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 ext4_discard_preallocations(inode); 1176 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); 1177 dquot_drop(inode); 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 bool ext4_has_stable_inodes(struct super_block *sb) 1349 { 1350 return ext4_has_feature_stable_inodes(sb); 1351 } 1352 1353 static void ext4_get_ino_and_lblk_bits(struct super_block *sb, 1354 int *ino_bits_ret, int *lblk_bits_ret) 1355 { 1356 *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count); 1357 *lblk_bits_ret = 8 * sizeof(ext4_lblk_t); 1358 } 1359 1360 static const struct fscrypt_operations ext4_cryptops = { 1361 .key_prefix = "ext4:", 1362 .get_context = ext4_get_context, 1363 .set_context = ext4_set_context, 1364 .dummy_context = ext4_dummy_context, 1365 .empty_dir = ext4_empty_dir, 1366 .max_namelen = EXT4_NAME_LEN, 1367 .has_stable_inodes = ext4_has_stable_inodes, 1368 .get_ino_and_lblk_bits = ext4_get_ino_and_lblk_bits, 1369 }; 1370 #endif 1371 1372 #ifdef CONFIG_QUOTA 1373 static const char * const quotatypes[] = INITQFNAMES; 1374 #define QTYPE2NAME(t) (quotatypes[t]) 1375 1376 static int ext4_write_dquot(struct dquot *dquot); 1377 static int ext4_acquire_dquot(struct dquot *dquot); 1378 static int ext4_release_dquot(struct dquot *dquot); 1379 static int ext4_mark_dquot_dirty(struct dquot *dquot); 1380 static int ext4_write_info(struct super_block *sb, int type); 1381 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 1382 const struct path *path); 1383 static int ext4_quota_on_mount(struct super_block *sb, int type); 1384 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 1385 size_t len, loff_t off); 1386 static ssize_t ext4_quota_write(struct super_block *sb, int type, 1387 const char *data, size_t len, loff_t off); 1388 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 1389 unsigned int flags); 1390 static int ext4_enable_quotas(struct super_block *sb); 1391 1392 static struct dquot **ext4_get_dquots(struct inode *inode) 1393 { 1394 return EXT4_I(inode)->i_dquot; 1395 } 1396 1397 static const struct dquot_operations ext4_quota_operations = { 1398 .get_reserved_space = ext4_get_reserved_space, 1399 .write_dquot = ext4_write_dquot, 1400 .acquire_dquot = ext4_acquire_dquot, 1401 .release_dquot = ext4_release_dquot, 1402 .mark_dirty = ext4_mark_dquot_dirty, 1403 .write_info = ext4_write_info, 1404 .alloc_dquot = dquot_alloc, 1405 .destroy_dquot = dquot_destroy, 1406 .get_projid = ext4_get_projid, 1407 .get_inode_usage = ext4_get_inode_usage, 1408 .get_next_id = dquot_get_next_id, 1409 }; 1410 1411 static const struct quotactl_ops ext4_qctl_operations = { 1412 .quota_on = ext4_quota_on, 1413 .quota_off = ext4_quota_off, 1414 .quota_sync = dquot_quota_sync, 1415 .get_state = dquot_get_state, 1416 .set_info = dquot_set_dqinfo, 1417 .get_dqblk = dquot_get_dqblk, 1418 .set_dqblk = dquot_set_dqblk, 1419 .get_nextdqblk = dquot_get_next_dqblk, 1420 }; 1421 #endif 1422 1423 static const struct super_operations ext4_sops = { 1424 .alloc_inode = ext4_alloc_inode, 1425 .free_inode = ext4_free_in_core_inode, 1426 .destroy_inode = ext4_destroy_inode, 1427 .write_inode = ext4_write_inode, 1428 .dirty_inode = ext4_dirty_inode, 1429 .drop_inode = ext4_drop_inode, 1430 .evict_inode = ext4_evict_inode, 1431 .put_super = ext4_put_super, 1432 .sync_fs = ext4_sync_fs, 1433 .freeze_fs = ext4_freeze, 1434 .unfreeze_fs = ext4_unfreeze, 1435 .statfs = ext4_statfs, 1436 .remount_fs = ext4_remount, 1437 .show_options = ext4_show_options, 1438 #ifdef CONFIG_QUOTA 1439 .quota_read = ext4_quota_read, 1440 .quota_write = ext4_quota_write, 1441 .get_dquots = ext4_get_dquots, 1442 #endif 1443 .bdev_try_to_free_page = bdev_try_to_free_page, 1444 }; 1445 1446 static const struct export_operations ext4_export_ops = { 1447 .fh_to_dentry = ext4_fh_to_dentry, 1448 .fh_to_parent = ext4_fh_to_parent, 1449 .get_parent = ext4_get_parent, 1450 .commit_metadata = ext4_nfs_commit_metadata, 1451 }; 1452 1453 enum { 1454 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid, 1455 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro, 1456 Opt_nouid32, Opt_debug, Opt_removed, 1457 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, 1458 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, 1459 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev, 1460 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit, 1461 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback, 1462 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, 1463 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota, 1464 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota, 1465 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, 1466 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax, 1467 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, 1468 Opt_nowarn_on_error, Opt_mblk_io_submit, 1469 Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize, 1470 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, 1471 Opt_inode_readahead_blks, Opt_journal_ioprio, 1472 Opt_dioread_nolock, Opt_dioread_lock, 1473 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, 1474 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache, 1475 }; 1476 1477 static const match_table_t tokens = { 1478 {Opt_bsd_df, "bsddf"}, 1479 {Opt_minix_df, "minixdf"}, 1480 {Opt_grpid, "grpid"}, 1481 {Opt_grpid, "bsdgroups"}, 1482 {Opt_nogrpid, "nogrpid"}, 1483 {Opt_nogrpid, "sysvgroups"}, 1484 {Opt_resgid, "resgid=%u"}, 1485 {Opt_resuid, "resuid=%u"}, 1486 {Opt_sb, "sb=%u"}, 1487 {Opt_err_cont, "errors=continue"}, 1488 {Opt_err_panic, "errors=panic"}, 1489 {Opt_err_ro, "errors=remount-ro"}, 1490 {Opt_nouid32, "nouid32"}, 1491 {Opt_debug, "debug"}, 1492 {Opt_removed, "oldalloc"}, 1493 {Opt_removed, "orlov"}, 1494 {Opt_user_xattr, "user_xattr"}, 1495 {Opt_nouser_xattr, "nouser_xattr"}, 1496 {Opt_acl, "acl"}, 1497 {Opt_noacl, "noacl"}, 1498 {Opt_noload, "norecovery"}, 1499 {Opt_noload, "noload"}, 1500 {Opt_removed, "nobh"}, 1501 {Opt_removed, "bh"}, 1502 {Opt_commit, "commit=%u"}, 1503 {Opt_min_batch_time, "min_batch_time=%u"}, 1504 {Opt_max_batch_time, "max_batch_time=%u"}, 1505 {Opt_journal_dev, "journal_dev=%u"}, 1506 {Opt_journal_path, "journal_path=%s"}, 1507 {Opt_journal_checksum, "journal_checksum"}, 1508 {Opt_nojournal_checksum, "nojournal_checksum"}, 1509 {Opt_journal_async_commit, "journal_async_commit"}, 1510 {Opt_abort, "abort"}, 1511 {Opt_data_journal, "data=journal"}, 1512 {Opt_data_ordered, "data=ordered"}, 1513 {Opt_data_writeback, "data=writeback"}, 1514 {Opt_data_err_abort, "data_err=abort"}, 1515 {Opt_data_err_ignore, "data_err=ignore"}, 1516 {Opt_offusrjquota, "usrjquota="}, 1517 {Opt_usrjquota, "usrjquota=%s"}, 1518 {Opt_offgrpjquota, "grpjquota="}, 1519 {Opt_grpjquota, "grpjquota=%s"}, 1520 {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, 1521 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, 1522 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, 1523 {Opt_grpquota, "grpquota"}, 1524 {Opt_noquota, "noquota"}, 1525 {Opt_quota, "quota"}, 1526 {Opt_usrquota, "usrquota"}, 1527 {Opt_prjquota, "prjquota"}, 1528 {Opt_barrier, "barrier=%u"}, 1529 {Opt_barrier, "barrier"}, 1530 {Opt_nobarrier, "nobarrier"}, 1531 {Opt_i_version, "i_version"}, 1532 {Opt_dax, "dax"}, 1533 {Opt_stripe, "stripe=%u"}, 1534 {Opt_delalloc, "delalloc"}, 1535 {Opt_warn_on_error, "warn_on_error"}, 1536 {Opt_nowarn_on_error, "nowarn_on_error"}, 1537 {Opt_lazytime, "lazytime"}, 1538 {Opt_nolazytime, "nolazytime"}, 1539 {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"}, 1540 {Opt_nodelalloc, "nodelalloc"}, 1541 {Opt_removed, "mblk_io_submit"}, 1542 {Opt_removed, "nomblk_io_submit"}, 1543 {Opt_block_validity, "block_validity"}, 1544 {Opt_noblock_validity, "noblock_validity"}, 1545 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, 1546 {Opt_journal_ioprio, "journal_ioprio=%u"}, 1547 {Opt_auto_da_alloc, "auto_da_alloc=%u"}, 1548 {Opt_auto_da_alloc, "auto_da_alloc"}, 1549 {Opt_noauto_da_alloc, "noauto_da_alloc"}, 1550 {Opt_dioread_nolock, "dioread_nolock"}, 1551 {Opt_dioread_lock, "dioread_lock"}, 1552 {Opt_discard, "discard"}, 1553 {Opt_nodiscard, "nodiscard"}, 1554 {Opt_init_itable, "init_itable=%u"}, 1555 {Opt_init_itable, "init_itable"}, 1556 {Opt_noinit_itable, "noinit_itable"}, 1557 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, 1558 {Opt_test_dummy_encryption, "test_dummy_encryption"}, 1559 {Opt_nombcache, "nombcache"}, 1560 {Opt_nombcache, "no_mbcache"}, /* for backward compatibility */ 1561 {Opt_removed, "check=none"}, /* mount option from ext2/3 */ 1562 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */ 1563 {Opt_removed, "reservation"}, /* mount option from ext2/3 */ 1564 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */ 1565 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */ 1566 {Opt_err, NULL}, 1567 }; 1568 1569 static ext4_fsblk_t get_sb_block(void **data) 1570 { 1571 ext4_fsblk_t sb_block; 1572 char *options = (char *) *data; 1573 1574 if (!options || strncmp(options, "sb=", 3) != 0) 1575 return 1; /* Default location */ 1576 1577 options += 3; 1578 /* TODO: use simple_strtoll with >32bit ext4 */ 1579 sb_block = simple_strtoul(options, &options, 0); 1580 if (*options && *options != ',') { 1581 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n", 1582 (char *) *data); 1583 return 1; 1584 } 1585 if (*options == ',') 1586 options++; 1587 *data = (void *) options; 1588 1589 return sb_block; 1590 } 1591 1592 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 1593 static const char deprecated_msg[] = 1594 "Mount option \"%s\" will be removed by %s\n" 1595 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n"; 1596 1597 #ifdef CONFIG_QUOTA 1598 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args) 1599 { 1600 struct ext4_sb_info *sbi = EXT4_SB(sb); 1601 char *qname, *old_qname = get_qf_name(sb, sbi, qtype); 1602 int ret = -1; 1603 1604 if (sb_any_quota_loaded(sb) && !old_qname) { 1605 ext4_msg(sb, KERN_ERR, 1606 "Cannot change journaled " 1607 "quota options when quota turned on"); 1608 return -1; 1609 } 1610 if (ext4_has_feature_quota(sb)) { 1611 ext4_msg(sb, KERN_INFO, "Journaled quota options " 1612 "ignored when QUOTA feature is enabled"); 1613 return 1; 1614 } 1615 qname = match_strdup(args); 1616 if (!qname) { 1617 ext4_msg(sb, KERN_ERR, 1618 "Not enough memory for storing quotafile name"); 1619 return -1; 1620 } 1621 if (old_qname) { 1622 if (strcmp(old_qname, qname) == 0) 1623 ret = 1; 1624 else 1625 ext4_msg(sb, KERN_ERR, 1626 "%s quota file already specified", 1627 QTYPE2NAME(qtype)); 1628 goto errout; 1629 } 1630 if (strchr(qname, '/')) { 1631 ext4_msg(sb, KERN_ERR, 1632 "quotafile must be on filesystem root"); 1633 goto errout; 1634 } 1635 rcu_assign_pointer(sbi->s_qf_names[qtype], qname); 1636 set_opt(sb, QUOTA); 1637 return 1; 1638 errout: 1639 kfree(qname); 1640 return ret; 1641 } 1642 1643 static int clear_qf_name(struct super_block *sb, int qtype) 1644 { 1645 1646 struct ext4_sb_info *sbi = EXT4_SB(sb); 1647 char *old_qname = get_qf_name(sb, sbi, qtype); 1648 1649 if (sb_any_quota_loaded(sb) && old_qname) { 1650 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options" 1651 " when quota turned on"); 1652 return -1; 1653 } 1654 rcu_assign_pointer(sbi->s_qf_names[qtype], NULL); 1655 synchronize_rcu(); 1656 kfree(old_qname); 1657 return 1; 1658 } 1659 #endif 1660 1661 #define MOPT_SET 0x0001 1662 #define MOPT_CLEAR 0x0002 1663 #define MOPT_NOSUPPORT 0x0004 1664 #define MOPT_EXPLICIT 0x0008 1665 #define MOPT_CLEAR_ERR 0x0010 1666 #define MOPT_GTE0 0x0020 1667 #ifdef CONFIG_QUOTA 1668 #define MOPT_Q 0 1669 #define MOPT_QFMT 0x0040 1670 #else 1671 #define MOPT_Q MOPT_NOSUPPORT 1672 #define MOPT_QFMT MOPT_NOSUPPORT 1673 #endif 1674 #define MOPT_DATAJ 0x0080 1675 #define MOPT_NO_EXT2 0x0100 1676 #define MOPT_NO_EXT3 0x0200 1677 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3) 1678 #define MOPT_STRING 0x0400 1679 1680 static const struct mount_opts { 1681 int token; 1682 int mount_opt; 1683 int flags; 1684 } ext4_mount_opts[] = { 1685 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET}, 1686 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR}, 1687 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET}, 1688 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR}, 1689 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET}, 1690 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR}, 1691 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, 1692 MOPT_EXT4_ONLY | MOPT_SET}, 1693 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, 1694 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1695 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET}, 1696 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR}, 1697 {Opt_delalloc, EXT4_MOUNT_DELALLOC, 1698 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1699 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, 1700 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1701 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET}, 1702 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR}, 1703 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1704 MOPT_EXT4_ONLY | MOPT_CLEAR}, 1705 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, 1706 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1707 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT | 1708 EXT4_MOUNT_JOURNAL_CHECKSUM), 1709 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT}, 1710 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET}, 1711 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR}, 1712 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR}, 1713 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR}, 1714 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, 1715 MOPT_NO_EXT2}, 1716 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, 1717 MOPT_NO_EXT2}, 1718 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET}, 1719 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR}, 1720 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, 1721 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, 1722 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, 1723 {Opt_commit, 0, MOPT_GTE0}, 1724 {Opt_max_batch_time, 0, MOPT_GTE0}, 1725 {Opt_min_batch_time, 0, MOPT_GTE0}, 1726 {Opt_inode_readahead_blks, 0, MOPT_GTE0}, 1727 {Opt_init_itable, 0, MOPT_GTE0}, 1728 {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET}, 