198 * Since we could still access attr_journal_task attribute via sysfs 1199 * path which could have sbi->s_journal->j_task as NULL 1200 * Unregister sysfs before flush sbi->s_error_work. 1201 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If 1202 * read metadata verify failed then will queue error work. 1203 * flush_stashed_error_work will call start_this_handle may trigger 1204 * BUG_ON. 1205 */ 1206 ext4_unregister_sysfs(sb); 1207 1208 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount")) 1209 ext4_msg(sb, KERN_INFO, "unmounting filesystem."); 1210 1211 ext4_unregister_li_request(sb); 1212 ext4_quota_off_umount(sb); 1213 1214 flush_work(&sbi->s_error_work); 1215 destroy_workqueue(sbi->rsv_conversion_wq); 1216 ext4_release_orphan_info(sb); 1217 1218 if (sbi->s_journal) { 1219 aborted = is_journal_aborted(sbi->s_journal); 1220 err = jbd2_journal_destroy(sbi->s_journal); 1221 sbi->s_journal = NULL; 1222 if ((err < 0) && !aborted) { 1223 ext4_abort(sb, -err, "Couldn't clean up the journal"); 1224 } 1225 } 1226 1227 ext4_es_unregister_shrinker(sbi); 1228 del_timer_sync(&sbi->s_err_report); 1229 ext4_release_system_zone(sb); 1230 ext4_mb_release(sb); 1231 ext4_ext_release(sb); 1232 1233 if (!sb_rdonly(sb) && !aborted) { 1234 ext4_clear_feature_journal_needs_recovery(sb); 1235 ext4_clear_feature_orphan_present(sb); 1236 es->s_state = cpu_to_le16(sbi->s_mount_state); 1237 } 1238 if (!sb_rdonly(sb)) 1239 ext4_commit_super(sb); 1240 1241 rcu_read_lock(); 1242 group_desc = rcu_dereference(sbi->s_group_desc); 1243 for (i = 0; i < sbi->s_gdb_count; i++) 1244 brelse(group_desc[i]); 1245 kvfree(group_desc); 1246 flex_groups = rcu_dereference(sbi->s_flex_groups); 1247 if (flex_groups) { 1248 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 1249 kvfree(flex_groups[i]); 1250 kvfree(flex_groups); 1251 } 1252 rcu_read_unlock(); 1253 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1254 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1255 percpu_counter_destroy(&sbi->s_dirs_counter); 1256 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1257 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 1258 percpu_free_rwsem(&sbi->s_writepages_rwsem); 1259 #ifdef CONFIG_QUOTA 1260 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1261 kfree(get_qf_name(sb, sbi, i)); 1262 #endif 1263 1264 /* Debugging code just in case the in-memory inode orphan list 1265 * isn't empty. The on-disk one can be non-empty if we've 1266 * detected an error and taken the fs readonly, but the 1267 * in-memory list had better be clean by this point. */ 1268 if (!list_empty(&sbi->s_orphan)) 1269 dump_orphan_list(sb, sbi); 1270 ASSERT(list_empty(&sbi->s_orphan)); 1271 1272 sync_blockdev(sb->s_bdev); 1273 invalidate_bdev(sb->s_bdev); 1274 if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) { 1275 /* 1276 * Invalidate the journal device's buffers. We don't want them 1277 * floating about in memory - the physical journal device may 1278 * hotswapped, and it breaks the `ro-after' testing code. 1279 */ 1280 sync_blockdev(sbi->s_journal_bdev); 1281 invalidate_bdev(sbi->s_journal_bdev); 1282 ext4_blkdev_remove(sbi); 1283 } 1284 1285 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1286 sbi->s_ea_inode_cache = NULL; 1287 1288 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1289 sbi->s_ea_block_cache = NULL; 1290 1291 ext4_stop_mmpd(sbi); 1292 1293 brelse(sbi->s_sbh); 1294 sb->s_fs_info = NULL; 1295 /* 1296 * Now that we are completely done shutting down the 1297 * superblock, we need to actually destroy the kobject. 1298 */ 1299 kobject_put(&sbi->s_kobj); 1300 wait_for_completion(&sbi->s_kobj_unregister); 1301 if (sbi->s_chksum_driver) 1302 crypto_free_shash(sbi->s_chksum_driver); 1303 kfree(sbi->s_blockgroup_lock); 1304 fs_put_dax(sbi->s_daxdev, NULL); 1305 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 1306 #if IS_ENABLED(CONFIG_UNICODE) 1307 utf8_unload(sb->s_encoding); 1308 #endif 1309 kfree(sbi); 1310 } 1311 1312 static struct kmem_cache *ext4_inode_cachep; 1313 1314 /* 1315 * Called inside transaction, so use GFP_NOFS 1316 */ 1317 static struct inode *ext4_alloc_inode(struct super_block *sb) 1318 { 1319 struct ext4_inode_info *ei; 1320 1321 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS); 1322 if (!ei) 1323 return NULL; 1324 1325 inode_set_iversion(&ei->vfs_inode, 1); 1326 spin_lock_init(&ei->i_raw_lock); 1327 INIT_LIST_HEAD(&ei->i_prealloc_list); 1328 atomic_set(&ei->i_prealloc_active, 0); 1329 spin_lock_init(&ei->i_prealloc_lock); 1330 ext4_es_init_tree(&ei->i_es_tree); 1331 rwlock_init(&ei->i_es_lock); 1332 INIT_LIST_HEAD(&ei->i_es_list); 1333 ei->i_es_all_nr = 0; 1334 ei->i_es_shk_nr = 0; 1335 ei->i_es_shrink_lblk = 0; 1336 ei->i_reserved_data_blocks = 0; 1337 spin_lock_init(&(ei->i_block_reservation_lock)); 1338 ext4_init_pending_tree(&ei->i_pending_tree); 1339 #ifdef CONFIG_QUOTA 1340 ei->i_reserved_quota = 0; 1341 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1342 #endif 1343 ei->jinode = NULL; 1344 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1345 spin_lock_init(&ei->i_completed_io_lock); 1346 ei->i_sync_tid = 0; 1347 ei->i_datasync_tid = 0; 1348 atomic_set(&ei->i_unwritten, 0); 1349 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1350 ext4_fc_init_inode(&ei->vfs_inode); 1351 mutex_init(&ei->i_fc_lock); 1352 return &ei->vfs_inode; 1353 } 1354 1355 static int ext4_drop_inode(struct inode *inode) 1356 { 1357 int drop = generic_drop_inode(inode); 1358 1359 if (!drop) 1360 drop = fscrypt_drop_inode(inode); 1361 1362 trace_ext4_drop_inode(inode, drop); 1363 return drop; 1364 } 1365 1366 static void ext4_free_in_core_inode(struct inode *inode) 1367 { 1368 fscrypt_free_inode(inode); 1369 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) { 1370 pr_warn("%s: inode %ld still in fc list", 1371 __func__, inode->i_ino); 1372 } 1373 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1374 } 1375 1376 static void ext4_destroy_inode(struct inode *inode) 1377 { 1378 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1379 ext4_msg(inode->i_sb, KERN_ERR, 1380 "Inode %lu (%p): orphan list check failed!", 1381 inode->i_ino, EXT4_I(inode)); 1382 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1383 EXT4_I(inode), sizeof(struct ext4_inode_info), 1384 true); 1385 dump_stack(); 1386 } 1387 1388 if (EXT4_I(inode)->i_reserved_data_blocks) 1389 ext4_msg(inode->i_sb, KERN_ERR, 1390 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!", 1391 inode->i_ino, EXT4_I(inode), 1392 EXT4_I(inode)->i_reserved_data_blocks); 1393 } 1394 1395 static void init_once(void *foo) 1396 { 1397 struct ext4_inode_info *ei = foo; 1398 1399 INIT_LIST_HEAD(&ei->i_orphan); 1400 init_rwsem(&ei->xattr_sem); 1401 init_rwsem(&ei->i_data_sem); 1402 inode_init_once(&ei->vfs_inode); 1403 ext4_fc_init_inode(&ei->vfs_inode); 1404 } 1405 1406 static int __init init_inodecache(void) 1407 { 1408 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1409 sizeof(struct ext4_inode_info), 0, 1410 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1411 SLAB_ACCOUNT), 1412 offsetof(struct ext4_inode_info, i_data), 1413 sizeof_field(struct ext4_inode_info, i_data), 1414 init_once); 1415 if (ext4_inode_cachep == NULL) 1416 return -ENOMEM; 1417 return 0; 1418 } 1419 1420 static void destroy_inodecache(void) 1421 { 1422 /* 1423 * Make sure all delayed rcu free inodes are flushed before we 1424 * destroy cache. 1425 */ 1426 rcu_barrier(); 1427 kmem_cache_destroy(ext4_inode_cachep); 1428 } 1429 1430 void ext4_clear_inode(struct inode *inode) 1431 { 1432 ext4_fc_del(inode); 1433 invalidate_inode_buffers(inode); 1434 clear_inode(inode); 1435 ext4_discard_preallocations(inode, 0); 1436 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1437 dquot_drop(inode); 1438 if (EXT4_I(inode)->jinode) { 1439 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1440 EXT4_I(inode)->jinode); 1441 jbd2_free_inode(EXT4_I(inode)->jinode); 1442 EXT4_I(inode)->jinode = NULL; 1443 } 1444 fscrypt_put_encryption_info(inode); 1445 fsverity_cleanup_inode(inode); 1446 } 1447 1448 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1449 u64 ino, u32 generation) 1450 { 1451 struct inode *inode; 1452 1453 /* 1454 * Currently we don't know the generation for parent directory, so 1455 * a generation of 0 means "accept any" 1456 */ 1457 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1458 if (IS_ERR(inode)) 1459 return ERR_CAST(inode); 1460 if (generation && inode->i_generation != generation) { 1461 iput(inode); 1462 return ERR_PTR(-ESTALE); 1463 } 1464 1465 return inode; 1466 } 1467 1468 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1469 int fh_len, int fh_type) 1470 { 1471 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1472 ext4_nfs_get_inode); 1473 } 1474 1475 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1476 int fh_len, int fh_type) 1477 { 1478 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1479 ext4_nfs_get_inode); 1480 } 1481 1482 static int ext4_nfs_commit_metadata(struct inode *inode) 1483 { 1484 struct writeback_control wbc = { 1485 .sync_mode = WB_SYNC_ALL 1486 }; 1487 1488 trace_ext4_nfs_commit_metadata(inode); 1489 return ext4_write_inode(inode, &wbc); 1490 } 1491 1492 #ifdef CONFIG_QUOTA 1493 static const char * const quotatypes[] = INITQFNAMES; 1494 #define QTYPE2NAME(t) (quotatypes[t]) 1495 1496 static int ext4_write_dquot(struct dquot *dquot); 1497 static int ext4_acquire_dquot(struct dquot *dquot); 1498 static int ext4_release_dquot(struct dquot *dquot); 1499 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1500 static int ext4_write_info(struct super_block *sb, int type); 1501 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1502 const struct path *path); 1503 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1504 size_t len, loff_t off); 1505 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1506 const char *data, size_t len, loff_t off); 1507 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1508 unsigned int flags); 1509 1510 static struct dquot **ext4_get_dquots(struct inode *inode) 1511 { 1512 return EXT4_I(inode)->i_dquot; 1513 } 1514 1515 static const struct dquot_operations ext4_quota_operations = { 1516 .get_reserved_space = ext4_get_reserved_space, 1517 .write_dquot = ext4_write_dquot, 1518 .acquire_dquot = ext4_acquire_dquot, 1519 .release_dquot = ext4_release_dquot, 1520 .mark_dirty = ext4_mark_dquot_dirty, 1521 .write_info = ext4_write_info, 1522 .alloc_dquot = dquot_alloc, 1523 .destroy_dquot = dquot_destroy, 1524 .get_projid = ext4_get_projid, 1525 .get_inode_usage = ext4_get_inode_usage, 1526 .get_next_id = dquot_get_next_id, 1527 }; 1528 1529 static const struct quotactl_ops ext4_qctl_operations = { 1530 .quota_on = ext4_quota_on, 1531 .quota_off = ext4_quota_off, 1532 .quota_sync = dquot_quota_sync, 1533 .get_state = dquot_get_state, 1534 .set_info = dquot_set_dqinfo, 1535 .get_dqblk = dquot_get_dqblk, 1536 .set_dqblk = dquot_set_dqblk, 1537 .get_nextdqblk = dquot_get_next_dqblk, 1538 }; 1539 #endif 1540 1541 static const struct super_operations ext4_sops = { 1542 .alloc_inode = ext4_alloc_inode, 1543 .free_inode = ext4_free_in_core_inode, 1544 .destroy_inode = ext4_destroy_inode, 1545 .write_inode = ext4_write_inode, 1546 .dirty_inode = ext4_dirty_inode, 1547 .drop_inode = ext4_drop_inode, 1548 .evict_inode = ext4_evict_inode, 1549 .put_super = ext4_put_super, 1550 .sync_fs = ext4_sync_fs, 1551 .freeze_fs = ext4_freeze, 1552 .unfreeze_fs = ext4_unfreeze, 1553 .statfs = ext4_statfs, 1554 .show_options = ext4_show_options, 1555 #ifdef CONFIG_QUOTA 1556 .quota_read = ext4_quota_read, 1557 .quota_write = ext4_quota_write, 1558 .get_dquots = ext4_get_dquots, 1559 #endif 1560 }; 1561 1562 static const struct export_operations ext4_export_ops = { 1563 .fh_to_dentry = ext4_fh_to_dentry, 1564 .fh_to_parent = ext4_fh_to_parent, 1565 .get_parent = ext4_get_parent, 1566 .commit_metadata = ext4_nfs_commit_metadata, 1567 }; 1568 1569 enum { 1570 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1571 Opt_resgid, Opt_resuid, Opt_sb, 1572 Opt_nouid32, Opt_debug, Opt_removed, 1573 Opt_user_xattr, Opt_acl, 1574 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1575 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1576 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1577 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1578 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1579 Opt_inlinecrypt, 1580 Opt_usrjquota, Opt_grpjquota, Opt_quota, 1581 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1582 Opt_usrquota, Opt_grpquota, Opt_prjquota, 1583 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, 1584 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1585 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize, 1586 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1587 Opt_inode_readahead_blks, Opt_journal_ioprio, 1588 Opt_dioread_nolock, Opt_dioread_lock, 1589 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1590 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1591 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan, 1592 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type, 1593 #ifdef CONFIG_EXT4_DEBUG 1594 Opt_fc_debug_max_replay, Opt_fc_debug_force 1595 #endif 1596 }; 1597 1598 static const struct constant_table ext4_param_errors[] = { 1599 {"continue", EXT4_MOUNT_ERRORS_CONT}, 1600 {"panic", EXT4_MOUNT_ERRORS_PANIC}, 1601 {"remount-ro", EXT4_MOUNT_ERRORS_RO}, 1602 {} 1603 }; 1604 1605 static const struct constant_table ext4_param_data[] = { 1606 {"journal", EXT4_MOUNT_JOURNAL_DATA}, 1607 {"ordered", EXT4_MOUNT_ORDERED_DATA}, 1608 {"writeback", EXT4_MOUNT_WRITEBACK_DATA}, 1609 {} 1610 }; 1611 1612 static const struct constant_table ext4_param_data_err[] = { 1613 {"abort", Opt_data_err_abort}, 1614 {"ignore", Opt_data_err_ignore}, 1615 {} 1616 }; 1617 1618 static const struct constant_table ext4_param_jqfmt[] = { 1619 {"vfsold", QFMT_VFS_OLD}, 1620 {"vfsv0", QFMT_VFS_V0}, 1621 {"vfsv1", QFMT_VFS_V1}, 1622 {} 1623 }; 1624 1625 static const struct constant_table ext4_param_dax[] = { 1626 {"always", Opt_dax_always}, 1627 {"inode", Opt_dax_inode}, 1628 {"never", Opt_dax_never}, 1629 {} 1630 }; 1631 1632 /* String parameter that allows empty argument */ 1633 #define fsparam_string_empty(NAME, OPT) \ 1634 __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL) 1635 1636 /* 1637 * Mount option specification 1638 * We don't use fsparam_flag_no because of the way we set the 1639 * options and the way we show them in _ext4_show_options(). To 1640 * keep the changes to a minimum, let's keep the negative options 1641 * separate for now. 1642 */ 1643 static const struct fs_parameter_spec ext4_param_specs[] = { 1644 fsparam_flag ("bsddf", Opt_bsd_df), 1645 fsparam_flag ("minixdf", Opt_minix_df), 1646 fsparam_flag ("grpid", Opt_grpid), 1647 fsparam_flag ("bsdgroups", Opt_grpid), 1648 fsparam_flag ("nogrpid", Opt_nogrpid), 1649 fsparam_flag ("sysvgroups", Opt_nogrpid), 1650 fsparam_u32 ("resgid", Opt_resgid), 1651 fsparam_u32 ("resuid", Opt_resuid), 1652 fsparam_u32 ("sb", Opt_sb), 1653 fsparam_enum ("errors", Opt_errors, ext4_param_errors), 1654 fsparam_flag ("nouid32", Opt_nouid32), 1655 fsparam_flag ("debug", Opt_debug), 1656 fsparam_flag ("oldalloc", Opt_removed), 1657 fsparam_flag ("orlov", Opt_removed), 1658 fsparam_flag ("user_xattr", Opt_user_xattr), 1659 fsparam_flag ("acl", Opt_acl), 1660 fsparam_flag ("norecovery", Opt_noload), 1661 fsparam_flag ("noload", Opt_noload), 1662 fsparam_flag ("bh", Opt_removed), 1663 fsparam_flag ("nobh", Opt_removed), 1664 fsparam_u32 ("commit", Opt_commit), 1665 fsparam_u32 ("min_batch_time", Opt_min_batch_time), 1666 fsparam_u32 ("max_batch_time", Opt_max_batch_time), 1667 fsparam_u32 ("journal_dev", Opt_journal_dev), 1668 fsparam_bdev ("journal_path", Opt_journal_path), 1669 fsparam_flag ("journal_checksum", Opt_journal_checksum), 1670 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum), 1671 fsparam_flag ("journal_async_commit",Opt_journal_async_commit), 1672 fsparam_flag ("abort", Opt_abort), 1673 fsparam_enum ("data", Opt_data, ext4_param_data), 1674 fsparam_enum ("data_err", Opt_data_err, 1675 ext4_param_data_err), 1676 fsparam_string_empty 1677 ("usrjquota", Opt_usrjquota), 1678 fsparam_string_empty 1679 ("grpjquota", Opt_grpjquota), 1680 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt), 1681 fsparam_flag ("grpquota", Opt_grpquota), 1682 fsparam_flag ("quota", Opt_quota), 1683 fsparam_flag ("noquota", Opt_noquota), 1684 fsparam_flag ("usrquota", Opt_usrquota), 1685 fsparam_flag ("prjquota", Opt_prjquota), 1686 fsparam_flag ("barrier", Opt_barrier), 1687 fsparam_u32 ("barrier", Opt_barrier), 1688 fsparam_flag ("nobarrier", Opt_nobarrier), 1689 fsparam_flag ("i_version", Opt_removed), 1690 fsparam_flag ("dax", Opt_dax), 1691 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax), 1692 fsparam_u32 ("stripe", Opt_stripe), 1693 fsparam_flag ("delalloc", Opt_delalloc), 1694 fsparam_flag ("nodelalloc", Opt_nodelalloc), 1695 fsparam_flag ("warn_on_error", Opt_warn_on_error), 1696 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), 1697 fsparam_u32 ("debug_want_extra_isize", 1698 Opt_debug_want_extra_isize), 1699 fsparam_flag ("mblk_io_submit", Opt_removed), 1700 fsparam_flag ("nomblk_io_submit", Opt_removed), 1701 fsparam_flag ("block_validity", Opt_block_validity), 1702 fsparam_flag ("noblock_validity", Opt_noblock_validity), 1703 fsparam_u32 ("inode_readahead_blks", 1704 Opt_inode_readahead_blks), 1705 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio), 1706 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc), 1707 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc), 1708 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc), 1709 fsparam_flag ("dioread_nolock", Opt_dioread_nolock), 1710 fsparam_flag ("nodioread_nolock", Opt_dioread_lock), 1711 fsparam_flag ("dioread_lock", Opt_dioread_lock), 1712 fsparam_flag ("discard", Opt_discard), 1713 fsparam_flag ("nodiscard", Opt_nodiscard), 1714 fsparam_u32 ("init_itable", Opt_init_itable), 1715 fsparam_flag ("init_itable", Opt_init_itable), 1716 fsparam_flag ("noinit_itable", Opt_noinit_itable), 1717 #ifdef CONFIG_EXT4_DEBUG 1718 fsparam_flag ("fc_debug_force", Opt_fc_debug_force), 1719 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay), 1720 #endif 1721 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb), 1722 fsparam_flag ("test_dummy_encryption", 1723 Opt_test_dummy_encryption), 1724 fsparam_string ("test_dummy_encryption", 1725 Opt_test_dummy_encryption), 1726 fsparam_flag ("inlinecrypt", Opt_inlinecrypt), 1727 fsparam_flag ("nombcache", Opt_nombcache), 1728 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */ 1729 fsparam_flag ("prefetch_block_bitmaps", 1730 Opt_removed), 1731 fsparam_flag ("no_prefetch_block_bitmaps", 1732 Opt_no_prefetch_block_bitmaps), 1733 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan), 1734 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */ 1735 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */ 1736 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */ 1737 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */ 1738 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */ 1739 {} 1740 }; 1741 1742 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1743 1744 #define MOPT_SET 0x0001 1745 #define MOPT_CLEAR 0x0002 1746 #define MOPT_NOSUPPORT 0x0004 1747 #define MOPT_EXPLICIT 0x0008 1748 #ifdef CONFIG_QUOTA 1749 #define MOPT_Q 0 1750 #define MOPT_QFMT 0x0010 1751 #else 1752 #define MOPT_Q MOPT_NOSUPPORT 1753 #define MOPT_QFMT MOPT_NOSUPPORT 1754 #endif 1755 #define MOPT_NO_EXT2 0x0020 1756 #define MOPT_NO_EXT3 0x0040 1757 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1758 #define MOPT_SKIP 0x0080 1759 #define MOPT_2 0x0100 1760 1761 static const struct mount_opts { 1762 int token; 1763 int mount_opt; 1764 int flags; 1765 } ext4_mount_opts[] = { 1766 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1767 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1768 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1769 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1770 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1771 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1772 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1773 MOPT_EXT4_ONLY | MOPT_SET}, 1774 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1775 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1776 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1777 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1778 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1779 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1780 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1781 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1782 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1783 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1784 {Opt_commit, 0, MOPT_NO_EXT2}, 1785 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1786 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1787 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1788 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1789 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1790 EXT4_MOUNT_JOURNAL_CHECKSUM), 1791 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1792 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1793 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2}, 1794 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1795 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1796 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1797 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1798 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1799 {Opt_dax_type, 0, MOPT_EXT4_ONLY}, 1800 {Opt_journal_dev, 0, MOPT_NO_EXT2}, 1801 {Opt_journal_path, 0, MOPT_NO_EXT2}, 1802 {Opt_journal_ioprio, 0, MOPT_NO_EXT2}, 1803 {Opt_data, 0, MOPT_NO_EXT2}, 1804 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1805 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1806 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1807 #else 1808 {Opt_acl, 0, MOPT_NOSUPPORT}, 1809 #endif 1810 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1811 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1812 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1813 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1814 MOPT_SET | MOPT_Q}, 1815 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1816 MOPT_SET | MOPT_Q}, 1817 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1818 MOPT_SET | MOPT_Q}, 1819 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1820 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1821 MOPT_CLEAR | MOPT_Q}, 1822 {Opt_usrjquota, 0, MOPT_Q}, 1823 {Opt_grpjquota, 0, MOPT_Q}, 1824 {Opt_jqfmt, 0, MOPT_QFMT}, 1825 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1826 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, 1827 MOPT_SET}, 1828 #ifdef CONFIG_EXT4_DEBUG 1829 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT, 1830 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY}, 1831 #endif 1832 {Opt_err, 0, 0} 1833 }; 1834 1835 #if IS_ENABLED(CONFIG_UNICODE) 1836 static const struct ext4_sb_encodings { 1837 __u16 magic; 1838 char *name; 1839 unsigned int version; 1840 } ext4_sb_encoding_map[] = { 1841 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)}, 1842 }; 1843 1844 static const struct ext4_sb_encodings * 1845 ext4_sb_read_encoding(const struct ext4_super_block *es) 1846 { 1847 __u16 magic = le16_to_cpu(es->s_encoding); 1848 int i; 1849 1850 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1851 if (magic == ext4_sb_encoding_map[i].magic) 1852 return &ext4_sb_encoding_map[i]; 1853 1854 return NULL; 1855 } 1856 #endif 1857 1858 #define EXT4_SPEC_JQUOTA (1 << 0) 1859 #define EXT4_SPEC_JQFMT (1 << 1) 1860 #define EXT4_SPEC_DATAJ (1 << 2) 1861 #define EXT4_SPEC_SB_BLOCK (1 << 3) 1862 #define EXT4_SPEC_JOURNAL_DEV (1 << 4) 1863 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5) 1864 #define EXT4_SPEC_s_want_extra_isize (1 << 7) 1865 #define EXT4_SPEC_s_max_batch_time (1 << 8) 1866 #define EXT4_SPEC_s_min_batch_time (1 << 9) 1867 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10) 1868 #define EXT4_SPEC_s_li_wait_mult (1 << 11) 1869 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12) 1870 #define EXT4_SPEC_s_stripe (1 << 13) 1871 #define EXT4_SPEC_s_resuid (1 << 14) 1872 #define EXT4_SPEC_s_resgid (1 << 15) 1873 #define EXT4_SPEC_s_commit_interval (1 << 16) 1874 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17) 1875 #define EXT4_SPEC_s_sb_block (1 << 18) 1876 #define EXT4_SPEC_mb_optimize_scan (1 << 19) 1877 1878 struct ext4_fs_context { 1879 char *s_qf_names[EXT4_MAXQUOTAS]; 1880 struct fscrypt_dummy_policy dummy_enc_policy; 1881 int s_jquota_fmt; /* Format of quota to use */ 1882 #ifdef CONFIG_EXT4_DEBUG 1883 int s_fc_debug_max_replay; 1884 #endif 1885 unsigned short qname_spec; 1886 unsigned long vals_s_flags; /* Bits to set in s_flags */ 1887 unsigned long mask_s_flags; /* Bits changed in s_flags */ 1888 unsigned long journal_devnum; 1889 unsigned long s_commit_interval; 1890 unsigned long s_stripe; 1891 unsigned int s_inode_readahead_blks; 1892 unsigned int s_want_extra_isize; 1893 unsigned int s_li_wait_mult; 1894 unsigned int s_max_dir_size_kb; 1895 unsigned int journal_ioprio; 1896 unsigned int vals_s_mount_opt; 1897 unsigned int mask_s_mount_opt; 1898 unsigned int vals_s_mount_opt2; 1899 unsigned int mask_s_mount_opt2; 1900 unsigned long vals_s_mount_flags; 1901 unsigned long mask_s_mount_flags; 1902 unsigned int opt_flags; /* MOPT flags */ 1903 unsigned int spec; 1904 u32 s_max_batch_time; 1905 u32 s_min_batch_time; 1906 kuid_t s_resuid; 1907 kgid_t s_resgid; 1908 ext4_fsblk_t s_sb_block; 1909 }; 1910 1911 static void ext4_fc_free(struct fs_context *fc) 1912 { 1913 struct ext4_fs_context *ctx = fc->fs_private; 1914 int i; 1915 1916 if (!ctx) 1917 return; 1918 1919 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1920 kfree(ctx->s_qf_names[i]); 1921 1922 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy); 1923 kfree(ctx); 1924 } 1925 1926 int ext4_init_fs_context(struct fs_context *fc) 1927 { 1928 struct ext4_fs_context *ctx; 1929 1930 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 1931 if (!ctx) 1932 return -ENOMEM; 1933 1934 fc->fs_private = ctx; 1935 fc->ops = &ext4_context_ops; 1936 1937 return 0; 1938 } 1939 1940 #ifdef CONFIG_QUOTA 1941 /* 1942 * Note the name of the specified quota file. 1943 */ 1944 static int note_qf_name(struct fs_context *fc, int qtype, 1945 struct fs_parameter *param) 1946 { 1947 struct ext4_fs_context *ctx = fc->fs_private; 1948 char *qname; 1949 1950 if (param->size < 1) { 1951 ext4_msg(NULL, KERN_ERR, "Missing quota name"); 1952 return -EINVAL; 1953 } 1954 if (strchr(param->string, '/')) { 1955 ext4_msg(NULL, KERN_ERR, 1956 "quotafile must be on filesystem root"); 1957 return -EINVAL; 1958 } 1959 if (ctx->s_qf_names[qtype]) { 1960 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) { 1961 ext4_msg(NULL, KERN_ERR, 1962 "%s quota file already specified", 1963 QTYPE2NAME(qtype)); 1964 return -EINVAL; 1965 } 1966 return 0; 1967 } 1968 1969 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL); 1970 if (!qname) { 1971 ext4_msg(NULL, KERN_ERR, 1972 "Not enough memory for storing quotafile name"); 1973 return -ENOMEM; 1974 } 1975 ctx->s_qf_names[qtype] = qname; 1976 ctx->qname_spec |= 1 << qtype; 1977 ctx->spec |= EXT4_SPEC_JQUOTA; 1978 return 0; 1979 } 1980 1981 /* 1982 * Clear the name of the specified quota file. 1983 */ 1984 static int unnote_qf_name(struct fs_context *fc, int qtype) 1985 { 1986 struct ext4_fs_context *ctx = fc->fs_private; 1987 1988 if (ctx->s_qf_names[qtype]) 1989 kfree(ctx->s_qf_names[qtype]); 1990 1991 ctx->s_qf_names[qtype] = NULL; 1992 ctx->qname_spec |= 1 << qtype; 1993 ctx->spec |= EXT4_SPEC_JQUOTA; 1994 return 0; 1995 } 1996 #endif 1997 1998 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param, 1999 struct ext4_fs_context *ctx) 2000 { 2001 int err; 2002 2003 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) { 2004 ext4_msg(NULL, KERN_WARNING, 2005 "test_dummy_encryption option not supported"); 2006 return -EINVAL; 2007 } 2008 err = fscrypt_parse_test_dummy_encryption(param, 2009 &ctx->dummy_enc_policy); 2010 if (err == -EINVAL) { 2011 ext4_msg(NULL, KERN_WARNING, 2012 "Value of option \"%s\" is unrecognized", param->key); 2013 } else if (err == -EEXIST) { 2014 ext4_msg(NULL, KERN_WARNING, 2015 "Conflicting test_dummy_encryption options"); 2016 return -EINVAL; 2017 } 2018 return err; 2019 } 2020 2021 #define EXT4_SET_CTX(name) \ 2022 static inline void ctx_set_##name(struct ext4_fs_context *ctx, \ 2023 unsigned long flag) \ 2024 { \ 2025 ctx->mask_s_##name |= flag; \ 2026 ctx->vals_s_##name |= flag; \ 2027 } 2028 2029 #define EXT4_CLEAR_CTX(name) \ 2030 static inline void ctx_clear_##name(struct ext4_fs_context *ctx, \ 2031 unsigned long flag) \ 2032 { \ 2033 ctx->mask_s_##name |= flag; \ 2034 ctx->vals_s_##name &= ~flag; \ 2035 } 2036 2037 #define EXT4_TEST_CTX(name) \ 2038 static inline unsigned long \ 2039 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \ 2040 { \ 2041 return (ctx->vals_s_##name & flag); \ 2042 } 2043 2044 EXT4_SET_CTX(flags); /* set only */ 2045 EXT4_SET_CTX(mount_opt); 2046 EXT4_CLEAR_CTX(mount_opt); 2047 EXT4_TEST_CTX(mount_opt); 2048 EXT4_SET_CTX(mount_opt2); 2049 EXT4_CLEAR_CTX(mount_opt2); 2050 EXT4_TEST_CTX(mount_opt2); 2051 2052 static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit) 2053 { 2054 set_bit(bit, &ctx->mask_s_mount_flags); 2055 set_bit(bit, &ctx->vals_s_mount_flags); 2056 } 2057 2058 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param) 2059 { 2060 struct ext4_fs_context *ctx = fc->fs_private; 2061 struct fs_parse_result result; 2062 const struct mount_opts *m; 2063 int is_remount; 2064 kuid_t uid; 2065 kgid_t gid; 2066 int token; 2067 2068 token = fs_parse(fc, ext4_param_specs, param, &result); 2069 if (token < 0) 2070 return token; 2071 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2072 2073 for (m = ext4_mount_opts; m->token != Opt_err; m++) 2074 if (token == m->token) 2075 break; 2076 2077 ctx->opt_flags |= m->flags; 2078 2079 if (m->flags & MOPT_EXPLICIT) { 2080 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 2081 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC); 2082 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 2083 ctx_set_mount_opt2(ctx, 2084 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM); 2085 } else 2086 return -EINVAL; 2087 } 2088 2089 if (m->flags & MOPT_NOSUPPORT) { 2090 ext4_msg(NULL, KERN_ERR, "%s option not supported", 2091 param->key); 2092 return 0; 2093 } 2094 2095 switch (token) { 2096 #ifdef CONFIG_QUOTA 2097 case Opt_usrjquota: 2098 if (!*param->string) 2099 return unnote_qf_name(fc, USRQUOTA); 2100 else 2101 return note_qf_name(fc, USRQUOTA, param); 2102 case Opt_grpjquota: 2103 if (!*param->string) 2104 return unnote_qf_name(fc, GRPQUOTA); 2105 else 2106 return note_qf_name(fc, GRPQUOTA, param); 2107 #endif 2108 case Opt_sb: 2109 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2110 ext4_msg(NULL, KERN_WARNING, 2111 "Ignoring %s option on remount", param->key); 2112 } else { 2113 ctx->s_sb_block = result.uint_32; 2114 ctx->spec |= EXT4_SPEC_s_sb_block; 2115 } 2116 return 0; 2117 case Opt_removed: 2118 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option", 2119 param->key); 2120 return 0; 2121 case Opt_abort: 2122 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED); 2123 return 0; 2124 case Opt_inlinecrypt: 2125 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT 2126 ctx_set_flags(ctx, SB_INLINECRYPT); 2127 #else 2128 ext4_msg(NULL, KERN_ERR, "inline encryption not supported"); 2129 #endif 2130 return 0; 2131 case Opt_errors: 2132 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK); 2133 ctx_set_mount_opt(ctx, result.uint_32); 2134 return 0; 2135 #ifdef CONFIG_QUOTA 2136 case Opt_jqfmt: 2137 ctx->s_jquota_fmt = result.uint_32; 2138 ctx->spec |= EXT4_SPEC_JQFMT; 2139 return 0; 2140 #endif 2141 case Opt_data: 2142 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2143 ctx_set_mount_opt(ctx, result.uint_32); 2144 ctx->spec |= EXT4_SPEC_DATAJ; 2145 return 0; 2146 case Opt_commit: 2147 if (result.uint_32 == 0) 2148 ctx->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE; 2149 else if (result.uint_32 > INT_MAX / HZ) { 2150 ext4_msg(NULL, KERN_ERR, 2151 "Invalid commit interval %d, " 2152 "must be smaller than %d", 2153 result.uint_32, INT_MAX / HZ); 2154 return -EINVAL; 2155 } 2156 ctx->s_commit_interval = HZ * result.uint_32; 2157 ctx->spec |= EXT4_SPEC_s_commit_interval; 2158 return 0; 2159 case Opt_debug_want_extra_isize: 2160 if ((result.uint_32 & 1) || (result.uint_32 < 4)) { 2161 ext4_msg(NULL, KERN_ERR, 2162 "Invalid want_extra_isize %d", result.uint_32); 2163 return -EINVAL; 2164 } 2165 ctx->s_want_extra_isize = result.uint_32; 2166 ctx->spec |= EXT4_SPEC_s_want_extra_isize; 2167 return 0; 2168 case Opt_max_batch_time: 2169 ctx->s_max_batch_time = result.uint_32; 2170 ctx->spec |= EXT4_SPEC_s_max_batch_time; 2171 return 0; 2172 case Opt_min_batch_time: 2173 ctx->s_min_batch_time = result.uint_32; 2174 ctx->spec |= EXT4_SPEC_s_min_batch_time; 2175 return 0; 2176 case Opt_inode_readahead_blks: 2177 if (result.uint_32 && 2178 (result.uint_32 > (1 << 30) || 2179 !is_power_of_2(result.uint_32))) { 2180 ext4_msg(NULL, KERN_ERR, 2181 "EXT4-fs: inode_readahead_blks must be " 2182 "0 or a power of 2 smaller than 2^31"); 2183 return -EINVAL; 2184 } 2185 ctx->s_inode_readahead_blks = result.uint_32; 2186 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks; 2187 return 0; 2188 case Opt_init_itable: 2189 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE); 2190 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 2191 if (param->type == fs_value_is_string) 2192 ctx->s_li_wait_mult = result.uint_32; 2193 ctx->spec |= EXT4_SPEC_s_li_wait_mult; 2194 return 0; 2195 case Opt_max_dir_size_kb: 2196 ctx->s_max_dir_size_kb = result.uint_32; 2197 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb; 2198 return 0; 2199 #ifdef CONFIG_EXT4_DEBUG 2200 case Opt_fc_debug_max_replay: 2201 ctx->s_fc_debug_max_replay = result.uint_32; 2202 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay; 2203 return 0; 2204 #endif 2205 case Opt_stripe: 2206 ctx->s_stripe = result.uint_32; 2207 ctx->spec |= EXT4_SPEC_s_stripe; 2208 return 0; 2209 case Opt_resuid: 2210 uid = make_kuid(current_user_ns(), result.uint_32); 2211 if (!uid_valid(uid)) { 2212 ext4_msg(NULL, KERN_ERR, "Invalid uid value %d", 2213 result.uint_32); 2214 return -EINVAL; 2215 } 2216 ctx->s_resuid = uid; 2217 ctx->spec |= EXT4_SPEC_s_resuid; 2218 return 0; 2219 case Opt_resgid: 2220 gid = make_kgid(current_user_ns(), result.uint_32); 2221 if (!gid_valid(gid)) { 2222 ext4_msg(NULL, KERN_ERR, "Invalid gid value %d", 2223 result.uint_32); 2224 return -EINVAL; 2225 } 2226 ctx->s_resgid = gid; 2227 ctx->spec |= EXT4_SPEC_s_resgid; 2228 return 0; 2229 case Opt_journal_dev: 2230 if (is_remount) { 2231 ext4_msg(NULL, KERN_ERR, 2232 "Cannot specify journal on remount"); 2233 return -EINVAL; 2234 } 2235 ctx->journal_devnum = result.uint_32; 2236 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2237 return 0; 2238 case Opt_journal_path: 2239 { 2240 struct inode *journal_inode; 2241 struct path path; 2242 int error; 2243 2244 if (is_remount) { 2245 ext4_msg(NULL, KERN_ERR, 2246 "Cannot specify journal on remount"); 2247 return -EINVAL; 2248 } 2249 2250 error = fs_lookup_param(fc, param, 1, &path); 2251 if (error) { 2252 ext4_msg(NULL, KERN_ERR, "error: could not find " 2253 "journal device path"); 2254 return -EINVAL; 2255 } 2256 2257 journal_inode = d_inode(path.dentry); 2258 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev); 2259 ctx->spec |= EXT4_SPEC_JOURNAL_DEV; 2260 path_put(&path); 2261 return 0; 2262 } 2263 case Opt_journal_ioprio: 2264 if (result.uint_32 > 7) { 2265 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority" 2266 " (must be 0-7)"); 2267 return -EINVAL; 2268 } 2269 ctx->journal_ioprio = 2270 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32); 2271 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO; 2272 return 0; 2273 case Opt_test_dummy_encryption: 2274 return ext4_parse_test_dummy_encryption(param, ctx); 2275 case Opt_dax: 2276 case Opt_dax_type: 2277 #ifdef CONFIG_FS_DAX 2278 { 2279 int type = (token == Opt_dax) ? 2280 Opt_dax : result.uint_32; 2281 2282 switch (type) { 2283 case Opt_dax: 2284 case Opt_dax_always: 2285 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2286 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2287 break; 2288 case Opt_dax_never: 2289 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2290 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2291 break; 2292 case Opt_dax_inode: 2293 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS); 2294 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER); 2295 /* Strictly for printing options */ 2296 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE); 2297 break; 2298 } 2299 return 0; 2300 } 2301 #else 2302 ext4_msg(NULL, KERN_INFO, "dax option not supported"); 2303 return -EINVAL; 2304 #endif 2305 case Opt_data_err: 2306 if (result.uint_32 == Opt_data_err_abort) 2307 ctx_set_mount_opt(ctx, m->mount_opt); 2308 else if (result.uint_32 == Opt_data_err_ignore) 2309 ctx_clear_mount_opt(ctx, m->mount_opt); 2310 return 0; 2311 case Opt_mb_optimize_scan: 2312 if (result.int_32 == 1) { 2313 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2314 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2315 } else if (result.int_32 == 0) { 2316 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN); 2317 ctx->spec |= EXT4_SPEC_mb_optimize_scan; 2318 } else { 2319 ext4_msg(NULL, KERN_WARNING, 2320 "mb_optimize_scan should be set to 0 or 1."); 2321 return -EINVAL; 2322 } 2323 return 0; 2324 } 2325 2326 /* 2327 * At this point we should only be getting options requiring MOPT_SET, 2328 * or MOPT_CLEAR. Anything else is a bug 2329 */ 2330 if (m->token == Opt_err) { 2331 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s", 2332 param->key); 2333 WARN_ON(1); 2334 return -EINVAL; 2335 } 2336 2337 else { 2338 unsigned int set = 0; 2339 2340 if ((param->type == fs_value_is_flag) || 2341 result.uint_32 > 0) 2342 set = 1; 2343 2344 if (m->flags & MOPT_CLEAR) 2345 set = !set; 2346 else if (unlikely(!(m->flags & MOPT_SET))) { 2347 ext4_msg(NULL, KERN_WARNING, 2348 "buggy handling of option %s", 2349 param->key); 2350 WARN_ON(1); 2351 return -EINVAL; 2352 } 2353 if (m->flags & MOPT_2) { 2354 if (set != 0) 2355 ctx_set_mount_opt2(ctx, m->mount_opt); 2356 else 2357 ctx_clear_mount_opt2(ctx, m->mount_opt); 2358 } else { 2359 if (set != 0) 2360 ctx_set_mount_opt(ctx, m->mount_opt); 2361 else 2362 ctx_clear_mount_opt(ctx, m->mount_opt); 2363 } 2364 } 2365 2366 return 0; 2367 } 2368 2369 static int parse_options(struct fs_context *fc, char *options) 2370 { 2371 struct fs_parameter param; 2372 int ret; 2373 char *key; 2374 2375 if (!options) 2376 return 0; 2377 2378 while ((key = strsep(&options, ",")) != NULL) { 2379 if (*key) { 2380 size_t v_len = 0; 2381 char *value = strchr(key, '='); 2382 2383 param.type = fs_value_is_flag; 2384 param.string = NULL; 2385 2386 if (value) { 2387 if (value == key) 2388 continue; 2389 2390 *value++ = 0; 2391 v_len = strlen(value); 2392 param.string = kmemdup_nul(value, v_len, 2393 GFP_KERNEL); 2394 if (!param.string) 2395 return -ENOMEM; 2396 param.type = fs_value_is_string; 2397 } 2398 2399 param.key = key; 2400 param.size = v_len; 2401 2402 ret = ext4_parse_param(fc, ¶m); 2403 if (param.string) 2404 kfree(param.string); 2405 if (ret < 0) 2406 return ret; 2407 } 2408 } 2409 2410 ret = ext4_validate_options(fc); 2411 if (ret < 0) 2412 return ret; 2413 2414 return 0; 2415 } 2416 2417 static int parse_apply_sb_mount_options(struct super_block *sb, 2418 struct ext4_fs_context *m_ctx) 2419 { 2420 struct ext4_sb_info *sbi = EXT4_SB(sb); 2421 char *s_mount_opts = NULL; 2422 struct ext4_fs_context *s_ctx = NULL; 2423 struct fs_context *fc = NULL; 2424 int ret = -ENOMEM; 2425 2426 if (!sbi->s_es->s_mount_opts[0]) 2427 return 0; 2428 2429 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 2430 sizeof(sbi->s_es->s_mount_opts), 2431 GFP_KERNEL); 2432 if (!s_mount_opts) 2433 return ret; 2434 2435 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL); 2436 if (!fc) 2437 goto out_free; 2438 2439 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL); 2440 if (!s_ctx) 2441 goto out_free; 2442 2443 fc->fs_private = s_ctx; 2444 fc->s_fs_info = sbi; 2445 2446 ret = parse_options(fc, s_mount_opts); 2447 if (ret < 0) 2448 goto parse_failed; 2449 2450 ret = ext4_check_opt_consistency(fc, sb); 2451 if (ret < 0) { 2452 parse_failed: 2453 ext4_msg(sb, KERN_WARNING, 2454 "failed to parse options in superblock: %s", 2455 s_mount_opts); 2456 ret = 0; 2457 goto out_free; 2458 } 2459 2460 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV) 2461 m_ctx->journal_devnum = s_ctx->journal_devnum; 2462 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO) 2463 m_ctx->journal_ioprio = s_ctx->journal_ioprio; 2464 2465 ext4_apply_options(fc, sb); 2466 ret = 0; 2467 2468 out_free: 2469 if (fc) { 2470 ext4_fc_free(fc); 2471 kfree(fc); 2472 } 2473 kfree(s_mount_opts); 2474 return ret; 2475 } 2476 2477 static void ext4_apply_quota_options(struct fs_context *fc, 2478 struct super_block *sb) 2479 { 2480 #ifdef CONFIG_QUOTA 2481 bool quota_feature = ext4_has_feature_quota(sb); 2482 struct ext4_fs_context *ctx = fc->fs_private; 2483 struct ext4_sb_info *sbi = EXT4_SB(sb); 2484 char *qname; 2485 int i; 2486 2487 if (quota_feature) 2488 return; 2489 2490 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2491 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2492 if (!(ctx->qname_spec & (1 << i))) 2493 continue; 2494 2495 qname = ctx->s_qf_names[i]; /* May be NULL */ 2496 if (qname) 2497 set_opt(sb, QUOTA); 2498 ctx->s_qf_names[i] = NULL; 2499 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname, 2500 lockdep_is_held(&sb->s_umount)); 2501 if (qname) 2502 kfree_rcu(qname); 2503 } 2504 } 2505 2506 if (ctx->spec & EXT4_SPEC_JQFMT) 2507 sbi->s_jquota_fmt = ctx->s_jquota_fmt; 2508 #endif 2509 } 2510 2511 /* 2512 * Check quota settings consistency. 2513 */ 2514 static int ext4_check_quota_consistency(struct fs_context *fc, 2515 struct super_block *sb) 2516 { 2517 #ifdef CONFIG_QUOTA 2518 struct ext4_fs_context *ctx = fc->fs_private; 2519 struct ext4_sb_info *sbi = EXT4_SB(sb); 2520 bool quota_feature = ext4_has_feature_quota(sb); 2521 bool quota_loaded = sb_any_quota_loaded(sb); 2522 bool usr_qf_name, grp_qf_name, usrquota, grpquota; 2523 int quota_flags, i; 2524 2525 /* 2526 * We do the test below only for project quotas. 'usrquota' and 2527 * 'grpquota' mount options are allowed even without quota feature 2528 * to support legacy quotas in quota files. 2529 */ 2530 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) && 2531 !ext4_has_feature_project(sb)) { 2532 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. " 2533 "Cannot enable project quota enforcement."); 2534 return -EINVAL; 2535 } 2536 2537 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 2538 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA; 2539 if (quota_loaded && 2540 ctx->mask_s_mount_opt & quota_flags && 2541 !ctx_test_mount_opt(ctx, quota_flags)) 2542 goto err_quota_change; 2543 2544 if (ctx->spec & EXT4_SPEC_JQUOTA) { 2545 2546 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2547 if (!(ctx->qname_spec & (1 << i))) 2548 continue; 2549 2550 if (quota_loaded && 2551 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i]) 2552 goto err_jquota_change; 2553 2554 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] && 2555 strcmp(get_qf_name(sb, sbi, i), 2556 ctx->s_qf_names[i]) != 0) 2557 goto err_jquota_specified; 2558 } 2559 2560 if (quota_feature) { 2561 ext4_msg(NULL, KERN_INFO, 2562 "Journaled quota options ignored when " 2563 "QUOTA feature is enabled"); 2564 return 0; 2565 } 2566 } 2567 2568 if (ctx->spec & EXT4_SPEC_JQFMT) { 2569 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded) 2570 goto err_jquota_change; 2571 if (quota_feature) { 2572 ext4_msg(NULL, KERN_INFO, "Quota format mount options " 2573 "ignored when QUOTA feature is enabled"); 2574 return 0; 2575 } 2576 } 2577 2578 /* Make sure we don't mix old and new quota format */ 2579 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) || 2580 ctx->s_qf_names[USRQUOTA]); 2581 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) || 2582 ctx->s_qf_names[GRPQUOTA]); 2583 2584 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2585 test_opt(sb, USRQUOTA)); 2586 2587 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) || 2588 test_opt(sb, GRPQUOTA)); 2589 2590 if (usr_qf_name) { 2591 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2592 usrquota = false; 2593 } 2594 if (grp_qf_name) { 2595 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2596 grpquota = false; 2597 } 2598 2599 if (usr_qf_name || grp_qf_name) { 2600 if (usrquota || grpquota) { 2601 ext4_msg(NULL, KERN_ERR, "old and new quota " 2602 "format mixing"); 2603 return -EINVAL; 2604 } 2605 2606 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) { 2607 ext4_msg(NULL, KERN_ERR, "journaled quota format " 2608 "not specified"); 2609 return -EINVAL; 2610 } 2611 } 2612 2613 return 0; 2614 2615 err_quota_change: 2616 ext4_msg(NULL, KERN_ERR, 2617 "Cannot change quota options when quota turned on"); 2618 return -EINVAL; 2619 err_jquota_change: 2620 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota " 2621 "options when quota turned on"); 2622 return -EINVAL; 2623 err_jquota_specified: 2624 ext4_msg(NULL, KERN_ERR, "%s quota file already specified", 2625 QTYPE2NAME(i)); 2626 return -EINVAL; 2627 #else 2628 return 0; 2629 #endif 2630 } 2631 2632 static int ext4_check_test_dummy_encryption(const struct fs_context *fc, 2633 struct super_block *sb) 2634 { 2635 const struct ext4_fs_context *ctx = fc->fs_private; 2636 const struct ext4_sb_info *sbi = EXT4_SB(sb); 2637 int err; 2638 2639 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy)) 2640 return 0; 2641 2642 if (!ext4_has_feature_encrypt(sb)) { 2643 ext4_msg(NULL, KERN_WARNING, 2644 "test_dummy_encryption requires encrypt feature"); 2645 return -EINVAL; 2646 } 2647 /* 2648 * This mount option is just for testing, and it's not worthwhile to 2649 * implement the extra complexity (e.g. RCU protection) that would be 2650 * needed to allow it to be set or changed during remount. We do allow 2651 * it to be specified during remount, but only if there is no change. 