1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/super.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Big-endian to little-endian byte-swapping/bitmaps by 17 * David S. Miller (davem@caip.rutgers.edu), 1995 18 */ 19 20 #include <linux/module.h> 21 #include <linux/string.h> 22 #include <linux/fs.h> 23 #include <linux/time.h> 24 #include <linux/vmalloc.h> 25 #include <linux/slab.h> 26 #include <linux/init.h> 27 #include <linux/blkdev.h> 28 #include <linux/backing-dev.h> 29 #include <linux/parser.h> 30 #include <linux/buffer_head.h> 31 #include <linux/exportfs.h> 32 #include <linux/vfs.h> 33 #include <linux/random.h> 34 #include <linux/mount.h> 35 #include <linux/namei.h> 36 #include <linux/quotaops.h> 37 #include <linux/seq_file.h> 38 #include <linux/ctype.h> 39 #include <linux/log2.h> 40 #include <linux/crc16.h> 41 #include <linux/dax.h> 42 #include <linux/cleancache.h> 43 #include <linux/uaccess.h> 44 #include <linux/iversion.h> 45 #include <linux/unicode.h> 46 47 #include <linux/kthread.h> 48 #include <linux/freezer.h> 49 50 #include "ext4.h" 51 #include "ext4_extents.h" /* Needed for trace points definition */ 52 #include "ext4_jbd2.h" 53 #include "xattr.h" 54 #include "acl.h" 55 #include "mballoc.h" 56 #include "fsmap.h" 57 58 #define CREATE_TRACE_POINTS 59 #include <trace/events/ext4.h> 60 61 static struct ext4_lazy_init *ext4_li_info; 62 static struct mutex ext4_li_mtx; 63 static struct ratelimit_state ext4_mount_msg_ratelimit; 64 65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 66 unsigned long journal_devnum); 67 static int ext4_show_options(struct seq_file *seq, struct dentry *root); 68 static int ext4_commit_super(struct super_block *sb, int sync); 69 static void ext4_mark_recovery_complete(struct super_block *sb, 70 struct ext4_super_block *es); 71 static void ext4_clear_journal_err(struct super_block *sb, 72 struct ext4_super_block *es); 73 static int ext4_sync_fs(struct super_block *sb, int wait); 74 static int ext4_remount(struct super_block *sb, int *flags, char *data); 75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf); 76 static int ext4_unfreeze(struct super_block *sb); 77 static int ext4_freeze(struct super_block *sb); 78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 79 const char *dev_name, void *data); 80 static inline int ext2_feature_set_ok(struct super_block *sb); 81 static inline int ext3_feature_set_ok(struct super_block *sb); 82 static int ext4_feature_set_ok(struct super_block *sb, int readonly); 83 static void ext4_destroy_lazyinit_thread(void); 84 static void ext4_unregister_li_request(struct super_block *sb); 85 static void ext4_clear_request_list(void); 86 static struct inode *ext4_get_journal_inode(struct super_block *sb, 87 unsigned int journal_inum); 88 89 /* 90 * Lock ordering 91 * 92 * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and 93 * i_mmap_rwsem (inode->i_mmap_rwsem)! 94 * 95 * page fault path: 96 * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start -> 97 * page lock -> i_data_sem (rw) 98 * 99 * buffered write path: 100 * sb_start_write -> i_mutex -> mmap_sem 101 * sb_start_write -> i_mutex -> transaction start -> page lock -> 102 * i_data_sem (rw) 103 * 104 * truncate: 105 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock 106 * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start -> 107 * i_data_sem (rw) 108 * 109 * direct IO: 110 * sb_start_write -> i_mutex -> mmap_sem 111 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw) 112 * 113 * writepages: 114 * transaction start -> page lock(s) -> i_data_sem (rw) 115 */ 116 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 118 static struct file_system_type ext2_fs_type = { 119 .owner = THIS_MODULE, 120 .name = "ext2", 121 .mount = ext4_mount, 122 .kill_sb = kill_block_super, 123 .fs_flags = FS_REQUIRES_DEV, 124 }; 125 MODULE_ALIAS_FS("ext2"); 126 MODULE_ALIAS("ext2"); 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type) 128 #else 129 #define IS_EXT2_SB(sb) (0) 130 #endif 131 132 133 static struct file_system_type ext3_fs_type = { 134 .owner = THIS_MODULE, 135 .name = "ext3", 136 .mount = ext4_mount, 137 .kill_sb = kill_block_super, 138 .fs_flags = FS_REQUIRES_DEV, 139 }; 140 MODULE_ALIAS_FS("ext3"); 141 MODULE_ALIAS("ext3"); 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type) 143 144 /* 145 * This works like sb_bread() except it uses ERR_PTR for error 146 * returns. Currently with sb_bread it's impossible to distinguish 147 * between ENOMEM and EIO situations (since both result in a NULL 148 * return. 149 */ 150 struct buffer_head * 151 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags) 152 { 153 struct buffer_head *bh = sb_getblk(sb, block); 154 155 if (bh == NULL) 156 return ERR_PTR(-ENOMEM); 157 if (buffer_uptodate(bh)) 158 return bh; 159 ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh); 160 wait_on_buffer(bh); 161 if (buffer_uptodate(bh)) 162 return bh; 163 put_bh(bh); 164 return ERR_PTR(-EIO); 165 } 166 167 static int ext4_verify_csum_type(struct super_block *sb, 168 struct ext4_super_block *es) 169 { 170 if (!ext4_has_feature_metadata_csum(sb)) 171 return 1; 172 173 return es->s_checksum_type == EXT4_CRC32C_CHKSUM; 174 } 175 176 static __le32 ext4_superblock_csum(struct super_block *sb, 177 struct ext4_super_block *es) 178 { 179 struct ext4_sb_info *sbi = EXT4_SB(sb); 180 int offset = offsetof(struct ext4_super_block, s_checksum); 181 __u32 csum; 182 183 csum = ext4_chksum(sbi, ~0, (char *)es, offset); 184 185 return cpu_to_le32(csum); 186 } 187 188 static int ext4_superblock_csum_verify(struct super_block *sb, 189 struct ext4_super_block *es) 190 { 191 if (!ext4_has_metadata_csum(sb)) 192 return 1; 193 194 return es->s_checksum == ext4_superblock_csum(sb, es); 195 } 196 197 void ext4_superblock_csum_set(struct super_block *sb) 198 { 199 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 200 201 if (!ext4_has_metadata_csum(sb)) 202 return; 203 204 es->s_checksum = ext4_superblock_csum(sb, es); 205 } 206 207 void *ext4_kvmalloc(size_t size, gfp_t flags) 208 { 209 void *ret; 210 211 ret = kmalloc(size, flags | __GFP_NOWARN); 212 if (!ret) 213 ret = __vmalloc(size, flags, PAGE_KERNEL); 214 return ret; 215 } 216 217 void *ext4_kvzalloc(size_t size, gfp_t flags) 218 { 219 void *ret; 220 221 ret = kzalloc(size, flags | __GFP_NOWARN); 222 if (!ret) 223 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL); 224 return ret; 225 } 226 227 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 228 struct ext4_group_desc *bg) 229 { 230 return le32_to_cpu(bg->bg_block_bitmap_lo) | 231 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 232 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0); 233 } 234 235 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb, 236 struct ext4_group_desc *bg) 237 { 238 return le32_to_cpu(bg->bg_inode_bitmap_lo) | 239 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 240 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0); 241 } 242 243 ext4_fsblk_t ext4_inode_table(struct super_block *sb, 244 struct ext4_group_desc *bg) 245 { 246 return le32_to_cpu(bg->bg_inode_table_lo) | 247 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 248 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0); 249 } 250 251 __u32 ext4_free_group_clusters(struct super_block *sb, 252 struct ext4_group_desc *bg) 253 { 254 return le16_to_cpu(bg->bg_free_blocks_count_lo) | 255 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 256 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0); 257 } 258 259 __u32 ext4_free_inodes_count(struct super_block *sb, 260 struct ext4_group_desc *bg) 261 { 262 return le16_to_cpu(bg->bg_free_inodes_count_lo) | 263 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 264 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0); 265 } 266 267 __u32 ext4_used_dirs_count(struct super_block *sb, 268 struct ext4_group_desc *bg) 269 { 270 return le16_to_cpu(bg->bg_used_dirs_count_lo) | 271 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 272 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0); 273 } 274 275 __u32 ext4_itable_unused_count(struct super_block *sb, 276 struct ext4_group_desc *bg) 277 { 278 return le16_to_cpu(bg->bg_itable_unused_lo) | 279 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ? 280 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0); 281 } 282 283 void ext4_block_bitmap_set(struct super_block *sb, 284 struct ext4_group_desc *bg, ext4_fsblk_t blk) 285 { 286 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk); 287 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 288 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32); 289 } 290 291 void ext4_inode_bitmap_set(struct super_block *sb, 292 struct ext4_group_desc *bg, ext4_fsblk_t blk) 293 { 294 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk); 295 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 296 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32); 297 } 298 299 void ext4_inode_table_set(struct super_block *sb, 300 struct ext4_group_desc *bg, ext4_fsblk_t blk) 301 { 302 bg->bg_inode_table_lo = cpu_to_le32((u32)blk); 303 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 304 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32); 305 } 306 307 void ext4_free_group_clusters_set(struct super_block *sb, 308 struct ext4_group_desc *bg, __u32 count) 309 { 310 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count); 311 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 312 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16); 313 } 314 315 void ext4_free_inodes_set(struct super_block *sb, 316 struct ext4_group_desc *bg, __u32 count) 317 { 318 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count); 319 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 320 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16); 321 } 322 323 void ext4_used_dirs_set(struct super_block *sb, 324 struct ext4_group_desc *bg, __u32 count) 325 { 326 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count); 327 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 328 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16); 329 } 330 331 void ext4_itable_unused_set(struct super_block *sb, 332 struct ext4_group_desc *bg, __u32 count) 333 { 334 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count); 335 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT) 336 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16); 337 } 338 339 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi) 340 { 341 time64_t now = ktime_get_real_seconds(); 342 343 now = clamp_val(now, 0, (1ull << 40) - 1); 344 345 *lo = cpu_to_le32(lower_32_bits(now)); 346 *hi = upper_32_bits(now); 347 } 348 349 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi) 350 { 351 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo); 352 } 353 #define ext4_update_tstamp(es, tstamp) \ 354 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 355 #define ext4_get_tstamp(es, tstamp) \ 356 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi) 357 358 static void __save_error_info(struct super_block *sb, const char *func, 359 unsigned int line) 360 { 361 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 362 363 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 364 if (bdev_read_only(sb->s_bdev)) 365 return; 366 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 367 ext4_update_tstamp(es, s_last_error_time); 368 strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func)); 369 es->s_last_error_line = cpu_to_le32(line); 370 if (!es->s_first_error_time) { 371 es->s_first_error_time = es->s_last_error_time; 372 es->s_first_error_time_hi = es->s_last_error_time_hi; 373 strncpy(es->s_first_error_func, func, 374 sizeof(es->s_first_error_func)); 375 es->s_first_error_line = cpu_to_le32(line); 376 es->s_first_error_ino = es->s_last_error_ino; 377 es->s_first_error_block = es->s_last_error_block; 378 } 379 /* 380 * Start the daily error reporting function if it hasn't been 381 * started already 382 */ 383 if (!es->s_error_count) 384 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ); 385 le32_add_cpu(&es->s_error_count, 1); 386 } 387 388 static void save_error_info(struct super_block *sb, const char *func, 389 unsigned int line) 390 { 391 __save_error_info(sb, func, line); 392 ext4_commit_super(sb, 1); 393 } 394 395 /* 396 * The del_gendisk() function uninitializes the disk-specific data 397 * structures, including the bdi structure, without telling anyone 398 * else. Once this happens, any attempt to call mark_buffer_dirty() 399 * (for example, by ext4_commit_super), will cause a kernel OOPS. 400 * This is a kludge to prevent these oops until we can put in a proper 401 * hook in del_gendisk() to inform the VFS and file system layers. 402 */ 403 static int block_device_ejected(struct super_block *sb) 404 { 405 struct inode *bd_inode = sb->s_bdev->bd_inode; 406 struct backing_dev_info *bdi = inode_to_bdi(bd_inode); 407 408 return bdi->dev == NULL; 409 } 410 411 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn) 412 { 413 struct super_block *sb = journal->j_private; 414 struct ext4_sb_info *sbi = EXT4_SB(sb); 415 int error = is_journal_aborted(journal); 416 struct ext4_journal_cb_entry *jce; 417 418 BUG_ON(txn->t_state == T_FINISHED); 419 420 ext4_process_freed_data(sb, txn->t_tid); 421 422 spin_lock(&sbi->s_md_lock); 423 while (!list_empty(&txn->t_private_list)) { 424 jce = list_entry(txn->t_private_list.next, 425 struct ext4_journal_cb_entry, jce_list); 426 list_del_init(&jce->jce_list); 427 spin_unlock(&sbi->s_md_lock); 428 jce->jce_func(sb, jce, error); 429 spin_lock(&sbi->s_md_lock); 430 } 431 spin_unlock(&sbi->s_md_lock); 432 } 433 434 static bool system_going_down(void) 435 { 436 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF 437 || system_state == SYSTEM_RESTART; 438 } 439 440 /* Deal with the reporting of failure conditions on a filesystem such as 441 * inconsistencies detected or read IO failures. 442 * 443 * On ext2, we can store the error state of the filesystem in the 444 * superblock. That is not possible on ext4, because we may have other 445 * write ordering constraints on the superblock which prevent us from 446 * writing it out straight away; and given that the journal is about to 447 * be aborted, we can't rely on the current, or future, transactions to 448 * write out the superblock safely. 449 * 450 * We'll just use the jbd2_journal_abort() error code to record an error in 451 * the journal instead. On recovery, the journal will complain about 452 * that error until we've noted it down and cleared it. 453 */ 454 455 static void ext4_handle_error(struct super_block *sb) 456 { 457 if (test_opt(sb, WARN_ON_ERROR)) 458 WARN_ON_ONCE(1); 459 460 if (sb_rdonly(sb)) 461 return; 462 463 if (!test_opt(sb, ERRORS_CONT)) { 464 journal_t *journal = EXT4_SB(sb)->s_journal; 465 466 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED; 467 if (journal) 468 jbd2_journal_abort(journal, -EIO); 469 } 470 /* 471 * We force ERRORS_RO behavior when system is rebooting. Otherwise we 472 * could panic during 'reboot -f' as the underlying device got already 473 * disabled. 474 */ 475 if (test_opt(sb, ERRORS_RO) || system_going_down()) { 476 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 477 /* 478 * Make sure updated value of ->s_mount_flags will be visible 479 * before ->s_flags update 480 */ 481 smp_wmb(); 482 sb->s_flags |= SB_RDONLY; 483 } else if (test_opt(sb, ERRORS_PANIC)) { 484 if (EXT4_SB(sb)->s_journal && 485 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR)) 486 return; 487 panic("EXT4-fs (device %s): panic forced after error\n", 488 sb->s_id); 489 } 490 } 491 492 #define ext4_error_ratelimit(sb) \ 493 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \ 494 "EXT4-fs error") 495 496 void __ext4_error(struct super_block *sb, const char *function, 497 unsigned int line, const char *fmt, ...) 498 { 499 struct va_format vaf; 500 va_list args; 501 502 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 503 return; 504 505 trace_ext4_error(sb, function, line); 506 if (ext4_error_ratelimit(sb)) { 507 va_start(args, fmt); 508 vaf.fmt = fmt; 509 vaf.va = &args; 510 printk(KERN_CRIT 511 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n", 512 sb->s_id, function, line, current->comm, &vaf); 513 va_end(args); 514 } 515 save_error_info(sb, function, line); 516 ext4_handle_error(sb); 517 } 518 519 void __ext4_error_inode(struct inode *inode, const char *function, 520 unsigned int line, ext4_fsblk_t block, 521 const char *fmt, ...) 522 { 523 va_list args; 524 struct va_format vaf; 525 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; 526 527 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 528 return; 529 530 trace_ext4_error(inode->i_sb, function, line); 531 es->s_last_error_ino = cpu_to_le32(inode->i_ino); 532 es->s_last_error_block = cpu_to_le64(block); 533 if (ext4_error_ratelimit(inode->i_sb)) { 534 va_start(args, fmt); 535 vaf.fmt = fmt; 536 vaf.va = &args; 537 if (block) 538 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 539 "inode #%lu: block %llu: comm %s: %pV\n", 540 inode->i_sb->s_id, function, line, inode->i_ino, 541 block, current->comm, &vaf); 542 else 543 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: " 544 "inode #%lu: comm %s: %pV\n", 545 inode->i_sb->s_id, function, line, inode->i_ino, 546 current->comm, &vaf); 547 va_end(args); 548 } 549 save_error_info(inode->i_sb, function, line); 550 ext4_handle_error(inode->i_sb); 551 } 552 553 void __ext4_error_file(struct file *file, const char *function, 554 unsigned int line, ext4_fsblk_t block, 555 const char *fmt, ...) 556 { 557 va_list args; 558 struct va_format vaf; 559 struct ext4_super_block *es; 560 struct inode *inode = file_inode(file); 561 char pathname[80], *path; 562 563 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 564 return; 565 566 trace_ext4_error(inode->i_sb, function, line); 567 es = EXT4_SB(inode->i_sb)->s_es; 568 es->s_last_error_ino = cpu_to_le32(inode->i_ino); 569 if (ext4_error_ratelimit(inode->i_sb)) { 570 path = file_path(file, pathname, sizeof(pathname)); 571 if (IS_ERR(path)) 572 path = "(unknown)"; 573 va_start(args, fmt); 574 vaf.fmt = fmt; 575 vaf.va = &args; 576 if (block) 577 printk(KERN_CRIT 578 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 579 "block %llu: comm %s: path %s: %pV\n", 580 inode->i_sb->s_id, function, line, inode->i_ino, 581 block, current->comm, path, &vaf); 582 else 583 printk(KERN_CRIT 584 "EXT4-fs error (device %s): %s:%d: inode #%lu: " 585 "comm %s: path %s: %pV\n", 586 inode->i_sb->s_id, function, line, inode->i_ino, 587 current->comm, path, &vaf); 588 va_end(args); 589 } 590 save_error_info(inode->i_sb, function, line); 591 ext4_handle_error(inode->i_sb); 592 } 593 594 const char *ext4_decode_error(struct super_block *sb, int errno, 595 char nbuf[16]) 596 { 597 char *errstr = NULL; 598 599 switch (errno) { 600 case -EFSCORRUPTED: 601 errstr = "Corrupt filesystem"; 602 break; 603 case -EFSBADCRC: 604 errstr = "Filesystem failed CRC"; 605 break; 606 case -EIO: 607 errstr = "IO failure"; 608 break; 609 case -ENOMEM: 610 errstr = "Out of memory"; 611 break; 612 case -EROFS: 613 if (!sb || (EXT4_SB(sb)->s_journal && 614 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)) 615 errstr = "Journal has aborted"; 616 else 617 errstr = "Readonly filesystem"; 618 break; 619 default: 620 /* If the caller passed in an extra buffer for unknown 621 * errors, textualise them now. Else we just return 622 * NULL. */ 623 if (nbuf) { 624 /* Check for truncated error codes... */ 625 if (snprintf(nbuf, 16, "error %d", -errno) >= 0) 626 errstr = nbuf; 627 } 628 break; 629 } 630 631 return errstr; 632 } 633 634 /* __ext4_std_error decodes expected errors from journaling functions 635 * automatically and invokes the appropriate error response. */ 636 637 void __ext4_std_error(struct super_block *sb, const char *function, 638 unsigned int line, int errno) 639 { 640 char nbuf[16]; 641 const char *errstr; 642 643 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 644 return; 645 646 /* Special case: if the error is EROFS, and we're not already 647 * inside a transaction, then there's really no point in logging 648 * an error. */ 649 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb)) 650 return; 651 652 if (ext4_error_ratelimit(sb)) { 653 errstr = ext4_decode_error(sb, errno, nbuf); 654 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n", 655 sb->s_id, function, line, errstr); 656 } 657 658 save_error_info(sb, function, line); 659 ext4_handle_error(sb); 660 } 661 662 /* 663 * ext4_abort is a much stronger failure handler than ext4_error. The 664 * abort function may be used to deal with unrecoverable failures such 665 * as journal IO errors or ENOMEM at a critical moment in log management. 666 * 667 * We unconditionally force the filesystem into an ABORT|READONLY state, 668 * unless the error response on the fs has been set to panic in which 669 * case we take the easy way out and panic immediately. 670 */ 671 672 void __ext4_abort(struct super_block *sb, const char *function, 673 unsigned int line, const char *fmt, ...) 674 { 675 struct va_format vaf; 676 va_list args; 677 678 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 679 return; 680 681 save_error_info(sb, function, line); 682 va_start(args, fmt); 683 vaf.fmt = fmt; 684 vaf.va = &args; 685 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n", 686 sb->s_id, function, line, &vaf); 687 va_end(args); 688 689 if (sb_rdonly(sb) == 0) { 690 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only"); 691 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED; 692 /* 693 * Make sure updated value of ->s_mount_flags will be visible 694 * before ->s_flags update 695 */ 696 smp_wmb(); 697 sb->s_flags |= SB_RDONLY; 698 if (EXT4_SB(sb)->s_journal) 699 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO); 700 save_error_info(sb, function, line); 701 } 702 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) { 703 if (EXT4_SB(sb)->s_journal && 704 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR)) 705 return; 706 panic("EXT4-fs panic from previous error\n"); 707 } 708 } 709 710 void __ext4_msg(struct super_block *sb, 711 const char *prefix, const char *fmt, ...) 712 { 713 struct va_format vaf; 714 va_list args; 715 716 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs")) 717 return; 718 719 va_start(args, fmt); 720 vaf.fmt = fmt; 721 vaf.va = &args; 722 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf); 723 va_end(args); 724 } 725 726 #define ext4_warning_ratelimit(sb) \ 727 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \ 728 "EXT4-fs warning") 729 730 void __ext4_warning(struct super_block *sb, const char *function, 731 unsigned int line, const char *fmt, ...) 732 { 733 struct va_format vaf; 734 va_list args; 735 736 if (!ext4_warning_ratelimit(sb)) 737 return; 738 739 va_start(args, fmt); 740 vaf.fmt = fmt; 741 vaf.va = &args; 742 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n", 743 sb->s_id, function, line, &vaf); 744 va_end(args); 745 } 746 747 void __ext4_warning_inode(const struct inode *inode, const char *function, 748 unsigned int line, const char *fmt, ...) 749 { 750 struct va_format vaf; 751 va_list args; 752 753 if (!ext4_warning_ratelimit(inode->i_sb)) 754 return; 755 756 va_start(args, fmt); 757 vaf.fmt = fmt; 758 vaf.va = &args; 759 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: " 760 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id, 761 function, line, inode->i_ino, current->comm, &vaf); 762 va_end(args); 763 } 764 765 void __ext4_grp_locked_error(const char *function, unsigned int line, 766 struct super_block *sb, ext4_group_t grp, 767 unsigned long ino, ext4_fsblk_t block, 768 const char *fmt, ...) 