1729 {Opt_stripe, 0, MOPT_GTE0}, 1730 {Opt_resuid, 0, MOPT_GTE0}, 1731 {Opt_resgid, 0, MOPT_GTE0}, 1732 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1733 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING}, 1734 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0}, 1735 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1736 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ}, 1737 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, 1738 MOPT_NO_EXT2 | MOPT_DATAJ}, 1739 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET}, 1740 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR}, 1741 #ifdef CONFIG_EXT4_FS_POSIX_ACL 1742 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET}, 1743 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR}, 1744 #else 1745 {Opt_acl, 0, MOPT_NOSUPPORT}, 1746 {Opt_noacl, 0, MOPT_NOSUPPORT}, 1747 #endif 1748 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET}, 1749 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET}, 1750 {Opt_debug_want_extra_isize, 0, MOPT_GTE0}, 1751 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q}, 1752 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, 1753 MOPT_SET | MOPT_Q}, 1754 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA, 1755 MOPT_SET | MOPT_Q}, 1756 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA, 1757 MOPT_SET | MOPT_Q}, 1758 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA | 1759 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA), 1760 MOPT_CLEAR | MOPT_Q}, 1761 {Opt_usrjquota, 0, MOPT_Q}, 1762 {Opt_grpjquota, 0, MOPT_Q}, 1763 {Opt_offusrjquota, 0, MOPT_Q}, 1764 {Opt_offgrpjquota, 0, MOPT_Q}, 1765 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT}, 1766 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT}, 1767 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT}, 1768 {Opt_max_dir_size_kb, 0, MOPT_GTE0}, 1769 {Opt_test_dummy_encryption, 0, MOPT_GTE0}, 1770 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, 1771 {Opt_err, 0, 0} 1772 }; 1773 1774 #ifdef CONFIG_UNICODE 1775 static const struct ext4_sb_encodings { 1776 __u16 magic; 1777 char *name; 1778 char *version; 1779 } ext4_sb_encoding_map[] = { 1780 {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"}, 1781 }; 1782 1783 static int ext4_sb_read_encoding(const struct ext4_super_block *es, 1784 const struct ext4_sb_encodings **encoding, 1785 __u16 *flags) 1786 { 1787 __u16 magic = le16_to_cpu(es->s_encoding); 1788 int i; 1789 1790 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++) 1791 if (magic == ext4_sb_encoding_map[i].magic) 1792 break; 1793 1794 if (i >= ARRAY_SIZE(ext4_sb_encoding_map)) 1795 return -EINVAL; 1796 1797 *encoding = &ext4_sb_encoding_map[i]; 1798 *flags = le16_to_cpu(es->s_encoding_flags); 1799 1800 return 0; 1801 } 1802 #endif 1803 1804 static int handle_mount_opt(struct super_block *sb, char *opt, int token, 1805 substring_t *args, unsigned long *journal_devnum, 1806 unsigned int *journal_ioprio, int is_remount) 1807 { 1808 struct ext4_sb_info *sbi = EXT4_SB(sb); 1809 const struct mount_opts *m; 1810 kuid_t uid; 1811 kgid_t gid; 1812 int arg = 0; 1813 1814 #ifdef CONFIG_QUOTA 1815 if (token == Opt_usrjquota) 1816 return set_qf_name(sb, USRQUOTA, &args[0]); 1817 else if (token == Opt_grpjquota) 1818 return set_qf_name(sb, GRPQUOTA, &args[0]); 1819 else if (token == Opt_offusrjquota) 1820 return clear_qf_name(sb, USRQUOTA); 1821 else if (token == Opt_offgrpjquota) 1822 return clear_qf_name(sb, GRPQUOTA); 1823 #endif 1824 switch (token) { 1825 case Opt_noacl: 1826 case Opt_nouser_xattr: 1827 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5"); 1828 break; 1829 case Opt_sb: 1830 return 1; /* handled by get_sb_block() */ 1831 case Opt_removed: 1832 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt); 1833 return 1; 1834 case Opt_abort: 1835 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED; 1836 return 1; 1837 case Opt_i_version: 1838 sb->s_flags |= SB_I_VERSION; 1839 return 1; 1840 case Opt_lazytime: 1841 sb->s_flags |= SB_LAZYTIME; 1842 return 1; 1843 case Opt_nolazytime: 1844 sb->s_flags &= ~SB_LAZYTIME; 1845 return 1; 1846 } 1847 1848 for (m = ext4_mount_opts; m->token != Opt_err; m++) 1849 if (token == m->token) 1850 break; 1851 1852 if (m->token == Opt_err) { 1853 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" " 1854 "or missing value", opt); 1855 return -1; 1856 } 1857 1858 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) { 1859 ext4_msg(sb, KERN_ERR, 1860 "Mount option \"%s\" incompatible with ext2", opt); 1861 return -1; 1862 } 1863 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) { 1864 ext4_msg(sb, KERN_ERR, 1865 "Mount option \"%s\" incompatible with ext3", opt); 1866 return -1; 1867 } 1868 1869 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg)) 1870 return -1; 1871 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0)) 1872 return -1; 1873 if (m->flags & MOPT_EXPLICIT) { 1874 if (m->mount_opt & EXT4_MOUNT_DELALLOC) { 1875 set_opt2(sb, EXPLICIT_DELALLOC); 1876 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) { 1877 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM); 1878 } else 1879 return -1; 1880 } 1881 if (m->flags & MOPT_CLEAR_ERR) 1882 clear_opt(sb, ERRORS_MASK); 1883 if (token == Opt_noquota && sb_any_quota_loaded(sb)) { 1884 ext4_msg(sb, KERN_ERR, "Cannot change quota " 1885 "options when quota turned on"); 1886 return -1; 1887 } 1888 1889 if (m->flags & MOPT_NOSUPPORT) { 1890 ext4_msg(sb, KERN_ERR, "%s option not supported", opt); 1891 } else if (token == Opt_commit) { 1892 if (arg == 0) 1893 arg = JBD2_DEFAULT_MAX_COMMIT_AGE; 1894 else if (arg > INT_MAX / HZ) { 1895 ext4_msg(sb, KERN_ERR, 1896 "Invalid commit interval %d, " 1897 "must be smaller than %d", 1898 arg, INT_MAX / HZ); 1899 return -1; 1900 } 1901 sbi->s_commit_interval = HZ * arg; 1902 } else if (token == Opt_debug_want_extra_isize) { 1903 sbi->s_want_extra_isize = arg; 1904 } else if (token == Opt_max_batch_time) { 1905 sbi->s_max_batch_time = arg; 1906 } else if (token == Opt_min_batch_time) { 1907 sbi->s_min_batch_time = arg; 1908 } else if (token == Opt_inode_readahead_blks) { 1909 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) { 1910 ext4_msg(sb, KERN_ERR, 1911 "EXT4-fs: inode_readahead_blks must be " 1912 "0 or a power of 2 smaller than 2^31"); 1913 return -1; 1914 } 1915 sbi->s_inode_readahead_blks = arg; 1916 } else if (token == Opt_init_itable) { 1917 set_opt(sb, INIT_INODE_TABLE); 1918 if (!args->from) 1919 arg = EXT4_DEF_LI_WAIT_MULT; 1920 sbi->s_li_wait_mult = arg; 1921 } else if (token == Opt_max_dir_size_kb) { 1922 sbi->s_max_dir_size_kb = arg; 1923 } else if (token == Opt_stripe) { 1924 sbi->s_stripe = arg; 1925 } else if (token == Opt_resuid) { 1926 uid = make_kuid(current_user_ns(), arg); 1927 if (!uid_valid(uid)) { 1928 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg); 1929 return -1; 1930 } 1931 sbi->s_resuid = uid; 1932 } else if (token == Opt_resgid) { 1933 gid = make_kgid(current_user_ns(), arg); 1934 if (!gid_valid(gid)) { 1935 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg); 1936 return -1; 1937 } 1938 sbi->s_resgid = gid; 1939 } else if (token == Opt_journal_dev) { 1940 if (is_remount) { 1941 ext4_msg(sb, KERN_ERR, 1942 "Cannot specify journal on remount"); 1943 return -1; 1944 } 1945 *journal_devnum = arg; 1946 } else if (token == Opt_journal_path) { 1947 char *journal_path; 1948 struct inode *journal_inode; 1949 struct path path; 1950 int error; 1951 1952 if (is_remount) { 1953 ext4_msg(sb, KERN_ERR, 1954 "Cannot specify journal on remount"); 1955 return -1; 1956 } 1957 journal_path = match_strdup(&args[0]); 1958 if (!journal_path) { 1959 ext4_msg(sb, KERN_ERR, "error: could not dup " 1960 "journal device string"); 1961 return -1; 1962 } 1963 1964 error = kern_path(journal_path, LOOKUP_FOLLOW, &path); 1965 if (error) { 1966 ext4_msg(sb, KERN_ERR, "error: could not find " 1967 "journal device path: error %d", error); 1968 kfree(journal_path); 1969 return -1; 1970 } 1971 1972 journal_inode = d_inode(path.dentry); 1973 if (!S_ISBLK(journal_inode->i_mode)) { 1974 ext4_msg(sb, KERN_ERR, "error: journal path %s " 1975 "is not a block device", journal_path); 1976 path_put(&path); 1977 kfree(journal_path); 1978 return -1; 1979 } 1980 1981 *journal_devnum = new_encode_dev(journal_inode->i_rdev); 1982 path_put(&path); 1983 kfree(journal_path); 1984 } else if (token == Opt_journal_ioprio) { 1985 if (arg > 7) { 1986 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority" 1987 " (must be 0-7)"); 1988 return -1; 1989 } 1990 *journal_ioprio = 1991 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg); 1992 } else if (token == Opt_test_dummy_encryption) { 1993 #ifdef CONFIG_FS_ENCRYPTION 1994 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION; 1995 ext4_msg(sb, KERN_WARNING, 1996 "Test dummy encryption mode enabled"); 1997 #else 1998 ext4_msg(sb, KERN_WARNING, 1999 "Test dummy encryption mount option ignored"); 2000 #endif 2001 } else if (m->flags & MOPT_DATAJ) { 2002 if (is_remount) { 2003 if (!sbi->s_journal) 2004 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option"); 2005 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) { 2006 ext4_msg(sb, KERN_ERR, 2007 "Cannot change data mode on remount"); 2008 return -1; 2009 } 2010 } else { 2011 clear_opt(sb, DATA_FLAGS); 2012 sbi->s_mount_opt |= m->mount_opt; 2013 } 2014 #ifdef CONFIG_QUOTA 2015 } else if (m->flags & MOPT_QFMT) { 2016 if (sb_any_quota_loaded(sb) && 2017 sbi->s_jquota_fmt != m->mount_opt) { 2018 ext4_msg(sb, KERN_ERR, "Cannot change journaled " 2019 "quota options when quota turned on"); 2020 return -1; 2021 } 2022 if (ext4_has_feature_quota(sb)) { 2023 ext4_msg(sb, KERN_INFO, 2024 "Quota format mount options ignored " 2025 "when QUOTA feature is enabled"); 2026 return 1; 2027 } 2028 sbi->s_jquota_fmt = m->mount_opt; 2029 #endif 2030 } else if (token == Opt_dax) { 2031 #ifdef CONFIG_FS_DAX 2032 ext4_msg(sb, KERN_WARNING, 2033 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 2034 sbi->s_mount_opt |= m->mount_opt; 2035 #else 2036 ext4_msg(sb, KERN_INFO, "dax option not supported"); 2037 return -1; 2038 #endif 2039 } else if (token == Opt_data_err_abort) { 2040 sbi->s_mount_opt |= m->mount_opt; 2041 } else if (token == Opt_data_err_ignore) { 2042 sbi->s_mount_opt &= ~m->mount_opt; 2043 } else { 2044 if (!args->from) 2045 arg = 1; 2046 if (m->flags & MOPT_CLEAR) 2047 arg = !arg; 2048 else if (unlikely(!(m->flags & MOPT_SET))) { 2049 ext4_msg(sb, KERN_WARNING, 2050 "buggy handling of option %s", opt); 2051 WARN_ON(1); 2052 return -1; 2053 } 2054 if (arg != 0) 2055 sbi->s_mount_opt |= m->mount_opt; 2056 else 2057 sbi->s_mount_opt &= ~m->mount_opt; 2058 } 2059 return 1; 2060 } 2061 2062 static int parse_options(char *options, struct super_block *sb, 2063 unsigned long *journal_devnum, 2064 unsigned int *journal_ioprio, 2065 int is_remount) 2066 { 2067 struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb); 2068 char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name; 2069 substring_t args[MAX_OPT_ARGS]; 2070 int token; 2071 2072 if (!options) 2073 return 1; 2074 2075 while ((p = strsep(&options, ",")) != NULL) { 2076 if (!*p) 2077 continue; 2078 /* 2079 * Initialize args struct so we know whether arg was 2080 * found; some options take optional arguments. 2081 */ 2082 args[0].to = args[0].from = NULL; 2083 token = match_token(p, tokens, args); 2084 if (handle_mount_opt(sb, p, token, args, journal_devnum, 2085 journal_ioprio, is_remount) < 0) 2086 return 0; 2087 } 2088 #ifdef CONFIG_QUOTA 2089 /* 2090 * We do the test below only for project quotas. 'usrquota' and 2091 * 'grpquota' mount options are allowed even without quota feature 2092 * to support legacy quotas in quota files. 2093 */ 2094 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) { 2095 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. " 2096 "Cannot enable project quota enforcement."); 2097 return 0; 2098 } 2099 usr_qf_name = get_qf_name(sb, sbi, USRQUOTA); 2100 grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA); 2101 if (usr_qf_name || grp_qf_name) { 2102 if (test_opt(sb, USRQUOTA) && usr_qf_name) 2103 clear_opt(sb, USRQUOTA); 2104 2105 if (test_opt(sb, GRPQUOTA) && grp_qf_name) 2106 clear_opt(sb, GRPQUOTA); 2107 2108 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) { 2109 ext4_msg(sb, KERN_ERR, "old and new quota " 2110 "format mixing"); 2111 return 0; 2112 } 2113 2114 if (!sbi->s_jquota_fmt) { 2115 ext4_msg(sb, KERN_ERR, "journaled quota format " 2116 "not specified"); 2117 return 0; 2118 } 2119 } 2120 #endif 2121 return 1; 2122 } 2123 2124 static inline void ext4_show_quota_options(struct seq_file *seq, 2125 struct super_block *sb) 2126 { 2127 #if defined(CONFIG_QUOTA) 2128 struct ext4_sb_info *sbi = EXT4_SB(sb); 2129 char *usr_qf_name, *grp_qf_name; 2130 2131 if (sbi->s_jquota_fmt) { 2132 char *fmtname = ""; 2133 2134 switch (sbi->s_jquota_fmt) { 2135 case QFMT_VFS_OLD: 2136 fmtname = "vfsold"; 2137 break; 2138 case QFMT_VFS_V0: 2139 fmtname = "vfsv0"; 2140 break; 2141 case QFMT_VFS_V1: 2142 fmtname = "vfsv1"; 2143 break; 2144 } 2145 seq_printf(seq, ",jqfmt=%s", fmtname); 2146 } 2147 2148 rcu_read_lock(); 2149 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]); 2150 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]); 2151 if (usr_qf_name) 2152 seq_show_option(seq, "usrjquota", usr_qf_name); 2153 if (grp_qf_name) 2154 seq_show_option(seq, "grpjquota", grp_qf_name); 2155 rcu_read_unlock(); 2156 #endif 2157 } 2158 2159 static const char *token2str(int token) 2160 { 2161 const struct match_token *t; 2162 2163 for (t = tokens; t->token != Opt_err; t++) 2164 if (t->token == token && !strchr(t->pattern, '=')) 2165 break; 2166 return t->pattern; 2167 } 2168 2169 /* 2170 * Show an option if 2171 * - it's set to a non-default value OR 2172 * - if the per-sb default is different from the global default 2173 */ 2174 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb, 2175 int nodefs) 2176 { 2177 struct ext4_sb_info *sbi = EXT4_SB(sb); 2178 struct ext4_super_block *es = sbi->s_es; 2179 int def_errors, def_mount_opt = sbi->s_def_mount_opt; 2180 const struct mount_opts *m; 2181 char sep = nodefs ? '\n' : ','; 2182 2183 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep) 2184 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg) 2185 2186 if (sbi->s_sb_block != 1) 2187 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block); 2188 2189 for (m = ext4_mount_opts; m->token != Opt_err; m++) { 2190 int want_set = m->flags & MOPT_SET; 2191 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) || 2192 (m->flags & MOPT_CLEAR_ERR)) 2193 continue; 2194 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt))) 2195 continue; /* skip if same as the default */ 2196 if ((want_set && 2197 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) || 2198 (!want_set && (sbi->s_mount_opt & m->mount_opt))) 2199 continue; /* select Opt_noFoo vs Opt_Foo */ 2200 SEQ_OPTS_PRINT("%s", token2str(m->token)); 2201 } 2202 2203 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) || 2204 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) 2205 SEQ_OPTS_PRINT("resuid=%u", 2206 from_kuid_munged(&init_user_ns, sbi->s_resuid)); 2207 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) || 2208 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) 2209 SEQ_OPTS_PRINT("resgid=%u", 2210 from_kgid_munged(&init_user_ns, sbi->s_resgid)); 2211 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors); 2212 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO) 2213 SEQ_OPTS_PUTS("errors=remount-ro"); 2214 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE) 2215 SEQ_OPTS_PUTS("errors=continue"); 2216 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC) 2217 SEQ_OPTS_PUTS("errors=panic"); 2218 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) 2219 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ); 2220 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) 2221 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time); 2222 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) 2223 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time); 2224 if (sb->s_flags & SB_I_VERSION) 2225 SEQ_OPTS_PUTS("i_version"); 2226 if (nodefs || sbi->s_stripe) 2227 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe); 2228 if (nodefs || EXT4_MOUNT_DATA_FLAGS & 2229 (sbi->s_mount_opt ^ def_mount_opt)) { 2230 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 2231 SEQ_OPTS_PUTS("data=journal"); 2232 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 2233 SEQ_OPTS_PUTS("data=ordered"); 2234 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 2235 SEQ_OPTS_PUTS("data=writeback"); 2236 } 2237 if (nodefs || 2238 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS) 2239 SEQ_OPTS_PRINT("inode_readahead_blks=%u", 2240 sbi->s_inode_readahead_blks); 2241 2242 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs || 2243 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT))) 2244 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult); 2245 if (nodefs || sbi->s_max_dir_size_kb) 2246 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb); 2247 if (test_opt(sb, DATA_ERR_ABORT)) 2248 SEQ_OPTS_PUTS("data_err=abort"); 2249 if (DUMMY_ENCRYPTION_ENABLED(sbi)) 2250 SEQ_OPTS_PUTS("test_dummy_encryption"); 2251 2252 ext4_show_quota_options(seq, sb); 2253 return 0; 2254 } 2255 2256 static int ext4_show_options(struct seq_file *seq, struct dentry *root) 2257 { 2258 return _ext4_show_options(seq, root->d_sb, 0); 2259 } 2260 2261 int ext4_seq_options_show(struct seq_file *seq, void *offset) 2262 { 2263 struct super_block *sb = seq->private; 2264 int rc; 2265 2266 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw"); 2267 rc = _ext4_show_options(seq, sb, 1); 2268 seq_puts(seq, "\n"); 2269 return rc; 2270 } 2271 2272 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es, 2273 int read_only) 2274 { 2275 struct ext4_sb_info *sbi = EXT4_SB(sb); 2276 int err = 0; 2277 2278 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) { 2279 ext4_msg(sb, KERN_ERR, "revision level too high, " 2280 "forcing read-only mode"); 2281 err = -EROFS; 2282 } 2283 if (read_only) 2284 goto done; 2285 if (!(sbi->s_mount_state & EXT4_VALID_FS)) 2286 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, " 2287 "running e2fsck is recommended"); 2288 else if (sbi->s_mount_state & EXT4_ERROR_FS) 2289 ext4_msg(sb, KERN_WARNING, 2290 "warning: mounting fs with errors, " 2291 "running e2fsck is recommended"); 2292 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 && 2293 le16_to_cpu(es->s_mnt_count) >= 2294 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count)) 2295 ext4_msg(sb, KERN_WARNING, 2296 "warning: maximal mount count reached, " 2297 "running e2fsck is recommended"); 2298 else if (le32_to_cpu(es->s_checkinterval) && 2299 (ext4_get_tstamp(es, s_lastcheck) + 2300 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds())) 2301 ext4_msg(sb, KERN_WARNING, 2302 "warning: checktime reached, " 2303 "running e2fsck is recommended"); 2304 if (!sbi->s_journal) 2305 es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 2306 if (!(__s16) le16_to_cpu(es->s_max_mnt_count)) 2307 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT); 2308 le16_add_cpu(&es->s_mnt_count, 1); 2309 ext4_update_tstamp(es, s_mtime); 2310 if (sbi->s_journal) 2311 ext4_set_feature_journal_needs_recovery(sb); 2312 2313 err = ext4_commit_super(sb, 1); 2314 done: 2315 if (test_opt(sb, DEBUG)) 2316 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, " 2317 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n", 2318 sb->s_blocksize, 2319 sbi->s_groups_count, 2320 EXT4_BLOCKS_PER_GROUP(sb), 2321 EXT4_INODES_PER_GROUP(sb), 2322 sbi->s_mount_opt, sbi->s_mount_opt2); 2323 2324 cleancache_init_fs(sb); 2325 return err; 2326 } 2327 2328 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup) 2329 { 2330 struct ext4_sb_info *sbi = EXT4_SB(sb); 2331 struct flex_groups *new_groups; 2332 int size; 2333 2334 if (!sbi->s_log_groups_per_flex) 2335 return 0; 2336 2337 size = ext4_flex_group(sbi, ngroup - 1) + 1; 2338 if (size <= sbi->s_flex_groups_allocated) 2339 return 0; 2340 2341 size = roundup_pow_of_two(size * sizeof(struct flex_groups)); 2342 new_groups = kvzalloc(size, GFP_KERNEL); 2343 if (!new_groups) { 2344 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups", 2345 size / (int) sizeof(struct flex_groups)); 2346 return -ENOMEM; 2347 } 2348 2349 if (sbi->s_flex_groups) { 2350 memcpy(new_groups, sbi->s_flex_groups, 2351 (sbi->s_flex_groups_allocated * 2352 sizeof(struct flex_groups))); 2353 kvfree(sbi->s_flex_groups); 2354 } 2355 sbi->s_flex_groups = new_groups; 2356 sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups); 2357 return 0; 2358 } 2359 2360 static int ext4_fill_flex_info(struct super_block *sb) 2361 { 2362 struct ext4_sb_info *sbi = EXT4_SB(sb); 2363 struct ext4_group_desc *gdp = NULL; 2364 ext4_group_t flex_group; 2365 int i, err; 2366 2367 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex; 2368 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) { 2369 sbi->s_log_groups_per_flex = 0; 2370 return 1; 2371 } 2372 2373 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count); 2374 if (err) 2375 goto failed; 2376 2377 for (i = 0; i < sbi->s_groups_count; i++) { 2378 gdp = ext4_get_group_desc(sb, i, NULL); 2379 2380 flex_group = ext4_flex_group(sbi, i); 2381 atomic_add(ext4_free_inodes_count(sb, gdp), 2382 &sbi->s_flex_groups[flex_group].free_inodes); 2383 atomic64_add(ext4_free_group_clusters(sb, gdp), 2384 &sbi->s_flex_groups[flex_group].free_clusters); 2385 atomic_add(ext4_used_dirs_count(sb, gdp), 2386 &sbi->s_flex_groups[flex_group].used_dirs); 2387 } 2388 2389 return 1; 2390 failed: 2391 return 0; 2392 } 2393 2394 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group, 2395 struct ext4_group_desc *gdp) 2396 { 2397 int offset = offsetof(struct ext4_group_desc, bg_checksum); 2398 __u16 crc = 0; 2399 __le32 le_group = cpu_to_le32(block_group); 2400 struct ext4_sb_info *sbi = EXT4_SB(sb); 2401 2402 if (ext4_has_metadata_csum(sbi->s_sb)) { 2403 /* Use new metadata_csum algorithm */ 2404 __u32 csum32; 2405 __u16 dummy_csum = 0; 2406 2407 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group, 2408 sizeof(le_group)); 2409 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset); 2410 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum, 2411 sizeof(dummy_csum)); 2412 offset += sizeof(dummy_csum); 2413 if (offset < sbi->s_desc_size) 2414 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset, 2415 sbi->s_desc_size - offset); 2416 2417 crc = csum32 & 0xFFFF; 2418 goto out; 2419 } 2420 2421 /* old crc16 code */ 2422 if (!ext4_has_feature_gdt_csum(sb)) 2423 return 0; 2424 2425 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid)); 2426 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group)); 2427 crc = crc16(crc, (__u8 *)gdp, offset); 2428 offset += sizeof(gdp->bg_checksum); /* skip checksum */ 2429 /* for checksum of struct ext4_group_desc do the rest...*/ 2430 if (ext4_has_feature_64bit(sb) && 2431 offset < le16_to_cpu(sbi->s_es->s_desc_size)) 2432 crc = crc16(crc, (__u8 *)gdp + offset, 2433 le16_to_cpu(sbi->s_es->s_desc_size) - 2434 offset); 2435 2436 out: 2437 return cpu_to_le16(crc); 2438 } 2439 2440 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group, 2441 struct ext4_group_desc *gdp) 2442 { 2443 if (ext4_has_group_desc_csum(sb) && 2444 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp))) 2445 return 0; 2446 2447 return 1; 2448 } 2449 2450 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group, 2451 struct ext4_group_desc *gdp) 2452 { 2453 if (!ext4_has_group_desc_csum(sb)) 2454 return; 2455 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp); 2456 } 2457 2458 /* Called at mount-time, super-block is locked */ 2459 static int ext4_check_descriptors(struct super_block *sb, 2460 ext4_fsblk_t sb_block, 2461 ext4_group_t *first_not_zeroed) 2462 { 2463 struct ext4_sb_info *sbi = EXT4_SB(sb); 2464 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 2465 ext4_fsblk_t last_block; 2466 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0); 2467 ext4_fsblk_t block_bitmap; 2468 ext4_fsblk_t inode_bitmap; 2469 ext4_fsblk_t inode_table; 2470 int flexbg_flag = 0; 2471 ext4_group_t i, grp = sbi->s_groups_count; 2472 2473 if (ext4_has_feature_flex_bg(sb)) 2474 flexbg_flag = 1; 2475 2476 ext4_debug("Checking group descriptors"); 2477 2478 for (i = 0; i < sbi->s_groups_count; i++) { 2479 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 2480 2481 if (i == sbi->s_groups_count - 1 || flexbg_flag) 2482 last_block = ext4_blocks_count(sbi->s_es) - 1; 2483 else 2484 last_block = first_block + 2485 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2486 2487 if ((grp == sbi->s_groups_count) && 2488 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 2489 grp = i; 2490 2491 block_bitmap = ext4_block_bitmap(sb, gdp); 2492 if (block_bitmap == sb_block) { 2493 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2494 "Block bitmap for group %u overlaps " 2495 "superblock", i); 2496 if (!sb_rdonly(sb)) 2497 return 0; 2498 } 2499 if (block_bitmap >= sb_block + 1 && 2500 block_bitmap <= last_bg_block) { 2501 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2502 "Block bitmap for group %u overlaps " 2503 "block group descriptors", i); 2504 if (!sb_rdonly(sb)) 2505 return 0; 2506 } 2507 if (block_bitmap < first_block || block_bitmap > last_block) { 2508 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2509 "Block bitmap for group %u not in group " 2510 "(block %llu)!", i, block_bitmap); 2511 return 0; 2512 } 2513 inode_bitmap = ext4_inode_bitmap(sb, gdp); 2514 if (inode_bitmap == sb_block) { 2515 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2516 "Inode bitmap for group %u overlaps " 2517 "superblock", i); 2518 if (!sb_rdonly(sb)) 2519 return 0; 2520 } 2521 if (inode_bitmap >= sb_block + 1 && 2522 inode_bitmap <= last_bg_block) { 2523 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2524 "Inode bitmap for group %u overlaps " 2525 "block group descriptors", i); 2526 if (!sb_rdonly(sb)) 2527 return 0; 2528 } 2529 if (inode_bitmap < first_block || inode_bitmap > last_block) { 2530 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2531 "Inode bitmap for group %u not in group " 2532 "(block %llu)!", i, inode_bitmap); 2533 return 0; 2534 } 2535 inode_table = ext4_inode_table(sb, gdp); 2536 if (inode_table == sb_block) { 2537 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2538 "Inode table for group %u overlaps " 2539 "superblock", i); 2540 if (!sb_rdonly(sb)) 2541 return 0; 2542 } 2543 if (inode_table >= sb_block + 1 && 2544 inode_table <= last_bg_block) { 2545 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2546 "Inode table for group %u overlaps " 2547 "block group descriptors", i); 2548 if (!sb_rdonly(sb)) 2549 return 0; 2550 } 2551 if (inode_table < first_block || 2552 inode_table + sbi->s_itb_per_group - 1 > last_block) { 2553 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2554 "Inode table for group %u not in group " 2555 "(block %llu)!", i, inode_table); 2556 return 0; 2557 } 2558 ext4_lock_group(sb, i); 2559 if (!ext4_group_desc_csum_verify(sb, i, gdp)) { 2560 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2561 "Checksum for group %u failed (%u!=%u)", 2562 i, le16_to_cpu(ext4_group_desc_csum(sb, i, 2563 gdp)), le16_to_cpu(gdp->bg_checksum)); 2564 if (!sb_rdonly(sb)) { 2565 ext4_unlock_group(sb, i); 2566 return 0; 2567 } 2568 } 2569 ext4_unlock_group(sb, i); 2570 if (!flexbg_flag) 2571 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2572 } 2573 if (NULL != first_not_zeroed) 2574 *first_not_zeroed = grp; 2575 return 1; 2576 } 2577 2578 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at 2579 * the superblock) which were deleted from all directories, but held open by 2580 * a process at the time of a crash. We walk the list and try to delete these 2581 * inodes at recovery time (only with a read-write filesystem). 2582 * 2583 * In order to keep the orphan inode chain consistent during traversal (in 2584 * case of crash during recovery), we link each inode into the superblock 2585 * orphan list_head and handle it the same way as an inode deletion during 2586 * normal operation (which journals the operations for us). 2587 * 2588 * We only do an iget() and an iput() on each inode, which is very safe if we 2589 * accidentally point at an in-use or already deleted inode. The worst that 2590 * can happen in this case is that we get a "bit already cleared" message from 2591 * ext4_free_inode(). The only reason we would point at a wrong inode is if 2592 * e2fsck was run on this filesystem, and it must have already done the orphan 2593 * inode cleanup for us, so we can safely abort without any further action. 