2652 */ 2653 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 2654 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2655 &ctx->dummy_enc_policy)) 2656 return 0; 2657 ext4_msg(NULL, KERN_WARNING, 2658 "Can't set or change test_dummy_encryption on remount"); 2659 return -EINVAL; 2660 } 2661 /* Also make sure s_mount_opts didn't contain a conflicting value. */ 2662 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) { 2663 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy, 2664 &ctx->dummy_enc_policy)) 2665 return 0; 2666 ext4_msg(NULL, KERN_WARNING, 2667 "Conflicting test_dummy_encryption options"); 2668 return -EINVAL; 2669 } 2670 /* 2671 * fscrypt_add_test_dummy_key() technically changes the super_block, so 2672 * technically it should be delayed until ext4_apply_options() like the 2673 * other changes. But since we never get here for remounts (see above), 2674 * and this is the last chance to report errors, we do it here. 2675 */ 2676 err = fscrypt_add_test_dummy_key(sb, &ctx->dummy_enc_policy); 2677 if (err) 2678 ext4_msg(NULL, KERN_WARNING, 2679 "Error adding test dummy encryption key [%d]", err); 2680 return err; 2681 } 2682 2683 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx, 2684 struct super_block *sb) 2685 { 2686 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) || 2687 /* if already set, it was already verified to be the same */ 2688 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy)) 2689 return; 2690 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy; 2691 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy)); 2692 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled"); 2693 } 2694 2695 static int ext4_check_opt_consistency(struct fs_context *fc, 2696 struct super_block *sb) 2697 { 2698 struct ext4_fs_context *ctx = fc->fs_private; 2699 struct ext4_sb_info *sbi = fc->s_fs_info; 2700 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; 2701 int err; 2702 2703 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 2704 ext4_msg(NULL, KERN_ERR, 2705 "Mount option(s) incompatible with ext2"); 2706 return -EINVAL; 2707 } 2708 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 2709 ext4_msg(NULL, KERN_ERR, 2710 "Mount option(s) incompatible with ext3"); 2711 return -EINVAL; 2712 } 2713 2714 if (ctx->s_want_extra_isize > 2715 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) { 2716 ext4_msg(NULL, KERN_ERR, 2717 "Invalid want_extra_isize %d", 2718 ctx->s_want_extra_isize); 2719 return -EINVAL; 2720 } 2721 2722 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) { 2723 int blocksize = 2724 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2725 if (blocksize < PAGE_SIZE) 2726 ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an " 2727 "experimental mount option 'dioread_nolock' " 2728 "for blocksize < PAGE_SIZE"); 2729 } 2730 2731 err = ext4_check_test_dummy_encryption(fc, sb); 2732 if (err) 2733 return err; 2734 2735 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) { 2736 if (!sbi->s_journal) { 2737 ext4_msg(NULL, KERN_WARNING, 2738 "Remounting file system with no journal " 2739 "so ignoring journalled data option"); 2740 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS); 2741 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) != 2742 test_opt(sb, DATA_FLAGS)) { 2743 ext4_msg(NULL, KERN_ERR, "Cannot change data mode " 2744 "on remount"); 2745 return -EINVAL; 2746 } 2747 } 2748 2749 if (is_remount) { 2750 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2751 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) { 2752 ext4_msg(NULL, KERN_ERR, "can't mount with " 2753 "both data=journal and dax"); 2754 return -EINVAL; 2755 } 2756 2757 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) && 2758 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2759 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) { 2760 fail_dax_change_remount: 2761 ext4_msg(NULL, KERN_ERR, "can't change " 2762 "dax mount option while remounting"); 2763 return -EINVAL; 2764 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) && 2765 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2766 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) { 2767 goto fail_dax_change_remount; 2768 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) && 2769 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) || 2770 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) || 2771 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) { 2772 goto fail_dax_change_remount; 2773 } 2774 } 2775 2776 return ext4_check_quota_consistency(fc, sb); 2777 } 2778 2779 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb) 2780 { 2781 struct ext4_fs_context *ctx = fc->fs_private; 2782 struct ext4_sb_info *sbi = fc->s_fs_info; 2783 2784 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt; 2785 sbi->s_mount_opt |= ctx->vals_s_mount_opt; 2786 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2; 2787 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2; 2788 sbi->s_mount_flags &= ~ctx->mask_s_mount_flags; 2789 sbi->s_mount_flags |= ctx->vals_s_mount_flags; 2790 sb->s_flags &= ~ctx->mask_s_flags; 2791 sb->s_flags |= ctx->vals_s_flags; 2792 2793 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; }) 2794 APPLY(s_commit_interval); 2795 APPLY(s_stripe); 2796 APPLY(s_max_batch_time); 2797 APPLY(s_min_batch_time); 2798 APPLY(s_want_extra_isize); 2799 APPLY(s_inode_readahead_blks); 2800 APPLY(s_max_dir_size_kb); 2801 APPLY(s_li_wait_mult); 2802 APPLY(s_resgid); 2803 APPLY(s_resuid); 2804 2805 #ifdef CONFIG_EXT4_DEBUG 2806 APPLY(s_fc_debug_max_replay); 2807 #endif 2808 2809 ext4_apply_quota_options(fc, sb); 2810 ext4_apply_test_dummy_encryption(ctx, sb); 2811 } 2812 2813 2814 static int ext4_validate_options(struct fs_context *fc) 2815 { 2816 #ifdef CONFIG_QUOTA 2817 struct ext4_fs_context *ctx = fc->fs_private; 2818 char *usr_qf_name, *grp_qf_name; 2819 2820 usr_qf_name = ctx->s_qf_names[USRQUOTA]; 2821 grp_qf_name = ctx->s_qf_names[GRPQUOTA]; 2822 2823 if (usr_qf_name || grp_qf_name) { 2824 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name) 2825 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA); 2826 2827 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name) 2828 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA); 2829 2830 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) || 2831 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) { 2832 ext4_msg(NULL, KERN_ERR, "old and new quota " 2833 "format mixing"); 2834 return -EINVAL; 2835 } 2836 } 2837 #endif 2838 return 1; 2839 } 2840 2841 static inline void ext4_show_quota_options(struct seq_file *seq, 2842 struct super_block *sb) 2843 { 2844 #if defined(CONFIG_QUOTA) 2845 struct ext4_sb_info *sbi = EXT4_SB(sb); 2846 char *usr_qf_name, *grp_qf_name; 2847 2848 if (sbi->s_jquota_fmt) { 2849 char *fmtname = ""; 2850 2851 switch (sbi->s_jquota_fmt) { 2852 case QFMT_VFS_OLD: 2853 fmtname = "vfsold"; 2854 break; 2855 case QFMT_VFS_V0: 2856 fmtname = "vfsv0"; 2857 break; 2858 case QFMT_VFS_V1: 2859 fmtname = "vfsv1"; 2860 break; 2861 } 2862 seq_printf(seq, ",jqfmt=%s", fmtname); 2863 } 2864 2865 rcu_read_lock(); 2866 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2867 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2868 if (usr_qf_name) 2869 seq_show_option(seq, "usrjquota", usr_qf_name); 2870 if (grp_qf_name) 2871 seq_show_option(seq, "grpjquota", grp_qf_name); 2872 rcu_read_unlock(); 2873 #endif 2874 } 2875 2876 static const char *token2str(int token) 2877 { 2878 const struct fs_parameter_spec *spec; 2879 2880 for (spec = ext4_param_specs; spec->name != NULL; spec++) 2881 if (spec->opt == token && !spec->type) 2882 break; 2883 return spec->name; 2884 } 2885 2886 /* 2887 * Show an option if 2888 * - it's set to a non-default value OR 2889 * - if the per-sb default is different from the global default 2890 */ 2891 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2892 int nodefs) 2893 { 2894 struct ext4_sb_info *sbi = EXT4_SB(sb); 2895 struct ext4_super_block *es = sbi->s_es; 2896 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2897 const struct mount_opts *m; 2898 char sep = nodefs ? '\n' : ','; 2899 2900 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2901 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2902 2903 if (sbi->s_sb_block != 1) 2904 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2905 2906 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2907 int want_set = m->flags & MOPT_SET; 2908 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2909 m->flags & MOPT_SKIP) 2910 continue; 2911 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2912 continue; /* skip if same as the default */ 2913 if ((want_set && 2914 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2915 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2916 continue; /* select Opt_noFoo vs Opt_Foo */ 2917 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2918 } 2919 2920 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2921 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2922 SEQ_OPTS_PRINT("resuid=%u", 2923 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2924 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2925 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2926 SEQ_OPTS_PRINT("resgid=%u", 2927 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2928 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2929 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2930 SEQ_OPTS_PUTS("errors=remount-ro"); 2931 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2932 SEQ_OPTS_PUTS("errors=continue"); 2933 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2934 SEQ_OPTS_PUTS("errors=panic"); 2935 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2936 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2937 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2938 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2939 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2940 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2941 if (nodefs || sbi->s_stripe) 2942 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2943 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2944 (sbi->s_mount_opt ^ def_mount_opt)) { 2945 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2946 SEQ_OPTS_PUTS("data=journal"); 2947 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2948 SEQ_OPTS_PUTS("data=ordered"); 2949 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2950 SEQ_OPTS_PUTS("data=writeback"); 2951 } 2952 if (nodefs || 2953 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2954 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2955 sbi->s_inode_readahead_blks); 2956 2957 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2958 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2959 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2960 if (nodefs || sbi->s_max_dir_size_kb) 2961 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2962 if (test_opt(sb, DATA_ERR_ABORT)) 2963 SEQ_OPTS_PUTS("data_err=abort"); 2964 2965 fscrypt_show_test_dummy_encryption(seq, sep, sb); 2966 2967 if (sb->s_flags & SB_INLINECRYPT) 2968 SEQ_OPTS_PUTS("inlinecrypt"); 2969 2970 if (test_opt(sb, DAX_ALWAYS)) { 2971 if (IS_EXT2_SB(sb)) 2972 SEQ_OPTS_PUTS("dax"); 2973 else 2974 SEQ_OPTS_PUTS("dax=always"); 2975 } else if (test_opt2(sb, DAX_NEVER)) { 2976 SEQ_OPTS_PUTS("dax=never"); 2977 } else if (test_opt2(sb, DAX_INODE)) { 2978 SEQ_OPTS_PUTS("dax=inode"); 2979 } 2980 2981 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 2982 !test_opt2(sb, MB_OPTIMIZE_SCAN)) { 2983 SEQ_OPTS_PUTS("mb_optimize_scan=0"); 2984 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD && 2985 test_opt2(sb, MB_OPTIMIZE_SCAN)) { 2986 SEQ_OPTS_PUTS("mb_optimize_scan=1"); 2987 } 2988 2989 ext4_show_quota_options(seq, sb); 2990 return 0; 2991 } 2992 2993 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2994 { 2995 return _ext4_show_options(seq, root->d_sb, 0); 2996 } 2997 2998 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2999 { 3000 struct super_block *sb = seq->private; 3001 int rc; 3002 3003 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 3004 rc = _ext4_show_options(seq, sb, 1); 3005 seq_puts(seq, "\n"); 3006 return rc; 3007 } 3008 3009 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 3010 int read_only) 3011 { 3012 struct ext4_sb_info *sbi = EXT4_SB(sb); 3013 int err = 0; 3014 3015 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 3016 ext4_msg(sb, KERN_ERR, "revision level too high, " 3017 "forcing read-only mode"); 3018 err = -EROFS; 3019 goto done; 3020 } 3021 if (read_only) 3022 goto done; 3023 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 3024 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 3025 "running e2fsck is recommended"); 3026 else if (sbi->s_mount_state & EXT4_ERROR_FS) 3027 ext4_msg(sb, KERN_WARNING, 3028 "warning: mounting fs with errors, " 3029 "running e2fsck is recommended"); 3030 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 3031 le16_to_cpu(es->s_mnt_count) >= 3032 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 3033 ext4_msg(sb, KERN_WARNING, 3034 "warning: maximal mount count reached, " 3035 "running e2fsck is recommended"); 3036 else if (le32_to_cpu(es->s_checkinterval) && 3037 (ext4_get_tstamp(es, s_lastcheck) + 3038 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 3039 ext4_msg(sb, KERN_WARNING, 3040 "warning: checktime reached, " 3041 "running e2fsck is recommended"); 3042 if (!sbi->s_journal) 3043 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 3044 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 3045 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 3046 le16_add_cpu(&es->s_mnt_count, 1); 3047 ext4_update_tstamp(es, s_mtime); 3048 if (sbi->s_journal) { 3049 ext4_set_feature_journal_needs_recovery(sb); 3050 if (ext4_has_feature_orphan_file(sb)) 3051 ext4_set_feature_orphan_present(sb); 3052 } 3053 3054 err = ext4_commit_super(sb); 3055 done: 3056 if (test_opt(sb, DEBUG)) 3057 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 3058 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 3059 sb->s_blocksize, 3060 sbi->s_groups_count, 3061 EXT4_BLOCKS_PER_GROUP(sb), 3062 EXT4_INODES_PER_GROUP(sb), 3063 sbi->s_mount_opt, sbi->s_mount_opt2); 3064 return err; 3065 } 3066 3067 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 3068 { 3069 struct ext4_sb_info *sbi = EXT4_SB(sb); 3070 struct flex_groups **old_groups, **new_groups; 3071 int size, i, j; 3072 3073 if (!sbi->s_log_groups_per_flex) 3074 return 0; 3075 3076 size = ext4_flex_group(sbi, ngroup - 1) + 1; 3077 if (size <= sbi->s_flex_groups_allocated) 3078 return 0; 3079 3080 new_groups = kvzalloc(roundup_pow_of_two(size * 3081 sizeof(*sbi->s_flex_groups)), GFP_KERNEL); 3082 if (!new_groups) { 3083 ext4_msg(sb, KERN_ERR, 3084 "not enough memory for %d flex group pointers", size); 3085 return -ENOMEM; 3086 } 3087 for (i = sbi->s_flex_groups_allocated; i < size; i++) { 3088 new_groups[i] = kvzalloc(roundup_pow_of_two( 3089 sizeof(struct flex_groups)), 3090 GFP_KERNEL); 3091 if (!new_groups[i]) { 3092 for (j = sbi->s_flex_groups_allocated; j < i; j++) 3093 kvfree(new_groups[j]); 3094 kvfree(new_groups); 3095 ext4_msg(sb, KERN_ERR, 3096 "not enough memory for %d flex groups", size); 3097 return -ENOMEM; 3098 } 3099 } 3100 rcu_read_lock(); 3101 old_groups = rcu_dereference(sbi->s_flex_groups); 3102 if (old_groups) 3103 memcpy(new_groups, old_groups, 3104 (sbi->s_flex_groups_allocated * 3105 sizeof(struct flex_groups *))); 3106 rcu_read_unlock(); 3107 rcu_assign_pointer(sbi->s_flex_groups, new_groups); 3108 sbi->s_flex_groups_allocated = size; 3109 if (old_groups) 3110 ext4_kvfree_array_rcu(old_groups); 3111 return 0; 3112 } 3113 3114 static int ext4_fill_flex_info(struct super_block *sb) 3115 { 3116 struct ext4_sb_info *sbi = EXT4_SB(sb); 3117 struct ext4_group_desc *gdp = NULL; 3118 struct flex_groups *fg; 3119 ext4_group_t flex_group; 3120 int i, err; 3121 3122 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 3123 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 3124 sbi->s_log_groups_per_flex = 0; 3125 return 1; 3126 } 3127 3128 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 3129 if (err) 3130 goto failed; 3131 3132 for (i = 0; i < sbi->s_groups_count; i++) { 3133 gdp = ext4_get_group_desc(sb, i, NULL); 3134 3135 flex_group = ext4_flex_group(sbi, i); 3136 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 3137 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes); 3138 atomic64_add(ext4_free_group_clusters(sb, gdp), 3139 &fg->free_clusters); 3140 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs); 3141 } 3142 3143 return 1; 3144 failed: 3145 return 0; 3146 } 3147 3148 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 3149 struct ext4_group_desc *gdp) 3150 { 3151 int offset = offsetof(struct ext4_group_desc, bg_checksum); 3152 __u16 crc = 0; 3153 __le32 le_group = cpu_to_le32(block_group); 3154 struct ext4_sb_info *sbi = EXT4_SB(sb); 3155 3156 if (ext4_has_metadata_csum(sbi->s_sb)) { 3157 /* Use new metadata_csum algorithm */ 3158 __u32 csum32; 3159 __u16 dummy_csum = 0; 3160 3161 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 3162 sizeof(le_group)); 3163 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 3164 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 3165 sizeof(dummy_csum)); 3166 offset += sizeof(dummy_csum); 3167 if (offset < sbi->s_desc_size) 3168 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 3169 sbi->s_desc_size - offset); 3170 3171 crc = csum32 & 0xFFFF; 3172 goto out; 3173 } 3174 3175 /* old crc16 code */ 3176 if (!ext4_has_feature_gdt_csum(sb)) 3177 return 0; 3178 3179 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 3180 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 3181 crc = crc16(crc, (__u8 *)gdp, offset); 3182 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 3183 /* for checksum of struct ext4_group_desc do the rest...*/ 3184 if (ext4_has_feature_64bit(sb) && 3185 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 3186 crc = crc16(crc, (__u8 *)gdp + offset, 3187 le16_to_cpu(sbi->s_es->s_desc_size) - 3188 offset); 3189 3190 out: 3191 return cpu_to_le16(crc); 3192 } 3193 3194 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 3195 struct ext4_group_desc *gdp) 3196 { 3197 if (ext4_has_group_desc_csum(sb) && 3198 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 3199 return 0; 3200 3201 return 1; 3202 } 3203 3204 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 3205 struct ext4_group_desc *gdp) 3206 { 3207 if (!ext4_has_group_desc_csum(sb)) 3208 return; 3209 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 3210 } 3211 3212 /* Called at mount-time, super-block is locked */ 3213 static int ext4_check_descriptors(struct super_block *sb, 3214 ext4_fsblk_t sb_block, 3215 ext4_group_t *first_not_zeroed) 3216 { 3217 struct ext4_sb_info *sbi = EXT4_SB(sb); 3218 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 3219 ext4_fsblk_t last_block; 3220 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 3221 ext4_fsblk_t block_bitmap; 3222 ext4_fsblk_t inode_bitmap; 3223 ext4_fsblk_t inode_table; 3224 int flexbg_flag = 0; 3225 ext4_group_t i, grp = sbi->s_groups_count; 3226 3227 if (ext4_has_feature_flex_bg(sb)) 3228 flexbg_flag = 1; 3229 3230 ext4_debug("Checking group descriptors"); 3231 3232 for (i = 0; i < sbi->s_groups_count; i++) { 3233 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 3234 3235 if (i == sbi->s_groups_count - 1 || flexbg_flag) 3236 last_block = ext4_blocks_count(sbi->s_es) - 1; 3237 else 3238 last_block = first_block + 3239 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 3240 3241 if ((grp == sbi->s_groups_count) && 3242 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3243 grp = i; 3244 3245 block_bitmap = ext4_block_bitmap(sb, gdp); 3246 if (block_bitmap == sb_block) { 3247 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3248 "Block bitmap for group %u overlaps " 3249 "superblock", i); 3250 if (!sb_rdonly(sb)) 3251 return 0; 3252 } 3253 if (block_bitmap >= sb_block + 1 && 3254 block_bitmap <= last_bg_block) { 3255 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3256 "Block bitmap for group %u overlaps " 3257 "block group descriptors", i); 3258 if (!sb_rdonly(sb)) 3259 return 0; 3260 } 3261 if (block_bitmap < first_block || block_bitmap > last_block) { 3262 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3263 "Block bitmap for group %u not in group " 3264 "(block %llu)!", i, block_bitmap); 3265 return 0; 3266 } 3267 inode_bitmap = ext4_inode_bitmap(sb, gdp); 3268 if (inode_bitmap == sb_block) { 3269 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3270 "Inode bitmap for group %u overlaps " 3271 "superblock", i); 3272 if (!sb_rdonly(sb)) 3273 return 0; 3274 } 3275 if (inode_bitmap >= sb_block + 1 && 3276 inode_bitmap <= last_bg_block) { 3277 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3278 "Inode bitmap for group %u overlaps " 3279 "block group descriptors", i); 3280 if (!sb_rdonly(sb)) 3281 return 0; 3282 } 3283 if (inode_bitmap < first_block || inode_bitmap > last_block) { 3284 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3285 "Inode bitmap for group %u not in group " 3286 "(block %llu)!", i, inode_bitmap); 3287 return 0; 3288 } 3289 inode_table = ext4_inode_table(sb, gdp); 3290 if (inode_table == sb_block) { 3291 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3292 "Inode table for group %u overlaps " 3293 "superblock", i); 3294 if (!sb_rdonly(sb)) 3295 return 0; 3296 } 3297 if (inode_table >= sb_block + 1 && 3298 inode_table <= last_bg_block) { 3299 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3300 "Inode table for group %u overlaps " 3301 "block group descriptors", i); 3302 if (!sb_rdonly(sb)) 3303 return 0; 3304 } 3305 if (inode_table < first_block || 3306 inode_table + sbi->s_itb_per_group - 1 > last_block) { 3307 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3308 "Inode table for group %u not in group " 3309 "(block %llu)!", i, inode_table); 3310 return 0; 3311 } 3312 ext4_lock_group(sb, i); 3313 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 3314 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 3315 "Checksum for group %u failed (%u!=%u)", 3316 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 3317 gdp)), le16_to_cpu(gdp->bg_checksum)); 3318 if (!sb_rdonly(sb)) { 3319 ext4_unlock_group(sb, i); 3320 return 0; 3321 } 3322 } 3323 ext4_unlock_group(sb, i); 3324 if (!flexbg_flag) 3325 first_block += EXT4_BLOCKS_PER_GROUP(sb); 3326 } 3327 if (NULL != first_not_zeroed) 3328 *first_not_zeroed = grp; 3329 return 1; 3330 } 3331 3332 /* 3333 * Maximal extent format file size. 3334 * Resulting logical blkno at s_maxbytes must fit in our on-disk 3335 * extent format containers, within a sector_t, and within i_blocks 3336 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 3337 * so that won't be a limiting factor. 3338 * 3339 * However there is other limiting factor. We do store extents in the form 3340 * of starting block and length, hence the resulting length of the extent 3341 * covering maximum file size must fit into on-disk format containers as 3342 * well. Given that length is always by 1 unit bigger than max unit (because 3343 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 3344 * 3345 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 3346 */ 3347 static loff_t ext4_max_size(int blkbits, int has_huge_files) 3348 { 3349 loff_t res; 3350 loff_t upper_limit = MAX_LFS_FILESIZE; 3351 3352 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 3353 3354 if (!has_huge_files) { 3355 upper_limit = (1LL << 32) - 1; 3356 3357 /* total blocks in file system block size */ 3358 upper_limit >>= (blkbits - 9); 3359 upper_limit <<= blkbits; 3360 } 3361 3362 /* 3363 * 32-bit extent-start container, ee_block. We lower the maxbytes 3364 * by one fs block, so ee_len can cover the extent of maximum file 3365 * size 3366 */ 3367 res = (1LL << 32) - 1; 3368 res <<= blkbits; 3369 3370 /* Sanity check against vm- & vfs- imposed limits */ 3371 if (res > upper_limit) 3372 res = upper_limit; 3373 3374 return res; 3375 } 3376 3377 /* 3378 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 3379 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 3380 * We need to be 1 filesystem block less than the 2^48 sector limit. 3381 */ 3382 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 3383 { 3384 loff_t upper_limit, res = EXT4_NDIR_BLOCKS; 3385 int meta_blocks; 3386 unsigned int ppb = 1 << (bits - 2); 3387 3388 /* 3389 * This is calculated to be the largest file size for a dense, block 3390 * mapped file such that the file's total number of 512-byte sectors, 3391 * including data and all indirect blocks, does not exceed (2^48 - 1). 