769 __releases(bitlock) 770 __acquires(bitlock) 771 { 772 struct va_format vaf; 773 va_list args; 774 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 775 776 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) 777 return; 778 779 trace_ext4_error(sb, function, line); 780 es->s_last_error_ino = cpu_to_le32(ino); 781 es->s_last_error_block = cpu_to_le64(block); 782 __save_error_info(sb, function, line); 783 784 if (ext4_error_ratelimit(sb)) { 785 va_start(args, fmt); 786 vaf.fmt = fmt; 787 vaf.va = &args; 788 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ", 789 sb->s_id, function, line, grp); 790 if (ino) 791 printk(KERN_CONT "inode %lu: ", ino); 792 if (block) 793 printk(KERN_CONT "block %llu:", 794 (unsigned long long) block); 795 printk(KERN_CONT "%pV\n", &vaf); 796 va_end(args); 797 } 798 799 if (test_opt(sb, WARN_ON_ERROR)) 800 WARN_ON_ONCE(1); 801 802 if (test_opt(sb, ERRORS_CONT)) { 803 ext4_commit_super(sb, 0); 804 return; 805 } 806 807 ext4_unlock_group(sb, grp); 808 ext4_commit_super(sb, 1); 809 ext4_handle_error(sb); 810 /* 811 * We only get here in the ERRORS_RO case; relocking the group 812 * may be dangerous, but nothing bad will happen since the 813 * filesystem will have already been marked read/only and the 814 * journal has been aborted. We return 1 as a hint to callers 815 * who might what to use the return value from 816 * ext4_grp_locked_error() to distinguish between the 817 * ERRORS_CONT and ERRORS_RO case, and perhaps return more 818 * aggressively from the ext4 function in question, with a 819 * more appropriate error code. 820 */ 821 ext4_lock_group(sb, grp); 822 return; 823 } 824 825 void ext4_mark_group_bitmap_corrupted(struct super_block *sb, 826 ext4_group_t group, 827 unsigned int flags) 828 { 829 struct ext4_sb_info *sbi = EXT4_SB(sb); 830 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 831 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 832 int ret; 833 834 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) { 835 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, 836 &grp->bb_state); 837 if (!ret) 838 percpu_counter_sub(&sbi->s_freeclusters_counter, 839 grp->bb_free); 840 } 841 842 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) { 843 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, 844 &grp->bb_state); 845 if (!ret && gdp) { 846 int count; 847 848 count = ext4_free_inodes_count(sb, gdp); 849 percpu_counter_sub(&sbi->s_freeinodes_counter, 850 count); 851 } 852 } 853 } 854 855 void ext4_update_dynamic_rev(struct super_block *sb) 856 { 857 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 858 859 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV) 860 return; 861 862 ext4_warning(sb, 863 "updating to rev %d because of new feature flag, " 864 "running e2fsck is recommended", 865 EXT4_DYNAMIC_REV); 866 867 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO); 868 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE); 869 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV); 870 /* leave es->s_feature_*compat flags alone */ 871 /* es->s_uuid will be set by e2fsck if empty */ 872 873 /* 874 * The rest of the superblock fields should be zero, and if not it 875 * means they are likely already in use, so leave them alone. We 876 * can leave it up to e2fsck to clean up any inconsistencies there. 877 */ 878 } 879 880 /* 881 * Open the external journal device 882 */ 883 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb) 884 { 885 struct block_device *bdev; 886 char b[BDEVNAME_SIZE]; 887 888 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb); 889 if (IS_ERR(bdev)) 890 goto fail; 891 return bdev; 892 893 fail: 894 ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld", 895 __bdevname(dev, b), PTR_ERR(bdev)); 896 return NULL; 897 } 898 899 /* 900 * Release the journal device 901 */ 902 static void ext4_blkdev_put(struct block_device *bdev) 903 { 904 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 905 } 906 907 static void ext4_blkdev_remove(struct ext4_sb_info *sbi) 908 { 909 struct block_device *bdev; 910 bdev = sbi->journal_bdev; 911 if (bdev) { 912 ext4_blkdev_put(bdev); 913 sbi->journal_bdev = NULL; 914 } 915 } 916 917 static inline struct inode *orphan_list_entry(struct list_head *l) 918 { 919 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode; 920 } 921 922 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi) 923 { 924 struct list_head *l; 925 926 ext4_msg(sb, KERN_ERR, "sb orphan head is %d", 927 le32_to_cpu(sbi->s_es->s_last_orphan)); 928 929 printk(KERN_ERR "sb_info orphan list:\n"); 930 list_for_each(l, &sbi->s_orphan) { 931 struct inode *inode = orphan_list_entry(l); 932 printk(KERN_ERR " " 933 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n", 934 inode->i_sb->s_id, inode->i_ino, inode, 935 inode->i_mode, inode->i_nlink, 936 NEXT_ORPHAN(inode)); 937 } 938 } 939 940 #ifdef CONFIG_QUOTA 941 static int ext4_quota_off(struct super_block *sb, int type); 942 943 static inline void ext4_quota_off_umount(struct super_block *sb) 944 { 945 int type; 946 947 /* Use our quota_off function to clear inode flags etc. */ 948 for (type = 0; type < EXT4_MAXQUOTAS; type++) 949 ext4_quota_off(sb, type); 950 } 951 952 /* 953 * This is a helper function which is used in the mount/remount 954 * codepaths (which holds s_umount) to fetch the quota file name. 955 */ 956 static inline char *get_qf_name(struct super_block *sb, 957 struct ext4_sb_info *sbi, 958 int type) 959 { 960 return rcu_dereference_protected(sbi->s_qf_names[type], 961 lockdep_is_held(&sb->s_umount)); 962 } 963 #else 964 static inline void ext4_quota_off_umount(struct super_block *sb) 965 { 966 } 967 #endif 968 969 static void ext4_put_super(struct super_block *sb) 970 { 971 struct ext4_sb_info *sbi = EXT4_SB(sb); 972 struct ext4_super_block *es = sbi->s_es; 973 int aborted = 0; 974 int i, err; 975 976 ext4_unregister_li_request(sb); 977 ext4_quota_off_umount(sb); 978 979 destroy_workqueue(sbi->rsv_conversion_wq); 980 981 if (sbi->s_journal) { 982 aborted = is_journal_aborted(sbi->s_journal); 983 err = jbd2_journal_destroy(sbi->s_journal); 984 sbi->s_journal = NULL; 985 if ((err < 0) && !aborted) 986 ext4_abort(sb, "Couldn't clean up the journal"); 987 } 988 989 ext4_unregister_sysfs(sb); 990 ext4_es_unregister_shrinker(sbi); 991 del_timer_sync(&sbi->s_err_report); 992 ext4_release_system_zone(sb); 993 ext4_mb_release(sb); 994 ext4_ext_release(sb); 995 996 if (!sb_rdonly(sb) && !aborted) { 997 ext4_clear_feature_journal_needs_recovery(sb); 998 es->s_state = cpu_to_le16(sbi->s_mount_state); 999 } 1000 if (!sb_rdonly(sb)) 1001 ext4_commit_super(sb, 1); 1002 1003 for (i = 0; i < sbi->s_gdb_count; i++) 1004 brelse(sbi->s_group_desc[i]); 1005 kvfree(sbi->s_group_desc); 1006 kvfree(sbi->s_flex_groups); 1007 percpu_counter_destroy(&sbi->s_freeclusters_counter); 1008 percpu_counter_destroy(&sbi->s_freeinodes_counter); 1009 percpu_counter_destroy(&sbi->s_dirs_counter); 1010 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 1011 percpu_free_rwsem(&sbi->s_journal_flag_rwsem); 1012 #ifdef CONFIG_QUOTA 1013 for (i = 0; i < EXT4_MAXQUOTAS; i++) 1014 kfree(get_qf_name(sb, sbi, i)); 1015 #endif 1016 1017 /* Debugging code just in case the in-memory inode orphan list 1018 * isn't empty. The on-disk one can be non-empty if we've 1019 * detected an error and taken the fs readonly, but the 1020 * in-memory list had better be clean by this point. */ 1021 if (!list_empty(&sbi->s_orphan)) 1022 dump_orphan_list(sb, sbi); 1023 J_ASSERT(list_empty(&sbi->s_orphan)); 1024 1025 sync_blockdev(sb->s_bdev); 1026 invalidate_bdev(sb->s_bdev); 1027 if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) { 1028 /* 1029 * Invalidate the journal device's buffers. We don't want them 1030 * floating about in memory - the physical journal device may 1031 * hotswapped, and it breaks the `ro-after' testing code. 1032 */ 1033 sync_blockdev(sbi->journal_bdev); 1034 invalidate_bdev(sbi->journal_bdev); 1035 ext4_blkdev_remove(sbi); 1036 } 1037 1038 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 1039 sbi->s_ea_inode_cache = NULL; 1040 1041 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 1042 sbi->s_ea_block_cache = NULL; 1043 1044 if (sbi->s_mmp_tsk) 1045 kthread_stop(sbi->s_mmp_tsk); 1046 brelse(sbi->s_sbh); 1047 sb->s_fs_info = NULL; 1048 /* 1049 * Now that we are completely done shutting down the 1050 * superblock, we need to actually destroy the kobject. 1051 */ 1052 kobject_put(&sbi->s_kobj); 1053 wait_for_completion(&sbi->s_kobj_unregister); 1054 if (sbi->s_chksum_driver) 1055 crypto_free_shash(sbi->s_chksum_driver); 1056 kfree(sbi->s_blockgroup_lock); 1057 fs_put_dax(sbi->s_daxdev); 1058 #ifdef CONFIG_UNICODE 1059 utf8_unload(sbi->s_encoding); 1060 #endif 1061 kfree(sbi); 1062 } 1063 1064 static struct kmem_cache *ext4_inode_cachep; 1065 1066 /* 1067 * Called inside transaction, so use GFP_NOFS 1068 */ 1069 static struct inode *ext4_alloc_inode(struct super_block *sb) 1070 { 1071 struct ext4_inode_info *ei; 1072 1073 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS); 1074 if (!ei) 1075 return NULL; 1076 1077 inode_set_iversion(&ei->vfs_inode, 1); 1078 spin_lock_init(&ei->i_raw_lock); 1079 INIT_LIST_HEAD(&ei->i_prealloc_list); 1080 spin_lock_init(&ei->i_prealloc_lock); 1081 ext4_es_init_tree(&ei->i_es_tree); 1082 rwlock_init(&ei->i_es_lock); 1083 INIT_LIST_HEAD(&ei->i_es_list); 1084 ei->i_es_all_nr = 0; 1085 ei->i_es_shk_nr = 0; 1086 ei->i_es_shrink_lblk = 0; 1087 ei->i_reserved_data_blocks = 0; 1088 ei->i_da_metadata_calc_len = 0; 1089 ei->i_da_metadata_calc_last_lblock = 0; 1090 spin_lock_init(&(ei->i_block_reservation_lock)); 1091 ext4_init_pending_tree(&ei->i_pending_tree); 1092 #ifdef CONFIG_QUOTA 1093 ei->i_reserved_quota = 0; 1094 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot)); 1095 #endif 1096 ei->jinode = NULL; 1097 INIT_LIST_HEAD(&ei->i_rsv_conversion_list); 1098 spin_lock_init(&ei->i_completed_io_lock); 1099 ei->i_sync_tid = 0; 1100 ei->i_datasync_tid = 0; 1101 atomic_set(&ei->i_unwritten, 0); 1102 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work); 1103 return &ei->vfs_inode; 1104 } 1105 1106 static int ext4_drop_inode(struct inode *inode) 1107 { 1108 int drop = generic_drop_inode(inode); 1109 1110 if (!drop) 1111 drop = fscrypt_drop_inode(inode); 1112 1113 trace_ext4_drop_inode(inode, drop); 1114 return drop; 1115 } 1116 1117 static void ext4_free_in_core_inode(struct inode *inode) 1118 { 1119 fscrypt_free_inode(inode); 1120 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode)); 1121 } 1122 1123 static void ext4_destroy_inode(struct inode *inode) 1124 { 1125 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 1126 ext4_msg(inode->i_sb, KERN_ERR, 1127 "Inode %lu (%p): orphan list check failed!", 1128 inode->i_ino, EXT4_I(inode)); 1129 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4, 1130 EXT4_I(inode), sizeof(struct ext4_inode_info), 1131 true); 1132 dump_stack(); 1133 } 1134 } 1135 1136 static void init_once(void *foo) 1137 { 1138 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo; 1139 1140 INIT_LIST_HEAD(&ei->i_orphan); 1141 init_rwsem(&ei->xattr_sem); 1142 init_rwsem(&ei->i_data_sem); 1143 init_rwsem(&ei->i_mmap_sem); 1144 inode_init_once(&ei->vfs_inode); 1145 } 1146 1147 static int __init init_inodecache(void) 1148 { 1149 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache", 1150 sizeof(struct ext4_inode_info), 0, 1151 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| 1152 SLAB_ACCOUNT), 1153 offsetof(struct ext4_inode_info, i_data), 1154 sizeof_field(struct ext4_inode_info, i_data), 1155 init_once); 1156 if (ext4_inode_cachep == NULL) 1157 return -ENOMEM; 1158 return 0; 1159 } 1160 1161 static void destroy_inodecache(void) 1162 { 1163 /* 1164 * Make sure all delayed rcu free inodes are flushed before we 1165 * destroy cache. 1166 */ 1167 rcu_barrier(); 1168 kmem_cache_destroy(ext4_inode_cachep); 1169 } 1170 1171 void ext4_clear_inode(struct inode *inode) 1172 { 1173 invalidate_inode_buffers(inode); 1174 clear_inode(inode); 1175 dquot_drop(inode); 1176 ext4_discard_preallocations(inode); 1177 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1178 if (EXT4_I(inode)->jinode) { 1179 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode), 1180 EXT4_I(inode)->jinode); 1181 jbd2_free_inode(EXT4_I(inode)->jinode); 1182 EXT4_I(inode)->jinode = NULL; 1183 } 1184 fscrypt_put_encryption_info(inode); 1185 fsverity_cleanup_inode(inode); 1186 } 1187 1188 static struct inode *ext4_nfs_get_inode(struct super_block *sb, 1189 u64 ino, u32 generation) 1190 { 1191 struct inode *inode; 1192 1193 /* 1194 * Currently we don't know the generation for parent directory, so 1195 * a generation of 0 means "accept any" 1196 */ 1197 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE); 1198 if (IS_ERR(inode)) 1199 return ERR_CAST(inode); 1200 if (generation && inode->i_generation != generation) { 1201 iput(inode); 1202 return ERR_PTR(-ESTALE); 1203 } 1204 1205 return inode; 1206 } 1207 1208 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid, 1209 int fh_len, int fh_type) 1210 { 1211 return generic_fh_to_dentry(sb, fid, fh_len, fh_type, 1212 ext4_nfs_get_inode); 1213 } 1214 1215 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid, 1216 int fh_len, int fh_type) 1217 { 1218 return generic_fh_to_parent(sb, fid, fh_len, fh_type, 1219 ext4_nfs_get_inode); 1220 } 1221 1222 static int ext4_nfs_commit_metadata(struct inode *inode) 1223 { 1224 struct writeback_control wbc = { 1225 .sync_mode = WB_SYNC_ALL 1226 }; 1227 1228 trace_ext4_nfs_commit_metadata(inode); 1229 return ext4_write_inode(inode, &wbc); 1230 } 1231 1232 /* 1233 * Try to release metadata pages (indirect blocks, directories) which are 1234 * mapped via the block device. Since these pages could have journal heads 1235 * which would prevent try_to_free_buffers() from freeing them, we must use 1236 * jbd2 layer's try_to_free_buffers() function to release them. 1237 */ 1238 static int bdev_try_to_free_page(struct super_block *sb, struct page *page, 1239 gfp_t wait) 1240 { 1241 journal_t *journal = EXT4_SB(sb)->s_journal; 1242 1243 WARN_ON(PageChecked(page)); 1244 if (!page_has_buffers(page)) 1245 return 0; 1246 if (journal) 1247 return jbd2_journal_try_to_free_buffers(journal, page, 1248 wait & ~__GFP_DIRECT_RECLAIM); 1249 return try_to_free_buffers(page); 1250 } 1251 1252 #ifdef CONFIG_FS_ENCRYPTION 1253 static int ext4_get_context(struct inode *inode, void *ctx, size_t len) 1254 { 1255 return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, 1256 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len); 1257 } 1258 1259 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len, 1260 void *fs_data) 1261 { 1262 handle_t *handle = fs_data; 1263 int res, res2, credits, retries = 0; 1264 1265 /* 1266 * Encrypting the root directory is not allowed because e2fsck expects 1267 * lost+found to exist and be unencrypted, and encrypting the root 1268 * directory would imply encrypting the lost+found directory as well as 1269 * the filename "lost+found" itself. 1270 */ 1271 if (inode->i_ino == EXT4_ROOT_INO) 1272 return -EPERM; 1273 1274 if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode))) 1275 return -EINVAL; 1276 1277 res = ext4_convert_inline_data(inode); 1278 if (res) 1279 return res; 1280 1281 /* 1282 * If a journal handle was specified, then the encryption context is 1283 * being set on a new inode via inheritance and is part of a larger 1284 * transaction to create the inode. Otherwise the encryption context is 1285 * being set on an existing inode in its own transaction. Only in the 1286 * latter case should the "retry on ENOSPC" logic be used. 1287 */ 1288 1289 if (handle) { 1290 res = ext4_xattr_set_handle(handle, inode, 1291 EXT4_XATTR_INDEX_ENCRYPTION, 1292 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1293 ctx, len, 0); 1294 if (!res) { 1295 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1296 ext4_clear_inode_state(inode, 1297 EXT4_STATE_MAY_INLINE_DATA); 1298 /* 1299 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1300 * S_DAX may be disabled 1301 */ 1302 ext4_set_inode_flags(inode); 1303 } 1304 return res; 1305 } 1306 1307 res = dquot_initialize(inode); 1308 if (res) 1309 return res; 1310 retry: 1311 res = ext4_xattr_set_credits(inode, len, false /* is_create */, 1312 &credits); 1313 if (res) 1314 return res; 1315 1316 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); 1317 if (IS_ERR(handle)) 1318 return PTR_ERR(handle); 1319 1320 res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION, 1321 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, 1322 ctx, len, 0); 1323 if (!res) { 1324 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); 1325 /* 1326 * Update inode->i_flags - S_ENCRYPTED will be enabled, 1327 * S_DAX may be disabled 1328 */ 1329 ext4_set_inode_flags(inode); 1330 res = ext4_mark_inode_dirty(handle, inode); 1331 if (res) 1332 EXT4_ERROR_INODE(inode, "Failed to mark inode dirty"); 1333 } 1334 res2 = ext4_journal_stop(handle); 1335 1336 if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 1337 goto retry; 1338 if (!res) 1339 res = res2; 1340 return res; 1341 } 1342 1343 static bool ext4_dummy_context(struct inode *inode) 1344 { 1345 return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb)); 1346 } 1347 1348 static const struct fscrypt_operations ext4_cryptops = { 1349 .key_prefix = "ext4:", 1350 .get_context = ext4_get_context, 1351 .set_context = ext4_set_context, 1352 .dummy_context = ext4_dummy_context, 1353 .empty_dir = ext4_empty_dir, 1354 .max_namelen = EXT4_NAME_LEN, 1355 }; 1356 #endif 1357 1358 #ifdef CONFIG_QUOTA 1359 static const char * const quotatypes[] = INITQFNAMES; 1360 #define QTYPE2NAME(t) (quotatypes[t]) 1361 1362 static int ext4_write_dquot(struct dquot *dquot); 1363 static int ext4_acquire_dquot(struct dquot *dquot); 1364 static int ext4_release_dquot(struct dquot *dquot); 1365 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1366 static int ext4_write_info(struct super_block *sb, int type); 1367 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1368 const struct path *path); 1369 static int ext4_quota_on_mount(struct super_block *sb, int type); 1370 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1371 size_t len, loff_t off); 1372 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1373 const char *data, size_t len, loff_t off); 1374 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1375 unsigned int flags); 1376 static int ext4_enable_quotas(struct super_block *sb); 1377 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid); 1378 1379 static struct dquot **ext4_get_dquots(struct inode *inode) 1380 { 1381 return EXT4_I(inode)->i_dquot; 1382 } 1383 1384 static const struct dquot_operations ext4_quota_operations = { 1385 .get_reserved_space = ext4_get_reserved_space, 1386 .write_dquot = ext4_write_dquot, 1387 .acquire_dquot = ext4_acquire_dquot, 1388 .release_dquot = ext4_release_dquot, 1389 .mark_dirty = ext4_mark_dquot_dirty, 1390 .write_info = ext4_write_info, 1391 .alloc_dquot = dquot_alloc, 1392 .destroy_dquot = dquot_destroy, 1393 .get_projid = ext4_get_projid, 1394 .get_inode_usage = ext4_get_inode_usage, 1395 .get_next_id = ext4_get_next_id, 1396 }; 1397 1398 static const struct quotactl_ops ext4_qctl_operations = { 1399 .quota_on = ext4_quota_on, 1400 .quota_off = ext4_quota_off, 1401 .quota_sync = dquot_quota_sync, 1402 .get_state = dquot_get_state, 1403 .set_info = dquot_set_dqinfo, 1404 .get_dqblk = dquot_get_dqblk, 1405 .set_dqblk = dquot_set_dqblk, 1406 .get_nextdqblk = dquot_get_next_dqblk, 1407 }; 1408 #endif 1409 1410 static const struct super_operations ext4_sops = { 1411 .alloc_inode = ext4_alloc_inode, 1412 .free_inode = ext4_free_in_core_inode, 1413 .destroy_inode = ext4_destroy_inode, 1414 .write_inode = ext4_write_inode, 1415 .dirty_inode = ext4_dirty_inode, 1416 .drop_inode = ext4_drop_inode, 1417 .evict_inode = ext4_evict_inode, 1418 .put_super = ext4_put_super, 1419 .sync_fs = ext4_sync_fs, 1420 .freeze_fs = ext4_freeze, 1421 .unfreeze_fs = ext4_unfreeze, 1422 .statfs = ext4_statfs, 1423 .remount_fs = ext4_remount, 1424 .show_options = ext4_show_options, 1425 #ifdef CONFIG_QUOTA 1426 .quota_read = ext4_quota_read, 1427 .quota_write = ext4_quota_write, 1428 .get_dquots = ext4_get_dquots, 1429 #endif 1430 .bdev_try_to_free_page = bdev_try_to_free_page, 1431 }; 1432 1433 static const struct export_operations ext4_export_ops = { 1434 .fh_to_dentry = ext4_fh_to_dentry, 1435 .fh_to_parent = ext4_fh_to_parent, 1436 .get_parent = ext4_get_parent, 1437 .commit_metadata = ext4_nfs_commit_metadata, 1438 }; 1439 1440 enum { 1441 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1442 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro, 1443 Opt_nouid32, Opt_debug, Opt_removed, 1444 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, 1445 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1446 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1447 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1448 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1449 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1450 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota, 1451 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota, 1452 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1453 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax, 1454 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1455 Opt_nowarn_on_error, Opt_mblk_io_submit, 1456 Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize, 1457 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1458 Opt_inode_readahead_blks, Opt_journal_ioprio, 1459 Opt_dioread_nolock, Opt_dioread_lock, 1460 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1461 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1462 }; 1463 1464 static const match_table_t tokens = { 1465 {Opt_bsd_df, "bsddf"}, 1466 {Opt_minix_df, "minixdf"}, 1467 {Opt_grpid, "grpid"}, 1468 {Opt_grpid, "bsdgroups"}, 1469 {Opt_nogrpid, "nogrpid"}, 1470 {Opt_nogrpid, "sysvgroups"}, 1471 {Opt_resgid, "resgid=%u"}, 1472 {Opt_resuid, "resuid=%u"}, 1473 {Opt_sb, "sb=%u"}, 1474 {Opt_err_cont, "errors=continue"}, 1475 {Opt_err_panic, "errors=panic"}, 1476 {Opt_err_ro, "errors=remount-ro"}, 1477 {Opt_nouid32, "nouid32"}, 1478 {Opt_debug, "debug"}, 1479 {Opt_removed, "oldalloc"}, 1480 {Opt_removed, "orlov"}, 1481 {Opt_user_xattr, "user_xattr"}, 1482 {Opt_nouser_xattr, "nouser_xattr"}, 1483 {Opt_acl, "acl"}, 1484 {Opt_noacl, "noacl"}, 1485 {Opt_noload, "norecovery"}, 1486 {Opt_noload, "noload"}, 1487 {Opt_removed, "nobh"}, 1488 {Opt_removed, "bh"}, 1489 {Opt_commit, "commit=%u"}, 1490 {Opt_min_batch_time, "min_batch_time=%u"}, 1491 {Opt_max_batch_time, "max_batch_time=%u"}, 1492 {Opt_journal_dev, "journal_dev=%u"}, 1493 {Opt_journal_path, "journal_path=%s"}, 1494 {Opt_journal_checksum, "journal_checksum"}, 1495 {Opt_nojournal_checksum, "nojournal_checksum"}, 1496 {Opt_journal_async_commit, "journal_async_commit"}, 1497 {Opt_abort, "abort"}, 1498 {Opt_data_journal, "data=journal"}, 1499 {Opt_data_ordered, "data=ordered"}, 1500 {Opt_data_writeback, "data=writeback"}, 1501 {Opt_data_err_abort, "data_err=abort"}, 1502 {Opt_data_err_ignore, "data_err=ignore"}, 1503 {Opt_offusrjquota, "usrjquota="}, 1504 {Opt_usrjquota, "usrjquota=%s"}, 1505 {Opt_offgrpjquota, "grpjquota="}, 1506 {Opt_grpjquota, "grpjquota=%s"}, 1507 {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, 1508 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, 1509 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, 1510 {Opt_grpquota, "grpquota"}, 1511 {Opt_noquota, "noquota"}, 1512 {Opt_quota, "quota"}, 1513 {Opt_usrquota, "usrquota"}, 1514 {Opt_prjquota, "prjquota"}, 1515 {Opt_barrier, "barrier=%u"}, 1516 {Opt_barrier, "barrier"}, 1517 {Opt_nobarrier, "nobarrier"}, 1518 {Opt_i_version, "i_version"}, 1519 {Opt_dax, "dax"}, 1520 {Opt_stripe, "stripe=%u"}, 1521 {Opt_delalloc, "delalloc"}, 1522 {Opt_warn_on_error, "warn_on_error"}, 1523 {Opt_nowarn_on_error, "nowarn_on_error"}, 1524 {Opt_lazytime, "lazytime"}, 1525 {Opt_nolazytime, "nolazytime"}, 1526 {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"}, 1527 {Opt_nodelalloc, "nodelalloc"}, 1528 {Opt_removed, "mblk_io_submit"}, 1529 {Opt_removed, "nomblk_io_submit"}, 1530 {Opt_block_validity, "block_validity"}, 1531 {Opt_noblock_validity, "noblock_validity"}, 1532 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, 1533 {Opt_journal_ioprio, "journal_ioprio=%u"}, 1534 {Opt_auto_da_alloc, "auto_da_alloc=%u"}, 1535 {Opt_auto_da_alloc, "auto_da_alloc"}, 1536 {Opt_noauto_da_alloc, "noauto_da_alloc"}, 1537 {Opt_dioread_nolock, "dioread_nolock"}, 1538 {Opt_dioread_lock, "dioread_lock"}, 1539 {Opt_discard, "discard"}, 1540 {Opt_nodiscard, "nodiscard"}, 1541 {Opt_init_itable, "init_itable=%u"}, 1542 {Opt_init_itable, "init_itable"}, 1543 {Opt_noinit_itable, "noinit_itable"}, 1544 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, 1545 {Opt_test_dummy_encryption, "test_dummy_encryption"}, 1546 {Opt_nombcache, "nombcache"}, 1547 {Opt_nombcache, "no_mbcache"}, /* for backward compatibility */ 1548 {Opt_removed, "check=none"}, /* mount option from ext2/3 */ 1549 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */ 1550 {Opt_removed, "reservation"}, /* mount option from ext2/3 */ 1551 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */ 1552 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */ 1553 {Opt_err, NULL}, 1554 }; 1555 1556 static ext4_fsblk_t get_sb_block(void **data) 1557 { 1558 ext4_fsblk_t sb_block; 1559 char *options = (char *) *data; 1560 1561 if (!options || strncmp(options, "sb=", 3) != 0) 1562 return 1; /* Default location */ 1563 1564 options += 3; 1565 /* TODO: use simple_strtoll with >32bit ext4 */ 1566 sb_block = simple_strtoul(options, &options, 0); 1567 if (*options && *options != ',') { 1568 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n", 1569 (char *) *data); 1570 return 1; 1571 } 1572 if (*options == ',') 1573 options++; 1574 *data = (void *) options; 1575 1576 return sb_block; 1577 } 1578 1579 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1580 static const char deprecated_msg[] = 1581 "Mount option \"%s\" will be removed by %s\n" 1582 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1583 1584 #ifdef CONFIG_QUOTA 1585 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args) 1586 { 1587 struct ext4_sb_info *sbi = EXT4_SB(sb); 1588 char *qname, *old_qname = get_qf_name(sb, sbi, qtype); 1589 int ret = -1; 1590 1591 if (sb_any_quota_loaded(sb) && !old_qname) { 1592 ext4_msg(sb, KERN_ERR, 1593 "Cannot change journaled " 1594 "quota options when quota turned on"); 1595 return -1; 1596 } 1597 if (ext4_has_feature_quota(sb)) { 1598 ext4_msg(sb, KERN_INFO, "Journaled quota options " 1599 "ignored when QUOTA feature is enabled"); 1600 return 1; 1601 } 1602 qname = match_strdup(args); 1603 if (!qname) { 1604 ext4_msg(sb, KERN_ERR, 1605 "Not enough memory for storing quotafile name"); 1606 return -1; 1607 } 1608 if (old_qname) { 1609 if (strcmp(old_qname, qname) == 0) 1610 ret = 1; 1611 else 1612 ext4_msg(sb, KERN_ERR, 1613 "%s quota file already specified", 1614 QTYPE2NAME(qtype)); 1615 goto errout; 1616 } 1617 if (strchr(qname, '/')) { 1618 ext4_msg(sb, KERN_ERR, 1619 "quotafile must be on filesystem root"); 1620 goto errout; 1621 } 1622 rcu_assign_pointer(sbi->s_qf_names[qtype], qname); 1623 set_opt(sb, QUOTA); 1624 return 1; 1625 errout: 1626 kfree(qname); 1627 return ret; 1628 } 1629 1630 static int clear_qf_name(struct super_block *sb, int qtype) 1631 { 1632 1633 struct ext4_sb_info *sbi = EXT4_SB(sb); 1634 char *old_qname = get_qf_name(sb, sbi, qtype); 1635 1636 if (sb_any_quota_loaded(sb) && old_qname) { 1637 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options" 1638 " when quota turned on"); 1639 return -1; 1640 } 1641 rcu_assign_pointer(sbi->s_qf_names[qtype], NULL); 1642 synchronize_rcu(); 1643 kfree(old_qname); 1644 return 1; 1645 } 1646 #endif 1647 1648 #define MOPT_SET 0x0001 1649 #define MOPT_CLEAR 0x0002 1650 #define MOPT_NOSUPPORT 0x0004 1651 #define MOPT_EXPLICIT 0x0008 1652 #define MOPT_CLEAR_ERR 0x0010 1653 #define MOPT_GTE0 0x0020 1654 #ifdef CONFIG_QUOTA 1655 #define MOPT_Q 0 1656 #define MOPT_QFMT 0x0040 1657 #else 1658 #define MOPT_Q MOPT_NOSUPPORT 1659 #define MOPT_QFMT MOPT_NOSUPPORT 1660 #endif 1661 #define MOPT_DATAJ 0x0080 1662 #define MOPT_NO_EXT2 0x0100 1663 #define MOPT_NO_EXT3 0x0200 1664 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1665 #define MOPT_STRING 0x0400 1666 1667 static const struct mount_opts { 1668 int token; 1669 int mount_opt; 1670 int flags; 1671 } ext4_mount_opts[] = { 1672 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1673 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1674 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1675 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1676 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1677 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1678 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1679 MOPT_EXT4_ONLY | MOPT_SET}, 1680 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1681 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1682 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1683 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1684 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1685 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1686 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1687 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1688 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1689 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1690 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1691 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1692 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1693 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1694 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1695 EXT4_MOUNT_JOURNAL_CHECKSUM), 1696 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1697 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1698 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR}, 1699 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR}, 1700 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR}, 1701 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, 1702 MOPT_NO_EXT2}, 1703 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, 1704 MOPT_NO_EXT2}, 1705 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1706 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1707 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1708 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1709 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1710 {Opt_commit, 0, MOPT_GTE0}, 1711 {Opt_max_batch_time, 0, MOPT_GTE0}, 1712 {Opt_min_batch_time, 0, MOPT_GTE0}, 1713 {Opt_inode_readahead_blks, 0, MOPT_GTE0}, 1714 {Opt_init_itable, 0, MOPT_GTE0}, 1715 {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET}, 1716 {Opt_stripe, 0, MOPT_GTE0}, 1717 {Opt_resuid, 0, MOPT_GTE0}, 1718 {Opt_resgid, 0, MOPT_GTE0}, 1719 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1720 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING}, 1721 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1722 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1723 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1724 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, 1725 MOPT_NO_EXT2 | MOPT_DATAJ}, 1726 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1727 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR}, 1728 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1729 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1730 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR}, 1731 #else 1732 {Opt_acl, 0, MOPT_NOSUPPORT}, 1733 {Opt_noacl, 0, MOPT_NOSUPPORT}, 1734 #endif 1735 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1736 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1737 {Opt_debug_want_extra_isize, 0, MOPT_GTE0}, 1738 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1739 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1740 MOPT_SET | MOPT_Q}, 1741 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1742 MOPT_SET | MOPT_Q}, 1743 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1744 MOPT_SET | MOPT_Q}, 1745 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1746 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1747 MOPT_CLEAR | MOPT_Q}, 1748 {Opt_usrjquota, 0, MOPT_Q}, 1749 {Opt_grpjquota, 0, MOPT_Q}, 1750 {Opt_offusrjquota, 0, MOPT_Q}, 1751 {Opt_offgrpjquota, 0, MOPT_Q}, 1752 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT}, 1753 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT}, 1754 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT}, 1755 {Opt_max_dir_size_kb, 0, MOPT_GTE0}, 1756 {Opt_test_dummy_encryption, 0, MOPT_GTE0}, 1757 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1758 {Opt_err, 0, 0} 1759 }; 1760 1761 #ifdef CONFIG_UNICODE 1762 static const struct ext4_sb_encodings { 1763 __u16 magic; 1764 char *name; 1765 char *version; 1766 } ext4_sb_encoding_map[] = { 1767 {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"}, 1768 }; 1769 1770 static int ext4_sb_read_encoding(const struct ext4_super_block *es, 1771 const struct ext4_sb_encodings **encoding, 1772 __u16 *flags) 1773 { 1774 __u16 magic = le16_to_cpu(es->s_encoding); 1775 int i; 1776 1777 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1778 if (magic == ext4_sb_encoding_map[i].magic) 1779 break; 1780 1781 if (i >= ARRAY_SIZE(ext4_sb_encoding_map)) 1782 return -EINVAL; 1783 1784 *encoding = &ext4_sb_encoding_map[i]; 1785 *flags = le16_to_cpu(es->s_encoding_flags); 1786 1787 return 0; 1788 } 1789 #endif 1790 1791 static int handle_mount_opt(struct super_block *sb, char *opt, int token, 1792 substring_t *args, unsigned long *journal_devnum, 1793 unsigned int *journal_ioprio, int is_remount) 1794 { 1795 struct ext4_sb_info *sbi = EXT4_SB(sb); 1796 const struct mount_opts *m; 1797 kuid_t uid; 1798 kgid_t gid; 1799 int arg = 0; 1800 1801 #ifdef CONFIG_QUOTA 1802 if (token == Opt_usrjquota) 1803 return set_qf_name(sb, USRQUOTA, &args[0]); 1804 else if (token == Opt_grpjquota) 1805 return set_qf_name(sb, GRPQUOTA, &args[0]); 1806 else if (token == Opt_offusrjquota) 1807 return clear_qf_name(sb, USRQUOTA); 1808 else if (token == Opt_offgrpjquota) 1809 return clear_qf_name(sb, GRPQUOTA); 1810 #endif 1811 switch (token) { 1812 case Opt_noacl: 1813 case Opt_nouser_xattr: 1814 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5"); 1815 break; 1816 case Opt_sb: 1817 return 1; /* handled by get_sb_block() */ 1818 case Opt_removed: 1819 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt); 1820 return 1; 1821 case Opt_abort: 1822 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED; 1823 return 1; 1824 case Opt_i_version: 1825 sb->s_flags |= SB_I_VERSION; 1826 return 1; 1827 case Opt_lazytime: 1828 sb->s_flags |= SB_LAZYTIME; 1829 return 1; 1830 case Opt_nolazytime: 1831 sb->s_flags &= ~SB_LAZYTIME; 1832 return 1; 1833 } 1834 1835 for (m = ext4_mount_opts; m->token != Opt_err; m++) 1836 if (token == m->token) 1837 break; 1838 1839 if (m->token == Opt_err) { 1840 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" " 1841 "or missing value", opt); 1842 return -1; 1843 } 1844 1845 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 1846 ext4_msg(sb, KERN_ERR, 1847 "Mount option \"%s\" incompatible with ext2", opt); 1848 return -1; 1849 } 1850 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 1851 ext4_msg(sb, KERN_ERR, 1852 "Mount option \"%s\" incompatible with ext3", opt); 1853 return -1; 1854 } 1855 1856 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg)) 1857 return -1; 1858 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0)) 1859 return -1; 1860 if (m->flags & MOPT_EXPLICIT) { 1861 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 1862 set_opt2(sb, EXPLICIT_DELALLOC); 1863 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 1864 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM); 1865 } else 1866 return -1; 1867 } 1868 if (m->flags & MOPT_CLEAR_ERR) 1869 clear_opt(sb, ERRORS_MASK); 1870 if (token == Opt_noquota && sb_any_quota_loaded(sb)) { 1871 ext4_msg(sb, KERN_ERR, "Cannot change quota " 1872 "options when quota turned on"); 1873 return -1; 1874 } 1875 1876 if (m->flags & MOPT_NOSUPPORT) { 1877 ext4_msg(sb, KERN_ERR, "%s option not supported", opt); 1878 } else if (token == Opt_commit) { 1879 if (arg == 0) 1880 arg = JBD2_DEFAULT_MAX_COMMIT_AGE; 1881 else if (arg > INT_MAX / HZ) { 1882 ext4_msg(sb, KERN_ERR, 1883 "Invalid commit interval %d, " 1884 "must be smaller than %d", 1885 arg, INT_MAX / HZ); 1886 return -1; 1887 } 1888 sbi->s_commit_interval = HZ * arg; 1889 } else if (token == Opt_debug_want_extra_isize) { 1890 sbi->s_want_extra_isize = arg; 1891 } else if (token == Opt_max_batch_time) { 1892 sbi->s_max_batch_time = arg; 1893 } else if (token == Opt_min_batch_time) { 1894 sbi->s_min_batch_time = arg; 1895 } else if (token == Opt_inode_readahead_blks) { 1896 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) { 1897 ext4_msg(sb, KERN_ERR, 1898 "EXT4-fs: inode_readahead_blks must be " 1899 "0 or a power of 2 smaller than 2^31"); 1900 return -1; 1901 } 1902 sbi->s_inode_readahead_blks = arg; 1903 } else if (token == Opt_init_itable) { 1904 set_opt(sb, INIT_INODE_TABLE); 1905 if (!args->from) 1906 arg = EXT4_DEF_LI_WAIT_MULT; 1907 sbi->s_li_wait_mult = arg; 1908 } else if (token == Opt_max_dir_size_kb) { 1909 sbi->s_max_dir_size_kb = arg; 1910 } else if (token == Opt_stripe) { 1911 sbi->s_stripe = arg; 1912 } else if (token == Opt_resuid) { 1913 uid = make_kuid(current_user_ns(), arg); 1914 if (!uid_valid(uid)) { 1915 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg); 1916 return -1; 1917 } 1918 sbi->s_resuid = uid; 1919 } else if (token == Opt_resgid) { 1920 gid = make_kgid(current_user_ns(), arg); 1921 if (!gid_valid(gid)) { 1922 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg); 1923 return -1; 1924 } 1925 sbi->s_resgid = gid; 1926 } else if (token == Opt_journal_dev) { 1927 if (is_remount) { 1928 ext4_msg(sb, KERN_ERR, 1929 "Cannot specify journal on remount"); 1930 return -1; 1931 } 1932 *journal_devnum = arg; 1933 } else if (token == Opt_journal_path) { 1934 char *journal_path; 1935 struct inode *journal_inode; 1936 struct path path; 1937 int error; 1938 1939 if (is_remount) { 1940 ext4_msg(sb, KERN_ERR, 1941 "Cannot specify journal on remount"); 1942 return -1; 1943 } 1944 journal_path = match_strdup(&args[0]); 1945 if (!journal_path) { 1946 ext4_msg(sb, KERN_ERR, "error: could not dup " 1947 "journal device string"); 1948 return -1; 1949 } 1950 1951 error = kern_path(journal_path, LOOKUP_FOLLOW, &path); 1952 if (error) { 1953 ext4_msg(sb, KERN_ERR, "error: could not find " 1954 "journal device path: error %d", error); 1955 kfree(journal_path); 1956 return -1; 1957 } 1958 1959 journal_inode = d_inode(path.dentry); 1960 if (!S_ISBLK(journal_inode->i_mode)) { 1961 ext4_msg(sb, KERN_ERR, "error: journal path %s " 1962 "is not a block device", journal_path); 1963 path_put(&path); 1964 kfree(journal_path); 1965 return -1; 1966 } 1967 1968 *journal_devnum = new_encode_dev(journal_inode->i_rdev); 1969 path_put(&path); 1970 kfree(journal_path); 1971 } else if (token == Opt_journal_ioprio) { 1972 if (arg > 7) { 1973 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority" 1974 " (must be 0-7)"); 1975 return -1; 1976 } 1977 *journal_ioprio = 1978 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg); 1979 } else if (token == Opt_test_dummy_encryption) { 1980 #ifdef CONFIG_FS_ENCRYPTION 1981 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION; 1982 ext4_msg(sb, KERN_WARNING, 1983 "Test dummy encryption mode enabled"); 1984 #else 1985 ext4_msg(sb, KERN_WARNING, 1986 "Test dummy encryption mount option ignored"); 1987 #endif 1988 } else if (m->flags & MOPT_DATAJ) { 1989 if (is_remount) { 1990 if (!sbi->s_journal) 1991 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option"); 1992 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) { 1993 ext4_msg(sb, KERN_ERR, 1994 "Cannot change data mode on remount"); 1995 return -1; 1996 } 1997 } else { 1998 clear_opt(sb, DATA_FLAGS); 1999 sbi->s_mount_opt |= m->mount_opt; 2000 } 2001 #ifdef CONFIG_QUOTA 2002 } else if (m->flags & MOPT_QFMT) { 2003 if (sb_any_quota_loaded(sb) && 2004 sbi->s_jquota_fmt != m->mount_opt) { 2005 ext4_msg(sb, KERN_ERR, "Cannot change journaled " 2006 "quota options when quota turned on"); 2007 return -1; 2008 } 2009 if (ext4_has_feature_quota(sb)) { 2010 ext4_msg(sb, KERN_INFO, 2011 "Quota format mount options ignored " 2012 "when QUOTA feature is enabled"); 2013 return 1; 2014 } 2015 sbi->s_jquota_fmt = m->mount_opt; 2016 #endif 2017 } else if (token == Opt_dax) { 2018 #ifdef CONFIG_FS_DAX 2019 ext4_msg(sb, KERN_WARNING, 2020 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 2021 sbi->s_mount_opt |= m->mount_opt; 2022 #else 2023 ext4_msg(sb, KERN_INFO, "dax option not supported"); 2024 return -1; 2025 #endif 2026 } else if (token == Opt_data_err_abort) { 2027 sbi->s_mount_opt |= m->mount_opt; 2028 } else if (token == Opt_data_err_ignore) { 2029 sbi->s_mount_opt &= ~m->mount_opt; 2030 } else { 2031 if (!args->from) 2032 arg = 1; 2033 if (m->flags & MOPT_CLEAR) 2034 arg = !arg; 2035 else if (unlikely(!(m->flags & MOPT_SET))) { 2036 ext4_msg(sb, KERN_WARNING, 2037 "buggy handling of option %s", opt); 2038 WARN_ON(1); 2039 return -1; 2040 } 2041 if (arg != 0) 2042 sbi->s_mount_opt |= m->mount_opt; 2043 else 2044 sbi->s_mount_opt &= ~m->mount_opt; 2045 } 2046 return 1; 2047 } 2048 2049 static int parse_options(char *options, struct super_block *sb, 2050 unsigned long *journal_devnum, 2051 unsigned int *journal_ioprio, 2052 int is_remount) 2053 { 2054 struct ext4_sb_info *sbi = EXT4_SB(sb); 2055 char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name; 2056 substring_t args[MAX_OPT_ARGS]; 2057 int token; 2058 2059 if (!options) 2060 return 1; 2061 2062 while ((p = strsep(&options, ",")) != NULL) { 2063 if (!*p) 2064 continue; 2065 /* 2066 * Initialize args struct so we know whether arg was 2067 * found; some options take optional arguments. 2068 */ 2069 args[0].to = args[0].from = NULL; 2070 token = match_token(p, tokens, args); 2071 if (handle_mount_opt(sb, p, token, args, journal_devnum, 2072 journal_ioprio, is_remount) < 0) 2073 return 0; 2074 } 2075 #ifdef CONFIG_QUOTA 2076 /* 2077 * We do the test below only for project quotas. 'usrquota' and 2078 * 'grpquota' mount options are allowed even without quota feature 2079 * to support legacy quotas in quota files. 2080 */ 2081 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) { 2082 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. " 2083 "Cannot enable project quota enforcement."); 2084 return 0; 2085 } 2086 usr_qf_name = get_qf_name(sb, sbi, USRQUOTA); 2087 grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA); 2088 if (usr_qf_name || grp_qf_name) { 2089 if (test_opt(sb, USRQUOTA) && usr_qf_name) 2090 clear_opt(sb, USRQUOTA); 2091 2092 if (test_opt(sb, GRPQUOTA) && grp_qf_name) 2093 clear_opt(sb, GRPQUOTA); 2094 2095 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) { 2096 ext4_msg(sb, KERN_ERR, "old and new quota " 2097 "format mixing"); 2098 return 0; 2099 } 2100 2101 if (!sbi->s_jquota_fmt) { 2102 ext4_msg(sb, KERN_ERR, "journaled quota format " 2103 "not specified"); 2104 return 0; 2105 } 2106 } 2107 #endif 2108 if (test_opt(sb, DIOREAD_NOLOCK)) { 2109 int blocksize = 2110 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size); 2111 2112 if (blocksize < PAGE_SIZE) { 2113 ext4_msg(sb, KERN_ERR, "can't mount with " 2114 "dioread_nolock if block size != PAGE_SIZE"); 2115 return 0; 2116 } 2117 } 2118 return 1; 2119 } 2120 2121 static inline void ext4_show_quota_options(struct seq_file *seq, 2122 struct super_block *sb) 2123 { 2124 #if defined(CONFIG_QUOTA) 2125 struct ext4_sb_info *sbi = EXT4_SB(sb); 2126 char *usr_qf_name, *grp_qf_name; 2127 2128 if (sbi->s_jquota_fmt) { 2129 char *fmtname = ""; 2130 2131 switch (sbi->s_jquota_fmt) { 2132 case QFMT_VFS_OLD: 2133 fmtname = "vfsold"; 2134 break; 2135 case QFMT_VFS_V0: 2136 fmtname = "vfsv0"; 2137 break; 2138 case QFMT_VFS_V1: 2139 fmtname = "vfsv1"; 2140 break; 2141 } 2142 seq_printf(seq, ",jqfmt=%s", fmtname); 2143 } 2144 2145 rcu_read_lock(); 2146 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2147 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2148 if (usr_qf_name) 2149 seq_show_option(seq, "usrjquota", usr_qf_name); 2150 if (grp_qf_name) 2151 seq_show_option(seq, "grpjquota", grp_qf_name); 2152 rcu_read_unlock(); 2153 #endif 2154 } 2155 2156 static const char *token2str(int token) 2157 { 2158 const struct match_token *t; 2159 2160 for (t = tokens; t->token != Opt_err; t++) 2161 if (t->token == token && !strchr(t->pattern, '=')) 2162 break; 2163 return t->pattern; 2164 } 2165 2166 /* 2167 * Show an option if 2168 * - it's set to a non-default value OR 2169 * - if the per-sb default is different from the global default 2170 */ 2171 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2172 int nodefs) 2173 { 2174 struct ext4_sb_info *sbi = EXT4_SB(sb); 2175 struct ext4_super_block *es = sbi->s_es; 2176 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2177 const struct mount_opts *m; 2178 char sep = nodefs ? '\n' : ','; 2179 2180 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2181 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2182 2183 if (sbi->s_sb_block != 1) 2184 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2185 2186 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2187 int want_set = m->flags & MOPT_SET; 2188 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2189 (m->flags & MOPT_CLEAR_ERR)) 2190 continue; 2191 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2192 continue; /* skip if same as the default */ 2193 if ((want_set && 2194 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2195 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2196 continue; /* select Opt_noFoo vs Opt_Foo */ 2197 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2198 } 2199 2200 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2201 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2202 SEQ_OPTS_PRINT("resuid=%u", 2203 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2204 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2205 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2206 SEQ_OPTS_PRINT("resgid=%u", 2207 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2208 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2209 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2210 SEQ_OPTS_PUTS("errors=remount-ro"); 2211 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2212 SEQ_OPTS_PUTS("errors=continue"); 2213 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2214 SEQ_OPTS_PUTS("errors=panic"); 2215 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2216 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2217 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2218 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2219 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2220 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2221 if (sb->s_flags & SB_I_VERSION) 2222 SEQ_OPTS_PUTS("i_version"); 2223 if (nodefs || sbi->s_stripe) 2224 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2225 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2226 (sbi->s_mount_opt ^ def_mount_opt)) { 2227 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2228 SEQ_OPTS_PUTS("data=journal"); 2229 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2230 SEQ_OPTS_PUTS("data=ordered"); 2231 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2232 SEQ_OPTS_PUTS("data=writeback"); 2233 } 2234 if (nodefs || 2235 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2236 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2237 sbi->s_inode_readahead_blks); 2238 2239 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2240 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2241 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2242 if (nodefs || sbi->s_max_dir_size_kb) 2243 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2244 if (test_opt(sb, DATA_ERR_ABORT)) 2245 SEQ_OPTS_PUTS("data_err=abort"); 2246 if (DUMMY_ENCRYPTION_ENABLED(sbi)) 2247 SEQ_OPTS_PUTS("test_dummy_encryption"); 2248 2249 ext4_show_quota_options(seq, sb); 2250 return 0; 2251 } 2252 2253 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2254 { 2255 return _ext4_show_options(seq, root->d_sb, 0); 2256 } 2257 2258 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2259 { 2260 struct super_block *sb = seq->private; 2261 int rc; 2262 2263 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2264 rc = _ext4_show_options(seq, sb, 1); 2265 seq_puts(seq, "\n"); 2266 return rc; 2267 } 2268 2269 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2270 int read_only) 2271 { 2272 struct ext4_sb_info *sbi = EXT4_SB(sb); 2273 int err = 0; 2274 2275 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2276 ext4_msg(sb, KERN_ERR, "revision level too high, " 2277 "forcing read-only mode"); 2278 err = -EROFS; 2279 } 2280 if (read_only) 2281 goto done; 2282 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2283 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2284 "running e2fsck is