2594 */ 2595 static void ext4_orphan_cleanup(struct super_block *sb, 2596 struct ext4_super_block *es) 2597 { 2598 unsigned int s_flags = sb->s_flags; 2599 int ret, nr_orphans = 0, nr_truncates = 0; 2600 #ifdef CONFIG_QUOTA 2601 int quota_update = 0; 2602 int i; 2603 #endif 2604 if (!es->s_last_orphan) { 2605 jbd_debug(4, "no orphan inodes to clean up\n"); 2606 return; 2607 } 2608 2609 if (bdev_read_only(sb->s_bdev)) { 2610 ext4_msg(sb, KERN_ERR, "write access " 2611 "unavailable, skipping orphan cleanup"); 2612 return; 2613 } 2614 2615 /* Check if feature set would not allow a r/w mount */ 2616 if (!ext4_feature_set_ok(sb, 0)) { 2617 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to " 2618 "unknown ROCOMPAT features"); 2619 return; 2620 } 2621 2622 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 2623 /* don't clear list on RO mount w/ errors */ 2624 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) { 2625 ext4_msg(sb, KERN_INFO, "Errors on filesystem, " 2626 "clearing orphan list.\n"); 2627 es->s_last_orphan = 0; 2628 } 2629 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 2630 return; 2631 } 2632 2633 if (s_flags & SB_RDONLY) { 2634 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs"); 2635 sb->s_flags &= ~SB_RDONLY; 2636 } 2637 #ifdef CONFIG_QUOTA 2638 /* Needed for iput() to work correctly and not trash data */ 2639 sb->s_flags |= SB_ACTIVE; 2640 2641 /* 2642 * Turn on quotas which were not enabled for read-only mounts if 2643 * filesystem has quota feature, so that they are updated correctly. 2644 */ 2645 if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) { 2646 int ret = ext4_enable_quotas(sb); 2647 2648 if (!ret) 2649 quota_update = 1; 2650 else 2651 ext4_msg(sb, KERN_ERR, 2652 "Cannot turn on quotas: error %d", ret); 2653 } 2654 2655 /* Turn on journaled quotas used for old sytle */ 2656 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2657 if (EXT4_SB(sb)->s_qf_names[i]) { 2658 int ret = ext4_quota_on_mount(sb, i); 2659 2660 if (!ret) 2661 quota_update = 1; 2662 else 2663 ext4_msg(sb, KERN_ERR, 2664 "Cannot turn on journaled " 2665 "quota: type %d: error %d", i, ret); 2666 } 2667 } 2668 #endif 2669 2670 while (es->s_last_orphan) { 2671 struct inode *inode; 2672 2673 /* 2674 * We may have encountered an error during cleanup; if 2675 * so, skip the rest. 2676 */ 2677 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 2678 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n"); 2679 es->s_last_orphan = 0; 2680 break; 2681 } 2682 2683 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)); 2684 if (IS_ERR(inode)) { 2685 es->s_last_orphan = 0; 2686 break; 2687 } 2688 2689 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); 2690 dquot_initialize(inode); 2691 if (inode->i_nlink) { 2692 if (test_opt(sb, DEBUG)) 2693 ext4_msg(sb, KERN_DEBUG, 2694 "%s: truncating inode %lu to %lld bytes", 2695 __func__, inode->i_ino, inode->i_size); 2696 jbd_debug(2, "truncating inode %lu to %lld bytes\n", 2697 inode->i_ino, inode->i_size); 2698 inode_lock(inode); 2699 truncate_inode_pages(inode->i_mapping, inode->i_size); 2700 ret = ext4_truncate(inode); 2701 if (ret) 2702 ext4_std_error(inode->i_sb, ret); 2703 inode_unlock(inode); 2704 nr_truncates++; 2705 } else { 2706 if (test_opt(sb, DEBUG)) 2707 ext4_msg(sb, KERN_DEBUG, 2708 "%s: deleting unreferenced inode %lu", 2709 __func__, inode->i_ino); 2710 jbd_debug(2, "deleting unreferenced inode %lu\n", 2711 inode->i_ino); 2712 nr_orphans++; 2713 } 2714 iput(inode); /* The delete magic happens here! */ 2715 } 2716 2717 #define PLURAL(x) (x), ((x) == 1) ? "" : "s" 2718 2719 if (nr_orphans) 2720 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted", 2721 PLURAL(nr_orphans)); 2722 if (nr_truncates) 2723 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", 2724 PLURAL(nr_truncates)); 2725 #ifdef CONFIG_QUOTA 2726 /* Turn off quotas if they were enabled for orphan cleanup */ 2727 if (quota_update) { 2728 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 2729 if (sb_dqopt(sb)->files[i]) 2730 dquot_quota_off(sb, i); 2731 } 2732 } 2733 #endif 2734 sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 2735 } 2736 2737 /* 2738 * Maximal extent format file size. 2739 * Resulting logical blkno at s_maxbytes must fit in our on-disk 2740 * extent format containers, within a sector_t, and within i_blocks 2741 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units, 2742 * so that won't be a limiting factor. 2743 * 2744 * However there is other limiting factor. We do store extents in the form 2745 * of starting block and length, hence the resulting length of the extent 2746 * covering maximum file size must fit into on-disk format containers as 2747 * well. Given that length is always by 1 unit bigger than max unit (because 2748 * we count 0 as well) we have to lower the s_maxbytes by one fs block. 2749 * 2750 * Note, this does *not* consider any metadata overhead for vfs i_blocks. 2751 */ 2752 static loff_t ext4_max_size(int blkbits, int has_huge_files) 2753 { 2754 loff_t res; 2755 loff_t upper_limit = MAX_LFS_FILESIZE; 2756 2757 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64)); 2758 2759 if (!has_huge_files) { 2760 upper_limit = (1LL << 32) - 1; 2761 2762 /* total blocks in file system block size */ 2763 upper_limit >>= (blkbits - 9); 2764 upper_limit <<= blkbits; 2765 } 2766 2767 /* 2768 * 32-bit extent-start container, ee_block. We lower the maxbytes 2769 * by one fs block, so ee_len can cover the extent of maximum file 2770 * size 2771 */ 2772 res = (1LL << 32) - 1; 2773 res <<= blkbits; 2774 2775 /* Sanity check against vm- & vfs- imposed limits */ 2776 if (res > upper_limit) 2777 res = upper_limit; 2778 2779 return res; 2780 } 2781 2782 /* 2783 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect 2784 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks. 2785 * We need to be 1 filesystem block less than the 2^48 sector limit. 2786 */ 2787 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) 2788 { 2789 loff_t res = EXT4_NDIR_BLOCKS; 2790 int meta_blocks; 2791 loff_t upper_limit; 2792 /* This is calculated to be the largest file size for a dense, block 2793 * mapped file such that the file's total number of 512-byte sectors, 2794 * including data and all indirect blocks, does not exceed (2^48 - 1). 2795 * 2796 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total 2797 * number of 512-byte sectors of the file. 2798 */ 2799 2800 if (!has_huge_files) { 2801 /* 2802 * !has_huge_files or implies that the inode i_block field 2803 * represents total file blocks in 2^32 512-byte sectors == 2804 * size of vfs inode i_blocks * 8 2805 */ 2806 upper_limit = (1LL << 32) - 1; 2807 2808 /* total blocks in file system block size */ 2809 upper_limit >>= (bits - 9); 2810 2811 } else { 2812 /* 2813 * We use 48 bit ext4_inode i_blocks 2814 * With EXT4_HUGE_FILE_FL set the i_blocks 2815 * represent total number of blocks in 2816 * file system block size 2817 */ 2818 upper_limit = (1LL << 48) - 1; 2819 2820 } 2821 2822 /* indirect blocks */ 2823 meta_blocks = 1; 2824 /* double indirect blocks */ 2825 meta_blocks += 1 + (1LL << (bits-2)); 2826 /* tripple indirect blocks */ 2827 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2))); 2828 2829 upper_limit -= meta_blocks; 2830 upper_limit <<= bits; 2831 2832 res += 1LL << (bits-2); 2833 res += 1LL << (2*(bits-2)); 2834 res += 1LL << (3*(bits-2)); 2835 res <<= bits; 2836 if (res > upper_limit) 2837 res = upper_limit; 2838 2839 if (res > MAX_LFS_FILESIZE) 2840 res = MAX_LFS_FILESIZE; 2841 2842 return res; 2843 } 2844 2845 static ext4_fsblk_t descriptor_loc(struct super_block *sb, 2846 ext4_fsblk_t logical_sb_block, int nr) 2847 { 2848 struct ext4_sb_info *sbi = EXT4_SB(sb); 2849 ext4_group_t bg, first_meta_bg; 2850 int has_super = 0; 2851 2852 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg); 2853 2854 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg) 2855 return logical_sb_block + nr + 1; 2856 bg = sbi->s_desc_per_block * nr; 2857 if (ext4_bg_has_super(sb, bg)) 2858 has_super = 1; 2859 2860 /* 2861 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at 2862 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled 2863 * on modern mke2fs or blksize > 1k on older mke2fs) then we must 2864 * compensate. 2865 */ 2866 if (sb->s_blocksize == 1024 && nr == 0 && 2867 le32_to_cpu(sbi->s_es->s_first_data_block) == 0) 2868 has_super++; 2869 2870 return (has_super + ext4_group_first_block_no(sb, bg)); 2871 } 2872 2873 /** 2874 * ext4_get_stripe_size: Get the stripe size. 2875 * @sbi: In memory super block info 2876 * 2877 * If we have specified it via mount option, then 2878 * use the mount option value. If the value specified at mount time is 2879 * greater than the blocks per group use the super block value. 2880 * If the super block value is greater than blocks per group return 0. 2881 * Allocator needs it be less than blocks per group. 2882 * 2883 */ 2884 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi) 2885 { 2886 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 2887 unsigned long stripe_width = 2888 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 2889 int ret; 2890 2891 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 2892 ret = sbi->s_stripe; 2893 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group) 2894 ret = stripe_width; 2895 else if (stride && stride <= sbi->s_blocks_per_group) 2896 ret = stride; 2897 else 2898 ret = 0; 2899 2900 /* 2901 * If the stripe width is 1, this makes no sense and 2902 * we set it to 0 to turn off stripe handling code. 2903 */ 2904 if (ret <= 1) 2905 ret = 0; 2906 2907 return ret; 2908 } 2909 2910 /* 2911 * Check whether this filesystem can be mounted based on 2912 * the features present and the RDONLY/RDWR mount requested. 2913 * Returns 1 if this filesystem can be mounted as requested, 2914 * 0 if it cannot be. 2915 */ 2916 static int ext4_feature_set_ok(struct super_block *sb, int readonly) 2917 { 2918 if (ext4_has_unknown_ext4_incompat_features(sb)) { 2919 ext4_msg(sb, KERN_ERR, 2920 "Couldn't mount because of " 2921 "unsupported optional features (%x)", 2922 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) & 2923 ~EXT4_FEATURE_INCOMPAT_SUPP)); 2924 return 0; 2925 } 2926 2927 #ifndef CONFIG_UNICODE 2928 if (ext4_has_feature_casefold(sb)) { 2929 ext4_msg(sb, KERN_ERR, 2930 "Filesystem with casefold feature cannot be " 2931 "mounted without CONFIG_UNICODE"); 2932 return 0; 2933 } 2934 #endif 2935 2936 if (readonly) 2937 return 1; 2938 2939 if (ext4_has_feature_readonly(sb)) { 2940 ext4_msg(sb, KERN_INFO, "filesystem is read-only"); 2941 sb->s_flags |= SB_RDONLY; 2942 return 1; 2943 } 2944 2945 /* Check that feature set is OK for a read-write mount */ 2946 if (ext4_has_unknown_ext4_ro_compat_features(sb)) { 2947 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of " 2948 "unsupported optional features (%x)", 2949 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) & 2950 ~EXT4_FEATURE_RO_COMPAT_SUPP)); 2951 return 0; 2952 } 2953 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) { 2954 ext4_msg(sb, KERN_ERR, 2955 "Can't support bigalloc feature without " 2956 "extents feature\n"); 2957 return 0; 2958 } 2959 2960 #ifndef CONFIG_QUOTA 2961 if (ext4_has_feature_quota(sb) && !readonly) { 2962 ext4_msg(sb, KERN_ERR, 2963 "Filesystem with quota feature cannot be mounted RDWR " 2964 "without CONFIG_QUOTA"); 2965 return 0; 2966 } 2967 if (ext4_has_feature_project(sb) && !readonly) { 2968 ext4_msg(sb, KERN_ERR, 2969 "Filesystem with project quota feature cannot be mounted RDWR " 2970 "without CONFIG_QUOTA"); 2971 return 0; 2972 } 2973 #endif /* CONFIG_QUOTA */ 2974 return 1; 2975 } 2976 2977 /* 2978 * This function is called once a day if we have errors logged 2979 * on the file system 2980 */ 2981 static void print_daily_error_info(struct timer_list *t) 2982 { 2983 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report); 2984 struct super_block *sb = sbi->s_sb; 2985 struct ext4_super_block *es = sbi->s_es; 2986 2987 if (es->s_error_count) 2988 /* fsck newer than v1.41.13 is needed to clean this condition. */ 2989 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u", 2990 le32_to_cpu(es->s_error_count)); 2991 if (es->s_first_error_time) { 2992 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d", 2993 sb->s_id, 2994 ext4_get_tstamp(es, s_first_error_time), 2995 (int) sizeof(es->s_first_error_func), 2996 es->s_first_error_func, 2997 le32_to_cpu(es->s_first_error_line)); 2998 if (es->s_first_error_ino) 2999 printk(KERN_CONT ": inode %u", 3000 le32_to_cpu(es->s_first_error_ino)); 3001 if (es->s_first_error_block) 3002 printk(KERN_CONT ": block %llu", (unsigned long long) 3003 le64_to_cpu(es->s_first_error_block)); 3004 printk(KERN_CONT "\n"); 3005 } 3006 if (es->s_last_error_time) { 3007 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d", 3008 sb->s_id, 3009 ext4_get_tstamp(es, s_last_error_time), 3010 (int) sizeof(es->s_last_error_func), 3011 es->s_last_error_func, 3012 le32_to_cpu(es->s_last_error_line)); 3013 if (es->s_last_error_ino) 3014 printk(KERN_CONT ": inode %u", 3015 le32_to_cpu(es->s_last_error_ino)); 3016 if (es->s_last_error_block) 3017 printk(KERN_CONT ": block %llu", (unsigned long long) 3018 le64_to_cpu(es->s_last_error_block)); 3019 printk(KERN_CONT "\n"); 3020 } 3021 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 3022 } 3023 3024 /* Find next suitable group and run ext4_init_inode_table */ 3025 static int ext4_run_li_request(struct ext4_li_request *elr) 3026 { 3027 struct ext4_group_desc *gdp = NULL; 3028 ext4_group_t group, ngroups; 3029 struct super_block *sb; 3030 unsigned long timeout = 0; 3031 int ret = 0; 3032 3033 sb = elr->lr_super; 3034 ngroups = EXT4_SB(sb)->s_groups_count; 3035 3036 for (group = elr->lr_next_group; group < ngroups; group++) { 3037 gdp = ext4_get_group_desc(sb, group, NULL); 3038 if (!gdp) { 3039 ret = 1; 3040 break; 3041 } 3042 3043 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3044 break; 3045 } 3046 3047 if (group >= ngroups) 3048 ret = 1; 3049 3050 if (!ret) { 3051 timeout = jiffies; 3052 ret = ext4_init_inode_table(sb, group, 3053 elr->lr_timeout ? 