3392 * 3393 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 3394 * number of 512-byte sectors of the file. 3395 */ 3396 if (!has_huge_files) { 3397 /* 3398 * !has_huge_files or implies that the inode i_block field 3399 * represents total file blocks in 2^32 512-byte sectors == 3400 * size of vfs inode i_blocks * 8 3401 */ 3402 upper_limit = (1LL << 32) - 1; 3403 3404 /* total blocks in file system block size */ 3405 upper_limit >>= (bits - 9); 3406 3407 } else { 3408 /* 3409 * We use 48 bit ext4_inode i_blocks 3410 * With EXT4_HUGE_FILE_FL set the i_blocks 3411 * represent total number of blocks in 3412 * file system block size 3413 */ 3414 upper_limit = (1LL << 48) - 1; 3415 3416 } 3417 3418 /* Compute how many blocks we can address by block tree */ 3419 res += ppb; 3420 res += ppb * ppb; 3421 res += ((loff_t)ppb) * ppb * ppb; 3422 /* Compute how many metadata blocks are needed */ 3423 meta_blocks = 1; 3424 meta_blocks += 1 + ppb; 3425 meta_blocks += 1 + ppb + ppb * ppb; 3426 /* Does block tree limit file size? */ 3427 if (res + meta_blocks <= upper_limit) 3428 goto check_lfs; 3429 3430 res = upper_limit; 3431 /* How many metadata blocks are needed for addressing upper_limit? */ 3432 upper_limit -= EXT4_NDIR_BLOCKS; 3433 /* indirect blocks */ 3434 meta_blocks = 1; 3435 upper_limit -= ppb; 3436 /* double indirect blocks */ 3437 if (upper_limit < ppb * ppb) { 3438 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb); 3439 res -= meta_blocks; 3440 goto check_lfs; 3441 } 3442 meta_blocks += 1 + ppb; 3443 upper_limit -= ppb * ppb; 3444 /* tripple indirect blocks for the rest */ 3445 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) + 3446 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb); 3447 res -= meta_blocks; 3448 check_lfs: 3449 res <<= bits; 3450 if (res > MAX_LFS_FILESIZE) 3451 res = MAX_LFS_FILESIZE; 3452 3453 return res; 3454 } 3455 3456 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 3457 ext4_fsblk_t logical_sb_block, int nr) 3458 { 3459 struct ext4_sb_info *sbi = EXT4_SB(sb); 3460 ext4_group_t bg, first_meta_bg; 3461 int has_super = 0; 3462 3463 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 3464 3465 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 3466 return logical_sb_block + nr + 1; 3467 bg = sbi->s_desc_per_block * nr; 3468 if (ext4_bg_has_super(sb, bg)) 3469 has_super = 1; 3470 3471 /* 3472 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 3473 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 3474 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 3475 * compensate. 3476 */ 3477 if (sb->s_blocksize == 1024 && nr == 0 && 3478 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 3479 has_super++; 3480 3481 return (has_super + ext4_group_first_block_no(sb, bg)); 3482 } 3483 3484 /** 3485 * ext4_get_stripe_size: Get the stripe size. 3486 * @sbi: In memory super block info 3487 * 3488 * If we have specified it via mount option, then 3489 * use the mount option value. If the value specified at mount time is 3490 * greater than the blocks per group use the super block value. 3491 * If the super block value is greater than blocks per group return 0. 3492 * Allocator needs it be less than blocks per group. 3493 * 3494 */ 3495 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 3496 { 3497 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 3498 unsigned long stripe_width = 3499 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 3500 int ret; 3501 3502 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 3503 ret = sbi->s_stripe; 3504 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 3505 ret = stripe_width; 3506 else if (stride && stride <= sbi->s_blocks_per_group) 3507 ret = stride; 3508 else 3509 ret = 0; 3510 3511 /* 3512 * If the stripe width is 1, this makes no sense and 3513 * we set it to 0 to turn off stripe handling code. 3514 */ 3515 if (ret <= 1) 3516 ret = 0; 3517 3518 return ret; 3519 } 3520 3521 /* 3522 * Check whether this filesystem can be mounted based on 3523 * the features present and the RDONLY/RDWR mount requested. 3524 * Returns 1 if this filesystem can be mounted as requested, 3525 * 0 if it cannot be. 3526 */ 3527 int ext4_feature_set_ok(struct super_block *sb, int readonly) 3528 { 3529 if (ext4_has_unknown_ext4_incompat_features(sb)) { 3530 ext4_msg(sb, KERN_ERR, 3531 "Couldn't mount because of " 3532 "unsupported optional features (%x)", 3533 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 3534 ~EXT4_FEATURE_INCOMPAT_SUPP)); 3535 return 0; 3536 } 3537 3538 #if !IS_ENABLED(CONFIG_UNICODE) 3539 if (ext4_has_feature_casefold(sb)) { 3540 ext4_msg(sb, KERN_ERR, 3541 "Filesystem with casefold feature cannot be " 3542 "mounted without CONFIG_UNICODE"); 3543 return 0; 3544 } 3545 #endif 3546 3547 if (readonly) 3548 return 1; 3549 3550 if (ext4_has_feature_readonly(sb)) { 3551 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 3552 sb->s_flags |= SB_RDONLY; 3553 return 1; 3554 } 3555 3556 /* Check that feature set is OK for a read-write mount */ 3557 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 3558 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 3559 "unsupported optional features (%x)", 3560 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 3561 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 3562 return 0; 3563 } 3564 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 3565 ext4_msg(sb, KERN_ERR, 3566 "Can't support bigalloc feature without " 3567 "extents feature\n"); 3568 return 0; 3569 } 3570 3571 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2) 3572 if (!readonly && (ext4_has_feature_quota(sb) || 3573 ext4_has_feature_project(sb))) { 3574 ext4_msg(sb, KERN_ERR, 3575 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2"); 3576 return 0; 3577 } 3578 #endif /* CONFIG_QUOTA */ 3579 return 1; 3580 } 3581 3582 /* 3583 * This function is called once a day if we have errors logged 3584 * on the file system 3585 */ 3586 static void print_daily_error_info(struct timer_list *t) 3587 { 3588 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 3589 struct super_block *sb = sbi->s_sb; 3590 struct ext4_super_block *es = sbi->s_es; 3591 3592 if (es->s_error_count) 3593 /* fsck newer than v1.41.13 is needed to clean this condition. */ 3594 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 3595 le32_to_cpu(es->s_error_count)); 3596 if (es->s_first_error_time) { 3597 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 3598 sb->s_id, 3599 ext4_get_tstamp(es, s_first_error_time), 3600 (int) sizeof(es->s_first_error_func), 3601 es->s_first_error_func, 3602 le32_to_cpu(es->s_first_error_line)); 3603 if (es->s_first_error_ino) 3604 printk(KERN_CONT ": inode %u", 3605 le32_to_cpu(es->s_first_error_ino)); 3606 if (es->s_first_error_block) 3607 printk(KERN_CONT ": block %llu", (unsigned long long) 3608 le64_to_cpu(es->s_first_error_block)); 3609 printk(KERN_CONT "\n"); 3610 } 3611 if (es->s_last_error_time) { 3612 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3613 sb->s_id, 3614 ext4_get_tstamp(es, s_last_error_time), 3615 (int) sizeof(es->s_last_error_func), 3616 es->s_last_error_func, 3617 le32_to_cpu(es->s_last_error_line)); 3618 if (es->s_last_error_ino) 3619 printk(KERN_CONT ": inode %u", 3620 le32_to_cpu(es->s_last_error_ino)); 3621 if (es->s_last_error_block) 3622 printk(KERN_CONT ": block %llu", (unsigned long long) 3623 le64_to_cpu(es->s_last_error_block)); 3624 printk(KERN_CONT "\n"); 3625 } 3626 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3627 } 3628 3629 /* Find next suitable group and run ext4_init_inode_table */ 3630 static int ext4_run_li_request(struct ext4_li_request *elr) 3631 { 3632 struct ext4_group_desc *gdp = NULL; 3633 struct super_block *sb = elr->lr_super; 3634 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; 3635 ext4_group_t group = elr->lr_next_group; 3636 unsigned int prefetch_ios = 0; 3637 int ret = 0; 3638 u64 start_time; 3639 3640 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) { 3641 elr->lr_next_group = ext4_mb_prefetch(sb, group, 3642 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios); 3643 if (prefetch_ios) 3644 ext4_mb_prefetch_fini(sb, elr->lr_next_group, 3645 prefetch_ios); 3646 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, 3647 prefetch_ios); 3648 if (group >= elr->lr_next_group) { 3649 ret = 1; 3650 if (elr->lr_first_not_zeroed != ngroups && 3651 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) { 3652 elr->lr_next_group = elr->lr_first_not_zeroed; 3653 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3654 ret = 0; 3655 } 3656 } 3657 return ret; 3658 } 3659 3660 for (; group < ngroups; group++) { 3661 gdp = ext4_get_group_desc(sb, group, NULL); 3662 if (!gdp) { 3663 ret = 1; 3664 break; 3665 } 3666 3667 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3668 break; 3669 } 3670 3671 if (group >= ngroups) 3672 ret = 1; 3673 3674 if (!ret) { 3675 start_time = ktime_get_real_ns(); 3676 ret = ext4_init_inode_table(sb, group, 3677 elr->lr_timeout ? 0 : 1); 3678 trace_ext4_lazy_itable_init(sb, group); 3679 if (elr->lr_timeout == 0) { 3680 elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) * 3681 EXT4_SB(elr->lr_super)->s_li_wait_mult); 3682 } 3683 elr->lr_next_sched = jiffies + elr->lr_timeout; 3684 elr->lr_next_group = group + 1; 3685 } 3686 return ret; 3687 } 3688 3689 /* 3690 * Remove lr_request from the list_request and free the 3691 * request structure. Should be called with li_list_mtx held 3692 */ 3693 static void ext4_remove_li_request(struct ext4_li_request *elr) 3694 { 3695 if (!elr) 3696 return; 3697 3698 list_del(&elr->lr_request); 3699 EXT4_SB(elr->lr_super)->s_li_request = NULL; 3700 kfree(elr); 3701 } 3702 3703 static void ext4_unregister_li_request(struct super_block *sb) 3704 { 3705 mutex_lock(&ext4_li_mtx); 3706 if (!ext4_li_info) { 3707 mutex_unlock(&ext4_li_mtx); 3708 return; 3709 } 3710 3711 mutex_lock(&ext4_li_info->li_list_mtx); 3712 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3713 mutex_unlock(&ext4_li_info->li_list_mtx); 3714 mutex_unlock(&ext4_li_mtx); 3715 } 3716 3717 static struct task_struct *ext4_lazyinit_task; 3718 3719 /* 3720 * This is the function where ext4lazyinit thread lives. It walks 3721 * through the request list searching for next scheduled filesystem. 3722 * When such a fs is found, run the lazy initialization request 3723 * (ext4_rn_li_request) and keep track of the time spend in this 3724 * function. Based on that time we compute next schedule time of 3725 * the request. When walking through the list is complete, compute 3726 * next waking time and put itself into sleep. 3727 */ 3728 static int ext4_lazyinit_thread(void *arg) 3729 { 3730 struct ext4_lazy_init *eli = arg; 3731 struct list_head *pos, *n; 3732 struct ext4_li_request *elr; 3733 unsigned long next_wakeup, cur; 3734 3735 BUG_ON(NULL == eli); 3736 set_freezable(); 3737 3738 cont_thread: 3739 while (true) { 3740 next_wakeup = MAX_JIFFY_OFFSET; 3741 3742 mutex_lock(&eli->li_list_mtx); 3743 if (list_empty(&eli->li_request_list)) { 3744 mutex_unlock(&eli->li_list_mtx); 3745 goto exit_thread; 3746 } 3747 list_for_each_safe(pos, n, &eli->li_request_list) { 3748 int err = 0; 3749 int progress = 0; 3750 elr = list_entry(pos, struct ext4_li_request, 3751 lr_request); 3752 3753 if (time_before(jiffies, elr->lr_next_sched)) { 3754 if (time_before(elr->lr_next_sched, next_wakeup)) 3755 next_wakeup = elr->lr_next_sched; 3756 continue; 3757 } 3758 if (down_read_trylock(&elr->lr_super->s_umount)) { 3759 if (sb_start_write_trylock(elr->lr_super)) { 3760 progress = 1; 3761 /* 3762 * We hold sb->s_umount, sb can not 3763 * be removed from the list, it is 3764 * now safe to drop li_list_mtx 3765 */ 3766 mutex_unlock(&eli->li_list_mtx); 3767 err = ext4_run_li_request(elr); 3768 sb_end_write(elr->lr_super); 3769 mutex_lock(&eli->li_list_mtx); 3770 n = pos->next; 3771 } 3772 up_read((&elr->lr_super->s_umount)); 3773 } 3774 /* error, remove the lazy_init job */ 3775 if (err) { 3776 ext4_remove_li_request(elr); 3777 continue; 3778 } 3779 if (!progress) { 3780 elr->lr_next_sched = jiffies + 3781 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3782 } 3783 if (time_before(elr->lr_next_sched, next_wakeup)) 3784 next_wakeup = elr->lr_next_sched; 3785 } 3786 mutex_unlock(&eli->li_list_mtx); 3787 3788 try_to_freeze(); 3789 3790 cur = jiffies; 3791 if ((time_after_eq(cur, next_wakeup)) || 3792 (MAX_JIFFY_OFFSET == next_wakeup)) { 3793 cond_resched(); 3794 continue; 3795 } 3796 3797 schedule_timeout_interruptible(next_wakeup - cur); 3798 3799 if (kthread_should_stop()) { 3800 ext4_clear_request_list(); 3801 goto exit_thread; 3802 } 3803 } 3804 3805 exit_thread: 3806 /* 3807 * It looks like the request list is empty, but we need 3808 * to check it under the li_list_mtx lock, to prevent any 3809 * additions into it, and of course we should lock ext4_li_mtx 3810 * to atomically free the list and ext4_li_info, because at 3811 * this point another ext4 filesystem could be registering 3812 * new one. 3813 */ 3814 mutex_lock(&ext4_li_mtx); 3815 mutex_lock(&eli->li_list_mtx); 3816 if (!list_empty(&eli->li_request_list)) { 3817 mutex_unlock(&eli->li_list_mtx); 3818 mutex_unlock(&ext4_li_mtx); 3819 goto cont_thread; 3820 } 3821 mutex_unlock(&eli->li_list_mtx); 3822 kfree(ext4_li_info); 3823 ext4_li_info = NULL; 3824 mutex_unlock(&ext4_li_mtx); 3825 3826 return 0; 3827 } 3828 3829 static void ext4_clear_request_list(void) 3830 { 3831 struct list_head *pos, *n; 3832 struct ext4_li_request *elr; 3833 3834 mutex_lock(&ext4_li_info->li_list_mtx); 3835 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3836 elr = list_entry(pos, struct ext4_li_request, 3837 lr_request); 3838 ext4_remove_li_request(elr); 3839 } 3840 mutex_unlock(&ext4_li_info->li_list_mtx); 3841 } 3842 3843 static int ext4_run_lazyinit_thread(void) 3844 { 3845 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3846 ext4_li_info, "ext4lazyinit"); 3847 if (IS_ERR(ext4_lazyinit_task)) { 3848 int err = PTR_ERR(ext4_lazyinit_task); 3849 ext4_clear_request_list(); 3850 kfree(ext4_li_info); 3851 ext4_li_info = NULL; 3852 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3853 "initialization thread\n", 3854 err); 3855 return err; 3856 } 3857 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3858 return 0; 3859 } 3860 3861 /* 3862 * Check whether it make sense to run itable init. thread or not. 3863 * If there is at least one uninitialized inode table, return 3864 * corresponding group number, else the loop goes through all 3865 * groups and return total number of groups. 3866 */ 3867 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3868 { 3869 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3870 struct ext4_group_desc *gdp = NULL; 3871 3872 if (!ext4_has_group_desc_csum(sb)) 3873 return ngroups; 3874 3875 for (group = 0; group < ngroups; group++) { 3876 gdp = ext4_get_group_desc(sb, group, NULL); 3877 if (!gdp) 3878 continue; 3879 3880 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3881 break; 3882 } 3883 3884 return group; 3885 } 3886 3887 static int ext4_li_info_new(void) 3888 { 3889 struct ext4_lazy_init *eli = NULL; 3890 3891 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3892 if (!eli) 3893 return -ENOMEM; 3894 3895 INIT_LIST_HEAD(&eli->li_request_list); 3896 mutex_init(&eli->li_list_mtx); 3897 3898 eli->li_state |= EXT4_LAZYINIT_QUIT; 3899 3900 ext4_li_info = eli; 3901 3902 return 0; 3903 } 3904 3905 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3906 ext4_group_t start) 3907 { 3908 struct ext4_li_request *elr; 3909 3910 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3911 if (!elr) 3912 return NULL; 3913 3914 elr->lr_super = sb; 3915 elr->lr_first_not_zeroed = start; 3916 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) { 3917 elr->lr_mode = EXT4_LI_MODE_ITABLE; 3918 elr->lr_next_group = start; 3919 } else { 3920 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP; 3921 } 3922 3923 /* 3924 * Randomize first schedule time of the request to 3925 * spread the inode table initialization requests 3926 * better. 3927 */ 3928 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ); 3929 return elr; 3930 } 3931 3932 int ext4_register_li_request(struct super_block *sb, 3933 ext4_group_t first_not_zeroed) 3934 { 3935 struct ext4_sb_info *sbi = EXT4_SB(sb); 3936 struct ext4_li_request *elr = NULL; 3937 ext4_group_t ngroups = sbi->s_groups_count; 3938 int ret = 0; 3939 3940 mutex_lock(&ext4_li_mtx); 3941 if (sbi->s_li_request != NULL) { 3942 /* 3943 * Reset timeout so it can be computed again, because 3944 * s_li_wait_mult might have changed. 3945 */ 3946 sbi->s_li_request->lr_timeout = 0; 3947 goto out; 3948 } 3949 3950 if (sb_rdonly(sb) || 3951 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) && 3952 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE)))) 3953 goto out; 3954 3955 elr = ext4_li_request_new(sb, first_not_zeroed); 3956 if (!elr) { 3957 ret = -ENOMEM; 3958 goto out; 3959 } 3960 3961 if (NULL == ext4_li_info) { 3962 ret = ext4_li_info_new(); 3963 if (ret) 3964 goto out; 3965 } 3966 3967 mutex_lock(&ext4_li_info->li_list_mtx); 3968 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3969 mutex_unlock(&ext4_li_info->li_list_mtx); 3970 3971 sbi->s_li_request = elr; 3972 /* 3973 * set elr to NULL here since it has been inserted to 3974 * the request_list and the removal and free of it is 3975 * handled by ext4_clear_request_list from now on. 3976 */ 3977 elr = NULL; 3978 3979 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3980 ret = ext4_run_lazyinit_thread(); 3981 if (ret) 3982 goto out; 3983 } 3984 out: 3985 mutex_unlock(&ext4_li_mtx); 3986 if (ret) 3987 kfree(elr); 3988 return ret; 3989 } 3990 3991 /* 3992 * We do not need to lock anything since this is called on 3993 * module unload. 3994 */ 3995 static void ext4_destroy_lazyinit_thread(void) 3996 { 3997 /* 3998 * If thread exited earlier 3999 * there's nothing to be done. 4000 */ 4001 if (!ext4_li_info || !ext4_lazyinit_task) 4002 return; 4003 4004 kthread_stop(ext4_lazyinit_task); 4005 } 4006 4007 static int set_journal_csum_feature_set(struct super_block *sb) 4008 { 4009 int ret = 1; 4010 int compat, incompat; 4011 struct ext4_sb_info *sbi = EXT4_SB(sb); 4012 4013 if (ext4_has_metadata_csum(sb)) { 4014 /* journal checksum v3 */ 4015 compat = 0; 4016 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 4017 } else { 4018 /* journal checksum v1 */ 4019 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 4020 incompat = 0; 4021 } 4022 4023 jbd2_journal_clear_features(sbi->s_journal, 4024 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 4025 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 4026 JBD2_FEATURE_INCOMPAT_CSUM_V2); 4027 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4028 ret = jbd2_journal_set_features(sbi->s_journal, 4029 compat, 0, 4030 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 4031 incompat); 4032 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 4033 ret = jbd2_journal_set_features(sbi->s_journal, 4034 compat, 0, 4035 incompat); 4036 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4037 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4038 } else { 4039 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 4040 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 4041 } 4042 4043 return ret; 4044 } 4045 4046 /* 4047 * Note: calculating the overhead so we can be compatible with 4048 * historical BSD practice is quite difficult in the face of 4049 * clusters/bigalloc. This is because multiple metadata blocks from 4050 * different block group can end up in the same allocation cluster. 4051 * Calculating the exact overhead in the face of clustered allocation 4052 * requires either O(all block bitmaps) in memory or O(number of block 4053 * groups**2) in time. We will still calculate the superblock for 4054 * older file systems --- and if we come across with a bigalloc file 4055 * system with zero in s_overhead_clusters the estimate will be close to 4056 * correct especially for very large cluster sizes --- but for newer 4057 * file systems, it's better to calculate this figure once at mkfs 4058 * time, and store it in the superblock. If the superblock value is 4059 * present (even for non-bigalloc file systems), we will use it. 4060 */ 4061 static int count_overhead(struct super_block *sb, ext4_group_t grp, 4062 char *buf) 4063 { 4064 struct ext4_sb_info *sbi = EXT4_SB(sb); 4065 struct ext4_group_desc *gdp; 4066 ext4_fsblk_t first_block, last_block, b; 4067 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4068 int s, j, count = 0; 4069 int has_super = ext4_bg_has_super(sb, grp); 4070 4071 if (!ext4_has_feature_bigalloc(sb)) 4072 return (has_super + ext4_bg_num_gdb(sb, grp) + 4073 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) + 4074 sbi->s_itb_per_group + 2); 4075 4076 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 4077 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 4078 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 4079 for (i = 0; i < ngroups; i++) { 4080 gdp = ext4_get_group_desc(sb, i, NULL); 4081 b = ext4_block_bitmap(sb, gdp); 4082 if (b >= first_block && b <= last_block) { 4083 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4084 count++; 4085 } 4086 b = ext4_inode_bitmap(sb, gdp); 4087 if (b >= first_block && b <= last_block) { 4088 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 4089 count++; 4090 } 4091 b = ext4_inode_table(sb, gdp); 4092 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 4093 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 4094 int c = EXT4_B2C(sbi, b - first_block); 4095 ext4_set_bit(c, buf); 4096 count++; 4097 } 4098 if (i != grp) 4099 continue; 4100 s = 0; 4101 if (ext4_bg_has_super(sb, grp)) { 4102 ext4_set_bit(s++, buf); 4103 count++; 4104 } 4105 j = ext4_bg_num_gdb(sb, grp); 4106 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 4107 ext4_error(sb, "Invalid number of block group " 4108 "descriptor blocks: %d", j); 4109 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 4110 } 4111 count += j; 4112 for (; j > 0; j--) 4113 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 4114 } 4115 if (!count) 4116 return 0; 4117 return EXT4_CLUSTERS_PER_GROUP(sb) - 4118 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 4119 } 4120 4121 /* 4122 * Compute the overhead and stash it in sbi->s_overhead 4123 */ 4124 int ext4_calculate_overhead(struct super_block *sb) 4125 { 4126 struct ext4_sb_info *sbi = EXT4_SB(sb); 4127 struct ext4_super_block *es = sbi->s_es; 4128 struct inode *j_inode; 4129 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 4130 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 4131 ext4_fsblk_t overhead = 0; 4132 char *buf = (char *) get_zeroed_page(GFP_NOFS); 4133 4134 if (!buf) 4135 return -ENOMEM; 4136 4137 /* 4138 * Compute the overhead (FS structures). This is constant 4139 * for a given filesystem unless the number of block groups 4140 * changes so we cache the previous value until it does. 4141 */ 4142 4143 /* 4144 * All of the blocks before first_data_block are overhead 4145 */ 4146 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 4147 4148 /* 4149 * Add the overhead found in each block group 4150 */ 4151 for (i = 0; i < ngroups; i++) { 4152 int blks; 4153 4154 blks = count_overhead(sb, i, buf); 4155 overhead += blks; 4156 if (blks) 4157 memset(buf, 0, PAGE_SIZE); 4158 cond_resched(); 4159 } 4160 4161 /* 4162 * Add the internal journal blocks whether the journal has been 4163 * loaded or not 4164 */ 4165 if (sbi->s_journal && !sbi->s_journal_bdev) 4166 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len); 4167 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) { 4168 /* j_inum for internal journal is non-zero */ 4169 j_inode = ext4_get_journal_inode(sb, j_inum); 4170 if (j_inode) { 4171 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 4172 overhead += EXT4_NUM_B2C(sbi, j_blocks); 4173 iput(j_inode); 4174 } else { 4175 ext4_msg(sb, KERN_ERR, "can't get journal size"); 4176 } 4177 } 4178 sbi->s_overhead = overhead; 4179 smp_wmb(); 4180 free_page((unsigned long) buf); 4181 return 0; 4182 } 4183 4184 static void ext4_set_resv_clusters(struct super_block *sb) 4185 { 4186 ext4_fsblk_t resv_clusters; 4187 struct ext4_sb_info *sbi = EXT4_SB(sb); 4188 4189 /* 4190 * There's no need to reserve anything when we aren't using extents. 