recommended"); 2285 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2286 ext4_msg(sb, KERN_WARNING, 2287 "warning: mounting fs with errors, " 2288 "running e2fsck is recommended"); 2289 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2290 le16_to_cpu(es->s_mnt_count) >= 2291 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2292 ext4_msg(sb, KERN_WARNING, 2293 "warning: maximal mount count reached, " 2294 "running e2fsck is recommended"); 2295 else if (le32_to_cpu(es->s_checkinterval) && 2296 (ext4_get_tstamp(es, s_lastcheck) + 2297 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2298 ext4_msg(sb, KERN_WARNING, 2299 "warning: checktime reached, " 2300 "running e2fsck is recommended"); 2301 if (!sbi->s_journal) 2302 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2303 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2304 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2305 le16_add_cpu(&es->s_mnt_count, 1); 2306 ext4_update_tstamp(es, s_mtime); 2307 if (sbi->s_journal) 2308 ext4_set_feature_journal_needs_recovery(sb); 2309 2310 err = ext4_commit_super(sb, 1); 2311 done: 2312 if (test_opt(sb, DEBUG)) 2313 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2314 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2315 sb->s_blocksize, 2316 sbi->s_groups_count, 2317 EXT4_BLOCKS_PER_GROUP(sb), 2318 EXT4_INODES_PER_GROUP(sb), 2319 sbi->s_mount_opt, sbi->s_mount_opt2); 2320 2321 cleancache_init_fs(sb); 2322 return err; 2323 } 2324 2325 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2326 { 2327 struct ext4_sb_info *sbi = EXT4_SB(sb); 2328 struct flex_groups *new_groups; 2329 int size; 2330 2331 if (!sbi->s_log_groups_per_flex) 2332 return 0; 2333 2334 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2335 if (size <= sbi->s_flex_groups_allocated) 2336 return 0; 2337 2338 size = roundup_pow_of_two(size * sizeof(struct flex_groups)); 2339 new_groups = kvzalloc(size, GFP_KERNEL); 2340 if (!new_groups) { 2341 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups", 2342 size / (int) sizeof(struct flex_groups)); 2343 return -ENOMEM; 2344 } 2345 2346 if (sbi->s_flex_groups) { 2347 memcpy(new_groups, sbi->s_flex_groups, 2348 (sbi->s_flex_groups_allocated * 2349 sizeof(struct flex_groups))); 2350 kvfree(sbi->s_flex_groups); 2351 } 2352 sbi->s_flex_groups = new_groups; 2353 sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups); 2354 return 0; 2355 } 2356 2357 static int ext4_fill_flex_info(struct super_block *sb) 2358 { 2359 struct ext4_sb_info *sbi = EXT4_SB(sb); 2360 struct ext4_group_desc *gdp = NULL; 2361 ext4_group_t flex_group; 2362 int i, err; 2363 2364 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2365 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2366 sbi->s_log_groups_per_flex = 0; 2367 return 1; 2368 } 2369 2370 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2371 if (err) 2372 goto failed; 2373 2374 for (i = 0; i < sbi->s_groups_count; i++) { 2375 gdp = ext4_get_group_desc(sb, i, NULL); 2376 2377 flex_group = ext4_flex_group(sbi, i); 2378 atomic_add(ext4_free_inodes_count(sb, gdp), 2379 &sbi->s_flex_groups[flex_group].free_inodes); 2380 atomic64_add(ext4_free_group_clusters(sb, gdp), 2381 &sbi->s_flex_groups[flex_group].free_clusters); 2382 atomic_add(ext4_used_dirs_count(sb, gdp), 2383 &sbi->s_flex_groups[flex_group].used_dirs); 2384 } 2385 2386 return 1; 2387 failed: 2388 return 0; 2389 } 2390 2391 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2392 struct ext4_group_desc *gdp) 2393 { 2394 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2395 __u16 crc = 0; 2396 __le32 le_group = cpu_to_le32(block_group); 2397 struct ext4_sb_info *sbi = EXT4_SB(sb); 2398 2399 if (ext4_has_metadata_csum(sbi->s_sb)) { 2400 /* Use new metadata_csum algorithm */ 2401 __u32 csum32; 2402 __u16 dummy_csum = 0; 2403 2404 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2405 sizeof(le_group)); 2406 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2407 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2408 sizeof(dummy_csum)); 2409 offset += sizeof(dummy_csum); 2410 if (offset < sbi->s_desc_size) 2411 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2412 sbi->s_desc_size - offset); 2413 2414 crc = csum32 & 0xFFFF; 2415 goto out; 2416 } 2417 2418 /* old crc16 code */ 2419 if (!ext4_has_feature_gdt_csum(sb)) 2420 return 0; 2421 2422 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2423 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2424 crc = crc16(crc, (__u8 *)gdp, offset); 2425 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2426 /* for checksum of struct ext4_group_desc do the rest...*/ 2427 if (ext4_has_feature_64bit(sb) && 2428 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2429 crc = crc16(crc, (__u8 *)gdp + offset, 2430 le16_to_cpu(sbi->s_es->s_desc_size) - 2431 offset); 2432 2433 out: 2434 return cpu_to_le16(crc); 2435 } 2436 2437 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2438 struct ext4_group_desc *gdp) 2439 { 2440 if (ext4_has_group_desc_csum(sb) && 2441 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2442 return 0; 2443 2444 return 1; 2445 } 2446 2447 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2448 struct ext4_group_desc *gdp) 2449 { 2450 if (!ext4_has_group_desc_csum(sb)) 2451 return; 2452 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2453 } 2454 2455 /* Called at mount-time, super-block is locked */ 2456 static int ext4_check_descriptors(struct super_block *sb, 2457 ext4_fsblk_t sb_block, 2458 ext4_group_t *first_not_zeroed) 2459 { 2460 struct ext4_sb_info *sbi = EXT4_SB(sb); 2461 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2462 ext4_fsblk_t last_block; 2463 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2464 ext4_fsblk_t block_bitmap; 2465 ext4_fsblk_t inode_bitmap; 2466 ext4_fsblk_t inode_table; 2467 int flexbg_flag = 0; 2468 ext4_group_t i, grp = sbi->s_groups_count; 2469 2470 if (ext4_has_feature_flex_bg(sb)) 2471 flexbg_flag = 1; 2472 2473 ext4_debug("Checking group descriptors"); 2474 2475 for (i = 0; i < sbi->s_groups_count; i++) { 2476 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2477 2478 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2479 last_block = ext4_blocks_count(sbi->s_es) - 1; 2480 else 2481 last_block = first_block + 2482 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2483 2484 if ((grp == sbi->s_groups_count) && 2485 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2486 grp = i; 2487 2488 block_bitmap = ext4_block_bitmap(sb, gdp); 2489 if (block_bitmap == sb_block) { 2490 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2491 "Block bitmap for group %u overlaps " 2492 "superblock", i); 2493 if (!sb_rdonly(sb)) 2494 return 0; 2495 } 2496 if (block_bitmap >= sb_block + 1 && 2497 block_bitmap <= last_bg_block) { 2498 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2499 "Block bitmap for group %u overlaps " 2500 "block group descriptors", i); 2501 if (!sb_rdonly(sb)) 2502 return 0; 2503 } 2504 if (block_bitmap < first_block || block_bitmap > last_block) { 2505 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2506 "Block bitmap for group %u not in group " 2507 "(block %llu)!", i, block_bitmap); 2508 return 0; 2509 } 2510 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2511 if (inode_bitmap == sb_block) { 2512 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2513 "Inode bitmap for group %u overlaps " 2514 "superblock", i); 2515 if (!sb_rdonly(sb)) 2516 return 0; 2517 } 2518 if (inode_bitmap >= sb_block + 1 && 2519 inode_bitmap <= last_bg_block) { 2520 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2521 "Inode bitmap for group %u overlaps " 2522 "block group descriptors", i); 2523 if (!sb_rdonly(sb)) 2524 return 0; 2525 } 2526 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2527 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2528 "Inode bitmap for group %u not in group " 2529 "(block %llu)!", i, inode_bitmap); 2530 return 0; 2531 } 2532 inode_table = ext4_inode_table(sb, gdp); 2533 if (inode_table == sb_block) { 2534 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2535 "Inode table for group %u overlaps " 2536 "superblock", i); 2537 if (!sb_rdonly(sb)) 2538 return 0; 2539 } 2540 if (inode_table >= sb_block + 1 && 2541 inode_table <= last_bg_block) { 2542 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2543 "Inode table for group %u overlaps " 2544 "block group descriptors", i); 2545 if (!sb_rdonly(sb)) 2546 return 0; 2547 } 2548 if (inode_table < first_block || 2549 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2550 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2551 "Inode table for group %u not in group " 2552 "(block %llu)!", i, inode_table); 2553 return 0; 2554 } 2555 ext4_lock_group(sb, i); 2556 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2557 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2558 "Checksum for group %u failed (%u!=%u)", 2559 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2560 gdp)), le16_to_cpu(gdp->bg_checksum)); 2561 if (!sb_rdonly(sb)) { 2562 ext4_unlock_group(sb, i); 2563 return 0; 2564 } 2565 } 2566 ext4_unlock_group(sb, i); 2567 if (!flexbg_flag) 2568 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2569 } 2570 if (NULL != first_not_zeroed) 2571 *first_not_zeroed = grp; 2572 return 1; 2573 } 2574 2575 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2576 * the superblock) which were deleted from all directories, but held open by 2577 * a process at the time of a crash. We walk the list and try to delete these 2578 * inodes at recovery time (only with a read-write filesystem). 2579 * 2580 * In order to keep the orphan inode chain consistent during traversal (in 2581 * case of crash during recovery), we link each inode into the superblock 2582 * orphan list_head and handle it the same way as an inode deletion during 2583 * normal operation (which journals the operations for us). 2584 * 2585 * We only do an iget() and an iput() on each inode, which is very safe if we 2586 * accidentally point at an in-use or already deleted inode. The worst that 2587 * can happen in this case is that we get a "bit already cleared" message from 2588 * ext4_free_inode(). The only reason we would point at a wrong inode is if 2589 * e2fsck was run on this filesystem, and it must have already done the orphan 2590 * inode cleanup for us, so we can safely abort without any further action. 2591 */ 2592 static void ext4_orphan_cleanup(struct super_block *sb, 2593 struct ext4_super_block *es) 2594 { 2595 unsigned int s_flags = sb->s_flags; 2596 int ret, nr_orphans = 0, nr_truncates = 0; 2597 #ifdef CONFIG_QUOTA 2598 int quota_update = 0; 2599 int i; 2600 #endif 2601 if (!es->s_last_orphan) { 2602 jbd_debug(4, "no orphan inodes to clean up\n"); 2603 return; 2604 } 2605 2606 if (bdev_read_only(sb->s_bdev)) { 2607 ext4_msg(sb, KERN_ERR, "write access " 2608 "unavailable, skipping orphan cleanup"); 2609 return; 2610 } 2611 2612 /* Check if feature set would not allow a r/w mount */ 2613 if (!ext4_feature_set_ok(sb, 0)) { 2614 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 2615 "unknown ROCOMPAT features"); 2616 return; 2617 } 2618 2619 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 2620 /* don't clear list on RO mount w/ errors */ 2621 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 2622 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 2623 "clearing orphan list.\n"); 2624 es->s_last_orphan = 0; 2625 } 2626 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 2627 return; 2628 } 2629 2630 if (s_flags & SB_RDONLY) { 2631 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 2632 sb->s_flags &= ~SB_RDONLY; 2633 } 2634 #ifdef CONFIG_QUOTA 2635 /* Needed for iput() to work correctly and not trash data */ 2636 sb->s_flags |= SB_ACTIVE; 2637 2638 /* 2639 * Turn on quotas which were not enabled for read-only mounts if 2640 * filesystem has quota feature, so that they are updated correctly. 2641 */ 2642 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 2643 int ret = ext4_enable_quotas(sb); 2644 2645 if (!ret) 2646 quota_update = 1; 2647 else 2648 ext4_msg(sb, KERN_ERR, 2649 "Cannot turn on quotas: error %d", ret); 2650 } 2651 2652 /* Turn on journaled quotas used for old sytle */ 2653 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2654 if (EXT4_SB(sb)->s_qf_names[i]) { 2655 int ret = ext4_quota_on_mount(sb, i); 2656 2657 if (!ret) 2658 quota_update = 1; 2659 else 2660 ext4_msg(sb, KERN_ERR, 2661 "Cannot turn on journaled " 2662 "quota: type %d: error %d", i, ret); 2663 } 2664 } 2665 #endif 2666 2667 while (es->s_last_orphan) { 2668 struct inode *inode; 2669 2670 /* 2671 * We may have encountered an error during cleanup; if 2672 * so, skip the rest. 2673 */ 2674 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 2675 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 2676 es->s_last_orphan = 0; 2677 break; 2678 } 2679 2680 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 2681 if (IS_ERR(inode)) { 2682 es->s_last_orphan = 0; 2683 break; 2684 } 2685 2686 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 2687 dquot_initialize(inode); 2688 if (inode->i_nlink) { 2689 if (test_opt(sb, DEBUG)) 2690 ext4_msg(sb, KERN_DEBUG, 2691 "%s: truncating inode %lu to %lld bytes", 2692 __func__, inode->i_ino, inode->i_size); 2693 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 2694 inode->i_ino, inode->i_size); 2695 inode_lock(inode); 2696 truncate_inode_pages(inode->i_mapping, inode->i_size); 2697 ret = ext4_truncate(inode); 2698 if (ret) 2699 ext4_std_error(inode->i_sb, ret); 2700 inode_unlock(inode); 2701 nr_truncates++; 2702 } else { 2703 if (test_opt(sb, DEBUG)) 2704 ext4_msg(sb, KERN_DEBUG, 2705 "%s: deleting unreferenced inode %lu", 2706 __func__, inode->i_ino); 2707 jbd_debug(2, "deleting unreferenced inode %lu\n", 2708 inode->i_ino); 2709 nr_orphans++; 2710 } 2711 iput(inode); /* The delete magic happens here! */ 2712 } 2713 2714 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 2715 2716 if (nr_orphans) 2717 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 2718 PLURAL(nr_orphans)); 2719 if (nr_truncates) 2720 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 2721 PLURAL(nr_truncates)); 2722 #ifdef CONFIG_QUOTA 2723 /* Turn off quotas if they were enabled for orphan cleanup */ 2724 if (quota_update) { 2725 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2726 if (sb_dqopt(sb)->files[i]) 2727 dquot_quota_off(sb, i); 2728 } 2729 } 2730 #endif 2731 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 2732 } 2733 2734 /* 2735 * Maximal extent format file size. 2736 * Resulting logical blkno at s_maxbytes must fit in our on-disk 2737 * extent format containers, within a sector_t, and within i_blocks 2738 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 2739 * so that won't be a limiting factor. 2740 * 2741 * However there is other limiting factor. We do store extents in the form 2742 * of starting block and length, hence the resulting length of the extent 2743 * covering maximum file size must fit into on-disk format containers as 2744 * well. Given that length is always by 1 unit bigger than max unit (because 2745 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 2746 * 2747 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 2748 */ 2749 static loff_t ext4_max_size(int blkbits, int has_huge_files) 2750 { 2751 loff_t res; 2752 loff_t upper_limit = MAX_LFS_FILESIZE; 2753 2754 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 2755 2756 if (!has_huge_files) { 2757 upper_limit = (1LL << 32) - 1; 2758 2759 /* total blocks in file system block size */ 2760 upper_limit >>= (blkbits - 9); 2761 upper_limit <<= blkbits; 2762 } 2763 2764 /* 2765 * 32-bit extent-start container, ee_block. We lower the maxbytes 2766 * by one fs block, so ee_len can cover the extent of maximum file 2767 * size 2768 */ 2769 res = (1LL << 32) - 1; 2770 res <<= blkbits; 2771 2772 /* Sanity check against vm- & vfs- imposed limits */ 2773 if (res > upper_limit) 2774 res = upper_limit; 2775 2776 return res; 2777 } 2778 2779 /* 2780 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 2781 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 2782 * We need to be 1 filesystem block less than the 2^48 sector limit. 2783 */ 2784 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 2785 { 2786 loff_t res = EXT4_NDIR_BLOCKS; 2787 int meta_blocks; 2788 loff_t upper_limit; 2789 /* This is calculated to be the largest file size for a dense, block 2790 * mapped file such that the file's total number of 512-byte sectors, 2791 * including data and all indirect blocks, does not exceed (2^48 - 1). 2792 * 2793 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 2794 * number of 512-byte sectors of the file. 2795 */ 2796 2797 if (!has_huge_files) { 2798 /* 2799 * !has_huge_files or implies that the inode i_block field 2800 * represents total file blocks in 2^32 512-byte sectors == 2801 * size of vfs inode i_blocks * 8 2802 */ 2803 upper_limit = (1LL << 32) - 1; 2804 2805 /* total blocks in file system block size */ 2806 upper_limit >>= (bits - 9); 2807 2808 } else { 2809 /* 2810 * We use 48 bit ext4_inode i_blocks 2811 * With EXT4_HUGE_FILE_FL set the i_blocks 2812 * represent total number of blocks in 2813 * file system block size 2814 */ 2815 upper_limit = (1LL << 48) - 1; 2816 2817 } 2818 2819 /* indirect blocks */ 2820 meta_blocks = 1; 2821 /* double indirect blocks */ 2822 meta_blocks += 1 + (1LL << (bits-2)); 2823 /* tripple indirect blocks */ 2824 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 2825 2826 upper_limit -= meta_blocks; 2827 upper_limit <<= bits; 2828 2829 res += 1LL << (bits-2); 2830 res += 1LL << (2*(bits-2)); 2831 res += 1LL << (3*(bits-2)); 2832 res <<= bits; 2833 if (res > upper_limit) 2834 res = upper_limit; 2835 2836 if (res > MAX_LFS_FILESIZE) 2837 res = MAX_LFS_FILESIZE; 2838 2839 return res; 2840 } 2841 2842 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 2843 ext4_fsblk_t logical_sb_block, int nr) 2844 { 2845 struct ext4_sb_info *sbi = EXT4_SB(sb); 2846 ext4_group_t bg, first_meta_bg; 2847 int has_super = 0; 2848 2849 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 2850 2851 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 2852 return logical_sb_block + nr + 1; 2853 bg = sbi->s_desc_per_block * nr; 2854 if (ext4_bg_has_super(sb, bg)) 2855 has_super = 1; 2856 2857 /* 2858 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 2859 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 2860 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 2861 * compensate. 2862 */ 2863 if (sb->s_blocksize == 1024 && nr == 0 && 2864 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 2865 has_super++; 2866 2867 return (has_super + ext4_group_first_block_no(sb, bg)); 2868 } 2869 2870 /** 2871 * ext4_get_stripe_size: Get the stripe size. 2872 * @sbi: In memory super block info 2873 * 2874 * If we have specified it via mount option, then 2875 * use the mount option value. If the value specified at mount time is 2876 * greater than the blocks per group use the super block value. 2877 * If the super block value is greater than blocks per group return 0. 2878 * Allocator needs it be less than blocks per group. 2879 * 2880 */ 2881 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 2882 { 2883 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 2884 unsigned long stripe_width = 2885 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 2886 int ret; 2887 2888 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 2889 ret = sbi->s_stripe; 2890 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 2891 ret = stripe_width; 2892 else if (stride && stride <= sbi->s_blocks_per_group) 2893 ret = stride; 2894 else 2895 ret = 0; 2896 2897 /* 2898 * If the stripe width is 1, this makes no sense and 2899 * we set it to 0 to turn off stripe handling code. 2900 */ 2901 if (ret <= 1) 2902 ret = 0; 2903 2904 return ret; 2905 } 2906 2907 /* 2908 * Check whether this filesystem can be mounted based on 2909 * the features present and the RDONLY/RDWR mount requested. 2910 * Returns 1 if this filesystem can be mounted as requested, 2911 * 0 if it cannot be. 2912 */ 2913 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 2914 { 2915 if (ext4_has_unknown_ext4_incompat_features(sb)) { 2916 ext4_msg(sb, KERN_ERR, 2917 "Couldn't mount because of " 2918 "unsupported optional features (%x)", 2919 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 2920 ~EXT4_FEATURE_INCOMPAT_SUPP)); 2921 return 0; 2922 } 2923 2924 #ifndef CONFIG_UNICODE 2925 if (ext4_has_feature_casefold(sb)) { 2926 ext4_msg(sb, KERN_ERR, 2927 "Filesystem with casefold feature cannot be " 2928 "mounted without CONFIG_UNICODE"); 2929 return 0; 2930 } 2931 #endif 2932 2933 if (readonly) 2934 return 1; 2935 2936 if (ext4_has_feature_readonly(sb)) { 2937 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 2938 sb->s_flags |= SB_RDONLY; 2939 return 1; 2940 } 2941 2942 /* Check that feature set is OK for a read-write mount */ 2943 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 2944 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 2945 "unsupported optional features (%x)", 2946 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 2947 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 2948 return 0; 2949 } 2950 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 2951 ext4_msg(sb, KERN_ERR, 2952 "Can't support bigalloc feature without " 2953 "extents feature\n"); 2954 return 0; 2955 } 2956 2957 #ifndef CONFIG_QUOTA 2958 if (ext4_has_feature_quota(sb) && !readonly) { 2959 ext4_msg(sb, KERN_ERR, 2960 "Filesystem with quota feature cannot be mounted RDWR " 2961 "without CONFIG_QUOTA"); 2962 return 0; 2963 } 2964 if (ext4_has_feature_project(sb) && !readonly) { 2965 ext4_msg(sb, KERN_ERR, 2966 "Filesystem with project quota feature cannot be mounted RDWR " 2967 "without CONFIG_QUOTA"); 2968 return 0; 2969 } 2970 #endif /* CONFIG_QUOTA */ 2971 return 1; 2972 } 2973 2974 /* 2975 * This function is called once a day if we have errors logged 2976 * on the file system 2977 */ 2978 static void print_daily_error_info(struct timer_list *t) 2979 { 2980 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 2981 struct super_block *sb = sbi->s_sb; 2982 struct ext4_super_block *es = sbi->s_es; 2983 2984 if (es->s_error_count) 2985 /* fsck newer than v1.41.13 is needed to clean this condition. */ 2986 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 2987 le32_to_cpu(es->s_error_count)); 2988 if (es->s_first_error_time) { 2989 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 2990 sb->s_id, 2991 ext4_get_tstamp(es, s_first_error_time), 2992 (int) sizeof(es->s_first_error_func), 2993 es->s_first_error_func, 2994 le32_to_cpu(es->s_first_error_line)); 2995 if (es->s_first_error_ino) 2996 printk(KERN_CONT ": inode %u", 2997 le32_to_cpu(es->s_first_error_ino)); 2998 if (es->s_first_error_block) 2999 printk(KERN_CONT ": block %llu", (unsigned long long) 3000 le64_to_cpu(es->s_first_error_block)); 3001 printk(KERN_CONT "\n"); 3002 } 3003 if (es->s_last_error_time) { 3004 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3005 sb->s_id, 3006 ext4_get_tstamp(es, s_last_error_time), 3007 (int) sizeof(es->s_last_error_func), 3008 es->s_last_error_func, 3009 le32_to_cpu(es->s_last_error_line)); 3010 if (es->s_last_error_ino) 3011 printk(KERN_CONT ": inode %u", 3012 le32_to_cpu(es->s_last_error_ino)); 3013 if (es->s_last_error_block) 3014 printk(KERN_CONT ": block %llu", (unsigned long long) 3015 le64_to_cpu(es->s_last_error_block)); 3016 printk(KERN_CONT "\n"); 3017 } 3018 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3019 } 3020 3021 /* Find next suitable group and run ext4_init_inode_table */ 3022 static int ext4_run_li_request(struct ext4_li_request *elr) 3023 { 3024 struct ext4_group_desc *gdp = NULL; 3025 ext4_group_t group, ngroups; 3026 struct super_block *sb; 3027 unsigned long timeout = 0; 3028 int ret = 0; 3029 3030 sb = elr->lr_super; 3031 ngroups = EXT4_SB(sb)->s_groups_count; 3032 3033 for (group = elr->lr_next_group; group < ngroups; group++) { 3034 gdp = ext4_get_group_desc(sb, group, NULL); 3035 if (!gdp) { 3036 ret = 1; 3037 break; 3038 } 3039 3040 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3041 break; 3042 } 3043 3044 if (group >= ngroups) 3045 ret = 1; 3046 3047 if (!ret) { 3048 timeout = jiffies; 3049 ret = ext4_init_inode_table(sb, group, 3050 elr->lr_timeout ? 