0 : 1); 3054 if (elr->lr_timeout == 0) { 3055 timeout = (jiffies - timeout) * 3056 elr->lr_sbi->s_li_wait_mult; 3057 elr->lr_timeout = timeout; 3058 } 3059 elr->lr_next_sched = jiffies + elr->lr_timeout; 3060 elr->lr_next_group = group + 1; 3061 } 3062 return ret; 3063 } 3064 3065 /* 3066 * Remove lr_request from the list_request and free the 3067 * request structure. Should be called with li_list_mtx held 3068 */ 3069 static void ext4_remove_li_request(struct ext4_li_request *elr) 3070 { 3071 struct ext4_sb_info *sbi; 3072 3073 if (!elr) 3074 return; 3075 3076 sbi = elr->lr_sbi; 3077 3078 list_del(&elr->lr_request); 3079 sbi->s_li_request = NULL; 3080 kfree(elr); 3081 } 3082 3083 static void ext4_unregister_li_request(struct super_block *sb) 3084 { 3085 mutex_lock(&ext4_li_mtx); 3086 if (!ext4_li_info) { 3087 mutex_unlock(&ext4_li_mtx); 3088 return; 3089 } 3090 3091 mutex_lock(&ext4_li_info->li_list_mtx); 3092 ext4_remove_li_request(EXT4_SB(sb)->s_li_request); 3093 mutex_unlock(&ext4_li_info->li_list_mtx); 3094 mutex_unlock(&ext4_li_mtx); 3095 } 3096 3097 static struct task_struct *ext4_lazyinit_task; 3098 3099 /* 3100 * This is the function where ext4lazyinit thread lives. It walks 3101 * through the request list searching for next scheduled filesystem. 3102 * When such a fs is found, run the lazy initialization request 3103 * (ext4_rn_li_request) and keep track of the time spend in this 3104 * function. Based on that time we compute next schedule time of 3105 * the request. When walking through the list is complete, compute 3106 * next waking time and put itself into sleep. 3107 */ 3108 static int ext4_lazyinit_thread(void *arg) 3109 { 3110 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg; 3111 struct list_head *pos, *n; 3112 struct ext4_li_request *elr; 3113 unsigned long next_wakeup, cur; 3114 3115 BUG_ON(NULL == eli); 3116 3117 cont_thread: 3118 while (true) { 3119 next_wakeup = MAX_JIFFY_OFFSET; 3120 3121 mutex_lock(&eli->li_list_mtx); 3122 if (list_empty(&eli->li_request_list)) { 3123 mutex_unlock(&eli->li_list_mtx); 3124 goto exit_thread; 3125 } 3126 list_for_each_safe(pos, n, &eli->li_request_list) { 3127 int err = 0; 3128 int progress = 0; 3129 elr = list_entry(pos, struct ext4_li_request, 3130 lr_request); 3131 3132 if (time_before(jiffies, elr->lr_next_sched)) { 3133 if (time_before(elr->lr_next_sched, next_wakeup)) 3134 next_wakeup = elr->lr_next_sched; 3135 continue; 3136 } 3137 if (down_read_trylock(&elr->lr_super->s_umount)) { 3138 if (sb_start_write_trylock(elr->lr_super)) { 3139 progress = 1; 3140 /* 3141 * We hold sb->s_umount, sb can not 3142 * be removed from the list, it is 3143 * now safe to drop li_list_mtx 3144 */ 3145 mutex_unlock(&eli->li_list_mtx); 3146 err = ext4_run_li_request(elr); 3147 sb_end_write(elr->lr_super); 3148 mutex_lock(&eli->li_list_mtx); 3149 n = pos->next; 3150 } 3151 up_read((&elr->lr_super->s_umount)); 3152 } 3153 /* error, remove the lazy_init job */ 3154 if (err) { 3155 ext4_remove_li_request(elr); 3156 continue; 3157 } 3158 if (!progress) { 3159 elr->lr_next_sched = jiffies + 3160 (prandom_u32() 3161 % (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3162 } 3163 if (time_before(elr->lr_next_sched, next_wakeup)) 3164 next_wakeup = elr->lr_next_sched; 3165 } 3166 mutex_unlock(&eli->li_list_mtx); 3167 3168 try_to_freeze(); 3169 3170 cur = jiffies; 3171 if ((time_after_eq(cur, next_wakeup)) || 3172 (MAX_JIFFY_OFFSET == next_wakeup)) { 3173 cond_resched(); 3174 continue; 3175 } 3176 3177 schedule_timeout_interruptible(next_wakeup - cur); 3178 3179 if (kthread_should_stop()) { 3180 ext4_clear_request_list(); 3181 goto exit_thread; 3182 } 3183 } 3184 3185 exit_thread: 3186 /* 3187 * It looks like the request list is empty, but we need 3188 * to check it under the li_list_mtx lock, to prevent any 3189 * additions into it, and of course we should lock ext4_li_mtx 3190 * to atomically free the list and ext4_li_info, because at 3191 * this point another ext4 filesystem could be registering 3192 * new one. 3193 */ 3194 mutex_lock(&ext4_li_mtx); 3195 mutex_lock(&eli->li_list_mtx); 3196 if (!list_empty(&eli->li_request_list)) { 3197 mutex_unlock(&eli->li_list_mtx); 3198 mutex_unlock(&ext4_li_mtx); 3199 goto cont_thread; 3200 } 3201 mutex_unlock(&eli->li_list_mtx); 3202 kfree(ext4_li_info); 3203 ext4_li_info = NULL; 3204 mutex_unlock(&ext4_li_mtx); 3205 3206 return 0; 3207 } 3208 3209 static void ext4_clear_request_list(void) 3210 { 3211 struct list_head *pos, *n; 3212 struct ext4_li_request *elr; 3213 3214 mutex_lock(&ext4_li_info->li_list_mtx); 3215 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) { 3216 elr = list_entry(pos, struct ext4_li_request, 3217 lr_request); 3218 ext4_remove_li_request(elr); 3219 } 3220 mutex_unlock(&ext4_li_info->li_list_mtx); 3221 } 3222 3223 static int ext4_run_lazyinit_thread(void) 3224 { 3225 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread, 3226 ext4_li_info, "ext4lazyinit"); 3227 if (IS_ERR(ext4_lazyinit_task)) { 3228 int err = PTR_ERR(ext4_lazyinit_task); 3229 ext4_clear_request_list(); 3230 kfree(ext4_li_info); 3231 ext4_li_info = NULL; 3232 printk(KERN_CRIT "EXT4-fs: error %d creating inode table " 3233 "initialization thread\n", 3234 err); 3235 return err; 3236 } 3237 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING; 3238 return 0; 3239 } 3240 3241 /* 3242 * Check whether it make sense to run itable init. thread or not. 3243 * If there is at least one uninitialized inode table, return 3244 * corresponding group number, else the loop goes through all 3245 * groups and return total number of groups. 3246 */ 3247 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb) 3248 { 3249 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count; 3250 struct ext4_group_desc *gdp = NULL; 3251 3252 if (!ext4_has_group_desc_csum(sb)) 3253 return ngroups; 3254 3255 for (group = 0; group < ngroups; group++) { 3256 gdp = ext4_get_group_desc(sb, group, NULL); 3257 if (!gdp) 3258 continue; 3259 3260 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))) 3261 break; 3262 } 3263 3264 return group; 3265 } 3266 3267 static int ext4_li_info_new(void) 3268 { 3269 struct ext4_lazy_init *eli = NULL; 3270 3271 eli = kzalloc(sizeof(*eli), GFP_KERNEL); 3272 if (!eli) 3273 return -ENOMEM; 3274 3275 INIT_LIST_HEAD(&eli->li_request_list); 3276 mutex_init(&eli->li_list_mtx); 3277 3278 eli->li_state |= EXT4_LAZYINIT_QUIT; 3279 3280 ext4_li_info = eli; 3281 3282 return 0; 3283 } 3284 3285 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb, 3286 ext4_group_t start) 3287 { 3288 struct ext4_sb_info *sbi = EXT4_SB(sb); 3289 struct ext4_li_request *elr; 3290 3291 elr = kzalloc(sizeof(*elr), GFP_KERNEL); 3292 if (!elr) 3293 return NULL; 3294 3295 elr->lr_super = sb; 3296 elr->lr_sbi = sbi; 3297 elr->lr_next_group = start; 3298 3299 /* 3300 * Randomize first schedule time of the request to 3301 * spread the inode table initialization requests 3302 * better. 3303 */ 3304 elr->lr_next_sched = jiffies + (prandom_u32() % 3305 (EXT4_DEF_LI_MAX_START_DELAY * HZ)); 3306 return elr; 3307 } 3308 3309 int ext4_register_li_request(struct super_block *sb, 3310 ext4_group_t first_not_zeroed) 3311 { 3312 struct ext4_sb_info *sbi = EXT4_SB(sb); 3313 struct ext4_li_request *elr = NULL; 3314 ext4_group_t ngroups = sbi->s_groups_count; 3315 int ret = 0; 3316 3317 mutex_lock(&ext4_li_mtx); 3318 if (sbi->s_li_request != NULL) { 3319 /* 3320 * Reset timeout so it can be computed again, because 3321 * s_li_wait_mult might have changed. 3322 */ 3323 sbi->s_li_request->lr_timeout = 0; 3324 goto out; 3325 } 3326 3327 if (first_not_zeroed == ngroups || sb_rdonly(sb) || 3328 !test_opt(sb, INIT_INODE_TABLE)) 3329 goto out; 3330 3331 elr = ext4_li_request_new(sb, first_not_zeroed); 3332 if (!elr) { 3333 ret = -ENOMEM; 3334 goto out; 3335 } 3336 3337 if (NULL == ext4_li_info) { 3338 ret = ext4_li_info_new(); 3339 if (ret) 3340 goto out; 3341 } 3342 3343 mutex_lock(&ext4_li_info->li_list_mtx); 3344 list_add(&elr->lr_request, &ext4_li_info->li_request_list); 3345 mutex_unlock(&ext4_li_info->li_list_mtx); 3346 3347 sbi->s_li_request = elr; 3348 /* 3349 * set elr to NULL here since it has been inserted to 3350 * the request_list and the removal and free of it is 3351 * handled by ext4_clear_request_list from now on. 3352 */ 3353 elr = NULL; 3354 3355 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) { 3356 ret = ext4_run_lazyinit_thread(); 3357 if (ret) 3358 goto out; 3359 } 3360 out: 3361 mutex_unlock(&ext4_li_mtx); 3362 if (ret) 3363 kfree(elr); 3364 return ret; 3365 } 3366 3367 /* 3368 * We do not need to lock anything since this is called on 3369 * module unload. 3370 */ 3371 static void ext4_destroy_lazyinit_thread(void) 3372 { 3373 /* 3374 * If thread exited earlier 3375 * there's nothing to be done. 3376 */ 3377 if (!ext4_li_info || !ext4_lazyinit_task) 3378 return; 3379 3380 kthread_stop(ext4_lazyinit_task); 3381 } 3382 3383 static int set_journal_csum_feature_set(struct super_block *sb) 3384 { 3385 int ret = 1; 3386 int compat, incompat; 3387 struct ext4_sb_info *sbi = EXT4_SB(sb); 3388 3389 if (ext4_has_metadata_csum(sb)) { 3390 /* journal checksum v3 */ 3391 compat = 0; 3392 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3; 3393 } else { 3394 /* journal checksum v1 */ 3395 compat = JBD2_FEATURE_COMPAT_CHECKSUM; 3396 incompat = 0; 3397 } 3398 3399 jbd2_journal_clear_features(sbi->s_journal, 3400 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 3401 JBD2_FEATURE_INCOMPAT_CSUM_V3 | 3402 JBD2_FEATURE_INCOMPAT_CSUM_V2); 3403 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 3404 ret = jbd2_journal_set_features(sbi->s_journal, 3405 compat, 0, 3406 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | 3407 incompat); 3408 } else if (test_opt(sb, JOURNAL_CHECKSUM)) { 3409 ret = jbd2_journal_set_features(sbi->s_journal, 3410 compat, 0, 3411 incompat); 3412 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3413 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3414 } else { 3415 jbd2_journal_clear_features(sbi->s_journal, 0, 0, 3416 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT); 3417 } 3418 3419 return ret; 3420 } 3421 3422 /* 3423 * Note: calculating the overhead so we can be compatible with 3424 * historical BSD practice is quite difficult in the face of 3425 * clusters/bigalloc. This is because multiple metadata blocks from 3426 * different block group can end up in the same allocation cluster. 3427 * Calculating the exact overhead in the face of clustered allocation 3428 * requires either O(all block bitmaps) in memory or O(number of block 3429 * groups**2) in time. We will still calculate the superblock for 3430 * older file systems --- and if we come across with a bigalloc file 3431 * system with zero in s_overhead_clusters the estimate will be close to 3432 * correct especially for very large cluster sizes --- but for newer 3433 * file systems, it's better to calculate this figure once at mkfs 3434 * time, and store it in the superblock. If the superblock value is 3435 * present (even for non-bigalloc file systems), we will use it. 3436 */ 3437 static int count_overhead(struct super_block *sb, ext4_group_t grp, 3438 char *buf) 3439 { 3440 struct ext4_sb_info *sbi = EXT4_SB(sb); 3441 struct ext4_group_desc *gdp; 3442 ext4_fsblk_t first_block, last_block, b; 3443 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3444 int s, j, count = 0; 3445 3446 if (!ext4_has_feature_bigalloc(sb)) 3447 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) + 3448 sbi->s_itb_per_group + 2); 3449 3450 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) + 3451 (grp * EXT4_BLOCKS_PER_GROUP(sb)); 3452 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1; 3453 for (i = 0; i < ngroups; i++) { 3454 gdp = ext4_get_group_desc(sb, i, NULL); 3455 b = ext4_block_bitmap(sb, gdp); 3456 if (b >= first_block && b <= last_block) { 3457 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3458 count++; 3459 } 3460 b = ext4_inode_bitmap(sb, gdp); 3461 if (b >= first_block && b <= last_block) { 3462 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf); 3463 count++; 3464 } 3465 b = ext4_inode_table(sb, gdp); 3466 if (b >= first_block && b + sbi->s_itb_per_group <= last_block) 3467 for (j = 0; j < sbi->s_itb_per_group; j++, b++) { 3468 int c = EXT4_B2C(sbi, b - first_block); 3469 ext4_set_bit(c, buf); 3470 count++; 3471 } 3472 if (i != grp) 3473 continue; 3474 s = 0; 3475 if (ext4_bg_has_super(sb, grp)) { 3476 ext4_set_bit(s++, buf); 3477 count++; 3478 } 3479 j = ext4_bg_num_gdb(sb, grp); 3480 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) { 3481 ext4_error(sb, "Invalid number of block group " 3482 "descriptor blocks: %d", j); 3483 j = EXT4_BLOCKS_PER_GROUP(sb) - s; 3484 } 3485 count += j; 3486 for (; j > 0; j--) 3487 ext4_set_bit(EXT4_B2C(sbi, s++), buf); 3488 } 3489 if (!count) 3490 return 0; 3491 return EXT4_CLUSTERS_PER_GROUP(sb) - 3492 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8); 3493 } 3494 3495 /* 3496 * Compute the overhead and stash it in sbi->s_overhead 3497 */ 3498 int ext4_calculate_overhead(struct super_block *sb) 3499 { 3500 struct ext4_sb_info *sbi = EXT4_SB(sb); 3501 struct ext4_super_block *es = sbi->s_es; 3502 struct inode *j_inode; 3503 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum); 3504 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 3505 ext4_fsblk_t overhead = 0; 3506 char *buf = (char *) get_zeroed_page(GFP_NOFS); 3507 3508 if (!buf) 3509 return -ENOMEM; 3510 3511 /* 3512 * Compute the overhead (FS structures). This is constant 3513 * for a given filesystem unless the number of block groups 3514 * changes so we cache the previous value until it does. 3515 */ 3516 3517 /* 3518 * All of the blocks before first_data_block are overhead 3519 */ 3520 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block)); 3521 3522 /* 3523 * Add the overhead found in each block group 3524 */ 3525 for (i = 0; i < ngroups; i++) { 3526 int blks; 3527 3528 blks = count_overhead(sb, i, buf); 3529 overhead += blks; 3530 if (blks) 3531 memset(buf, 0, PAGE_SIZE); 3532 cond_resched(); 3533 } 3534 3535 /* 3536 * Add the internal journal blocks whether the journal has been 3537 * loaded or not 3538 */ 3539 if (sbi->s_journal && !sbi->journal_bdev) 3540 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen); 3541 else if (ext4_has_feature_journal(sb) && !sbi->s_journal) { 3542 j_inode = ext4_get_journal_inode(sb, j_inum); 3543 if (j_inode) { 3544 j_blocks = j_inode->i_size >> sb->s_blocksize_bits; 3545 overhead += EXT4_NUM_B2C(sbi, j_blocks); 3546 iput(j_inode); 3547 } else { 3548 ext4_msg(sb, KERN_ERR, "can't get journal size"); 3549 } 3550 } 3551 sbi->s_overhead = overhead; 3552 smp_wmb(); 3553 free_page((unsigned long) buf); 3554 return 0; 3555 } 3556 3557 static void ext4_clamp_want_extra_isize(struct super_block *sb) 3558 { 3559 struct ext4_sb_info *sbi = EXT4_SB(sb); 3560 struct ext4_super_block *es = sbi->s_es; 3561 unsigned def_extra_isize = sizeof(struct ext4_inode) - 3562 EXT4_GOOD_OLD_INODE_SIZE; 3563 3564 if (sbi->s_inode_size == EXT4_GOOD_OLD_INODE_SIZE) { 3565 sbi->s_want_extra_isize = 0; 3566 return; 3567 } 3568 if (sbi->s_want_extra_isize < 4) { 3569 sbi->s_want_extra_isize = def_extra_isize; 3570 if (ext4_has_feature_extra_isize(sb)) { 3571 if (sbi->s_want_extra_isize < 3572 le16_to_cpu(es->s_want_extra_isize)) 3573 sbi->s_want_extra_isize = 3574 le16_to_cpu(es->s_want_extra_isize); 3575 if (sbi->s_want_extra_isize < 3576 le16_to_cpu(es->s_min_extra_isize)) 3577 sbi->s_want_extra_isize = 3578 le16_to_cpu(es->s_min_extra_isize); 3579 } 3580 } 3581 /* Check if enough inode space is available */ 3582 if ((sbi->s_want_extra_isize > sbi->s_inode_size) || 3583 (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize > 3584 sbi->s_inode_size)) { 3585 sbi->s_want_extra_isize = def_extra_isize; 3586 ext4_msg(sb, KERN_INFO, 3587 "required extra inode space not available"); 3588 } 3589 } 3590 3591 static void ext4_set_resv_clusters(struct super_block *sb) 3592 { 3593 ext4_fsblk_t resv_clusters; 3594 struct ext4_sb_info *sbi = EXT4_SB(sb); 3595 3596 /* 3597 * There's no need to reserve anything when we aren't using extents. 