4191 * The space estimates are exact, there are no unwritten extents, 4192 * hole punching doesn't need new metadata... This is needed especially 4193 * to keep ext2/3 backward compatibility. 4194 */ 4195 if (!ext4_has_feature_extents(sb)) 4196 return; 4197 /* 4198 * By default we reserve 2% or 4096 clusters, whichever is smaller. 4199 * This should cover the situations where we can not afford to run 4200 * out of space like for example punch hole, or converting 4201 * unwritten extents in delalloc path. In most cases such 4202 * allocation would require 1, or 2 blocks, higher numbers are 4203 * very rare. 4204 */ 4205 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 4206 sbi->s_cluster_bits); 4207 4208 do_div(resv_clusters, 50); 4209 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 4210 4211 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 4212 } 4213 4214 static const char *ext4_quota_mode(struct super_block *sb) 4215 { 4216 #ifdef CONFIG_QUOTA 4217 if (!ext4_quota_capable(sb)) 4218 return "none"; 4219 4220 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb)) 4221 return "journalled"; 4222 else 4223 return "writeback"; 4224 #else 4225 return "disabled"; 4226 #endif 4227 } 4228 4229 static void ext4_setup_csum_trigger(struct super_block *sb, 4230 enum ext4_journal_trigger_type type, 4231 void (*trigger)( 4232 struct jbd2_buffer_trigger_type *type, 4233 struct buffer_head *bh, 4234 void *mapped_data, 4235 size_t size)) 4236 { 4237 struct ext4_sb_info *sbi = EXT4_SB(sb); 4238 4239 sbi->s_journal_triggers[type].sb = sb; 4240 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger; 4241 } 4242 4243 static void ext4_free_sbi(struct ext4_sb_info *sbi) 4244 { 4245 if (!sbi) 4246 return; 4247 4248 kfree(sbi->s_blockgroup_lock); 4249 fs_put_dax(sbi->s_daxdev, NULL); 4250 kfree(sbi); 4251 } 4252 4253 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb) 4254 { 4255 struct ext4_sb_info *sbi; 4256 4257 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 4258 if (!sbi) 4259 return NULL; 4260 4261 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off, 4262 NULL, NULL); 4263 4264 sbi->s_blockgroup_lock = 4265 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 4266 4267 if (!sbi->s_blockgroup_lock) 4268 goto err_out; 4269 4270 sb->s_fs_info = sbi; 4271 sbi->s_sb = sb; 4272 return sbi; 4273 err_out: 4274 fs_put_dax(sbi->s_daxdev, NULL); 4275 kfree(sbi); 4276 return NULL; 4277 } 4278 4279 static void ext4_set_def_opts(struct super_block *sb, 4280 struct ext4_super_block *es) 4281 { 4282 unsigned long def_mount_opts; 4283 4284 /* Set defaults before we parse the mount options */ 4285 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 4286 set_opt(sb, INIT_INODE_TABLE); 4287 if (def_mount_opts & EXT4_DEFM_DEBUG) 4288 set_opt(sb, DEBUG); 4289 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 4290 set_opt(sb, GRPID); 4291 if (def_mount_opts & EXT4_DEFM_UID16) 4292 set_opt(sb, NO_UID32); 4293 /* xattr user namespace & acls are now defaulted on */ 4294 set_opt(sb, XATTR_USER); 4295 #ifdef CONFIG_EXT4_FS_POSIX_ACL 4296 set_opt(sb, POSIX_ACL); 4297 #endif 4298 if (ext4_has_feature_fast_commit(sb)) 4299 set_opt2(sb, JOURNAL_FAST_COMMIT); 4300 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 4301 if (ext4_has_metadata_csum(sb)) 4302 set_opt(sb, JOURNAL_CHECKSUM); 4303 4304 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 4305 set_opt(sb, JOURNAL_DATA); 4306 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 4307 set_opt(sb, ORDERED_DATA); 4308 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 4309 set_opt(sb, WRITEBACK_DATA); 4310 4311 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC) 4312 set_opt(sb, ERRORS_PANIC); 4313 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE) 4314 set_opt(sb, ERRORS_CONT); 4315 else 4316 set_opt(sb, ERRORS_RO); 4317 /* block_validity enabled by default; disable with noblock_validity */ 4318 set_opt(sb, BLOCK_VALIDITY); 4319 if (def_mount_opts & EXT4_DEFM_DISCARD) 4320 set_opt(sb, DISCARD); 4321 4322 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 4323 set_opt(sb, BARRIER); 4324 4325 /* 4326 * enable delayed allocation by default 4327 * Use -o nodelalloc to turn it off 4328 */ 4329 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 4330 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 4331 set_opt(sb, DELALLOC); 4332 4333 if (sb->s_blocksize == PAGE_SIZE) 4334 set_opt(sb, DIOREAD_NOLOCK); 4335 } 4336 4337 static int ext4_handle_clustersize(struct super_block *sb) 4338 { 4339 struct ext4_sb_info *sbi = EXT4_SB(sb); 4340 struct ext4_super_block *es = sbi->s_es; 4341 int clustersize; 4342 4343 /* Handle clustersize */ 4344 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4345 if (ext4_has_feature_bigalloc(sb)) { 4346 if (clustersize < sb->s_blocksize) { 4347 ext4_msg(sb, KERN_ERR, 4348 "cluster size (%d) smaller than " 4349 "block size (%lu)", clustersize, sb->s_blocksize); 4350 return -EINVAL; 4351 } 4352 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4353 le32_to_cpu(es->s_log_block_size); 4354 sbi->s_clusters_per_group = 4355 le32_to_cpu(es->s_clusters_per_group); 4356 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) { 4357 ext4_msg(sb, KERN_ERR, 4358 "#clusters per group too big: %lu", 4359 sbi->s_clusters_per_group); 4360 return -EINVAL; 4361 } 4362 if (sbi->s_blocks_per_group != 4363 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) { 4364 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4365 "clusters per group (%lu) inconsistent", 4366 sbi->s_blocks_per_group, 4367 sbi->s_clusters_per_group); 4368 return -EINVAL; 4369 } 4370 } else { 4371 if (clustersize != sb->s_blocksize) { 4372 ext4_msg(sb, KERN_ERR, 4373 "fragment/cluster size (%d) != " 4374 "block size (%lu)", clustersize, sb->s_blocksize); 4375 return -EINVAL; 4376 } 4377 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) { 4378 ext4_msg(sb, KERN_ERR, 4379 "#blocks per group too big: %lu", 4380 sbi->s_blocks_per_group); 4381 return -EINVAL; 4382 } 4383 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4384 sbi->s_cluster_bits = 0; 4385 } 4386 sbi->s_cluster_ratio = clustersize / sb->s_blocksize; 4387 4388 /* Do we have standard group size of clustersize * 8 blocks ? */ 4389 if (sbi->s_blocks_per_group == clustersize << 3) 4390 set_opt2(sb, STD_GROUP_SIZE); 4391 4392 return 0; 4393 } 4394 4395 static void ext4_fast_commit_init(struct super_block *sb) 4396 { 4397 struct ext4_sb_info *sbi = EXT4_SB(sb); 4398 4399 /* Initialize fast commit stuff */ 4400 atomic_set(&sbi->s_fc_subtid, 0); 4401 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]); 4402 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]); 4403 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]); 4404 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]); 4405 sbi->s_fc_bytes = 0; 4406 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE); 4407 sbi->s_fc_ineligible_tid = 0; 4408 spin_lock_init(&sbi->s_fc_lock); 4409 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats)); 4410 sbi->s_fc_replay_state.fc_regions = NULL; 4411 sbi->s_fc_replay_state.fc_regions_size = 0; 4412 sbi->s_fc_replay_state.fc_regions_used = 0; 4413 sbi->s_fc_replay_state.fc_regions_valid = 0; 4414 sbi->s_fc_replay_state.fc_modified_inodes = NULL; 4415 sbi->s_fc_replay_state.fc_modified_inodes_size = 0; 4416 sbi->s_fc_replay_state.fc_modified_inodes_used = 0; 4417 } 4418 4419 static int ext4_inode_info_init(struct super_block *sb, 4420 struct ext4_super_block *es) 4421 { 4422 struct ext4_sb_info *sbi = EXT4_SB(sb); 4423 4424 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4425 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4426 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4427 } else { 4428 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4429 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4430 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4431 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4432 sbi->s_first_ino); 4433 return -EINVAL; 4434 } 4435 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4436 (!is_power_of_2(sbi->s_inode_size)) || 4437 (sbi->s_inode_size > sb->s_blocksize)) { 4438 ext4_msg(sb, KERN_ERR, 4439 "unsupported inode size: %d", 4440 sbi->s_inode_size); 4441 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize); 4442 return -EINVAL; 4443 } 4444 /* 4445 * i_atime_extra is the last extra field available for 4446 * [acm]times in struct ext4_inode. Checking for that 4447 * field should suffice to ensure we have extra space 4448 * for all three. 4449 */ 4450 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4451 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4452 sb->s_time_gran = 1; 4453 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4454 } else { 4455 sb->s_time_gran = NSEC_PER_SEC; 4456 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4457 } 4458 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4459 } 4460 4461 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) { 4462 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 4463 EXT4_GOOD_OLD_INODE_SIZE; 4464 if (ext4_has_feature_extra_isize(sb)) { 4465 unsigned v, max = (sbi->s_inode_size - 4466 EXT4_GOOD_OLD_INODE_SIZE); 4467 4468 v = le16_to_cpu(es->s_want_extra_isize); 4469 if (v > max) { 4470 ext4_msg(sb, KERN_ERR, 4471 "bad s_want_extra_isize: %d", v); 4472 return -EINVAL; 4473 } 4474 if (sbi->s_want_extra_isize < v) 4475 sbi->s_want_extra_isize = v; 4476 4477 v = le16_to_cpu(es->s_min_extra_isize); 4478 if (v > max) { 4479 ext4_msg(sb, KERN_ERR, 4480 "bad s_min_extra_isize: %d", v); 4481 return -EINVAL; 4482 } 4483 if (sbi->s_want_extra_isize < v) 4484 sbi->s_want_extra_isize = v; 4485 } 4486 } 4487 4488 return 0; 4489 } 4490 4491 #if IS_ENABLED(CONFIG_UNICODE) 4492 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4493 { 4494 const struct ext4_sb_encodings *encoding_info; 4495 struct unicode_map *encoding; 4496 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags); 4497 4498 if (!ext4_has_feature_casefold(sb) || sb->s_encoding) 4499 return 0; 4500 4501 encoding_info = ext4_sb_read_encoding(es); 4502 if (!encoding_info) { 4503 ext4_msg(sb, KERN_ERR, 4504 "Encoding requested by superblock is unknown"); 4505 return -EINVAL; 4506 } 4507 4508 encoding = utf8_load(encoding_info->version); 4509 if (IS_ERR(encoding)) { 4510 ext4_msg(sb, KERN_ERR, 4511 "can't mount with superblock charset: %s-%u.%u.%u " 4512 "not supported by the kernel. flags: 0x%x.", 4513 encoding_info->name, 4514 unicode_major(encoding_info->version), 4515 unicode_minor(encoding_info->version), 4516 unicode_rev(encoding_info->version), 4517 encoding_flags); 4518 return -EINVAL; 4519 } 4520 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 4521 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name, 4522 unicode_major(encoding_info->version), 4523 unicode_minor(encoding_info->version), 4524 unicode_rev(encoding_info->version), 4525 encoding_flags); 4526 4527 sb->s_encoding = encoding; 4528 sb->s_encoding_flags = encoding_flags; 4529 4530 return 0; 4531 } 4532 #else 4533 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es) 4534 { 4535 return 0; 4536 } 4537 #endif 4538 4539 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es) 4540 { 4541 struct ext4_sb_info *sbi = EXT4_SB(sb); 4542 4543 /* Warn if metadata_csum and gdt_csum are both set. */ 4544 if (ext4_has_feature_metadata_csum(sb) && 4545 ext4_has_feature_gdt_csum(sb)) 4546 ext4_warning(sb, "metadata_csum and uninit_bg are " 4547 "redundant flags; please run fsck."); 4548 4549 /* Check for a known checksum algorithm */ 4550 if (!ext4_verify_csum_type(sb, es)) { 4551 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4552 "unknown checksum algorithm."); 4553 return -EINVAL; 4554 } 4555 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE, 4556 ext4_orphan_file_block_trigger); 4557 4558 /* Load the checksum driver */ 4559 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 4560 if (IS_ERR(sbi->s_chksum_driver)) { 4561 int ret = PTR_ERR(sbi->s_chksum_driver); 4562 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 4563 sbi->s_chksum_driver = NULL; 4564 return ret; 4565 } 4566 4567 /* Check superblock checksum */ 4568 if (!ext4_superblock_csum_verify(sb, es)) { 4569 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 4570 "invalid superblock checksum. Run e2fsck?"); 4571 return -EFSBADCRC; 4572 } 4573 4574 /* Precompute checksum seed for all metadata */ 4575 if (ext4_has_feature_csum_seed(sb)) 4576 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 4577 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 4578 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 4579 sizeof(es->s_uuid)); 4580 return 0; 4581 } 4582 4583 static int ext4_check_feature_compatibility(struct super_block *sb, 4584 struct ext4_super_block *es, 4585 int silent) 4586 { 4587 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 4588 (ext4_has_compat_features(sb) || 4589 ext4_has_ro_compat_features(sb) || 4590 ext4_has_incompat_features(sb))) 4591 ext4_msg(sb, KERN_WARNING, 4592 "feature flags set on rev 0 fs, " 4593 "running e2fsck is recommended"); 4594 4595 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 4596 set_opt2(sb, HURD_COMPAT); 4597 if (ext4_has_feature_64bit(sb)) { 4598 ext4_msg(sb, KERN_ERR, 4599 "The Hurd can't support 64-bit file systems"); 4600 return -EINVAL; 4601 } 4602 4603 /* 4604 * ea_inode feature uses l_i_version field which is not 4605 * available in HURD_COMPAT mode. 4606 */ 4607 if (ext4_has_feature_ea_inode(sb)) { 4608 ext4_msg(sb, KERN_ERR, 4609 "ea_inode feature is not supported for Hurd"); 4610 return -EINVAL; 4611 } 4612 } 4613 4614 if (IS_EXT2_SB(sb)) { 4615 if (ext2_feature_set_ok(sb)) 4616 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 4617 "using the ext4 subsystem"); 4618 else { 4619 /* 4620 * If we're probing be silent, if this looks like 4621 * it's actually an ext[34] filesystem. 4622 */ 4623 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4624 return -EINVAL; 4625 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 4626 "to feature incompatibilities"); 4627 return -EINVAL; 4628 } 4629 } 4630 4631 if (IS_EXT3_SB(sb)) { 4632 if (ext3_feature_set_ok(sb)) 4633 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 4634 "using the ext4 subsystem"); 4635 else { 4636 /* 4637 * If we're probing be silent, if this looks like 4638 * it's actually an ext4 filesystem. 4639 */ 4640 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 4641 return -EINVAL; 4642 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 4643 "to feature incompatibilities"); 4644 return -EINVAL; 4645 } 4646 } 4647 4648 /* 4649 * Check feature flags regardless of the revision level, since we 4650 * previously didn't change the revision level when setting the flags, 4651 * so there is a chance incompat flags are set on a rev 0 filesystem. 4652 */ 4653 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 4654 return -EINVAL; 4655 4656 return 0; 4657 } 4658 4659 static int ext4_geometry_check(struct super_block *sb, 4660 struct ext4_super_block *es) 4661 { 4662 struct ext4_sb_info *sbi = EXT4_SB(sb); 4663 __u64 blocks_count; 4664 4665 /* check blocks count against device size */ 4666 blocks_count = sb_bdev_nr_blocks(sb); 4667 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4668 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4669 "exceeds size of device (%llu blocks)", 4670 ext4_blocks_count(es), blocks_count); 4671 return -EINVAL; 4672 } 4673 4674 /* 4675 * It makes no sense for the first data block to be beyond the end 4676 * of the filesystem. 4677 */ 4678 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4679 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4680 "block %u is beyond end of filesystem (%llu)", 4681 le32_to_cpu(es->s_first_data_block), 4682 ext4_blocks_count(es)); 4683 return -EINVAL; 4684 } 4685 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4686 (sbi->s_cluster_ratio == 1)) { 4687 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4688 "block is 0 with a 1k block and cluster size"); 4689 return -EINVAL; 4690 } 4691 4692 blocks_count = (ext4_blocks_count(es) - 4693 le32_to_cpu(es->s_first_data_block) + 4694 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4695 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4696 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4697 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu " 4698 "(block count %llu, first data block %u, " 4699 "blocks per group %lu)", blocks_count, 4700 ext4_blocks_count(es), 4701 le32_to_cpu(es->s_first_data_block), 4702 EXT4_BLOCKS_PER_GROUP(sb)); 4703 return -EINVAL; 4704 } 4705 sbi->s_groups_count = blocks_count; 4706 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4707 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4708 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4709 le32_to_cpu(es->s_inodes_count)) { 4710 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4711 le32_to_cpu(es->s_inodes_count), 4712 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4713 return -EINVAL; 4714 } 4715 4716 return 0; 4717 } 4718 4719 static void ext4_group_desc_free(struct ext4_sb_info *sbi) 4720 { 4721 struct buffer_head **group_desc; 4722 int i; 4723 4724 rcu_read_lock(); 4725 group_desc = rcu_dereference(sbi->s_group_desc); 4726 for (i = 0; i < sbi->s_gdb_count; i++) 4727 brelse(group_desc[i]); 4728 kvfree(group_desc); 4729 rcu_read_unlock(); 4730 } 4731 4732 static int ext4_group_desc_init(struct super_block *sb, 4733 struct ext4_super_block *es, 4734 ext4_fsblk_t logical_sb_block, 4735 ext4_group_t *first_not_zeroed) 4736 { 4737 struct ext4_sb_info *sbi = EXT4_SB(sb); 4738 unsigned int db_count; 4739 ext4_fsblk_t block; 4740 int ret; 4741 int i; 4742 4743 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4744 EXT4_DESC_PER_BLOCK(sb); 4745 if (ext4_has_feature_meta_bg(sb)) { 4746 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4747 ext4_msg(sb, KERN_WARNING, 4748 "first meta block group too large: %u " 4749 "(group descriptor block count %u)", 4750 le32_to_cpu(es->s_first_meta_bg), db_count); 4751 return -EINVAL; 4752 } 4753 } 4754 rcu_assign_pointer(sbi->s_group_desc, 4755 kvmalloc_array(db_count, 4756 sizeof(struct buffer_head *), 4757 GFP_KERNEL)); 4758 if (sbi->s_group_desc == NULL) { 4759 ext4_msg(sb, KERN_ERR, "not enough memory"); 4760 return -ENOMEM; 4761 } 4762 4763 bgl_lock_init(sbi->s_blockgroup_lock); 4764 4765 /* Pre-read the descriptors into the buffer cache */ 4766 for (i = 0; i < db_count; i++) { 4767 block = descriptor_loc(sb, logical_sb_block, i); 4768 ext4_sb_breadahead_unmovable(sb, block); 4769 } 4770 4771 for (i = 0; i < db_count; i++) { 4772 struct buffer_head *bh; 4773 4774 block = descriptor_loc(sb, logical_sb_block, i); 4775 bh = ext4_sb_bread_unmovable(sb, block); 4776 if (IS_ERR(bh)) { 4777 ext4_msg(sb, KERN_ERR, 4778 "can't read group descriptor %d", i); 4779 sbi->s_gdb_count = i; 4780 ret = PTR_ERR(bh); 4781 goto out; 4782 } 4783 rcu_read_lock(); 4784 rcu_dereference(sbi->s_group_desc)[i] = bh; 4785 rcu_read_unlock(); 4786 } 4787 sbi->s_gdb_count = db_count; 4788 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) { 4789 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4790 ret = -EFSCORRUPTED; 4791 goto out; 4792 } 4793 return 0; 4794 out: 4795 ext4_group_desc_free(sbi); 4796 return ret; 4797 } 4798 4799 static int ext4_load_and_init_journal(struct super_block *sb, 4800 struct ext4_super_block *es, 4801 struct ext4_fs_context *ctx) 4802 { 4803 struct ext4_sb_info *sbi = EXT4_SB(sb); 4804 int err; 4805 4806 err = ext4_load_journal(sb, es, ctx->journal_devnum); 4807 if (err) 4808 return err; 4809 4810 if (ext4_has_feature_64bit(sb) && 4811 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4812 JBD2_FEATURE_INCOMPAT_64BIT)) { 4813 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4814 goto out; 4815 } 4816 4817 if (!set_journal_csum_feature_set(sb)) { 4818 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4819 "feature set"); 4820 goto out; 4821 } 4822 4823 if (test_opt2(sb, JOURNAL_FAST_COMMIT) && 4824 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4825 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) { 4826 ext4_msg(sb, KERN_ERR, 4827 "Failed to set fast commit journal feature"); 4828 goto out; 4829 } 4830 4831 /* We have now updated the journal if required, so we can 4832 * validate the data journaling mode. */ 4833 switch (test_opt(sb, DATA_FLAGS)) { 4834 case 0: 4835 /* No mode set, assume a default based on the journal 4836 * capabilities: ORDERED_DATA if the journal can 4837 * cope, else JOURNAL_DATA 4838 */ 4839 if (jbd2_journal_check_available_features 4840 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4841 set_opt(sb, ORDERED_DATA); 4842 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4843 } else { 4844 set_opt(sb, JOURNAL_DATA); 4845 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4846 } 4847 break; 4848 4849 case EXT4_MOUNT_ORDERED_DATA: 4850 case EXT4_MOUNT_WRITEBACK_DATA: 4851 if (!jbd2_journal_check_available_features 4852 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4853 ext4_msg(sb, KERN_ERR, "Journal does not support " 4854 "requested data journaling mode"); 4855 goto out; 4856 } 4857 break; 4858 default: 4859 break; 4860 } 4861 4862 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4863 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4864 ext4_msg(sb, KERN_ERR, "can't mount with " 4865 "journal_async_commit in data=ordered mode"); 4866 goto out; 4867 } 4868 4869 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 4870 4871 sbi->s_journal->j_submit_inode_data_buffers = 4872 ext4_journal_submit_inode_data_buffers; 4873 sbi->s_journal->j_finish_inode_data_buffers = 4874 ext4_journal_finish_inode_data_buffers; 4875 4876 return 0; 4877 4878 out: 4879 /* flush s_error_work before journal destroy. */ 4880 flush_work(&sbi->s_error_work); 4881 jbd2_journal_destroy(sbi->s_journal); 4882 sbi->s_journal = NULL; 4883 return -EINVAL; 4884 } 4885 4886 static int ext4_journal_data_mode_check(struct super_block *sb) 4887 { 4888 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 4889 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with " 4890 "data=journal disables delayed allocation, " 4891 "dioread_nolock, O_DIRECT and fast_commit support!\n"); 4892 /* can't mount with both data=journal and dioread_nolock. */ 4893 clear_opt(sb, DIOREAD_NOLOCK); 4894 clear_opt2(sb, JOURNAL_FAST_COMMIT); 4895 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 4896 ext4_msg(sb, KERN_ERR, "can't mount with " 4897 "both data=journal and delalloc"); 4898 return -EINVAL; 4899 } 4900 if (test_opt(sb, DAX_ALWAYS)) { 4901 ext4_msg(sb, KERN_ERR, "can't mount with " 4902 "both data=journal and dax"); 4903 return -EINVAL; 4904 } 4905 if (ext4_has_feature_encrypt(sb)) { 4906 ext4_msg(sb, KERN_WARNING, 4907 "encrypted files will use data=ordered " 4908 "instead of data journaling mode"); 4909 } 4910 if (test_opt(sb, DELALLOC)) 4911 clear_opt(sb, DELALLOC); 4912 } else { 4913 sb->s_iflags |= SB_I_CGROUPWB; 4914 } 4915 4916 return 0; 4917 } 4918 4919 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb, 4920 int silent) 4921 { 4922 struct ext4_sb_info *sbi = EXT4_SB(sb); 4923 struct ext4_super_block *es; 4924 ext4_fsblk_t logical_sb_block; 4925 unsigned long offset = 0; 4926 struct buffer_head *bh; 4927 int ret = -EINVAL; 4928 int blocksize; 4929 4930 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 4931 if (!blocksize) { 4932 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 4933 return -EINVAL; 4934 } 4935 4936 /* 4937 * The ext4 superblock will not be buffer aligned for other than 1kB 4938 * block sizes. We need to calculate the offset from buffer start. 