0 : 1); 3051 if (elr->lr_timeout == 0) { 3052 timeout = (jiffies - timeout) * 3053 elr->lr_sbi->s_li_wait_mult; 3054 elr->lr_timeout = timeout; 3055 } 3056 elr->lr_next_sched = jiffies + elr->lr_timeout; 3057 elr->lr_next_group = group + 1; 3058 } 3059 return ret; 3060 } 3061 3062 /* 3063 * Remove lr_request from the list_request and free the 3064 * request structure. Should be called with li_list_mtx held 3065 */ 3066 static void ext4_remove_li_request(struct ext4_li_request *elr) 3067 { 3068 struct ext4_sb_info *sbi; 3069 3070 if (!elr) 3071 return; 3072 3073 sbi = elr->lr_sbi; 3074 3075 list_del(&elr->lr_request); 3076 sbi->s_li_request = NULL; 3077 kfree(elr); 3078 } 3079 3080 static void ext4_unregister_li_request(struct super_block *sb) 3081 { 3082 mutex_lock(&ext4_li_mtx); 3083 if (!ext4_li_info) { 3084 mutex_unlock(&ext4_li_mtx); 3085 return; 3086 } 3087 3088 mutex_lock(&ext4_li_info->li_list_mtx); 3089 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3090 mutex_unlock(&ext4_li_info->li_list_mtx); 3091 mutex_unlock(&ext4_li_mtx); 3092 } 3093 3094 static struct task_struct *ext4_lazyinit_task; 3095 3096 /* 3097 * This is the function where ext4lazyinit thread lives. It walks 3098 * through the request list searching for next scheduled filesystem. 3099 * When such a fs is found, run the lazy initialization request 3100 * (ext4_rn_li_request) and keep track of the time spend in this 3101 * function. Based on that time we compute next schedule time of 3102 * the request. When walking through the list is complete, compute 3103 * next waking time and put itself into sleep. 3104 */ 3105 static int ext4_lazyinit_thread(void *arg) 3106 { 3107 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3108 struct list_head *pos, *n; 3109 struct ext4_li_request *elr; 3110 unsigned long next_wakeup, cur; 3111 3112 BUG_ON(NULL == eli); 3113 3114 cont_thread: 3115 while (true) { 3116 next_wakeup = MAX_JIFFY_OFFSET; 3117 3118 mutex_lock(&eli->li_list_mtx); 3119 if (list_empty(&eli->li_request_list)) { 3120 mutex_unlock(&eli->li_list_mtx); 3121 goto exit_thread; 3122 } 3123 list_for_each_safe(pos, n, &eli->li_request_list) { 3124 int err = 0; 3125 int progress = 0; 3126 elr = list_entry(pos, struct ext4_li_request, 3127 lr_request); 3128 3129 if (time_before(jiffies, elr->lr_next_sched)) { 3130 if (time_before(elr->lr_next_sched, next_wakeup)) 3131 next_wakeup = elr->lr_next_sched; 3132 continue; 3133 } 3134 if (down_read_trylock(&elr->lr_super->s_umount)) { 3135 if (sb_start_write_trylock(elr->lr_super)) { 3136 progress = 1; 3137 /* 3138 * We hold sb->s_umount, sb can not 3139 * be removed from the list, it is 3140 * now safe to drop li_list_mtx 3141 */ 3142 mutex_unlock(&eli->li_list_mtx); 3143 err = ext4_run_li_request(elr); 3144 sb_end_write(elr->lr_super); 3145 mutex_lock(&eli->li_list_mtx); 3146 n = pos->next; 3147 } 3148 up_read((&elr->lr_super->s_umount)); 3149 } 3150 /* error, remove the lazy_init job */ 3151 if (err) { 3152 ext4_remove_li_request(elr); 3153 continue; 3154 } 3155 if (!progress) { 3156 elr->lr_next_sched = jiffies + 3157 (prandom_u32() 3158 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3159 } 3160 if (time_before(elr->lr_next_sched, next_wakeup)) 3161 next_wakeup = elr->lr_next_sched; 3162 } 3163 mutex_unlock(&eli->li_list_mtx); 3164 3165 try_to_freeze(); 3166 3167 cur = jiffies; 3168 if ((time_after_eq(cur, next_wakeup)) || 3169 (MAX_JIFFY_OFFSET == next_wakeup)) { 3170 cond_resched(); 3171 continue; 3172 } 3173 3174 schedule_timeout_interruptible(next_wakeup - cur); 3175 3176 if (kthread_should_stop()) { 3177 ext4_clear_request_list(); 3178 goto exit_thread; 3179 } 3180 } 3181 3182 exit_thread: 3183 /* 3184 * It looks like the request list is empty, but we need 3185 * to check it under the li_list_mtx lock, to prevent any 3186 * additions into it, and of course we should lock ext4_li_mtx 3187 * to atomically free the list and ext4_li_info, because at 3188 * this point another ext4 filesystem could be registering 3189 * new one. 3190 */ 3191 mutex_lock(&ext4_li_mtx); 3192 mutex_lock(&eli->li_list_mtx); 3193 if (!list_empty(&eli->li_request_list)) { 3194 mutex_unlock(&eli->li_list_mtx); 3195 mutex_unlock(&ext4_li_mtx); 3196 goto cont_thread; 3197 } 3198 mutex_unlock(&eli->li_list_mtx); 3199 kfree(ext4_li_info); 3200 ext4_li_info = NULL; 3201 mutex_unlock(&ext4_li_mtx); 3202 3203 return 0; 3204 } 3205 3206 static void ext4_clear_request_list(void) 3207 { 3208 struct list_head *pos, *n; 3209 struct ext4_li_request *elr; 3210 3211 mutex_lock(&ext4_li_info->li_list_mtx); 3212 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3213 elr = list_entry(pos, struct ext4_li_request, 3214 lr_request); 3215 ext4_remove_li_request(elr); 3216 } 3217 mutex_unlock(&ext4_li_info->li_list_mtx); 3218 } 3219 3220 static int ext4_run_lazyinit_thread(void) 3221 { 3222 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3223 ext4_li_info, "ext4lazyinit"); 3224 if (IS_ERR(ext4_lazyinit_task)) { 3225 int err = PTR_ERR(ext4_lazyinit_task); 3226 ext4_clear_request_list(); 3227 kfree(ext4_li_info); 3228 ext4_li_info = NULL; 3229 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3230 "initialization thread\n", 3231 err); 3232 return err; 3233 } 3234 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3235 return 0; 3236 } 3237 3238 /* 3239 * Check whether it make sense to run itable init. thread or not. 3240 * If there is at least one uninitialized inode table, return 3241 * corresponding group number, else the loop goes through all 3242 * groups and return total number of groups. 3243 */ 3244 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3245 { 3246 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3247 struct ext4_group_desc *gdp = NULL; 3248 3249 if (!ext4_has_group_desc_csum(sb)) 3250 return ngroups; 3251 3252 for (group = 0; group < ngroups; group++) { 3253 gdp = ext4_get_group_desc(sb, group, NULL); 3254 if (!gdp) 3255 continue; 3256 3257 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3258 break; 3259 } 3260 3261 return group; 3262 } 3263 3264 static int ext4_li_info_new(void) 3265 { 3266 struct ext4_lazy_init *eli = NULL; 3267 3268 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3269 if (!eli) 3270 return -ENOMEM; 3271 3272 INIT_LIST_HEAD(&eli->li_request_list); 3273 mutex_init(&eli->li_list_mtx); 3274 3275 eli->li_state |= EXT4_LAZYINIT_QUIT; 3276 3277 ext4_li_info = eli; 3278 3279 return 0; 3280 } 3281 3282 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3283 ext4_group_t start) 3284 { 3285 struct ext4_sb_info *sbi = EXT4_SB(sb); 3286 struct ext4_li_request *elr; 3287 3288 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3289 if (!elr) 3290 return NULL; 3291 3292 elr->lr_super = sb; 3293 elr->lr_sbi = sbi; 3294 elr->lr_next_group = start; 3295 3296 /* 3297 * Randomize first schedule time of the request to 3298 * spread the inode table initialization requests 3299 * better. 3300 */ 3301 elr->lr_next_sched = jiffies + (prandom_u32() % 3302 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3303 return elr; 3304 } 3305 3306 int ext4_register_li_request(struct super_block *sb, 3307 ext4_group_t first_not_zeroed) 3308 { 3309 struct ext4_sb_info *sbi = EXT4_SB(sb); 3310 struct ext4_li_request *elr = NULL; 3311 ext4_group_t ngroups = sbi->s_groups_count; 3312 int ret = 0; 3313 3314 mutex_lock(&ext4_li_mtx); 3315 if (sbi->s_li_request != NULL) { 3316 /* 3317 * Reset timeout so it can be computed again, because 3318 * s_li_wait_mult might have changed. 3319 */ 3320 sbi->s_li_request->lr_timeout = 0; 3321 goto out; 3322 } 3323 3324 if (first_not_zeroed == ngroups || sb_rdonly(sb) || 3325 !test_opt(sb, INIT_INODE_TABLE)) 3326 goto out; 3327 3328 elr = ext4_li_request_new(sb, first_not_zeroed); 3329 if (!elr) { 3330 ret = -ENOMEM; 3331 goto out; 3332 } 3333 3334 if (NULL == ext4_li_info) { 3335 ret = ext4_li_info_new(); 3336 if (ret) 3337 goto out; 3338 } 3339 3340 mutex_lock(&ext4_li_info->li_list_mtx); 3341 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3342 mutex_unlock(&ext4_li_info->li_list_mtx); 3343 3344 sbi->s_li_request = elr; 3345 /* 3346 * set elr to NULL here since it has been inserted to 3347 * the request_list and the removal and free of it is 3348 * handled by ext4_clear_request_list from now on. 3349 */ 3350 elr = NULL; 3351 3352 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3353 ret = ext4_run_lazyinit_thread(); 3354 if (ret) 3355 goto out; 3356 } 3357 out: 3358 mutex_unlock(&ext4_li_mtx); 3359 if (ret) 3360 kfree(elr); 3361 return ret; 3362 } 3363 3364 /* 3365 * We do not need to lock anything since this is called on 3366 * module unload. 3367 */ 3368 static void ext4_destroy_lazyinit_thread(void) 3369 { 3370 /* 3371 * If thread exited earlier 3372 * there's nothing to be done. 3373 */ 3374 if (!ext4_li_info || !ext4_lazyinit_task) 3375 return; 3376 3377 kthread_stop(ext4_lazyinit_task); 3378 } 3379 3380 static int set_journal_csum_feature_set(struct super_block *sb) 3381 { 3382 int ret = 1; 3383 int compat, incompat; 3384 struct ext4_sb_info *sbi = EXT4_SB(sb); 3385 3386 if (ext4_has_metadata_csum(sb)) { 3387 /* journal checksum v3 */ 3388 compat = 0; 3389 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3390 } else { 3391 /* journal checksum v1 */ 3392 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3393 incompat = 0; 3394 } 3395 3396 jbd2_journal_clear_features(sbi->s_journal, 3397 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3398 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3399 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3400 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3401 ret = jbd2_journal_set_features(sbi->s_journal, 3402 compat, 0, 3403 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3404 incompat); 3405 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3406 ret = jbd2_journal_set_features(sbi->s_journal, 3407 compat, 0, 3408 incompat); 3409 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3410 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3411 } else { 3412 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3413 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3414 } 3415 3416 return ret; 3417 } 3418 3419 /* 3420 * Note: calculating the overhead so we can be compatible with 3421 * historical BSD practice is quite difficult in the face of 3422 * clusters/bigalloc. This is because multiple metadata blocks from 3423 * different block group can end up in the same allocation cluster. 3424 * Calculating the exact overhead in the face of clustered allocation 3425 * requires either O(all block bitmaps) in memory or O(number of block 3426 * groups**2) in time. We will still calculate the superblock for 3427 * older file systems --- and if we come across with a bigalloc file 3428 * system with zero in s_overhead_clusters the estimate will be close to 3429 * correct especially for very large cluster sizes --- but for newer 3430 * file systems, it's better to calculate this figure once at mkfs 3431 * time, and store it in the superblock. If the superblock value is 3432 * present (even for non-bigalloc file systems), we will use it. 3433 */ 3434 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3435 char *buf) 3436 { 3437 struct ext4_sb_info *sbi = EXT4_SB(sb); 3438 struct ext4_group_desc *gdp; 3439 ext4_fsblk_t first_block, last_block, b; 3440 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3441 int s, j, count = 0; 3442 3443 if (!ext4_has_feature_bigalloc(sb)) 3444 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3445 sbi->s_itb_per_group + 2); 3446 3447 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3448 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3449 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3450 for (i = 0; i < ngroups; i++) { 3451 gdp = ext4_get_group_desc(sb, i, NULL); 3452 b = ext4_block_bitmap(sb, gdp); 3453 if (b >= first_block && b <= last_block) { 3454 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3455 count++; 3456 } 3457 b = ext4_inode_bitmap(sb, gdp); 3458 if (b >= first_block && b <= last_block) { 3459 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3460 count++; 3461 } 3462 b = ext4_inode_table(sb, gdp); 3463 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3464 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3465 int c = EXT4_B2C(sbi, b - first_block); 3466 ext4_set_bit(c, buf); 3467 count++; 3468 } 3469 if (i != grp) 3470 continue; 3471 s = 0; 3472 if (ext4_bg_has_super(sb, grp)) { 3473 ext4_set_bit(s++, buf); 3474 count++; 3475 } 3476 j = ext4_bg_num_gdb(sb, grp); 3477 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3478 ext4_error(sb, "Invalid number of block group " 3479 "descriptor blocks: %d", j); 3480 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3481 } 3482 count += j; 3483 for (; j > 0; j--) 3484 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3485 } 3486 if (!count) 3487 return 0; 3488 return EXT4_CLUSTERS_PER_GROUP(sb) - 3489 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3490 } 3491 3492 /* 3493 * Compute the overhead and stash it in sbi->s_overhead 3494 */ 3495 int ext4_calculate_overhead(struct super_block *sb) 3496 { 3497 struct ext4_sb_info *sbi = EXT4_SB(sb); 3498 struct ext4_super_block *es = sbi->s_es; 3499 struct inode *j_inode; 3500 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3501 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3502 ext4_fsblk_t overhead = 0; 3503 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3504 3505 if (!buf) 3506 return -ENOMEM; 3507 3508 /* 3509 * Compute the overhead (FS structures). This is constant 3510 * for a given filesystem unless the number of block groups 3511 * changes so we cache the previous value until it does. 3512 */ 3513 3514 /* 3515 * All of the blocks before first_data_block are overhead 3516 */ 3517 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3518 3519 /* 3520 * Add the overhead found in each block group 3521 */ 3522 for (i = 0; i < ngroups; i++) { 3523 int blks; 3524 3525 blks = count_overhead(sb, i, buf); 3526 overhead += blks; 3527 if (blks) 3528 memset(buf, 0, PAGE_SIZE); 3529 cond_resched(); 3530 } 3531 3532 /* 3533 * Add the internal journal blocks whether the journal has been 3534 * loaded or not 3535 */ 3536 if (sbi->s_journal && !sbi->journal_bdev) 3537 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen); 3538 else if (ext4_has_feature_journal(sb) && !sbi->s_journal) { 3539 j_inode = ext4_get_journal_inode(sb, j_inum); 3540 if (j_inode) { 3541 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3542 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3543 iput(j_inode); 3544 } else { 3545 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3546 } 3547 } 3548 sbi->s_overhead = overhead; 3549 smp_wmb(); 3550 free_page((unsigned long) buf); 3551 return 0; 3552 } 3553 3554 static void ext4_clamp_want_extra_isize(struct super_block *sb) 3555 { 3556 struct ext4_sb_info *sbi = EXT4_SB(sb); 3557 struct ext4_super_block *es = sbi->s_es; 3558 3559 /* determine the minimum size of new large inodes, if present */ 3560 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE && 3561 sbi->s_want_extra_isize == 0) { 3562 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 3563 EXT4_GOOD_OLD_INODE_SIZE; 3564 if (ext4_has_feature_extra_isize(sb)) { 3565 if (sbi->s_want_extra_isize < 3566 le16_to_cpu(es->s_want_extra_isize)) 3567 sbi->s_want_extra_isize = 3568 le16_to_cpu(es->s_want_extra_isize); 3569 if (sbi->s_want_extra_isize < 3570 le16_to_cpu(es->s_min_extra_isize)) 3571 sbi->s_want_extra_isize = 3572 le16_to_cpu(es->s_min_extra_isize); 3573 } 3574 } 3575 /* Check if enough inode space is available */ 3576 if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize > 3577 sbi->s_inode_size) { 3578 sbi->s_want_extra_isize = sizeof(struct ext4_inode) - 3579 EXT4_GOOD_OLD_INODE_SIZE; 3580 ext4_msg(sb, KERN_INFO, 3581 "required extra inode space not available"); 3582 } 3583 } 3584 3585 static void ext4_set_resv_clusters(struct super_block *sb) 3586 { 3587 ext4_fsblk_t resv_clusters; 3588 struct ext4_sb_info *sbi = EXT4_SB(sb); 3589 3590 /* 3591 * There's no need to reserve anything when we aren't using extents. 3592 * The space estimates are exact, there are no unwritten extents, 3593 * hole punching doesn't need new metadata... This is needed especially 3594 * to keep ext2/3 backward compatibility. 3595 */ 3596 if (!ext4_has_feature_extents(sb)) 3597 return; 3598 /* 3599 * By default we reserve 2% or 4096 clusters, whichever is smaller. 3600 * This should cover the situations where we can not afford to run 3601 * out of space like for example punch hole, or converting 3602 * unwritten extents in delalloc path. In most cases such 3603 * allocation would require 1, or 2 blocks, higher numbers are 3604 * very rare. 3605 */ 3606 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 3607 sbi->s_cluster_bits); 3608 3609 do_div(resv_clusters, 50); 3610 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 3611 3612 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 3613 } 3614 3615 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 3616 { 3617 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 3618 char *orig_data = kstrdup(data, GFP_KERNEL); 3619 struct buffer_head *bh; 3620 struct ext4_super_block *es = NULL; 3621 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 3622 ext4_fsblk_t block; 3623 ext4_fsblk_t sb_block = get_sb_block(&data); 3624 ext4_fsblk_t logical_sb_block; 3625 unsigned long offset = 0; 3626 unsigned long journal_devnum = 0; 3627 unsigned long def_mount_opts; 3628 struct inode *root; 3629 const char *descr; 3630 int ret = -ENOMEM; 3631 int blocksize, clustersize; 3632 unsigned int db_count; 3633 unsigned int i; 3634 int needs_recovery, has_huge_files, has_bigalloc; 3635 __u64 blocks_count; 3636 int err = 0; 3637 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 3638 ext4_group_t first_not_zeroed; 3639 3640 if ((data && !orig_data) || !sbi) 3641 goto out_free_base; 3642 3643 sbi->s_daxdev = dax_dev; 3644 sbi->s_blockgroup_lock = 3645 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 3646 if (!sbi->s_blockgroup_lock) 3647 goto out_free_base; 3648 3649 sb->s_fs_info = sbi; 3650 sbi->s_sb = sb; 3651 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 3652 sbi->s_sb_block = sb_block; 3653 if (sb->s_bdev->bd_part) 3654 sbi->s_sectors_written_start = 3655 part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]); 3656 3657 /* Cleanup superblock name */ 3658 strreplace(sb->s_id, '/', '!'); 3659 3660 /* -EINVAL is default */ 3661 ret = -EINVAL; 3662 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 3663 if (!blocksize) { 3664 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 3665 goto out_fail; 3666 } 3667 3668 /* 3669 * The ext4 superblock will not be buffer aligned for other than 1kB 3670 * block sizes. We need to calculate the offset from buffer start. 3671 */ 3672 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 3673 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 3674 offset = do_div(logical_sb_block, blocksize); 3675 } else { 3676 logical_sb_block = sb_block; 3677 } 3678 3679 if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) { 3680 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 3681 goto out_fail; 3682 } 3683 /* 3684 * Note: s_es must be initialized as soon as possible because 3685 * some ext4 macro-instructions depend on its value 3686 */ 3687 es = (struct ext4_super_block *) (bh->b_data + offset); 3688 sbi->s_es = es; 3689 sb->s_magic = le16_to_cpu(es->s_magic); 3690 if (sb->s_magic != EXT4_SUPER_MAGIC) 3691 goto cantfind_ext4; 3692 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 3693 3694 /* Warn if metadata_csum and gdt_csum are both set. */ 3695 if (ext4_has_feature_metadata_csum(sb) && 3696 ext4_has_feature_gdt_csum(sb)) 3697 ext4_warning(sb, "metadata_csum and uninit_bg are " 3698 "redundant flags; please run fsck."); 3699 3700 /* Check for a known checksum algorithm */ 3701 if (!ext4_verify_csum_type(sb, es)) { 3702 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 3703 "unknown checksum algorithm."); 3704 silent = 1; 3705 goto cantfind_ext4; 3706 } 3707 3708 /* Load the checksum driver */ 3709 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 3710 if (IS_ERR(sbi->s_chksum_driver)) { 3711 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 3712 ret = PTR_ERR(sbi->s_chksum_driver); 3713 sbi->s_chksum_driver = NULL; 3714 goto failed_mount; 3715 } 3716 3717 /* Check superblock checksum */ 3718 if (!ext4_superblock_csum_verify(sb, es)) { 3719 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 3720 "invalid superblock checksum. Run e2fsck?"); 3721 silent = 1; 3722 ret = -EFSBADCRC; 3723 goto cantfind_ext4; 3724 } 3725 3726 /* Precompute checksum seed for all metadata */ 3727 if (ext4_has_feature_csum_seed(sb)) 3728 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 3729 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 3730 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 3731 sizeof(es->s_uuid)); 3732 3733 /* Set defaults before we parse the mount options */ 3734 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 3735 set_opt(sb, INIT_INODE_TABLE); 3736 if (def_mount_opts & EXT4_DEFM_DEBUG) 3737 set_opt(sb, DEBUG); 3738 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 3739 set_opt(sb, GRPID); 3740 if (def_mount_opts & EXT4_DEFM_UID16) 3741 set_opt(sb, NO_UID32); 3742 /* xattr user namespace & acls are now defaulted on */ 3743 set_opt(sb, XATTR_USER); 3744 #ifdef CONFIG_EXT4_FS_POSIX_ACL 3745 set_opt(sb, POSIX_ACL); 3746 #endif 3747 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 3748 if (ext4_has_metadata_csum(sb)) 3749 set_opt(sb, JOURNAL_CHECKSUM); 3750 3751 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 3752 set_opt(sb, JOURNAL_DATA); 3753 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 3754 set_opt(sb, ORDERED_DATA); 3755 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 3756 set_opt(sb, WRITEBACK_DATA); 3757 3758 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 3759 set_opt(sb, ERRORS_PANIC); 3760 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 3761 set_opt(sb, ERRORS_CONT); 3762 else 3763 set_opt(sb, ERRORS_RO); 3764 /* block_validity enabled by default; disable with noblock_validity */ 3765 set_opt(sb, BLOCK_VALIDITY); 3766 if (def_mount_opts & EXT4_DEFM_DISCARD) 3767 set_opt(sb, DISCARD); 3768 3769 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 3770 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 3771 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 3772 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 3773 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 3774 3775 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 3776 set_opt(sb, BARRIER); 3777 3778 /* 3779 * enable delayed allocation by default 3780 * Use -o nodelalloc to turn it off 3781 */ 3782 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 3783 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 3784 set_opt(sb, DELALLOC); 3785 3786 /* 3787 * set default s_li_wait_mult for lazyinit, for the case there is 3788 * no mount option specified. 3789 */ 3790 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 3791 3792 if (sbi->s_es->s_mount_opts[0]) { 3793 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 3794 sizeof(sbi->s_es->s_mount_opts), 3795 GFP_KERNEL); 3796 if (!s_mount_opts) 3797 goto failed_mount; 3798 if (!parse_options(s_mount_opts, sb, &journal_devnum, 3799 &journal_ioprio, 0)) { 3800 ext4_msg(sb, KERN_WARNING, 3801 "failed to parse options in superblock: %s", 3802 s_mount_opts); 3803 } 3804 kfree(s_mount_opts); 3805 } 3806 sbi->s_def_mount_opt = sbi->s_mount_opt; 3807 if (!parse_options((char *) data, sb, &journal_devnum, 3808 &journal_ioprio, 0)) 3809 goto failed_mount; 3810 3811 #ifdef CONFIG_UNICODE 3812 if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) { 3813 const struct ext4_sb_encodings *encoding_info; 3814 struct unicode_map *encoding; 3815 __u16 encoding_flags; 3816 3817 if (ext4_has_feature_encrypt(sb)) { 3818 ext4_msg(sb, KERN_ERR, 3819 "Can't mount with encoding and encryption"); 3820 goto failed_mount; 3821 } 3822 3823 if (ext4_sb_read_encoding(es, &encoding_info, 3824 &encoding_flags)) { 3825 ext4_msg(sb, KERN_ERR, 3826 "Encoding requested by superblock is unknown"); 3827 goto failed_mount; 3828 } 3829 3830 encoding = utf8_load(encoding_info->version); 3831 if (IS_ERR(encoding)) { 3832 ext4_msg(sb, KERN_ERR, 3833 "can't mount with superblock charset: %s-%s " 3834 "not supported by the kernel. flags: 0x%x.", 3835 encoding_info->name, encoding_info->version, 3836 encoding_flags); 3837 goto failed_mount; 3838 } 3839 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 3840 "%s-%s with flags 0x%hx", encoding_info->name, 3841 encoding_info->version?:"\b", encoding_flags); 3842 3843 sbi->s_encoding = encoding; 3844 sbi->s_encoding_flags = encoding_flags; 3845 } 3846 #endif 3847 3848 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 3849 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting " 3850 "with data=journal disables delayed " 3851 "allocation and O_DIRECT support!