3598 * The space estimates are exact, there are no unwritten extents, 3599 * hole punching doesn't need new metadata... This is needed especially 3600 * to keep ext2/3 backward compatibility. 3601 */ 3602 if (!ext4_has_feature_extents(sb)) 3603 return; 3604 /* 3605 * By default we reserve 2% or 4096 clusters, whichever is smaller. 3606 * This should cover the situations where we can not afford to run 3607 * out of space like for example punch hole, or converting 3608 * unwritten extents in delalloc path. In most cases such 3609 * allocation would require 1, or 2 blocks, higher numbers are 3610 * very rare. 3611 */ 3612 resv_clusters = (ext4_blocks_count(sbi->s_es) >> 3613 sbi->s_cluster_bits); 3614 3615 do_div(resv_clusters, 50); 3616 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096); 3617 3618 atomic64_set(&sbi->s_resv_clusters, resv_clusters); 3619 } 3620 3621 static int ext4_fill_super(struct super_block *sb, void *data, int silent) 3622 { 3623 struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev); 3624 char *orig_data = kstrdup(data, GFP_KERNEL); 3625 struct buffer_head *bh; 3626 struct ext4_super_block *es = NULL; 3627 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 3628 ext4_fsblk_t block; 3629 ext4_fsblk_t sb_block = get_sb_block(&data); 3630 ext4_fsblk_t logical_sb_block; 3631 unsigned long offset = 0; 3632 unsigned long journal_devnum = 0; 3633 unsigned long def_mount_opts; 3634 struct inode *root; 3635 const char *descr; 3636 int ret = -ENOMEM; 3637 int blocksize, clustersize; 3638 unsigned int db_count; 3639 unsigned int i; 3640 int needs_recovery, has_huge_files, has_bigalloc; 3641 __u64 blocks_count; 3642 int err = 0; 3643 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 3644 ext4_group_t first_not_zeroed; 3645 3646 if ((data && !orig_data) || !sbi) 3647 goto out_free_base; 3648 3649 sbi->s_daxdev = dax_dev; 3650 sbi->s_blockgroup_lock = 3651 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL); 3652 if (!sbi->s_blockgroup_lock) 3653 goto out_free_base; 3654 3655 sb->s_fs_info = sbi; 3656 sbi->s_sb = sb; 3657 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS; 3658 sbi->s_sb_block = sb_block; 3659 if (sb->s_bdev->bd_part) 3660 sbi->s_sectors_written_start = 3661 part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]); 3662 3663 /* Cleanup superblock name */ 3664 strreplace(sb->s_id, '/', '!'); 3665 3666 /* -EINVAL is default */ 3667 ret = -EINVAL; 3668 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE); 3669 if (!blocksize) { 3670 ext4_msg(sb, KERN_ERR, "unable to set blocksize"); 3671 goto out_fail; 3672 } 3673 3674 /* 3675 * The ext4 superblock will not be buffer aligned for other than 1kB 3676 * block sizes. We need to calculate the offset from buffer start. 3677 */ 3678 if (blocksize != EXT4_MIN_BLOCK_SIZE) { 3679 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 3680 offset = do_div(logical_sb_block, blocksize); 3681 } else { 3682 logical_sb_block = sb_block; 3683 } 3684 3685 if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) { 3686 ext4_msg(sb, KERN_ERR, "unable to read superblock"); 3687 goto out_fail; 3688 } 3689 /* 3690 * Note: s_es must be initialized as soon as possible because 3691 * some ext4 macro-instructions depend on its value 3692 */ 3693 es = (struct ext4_super_block *) (bh->b_data + offset); 3694 sbi->s_es = es; 3695 sb->s_magic = le16_to_cpu(es->s_magic); 3696 if (sb->s_magic != EXT4_SUPER_MAGIC) 3697 goto cantfind_ext4; 3698 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written); 3699 3700 /* Warn if metadata_csum and gdt_csum are both set. */ 3701 if (ext4_has_feature_metadata_csum(sb) && 3702 ext4_has_feature_gdt_csum(sb)) 3703 ext4_warning(sb, "metadata_csum and uninit_bg are " 3704 "redundant flags; please run fsck."); 3705 3706 /* Check for a known checksum algorithm */ 3707 if (!ext4_verify_csum_type(sb, es)) { 3708 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 3709 "unknown checksum algorithm."); 3710 silent = 1; 3711 goto cantfind_ext4; 3712 } 3713 3714 /* Load the checksum driver */ 3715 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 3716 if (IS_ERR(sbi->s_chksum_driver)) { 3717 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver."); 3718 ret = PTR_ERR(sbi->s_chksum_driver); 3719 sbi->s_chksum_driver = NULL; 3720 goto failed_mount; 3721 } 3722 3723 /* Check superblock checksum */ 3724 if (!ext4_superblock_csum_verify(sb, es)) { 3725 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with " 3726 "invalid superblock checksum. Run e2fsck?"); 3727 silent = 1; 3728 ret = -EFSBADCRC; 3729 goto cantfind_ext4; 3730 } 3731 3732 /* Precompute checksum seed for all metadata */ 3733 if (ext4_has_feature_csum_seed(sb)) 3734 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed); 3735 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb)) 3736 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid, 3737 sizeof(es->s_uuid)); 3738 3739 /* Set defaults before we parse the mount options */ 3740 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 3741 set_opt(sb, INIT_INODE_TABLE); 3742 if (def_mount_opts & EXT4_DEFM_DEBUG) 3743 set_opt(sb, DEBUG); 3744 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) 3745 set_opt(sb, GRPID); 3746 if (def_mount_opts & EXT4_DEFM_UID16) 3747 set_opt(sb, NO_UID32); 3748 /* xattr user namespace & acls are now defaulted on */ 3749 set_opt(sb, XATTR_USER); 3750 #ifdef CONFIG_EXT4_FS_POSIX_ACL 3751 set_opt(sb, POSIX_ACL); 3752 #endif 3753 /* don't forget to enable journal_csum when metadata_csum is enabled. */ 3754 if (ext4_has_metadata_csum(sb)) 3755 set_opt(sb, JOURNAL_CHECKSUM); 3756 3757 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA) 3758 set_opt(sb, JOURNAL_DATA); 3759 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED) 3760 set_opt(sb, ORDERED_DATA); 3761 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK) 3762 set_opt(sb, WRITEBACK_DATA); 3763 3764 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC) 3765 set_opt(sb, ERRORS_PANIC); 3766 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE) 3767 set_opt(sb, ERRORS_CONT); 3768 else 3769 set_opt(sb, ERRORS_RO); 3770 /* block_validity enabled by default; disable with noblock_validity */ 3771 set_opt(sb, BLOCK_VALIDITY); 3772 if (def_mount_opts & EXT4_DEFM_DISCARD) 3773 set_opt(sb, DISCARD); 3774 3775 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid)); 3776 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid)); 3777 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ; 3778 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME; 3779 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME; 3780 3781 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0) 3782 set_opt(sb, BARRIER); 3783 3784 /* 3785 * enable delayed allocation by default 3786 * Use -o nodelalloc to turn it off 3787 */ 3788 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) && 3789 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0)) 3790 set_opt(sb, DELALLOC); 3791 3792 /* 3793 * set default s_li_wait_mult for lazyinit, for the case there is 3794 * no mount option specified. 3795 */ 3796 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT; 3797 3798 if (sbi->s_es->s_mount_opts[0]) { 3799 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts, 3800 sizeof(sbi->s_es->s_mount_opts), 3801 GFP_KERNEL); 3802 if (!s_mount_opts) 3803 goto failed_mount; 3804 if (!parse_options(s_mount_opts, sb, &journal_devnum, 3805 &journal_ioprio, 0)) { 3806 ext4_msg(sb, KERN_WARNING, 3807 "failed to parse options in superblock: %s", 3808 s_mount_opts); 3809 } 3810 kfree(s_mount_opts); 3811 } 3812 sbi->s_def_mount_opt = sbi->s_mount_opt; 3813 if (!parse_options((char *) data, sb, &journal_devnum, 3814 &journal_ioprio, 0)) 3815 goto failed_mount; 3816 3817 #ifdef CONFIG_UNICODE 3818 if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) { 3819 const struct ext4_sb_encodings *encoding_info; 3820 struct unicode_map *encoding; 3821 __u16 encoding_flags; 3822 3823 if (ext4_has_feature_encrypt(sb)) { 3824 ext4_msg(sb, KERN_ERR, 3825 "Can't mount with encoding and encryption"); 3826 goto failed_mount; 3827 } 3828 3829 if (ext4_sb_read_encoding(es, &encoding_info, 3830 &encoding_flags)) { 3831 ext4_msg(sb, KERN_ERR, 3832 "Encoding requested by superblock is unknown"); 3833 goto failed_mount; 3834 } 3835 3836 encoding = utf8_load(encoding_info->version); 3837 if (IS_ERR(encoding)) { 3838 ext4_msg(sb, KERN_ERR, 3839 "can't mount with superblock charset: %s-%s " 3840 "not supported by the kernel. flags: 0x%x.", 3841 encoding_info->name, encoding_info->version, 3842 encoding_flags); 3843 goto failed_mount; 3844 } 3845 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: " 3846 "%s-%s with flags 0x%hx", encoding_info->name, 3847 encoding_info->version?:"\b", encoding_flags); 3848 3849 sbi->s_encoding = encoding; 3850 sbi->s_encoding_flags = encoding_flags; 3851 } 3852 #endif 3853 3854 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 3855 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting " 3856 "with data=journal disables delayed " 3857 "allocation and O_DIRECT support!\n"); 3858 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 3859 ext4_msg(sb, KERN_ERR, "can't mount with " 3860 "both data=journal and delalloc"); 3861 goto failed_mount; 3862 } 3863 if (test_opt(sb, DIOREAD_NOLOCK)) { 3864 ext4_msg(sb, KERN_ERR, "can't mount with " 3865 "both data=journal and dioread_nolock"); 3866 goto failed_mount; 3867 } 3868 if (test_opt(sb, DAX)) { 3869 ext4_msg(sb, KERN_ERR, "can't mount with " 3870 "both data=journal and dax"); 3871 goto failed_mount; 3872 } 3873 if (ext4_has_feature_encrypt(sb)) { 3874 ext4_msg(sb, KERN_WARNING, 3875 "encrypted files will use data=ordered " 3876 "instead of data journaling mode"); 3877 } 3878 if (test_opt(sb, DELALLOC)) 3879 clear_opt(sb, DELALLOC); 3880 } else { 3881 sb->s_iflags |= SB_I_CGROUPWB; 3882 } 3883 3884 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 3885 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 3886 3887 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV && 3888 (ext4_has_compat_features(sb) || 3889 ext4_has_ro_compat_features(sb) || 3890 ext4_has_incompat_features(sb))) 3891 ext4_msg(sb, KERN_WARNING, 3892 "feature flags set on rev 0 fs, " 3893 "running e2fsck is recommended"); 3894 3895 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) { 3896 set_opt2(sb, HURD_COMPAT); 3897 if (ext4_has_feature_64bit(sb)) { 3898 ext4_msg(sb, KERN_ERR, 3899 "The Hurd can't support 64-bit file systems"); 3900 goto failed_mount; 3901 } 3902 3903 /* 3904 * ea_inode feature uses l_i_version field which is not 3905 * available in HURD_COMPAT mode. 3906 */ 3907 if (ext4_has_feature_ea_inode(sb)) { 3908 ext4_msg(sb, KERN_ERR, 3909 "ea_inode feature is not supported for Hurd"); 3910 goto failed_mount; 3911 } 3912 } 3913 3914 if (IS_EXT2_SB(sb)) { 3915 if (ext2_feature_set_ok(sb)) 3916 ext4_msg(sb, KERN_INFO, "mounting ext2 file system " 3917 "using the ext4 subsystem"); 3918 else { 3919 /* 3920 * If we're probing be silent, if this looks like 3921 * it's actually an ext[34] filesystem. 3922 */ 3923 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 3924 goto failed_mount; 3925 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due " 3926 "to feature incompatibilities"); 3927 goto failed_mount; 3928 } 3929 } 3930 3931 if (IS_EXT3_SB(sb)) { 3932 if (ext3_feature_set_ok(sb)) 3933 ext4_msg(sb, KERN_INFO, "mounting ext3 file system " 3934 "using the ext4 subsystem"); 3935 else { 3936 /* 3937 * If we're probing be silent, if this looks like 3938 * it's actually an ext4 filesystem. 3939 */ 3940 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb))) 3941 goto failed_mount; 3942 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due " 3943 "to feature incompatibilities"); 3944 goto failed_mount; 3945 } 3946 } 3947 3948 /* 3949 * Check feature flags regardless of the revision level, since we 3950 * previously didn't change the revision level when setting the flags, 3951 * so there is a chance incompat flags are set on a rev 0 filesystem. 