4939 */ 4940 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 4941 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 4942 offset = do_div(logical_sb_block, blocksize); 4943 } else { 4944 logical_sb_block = sbi->s_sb_block; 4945 } 4946 4947 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 4948 if (IS_ERR(bh)) { 4949 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 4950 return PTR_ERR(bh); 4951 } 4952 /* 4953 * Note: s_es must be initialized as soon as possible because 4954 * some ext4 macro-instructions depend on its value 4955 */ 4956 es = (struct ext4_super_block *) (bh->b_data + offset); 4957 sbi->s_es = es; 4958 sb->s_magic = le16_to_cpu(es->s_magic); 4959 if (sb->s_magic != EXT4_SUPER_MAGIC) { 4960 if (!silent) 4961 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 4962 goto out; 4963 } 4964 4965 if (le32_to_cpu(es->s_log_block_size) > 4966 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4967 ext4_msg(sb, KERN_ERR, 4968 "Invalid log block size: %u", 4969 le32_to_cpu(es->s_log_block_size)); 4970 goto out; 4971 } 4972 if (le32_to_cpu(es->s_log_cluster_size) > 4973 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 4974 ext4_msg(sb, KERN_ERR, 4975 "Invalid log cluster size: %u", 4976 le32_to_cpu(es->s_log_cluster_size)); 4977 goto out; 4978 } 4979 4980 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 4981 4982 /* 4983 * If the default block size is not the same as the real block size, 4984 * we need to reload it. 4985 */ 4986 if (sb->s_blocksize == blocksize) { 4987 *lsb = logical_sb_block; 4988 sbi->s_sbh = bh; 4989 return 0; 4990 } 4991 4992 /* 4993 * bh must be released before kill_bdev(), otherwise 4994 * it won't be freed and its page also. kill_bdev() 4995 * is called by sb_set_blocksize(). 4996 */ 4997 brelse(bh); 4998 /* Validate the filesystem blocksize */ 4999 if (!sb_set_blocksize(sb, blocksize)) { 5000 ext4_msg(sb, KERN_ERR, "bad block size %d", 5001 blocksize); 5002 bh = NULL; 5003 goto out; 5004 } 5005 5006 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE; 5007 offset = do_div(logical_sb_block, blocksize); 5008 bh = ext4_sb_bread_unmovable(sb, logical_sb_block); 5009 if (IS_ERR(bh)) { 5010 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try"); 5011 ret = PTR_ERR(bh); 5012 bh = NULL; 5013 goto out; 5014 } 5015 es = (struct ext4_super_block *)(bh->b_data + offset); 5016 sbi->s_es = es; 5017 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 5018 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!"); 5019 goto out; 5020 } 5021 *lsb = logical_sb_block; 5022 sbi->s_sbh = bh; 5023 return 0; 5024 out: 5025 brelse(bh); 5026 return ret; 5027 } 5028 5029 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb) 5030 { 5031 struct ext4_super_block *es = NULL; 5032 struct ext4_sb_info *sbi = EXT4_SB(sb); 5033 struct flex_groups **flex_groups; 5034 ext4_fsblk_t block; 5035 ext4_fsblk_t logical_sb_block; 5036 struct inode *root; 5037 int ret = -ENOMEM; 5038 unsigned int i; 5039 int needs_recovery, has_huge_files; 5040 int err = 0; 5041 ext4_group_t first_not_zeroed; 5042 struct ext4_fs_context *ctx = fc->fs_private; 5043 int silent = fc->sb_flags & SB_SILENT; 5044 5045 /* Set defaults for the variables that will be set during parsing */ 5046 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) 5047 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5048 5049 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 5050 sbi->s_sectors_written_start = 5051 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]); 5052 5053 /* -EINVAL is default */ 5054 ret = -EINVAL; 5055 err = ext4_load_super(sb, &logical_sb_block, silent); 5056 if (err) 5057 goto out_fail; 5058 5059 es = sbi->s_es; 5060 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 5061 5062 err = ext4_init_metadata_csum(sb, es); 5063 if (err) 5064 goto failed_mount; 5065 5066 ext4_set_def_opts(sb, es); 5067 5068 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 5069 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 5070 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 5071 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 5072 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 5073 5074 /* 5075 * set default s_li_wait_mult for lazyinit, for the case there is 5076 * no mount option specified. 5077 */ 5078 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 5079 5080 if (ext4_inode_info_init(sb, es)) 5081 goto failed_mount; 5082 5083 err = parse_apply_sb_mount_options(sb, ctx); 5084 if (err < 0) 5085 goto failed_mount; 5086 5087 sbi->s_def_mount_opt = sbi->s_mount_opt; 5088 5089 err = ext4_check_opt_consistency(fc, sb); 5090 if (err < 0) 5091 goto failed_mount; 5092 5093 ext4_apply_options(fc, sb); 5094 5095 if (ext4_encoding_init(sb, es)) 5096 goto failed_mount; 5097 5098 if (ext4_journal_data_mode_check(sb)) 5099 goto failed_mount; 5100 5101 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5102 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5103 5104 /* i_version is always enabled now */ 5105 sb->s_flags |= SB_I_VERSION; 5106 5107 if (ext4_check_feature_compatibility(sb, es, silent)) 5108 goto failed_mount; 5109 5110 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) { 5111 ext4_msg(sb, KERN_ERR, 5112 "Number of reserved GDT blocks insanely large: %d", 5113 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 5114 goto failed_mount; 5115 } 5116 5117 if (sbi->s_daxdev) { 5118 if (sb->s_blocksize == PAGE_SIZE) 5119 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags); 5120 else 5121 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n"); 5122 } 5123 5124 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) { 5125 if (ext4_has_feature_inline_data(sb)) { 5126 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 5127 " that may contain inline data"); 5128 goto failed_mount; 5129 } 5130 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) { 5131 ext4_msg(sb, KERN_ERR, 5132 "DAX unsupported by block device."); 5133 goto failed_mount; 5134 } 5135 } 5136 5137 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 5138 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 5139 es->s_encryption_level); 5140 goto failed_mount; 5141 } 5142 5143 has_huge_files = ext4_has_feature_huge_file(sb); 5144 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 5145 has_huge_files); 5146 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 5147 5148 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 5149 if (ext4_has_feature_64bit(sb)) { 5150 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 5151 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 5152 !is_power_of_2(sbi->s_desc_size)) { 5153 ext4_msg(sb, KERN_ERR, 5154 "unsupported descriptor size %lu", 5155 sbi->s_desc_size); 5156 goto failed_mount; 5157 } 5158 } else 5159 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 5160 5161 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 5162 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 5163 5164 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb); 5165 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) { 5166 if (!silent) 5167 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 5168 goto failed_mount; 5169 } 5170 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 5171 sbi->s_inodes_per_group > sb->s_blocksize * 8) { 5172 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 5173 sbi->s_inodes_per_group); 5174 goto failed_mount; 5175 } 5176 sbi->s_itb_per_group = sbi->s_inodes_per_group / 5177 sbi->s_inodes_per_block; 5178 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb); 5179 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY; 5180 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 5181 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 5182 5183 for (i = 0; i < 4; i++) 5184 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 5185 sbi->s_def_hash_version = es->s_def_hash_version; 5186 if (ext4_has_feature_dir_index(sb)) { 5187 i = le32_to_cpu(es->s_flags); 5188 if (i & EXT2_FLAGS_UNSIGNED_HASH) 5189 sbi->s_hash_unsigned = 3; 5190 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 5191 #ifdef __CHAR_UNSIGNED__ 5192 if (!sb_rdonly(sb)) 5193 es->s_flags |= 5194 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 5195 sbi->s_hash_unsigned = 3; 5196 #else 5197 if (!sb_rdonly(sb)) 5198 es->s_flags |= 5199 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 5200 #endif 5201 } 5202 } 5203 5204 if (ext4_handle_clustersize(sb)) 5205 goto failed_mount; 5206 5207 /* 5208 * Test whether we have more sectors than will fit in sector_t, 5209 * and whether the max offset is addressable by the page cache. 5210 */ 5211 err = generic_check_addressable(sb->s_blocksize_bits, 5212 ext4_blocks_count(es)); 5213 if (err) { 5214 ext4_msg(sb, KERN_ERR, "filesystem" 5215 " too large to mount safely on this system"); 5216 goto failed_mount; 5217 } 5218 5219 if (ext4_geometry_check(sb, es)) 5220 goto failed_mount; 5221 5222 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed); 5223 if (err) 5224 goto failed_mount; 5225 5226 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 5227 spin_lock_init(&sbi->s_error_lock); 5228 INIT_WORK(&sbi->s_error_work, flush_stashed_error_work); 5229 5230 /* Register extent status tree shrinker */ 5231 if (ext4_es_register_shrinker(sbi)) 5232 goto failed_mount3; 5233 5234 sbi->s_stripe = ext4_get_stripe_size(sbi); 5235 sbi->s_extent_max_zeroout_kb = 32; 5236 5237 /* 5238 * set up enough so that it can read an inode 5239 */ 5240 sb->s_op = &ext4_sops; 5241 sb->s_export_op = &ext4_export_ops; 5242 sb->s_xattr = ext4_xattr_handlers; 5243 #ifdef CONFIG_FS_ENCRYPTION 5244 sb->s_cop = &ext4_cryptops; 5245 #endif 5246 #ifdef CONFIG_FS_VERITY 5247 sb->s_vop = &ext4_verityops; 5248 #endif 5249 #ifdef CONFIG_QUOTA 5250 sb->dq_op = &ext4_quota_operations; 5251 if (ext4_has_feature_quota(sb)) 5252 sb->s_qcop = &dquot_quotactl_sysfile_ops; 5253 else 5254 sb->s_qcop = &ext4_qctl_operations; 5255 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 5256 #endif 5257 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 5258 5259 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 5260 mutex_init(&sbi->s_orphan_lock); 5261 5262 ext4_fast_commit_init(sb); 5263 5264 sb->s_root = NULL; 5265 5266 needs_recovery = (es->s_last_orphan != 0 || 5267 ext4_has_feature_orphan_present(sb) || 5268 ext4_has_feature_journal_needs_recovery(sb)); 5269 5270 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 5271 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 5272 goto failed_mount3a; 5273 5274 /* 5275 * The first inode we look at is the journal inode. Don't try 5276 * root first: it may be modified in the journal! 5277 */ 5278 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 5279 err = ext4_load_and_init_journal(sb, es, ctx); 5280 if (err) 5281 goto failed_mount3a; 5282 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 5283 ext4_has_feature_journal_needs_recovery(sb)) { 5284 ext4_msg(sb, KERN_ERR, "required journal recovery " 5285 "suppressed and not mounted read-only"); 5286 goto failed_mount3a; 5287 } else { 5288 /* Nojournal mode, all journal mount options are illegal */ 5289 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 5290 ext4_msg(sb, KERN_ERR, "can't mount with " 5291 "journal_checksum, fs mounted w/o journal"); 5292 goto failed_mount3a; 5293 } 5294 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5295 ext4_msg(sb, KERN_ERR, "can't mount with " 5296 "journal_async_commit, fs mounted w/o journal"); 5297 goto failed_mount3a; 5298 } 5299 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 5300 ext4_msg(sb, KERN_ERR, "can't mount with " 5301 "commit=%lu, fs mounted w/o journal", 5302 sbi->s_commit_interval / HZ); 5303 goto failed_mount3a; 5304 } 5305 if (EXT4_MOUNT_DATA_FLAGS & 5306 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 5307 ext4_msg(sb, KERN_ERR, "can't mount with " 5308 "data=, fs mounted w/o journal"); 5309 goto failed_mount3a; 5310 } 5311 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 5312 clear_opt(sb, JOURNAL_CHECKSUM); 5313 clear_opt(sb, DATA_FLAGS); 5314 clear_opt2(sb, JOURNAL_FAST_COMMIT); 5315 sbi->s_journal = NULL; 5316 needs_recovery = 0; 5317 } 5318 5319 if (!test_opt(sb, NO_MBCACHE)) { 5320 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 5321 if (!sbi->s_ea_block_cache) { 5322 ext4_msg(sb, KERN_ERR, 5323 "Failed to create ea_block_cache"); 5324 goto failed_mount_wq; 5325 } 5326 5327 if (ext4_has_feature_ea_inode(sb)) { 5328 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 5329 if (!sbi->s_ea_inode_cache) { 5330 ext4_msg(sb, KERN_ERR, 5331 "Failed to create ea_inode_cache"); 5332 goto failed_mount_wq; 5333 } 5334 } 5335 } 5336 5337 if (ext4_has_feature_verity(sb) && sb->s_blocksize != PAGE_SIZE) { 5338 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 5339 goto failed_mount_wq; 5340 } 5341 5342 /* 5343 * Get the # of file system overhead blocks from the 5344 * superblock if present. 5345 */ 5346 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 5347 /* ignore the precalculated value if it is ridiculous */ 5348 if (sbi->s_overhead > ext4_blocks_count(es)) 5349 sbi->s_overhead = 0; 5350 /* 5351 * If the bigalloc feature is not enabled recalculating the 5352 * overhead doesn't take long, so we might as well just redo 5353 * it to make sure we are using the correct value. 5354 */ 5355 if (!ext4_has_feature_bigalloc(sb)) 5356 sbi->s_overhead = 0; 5357 if (sbi->s_overhead == 0) { 5358 err = ext4_calculate_overhead(sb); 5359 if (err) 5360 goto failed_mount_wq; 5361 } 5362 5363 /* 5364 * The maximum number of concurrent works can be high and 5365 * concurrency isn't really necessary. Limit it to 1. 5366 */ 5367 EXT4_SB(sb)->rsv_conversion_wq = 5368 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 5369 if (!EXT4_SB(sb)->rsv_conversion_wq) { 5370 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 5371 ret = -ENOMEM; 5372 goto failed_mount4; 5373 } 5374 5375 /* 5376 * The jbd2_journal_load will have done any necessary log recovery, 5377 * so we can safely mount the rest of the filesystem now. 5378 */ 5379 5380 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 5381 if (IS_ERR(root)) { 5382 ext4_msg(sb, KERN_ERR, "get root inode failed"); 5383 ret = PTR_ERR(root); 5384 root = NULL; 5385 goto failed_mount4; 5386 } 5387 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 5388 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 5389 iput(root); 5390 goto failed_mount4; 5391 } 5392 5393 sb->s_root = d_make_root(root); 5394 if (!sb->s_root) { 5395 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 5396 ret = -ENOMEM; 5397 goto failed_mount4; 5398 } 5399 5400 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 5401 if (ret == -EROFS) { 5402 sb->s_flags |= SB_RDONLY; 5403 ret = 0; 5404 } else if (ret) 5405 goto failed_mount4a; 5406 5407 ext4_set_resv_clusters(sb); 5408 5409 if (test_opt(sb, BLOCK_VALIDITY)) { 5410 err = ext4_setup_system_zone(sb); 5411 if (err) { 5412 ext4_msg(sb, KERN_ERR, "failed to initialize system " 5413 "zone (%d)", err); 5414 goto failed_mount4a; 5415 } 5416 } 5417 ext4_fc_replay_cleanup(sb); 5418 5419 ext4_ext_init(sb); 5420 5421 /* 5422 * Enable optimize_scan if number of groups is > threshold. This can be 5423 * turned off by passing "mb_optimize_scan=0". This can also be 5424 * turned on forcefully by passing "mb_optimize_scan=1". 5425 */ 5426 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) { 5427 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD) 5428 set_opt2(sb, MB_OPTIMIZE_SCAN); 5429 else 5430 clear_opt2(sb, MB_OPTIMIZE_SCAN); 5431 } 5432 5433 err = ext4_mb_init(sb); 5434 if (err) { 5435 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 5436 err); 5437 goto failed_mount5; 5438 } 5439 5440 /* 5441 * We can only set up the journal commit callback once 5442 * mballoc is initialized 5443 */ 5444 if (sbi->s_journal) 5445 sbi->s_journal->j_commit_callback = 5446 ext4_journal_commit_callback; 5447 5448 block = ext4_count_free_clusters(sb); 5449 ext4_free_blocks_count_set(sbi->s_es, 5450 EXT4_C2B(sbi, block)); 5451 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 5452 GFP_KERNEL); 5453 if (!err) { 5454 unsigned long freei = ext4_count_free_inodes(sb); 5455 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 5456 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 5457 GFP_KERNEL); 5458 } 5459 if (!err) 5460 err = percpu_counter_init(&sbi->s_dirs_counter, 5461 ext4_count_dirs(sb), GFP_KERNEL); 5462 if (!err) 5463 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 5464 GFP_KERNEL); 5465 if (!err) 5466 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0, 5467 GFP_KERNEL); 5468 if (!err) 5469 err = percpu_init_rwsem(&sbi->s_writepages_rwsem); 5470 5471 if (err) { 5472 ext4_msg(sb, KERN_ERR, "insufficient memory"); 5473 goto failed_mount6; 5474 } 5475 5476 if (ext4_has_feature_flex_bg(sb)) 5477 if (!ext4_fill_flex_info(sb)) { 5478 ext4_msg(sb, KERN_ERR, 5479 "unable to initialize " 5480 "flex_bg meta info!"); 5481 ret = -ENOMEM; 5482 goto failed_mount6; 5483 } 5484 5485 err = ext4_register_li_request(sb, first_not_zeroed); 5486 if (err) 5487 goto failed_mount6; 5488 5489 err = ext4_register_sysfs(sb); 5490 if (err) 5491 goto failed_mount7; 5492 5493 err = ext4_init_orphan_info(sb); 5494 if (err) 5495 goto failed_mount8; 5496 #ifdef CONFIG_QUOTA 5497 /* Enable quota usage during mount. */ 5498 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 5499 err = ext4_enable_quotas(sb); 5500 if (err) 5501 goto failed_mount9; 5502 } 5503 #endif /* CONFIG_QUOTA */ 5504 5505 /* 5506 * Save the original bdev mapping's wb_err value which could be 5507 * used to detect the metadata async write error. 5508 */ 5509 spin_lock_init(&sbi->s_bdev_wb_lock); 5510 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err, 5511 &sbi->s_bdev_wb_err); 5512 sb->s_bdev->bd_super = sb; 5513 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 5514 ext4_orphan_cleanup(sb, es); 5515 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 5516 /* 5517 * Update the checksum after updating free space/inode counters and 5518 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect 5519 * checksum in the buffer cache until it is written out and 5520 * e2fsprogs programs trying to open a file system immediately 5521 * after it is mounted can fail. 5522 */ 5523 ext4_superblock_csum_set(sb); 5524 if (needs_recovery) { 5525 ext4_msg(sb, KERN_INFO, "recovery complete"); 5526 err = ext4_mark_recovery_complete(sb, es); 5527 if (err) 5528 goto failed_mount9; 5529 } 5530 5531 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev)) 5532 ext4_msg(sb, KERN_WARNING, 5533 "mounting with \"discard\" option, but the device does not support discard"); 5534 5535 if (es->s_error_count) 5536 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 5537 5538 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 5539 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 5540 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 5541 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 5542 atomic_set(&sbi->s_warning_count, 0); 5543 atomic_set(&sbi->s_msg_count, 0); 5544 5545 return 0; 5546 5547 failed_mount9: 5548 ext4_release_orphan_info(sb); 5549 failed_mount8: 5550 ext4_unregister_sysfs(sb); 5551 kobject_put(&sbi->s_kobj); 5552 failed_mount7: 5553 ext4_unregister_li_request(sb); 5554 failed_mount6: 5555 ext4_mb_release(sb); 5556 rcu_read_lock(); 5557 flex_groups = rcu_dereference(sbi->s_flex_groups); 5558 if (flex_groups) { 5559 for (i = 0; i < sbi->s_flex_groups_allocated; i++) 5560 kvfree(flex_groups[i]); 5561 kvfree(flex_groups); 5562 } 5563 rcu_read_unlock(); 5564 percpu_counter_destroy(&sbi->s_freeclusters_counter); 5565 percpu_counter_destroy(&sbi->s_freeinodes_counter); 5566 percpu_counter_destroy(&sbi->s_dirs_counter); 5567 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 5568 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit); 5569 percpu_free_rwsem(&sbi->s_writepages_rwsem); 5570 failed_mount5: 5571 ext4_ext_release(sb); 5572 ext4_release_system_zone(sb); 5573 failed_mount4a: 5574 dput(sb->s_root); 5575 sb->s_root = NULL; 5576 failed_mount4: 5577 ext4_msg(sb, KERN_ERR, "mount failed"); 5578 if (EXT4_SB(sb)->rsv_conversion_wq) 5579 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 5580 failed_mount_wq: 5581 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 5582 sbi->s_ea_inode_cache = NULL; 5583 5584 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 5585 sbi->s_ea_block_cache = NULL; 5586 5587 if (sbi->s_journal) { 5588 /* flush s_error_work before journal destroy. */ 5589 flush_work(&sbi->s_error_work); 5590 jbd2_journal_destroy(sbi->s_journal); 5591 sbi->s_journal = NULL; 5592 } 5593 failed_mount3a: 5594 ext4_es_unregister_shrinker(sbi); 5595 failed_mount3: 5596 /* flush s_error_work before sbi destroy */ 5597 flush_work(&sbi->s_error_work); 5598 del_timer_sync(&sbi->s_err_report); 5599 ext4_stop_mmpd(sbi); 5600 ext4_group_desc_free(sbi); 5601 failed_mount: 5602 if (sbi->s_chksum_driver) 5603 crypto_free_shash(sbi->s_chksum_driver); 5604 5605 #if IS_ENABLED(CONFIG_UNICODE) 5606 utf8_unload(sb->s_encoding); 5607 #endif 5608 5609 #ifdef CONFIG_QUOTA 5610 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5611 kfree(get_qf_name(sb, sbi, i)); 5612 #endif 5613 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy); 5614 /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */ 5615 brelse(sbi->s_sbh); 5616 ext4_blkdev_remove(sbi); 5617 out_fail: 5618 sb->s_fs_info = NULL; 5619 return err ? err : ret; 5620 } 5621 5622 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc) 5623 { 5624 struct ext4_fs_context *ctx = fc->fs_private; 5625 struct ext4_sb_info *sbi; 5626 const char *descr; 5627 int ret; 5628 5629 sbi = ext4_alloc_sbi(sb); 5630 if (!sbi) 5631 return -ENOMEM; 5632 5633 fc->s_fs_info = sbi; 5634 5635 /* Cleanup superblock name */ 5636 strreplace(sb->s_id, '/', '!'); 5637 5638 sbi->s_sb_block = 1; /* Default super block location */ 5639 if (ctx->spec & EXT4_SPEC_s_sb_block) 5640 sbi->s_sb_block = ctx->s_sb_block; 5641 5642 ret = __ext4_fill_super(fc, sb); 5643 if (ret < 0) 5644 goto free_sbi; 5645 5646 if (sbi->s_journal) { 5647 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 5648 descr = " journalled data mode"; 5649 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 5650 descr = " ordered data mode"; 5651 else 5652 descr = " writeback data mode"; 5653 } else 5654 descr = "out journal"; 5655 5656 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 5657 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 5658 "Quota mode: %s.", descr, ext4_quota_mode(sb)); 5659 5660 /* Update the s_overhead_clusters if necessary */ 5661 ext4_update_overhead(sb, false); 5662 return 0; 5663 5664 free_sbi: 5665 ext4_free_sbi(sbi); 5666 fc->s_fs_info = NULL; 5667 return ret; 5668 } 5669 5670 static int ext4_get_tree(struct fs_context *fc) 5671 { 5672 return get_tree_bdev(fc, ext4_fill_super); 5673 } 5674 5675 /* 5676 * Setup any per-fs journal parameters now. We'll do this both on 5677 * initial mount, once the journal has been initialised but before we've 5678 * done any recovery; and again on any subsequent remount. 