\n"); 3852 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 3853 ext4_msg(sb, KERN_ERR, "can't mount with " 3854 "both data=journal and delalloc"); 3855 goto failed_mount; 3856 } 3857 if (test_opt(sb, DIOREAD_NOLOCK)) { 3858 ext4_msg(sb, KERN_ERR, "can't mount with " 3859 "both data=journal and dioread_nolock"); 3860 goto failed_mount; 3861 } 3862 if (test_opt(sb, DAX)) { 3863 ext4_msg(sb, KERN_ERR, "can't mount with " 3864 "both data=journal and dax"); 3865 goto failed_mount; 3866 } 3867 if (ext4_has_feature_encrypt(sb)) { 3868 ext4_msg(sb, KERN_WARNING, 3869 "encrypted files will use data=ordered " 3870 "instead of data journaling mode"); 3871 } 3872 if (test_opt(sb, DELALLOC)) 3873 clear_opt(sb, DELALLOC); 3874 } else { 3875 sb->s_iflags |= SB_I_CGROUPWB; 3876 } 3877 3878 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 3879 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 3880 3881 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 3882 (ext4_has_compat_features(sb) || 3883 ext4_has_ro_compat_features(sb) || 3884 ext4_has_incompat_features(sb))) 3885 ext4_msg(sb, KERN_WARNING, 3886 "feature flags set on rev 0 fs, " 3887 "running e2fsck is recommended"); 3888 3889 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 3890 set_opt2(sb, HURD_COMPAT); 3891 if (ext4_has_feature_64bit(sb)) { 3892 ext4_msg(sb, KERN_ERR, 3893 "The Hurd can't support 64-bit file systems"); 3894 goto failed_mount; 3895 } 3896 3897 /* 3898 * ea_inode feature uses l_i_version field which is not 3899 * available in HURD_COMPAT mode. 3900 */ 3901 if (ext4_has_feature_ea_inode(sb)) { 3902 ext4_msg(sb, KERN_ERR, 3903 "ea_inode feature is not supported for Hurd"); 3904 goto failed_mount; 3905 } 3906 } 3907 3908 if (IS_EXT2_SB(sb)) { 3909 if (ext2_feature_set_ok(sb)) 3910 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 3911 "using the ext4 subsystem"); 3912 else { 3913 /* 3914 * If we're probing be silent, if this looks like 3915 * it's actually an ext[34] filesystem. 3916 */ 3917 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 3918 goto failed_mount; 3919 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 3920 "to feature incompatibilities"); 3921 goto failed_mount; 3922 } 3923 } 3924 3925 if (IS_EXT3_SB(sb)) { 3926 if (ext3_feature_set_ok(sb)) 3927 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 3928 "using the ext4 subsystem"); 3929 else { 3930 /* 3931 * If we're probing be silent, if this looks like 3932 * it's actually an ext4 filesystem. 3933 */ 3934 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 3935 goto failed_mount; 3936 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 3937 "to feature incompatibilities"); 3938 goto failed_mount; 3939 } 3940 } 3941 3942 /* 3943 * Check feature flags regardless of the revision level, since we 3944 * previously didn't change the revision level when setting the flags, 3945 * so there is a chance incompat flags are set on a rev 0 filesystem. 3946 */ 3947 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 3948 goto failed_mount; 3949 3950 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 3951 if (blocksize < EXT4_MIN_BLOCK_SIZE || 3952 blocksize > EXT4_MAX_BLOCK_SIZE) { 3953 ext4_msg(sb, KERN_ERR, 3954 "Unsupported filesystem blocksize %d (%d log_block_size)", 3955 blocksize, le32_to_cpu(es->s_log_block_size)); 3956 goto failed_mount; 3957 } 3958 if (le32_to_cpu(es->s_log_block_size) > 3959 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 3960 ext4_msg(sb, KERN_ERR, 3961 "Invalid log block size: %u", 3962 le32_to_cpu(es->s_log_block_size)); 3963 goto failed_mount; 3964 } 3965 if (le32_to_cpu(es->s_log_cluster_size) > 3966 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 3967 ext4_msg(sb, KERN_ERR, 3968 "Invalid log cluster size: %u", 3969 le32_to_cpu(es->s_log_cluster_size)); 3970 goto failed_mount; 3971 } 3972 3973 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 3974 ext4_msg(sb, KERN_ERR, 3975 "Number of reserved GDT blocks insanely large: %d", 3976 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 3977 goto failed_mount; 3978 } 3979 3980 if (sbi->s_mount_opt & EXT4_MOUNT_DAX) { 3981 if (ext4_has_feature_inline_data(sb)) { 3982 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 3983 " that may contain inline data"); 3984 goto failed_mount; 3985 } 3986 if (!bdev_dax_supported(sb->s_bdev, blocksize)) { 3987 ext4_msg(sb, KERN_ERR, 3988 "DAX unsupported by block device."); 3989 goto failed_mount; 3990 } 3991 } 3992 3993 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 3994 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 3995 es->s_encryption_level); 3996 goto failed_mount; 3997 } 3998 3999 if (sb->s_blocksize != blocksize) { 4000 /* Validate the filesystem blocksize */ 4001 if (!sb_set_blocksize(sb, blocksize)) { 4002 ext4_msg(sb, KERN_ERR, "bad block size %d", 4003 blocksize); 4004 goto failed_mount; 4005 } 4006 4007 brelse(bh); 4008 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4009 offset = do_div(logical_sb_block, blocksize); 4010 bh = sb_bread_unmovable(sb, logical_sb_block); 4011 if (!bh) { 4012 ext4_msg(sb, KERN_ERR, 4013 "Can't read superblock on 2nd try"); 4014 goto failed_mount; 4015 } 4016 es = (struct ext4_super_block *)(bh->b_data + offset); 4017 sbi->s_es = es; 4018 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4019 ext4_msg(sb, KERN_ERR, 4020 "Magic mismatch, very weird!"); 4021 goto failed_mount; 4022 } 4023 } 4024 4025 has_huge_files = ext4_has_feature_huge_file(sb); 4026 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4027 has_huge_files); 4028 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4029 4030 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4031 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4032 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4033 } else { 4034 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4035 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4036 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4037 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4038 sbi->s_first_ino); 4039 goto failed_mount; 4040 } 4041 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4042 (!is_power_of_2(sbi->s_inode_size)) || 4043 (sbi->s_inode_size > blocksize)) { 4044 ext4_msg(sb, KERN_ERR, 4045 "unsupported inode size: %d", 4046 sbi->s_inode_size); 4047 goto failed_mount; 4048 } 4049 /* 4050 * i_atime_extra is the last extra field available for [acm]times in 4051 * struct ext4_inode. Checking for that field should suffice to ensure 4052 * we have extra space for all three. 4053 */ 4054 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4055 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4056 sb->s_time_gran = 1; 4057 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4058 } else { 4059 sb->s_time_gran = NSEC_PER_SEC; 4060 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4061 } 4062 4063 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4064 } 4065 4066 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4067 if (ext4_has_feature_64bit(sb)) { 4068 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4069 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4070 !is_power_of_2(sbi->s_desc_size)) { 4071 ext4_msg(sb, KERN_ERR, 4072 "unsupported descriptor size %lu", 4073 sbi->s_desc_size); 4074 goto failed_mount; 4075 } 4076 } else 4077 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4078 4079 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4080 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4081 4082 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4083 if (sbi->s_inodes_per_block == 0) 4084 goto cantfind_ext4; 4085 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4086 sbi->s_inodes_per_group > blocksize * 8) { 4087 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4088 sbi->s_blocks_per_group); 4089 goto failed_mount; 4090 } 4091 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4092 sbi->s_inodes_per_block; 4093 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4094 sbi->s_sbh = bh; 4095 sbi->s_mount_state = le16_to_cpu(es->s_state); 4096 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4097 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4098 4099 for (i = 0; i < 4; i++) 4100 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4101 sbi->s_def_hash_version = es->s_def_hash_version; 4102 if (ext4_has_feature_dir_index(sb)) { 4103 i = le32_to_cpu(es->s_flags); 4104 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4105 sbi->s_hash_unsigned = 3; 4106 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4107 #ifdef __CHAR_UNSIGNED__ 4108 if (!sb_rdonly(sb)) 4109 es->s_flags |= 4110 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4111 sbi->s_hash_unsigned = 3; 4112 #else 4113 if (!sb_rdonly(sb)) 4114 es->s_flags |= 4115 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4116 #endif 4117 } 4118 } 4119 4120 /* Handle clustersize */ 4121 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4122 has_bigalloc = ext4_has_feature_bigalloc(sb); 4123 if (has_bigalloc) { 4124 if (clustersize < blocksize) { 4125 ext4_msg(sb, KERN_ERR, 4126 "cluster size (%d) smaller than " 4127 "block size (%d)", clustersize, blocksize); 4128 goto failed_mount; 4129 } 4130 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4131 le32_to_cpu(es->s_log_block_size); 4132 sbi->s_clusters_per_group = 4133 le32_to_cpu(es->s_clusters_per_group); 4134 if (sbi->s_clusters_per_group > blocksize * 8) { 4135 ext4_msg(sb, KERN_ERR, 4136 "#clusters per group too big: %lu", 4137 sbi->s_clusters_per_group); 4138 goto failed_mount; 4139 } 4140 if (sbi->s_blocks_per_group != 4141 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4142 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4143 "clusters per group (%lu) inconsistent", 4144 sbi->s_blocks_per_group, 4145 sbi->s_clusters_per_group); 4146 goto failed_mount; 4147 } 4148 } else { 4149 if (clustersize != blocksize) { 4150 ext4_msg(sb, KERN_ERR, 4151 "fragment/cluster size (%d) != " 4152 "block size (%d)", clustersize, blocksize); 4153 goto failed_mount; 4154 } 4155 if (sbi->s_blocks_per_group > blocksize * 8) { 4156 ext4_msg(sb, KERN_ERR, 4157 "#blocks per group too big: %lu", 4158 sbi->s_blocks_per_group); 4159 goto failed_mount; 4160 } 4161 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4162 sbi->s_cluster_bits = 0; 4163 } 4164 sbi->s_cluster_ratio = clustersize / blocksize; 4165 4166 /* Do we have standard group size of clustersize * 8 blocks ? */ 4167 if (sbi->s_blocks_per_group == clustersize << 3) 4168 set_opt2(sb, STD_GROUP_SIZE); 4169 4170 /* 4171 * Test whether we have more sectors than will fit in sector_t, 4172 * and whether the max offset is addressable by the page cache. 4173 */ 4174 err = generic_check_addressable(sb->s_blocksize_bits, 4175 ext4_blocks_count(es)); 4176 if (err) { 4177 ext4_msg(sb, KERN_ERR, "filesystem" 4178 " too large to mount safely on this system"); 4179 goto failed_mount; 4180 } 4181 4182 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4183 goto cantfind_ext4; 4184 4185 /* check blocks count against device size */ 4186 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4187 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4188 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4189 "exceeds size of device (%llu blocks)", 4190 ext4_blocks_count(es), blocks_count); 4191 goto failed_mount; 4192 } 4193 4194 /* 4195 * It makes no sense for the first data block to be beyond the end 4196 * of the filesystem. 4197 */ 4198 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4199 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4200 "block %u is beyond end of filesystem (%llu)", 4201 le32_to_cpu(es->s_first_data_block), 4202 ext4_blocks_count(es)); 4203 goto failed_mount; 4204 } 4205 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4206 (sbi->s_cluster_ratio == 1)) { 4207 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4208 "block is 0 with a 1k block and cluster size"); 4209 goto failed_mount; 4210 } 4211 4212 blocks_count = (ext4_blocks_count(es) - 4213 le32_to_cpu(es->s_first_data_block) + 4214 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4215 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4216 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4217 ext4_msg(sb, KERN_WARNING, "groups count too large: %u " 4218 "(block count %llu, first data block %u, " 4219 "blocks per group %lu)", sbi->s_groups_count, 4220 ext4_blocks_count(es), 4221 le32_to_cpu(es->s_first_data_block), 4222 EXT4_BLOCKS_PER_GROUP(sb)); 4223 goto failed_mount; 4224 } 4225 sbi->s_groups_count = blocks_count; 4226 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4227 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4228 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4229 le32_to_cpu(es->s_inodes_count)) { 4230 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4231 le32_to_cpu(es->s_inodes_count), 4232 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4233 ret = -EINVAL; 4234 goto failed_mount; 4235 } 4236 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4237 EXT4_DESC_PER_BLOCK(sb); 4238 if (ext4_has_feature_meta_bg(sb)) { 4239 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4240 ext4_msg(sb, KERN_WARNING, 4241 "first meta block group too large: %u " 4242 "(group descriptor block count %u)", 4243 le32_to_cpu(es->s_first_meta_bg), db_count); 4244 goto failed_mount; 4245 } 4246 } 4247 sbi->s_group_desc = kvmalloc_array(db_count, 4248 sizeof(struct buffer_head *), 4249 GFP_KERNEL); 4250 if (sbi->s_group_desc == NULL) { 4251 ext4_msg(sb, KERN_ERR, "not enough memory"); 4252 ret = -ENOMEM; 4253 goto failed_mount; 4254 } 4255 4256 bgl_lock_init(sbi->s_blockgroup_lock); 4257 4258 /* Pre-read the descriptors into the buffer cache */ 4259 for (i = 0; i < db_count; i++) { 4260 block = descriptor_loc(sb, logical_sb_block, i); 4261 sb_breadahead(sb, block); 4262 } 4263 4264 for (i = 0; i < db_count; i++) { 4265 block = descriptor_loc(sb, logical_sb_block, i); 4266 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block); 4267 if (!sbi->s_group_desc[i]) { 4268 ext4_msg(sb, KERN_ERR, 4269 "can't read group descriptor %d", i); 4270 db_count = i; 4271 goto failed_mount2; 4272 } 4273 } 4274 sbi->s_gdb_count = db_count; 4275 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4276 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4277 ret = -EFSCORRUPTED; 4278 goto failed_mount2; 4279 } 4280 4281 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4282 4283 /* Register extent status tree shrinker */ 4284 if (ext4_es_register_shrinker(sbi)) 4285 goto failed_mount3; 4286 4287 sbi->s_stripe = ext4_get_stripe_size(sbi); 4288 sbi->s_extent_max_zeroout_kb = 32; 4289 4290 /* 4291 * set up enough so that it can read an inode 4292 */ 4293 sb->s_op = &ext4_sops; 4294 sb->s_export_op = &ext4_export_ops; 4295 sb->s_xattr = ext4_xattr_handlers; 4296 #ifdef CONFIG_FS_ENCRYPTION 4297 sb->s_cop = &ext4_cryptops; 4298 #endif 4299 #ifdef CONFIG_FS_VERITY 4300 sb->s_vop = &ext4_verityops; 4301 #endif 4302 #ifdef CONFIG_QUOTA 4303 sb->dq_op = &ext4_quota_operations; 4304 if (ext4_has_feature_quota(sb)) 4305 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4306 else 4307 sb->s_qcop = &ext4_qctl_operations; 4308 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4309 #endif 4310 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4311 4312 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4313 mutex_init(&sbi->s_orphan_lock); 4314 4315 sb->s_root = NULL; 4316 4317 needs_recovery = (es->s_last_orphan != 0 || 4318 ext4_has_feature_journal_needs_recovery(sb)); 4319 4320 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4321 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4322 goto failed_mount3a; 4323 4324 /* 4325 * The first inode we look at is the journal inode. Don't try 4326 * root first: it may be modified in the journal! 4327 */ 4328 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4329 err = ext4_load_journal(sb, es, journal_devnum); 4330 if (err) 4331 goto failed_mount3a; 4332 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4333 ext4_has_feature_journal_needs_recovery(sb)) { 4334 ext4_msg(sb, KERN_ERR, "required journal recovery " 4335 "suppressed and not mounted read-only"); 4336 goto failed_mount_wq; 4337 } else { 4338 /* Nojournal mode, all journal mount options are illegal */ 4339 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4340 ext4_msg(sb, KERN_ERR, "can't mount with " 4341 "journal_checksum, fs mounted w/o journal"); 4342 goto failed_mount_wq; 4343 } 4344 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4345 ext4_msg(sb, KERN_ERR, "can't mount with " 4346 "journal_async_commit, fs mounted w/o journal"); 4347 goto failed_mount_wq; 4348 } 4349 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4350 ext4_msg(sb, KERN_ERR, "can't mount with " 4351 "commit=%lu, fs mounted w/o journal", 4352 sbi->s_commit_interval / HZ); 4353 goto failed_mount_wq; 4354 } 4355 if (EXT4_MOUNT_DATA_FLAGS & 4356 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4357 ext4_msg(sb, KERN_ERR, "can't mount with " 4358 "data=, fs mounted w/o journal"); 4359 goto failed_mount_wq; 4360 } 4361 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4362 clear_opt(sb, JOURNAL_CHECKSUM); 4363 clear_opt(sb, DATA_FLAGS); 4364 sbi->s_journal = NULL; 4365 needs_recovery = 0; 4366 goto no_journal; 4367 } 4368 4369 if (ext4_has_feature_64bit(sb) && 4370 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4371 JBD2_FEATURE_INCOMPAT_64BIT)) { 4372 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4373 goto failed_mount_wq; 4374 } 4375 4376 if (!set_journal_csum_feature_set(sb)) { 4377 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4378 "feature set"); 4379 goto failed_mount_wq; 4380 } 4381 4382 /* We have now updated the journal if required, so we can 4383 * validate the data journaling mode. */ 4384 switch (test_opt(sb, DATA_FLAGS)) { 4385 case 0: 4386 /* No mode set, assume a default based on the journal 4387 * capabilities: ORDERED_DATA if the journal can 4388 * cope, else JOURNAL_DATA 4389 */ 4390 if (jbd2_journal_check_available_features 4391 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4392 set_opt(sb, ORDERED_DATA); 4393 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4394 } else { 4395 set_opt(sb, JOURNAL_DATA); 4396 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4397 } 4398 break; 4399 4400 case EXT4_MOUNT_ORDERED_DATA: 4401 case EXT4_MOUNT_WRITEBACK_DATA: 4402 if (!jbd2_journal_check_available_features 4403 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4404 ext4_msg(sb, KERN_ERR, "Journal does not support " 4405 "requested data journaling mode"); 4406 goto failed_mount_wq; 4407 } 4408 default: 4409 break; 4410 } 4411 4412 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4413 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4414 ext4_msg(sb, KERN_ERR, "can't mount with " 4415 "journal_async_commit in data=ordered mode"); 4416 goto failed_mount_wq; 4417 } 4418 4419 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 4420 4421 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback; 4422 4423 no_journal: 4424 if (!test_opt(sb, NO_MBCACHE)) { 4425 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4426 if (!sbi->s_ea_block_cache) { 4427 ext4_msg(sb, KERN_ERR, 4428 "Failed to create ea_block_cache"); 4429 goto failed_mount_wq; 4430 } 4431 4432 if (ext4_has_feature_ea_inode(sb)) { 4433 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4434 if (!sbi->s_ea_inode_cache) { 4435 ext4_msg(sb, KERN_ERR, 4436 "Failed to create ea_inode_cache"); 4437 goto failed_mount_wq; 4438 } 4439 } 4440 } 4441 4442 if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) && 4443 (blocksize != PAGE_SIZE)) { 4444 ext4_msg(sb, KERN_ERR, 4445 "Unsupported blocksize for fs encryption"); 4446 goto failed_mount_wq; 4447 } 4448 4449 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4450 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4451 goto failed_mount_wq; 4452 } 4453 4454 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4455 !ext4_has_feature_encrypt(sb)) { 4456 ext4_set_feature_encrypt(sb); 4457 ext4_commit_super(sb, 1); 4458 } 4459 4460 /* 4461 * Get the # of file system overhead blocks from the 4462 * superblock if present. 4463 */ 4464 if (es->s_overhead_clusters) 4465 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4466 else { 4467 err = ext4_calculate_overhead(sb); 4468 if (err) 4469 goto failed_mount_wq; 4470 } 4471 4472 /* 4473 * The maximum number of concurrent works can be high and 4474 * concurrency isn't really necessary. Limit it to 1. 4475 */ 4476 EXT4_SB(sb)->rsv_conversion_wq = 4477 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4478 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4479 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4480 ret = -ENOMEM; 4481 goto failed_mount4; 4482 } 4483 4484 /* 4485 * The jbd2_journal_load will have done any necessary log recovery, 4486 * so we can safely mount the rest of the filesystem now. 4487 */ 4488 4489 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4490 if (IS_ERR(root)) { 4491 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4492 ret = PTR_ERR(root); 4493 root = NULL; 4494 goto failed_mount4; 4495 } 4496 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4497 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4498 iput(root); 4499 goto failed_mount4; 4500 } 4501 4502 #ifdef CONFIG_UNICODE 4503 if (sbi->s_encoding) 4504 sb->s_d_op = &ext4_dentry_ops; 4505 #endif 4506 4507 sb->s_root = d_make_root(root); 4508 if (!sb->s_root) { 4509 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4510 ret = -ENOMEM; 4511 goto failed_mount4; 4512 } 4513 4514 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4515 if (ret == -EROFS) { 4516 sb->s_flags |= SB_RDONLY; 4517 ret = 0; 4518 } else if (ret) 4519 goto failed_mount4a; 4520 4521 ext4_clamp_want_extra_isize(sb); 4522 4523 ext4_set_resv_clusters(sb); 4524 4525 err = ext4_setup_system_zone(sb); 4526 if (err) { 4527 ext4_msg(sb, KERN_ERR, "failed to initialize system " 4528 "zone (%d)", err); 4529 goto failed_mount4a; 4530 } 4531 4532 ext4_ext_init(sb); 4533 err = ext4_mb_init(sb); 4534 if (err) { 4535 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 4536 err); 4537 goto failed_mount5; 4538 } 4539 4540 block = ext4_count_free_clusters(sb); 4541 ext4_free_blocks_count_set(sbi->s_es, 4542 EXT4_C2B(sbi, block)); 4543 ext4_superblock_csum_set(sb); 4544 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 4545 GFP_KERNEL); 4546 if (!err) { 4547 unsigned long freei = ext4_count_free_inodes(sb); 4548 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 4549 ext4_superblock_csum_set(sb); 4550 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 4551 GFP_KERNEL); 4552 } 4553 if (!err) 4554 err = percpu_counter_init(&sbi->s_dirs_counter, 4555 ext4_count_dirs(sb), GFP_KERNEL); 4556 if (!err) 4557 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 4558 GFP_KERNEL); 4559 if (!err) 4560 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem); 4561 4562 if (err) { 4563 ext4_msg(sb, KERN_ERR, "insufficient memory"); 4564 goto failed_mount6; 4565 } 4566 4567 if (ext4_has_feature_flex_bg(sb)) 4568 if (!ext4_fill_flex_info(sb)) { 4569 ext4_msg(sb, KERN_ERR, 4570 "unable to initialize " 4571 "flex_bg meta info!"); 4572 goto failed_mount6; 4573 } 4574 4575 err = ext4_register_li_request(sb, first_not_zeroed); 4576 if (err) 4577 goto failed_mount6; 4578 4579 err = ext4_register_sysfs(sb); 4580 if (err) 4581 goto failed_mount7; 4582 4583 #ifdef CONFIG_QUOTA 4584 /* Enable quota usage during mount. */ 4585 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 4586 err = ext4_enable_quotas(sb); 4587 if (err) 4588 goto failed_mount8; 4589 } 4590 #endif /* CONFIG_QUOTA */ 4591 4592 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 4593 ext4_orphan_cleanup(sb, es); 4594 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 4595 if (needs_recovery) { 4596 ext4_msg(sb, KERN_INFO, "recovery complete"); 4597 ext4_mark_recovery_complete(sb, es); 4598 } 4599 if (EXT4_SB(sb)->s_journal) { 4600 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 4601 descr = " journalled data mode"; 4602 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 4603 descr = " ordered data mode"; 4604 else 4605 descr = " writeback data mode"; 4606 } else 4607 descr = "out journal"; 4608 4609 if (test_opt(sb, DISCARD)) { 4610 struct request_queue *q = bdev_get_queue(sb->s_bdev); 4611 if (!blk_queue_discard(q)) 4612 ext4_msg(sb, KERN_WARNING, 4613 "mounting with \"discard\" option, but " 4614 "the device does not support discard"); 4615 } 4616 4617 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 4618 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 4619 "Opts: %.