3952 */ 3953 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb)))) 3954 goto failed_mount; 3955 3956 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size); 3957 if (blocksize < EXT4_MIN_BLOCK_SIZE || 3958 blocksize > EXT4_MAX_BLOCK_SIZE) { 3959 ext4_msg(sb, KERN_ERR, 3960 "Unsupported filesystem blocksize %d (%d log_block_size)", 3961 blocksize, le32_to_cpu(es->s_log_block_size)); 3962 goto failed_mount; 3963 } 3964 if (le32_to_cpu(es->s_log_block_size) > 3965 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 3966 ext4_msg(sb, KERN_ERR, 3967 "Invalid log block size: %u", 3968 le32_to_cpu(es->s_log_block_size)); 3969 goto failed_mount; 3970 } 3971 if (le32_to_cpu(es->s_log_cluster_size) > 3972 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) { 3973 ext4_msg(sb, KERN_ERR, 3974 "Invalid log cluster size: %u", 3975 le32_to_cpu(es->s_log_cluster_size)); 3976 goto failed_mount; 3977 } 3978 3979 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) { 3980 ext4_msg(sb, KERN_ERR, 3981 "Number of reserved GDT blocks insanely large: %d", 3982 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks)); 3983 goto failed_mount; 3984 } 3985 3986 if (sbi->s_mount_opt & EXT4_MOUNT_DAX) { 3987 if (ext4_has_feature_inline_data(sb)) { 3988 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem" 3989 " that may contain inline data"); 3990 goto failed_mount; 3991 } 3992 if (!bdev_dax_supported(sb->s_bdev, blocksize)) { 3993 ext4_msg(sb, KERN_ERR, 3994 "DAX unsupported by block device."); 3995 goto failed_mount; 3996 } 3997 } 3998 3999 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) { 4000 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d", 4001 es->s_encryption_level); 4002 goto failed_mount; 4003 } 4004 4005 if (sb->s_blocksize != blocksize) { 4006 /* Validate the filesystem blocksize */ 4007 if (!sb_set_blocksize(sb, blocksize)) { 4008 ext4_msg(sb, KERN_ERR, "bad block size %d", 4009 blocksize); 4010 goto failed_mount; 4011 } 4012 4013 brelse(bh); 4014 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE; 4015 offset = do_div(logical_sb_block, blocksize); 4016 bh = sb_bread_unmovable(sb, logical_sb_block); 4017 if (!bh) { 4018 ext4_msg(sb, KERN_ERR, 4019 "Can't read superblock on 2nd try"); 4020 goto failed_mount; 4021 } 4022 es = (struct ext4_super_block *)(bh->b_data + offset); 4023 sbi->s_es = es; 4024 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) { 4025 ext4_msg(sb, KERN_ERR, 4026 "Magic mismatch, very weird!"); 4027 goto failed_mount; 4028 } 4029 } 4030 4031 has_huge_files = ext4_has_feature_huge_file(sb); 4032 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits, 4033 has_huge_files); 4034 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files); 4035 4036 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) { 4037 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE; 4038 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO; 4039 } else { 4040 sbi->s_inode_size = le16_to_cpu(es->s_inode_size); 4041 sbi->s_first_ino = le32_to_cpu(es->s_first_ino); 4042 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) { 4043 ext4_msg(sb, KERN_ERR, "invalid first ino: %u", 4044 sbi->s_first_ino); 4045 goto failed_mount; 4046 } 4047 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) || 4048 (!is_power_of_2(sbi->s_inode_size)) || 4049 (sbi->s_inode_size > blocksize)) { 4050 ext4_msg(sb, KERN_ERR, 4051 "unsupported inode size: %d", 4052 sbi->s_inode_size); 4053 goto failed_mount; 4054 } 4055 /* 4056 * i_atime_extra is the last extra field available for [acm]times in 4057 * struct ext4_inode. Checking for that field should suffice to ensure 4058 * we have extra space for all three. 4059 */ 4060 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) + 4061 sizeof(((struct ext4_inode *)0)->i_atime_extra)) { 4062 sb->s_time_gran = 1; 4063 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX; 4064 } else { 4065 sb->s_time_gran = NSEC_PER_SEC; 4066 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX; 4067 } 4068 4069 sb->s_time_min = EXT4_TIMESTAMP_MIN; 4070 } 4071 4072 sbi->s_desc_size = le16_to_cpu(es->s_desc_size); 4073 if (ext4_has_feature_64bit(sb)) { 4074 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT || 4075 sbi->s_desc_size > EXT4_MAX_DESC_SIZE || 4076 !is_power_of_2(sbi->s_desc_size)) { 4077 ext4_msg(sb, KERN_ERR, 4078 "unsupported descriptor size %lu", 4079 sbi->s_desc_size); 4080 goto failed_mount; 4081 } 4082 } else 4083 sbi->s_desc_size = EXT4_MIN_DESC_SIZE; 4084 4085 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group); 4086 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group); 4087 4088 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb); 4089 if (sbi->s_inodes_per_block == 0) 4090 goto cantfind_ext4; 4091 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block || 4092 sbi->s_inodes_per_group > blocksize * 8) { 4093 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n", 4094 sbi->s_blocks_per_group); 4095 goto failed_mount; 4096 } 4097 sbi->s_itb_per_group = sbi->s_inodes_per_group / 4098 sbi->s_inodes_per_block; 4099 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb); 4100 sbi->s_sbh = bh; 4101 sbi->s_mount_state = le16_to_cpu(es->s_state); 4102 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb)); 4103 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb)); 4104 4105 for (i = 0; i < 4; i++) 4106 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]); 4107 sbi->s_def_hash_version = es->s_def_hash_version; 4108 if (ext4_has_feature_dir_index(sb)) { 4109 i = le32_to_cpu(es->s_flags); 4110 if (i & EXT2_FLAGS_UNSIGNED_HASH) 4111 sbi->s_hash_unsigned = 3; 4112 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) { 4113 #ifdef __CHAR_UNSIGNED__ 4114 if (!sb_rdonly(sb)) 4115 es->s_flags |= 4116 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH); 4117 sbi->s_hash_unsigned = 3; 4118 #else 4119 if (!sb_rdonly(sb)) 4120 es->s_flags |= 4121 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH); 4122 #endif 4123 } 4124 } 4125 4126 /* Handle clustersize */ 4127 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size); 4128 has_bigalloc = ext4_has_feature_bigalloc(sb); 4129 if (has_bigalloc) { 4130 if (clustersize < blocksize) { 4131 ext4_msg(sb, KERN_ERR, 4132 "cluster size (%d) smaller than " 4133 "block size (%d)", clustersize, blocksize); 4134 goto failed_mount; 4135 } 4136 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) - 4137 le32_to_cpu(es->s_log_block_size); 4138 sbi->s_clusters_per_group = 4139 le32_to_cpu(es->s_clusters_per_group); 4140 if (sbi->s_clusters_per_group > blocksize * 8) { 4141 ext4_msg(sb, KERN_ERR, 4142 "#clusters per group too big: %lu", 4143 sbi->s_clusters_per_group); 4144 goto failed_mount; 4145 } 4146 if (sbi->s_blocks_per_group != 4147 (sbi->s_clusters_per_group * (clustersize / blocksize))) { 4148 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and " 4149 "clusters per group (%lu) inconsistent", 4150 sbi->s_blocks_per_group, 4151 sbi->s_clusters_per_group); 4152 goto failed_mount; 4153 } 4154 } else { 4155 if (clustersize != blocksize) { 4156 ext4_msg(sb, KERN_ERR, 4157 "fragment/cluster size (%d) != " 4158 "block size (%d)", clustersize, blocksize); 4159 goto failed_mount; 4160 } 4161 if (sbi->s_blocks_per_group > blocksize * 8) { 4162 ext4_msg(sb, KERN_ERR, 4163 "#blocks per group too big: %lu", 4164 sbi->s_blocks_per_group); 4165 goto failed_mount; 4166 } 4167 sbi->s_clusters_per_group = sbi->s_blocks_per_group; 4168 sbi->s_cluster_bits = 0; 4169 } 4170 sbi->s_cluster_ratio = clustersize / blocksize; 4171 4172 /* Do we have standard group size of clustersize * 8 blocks ? */ 4173 if (sbi->s_blocks_per_group == clustersize << 3) 4174 set_opt2(sb, STD_GROUP_SIZE); 4175 4176 /* 4177 * Test whether we have more sectors than will fit in sector_t, 4178 * and whether the max offset is addressable by the page cache. 4179 */ 4180 err = generic_check_addressable(sb->s_blocksize_bits, 4181 ext4_blocks_count(es)); 4182 if (err) { 4183 ext4_msg(sb, KERN_ERR, "filesystem" 4184 " too large to mount safely on this system"); 4185 goto failed_mount; 4186 } 4187 4188 if (EXT4_BLOCKS_PER_GROUP(sb) == 0) 4189 goto cantfind_ext4; 4190 4191 /* check blocks count against device size */ 4192 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits; 4193 if (blocks_count && ext4_blocks_count(es) > blocks_count) { 4194 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu " 4195 "exceeds size of device (%llu blocks)", 4196 ext4_blocks_count(es), blocks_count); 4197 goto failed_mount; 4198 } 4199 4200 /* 4201 * It makes no sense for the first data block to be beyond the end 4202 * of the filesystem. 4203 */ 4204 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) { 4205 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4206 "block %u is beyond end of filesystem (%llu)", 4207 le32_to_cpu(es->s_first_data_block), 4208 ext4_blocks_count(es)); 4209 goto failed_mount; 4210 } 4211 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) && 4212 (sbi->s_cluster_ratio == 1)) { 4213 ext4_msg(sb, KERN_WARNING, "bad geometry: first data " 4214 "block is 0 with a 1k block and cluster size"); 4215 goto failed_mount; 4216 } 4217 4218 blocks_count = (ext4_blocks_count(es) - 4219 le32_to_cpu(es->s_first_data_block) + 4220 EXT4_BLOCKS_PER_GROUP(sb) - 1); 4221 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb)); 4222 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) { 4223 ext4_msg(sb, KERN_WARNING, "groups count too large: %u " 4224 "(block count %llu, first data block %u, " 4225 "blocks per group %lu)", sbi->s_groups_count, 4226 ext4_blocks_count(es), 4227 le32_to_cpu(es->s_first_data_block), 4228 EXT4_BLOCKS_PER_GROUP(sb)); 4229 goto failed_mount; 4230 } 4231 sbi->s_groups_count = blocks_count; 4232 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 4233 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 4234 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) != 4235 le32_to_cpu(es->s_inodes_count)) { 4236 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu", 4237 le32_to_cpu(es->s_inodes_count), 4238 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group)); 4239 ret = -EINVAL; 4240 goto failed_mount; 4241 } 4242 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 4243 EXT4_DESC_PER_BLOCK(sb); 4244 if (ext4_has_feature_meta_bg(sb)) { 4245 if (le32_to_cpu(es->s_first_meta_bg) > db_count) { 4246 ext4_msg(sb, KERN_WARNING, 4247 "first meta block group too large: %u " 4248 "(group descriptor block count %u)", 4249 le32_to_cpu(es->s_first_meta_bg), db_count); 4250 goto failed_mount; 4251 } 4252 } 4253 sbi->s_group_desc = kvmalloc_array(db_count, 4254 sizeof(struct buffer_head *), 4255 GFP_KERNEL); 4256 if (sbi->s_group_desc == NULL) { 4257 ext4_msg(sb, KERN_ERR, "not enough memory"); 4258 ret = -ENOMEM; 4259 goto failed_mount; 4260 } 4261 4262 bgl_lock_init(sbi->s_blockgroup_lock); 4263 4264 /* Pre-read the descriptors into the buffer cache */ 4265 for (i = 0; i < db_count; i++) { 4266 block = descriptor_loc(sb, logical_sb_block, i); 4267 sb_breadahead(sb, block); 4268 } 4269 4270 for (i = 0; i < db_count; i++) { 4271 block = descriptor_loc(sb, logical_sb_block, i); 4272 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block); 4273 if (!sbi->s_group_desc[i]) { 4274 ext4_msg(sb, KERN_ERR, 4275 "can't read group descriptor %d", i); 4276 db_count = i; 4277 goto failed_mount2; 4278 } 4279 } 4280 sbi->s_gdb_count = db_count; 4281 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) { 4282 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 4283 ret = -EFSCORRUPTED; 4284 goto failed_mount2; 4285 } 4286 4287 timer_setup(&sbi->s_err_report, print_daily_error_info, 0); 4288 4289 /* Register extent status tree shrinker */ 4290 if (ext4_es_register_shrinker(sbi)) 4291 goto failed_mount3; 4292 4293 sbi->s_stripe = ext4_get_stripe_size(sbi); 4294 sbi->s_extent_max_zeroout_kb = 32; 4295 4296 /* 4297 * set up enough so that it can read an inode 4298 */ 4299 sb->s_op = &ext4_sops; 4300 sb->s_export_op = &ext4_export_ops; 4301 sb->s_xattr = ext4_xattr_handlers; 4302 #ifdef CONFIG_FS_ENCRYPTION 4303 sb->s_cop = &ext4_cryptops; 4304 #endif 4305 #ifdef CONFIG_FS_VERITY 4306 sb->s_vop = &ext4_verityops; 4307 #endif 4308 #ifdef CONFIG_QUOTA 4309 sb->dq_op = &ext4_quota_operations; 4310 if (ext4_has_feature_quota(sb)) 4311 sb->s_qcop = &dquot_quotactl_sysfile_ops; 4312 else 4313 sb->s_qcop = &ext4_qctl_operations; 4314 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 4315 #endif 4316 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid)); 4317 4318 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 4319 mutex_init(&sbi->s_orphan_lock); 4320 4321 sb->s_root = NULL; 4322 4323 needs_recovery = (es->s_last_orphan != 0 || 4324 ext4_has_feature_journal_needs_recovery(sb)); 4325 4326 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) 4327 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block))) 4328 goto failed_mount3a; 4329 4330 /* 4331 * The first inode we look at is the journal inode. Don't try 4332 * root first: it may be modified in the journal! 4333 */ 4334 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) { 4335 err = ext4_load_journal(sb, es, journal_devnum); 4336 if (err) 4337 goto failed_mount3a; 4338 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) && 4339 ext4_has_feature_journal_needs_recovery(sb)) { 4340 ext4_msg(sb, KERN_ERR, "required journal recovery " 4341 "suppressed and not mounted read-only"); 4342 goto failed_mount_wq; 4343 } else { 4344 /* Nojournal mode, all journal mount options are illegal */ 4345 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) { 4346 ext4_msg(sb, KERN_ERR, "can't mount with " 4347 "journal_checksum, fs mounted w/o journal"); 4348 goto failed_mount_wq; 4349 } 4350 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4351 ext4_msg(sb, KERN_ERR, "can't mount with " 4352 "journal_async_commit, fs mounted w/o journal"); 4353 goto failed_mount_wq; 4354 } 4355 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) { 4356 ext4_msg(sb, KERN_ERR, "can't mount with " 4357 "commit=%lu, fs mounted w/o journal", 4358 sbi->s_commit_interval / HZ); 4359 goto failed_mount_wq; 4360 } 4361 if (EXT4_MOUNT_DATA_FLAGS & 4362 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) { 4363 ext4_msg(sb, KERN_ERR, "can't mount with " 4364 "data=, fs mounted w/o journal"); 4365 goto failed_mount_wq; 4366 } 4367 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM; 4368 clear_opt(sb, JOURNAL_CHECKSUM); 4369 clear_opt(sb, DATA_FLAGS); 4370 sbi->s_journal = NULL; 4371 needs_recovery = 0; 4372 goto no_journal; 4373 } 4374 4375 if (ext4_has_feature_64bit(sb) && 4376 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0, 4377 JBD2_FEATURE_INCOMPAT_64BIT)) { 4378 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature"); 4379 goto failed_mount_wq; 4380 } 4381 4382 if (!set_journal_csum_feature_set(sb)) { 4383 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum " 4384 "feature set"); 4385 goto failed_mount_wq; 4386 } 4387 4388 /* We have now updated the journal if required, so we can 4389 * validate the data journaling mode. */ 4390 switch (test_opt(sb, DATA_FLAGS)) { 4391 case 0: 4392 /* No mode set, assume a default based on the journal 4393 * capabilities: ORDERED_DATA if the journal can 4394 * cope, else JOURNAL_DATA 4395 */ 4396 if (jbd2_journal_check_available_features 4397 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4398 set_opt(sb, ORDERED_DATA); 4399 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA; 4400 } else { 4401 set_opt(sb, JOURNAL_DATA); 4402 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA; 4403 } 4404 break; 4405 4406 case EXT4_MOUNT_ORDERED_DATA: 4407 case EXT4_MOUNT_WRITEBACK_DATA: 4408 if (!jbd2_journal_check_available_features 4409 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) { 4410 ext4_msg(sb, KERN_ERR, "Journal does not support " 4411 "requested data journaling mode"); 4412 goto failed_mount_wq; 4413 } 4414 default: 4415 break; 4416 } 4417 4418 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA && 4419 test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 4420 ext4_msg(sb, KERN_ERR, "can't mount with " 4421 "journal_async_commit in data=ordered mode"); 4422 goto failed_mount_wq; 4423 } 4424 4425 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 4426 4427 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback; 4428 4429 no_journal: 4430 if (!test_opt(sb, NO_MBCACHE)) { 4431 sbi->s_ea_block_cache = ext4_xattr_create_cache(); 4432 if (!sbi->s_ea_block_cache) { 4433 ext4_msg(sb, KERN_ERR, 4434 "Failed to create ea_block_cache"); 4435 goto failed_mount_wq; 4436 } 4437 4438 if (ext4_has_feature_ea_inode(sb)) { 4439 sbi->s_ea_inode_cache = ext4_xattr_create_cache(); 4440 if (!sbi->s_ea_inode_cache) { 4441 ext4_msg(sb, KERN_ERR, 4442 "Failed to create ea_inode_cache"); 4443 goto failed_mount_wq; 4444 } 4445 } 4446 } 4447 4448 if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) { 4449 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity"); 4450 goto failed_mount_wq; 4451 } 4452 4453 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) && 4454 !ext4_has_feature_encrypt(sb)) { 4455 ext4_set_feature_encrypt(sb); 4456 ext4_commit_super(sb, 1); 4457 } 4458 4459 /* 4460 * Get the # of file system overhead blocks from the 4461 * superblock if present. 