5679 */ 5680 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 5681 { 5682 struct ext4_sb_info *sbi = EXT4_SB(sb); 5683 5684 journal->j_commit_interval = sbi->s_commit_interval; 5685 journal->j_min_batch_time = sbi->s_min_batch_time; 5686 journal->j_max_batch_time = sbi->s_max_batch_time; 5687 ext4_fc_init(sb, journal); 5688 5689 write_lock(&journal->j_state_lock); 5690 if (test_opt(sb, BARRIER)) 5691 journal->j_flags |= JBD2_BARRIER; 5692 else 5693 journal->j_flags &= ~JBD2_BARRIER; 5694 if (test_opt(sb, DATA_ERR_ABORT)) 5695 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 5696 else 5697 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 5698 write_unlock(&journal->j_state_lock); 5699 } 5700 5701 static struct inode *ext4_get_journal_inode(struct super_block *sb, 5702 unsigned int journal_inum) 5703 { 5704 struct inode *journal_inode; 5705 5706 /* 5707 * Test for the existence of a valid inode on disk. Bad things 5708 * happen if we iget() an unused inode, as the subsequent iput() 5709 * will try to delete it. 5710 */ 5711 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 5712 if (IS_ERR(journal_inode)) { 5713 ext4_msg(sb, KERN_ERR, "no journal found"); 5714 return NULL; 5715 } 5716 if (!journal_inode->i_nlink) { 5717 make_bad_inode(journal_inode); 5718 iput(journal_inode); 5719 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 5720 return NULL; 5721 } 5722 5723 ext4_debug("Journal inode found at %p: %lld bytes\n", 5724 journal_inode, journal_inode->i_size); 5725 if (!S_ISREG(journal_inode->i_mode)) { 5726 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 5727 iput(journal_inode); 5728 return NULL; 5729 } 5730 return journal_inode; 5731 } 5732 5733 static journal_t *ext4_get_journal(struct super_block *sb, 5734 unsigned int journal_inum) 5735 { 5736 struct inode *journal_inode; 5737 journal_t *journal; 5738 5739 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5740 return NULL; 5741 5742 journal_inode = ext4_get_journal_inode(sb, journal_inum); 5743 if (!journal_inode) 5744 return NULL; 5745 5746 journal = jbd2_journal_init_inode(journal_inode); 5747 if (!journal) { 5748 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 5749 iput(journal_inode); 5750 return NULL; 5751 } 5752 journal->j_private = sb; 5753 ext4_init_journal_params(sb, journal); 5754 return journal; 5755 } 5756 5757 static journal_t *ext4_get_dev_journal(struct super_block *sb, 5758 dev_t j_dev) 5759 { 5760 struct buffer_head *bh; 5761 journal_t *journal; 5762 ext4_fsblk_t start; 5763 ext4_fsblk_t len; 5764 int hblock, blocksize; 5765 ext4_fsblk_t sb_block; 5766 unsigned long offset; 5767 struct ext4_super_block *es; 5768 struct block_device *bdev; 5769 5770 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5771 return NULL; 5772 5773 bdev = ext4_blkdev_get(j_dev, sb); 5774 if (bdev == NULL) 5775 return NULL; 5776 5777 blocksize = sb->s_blocksize; 5778 hblock = bdev_logical_block_size(bdev); 5779 if (blocksize < hblock) { 5780 ext4_msg(sb, KERN_ERR, 5781 "blocksize too small for journal device"); 5782 goto out_bdev; 5783 } 5784 5785 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 5786 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 5787 set_blocksize(bdev, blocksize); 5788 if (!(bh = __bread(bdev, sb_block, blocksize))) { 5789 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 5790 "external journal"); 5791 goto out_bdev; 5792 } 5793 5794 es = (struct ext4_super_block *) (bh->b_data + offset); 5795 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 5796 !(le32_to_cpu(es->s_feature_incompat) & 5797 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 5798 ext4_msg(sb, KERN_ERR, "external journal has " 5799 "bad superblock"); 5800 brelse(bh); 5801 goto out_bdev; 5802 } 5803 5804 if ((le32_to_cpu(es->s_feature_ro_compat) & 5805 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 5806 es->s_checksum != ext4_superblock_csum(sb, es)) { 5807 ext4_msg(sb, KERN_ERR, "external journal has " 5808 "corrupt superblock"); 5809 brelse(bh); 5810 goto out_bdev; 5811 } 5812 5813 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 5814 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 5815 brelse(bh); 5816 goto out_bdev; 5817 } 5818 5819 len = ext4_blocks_count(es); 5820 start = sb_block + 1; 5821 brelse(bh); /* we're done with the superblock */ 5822 5823 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 5824 start, len, blocksize); 5825 if (!journal) { 5826 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 5827 goto out_bdev; 5828 } 5829 journal->j_private = sb; 5830 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) { 5831 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 5832 goto out_journal; 5833 } 5834 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 5835 ext4_msg(sb, KERN_ERR, "External journal has more than one " 5836 "user (unsupported) - %d", 5837 be32_to_cpu(journal->j_superblock->s_nr_users)); 5838 goto out_journal; 5839 } 5840 EXT4_SB(sb)->s_journal_bdev = bdev; 5841 ext4_init_journal_params(sb, journal); 5842 return journal; 5843 5844 out_journal: 5845 jbd2_journal_destroy(journal); 5846 out_bdev: 5847 ext4_blkdev_put(bdev); 5848 return NULL; 5849 } 5850 5851 static int ext4_load_journal(struct super_block *sb, 5852 struct ext4_super_block *es, 5853 unsigned long journal_devnum) 5854 { 5855 journal_t *journal; 5856 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 5857 dev_t journal_dev; 5858 int err = 0; 5859 int really_read_only; 5860 int journal_dev_ro; 5861 5862 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb))) 5863 return -EFSCORRUPTED; 5864 5865 if (journal_devnum && 5866 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5867 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 5868 "numbers have changed"); 5869 journal_dev = new_decode_dev(journal_devnum); 5870 } else 5871 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 5872 5873 if (journal_inum && journal_dev) { 5874 ext4_msg(sb, KERN_ERR, 5875 "filesystem has both journal inode and journal device!"); 5876 return -EINVAL; 5877 } 5878 5879 if (journal_inum) { 5880 journal = ext4_get_journal(sb, journal_inum); 5881 if (!journal) 5882 return -EINVAL; 5883 } else { 5884 journal = ext4_get_dev_journal(sb, journal_dev); 5885 if (!journal) 5886 return -EINVAL; 5887 } 5888 5889 journal_dev_ro = bdev_read_only(journal->j_dev); 5890 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro; 5891 5892 if (journal_dev_ro && !sb_rdonly(sb)) { 5893 ext4_msg(sb, KERN_ERR, 5894 "journal device read-only, try mounting with '-o ro'"); 5895 err = -EROFS; 5896 goto err_out; 5897 } 5898 5899 /* 5900 * Are we loading a blank journal or performing recovery after a 5901 * crash? For recovery, we need to check in advance whether we 5902 * can get read-write access to the device. 5903 */ 5904 if (ext4_has_feature_journal_needs_recovery(sb)) { 5905 if (sb_rdonly(sb)) { 5906 ext4_msg(sb, KERN_INFO, "INFO: recovery " 5907 "required on readonly filesystem"); 5908 if (really_read_only) { 5909 ext4_msg(sb, KERN_ERR, "write access " 5910 "unavailable, cannot proceed " 5911 "(try mounting with noload)"); 5912 err = -EROFS; 5913 goto err_out; 5914 } 5915 ext4_msg(sb, KERN_INFO, "write access will " 5916 "be enabled during recovery"); 5917 } 5918 } 5919 5920 if (!(journal->j_flags & JBD2_BARRIER)) 5921 ext4_msg(sb, KERN_INFO, "barriers disabled"); 5922 5923 if (!ext4_has_feature_journal_needs_recovery(sb)) 5924 err = jbd2_journal_wipe(journal, !really_read_only); 5925 if (!err) { 5926 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 5927 if (save) 5928 memcpy(save, ((char *) es) + 5929 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 5930 err = jbd2_journal_load(journal); 5931 if (save) 5932 memcpy(((char *) es) + EXT4_S_ERR_START, 5933 save, EXT4_S_ERR_LEN); 5934 kfree(save); 5935 } 5936 5937 if (err) { 5938 ext4_msg(sb, KERN_ERR, "error loading journal"); 5939 goto err_out; 5940 } 5941 5942 EXT4_SB(sb)->s_journal = journal; 5943 err = ext4_clear_journal_err(sb, es); 5944 if (err) { 5945 EXT4_SB(sb)->s_journal = NULL; 5946 jbd2_journal_destroy(journal); 5947 return err; 5948 } 5949 5950 if (!really_read_only && journal_devnum && 5951 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 5952 es->s_journal_dev = cpu_to_le32(journal_devnum); 5953 5954 /* Make sure we flush the recovery flag to disk. */ 5955 ext4_commit_super(sb); 5956 } 5957 5958 return 0; 5959 5960 err_out: 5961 jbd2_journal_destroy(journal); 5962 return err; 5963 } 5964 5965 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */ 5966 static void ext4_update_super(struct super_block *sb) 5967 { 5968 struct ext4_sb_info *sbi = EXT4_SB(sb); 5969 struct ext4_super_block *es = sbi->s_es; 5970 struct buffer_head *sbh = sbi->s_sbh; 5971 5972 lock_buffer(sbh); 5973 /* 5974 * If the file system is mounted read-only, don't update the 5975 * superblock write time. This avoids updating the superblock 5976 * write time when we are mounting the root file system 5977 * read/only but we need to replay the journal; at that point, 5978 * for people who are east of GMT and who make their clock 5979 * tick in localtime for Windows bug-for-bug compatibility, 5980 * the clock is set in the future, and this will cause e2fsck 5981 * to complain and force a full file system check. 5982 */ 5983 if (!(sb->s_flags & SB_RDONLY)) 5984 ext4_update_tstamp(es, s_wtime); 5985 es->s_kbytes_written = 5986 cpu_to_le64(sbi->s_kbytes_written + 5987 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) - 5988 sbi->s_sectors_written_start) >> 1)); 5989 if (percpu_counter_initialized(&sbi->s_freeclusters_counter)) 5990 ext4_free_blocks_count_set(es, 5991 EXT4_C2B(sbi, percpu_counter_sum_positive( 5992 &sbi->s_freeclusters_counter))); 5993 if (percpu_counter_initialized(&sbi->s_freeinodes_counter)) 5994 es->s_free_inodes_count = 5995 cpu_to_le32(percpu_counter_sum_positive( 5996 &sbi->s_freeinodes_counter)); 5997 /* Copy error information to the on-disk superblock */ 5998 spin_lock(&sbi->s_error_lock); 5999 if (sbi->s_add_error_count > 0) { 6000 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6001 if (!es->s_first_error_time && !es->s_first_error_time_hi) { 6002 __ext4_update_tstamp(&es->s_first_error_time, 6003 &es->s_first_error_time_hi, 6004 sbi->s_first_error_time); 6005 strncpy(es->s_first_error_func, sbi->s_first_error_func, 6006 sizeof(es->s_first_error_func)); 6007 es->s_first_error_line = 6008 cpu_to_le32(sbi->s_first_error_line); 6009 es->s_first_error_ino = 6010 cpu_to_le32(sbi->s_first_error_ino); 6011 es->s_first_error_block = 6012 cpu_to_le64(sbi->s_first_error_block); 6013 es->s_first_error_errcode = 6014 ext4_errno_to_code(sbi->s_first_error_code); 6015 } 6016 __ext4_update_tstamp(&es->s_last_error_time, 6017 &es->s_last_error_time_hi, 6018 sbi->s_last_error_time); 6019 strncpy(es->s_last_error_func, sbi->s_last_error_func, 6020 sizeof(es->s_last_error_func)); 6021 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line); 6022 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino); 6023 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block); 6024 es->s_last_error_errcode = 6025 ext4_errno_to_code(sbi->s_last_error_code); 6026 /* 6027 * Start the daily error reporting function if it hasn't been 6028 * started already 6029 */ 6030 if (!es->s_error_count) 6031 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); 6032 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count); 6033 sbi->s_add_error_count = 0; 6034 } 6035 spin_unlock(&sbi->s_error_lock); 6036 6037 ext4_superblock_csum_set(sb); 6038 unlock_buffer(sbh); 6039 } 6040 6041 static int ext4_commit_super(struct super_block *sb) 6042 { 6043 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 6044 6045 if (!sbh) 6046 return -EINVAL; 6047 if (block_device_ejected(sb)) 6048 return -ENODEV; 6049 6050 ext4_update_super(sb); 6051 6052 lock_buffer(sbh); 6053 /* Buffer got discarded which means block device got invalidated */ 6054 if (!buffer_mapped(sbh)) { 6055 unlock_buffer(sbh); 6056 return -EIO; 6057 } 6058 6059 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 6060 /* 6061 * Oh, dear. A previous attempt to write the 6062 * superblock failed. This could happen because the 6063 * USB device was yanked out. Or it could happen to 6064 * be a transient write error and maybe the block will 6065 * be remapped. Nothing we can do but to retry the 6066 * write and hope for the best. 6067 */ 6068 ext4_msg(sb, KERN_ERR, "previous I/O error to " 6069 "superblock detected"); 6070 clear_buffer_write_io_error(sbh); 6071 set_buffer_uptodate(sbh); 6072 } 6073 get_bh(sbh); 6074 /* Clear potential dirty bit if it was journalled update */ 6075 clear_buffer_dirty(sbh); 6076 sbh->b_end_io = end_buffer_write_sync; 6077 submit_bh(REQ_OP_WRITE | REQ_SYNC | 6078 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh); 6079 wait_on_buffer(sbh); 6080 if (buffer_write_io_error(sbh)) { 6081 ext4_msg(sb, KERN_ERR, "I/O error while writing " 6082 "superblock"); 6083 clear_buffer_write_io_error(sbh); 6084 set_buffer_uptodate(sbh); 6085 return -EIO; 6086 } 6087 return 0; 6088 } 6089 6090 /* 6091 * Have we just finished recovery? If so, and if we are mounting (or 6092 * remounting) the filesystem readonly, then we will end up with a 6093 * consistent fs on disk. Record that fact. 6094 */ 6095 static int ext4_mark_recovery_complete(struct super_block *sb, 6096 struct ext4_super_block *es) 6097 { 6098 int err; 6099 journal_t *journal = EXT4_SB(sb)->s_journal; 6100 6101 if (!ext4_has_feature_journal(sb)) { 6102 if (journal != NULL) { 6103 ext4_error(sb, "Journal got removed while the fs was " 6104 "mounted!"); 6105 return -EFSCORRUPTED; 6106 } 6107 return 0; 6108 } 6109 jbd2_journal_lock_updates(journal); 6110 err = jbd2_journal_flush(journal, 0); 6111 if (err < 0) 6112 goto out; 6113 6114 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) || 6115 ext4_has_feature_orphan_present(sb))) { 6116 if (!ext4_orphan_file_empty(sb)) { 6117 ext4_error(sb, "Orphan file not empty on read-only fs."); 6118 err = -EFSCORRUPTED; 6119 goto out; 6120 } 6121 ext4_clear_feature_journal_needs_recovery(sb); 6122 ext4_clear_feature_orphan_present(sb); 6123 ext4_commit_super(sb); 6124 } 6125 out: 6126 jbd2_journal_unlock_updates(journal); 6127 return err; 6128 } 6129 6130 /* 6131 * If we are mounting (or read-write remounting) a filesystem whose journal 6132 * has recorded an error from a previous lifetime, move that error to the 6133 * main filesystem now. 6134 */ 6135 static int ext4_clear_journal_err(struct super_block *sb, 6136 struct ext4_super_block *es) 6137 { 6138 journal_t *journal; 6139 int j_errno; 6140 const char *errstr; 6141 6142 if (!ext4_has_feature_journal(sb)) { 6143 ext4_error(sb, "Journal got removed while the fs was mounted!"); 6144 return -EFSCORRUPTED; 6145 } 6146 6147 journal = EXT4_SB(sb)->s_journal; 6148 6149 /* 6150 * Now check for any error status which may have been recorded in the 6151 * journal by a prior ext4_error() or ext4_abort() 6152 */ 6153 6154 j_errno = jbd2_journal_errno(journal); 6155 if (j_errno) { 6156 char nbuf[16]; 6157 6158 errstr = ext4_decode_error(sb, j_errno, nbuf); 6159 ext4_warning(sb, "Filesystem error recorded " 6160 "from previous mount: %s", errstr); 6161 ext4_warning(sb, "Marking fs in need of filesystem check."); 6162 6163 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 6164 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 6165 ext4_commit_super(sb); 6166 6167 jbd2_journal_clear_err(journal); 6168 jbd2_journal_update_sb_errno(journal); 6169 } 6170 return 0; 6171 } 6172 6173 /* 6174 * Force the running and committing transactions to commit, 6175 * and wait on the commit. 6176 */ 6177 int ext4_force_commit(struct super_block *sb) 6178 { 6179 journal_t *journal; 6180 6181 if (sb_rdonly(sb)) 6182 return 0; 6183 6184 journal = EXT4_SB(sb)->s_journal; 6185 return ext4_journal_force_commit(journal); 6186 } 6187 6188 static int ext4_sync_fs(struct super_block *sb, int wait) 6189 { 6190 int ret = 0; 6191 tid_t target; 6192 bool needs_barrier = false; 6193 struct ext4_sb_info *sbi = EXT4_SB(sb); 6194 6195 if (unlikely(ext4_forced_shutdown(sbi))) 6196 return 0; 6197 6198 trace_ext4_sync_fs(sb, wait); 6199 flush_workqueue(sbi->rsv_conversion_wq); 6200 /* 6201 * Writeback quota in non-journalled quota case - journalled quota has 6202 * no dirty dquots 6203 */ 6204 dquot_writeback_dquots(sb, -1); 6205 /* 6206 * Data writeback is possible w/o journal transaction, so barrier must 6207 * being sent at the end of the function. But we can skip it if 6208 * transaction_commit will do it for us. 6209 */ 6210 if (sbi->s_journal) { 6211 target = jbd2_get_latest_transaction(sbi->s_journal); 6212 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 6213 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 6214 needs_barrier = true; 6215 6216 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 6217 if (wait) 6218 ret = jbd2_log_wait_commit(sbi->s_journal, 6219 target); 6220 } 6221 } else if (wait && test_opt(sb, BARRIER)) 6222 needs_barrier = true; 6223 if (needs_barrier) { 6224 int err; 6225 err = blkdev_issue_flush(sb->s_bdev); 6226 if (!ret) 6227 ret = err; 6228 } 6229 6230 return ret; 6231 } 6232 6233 /* 6234 * LVM calls this function before a (read-only) snapshot is created. This 6235 * gives us a chance to flush the journal completely and mark the fs clean. 6236 * 6237 * Note that only this function cannot bring a filesystem to be in a clean 6238 * state independently. It relies on upper layer to stop all data & metadata 6239 * modifications. 6240 */ 6241 static int ext4_freeze(struct super_block *sb) 6242 { 6243 int error = 0; 6244 journal_t *journal; 6245 6246 if (sb_rdonly(sb)) 6247 return 0; 6248 6249 journal = EXT4_SB(sb)->s_journal; 6250 6251 if (journal) { 6252 /* Now we set up the journal barrier. */ 6253 jbd2_journal_lock_updates(journal); 6254 6255 /* 6256 * Don't clear the needs_recovery flag if we failed to 6257 * flush the journal. 6258 */ 6259 error = jbd2_journal_flush(journal, 0); 6260 if (error < 0) 6261 goto out; 6262 6263 /* Journal blocked and flushed, clear needs_recovery flag. */ 6264 ext4_clear_feature_journal_needs_recovery(sb); 6265 if (ext4_orphan_file_empty(sb)) 6266 ext4_clear_feature_orphan_present(sb); 6267 } 6268 6269 error = ext4_commit_super(sb); 6270 out: 6271 if (journal) 6272 /* we rely on upper layer to stop further updates */ 6273 jbd2_journal_unlock_updates(journal); 6274 return error; 6275 } 6276 6277 /* 6278 * Called by LVM after the snapshot is done. We need to reset the RECOVER 6279 * flag here, even though the filesystem is not technically dirty yet. 6280 */ 6281 static int ext4_unfreeze(struct super_block *sb) 6282 { 6283 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 6284 return 0; 6285 6286 if (EXT4_SB(sb)->s_journal) { 6287 /* Reset the needs_recovery flag before the fs is unlocked. */ 6288 ext4_set_feature_journal_needs_recovery(sb); 6289 if (ext4_has_feature_orphan_file(sb)) 6290 ext4_set_feature_orphan_present(sb); 6291 } 6292 6293 ext4_commit_super(sb); 6294 return 0; 6295 } 6296 6297 /* 6298 * Structure to save mount options for ext4_remount's benefit 6299 */ 6300 struct ext4_mount_options { 6301 unsigned long s_mount_opt; 6302 unsigned long s_mount_opt2; 6303 kuid_t s_resuid; 6304 kgid_t s_resgid; 6305 unsigned long s_commit_interval; 6306 u32 s_min_batch_time, s_max_batch_time; 6307 #ifdef CONFIG_QUOTA 6308 int s_jquota_fmt; 6309 char *s_qf_names[EXT4_MAXQUOTAS]; 6310 #endif 6311 }; 6312 6313 static int __ext4_remount(struct fs_context *fc, struct super_block *sb) 6314 { 6315 struct ext4_fs_context *ctx = fc->fs_private; 6316 struct ext4_super_block *es; 6317 struct ext4_sb_info *sbi = EXT4_SB(sb); 6318 unsigned long old_sb_flags; 6319 struct ext4_mount_options old_opts; 6320 ext4_group_t g; 6321 int err = 0; 6322 #ifdef CONFIG_QUOTA 6323 int enable_quota = 0; 6324 int i, j; 6325 char *to_free[EXT4_MAXQUOTAS]; 6326 #endif 6327 6328 6329 /* Store the original options */ 6330 old_sb_flags = sb->s_flags; 6331 old_opts.s_mount_opt = sbi->s_mount_opt; 6332 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 6333 old_opts.s_resuid = sbi->s_resuid; 6334 old_opts.s_resgid = sbi->s_resgid; 6335 old_opts.s_commit_interval = sbi->s_commit_interval; 6336 old_opts.s_min_batch_time = sbi->s_min_batch_time; 6337 old_opts.s_max_batch_time = sbi->s_max_batch_time; 6338 #ifdef CONFIG_QUOTA 6339 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 6340 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6341 if (sbi->s_qf_names[i]) { 6342 char *qf_name = get_qf_name(sb, sbi, i); 6343 6344 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 6345 if (!old_opts.s_qf_names[i]) { 6346 for (j = 0; j < i; j++) 6347 kfree(old_opts.s_qf_names[j]); 6348 return -ENOMEM; 6349 } 6350 } else 6351 old_opts.s_qf_names[i] = NULL; 6352 #endif 6353 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) { 6354 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 6355 ctx->journal_ioprio = 6356 sbi->s_journal->j_task->io_context->ioprio; 6357 else 6358 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 6359 6360 } 6361 6362 ext4_apply_options(fc, sb); 6363 6364 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 6365 test_opt(sb, JOURNAL_CHECKSUM)) { 6366 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 6367 "during remount not supported; ignoring"); 6368 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 6369 } 6370 6371 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 6372 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 6373 ext4_msg(sb, KERN_ERR, "can't mount with " 6374 "both data=journal and delalloc"); 6375 err = -EINVAL; 6376 goto restore_opts; 6377 } 6378 if (test_opt(sb, DIOREAD_NOLOCK)) { 6379 ext4_msg(sb, KERN_ERR, "can't mount with " 6380 "both data=journal and dioread_nolock"); 6381 err = -EINVAL; 6382 goto restore_opts; 6383 } 6384 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 6385 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 6386 ext4_msg(sb, KERN_ERR, "can't mount with " 6387 "journal_async_commit in data=ordered mode"); 6388 err = -EINVAL; 6389 goto restore_opts; 6390 } 6391 } 6392 6393 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 6394 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 6395 err = -EINVAL; 6396 goto restore_opts; 6397 } 6398 6399 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) 6400 ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); 6401 6402 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 6403 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 6404 6405 es = sbi->s_es; 6406 6407 if (sbi->s_journal) { 6408 ext4_init_journal_params(sb, sbi->s_journal); 6409 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio); 6410 } 6411 6412 /* Flush outstanding errors before changing fs state */ 6413 flush_work(&sbi->s_error_work); 6414 6415 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) { 6416 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) { 6417 err = -EROFS; 6418 goto restore_opts; 6419 } 6420 6421 if (fc->sb_flags & SB_RDONLY) { 6422 err = sync_filesystem(sb); 6423 if (err < 0) 6424 goto restore_opts; 6425 err = dquot_suspend(sb, -1); 6426 if (err < 0) 6427 goto restore_opts; 6428 6429 /* 6430 * First of all, the unconditional stuff we have to do 6431 * to disable replay of the journal when we next remount 6432 */ 6433 sb->s_flags |= SB_RDONLY; 6434 6435 /* 6436 * OK, test if we are remounting a valid rw partition 6437 * readonly, and if so set the rdonly flag and then 6438 * mark the partition as valid again. 6439 */ 6440 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 6441 (sbi->s_mount_state & EXT4_VALID_FS)) 6442 es->s_state = cpu_to_le16(sbi->s_mount_state); 6443 6444 if (sbi->s_journal) { 6445 /* 6446 * We let remount-ro finish even if marking fs 6447 * as clean failed... 