*s%s%s", descr, 4620 (int) sizeof(sbi->s_es->s_mount_opts), 4621 sbi->s_es->s_mount_opts, 4622 *sbi->s_es->s_mount_opts ? "; " : "", orig_data); 4623 4624 if (es->s_error_count) 4625 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 4626 4627 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 4628 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 4629 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 4630 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 4631 4632 kfree(orig_data); 4633 return 0; 4634 4635 cantfind_ext4: 4636 if (!silent) 4637 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 4638 goto failed_mount; 4639 4640 #ifdef CONFIG_QUOTA 4641 failed_mount8: 4642 ext4_unregister_sysfs(sb); 4643 #endif 4644 failed_mount7: 4645 ext4_unregister_li_request(sb); 4646 failed_mount6: 4647 ext4_mb_release(sb); 4648 if (sbi->s_flex_groups) 4649 kvfree(sbi->s_flex_groups); 4650 percpu_counter_destroy(&sbi->s_freeclusters_counter); 4651 percpu_counter_destroy(&sbi->s_freeinodes_counter); 4652 percpu_counter_destroy(&sbi->s_dirs_counter); 4653 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 4654 percpu_free_rwsem(&sbi->s_journal_flag_rwsem); 4655 failed_mount5: 4656 ext4_ext_release(sb); 4657 ext4_release_system_zone(sb); 4658 failed_mount4a: 4659 dput(sb->s_root); 4660 sb->s_root = NULL; 4661 failed_mount4: 4662 ext4_msg(sb, KERN_ERR, "mount failed"); 4663 if (EXT4_SB(sb)->rsv_conversion_wq) 4664 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 4665 failed_mount_wq: 4666 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 4667 sbi->s_ea_inode_cache = NULL; 4668 4669 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 4670 sbi->s_ea_block_cache = NULL; 4671 4672 if (sbi->s_journal) { 4673 jbd2_journal_destroy(sbi->s_journal); 4674 sbi->s_journal = NULL; 4675 } 4676 failed_mount3a: 4677 ext4_es_unregister_shrinker(sbi); 4678 failed_mount3: 4679 del_timer_sync(&sbi->s_err_report); 4680 if (sbi->s_mmp_tsk) 4681 kthread_stop(sbi->s_mmp_tsk); 4682 failed_mount2: 4683 for (i = 0; i < db_count; i++) 4684 brelse(sbi->s_group_desc[i]); 4685 kvfree(sbi->s_group_desc); 4686 failed_mount: 4687 if (sbi->s_chksum_driver) 4688 crypto_free_shash(sbi->s_chksum_driver); 4689 4690 #ifdef CONFIG_UNICODE 4691 utf8_unload(sbi->s_encoding); 4692 #endif 4693 4694 #ifdef CONFIG_QUOTA 4695 for (i = 0; i < EXT4_MAXQUOTAS; i++) 4696 kfree(get_qf_name(sb, sbi, i)); 4697 #endif 4698 ext4_blkdev_remove(sbi); 4699 brelse(bh); 4700 out_fail: 4701 sb->s_fs_info = NULL; 4702 kfree(sbi->s_blockgroup_lock); 4703 out_free_base: 4704 kfree(sbi); 4705 kfree(orig_data); 4706 fs_put_dax(dax_dev); 4707 return err ? err : ret; 4708 } 4709 4710 /* 4711 * Setup any per-fs journal parameters now. We'll do this both on 4712 * initial mount, once the journal has been initialised but before we've 4713 * done any recovery; and again on any subsequent remount. 4714 */ 4715 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 4716 { 4717 struct ext4_sb_info *sbi = EXT4_SB(sb); 4718 4719 journal->j_commit_interval = sbi->s_commit_interval; 4720 journal->j_min_batch_time = sbi->s_min_batch_time; 4721 journal->j_max_batch_time = sbi->s_max_batch_time; 4722 4723 write_lock(&journal->j_state_lock); 4724 if (test_opt(sb, BARRIER)) 4725 journal->j_flags |= JBD2_BARRIER; 4726 else 4727 journal->j_flags &= ~JBD2_BARRIER; 4728 if (test_opt(sb, DATA_ERR_ABORT)) 4729 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 4730 else 4731 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 4732 write_unlock(&journal->j_state_lock); 4733 } 4734 4735 static struct inode *ext4_get_journal_inode(struct super_block *sb, 4736 unsigned int journal_inum) 4737 { 4738 struct inode *journal_inode; 4739 4740 /* 4741 * Test for the existence of a valid inode on disk. Bad things 4742 * happen if we iget() an unused inode, as the subsequent iput() 4743 * will try to delete it. 4744 */ 4745 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 4746 if (IS_ERR(journal_inode)) { 4747 ext4_msg(sb, KERN_ERR, "no journal found"); 4748 return NULL; 4749 } 4750 if (!journal_inode->i_nlink) { 4751 make_bad_inode(journal_inode); 4752 iput(journal_inode); 4753 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 4754 return NULL; 4755 } 4756 4757 jbd_debug(2, "Journal inode found at %p: %lld bytes\n", 4758 journal_inode, journal_inode->i_size); 4759 if (!S_ISREG(journal_inode->i_mode)) { 4760 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 4761 iput(journal_inode); 4762 return NULL; 4763 } 4764 return journal_inode; 4765 } 4766 4767 static journal_t *ext4_get_journal(struct super_block *sb, 4768 unsigned int journal_inum) 4769 { 4770 struct inode *journal_inode; 4771 journal_t *journal; 4772 4773 BUG_ON(!ext4_has_feature_journal(sb)); 4774 4775 journal_inode = ext4_get_journal_inode(sb, journal_inum); 4776 if (!journal_inode) 4777 return NULL; 4778 4779 journal = jbd2_journal_init_inode(journal_inode); 4780 if (!journal) { 4781 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 4782 iput(journal_inode); 4783 return NULL; 4784 } 4785 journal->j_private = sb; 4786 ext4_init_journal_params(sb, journal); 4787 return journal; 4788 } 4789 4790 static journal_t *ext4_get_dev_journal(struct super_block *sb, 4791 dev_t j_dev) 4792 { 4793 struct buffer_head *bh; 4794 journal_t *journal; 4795 ext4_fsblk_t start; 4796 ext4_fsblk_t len; 4797 int hblock, blocksize; 4798 ext4_fsblk_t sb_block; 4799 unsigned long offset; 4800 struct ext4_super_block *es; 4801 struct block_device *bdev; 4802 4803 BUG_ON(!ext4_has_feature_journal(sb)); 4804 4805 bdev = ext4_blkdev_get(j_dev, sb); 4806 if (bdev == NULL) 4807 return NULL; 4808 4809 blocksize = sb->s_blocksize; 4810 hblock = bdev_logical_block_size(bdev); 4811 if (blocksize < hblock) { 4812 ext4_msg(sb, KERN_ERR, 4813 "blocksize too small for journal device"); 4814 goto out_bdev; 4815 } 4816 4817 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 4818 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 4819 set_blocksize(bdev, blocksize); 4820 if (!(bh = __bread(bdev, sb_block, blocksize))) { 4821 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 4822 "external journal"); 4823 goto out_bdev; 4824 } 4825 4826 es = (struct ext4_super_block *) (bh->b_data + offset); 4827 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 4828 !(le32_to_cpu(es->s_feature_incompat) & 4829 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 4830 ext4_msg(sb, KERN_ERR, "external journal has " 4831 "bad superblock"); 4832 brelse(bh); 4833 goto out_bdev; 4834 } 4835 4836 if ((le32_to_cpu(es->s_feature_ro_compat) & 4837 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 4838 es->s_checksum != ext4_superblock_csum(sb, es)) { 4839 ext4_msg(sb, KERN_ERR, "external journal has " 4840 "corrupt superblock"); 4841 brelse(bh); 4842 goto out_bdev; 4843 } 4844 4845 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 4846 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 4847 brelse(bh); 4848 goto out_bdev; 4849 } 4850 4851 len = ext4_blocks_count(es); 4852 start = sb_block + 1; 4853 brelse(bh); /* we're done with the superblock */ 4854 4855 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 4856 start, len, blocksize); 4857 if (!journal) { 4858 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 4859 goto out_bdev; 4860 } 4861 journal->j_private = sb; 4862 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer); 4863 wait_on_buffer(journal->j_sb_buffer); 4864 if (!buffer_uptodate(journal->j_sb_buffer)) { 4865 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 4866 goto out_journal; 4867 } 4868 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 4869 ext4_msg(sb, KERN_ERR, "External journal has more than one " 4870 "user (unsupported) - %d", 4871 be32_to_cpu(journal->j_superblock->s_nr_users)); 4872 goto out_journal; 4873 } 4874 EXT4_SB(sb)->journal_bdev = bdev; 4875 ext4_init_journal_params(sb, journal); 4876 return journal; 4877 4878 out_journal: 4879 jbd2_journal_destroy(journal); 4880 out_bdev: 4881 ext4_blkdev_put(bdev); 4882 return NULL; 4883 } 4884 4885 static int ext4_load_journal(struct super_block *sb, 4886 struct ext4_super_block *es, 4887 unsigned long journal_devnum) 4888 { 4889 journal_t *journal; 4890 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 4891 dev_t journal_dev; 4892 int err = 0; 4893 int really_read_only; 4894 4895 BUG_ON(!ext4_has_feature_journal(sb)); 4896 4897 if (journal_devnum && 4898 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 4899 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 4900 "numbers have changed"); 4901 journal_dev = new_decode_dev(journal_devnum); 4902 } else 4903 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 4904 4905 really_read_only = bdev_read_only(sb->s_bdev); 4906 4907 /* 4908 * Are we loading a blank journal or performing recovery after a 4909 * crash? For recovery, we need to check in advance whether we 4910 * can get read-write access to the device. 4911 */ 4912 if (ext4_has_feature_journal_needs_recovery(sb)) { 4913 if (sb_rdonly(sb)) { 4914 ext4_msg(sb, KERN_INFO, "INFO: recovery " 4915 "required on readonly filesystem"); 4916 if (really_read_only) { 4917 ext4_msg(sb, KERN_ERR, "write access " 4918 "unavailable, cannot proceed " 4919 "(try mounting with noload)"); 4920 return -EROFS; 4921 } 4922 ext4_msg(sb, KERN_INFO, "write access will " 4923 "be enabled during recovery"); 4924 } 4925 } 4926 4927 if (journal_inum && journal_dev) { 4928 ext4_msg(sb, KERN_ERR, "filesystem has both journal " 4929 "and inode journals!"); 4930 return -EINVAL; 4931 } 4932 4933 if (journal_inum) { 4934 if (!(journal = ext4_get_journal(sb, journal_inum))) 4935 return -EINVAL; 4936 } else { 4937 if (!(journal = ext4_get_dev_journal(sb, journal_dev))) 4938 return -EINVAL; 4939 } 4940 4941 if (!(journal->j_flags & JBD2_BARRIER)) 4942 ext4_msg(sb, KERN_INFO, "barriers disabled"); 4943 4944 if (!ext4_has_feature_journal_needs_recovery(sb)) 4945 err = jbd2_journal_wipe(journal, !really_read_only); 4946 if (!err) { 4947 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 4948 if (save) 4949 memcpy(save, ((char *) es) + 4950 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 4951 err = jbd2_journal_load(journal); 4952 if (save) 4953 memcpy(((char *) es) + EXT4_S_ERR_START, 4954 save, EXT4_S_ERR_LEN); 4955 kfree(save); 4956 } 4957 4958 if (err) { 4959 ext4_msg(sb, KERN_ERR, "error loading journal"); 4960 jbd2_journal_destroy(journal); 4961 return err; 4962 } 4963 4964 EXT4_SB(sb)->s_journal = journal; 4965 ext4_clear_journal_err(sb, es); 4966 4967 if (!really_read_only && journal_devnum && 4968 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 4969 es->s_journal_dev = cpu_to_le32(journal_devnum); 4970 4971 /* Make sure we flush the recovery flag to disk. */ 4972 ext4_commit_super(sb, 1); 4973 } 4974 4975 return 0; 4976 } 4977 4978 static int ext4_commit_super(struct super_block *sb, int sync) 4979 { 4980 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 4981 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 4982 int error = 0; 4983 4984 if (!sbh || block_device_ejected(sb)) 4985 return error; 4986 4987 /* 4988 * The superblock bh should be mapped, but it might not be if the 4989 * device was hot-removed. Not much we can do but fail the I/O. 4990 */ 4991 if (!buffer_mapped(sbh)) 4992 return error; 4993 4994 /* 4995 * If the file system is mounted read-only, don't update the 4996 * superblock write time. This avoids updating the superblock 4997 * write time when we are mounting the root file system 4998 * read/only but we need to replay the journal; at that point, 4999 * for people who are east of GMT and who make their clock 5000 * tick in localtime for Windows bug-for-bug compatibility, 5001 * the clock is set in the future, and this will cause e2fsck 5002 * to complain and force a full file system check. 5003 */ 5004 if (!(sb->s_flags & SB_RDONLY)) 5005 ext4_update_tstamp(es, s_wtime); 5006 if (sb->s_bdev->bd_part) 5007 es->s_kbytes_written = 5008 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 5009 ((part_stat_read(sb->s_bdev->bd_part, 5010 sectors[STAT_WRITE]) - 5011 EXT4_SB(sb)->s_sectors_written_start) >> 1)); 5012 else 5013 es->s_kbytes_written = 5014 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written); 5015 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter)) 5016 ext4_free_blocks_count_set(es, 5017 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive( 5018 &EXT4_SB(sb)->s_freeclusters_counter))); 5019 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter)) 5020 es->s_free_inodes_count = 5021 cpu_to_le32(percpu_counter_sum_positive( 5022 &EXT4_SB(sb)->s_freeinodes_counter)); 5023 BUFFER_TRACE(sbh, "marking dirty"); 5024 ext4_superblock_csum_set(sb); 5025 if (sync) 5026 lock_buffer(sbh); 5027 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 5028 /* 5029 * Oh, dear. A previous attempt to write the 5030 * superblock failed. This could happen because the 5031 * USB device was yanked out. Or it could happen to 5032 * be a transient write error and maybe the block will 5033 * be remapped. Nothing we can do but to retry the 5034 * write and hope for the best. 5035 */ 5036 ext4_msg(sb, KERN_ERR, "previous I/O error to " 5037 "superblock detected"); 5038 clear_buffer_write_io_error(sbh); 5039 set_buffer_uptodate(sbh); 5040 } 5041 mark_buffer_dirty(sbh); 5042 if (sync) { 5043 unlock_buffer(sbh); 5044 error = __sync_dirty_buffer(sbh, 5045 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0)); 5046 if (buffer_write_io_error(sbh)) { 5047 ext4_msg(sb, KERN_ERR, "I/O error while writing " 5048 "superblock"); 5049 clear_buffer_write_io_error(sbh); 5050 set_buffer_uptodate(sbh); 5051 } 5052 } 5053 return error; 5054 } 5055 5056 /* 5057 * Have we just finished recovery? If so, and if we are mounting (or 5058 * remounting) the filesystem readonly, then we will end up with a 5059 * consistent fs on disk. Record that fact. 5060 */ 5061 static void ext4_mark_recovery_complete(struct super_block *sb, 5062 struct ext4_super_block *es) 5063 { 5064 journal_t *journal = EXT4_SB(sb)->s_journal; 5065 5066 if (!ext4_has_feature_journal(sb)) { 5067 BUG_ON(journal != NULL); 5068 return; 5069 } 5070 jbd2_journal_lock_updates(journal); 5071 if (jbd2_journal_flush(journal) < 0) 5072 goto out; 5073 5074 if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) { 5075 ext4_clear_feature_journal_needs_recovery(sb); 5076 ext4_commit_super(sb, 1); 5077 } 5078 5079 out: 5080 jbd2_journal_unlock_updates(journal); 5081 } 5082 5083 /* 5084 * If we are mounting (or read-write remounting) a filesystem whose journal 5085 * has recorded an error from a previous lifetime, move that error to the 5086 * main filesystem now. 5087 */ 5088 static void ext4_clear_journal_err(struct super_block *sb, 5089 struct ext4_super_block *es) 5090 { 5091 journal_t *journal; 5092 int j_errno; 5093 const char *errstr; 5094 5095 BUG_ON(!ext4_has_feature_journal(sb)); 5096 5097 journal = EXT4_SB(sb)->s_journal; 5098 5099 /* 5100 * Now check for any error status which may have been recorded in the 5101 * journal by a prior ext4_error() or ext4_abort() 5102 */ 5103 5104 j_errno = jbd2_journal_errno(journal); 5105 if (j_errno) { 5106 char nbuf[16]; 5107 5108 errstr = ext4_decode_error(sb, j_errno, nbuf); 5109 ext4_warning(sb, "Filesystem error recorded " 5110 "from previous mount: %s", errstr); 5111 ext4_warning(sb, "Marking fs in need of filesystem check."); 5112 5113 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 5114 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5115 ext4_commit_super(sb, 1); 5116 5117 jbd2_journal_clear_err(journal); 5118 jbd2_journal_update_sb_errno(journal); 5119 } 5120 } 5121 5122 /* 5123 * Force the running and committing transactions to commit, 5124 * and wait on the commit. 5125 */ 5126 int ext4_force_commit(struct super_block *sb) 5127 { 5128 journal_t *journal; 5129 5130 if (sb_rdonly(sb)) 5131 return 0; 5132 5133 journal = EXT4_SB(sb)->s_journal; 5134 return ext4_journal_force_commit(journal); 5135 } 5136 5137 static int ext4_sync_fs(struct super_block *sb, int wait) 5138 { 5139 int ret = 0; 5140 tid_t target; 5141 bool needs_barrier = false; 5142 struct ext4_sb_info *sbi = EXT4_SB(sb); 5143 5144 if (unlikely(ext4_forced_shutdown(sbi))) 5145 return 0; 5146 5147 trace_ext4_sync_fs(sb, wait); 5148 flush_workqueue(sbi->rsv_conversion_wq); 5149 /* 5150 * Writeback quota in non-journalled quota case - journalled quota has 5151 * no dirty dquots 5152 */ 5153 dquot_writeback_dquots(sb, -1); 5154 /* 5155 * Data writeback is possible w/o journal transaction, so barrier must 5156 * being sent at the end of the function. But we can skip it if 5157 * transaction_commit will do it for us. 5158 */ 5159 if (sbi->s_journal) { 5160 target = jbd2_get_latest_transaction(sbi->s_journal); 5161 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 5162 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 5163 needs_barrier = true; 5164 5165 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 5166 if (wait) 5167 ret = jbd2_log_wait_commit(sbi->s_journal, 5168 target); 5169 } 5170 } else if (wait && test_opt(sb, BARRIER)) 5171 needs_barrier = true; 5172 if (needs_barrier) { 5173 int err; 5174 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL); 5175 if (!ret) 5176 ret = err; 5177 } 5178 5179 return ret; 5180 } 5181 5182 /* 5183 * LVM calls this function before a (read-only) snapshot is created. This 5184 * gives us a chance to flush the journal completely and mark the fs clean. 5185 * 5186 * Note that only this function cannot bring a filesystem to be in a clean 5187 * state independently. It relies on upper layer to stop all data & metadata 5188 * modifications. 5189 */ 5190 static int ext4_freeze(struct super_block *sb) 5191 { 5192 int error = 0; 5193 journal_t *journal; 5194 5195 if (sb_rdonly(sb)) 5196 return 0; 5197 5198 journal = EXT4_SB(sb)->s_journal; 5199 5200 if (journal) { 5201 /* Now we set up the journal barrier. */ 5202 jbd2_journal_lock_updates(journal); 5203 5204 /* 5205 * Don't clear the needs_recovery flag if we failed to 5206 * flush the journal. 5207 */ 5208 error = jbd2_journal_flush(journal); 5209 if (error < 0) 5210 goto out; 5211 5212 /* Journal blocked and flushed, clear needs_recovery flag. */ 5213 ext4_clear_feature_journal_needs_recovery(sb); 5214 } 5215 5216 error = ext4_commit_super(sb, 1); 5217 out: 5218 if (journal) 5219 /* we rely on upper layer to stop further updates */ 5220 jbd2_journal_unlock_updates(journal); 5221 return error; 5222 } 5223 5224 /* 5225 * Called by LVM after the snapshot is done. We need to reset the RECOVER 5226 * flag here, even though the filesystem is not technically dirty yet. 5227 */ 5228 static int ext4_unfreeze(struct super_block *sb) 5229 { 5230 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 5231 return 0; 5232 5233 if (EXT4_SB(sb)->s_journal) { 5234 /* Reset the needs_recovery flag before the fs is unlocked. */ 5235 ext4_set_feature_journal_needs_recovery(sb); 5236 } 5237 5238 ext4_commit_super(sb, 1); 5239 return 0; 5240 } 5241 5242 /* 5243 * Structure to save mount options for ext4_remount's benefit 5244 */ 5245 struct ext4_mount_options { 5246 unsigned long s_mount_opt; 5247 unsigned long s_mount_opt2; 5248 kuid_t s_resuid; 5249 kgid_t s_resgid; 5250 unsigned long s_commit_interval; 5251 u32 s_min_batch_time, s_max_batch_time; 5252 #ifdef CONFIG_QUOTA 5253 int s_jquota_fmt; 5254 char *s_qf_names[EXT4_MAXQUOTAS]; 5255 #endif 5256 }; 5257 5258 static int ext4_remount(struct super_block *sb, int *flags, char *data) 5259 { 5260 struct ext4_super_block *es; 5261 struct ext4_sb_info *sbi = EXT4_SB(sb); 5262 unsigned long old_sb_flags; 5263 struct ext4_mount_options old_opts; 5264 int enable_quota = 0; 5265 ext4_group_t g; 5266 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5267 int err = 0; 5268 #ifdef CONFIG_QUOTA 5269 int i, j; 5270 char *to_free[EXT4_MAXQUOTAS]; 5271 #endif 5272 char *orig_data = kstrdup(data, GFP_KERNEL); 5273 5274 if (data && !orig_data) 5275 return -ENOMEM; 5276 5277 /* Store the original options */ 5278 old_sb_flags = sb->s_flags; 5279 old_opts.s_mount_opt = sbi->s_mount_opt; 5280 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 5281 old_opts.s_resuid = sbi->s_resuid; 5282 old_opts.s_resgid = sbi->s_resgid; 5283 old_opts.s_commit_interval = sbi->s_commit_interval; 5284 old_opts.s_min_batch_time = sbi->s_min_batch_time; 5285 old_opts.s_max_batch_time = sbi->s_max_batch_time; 5286 #ifdef CONFIG_QUOTA 5287 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 5288 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5289 if (sbi->s_qf_names[i]) { 5290 char *qf_name = get_qf_name(sb, sbi, i); 5291 5292 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 5293 if (!old_opts.s_qf_names[i]) { 5294 for (j = 0; j < i; j++) 5295 kfree(old_opts.s_qf_names[j]); 5296 kfree(orig_data); 5297 return -ENOMEM; 5298 } 5299 } else 5300 old_opts.s_qf_names[i] = NULL; 5301 #endif 5302 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 5303 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio; 5304 5305 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) { 5306 err = -EINVAL; 5307 goto restore_opts; 5308 } 5309 5310 ext4_clamp_want_extra_isize(sb); 5311 5312 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 5313 test_opt(sb, JOURNAL_CHECKSUM)) { 5314 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 5315 "during remount not supported; ignoring"); 5316 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 5317 } 5318 5319 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5320 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5321 ext4_msg(sb, KERN_ERR, "can't mount with " 5322 "both data=journal and delalloc"); 5323 err = -EINVAL; 5324 goto restore_opts; 5325 } 5326 if (test_opt(sb, DIOREAD_NOLOCK)) { 5327 ext4_msg(sb, KERN_ERR, "can't mount with " 5328 "both data=journal and dioread_nolock"); 5329 err = -EINVAL; 5330 goto restore_opts; 5331 } 5332 if (test_opt(sb, DAX)) { 5333 ext4_msg(sb, KERN_ERR, "can't mount with " 5334 "both data=journal and dax"); 5335 err = -EINVAL; 5336 goto restore_opts; 5337 } 5338 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 5339 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5340 ext4_msg(sb, KERN_ERR, "can't mount with " 5341 "journal_async_commit in data=ordered mode"); 5342 err = -EINVAL; 5343 goto restore_opts; 5344 } 5345 } 5346 5347 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 5348 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 5349 err = -EINVAL; 5350 goto restore_opts; 5351 } 5352 5353 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) { 5354 ext4_msg(sb, KERN_WARNING, "warning: refusing change of " 5355 "dax flag with busy inodes while remounting"); 5356 sbi->s_mount_opt ^= EXT4_MOUNT_DAX; 5357 } 5358 5359 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) 5360 ext4_abort(sb, "Abort forced by user"); 5361 5362 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5363 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5364 5365 es = sbi->s_es; 5366 5367 if (sbi->s_journal) { 5368 ext4_init_journal_params(sb, sbi->s_journal); 5369 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 5370 } 5371 5372 if (*flags & SB_LAZYTIME) 5373 sb->s_flags |= SB_LAZYTIME; 5374 5375 if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) { 5376 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) { 5377 err = -EROFS; 5378 goto restore_opts; 5379 } 5380 5381 if (*flags & SB_RDONLY) { 5382 err = sync_filesystem(sb); 5383 if (err < 0) 5384 goto restore_opts; 5385 err = dquot_suspend(sb, -1); 5386 if (err < 0) 5387 goto restore_opts; 5388 5389 /* 5390 * First of all, the unconditional stuff we have to do 5391 * to disable replay of the journal when we next remount 5392 */ 5393 sb->s_flags |= SB_RDONLY; 5394 5395 /* 5396 * OK, test if we are remounting a valid rw partition 5397 * readonly, and if so set the rdonly flag and then 5398 * mark the partition as valid again. 