6448 */ 6449 ext4_mark_recovery_complete(sb, es); 6450 } 6451 } else { 6452 /* Make sure we can mount this feature set readwrite */ 6453 if (ext4_has_feature_readonly(sb) || 6454 !ext4_feature_set_ok(sb, 0)) { 6455 err = -EROFS; 6456 goto restore_opts; 6457 } 6458 /* 6459 * Make sure the group descriptor checksums 6460 * are sane. If they aren't, refuse to remount r/w. 6461 */ 6462 for (g = 0; g < sbi->s_groups_count; g++) { 6463 struct ext4_group_desc *gdp = 6464 ext4_get_group_desc(sb, g, NULL); 6465 6466 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 6467 ext4_msg(sb, KERN_ERR, 6468 "ext4_remount: Checksum for group %u failed (%u!=%u)", 6469 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 6470 le16_to_cpu(gdp->bg_checksum)); 6471 err = -EFSBADCRC; 6472 goto restore_opts; 6473 } 6474 } 6475 6476 /* 6477 * If we have an unprocessed orphan list hanging 6478 * around from a previously readonly bdev mount, 6479 * require a full umount/remount for now. 6480 */ 6481 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) { 6482 ext4_msg(sb, KERN_WARNING, "Couldn't " 6483 "remount RDWR because of unprocessed " 6484 "orphan inode list. Please " 6485 "umount/remount instead"); 6486 err = -EINVAL; 6487 goto restore_opts; 6488 } 6489 6490 /* 6491 * Mounting a RDONLY partition read-write, so reread 6492 * and store the current valid flag. (It may have 6493 * been changed by e2fsck since we originally mounted 6494 * the partition.) 6495 */ 6496 if (sbi->s_journal) { 6497 err = ext4_clear_journal_err(sb, es); 6498 if (err) 6499 goto restore_opts; 6500 } 6501 sbi->s_mount_state = (le16_to_cpu(es->s_state) & 6502 ~EXT4_FC_REPLAY); 6503 6504 err = ext4_setup_super(sb, es, 0); 6505 if (err) 6506 goto restore_opts; 6507 6508 sb->s_flags &= ~SB_RDONLY; 6509 if (ext4_has_feature_mmp(sb)) 6510 if (ext4_multi_mount_protect(sb, 6511 le64_to_cpu(es->s_mmp_block))) { 6512 err = -EROFS; 6513 goto restore_opts; 6514 } 6515 #ifdef CONFIG_QUOTA 6516 enable_quota = 1; 6517 #endif 6518 } 6519 } 6520 6521 /* 6522 * Reinitialize lazy itable initialization thread based on 6523 * current settings 6524 */ 6525 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 6526 ext4_unregister_li_request(sb); 6527 else { 6528 ext4_group_t first_not_zeroed; 6529 first_not_zeroed = ext4_has_uninit_itable(sb); 6530 ext4_register_li_request(sb, first_not_zeroed); 6531 } 6532 6533 /* 6534 * Handle creation of system zone data early because it can fail. 6535 * Releasing of existing data is done when we are sure remount will 6536 * succeed. 6537 */ 6538 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) { 6539 err = ext4_setup_system_zone(sb); 6540 if (err) 6541 goto restore_opts; 6542 } 6543 6544 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 6545 err = ext4_commit_super(sb); 6546 if (err) 6547 goto restore_opts; 6548 } 6549 6550 #ifdef CONFIG_QUOTA 6551 /* Release old quota file names */ 6552 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6553 kfree(old_opts.s_qf_names[i]); 6554 if (enable_quota) { 6555 if (sb_any_quota_suspended(sb)) 6556 dquot_resume(sb, -1); 6557 else if (ext4_has_feature_quota(sb)) { 6558 err = ext4_enable_quotas(sb); 6559 if (err) 6560 goto restore_opts; 6561 } 6562 } 6563 #endif 6564 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6565 ext4_release_system_zone(sb); 6566 6567 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6568 ext4_stop_mmpd(sbi); 6569 6570 return 0; 6571 6572 restore_opts: 6573 sb->s_flags = old_sb_flags; 6574 sbi->s_mount_opt = old_opts.s_mount_opt; 6575 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 6576 sbi->s_resuid = old_opts.s_resuid; 6577 sbi->s_resgid = old_opts.s_resgid; 6578 sbi->s_commit_interval = old_opts.s_commit_interval; 6579 sbi->s_min_batch_time = old_opts.s_min_batch_time; 6580 sbi->s_max_batch_time = old_opts.s_max_batch_time; 6581 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks) 6582 ext4_release_system_zone(sb); 6583 #ifdef CONFIG_QUOTA 6584 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 6585 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 6586 to_free[i] = get_qf_name(sb, sbi, i); 6587 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 6588 } 6589 synchronize_rcu(); 6590 for (i = 0; i < EXT4_MAXQUOTAS; i++) 6591 kfree(to_free[i]); 6592 #endif 6593 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) 6594 ext4_stop_mmpd(sbi); 6595 return err; 6596 } 6597 6598 static int ext4_reconfigure(struct fs_context *fc) 6599 { 6600 struct super_block *sb = fc->root->d_sb; 6601 int ret; 6602 6603 fc->s_fs_info = EXT4_SB(sb); 6604 6605 ret = ext4_check_opt_consistency(fc, sb); 6606 if (ret < 0) 6607 return ret; 6608 6609 ret = __ext4_remount(fc, sb); 6610 if (ret < 0) 6611 return ret; 6612 6613 ext4_msg(sb, KERN_INFO, "re-mounted. Quota mode: %s.", 6614 ext4_quota_mode(sb)); 6615 6616 return 0; 6617 } 6618 6619 #ifdef CONFIG_QUOTA 6620 static int ext4_statfs_project(struct super_block *sb, 6621 kprojid_t projid, struct kstatfs *buf) 6622 { 6623 struct kqid qid; 6624 struct dquot *dquot; 6625 u64 limit; 6626 u64 curblock; 6627 6628 qid = make_kqid_projid(projid); 6629 dquot = dqget(sb, qid); 6630 if (IS_ERR(dquot)) 6631 return PTR_ERR(dquot); 6632 spin_lock(&dquot->dq_dqb_lock); 6633 6634 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit, 6635 dquot->dq_dqb.dqb_bhardlimit); 6636 limit >>= sb->s_blocksize_bits; 6637 6638 if (limit && buf->f_blocks > limit) { 6639 curblock = (dquot->dq_dqb.dqb_curspace + 6640 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 6641 buf->f_blocks = limit; 6642 buf->f_bfree = buf->f_bavail = 6643 (buf->f_blocks > curblock) ? 6644 (buf->f_blocks - curblock) : 0; 6645 } 6646 6647 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit, 6648 dquot->dq_dqb.dqb_ihardlimit); 6649 if (limit && buf->f_files > limit) { 6650 buf->f_files = limit; 6651 buf->f_ffree = 6652 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 6653 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 6654 } 6655 6656 spin_unlock(&dquot->dq_dqb_lock); 6657 dqput(dquot); 6658 return 0; 6659 } 6660 #endif 6661 6662 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 6663 { 6664 struct super_block *sb = dentry->d_sb; 6665 struct ext4_sb_info *sbi = EXT4_SB(sb); 6666 struct ext4_super_block *es = sbi->s_es; 6667 ext4_fsblk_t overhead = 0, resv_blocks; 6668 s64 bfree; 6669 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 6670 6671 if (!test_opt(sb, MINIX_DF)) 6672 overhead = sbi->s_overhead; 6673 6674 buf->f_type = EXT4_SUPER_MAGIC; 6675 buf->f_bsize = sb->s_blocksize; 6676 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 6677 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 6678 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 6679 /* prevent underflow in case that few free space is available */ 6680 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 6681 buf->f_bavail = buf->f_bfree - 6682 (ext4_r_blocks_count(es) + resv_blocks); 6683 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 6684 buf->f_bavail = 0; 6685 buf->f_files = le32_to_cpu(es->s_inodes_count); 6686 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 6687 buf->f_namelen = EXT4_NAME_LEN; 6688 buf->f_fsid = uuid_to_fsid(es->s_uuid); 6689 6690 #ifdef CONFIG_QUOTA 6691 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 6692 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 6693 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 6694 #endif 6695 return 0; 6696 } 6697 6698 6699 #ifdef CONFIG_QUOTA 6700 6701 /* 6702 * Helper functions so that transaction is started before we acquire dqio_sem 6703 * to keep correct lock ordering of transaction > dqio_sem 6704 */ 6705 static inline struct inode *dquot_to_inode(struct dquot *dquot) 6706 { 6707 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 6708 } 6709 6710 static int ext4_write_dquot(struct dquot *dquot) 6711 { 6712 int ret, err; 6713 handle_t *handle; 6714 struct inode *inode; 6715 6716 inode = dquot_to_inode(dquot); 6717 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 6718 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 6719 if (IS_ERR(handle)) 6720 return PTR_ERR(handle); 6721 ret = dquot_commit(dquot); 6722 err = ext4_journal_stop(handle); 6723 if (!ret) 6724 ret = err; 6725 return ret; 6726 } 6727 6728 static int ext4_acquire_dquot(struct dquot *dquot) 6729 { 6730 int ret, err; 6731 handle_t *handle; 6732 6733 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6734 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 6735 if (IS_ERR(handle)) 6736 return PTR_ERR(handle); 6737 ret = dquot_acquire(dquot); 6738 err = ext4_journal_stop(handle); 6739 if (!ret) 6740 ret = err; 6741 return ret; 6742 } 6743 6744 static int ext4_release_dquot(struct dquot *dquot) 6745 { 6746 int ret, err; 6747 handle_t *handle; 6748 6749 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 6750 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 6751 if (IS_ERR(handle)) { 6752 /* Release dquot anyway to avoid endless cycle in dqput() */ 6753 dquot_release(dquot); 6754 return PTR_ERR(handle); 6755 } 6756 ret = dquot_release(dquot); 6757 err = ext4_journal_stop(handle); 6758 if (!ret) 6759 ret = err; 6760 return ret; 6761 } 6762 6763 static int ext4_mark_dquot_dirty(struct dquot *dquot) 6764 { 6765 struct super_block *sb = dquot->dq_sb; 6766 6767 if (ext4_is_quota_journalled(sb)) { 6768 dquot_mark_dquot_dirty(dquot); 6769 return ext4_write_dquot(dquot); 6770 } else { 6771 return dquot_mark_dquot_dirty(dquot); 6772 } 6773 } 6774 6775 static int ext4_write_info(struct super_block *sb, int type) 6776 { 6777 int ret, err; 6778 handle_t *handle; 6779 6780 /* Data block + inode block */ 6781 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2); 6782 if (IS_ERR(handle)) 6783 return PTR_ERR(handle); 6784 ret = dquot_commit_info(sb, type); 6785 err = ext4_journal_stop(handle); 6786 if (!ret) 6787 ret = err; 6788 return ret; 6789 } 6790 6791 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 6792 { 6793 struct ext4_inode_info *ei = EXT4_I(inode); 6794 6795 /* The first argument of lockdep_set_subclass has to be 6796 * *exactly* the same as the argument to init_rwsem() --- in 6797 * this case, in init_once() --- or lockdep gets unhappy 6798 * because the name of the lock is set using the 6799 * stringification of the argument to init_rwsem(). 6800 */ 6801 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 6802 lockdep_set_subclass(&ei->i_data_sem, subclass); 6803 } 6804 6805 /* 6806 * Standard function to be called on quota_on 6807 */ 6808 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 6809 const struct path *path) 6810 { 6811 int err; 6812 6813 if (!test_opt(sb, QUOTA)) 6814 return -EINVAL; 6815 6816 /* Quotafile not on the same filesystem? */ 6817 if (path->dentry->d_sb != sb) 6818 return -EXDEV; 6819 6820 /* Quota already enabled for this file? */ 6821 if (IS_NOQUOTA(d_inode(path->dentry))) 6822 return -EBUSY; 6823 6824 /* Journaling quota? */ 6825 if (EXT4_SB(sb)->s_qf_names[type]) { 6826 /* Quotafile not in fs root? */ 6827 if (path->dentry->d_parent != sb->s_root) 6828 ext4_msg(sb, KERN_WARNING, 6829 "Quota file not on filesystem root. " 6830 "Journaled quota will not work"); 6831 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 6832 } else { 6833 /* 6834 * Clear the flag just in case mount options changed since 6835 * last time. 6836 */ 6837 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 6838 } 6839 6840 /* 6841 * When we journal data on quota file, we have to flush journal to see 6842 * all updates to the file when we bypass pagecache... 6843 */ 6844 if (EXT4_SB(sb)->s_journal && 6845 ext4_should_journal_data(d_inode(path->dentry))) { 6846 /* 6847 * We don't need to lock updates but journal_flush() could 6848 * otherwise be livelocked... 6849 */ 6850 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 6851 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0); 6852 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 6853 if (err) 6854 return err; 6855 } 6856 6857 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 6858 err = dquot_quota_on(sb, type, format_id, path); 6859 if (!err) { 6860 struct inode *inode = d_inode(path->dentry); 6861 handle_t *handle; 6862 6863 /* 6864 * Set inode flags to prevent userspace from messing with quota 6865 * files. If this fails, we return success anyway since quotas 6866 * are already enabled and this is not a hard failure. 6867 */ 6868 inode_lock(inode); 6869 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6870 if (IS_ERR(handle)) 6871 goto unlock_inode; 6872 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 6873 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 6874 S_NOATIME | S_IMMUTABLE); 6875 err = ext4_mark_inode_dirty(handle, inode); 6876 ext4_journal_stop(handle); 6877 unlock_inode: 6878 inode_unlock(inode); 6879 if (err) 6880 dquot_quota_off(sb, type); 6881 } 6882 if (err) 6883 lockdep_set_quota_inode(path->dentry->d_inode, 6884 I_DATA_SEM_NORMAL); 6885 return err; 6886 } 6887 6888 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 6889 unsigned int flags) 6890 { 6891 int err; 6892 struct inode *qf_inode; 6893 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6894 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6895 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6896 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6897 }; 6898 6899 BUG_ON(!ext4_has_feature_quota(sb)); 6900 6901 if (!qf_inums[type]) 6902 return -EPERM; 6903 6904 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 6905 if (IS_ERR(qf_inode)) { 6906 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 6907 return PTR_ERR(qf_inode); 6908 } 6909 6910 /* Don't account quota for quota files to avoid recursion */ 6911 qf_inode->i_flags |= S_NOQUOTA; 6912 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 6913 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 6914 if (err) 6915 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 6916 iput(qf_inode); 6917 6918 return err; 6919 } 6920 6921 /* Enable usage tracking for all quota types. */ 6922 int ext4_enable_quotas(struct super_block *sb) 6923 { 6924 int type, err = 0; 6925 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 6926 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 6927 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 6928 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 6929 }; 6930 bool quota_mopt[EXT4_MAXQUOTAS] = { 6931 test_opt(sb, USRQUOTA), 6932 test_opt(sb, GRPQUOTA), 6933 test_opt(sb, PRJQUOTA), 6934 }; 6935 6936 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 6937 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 6938 if (qf_inums[type]) { 6939 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 6940 DQUOT_USAGE_ENABLED | 6941 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 6942 if (err) { 6943 ext4_warning(sb, 6944 "Failed to enable quota tracking " 6945 "(type=%d, err=%d). Please run " 6946 "e2fsck to fix.", type, err); 6947 for (type--; type >= 0; type--) { 6948 struct inode *inode; 6949 6950 inode = sb_dqopt(sb)->files[type]; 6951 if (inode) 6952 inode = igrab(inode); 6953 dquot_quota_off(sb, type); 6954 if (inode) { 6955 lockdep_set_quota_inode(inode, 6956 I_DATA_SEM_NORMAL); 6957 iput(inode); 6958 } 6959 } 6960 6961 return err; 6962 } 6963 } 6964 } 6965 return 0; 6966 } 6967 6968 static int ext4_quota_off(struct super_block *sb, int type) 6969 { 6970 struct inode *inode = sb_dqopt(sb)->files[type]; 6971 handle_t *handle; 6972 int err; 6973 6974 /* Force all delayed allocation blocks to be allocated. 6975 * Caller already holds s_umount sem */ 6976 if (test_opt(sb, DELALLOC)) 6977 sync_filesystem(sb); 6978 6979 if (!inode || !igrab(inode)) 6980 goto out; 6981 6982 err = dquot_quota_off(sb, type); 6983 if (err || ext4_has_feature_quota(sb)) 6984 goto out_put; 6985 6986 inode_lock(inode); 6987 /* 6988 * Update modification times of quota files when userspace can 6989 * start looking at them. If we fail, we return success anyway since 6990 * this is not a hard failure and quotas are already disabled. 6991 */ 6992 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 6993 if (IS_ERR(handle)) { 6994 err = PTR_ERR(handle); 6995 goto out_unlock; 6996 } 6997 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 6998 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 6999 inode->i_mtime = inode->i_ctime = current_time(inode); 7000 err = ext4_mark_inode_dirty(handle, inode); 7001 ext4_journal_stop(handle); 7002 out_unlock: 7003 inode_unlock(inode); 7004 out_put: 7005 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 7006 iput(inode); 7007 return err; 7008 out: 7009 return dquot_quota_off(sb, type); 7010 } 7011 7012 /* Read data from quotafile - avoid pagecache and such because we cannot afford 7013 * acquiring the locks... As quota files are never truncated and quota code 7014 * itself serializes the operations (and no one else should touch the files) 7015 * we don't have to be afraid of races */ 7016 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 7017 size_t len, loff_t off) 7018 { 7019 struct inode *inode = sb_dqopt(sb)->files[type]; 7020 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7021 int offset = off & (sb->s_blocksize - 1); 7022 int tocopy; 7023 size_t toread; 7024 struct buffer_head *bh; 7025 loff_t i_size = i_size_read(inode); 7026 7027 if (off > i_size) 7028 return 0; 7029 if (off+len > i_size) 7030 len = i_size-off; 7031 toread = len; 7032 while (toread > 0) { 7033 tocopy = sb->s_blocksize - offset < toread ? 7034 sb->s_blocksize - offset : toread; 7035 bh = ext4_bread(NULL, inode, blk, 0); 7036 if (IS_ERR(bh)) 7037 return PTR_ERR(bh); 7038 if (!bh) /* A hole? */ 7039 memset(data, 0, tocopy); 7040 else 7041 memcpy(data, bh->b_data+offset, tocopy); 7042 brelse(bh); 7043 offset = 0; 7044 toread -= tocopy; 7045 data += tocopy; 7046 blk++; 7047 } 7048 return len; 7049 } 7050 7051 /* Write to quotafile (we know the transaction is already started and has 7052 * enough credits) */ 7053 static ssize_t ext4_quota_write(struct super_block *sb, int type, 7054 const char *data, size_t len, loff_t off) 7055 { 7056 struct inode *inode = sb_dqopt(sb)->files[type]; 7057 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 7058 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1); 7059 int retries = 0; 7060 struct buffer_head *bh; 7061 handle_t *handle = journal_current_handle(); 7062 7063 if (!handle) { 7064 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7065 " cancelled because transaction is not started", 7066 (unsigned long long)off, (unsigned long long)len); 7067 return -EIO; 7068 } 7069 /* 7070 * Since we account only one data block in transaction credits, 7071 * then it is impossible to cross a block boundary. 7072 */ 7073 if (sb->s_blocksize - offset < len) { 7074 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 7075 " cancelled because not block aligned", 7076 (unsigned long long)off, (unsigned long long)len); 7077 return -EIO; 7078 } 7079 7080 do { 7081 bh = ext4_bread(handle, inode, blk, 7082 EXT4_GET_BLOCKS_CREATE | 7083 EXT4_GET_BLOCKS_METADATA_NOFAIL); 7084 } while (PTR_ERR(bh) == -ENOSPC && 7085 ext4_should_retry_alloc(inode->i_sb, &retries)); 7086 if (IS_ERR(bh)) 7087 return PTR_ERR(bh); 7088 if (!bh) 7089 goto out; 7090 BUFFER_TRACE(bh, "get write access"); 7091 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 7092 if (err) { 7093 brelse(bh); 7094 return err; 7095 } 7096 lock_buffer(bh); 7097 memcpy(bh->b_data+offset, data, len); 7098 flush_dcache_page(bh->b_page); 7099 unlock_buffer(bh); 7100 err = ext4_handle_dirty_metadata(handle, NULL, bh); 7101 brelse(bh); 7102 out: 7103 if (inode->i_size < off + len) { 7104 i_size_write(inode, off + len); 7105 EXT4_I(inode)->i_disksize = inode->i_size; 7106 err2 = ext4_mark_inode_dirty(handle, inode); 7107 if (unlikely(err2 && !err)) 7108 err = err2; 7109 } 7110 return err ? err : len; 7111 } 7112 #endif 7113 7114 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 7115 static inline void register_as_ext2(void) 7116 { 7117 int err = register_filesystem(&ext2_fs_type); 7118 if (err) 7119 printk(KERN_WARNING 7120 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 7121 } 7122 7123 static inline void unregister_as_ext2(void) 7124 { 7125 unregister_filesystem(&ext2_fs_type); 7126 } 7127 7128 static inline int ext2_feature_set_ok(struct super_block *sb) 7129 { 7130 if (ext4_has_unknown_ext2_incompat_features(sb)) 7131 return 0; 7132 if (sb_rdonly(sb)) 7133 return 1; 7134 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 7135 return 0; 7136 return 1; 7137 } 7138 #else 7139 static inline void register_as_ext2(void) { } 7140 static inline void unregister_as_ext2(void) { } 7141 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 7142 #endif 7143 7144 static inline void register_as_ext3(void) 7145 { 7146 int err = register_filesystem(&ext3_fs_type); 7147 if (err) 7148 printk(KERN_WARNING 7149 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 7150 } 7151 7152 static inline void unregister_as_ext3(void) 7153 { 7154 unregister_filesystem(&ext3_fs_type); 7155 } 7156 7157 static inline int ext3_feature_set_ok(struct super_block *sb) 7158 { 7159 if (ext4_has_unknown_ext3_incompat_features(sb)) 7160 return 0; 7161 if (!ext4_has_feature_journal(sb)) 7162 return 0; 7163 if (sb_rdonly(sb)) 7164 return 1; 7165 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 7166 return 0; 7167 return 1; 7168 } 7169 7170 static struct file_system_type ext4_fs_type = { 7171 .owner = THIS_MODULE, 7172 .name = "ext4", 7173 .init_fs_context = ext4_init_fs_context, 7174 .parameters = ext4_param_specs, 7175 .kill_sb = kill_block_super, 7176 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP, 7177 }; 7178 MODULE_ALIAS_FS("ext4"); 7179 7180 /* Shared across all ext4 file systems */ 7181 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 7182 7183 static int __init ext4_init_fs(void) 7184 { 7185 int i, err; 7186 7187 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 7188 ext4_li_info = NULL; 7189 7190 /* Build-time check for flags consistency */ 7191 ext4_check_flag_values(); 7192 7193 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 7194 init_waitqueue_head(&ext4__ioend_wq[i]); 7195 7196 err = ext4_init_es(); 7197 if (err) 7198 return err; 7199 7200 err = ext4_init_pending(); 7201 if (err) 7202 goto out7; 7203 7204 err = ext4_init_post_read_processing(); 7205 if (err) 7206 goto out6; 7207 7208 err = ext4_init_pageio(); 7209 if (err) 7210 goto out5; 7211 7212 err = ext4_init_system_zone(); 7213 if (err) 7214 goto out4; 7215 7216 err = ext4_init_sysfs(); 7217 if (err) 7218 goto out3; 7219 7220 err = ext4_init_mballoc(); 7221 if (err) 7222 goto out2; 7223 err = init_inodecache(); 7224 if (err) 7225 goto out1; 7226 7227 err = ext4_fc_init_dentry_cache(); 7228 if (err) 7229 goto out05; 7230 7231 register_as_ext3(); 7232 register_as_ext2(); 7233 err = register_filesystem(&ext4_fs_type); 7234 if (err) 7235 goto out; 7236 7237 return 0; 7238 out: 7239 unregister_as_ext2(); 7240 unregister_as_ext3(); 7241 ext4_fc_destroy_dentry_cache(); 7242 out05: 7243 destroy_inodecache(); 7244 out1: 7245 ext4_exit_mballoc(); 7246 out2: 7247 ext4_exit_sysfs(); 7248 out3: 7249 ext4_exit_system_zone(); 7250 out4: 7251 ext4_exit_pageio(); 7252 out5: 7253 ext4_exit_post_read_processing(); 7254 out6: 7255 ext4_exit_pending(); 7256 out7: 7257 ext4_exit_es(); 7258 7259 return err; 7260 } 7261 7262 static void __exit ext4_exit_fs(void) 7263 { 7264 ext4_destroy_lazyinit_thread(); 7265 unregister_as_ext2(); 7266 unregister_as_ext3(); 7267 unregister_filesystem(&ext4_fs_type); 7268 ext4_fc_destroy_dentry_cache(); 7269 destroy_inodecache(); 7270 ext4_exit_mballoc(); 7271 ext4_exit_sysfs(); 7272 ext4_exit_system_zone(); 7273 ext4_exit_pageio(); 7274 ext4_exit_post_read_processing(); 7275 ext4_exit_es(); 7276 ext4_exit_pending(); 7277 } 7278 7279 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 7280 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 7281 MODULE_LICENSE("GPL"); 7282 MODULE_SOFTDEP("pre: crc32c"); 7283 module_init(ext4_init_fs) 7284 module_exit(ext4_exit_fs) 7285