5399 */ 5400 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 5401 (sbi->s_mount_state & EXT4_VALID_FS)) 5402 es->s_state = cpu_to_le16(sbi->s_mount_state); 5403 5404 if (sbi->s_journal) 5405 ext4_mark_recovery_complete(sb, es); 5406 if (sbi->s_mmp_tsk) 5407 kthread_stop(sbi->s_mmp_tsk); 5408 } else { 5409 /* Make sure we can mount this feature set readwrite */ 5410 if (ext4_has_feature_readonly(sb) || 5411 !ext4_feature_set_ok(sb, 0)) { 5412 err = -EROFS; 5413 goto restore_opts; 5414 } 5415 /* 5416 * Make sure the group descriptor checksums 5417 * are sane. If they aren't, refuse to remount r/w. 5418 */ 5419 for (g = 0; g < sbi->s_groups_count; g++) { 5420 struct ext4_group_desc *gdp = 5421 ext4_get_group_desc(sb, g, NULL); 5422 5423 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 5424 ext4_msg(sb, KERN_ERR, 5425 "ext4_remount: Checksum for group %u failed (%u!=%u)", 5426 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 5427 le16_to_cpu(gdp->bg_checksum)); 5428 err = -EFSBADCRC; 5429 goto restore_opts; 5430 } 5431 } 5432 5433 /* 5434 * If we have an unprocessed orphan list hanging 5435 * around from a previously readonly bdev mount, 5436 * require a full umount/remount for now. 5437 */ 5438 if (es->s_last_orphan) { 5439 ext4_msg(sb, KERN_WARNING, "Couldn't " 5440 "remount RDWR because of unprocessed " 5441 "orphan inode list. Please " 5442 "umount/remount instead"); 5443 err = -EINVAL; 5444 goto restore_opts; 5445 } 5446 5447 /* 5448 * Mounting a RDONLY partition read-write, so reread 5449 * and store the current valid flag. (It may have 5450 * been changed by e2fsck since we originally mounted 5451 * the partition.) 5452 */ 5453 if (sbi->s_journal) 5454 ext4_clear_journal_err(sb, es); 5455 sbi->s_mount_state = le16_to_cpu(es->s_state); 5456 5457 err = ext4_setup_super(sb, es, 0); 5458 if (err) 5459 goto restore_opts; 5460 5461 sb->s_flags &= ~SB_RDONLY; 5462 if (ext4_has_feature_mmp(sb)) 5463 if (ext4_multi_mount_protect(sb, 5464 le64_to_cpu(es->s_mmp_block))) { 5465 err = -EROFS; 5466 goto restore_opts; 5467 } 5468 enable_quota = 1; 5469 } 5470 } 5471 5472 /* 5473 * Reinitialize lazy itable initialization thread based on 5474 * current settings 5475 */ 5476 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 5477 ext4_unregister_li_request(sb); 5478 else { 5479 ext4_group_t first_not_zeroed; 5480 first_not_zeroed = ext4_has_uninit_itable(sb); 5481 ext4_register_li_request(sb, first_not_zeroed); 5482 } 5483 5484 ext4_setup_system_zone(sb); 5485 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 5486 err = ext4_commit_super(sb, 1); 5487 if (err) 5488 goto restore_opts; 5489 } 5490 5491 #ifdef CONFIG_QUOTA 5492 /* Release old quota file names */ 5493 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5494 kfree(old_opts.s_qf_names[i]); 5495 if (enable_quota) { 5496 if (sb_any_quota_suspended(sb)) 5497 dquot_resume(sb, -1); 5498 else if (ext4_has_feature_quota(sb)) { 5499 err = ext4_enable_quotas(sb); 5500 if (err) 5501 goto restore_opts; 5502 } 5503 } 5504 #endif 5505 5506 *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME); 5507 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data); 5508 kfree(orig_data); 5509 return 0; 5510 5511 restore_opts: 5512 sb->s_flags = old_sb_flags; 5513 sbi->s_mount_opt = old_opts.s_mount_opt; 5514 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 5515 sbi->s_resuid = old_opts.s_resuid; 5516 sbi->s_resgid = old_opts.s_resgid; 5517 sbi->s_commit_interval = old_opts.s_commit_interval; 5518 sbi->s_min_batch_time = old_opts.s_min_batch_time; 5519 sbi->s_max_batch_time = old_opts.s_max_batch_time; 5520 #ifdef CONFIG_QUOTA 5521 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 5522 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 5523 to_free[i] = get_qf_name(sb, sbi, i); 5524 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 5525 } 5526 synchronize_rcu(); 5527 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5528 kfree(to_free[i]); 5529 #endif 5530 kfree(orig_data); 5531 return err; 5532 } 5533 5534 #ifdef CONFIG_QUOTA 5535 static int ext4_statfs_project(struct super_block *sb, 5536 kprojid_t projid, struct kstatfs *buf) 5537 { 5538 struct kqid qid; 5539 struct dquot *dquot; 5540 u64 limit; 5541 u64 curblock; 5542 5543 qid = make_kqid_projid(projid); 5544 dquot = dqget(sb, qid); 5545 if (IS_ERR(dquot)) 5546 return PTR_ERR(dquot); 5547 spin_lock(&dquot->dq_dqb_lock); 5548 5549 limit = (dquot->dq_dqb.dqb_bsoftlimit ? 5550 dquot->dq_dqb.dqb_bsoftlimit : 5551 dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits; 5552 if (limit && buf->f_blocks > limit) { 5553 curblock = (dquot->dq_dqb.dqb_curspace + 5554 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 5555 buf->f_blocks = limit; 5556 buf->f_bfree = buf->f_bavail = 5557 (buf->f_blocks > curblock) ? 5558 (buf->f_blocks - curblock) : 0; 5559 } 5560 5561 limit = dquot->dq_dqb.dqb_isoftlimit ? 5562 dquot->dq_dqb.dqb_isoftlimit : 5563 dquot->dq_dqb.dqb_ihardlimit; 5564 if (limit && buf->f_files > limit) { 5565 buf->f_files = limit; 5566 buf->f_ffree = 5567 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 5568 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 5569 } 5570 5571 spin_unlock(&dquot->dq_dqb_lock); 5572 dqput(dquot); 5573 return 0; 5574 } 5575 #endif 5576 5577 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 5578 { 5579 struct super_block *sb = dentry->d_sb; 5580 struct ext4_sb_info *sbi = EXT4_SB(sb); 5581 struct ext4_super_block *es = sbi->s_es; 5582 ext4_fsblk_t overhead = 0, resv_blocks; 5583 u64 fsid; 5584 s64 bfree; 5585 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 5586 5587 if (!test_opt(sb, MINIX_DF)) 5588 overhead = sbi->s_overhead; 5589 5590 buf->f_type = EXT4_SUPER_MAGIC; 5591 buf->f_bsize = sb->s_blocksize; 5592 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 5593 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 5594 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 5595 /* prevent underflow in case that few free space is available */ 5596 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 5597 buf->f_bavail = buf->f_bfree - 5598 (ext4_r_blocks_count(es) + resv_blocks); 5599 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 5600 buf->f_bavail = 0; 5601 buf->f_files = le32_to_cpu(es->s_inodes_count); 5602 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 5603 buf->f_namelen = EXT4_NAME_LEN; 5604 fsid = le64_to_cpup((void *)es->s_uuid) ^ 5605 le64_to_cpup((void *)es->s_uuid + sizeof(u64)); 5606 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL; 5607 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL; 5608 5609 #ifdef CONFIG_QUOTA 5610 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 5611 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 5612 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 5613 #endif 5614 return 0; 5615 } 5616 5617 5618 #ifdef CONFIG_QUOTA 5619 5620 /* 5621 * Helper functions so that transaction is started before we acquire dqio_sem 5622 * to keep correct lock ordering of transaction > dqio_sem 5623 */ 5624 static inline struct inode *dquot_to_inode(struct dquot *dquot) 5625 { 5626 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 5627 } 5628 5629 static int ext4_write_dquot(struct dquot *dquot) 5630 { 5631 int ret, err; 5632 handle_t *handle; 5633 struct inode *inode; 5634 5635 inode = dquot_to_inode(dquot); 5636 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 5637 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 5638 if (IS_ERR(handle)) 5639 return PTR_ERR(handle); 5640 ret = dquot_commit(dquot); 5641 err = ext4_journal_stop(handle); 5642 if (!ret) 5643 ret = err; 5644 return ret; 5645 } 5646 5647 static int ext4_acquire_dquot(struct dquot *dquot) 5648 { 5649 int ret, err; 5650 handle_t *handle; 5651 5652 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 5653 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 5654 if (IS_ERR(handle)) 5655 return PTR_ERR(handle); 5656 ret = dquot_acquire(dquot); 5657 err = ext4_journal_stop(handle); 5658 if (!ret) 5659 ret = err; 5660 return ret; 5661 } 5662 5663 static int ext4_release_dquot(struct dquot *dquot) 5664 { 5665 int ret, err; 5666 handle_t *handle; 5667 5668 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 5669 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 5670 if (IS_ERR(handle)) { 5671 /* Release dquot anyway to avoid endless cycle in dqput() */ 5672 dquot_release(dquot); 5673 return PTR_ERR(handle); 5674 } 5675 ret = dquot_release(dquot); 5676 err = ext4_journal_stop(handle); 5677 if (!ret) 5678 ret = err; 5679 return ret; 5680 } 5681 5682 static int ext4_mark_dquot_dirty(struct dquot *dquot) 5683 { 5684 struct super_block *sb = dquot->dq_sb; 5685 struct ext4_sb_info *sbi = EXT4_SB(sb); 5686 5687 /* Are we journaling quotas? */ 5688 if (ext4_has_feature_quota(sb) || 5689 sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) { 5690 dquot_mark_dquot_dirty(dquot); 5691 return ext4_write_dquot(dquot); 5692 } else { 5693 return dquot_mark_dquot_dirty(dquot); 5694 } 5695 } 5696 5697 static int ext4_write_info(struct super_block *sb, int type) 5698 { 5699 int ret, err; 5700 handle_t *handle; 5701 5702 /* Data block + inode block */ 5703 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 5704 if (IS_ERR(handle)) 5705 return PTR_ERR(handle); 5706 ret = dquot_commit_info(sb, type); 5707 err = ext4_journal_stop(handle); 5708 if (!ret) 5709 ret = err; 5710 return ret; 5711 } 5712 5713 /* 5714 * Turn on quotas during mount time - we need to find 5715 * the quota file and such... 5716 */ 5717 static int ext4_quota_on_mount(struct super_block *sb, int type) 5718 { 5719 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 5720 EXT4_SB(sb)->s_jquota_fmt, type); 5721 } 5722 5723 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 5724 { 5725 struct ext4_inode_info *ei = EXT4_I(inode); 5726 5727 /* The first argument of lockdep_set_subclass has to be 5728 * *exactly* the same as the argument to init_rwsem() --- in 5729 * this case, in init_once() --- or lockdep gets unhappy 5730 * because the name of the lock is set using the 5731 * stringification of the argument to init_rwsem(). 5732 */ 5733 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 5734 lockdep_set_subclass(&ei->i_data_sem, subclass); 5735 } 5736 5737 /* 5738 * Standard function to be called on quota_on 5739 */ 5740 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 5741 const struct path *path) 5742 { 5743 int err; 5744 5745 if (!test_opt(sb, QUOTA)) 5746 return -EINVAL; 5747 5748 /* Quotafile not on the same filesystem? */ 5749 if (path->dentry->d_sb != sb) 5750 return -EXDEV; 5751 /* Journaling quota? */ 5752 if (EXT4_SB(sb)->s_qf_names[type]) { 5753 /* Quotafile not in fs root? */ 5754 if (path->dentry->d_parent != sb->s_root) 5755 ext4_msg(sb, KERN_WARNING, 5756 "Quota file not on filesystem root. " 5757 "Journaled quota will not work"); 5758 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 5759 } else { 5760 /* 5761 * Clear the flag just in case mount options changed since 5762 * last time. 5763 */ 5764 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 5765 } 5766 5767 /* 5768 * When we journal data on quota file, we have to flush journal to see 5769 * all updates to the file when we bypass pagecache... 5770 */ 5771 if (EXT4_SB(sb)->s_journal && 5772 ext4_should_journal_data(d_inode(path->dentry))) { 5773 /* 5774 * We don't need to lock updates but journal_flush() could 5775 * otherwise be livelocked... 5776 */ 5777 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 5778 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 5779 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 5780 if (err) 5781 return err; 5782 } 5783 5784 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 5785 err = dquot_quota_on(sb, type, format_id, path); 5786 if (err) { 5787 lockdep_set_quota_inode(path->dentry->d_inode, 5788 I_DATA_SEM_NORMAL); 5789 } else { 5790 struct inode *inode = d_inode(path->dentry); 5791 handle_t *handle; 5792 5793 /* 5794 * Set inode flags to prevent userspace from messing with quota 5795 * files. If this fails, we return success anyway since quotas 5796 * are already enabled and this is not a hard failure. 5797 */ 5798 inode_lock(inode); 5799 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 5800 if (IS_ERR(handle)) 5801 goto unlock_inode; 5802 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 5803 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 5804 S_NOATIME | S_IMMUTABLE); 5805 ext4_mark_inode_dirty(handle, inode); 5806 ext4_journal_stop(handle); 5807 unlock_inode: 5808 inode_unlock(inode); 5809 } 5810 return err; 5811 } 5812 5813 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 5814 unsigned int flags) 5815 { 5816 int err; 5817 struct inode *qf_inode; 5818 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 5819 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 5820 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 5821 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 5822 }; 5823 5824 BUG_ON(!ext4_has_feature_quota(sb)); 5825 5826 if (!qf_inums[type]) 5827 return -EPERM; 5828 5829 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 5830 if (IS_ERR(qf_inode)) { 5831 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 5832 return PTR_ERR(qf_inode); 5833 } 5834 5835 /* Don't account quota for quota files to avoid recursion */ 5836 qf_inode->i_flags |= S_NOQUOTA; 5837 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 5838 err = dquot_enable(qf_inode, type, format_id, flags); 5839 if (err) 5840 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 5841 iput(qf_inode); 5842 5843 return err; 5844 } 5845 5846 /* Enable usage tracking for all quota types. */ 5847 static int ext4_enable_quotas(struct super_block *sb) 5848 { 5849 int type, err = 0; 5850 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 5851 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 5852 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 5853 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 5854 }; 5855 bool quota_mopt[EXT4_MAXQUOTAS] = { 5856 test_opt(sb, USRQUOTA), 5857 test_opt(sb, GRPQUOTA), 5858 test_opt(sb, PRJQUOTA), 5859 }; 5860 5861 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 5862 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 5863 if (qf_inums[type]) { 5864 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 5865 DQUOT_USAGE_ENABLED | 5866 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 5867 if (err) { 5868 ext4_warning(sb, 5869 "Failed to enable quota tracking " 5870 "(type=%d, err=%d). Please run " 5871 "e2fsck to fix.", type, err); 5872 for (type--; type >= 0; type--) 5873 dquot_quota_off(sb, type); 5874 5875 return err; 5876 } 5877 } 5878 } 5879 return 0; 5880 } 5881 5882 static int ext4_quota_off(struct super_block *sb, int type) 5883 { 5884 struct inode *inode = sb_dqopt(sb)->files[type]; 5885 handle_t *handle; 5886 int err; 5887 5888 /* Force all delayed allocation blocks to be allocated. 5889 * Caller already holds s_umount sem */ 5890 if (test_opt(sb, DELALLOC)) 5891 sync_filesystem(sb); 5892 5893 if (!inode || !igrab(inode)) 5894 goto out; 5895 5896 err = dquot_quota_off(sb, type); 5897 if (err || ext4_has_feature_quota(sb)) 5898 goto out_put; 5899 5900 inode_lock(inode); 5901 /* 5902 * Update modification times of quota files when userspace can 5903 * start looking at them. If we fail, we return success anyway since 5904 * this is not a hard failure and quotas are already disabled. 5905 */ 5906 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 5907 if (IS_ERR(handle)) 5908 goto out_unlock; 5909 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 5910 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 5911 inode->i_mtime = inode->i_ctime = current_time(inode); 5912 ext4_mark_inode_dirty(handle, inode); 5913 ext4_journal_stop(handle); 5914 out_unlock: 5915 inode_unlock(inode); 5916 out_put: 5917 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 5918 iput(inode); 5919 return err; 5920 out: 5921 return dquot_quota_off(sb, type); 5922 } 5923 5924 /* Read data from quotafile - avoid pagecache and such because we cannot afford 5925 * acquiring the locks... As quota files are never truncated and quota code 5926 * itself serializes the operations (and no one else should touch the files) 5927 * we don't have to be afraid of races */ 5928 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 5929 size_t len, loff_t off) 5930 { 5931 struct inode *inode = sb_dqopt(sb)->files[type]; 5932 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 5933 int offset = off & (sb->s_blocksize - 1); 5934 int tocopy; 5935 size_t toread; 5936 struct buffer_head *bh; 5937 loff_t i_size = i_size_read(inode); 5938 5939 if (off > i_size) 5940 return 0; 5941 if (off+len > i_size) 5942 len = i_size-off; 5943 toread = len; 5944 while (toread > 0) { 5945 tocopy = sb->s_blocksize - offset < toread ? 5946 sb->s_blocksize - offset : toread; 5947 bh = ext4_bread(NULL, inode, blk, 0); 5948 if (IS_ERR(bh)) 5949 return PTR_ERR(bh); 5950 if (!bh) /* A hole? */ 5951 memset(data, 0, tocopy); 5952 else 5953 memcpy(data, bh->b_data+offset, tocopy); 5954 brelse(bh); 5955 offset = 0; 5956 toread -= tocopy; 5957 data += tocopy; 5958 blk++; 5959 } 5960 return len; 5961 } 5962 5963 /* Write to quotafile (we know the transaction is already started and has 5964 * enough credits) */ 5965 static ssize_t ext4_quota_write(struct super_block *sb, int type, 5966 const char *data, size_t len, loff_t off) 5967 { 5968 struct inode *inode = sb_dqopt(sb)->files[type]; 5969 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 5970 int err, offset = off & (sb->s_blocksize - 1); 5971 int retries = 0; 5972 struct buffer_head *bh; 5973 handle_t *handle = journal_current_handle(); 5974 5975 if (EXT4_SB(sb)->s_journal && !handle) { 5976 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 5977 " cancelled because transaction is not started", 5978 (unsigned long long)off, (unsigned long long)len); 5979 return -EIO; 5980 } 5981 /* 5982 * Since we account only one data block in transaction credits, 5983 * then it is impossible to cross a block boundary. 5984 */ 5985 if (sb->s_blocksize - offset < len) { 5986 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 5987 " cancelled because not block aligned", 5988 (unsigned long long)off, (unsigned long long)len); 5989 return -EIO; 5990 } 5991 5992 do { 5993 bh = ext4_bread(handle, inode, blk, 5994 EXT4_GET_BLOCKS_CREATE | 5995 EXT4_GET_BLOCKS_METADATA_NOFAIL); 5996 } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) && 5997 ext4_should_retry_alloc(inode->i_sb, &retries)); 5998 if (IS_ERR(bh)) 5999 return PTR_ERR(bh); 6000 if (!bh) 6001 goto out; 6002 BUFFER_TRACE(bh, "get write access"); 6003 err = ext4_journal_get_write_access(handle, bh); 6004 if (err) { 6005 brelse(bh); 6006 return err; 6007 } 6008 lock_buffer(bh); 6009 memcpy(bh->b_data+offset, data, len); 6010 flush_dcache_page(bh->b_page); 6011 unlock_buffer(bh); 6012 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6013 brelse(bh); 6014 out: 6015 if (inode->i_size < off + len) { 6016 i_size_write(inode, off + len); 6017 EXT4_I(inode)->i_disksize = inode->i_size; 6018 ext4_mark_inode_dirty(handle, inode); 6019 } 6020 return len; 6021 } 6022 6023 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid) 6024 { 6025 const struct quota_format_ops *ops; 6026 6027 if (!sb_has_quota_loaded(sb, qid->type)) 6028 return -ESRCH; 6029 ops = sb_dqopt(sb)->ops[qid->type]; 6030 if (!ops || !ops->get_next_id) 6031 return -ENOSYS; 6032 return dquot_get_next_id(sb, qid); 6033 } 6034 #endif 6035 6036 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6037 const char *dev_name, void *data) 6038 { 6039 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6040 } 6041 6042 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6043 static inline void register_as_ext2(void) 6044 { 6045 int err = register_filesystem(&ext2_fs_type); 6046 if (err) 6047 printk(KERN_WARNING 6048 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6049 } 6050 6051 static inline void unregister_as_ext2(void) 6052 { 6053 unregister_filesystem(&ext2_fs_type); 6054 } 6055 6056 static inline int ext2_feature_set_ok(struct super_block *sb) 6057 { 6058 if (ext4_has_unknown_ext2_incompat_features(sb)) 6059 return 0; 6060 if (sb_rdonly(sb)) 6061 return 1; 6062 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6063 return 0; 6064 return 1; 6065 } 6066 #else 6067 static inline void register_as_ext2(void) { } 6068 static inline void unregister_as_ext2(void) { } 6069 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6070 #endif 6071 6072 static inline void register_as_ext3(void) 6073 { 6074 int err = register_filesystem(&ext3_fs_type); 6075 if (err) 6076 printk(KERN_WARNING 6077 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6078 } 6079 6080 static inline void unregister_as_ext3(void) 6081 { 6082 unregister_filesystem(&ext3_fs_type); 6083 } 6084 6085 static inline int ext3_feature_set_ok(struct super_block *sb) 6086 { 6087 if (ext4_has_unknown_ext3_incompat_features(sb)) 6088 return 0; 6089 if (!ext4_has_feature_journal(sb)) 6090 return 0; 6091 if (sb_rdonly(sb)) 6092 return 1; 6093 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6094 return 0; 6095 return 1; 6096 } 6097 6098 static struct file_system_type ext4_fs_type = { 6099 .owner = THIS_MODULE, 6100 .name = "ext4", 6101 .mount = ext4_mount, 6102 .kill_sb = kill_block_super, 6103 .fs_flags = FS_REQUIRES_DEV, 6104 }; 6105 MODULE_ALIAS_FS("ext4"); 6106 6107 /* Shared across all ext4 file systems */ 6108 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6109 6110 static int __init ext4_init_fs(void) 6111 { 6112 int i, err; 6113 6114 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6115 ext4_li_info = NULL; 6116 mutex_init(&ext4_li_mtx); 6117 6118 /* Build-time check for flags consistency */ 6119 ext4_check_flag_values(); 6120 6121 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6122 init_waitqueue_head(&ext4__ioend_wq[i]); 6123 6124 err = ext4_init_es(); 6125 if (err) 6126 return err; 6127 6128 err = ext4_init_pending(); 6129 if (err) 6130 goto out7; 6131 6132 err = ext4_init_post_read_processing(); 6133 if (err) 6134 goto out6; 6135 6136 err = ext4_init_pageio(); 6137 if (err) 6138 goto out5; 6139 6140 err = ext4_init_system_zone(); 6141 if (err) 6142 goto out4; 6143 6144 err = ext4_init_sysfs(); 6145 if (err) 6146 goto out3; 6147 6148 err = ext4_init_mballoc(); 6149 if (err) 6150 goto out2; 6151 err = init_inodecache(); 6152 if (err) 6153 goto out1; 6154 register_as_ext3(); 6155 register_as_ext2(); 6156 err = register_filesystem(&ext4_fs_type); 6157 if (err) 6158 goto out; 6159 6160 return 0; 6161 out: 6162 unregister_as_ext2(); 6163 unregister_as_ext3(); 6164 destroy_inodecache(); 6165 out1: 6166 ext4_exit_mballoc(); 6167 out2: 6168 ext4_exit_sysfs(); 6169 out3: 6170 ext4_exit_system_zone(); 6171 out4: 6172 ext4_exit_pageio(); 6173 out5: 6174 ext4_exit_post_read_processing(); 6175 out6: 6176 ext4_exit_pending(); 6177 out7: 6178 ext4_exit_es(); 6179 6180 return err; 6181 } 6182 6183 static void __exit ext4_exit_fs(void) 6184 { 6185 ext4_destroy_lazyinit_thread(); 6186 unregister_as_ext2(); 6187 unregister_as_ext3(); 6188 unregister_filesystem(&ext4_fs_type); 6189 destroy_inodecache(); 6190 ext4_exit_mballoc(); 6191 ext4_exit_sysfs(); 6192 ext4_exit_system_zone(); 6193 ext4_exit_pageio(); 6194 ext4_exit_post_read_processing(); 6195 ext4_exit_es(); 6196 ext4_exit_pending(); 6197 } 6198 6199 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6200 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6201 MODULE_LICENSE("GPL"); 6202 MODULE_SOFTDEP("pre: crc32c"); 6203 module_init(ext4_init_fs) 6204 module_exit(ext4_exit_fs) 6205