4462 */ 4463 if (es->s_overhead_clusters) 4464 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters); 4465 else { 4466 err = ext4_calculate_overhead(sb); 4467 if (err) 4468 goto failed_mount_wq; 4469 } 4470 4471 /* 4472 * The maximum number of concurrent works can be high and 4473 * concurrency isn't really necessary. Limit it to 1. 4474 */ 4475 EXT4_SB(sb)->rsv_conversion_wq = 4476 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1); 4477 if (!EXT4_SB(sb)->rsv_conversion_wq) { 4478 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n"); 4479 ret = -ENOMEM; 4480 goto failed_mount4; 4481 } 4482 4483 /* 4484 * The jbd2_journal_load will have done any necessary log recovery, 4485 * so we can safely mount the rest of the filesystem now. 4486 */ 4487 4488 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL); 4489 if (IS_ERR(root)) { 4490 ext4_msg(sb, KERN_ERR, "get root inode failed"); 4491 ret = PTR_ERR(root); 4492 root = NULL; 4493 goto failed_mount4; 4494 } 4495 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) { 4496 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck"); 4497 iput(root); 4498 goto failed_mount4; 4499 } 4500 4501 #ifdef CONFIG_UNICODE 4502 if (sbi->s_encoding) 4503 sb->s_d_op = &ext4_dentry_ops; 4504 #endif 4505 4506 sb->s_root = d_make_root(root); 4507 if (!sb->s_root) { 4508 ext4_msg(sb, KERN_ERR, "get root dentry failed"); 4509 ret = -ENOMEM; 4510 goto failed_mount4; 4511 } 4512 4513 ret = ext4_setup_super(sb, es, sb_rdonly(sb)); 4514 if (ret == -EROFS) { 4515 sb->s_flags |= SB_RDONLY; 4516 ret = 0; 4517 } else if (ret) 4518 goto failed_mount4a; 4519 4520 ext4_clamp_want_extra_isize(sb); 4521 4522 ext4_set_resv_clusters(sb); 4523 4524 err = ext4_setup_system_zone(sb); 4525 if (err) { 4526 ext4_msg(sb, KERN_ERR, "failed to initialize system " 4527 "zone (%d)", err); 4528 goto failed_mount4a; 4529 } 4530 4531 ext4_ext_init(sb); 4532 err = ext4_mb_init(sb); 4533 if (err) { 4534 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)", 4535 err); 4536 goto failed_mount5; 4537 } 4538 4539 block = ext4_count_free_clusters(sb); 4540 ext4_free_blocks_count_set(sbi->s_es, 4541 EXT4_C2B(sbi, block)); 4542 ext4_superblock_csum_set(sb); 4543 err = percpu_counter_init(&sbi->s_freeclusters_counter, block, 4544 GFP_KERNEL); 4545 if (!err) { 4546 unsigned long freei = ext4_count_free_inodes(sb); 4547 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei); 4548 ext4_superblock_csum_set(sb); 4549 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei, 4550 GFP_KERNEL); 4551 } 4552 if (!err) 4553 err = percpu_counter_init(&sbi->s_dirs_counter, 4554 ext4_count_dirs(sb), GFP_KERNEL); 4555 if (!err) 4556 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0, 4557 GFP_KERNEL); 4558 if (!err) 4559 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem); 4560 4561 if (err) { 4562 ext4_msg(sb, KERN_ERR, "insufficient memory"); 4563 goto failed_mount6; 4564 } 4565 4566 if (ext4_has_feature_flex_bg(sb)) 4567 if (!ext4_fill_flex_info(sb)) { 4568 ext4_msg(sb, KERN_ERR, 4569 "unable to initialize " 4570 "flex_bg meta info!"); 4571 goto failed_mount6; 4572 } 4573 4574 err = ext4_register_li_request(sb, first_not_zeroed); 4575 if (err) 4576 goto failed_mount6; 4577 4578 err = ext4_register_sysfs(sb); 4579 if (err) 4580 goto failed_mount7; 4581 4582 #ifdef CONFIG_QUOTA 4583 /* Enable quota usage during mount. */ 4584 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) { 4585 err = ext4_enable_quotas(sb); 4586 if (err) 4587 goto failed_mount8; 4588 } 4589 #endif /* CONFIG_QUOTA */ 4590 4591 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS; 4592 ext4_orphan_cleanup(sb, es); 4593 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS; 4594 if (needs_recovery) { 4595 ext4_msg(sb, KERN_INFO, "recovery complete"); 4596 ext4_mark_recovery_complete(sb, es); 4597 } 4598 if (EXT4_SB(sb)->s_journal) { 4599 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 4600 descr = " journalled data mode"; 4601 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 4602 descr = " ordered data mode"; 4603 else 4604 descr = " writeback data mode"; 4605 } else 4606 descr = "out journal"; 4607 4608 if (test_opt(sb, DISCARD)) { 4609 struct request_queue *q = bdev_get_queue(sb->s_bdev); 4610 if (!blk_queue_discard(q)) 4611 ext4_msg(sb, KERN_WARNING, 4612 "mounting with \"discard\" option, but " 4613 "the device does not support discard"); 4614 } 4615 4616 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount")) 4617 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. " 4618 "Opts: %.*s%s%s", descr, 4619 (int) sizeof(sbi->s_es->s_mount_opts), 4620 sbi->s_es->s_mount_opts, 4621 *sbi->s_es->s_mount_opts ? "; " : "", orig_data); 4622 4623 if (es->s_error_count) 4624 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */ 4625 4626 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */ 4627 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10); 4628 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10); 4629 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10); 4630 4631 kfree(orig_data); 4632 return 0; 4633 4634 cantfind_ext4: 4635 if (!silent) 4636 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem"); 4637 goto failed_mount; 4638 4639 #ifdef CONFIG_QUOTA 4640 failed_mount8: 4641 ext4_unregister_sysfs(sb); 4642 #endif 4643 failed_mount7: 4644 ext4_unregister_li_request(sb); 4645 failed_mount6: 4646 ext4_mb_release(sb); 4647 if (sbi->s_flex_groups) 4648 kvfree(sbi->s_flex_groups); 4649 percpu_counter_destroy(&sbi->s_freeclusters_counter); 4650 percpu_counter_destroy(&sbi->s_freeinodes_counter); 4651 percpu_counter_destroy(&sbi->s_dirs_counter); 4652 percpu_counter_destroy(&sbi->s_dirtyclusters_counter); 4653 percpu_free_rwsem(&sbi->s_journal_flag_rwsem); 4654 failed_mount5: 4655 ext4_ext_release(sb); 4656 ext4_release_system_zone(sb); 4657 failed_mount4a: 4658 dput(sb->s_root); 4659 sb->s_root = NULL; 4660 failed_mount4: 4661 ext4_msg(sb, KERN_ERR, "mount failed"); 4662 if (EXT4_SB(sb)->rsv_conversion_wq) 4663 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq); 4664 failed_mount_wq: 4665 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache); 4666 sbi->s_ea_inode_cache = NULL; 4667 4668 ext4_xattr_destroy_cache(sbi->s_ea_block_cache); 4669 sbi->s_ea_block_cache = NULL; 4670 4671 if (sbi->s_journal) { 4672 jbd2_journal_destroy(sbi->s_journal); 4673 sbi->s_journal = NULL; 4674 } 4675 failed_mount3a: 4676 ext4_es_unregister_shrinker(sbi); 4677 failed_mount3: 4678 del_timer_sync(&sbi->s_err_report); 4679 if (sbi->s_mmp_tsk) 4680 kthread_stop(sbi->s_mmp_tsk); 4681 failed_mount2: 4682 for (i = 0; i < db_count; i++) 4683 brelse(sbi->s_group_desc[i]); 4684 kvfree(sbi->s_group_desc); 4685 failed_mount: 4686 if (sbi->s_chksum_driver) 4687 crypto_free_shash(sbi->s_chksum_driver); 4688 4689 #ifdef CONFIG_UNICODE 4690 utf8_unload(sbi->s_encoding); 4691 #endif 4692 4693 #ifdef CONFIG_QUOTA 4694 for (i = 0; i < EXT4_MAXQUOTAS; i++) 4695 kfree(get_qf_name(sb, sbi, i)); 4696 #endif 4697 ext4_blkdev_remove(sbi); 4698 brelse(bh); 4699 out_fail: 4700 sb->s_fs_info = NULL; 4701 kfree(sbi->s_blockgroup_lock); 4702 out_free_base: 4703 kfree(sbi); 4704 kfree(orig_data); 4705 fs_put_dax(dax_dev); 4706 return err ? err : ret; 4707 } 4708 4709 /* 4710 * Setup any per-fs journal parameters now. We'll do this both on 4711 * initial mount, once the journal has been initialised but before we've 4712 * done any recovery; and again on any subsequent remount. 4713 */ 4714 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) 4715 { 4716 struct ext4_sb_info *sbi = EXT4_SB(sb); 4717 4718 journal->j_commit_interval = sbi->s_commit_interval; 4719 journal->j_min_batch_time = sbi->s_min_batch_time; 4720 journal->j_max_batch_time = sbi->s_max_batch_time; 4721 4722 write_lock(&journal->j_state_lock); 4723 if (test_opt(sb, BARRIER)) 4724 journal->j_flags |= JBD2_BARRIER; 4725 else 4726 journal->j_flags &= ~JBD2_BARRIER; 4727 if (test_opt(sb, DATA_ERR_ABORT)) 4728 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR; 4729 else 4730 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR; 4731 write_unlock(&journal->j_state_lock); 4732 } 4733 4734 static struct inode *ext4_get_journal_inode(struct super_block *sb, 4735 unsigned int journal_inum) 4736 { 4737 struct inode *journal_inode; 4738 4739 /* 4740 * Test for the existence of a valid inode on disk. Bad things 4741 * happen if we iget() an unused inode, as the subsequent iput() 4742 * will try to delete it. 4743 */ 4744 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL); 4745 if (IS_ERR(journal_inode)) { 4746 ext4_msg(sb, KERN_ERR, "no journal found"); 4747 return NULL; 4748 } 4749 if (!journal_inode->i_nlink) { 4750 make_bad_inode(journal_inode); 4751 iput(journal_inode); 4752 ext4_msg(sb, KERN_ERR, "journal inode is deleted"); 4753 return NULL; 4754 } 4755 4756 jbd_debug(2, "Journal inode found at %p: %lld bytes\n", 4757 journal_inode, journal_inode->i_size); 4758 if (!S_ISREG(journal_inode->i_mode)) { 4759 ext4_msg(sb, KERN_ERR, "invalid journal inode"); 4760 iput(journal_inode); 4761 return NULL; 4762 } 4763 return journal_inode; 4764 } 4765 4766 static journal_t *ext4_get_journal(struct super_block *sb, 4767 unsigned int journal_inum) 4768 { 4769 struct inode *journal_inode; 4770 journal_t *journal; 4771 4772 BUG_ON(!ext4_has_feature_journal(sb)); 4773 4774 journal_inode = ext4_get_journal_inode(sb, journal_inum); 4775 if (!journal_inode) 4776 return NULL; 4777 4778 journal = jbd2_journal_init_inode(journal_inode); 4779 if (!journal) { 4780 ext4_msg(sb, KERN_ERR, "Could not load journal inode"); 4781 iput(journal_inode); 4782 return NULL; 4783 } 4784 journal->j_private = sb; 4785 ext4_init_journal_params(sb, journal); 4786 return journal; 4787 } 4788 4789 static journal_t *ext4_get_dev_journal(struct super_block *sb, 4790 dev_t j_dev) 4791 { 4792 struct buffer_head *bh; 4793 journal_t *journal; 4794 ext4_fsblk_t start; 4795 ext4_fsblk_t len; 4796 int hblock, blocksize; 4797 ext4_fsblk_t sb_block; 4798 unsigned long offset; 4799 struct ext4_super_block *es; 4800 struct block_device *bdev; 4801 4802 BUG_ON(!ext4_has_feature_journal(sb)); 4803 4804 bdev = ext4_blkdev_get(j_dev, sb); 4805 if (bdev == NULL) 4806 return NULL; 4807 4808 blocksize = sb->s_blocksize; 4809 hblock = bdev_logical_block_size(bdev); 4810 if (blocksize < hblock) { 4811 ext4_msg(sb, KERN_ERR, 4812 "blocksize too small for journal device"); 4813 goto out_bdev; 4814 } 4815 4816 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize; 4817 offset = EXT4_MIN_BLOCK_SIZE % blocksize; 4818 set_blocksize(bdev, blocksize); 4819 if (!(bh = __bread(bdev, sb_block, blocksize))) { 4820 ext4_msg(sb, KERN_ERR, "couldn't read superblock of " 4821 "external journal"); 4822 goto out_bdev; 4823 } 4824 4825 es = (struct ext4_super_block *) (bh->b_data + offset); 4826 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) || 4827 !(le32_to_cpu(es->s_feature_incompat) & 4828 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) { 4829 ext4_msg(sb, KERN_ERR, "external journal has " 4830 "bad superblock"); 4831 brelse(bh); 4832 goto out_bdev; 4833 } 4834 4835 if ((le32_to_cpu(es->s_feature_ro_compat) & 4836 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) && 4837 es->s_checksum != ext4_superblock_csum(sb, es)) { 4838 ext4_msg(sb, KERN_ERR, "external journal has " 4839 "corrupt superblock"); 4840 brelse(bh); 4841 goto out_bdev; 4842 } 4843 4844 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) { 4845 ext4_msg(sb, KERN_ERR, "journal UUID does not match"); 4846 brelse(bh); 4847 goto out_bdev; 4848 } 4849 4850 len = ext4_blocks_count(es); 4851 start = sb_block + 1; 4852 brelse(bh); /* we're done with the superblock */ 4853 4854 journal = jbd2_journal_init_dev(bdev, sb->s_bdev, 4855 start, len, blocksize); 4856 if (!journal) { 4857 ext4_msg(sb, KERN_ERR, "failed to create device journal"); 4858 goto out_bdev; 4859 } 4860 journal->j_private = sb; 4861 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer); 4862 wait_on_buffer(journal->j_sb_buffer); 4863 if (!buffer_uptodate(journal->j_sb_buffer)) { 4864 ext4_msg(sb, KERN_ERR, "I/O error on journal device"); 4865 goto out_journal; 4866 } 4867 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) { 4868 ext4_msg(sb, KERN_ERR, "External journal has more than one " 4869 "user (unsupported) - %d", 4870 be32_to_cpu(journal->j_superblock->s_nr_users)); 4871 goto out_journal; 4872 } 4873 EXT4_SB(sb)->journal_bdev = bdev; 4874 ext4_init_journal_params(sb, journal); 4875 return journal; 4876 4877 out_journal: 4878 jbd2_journal_destroy(journal); 4879 out_bdev: 4880 ext4_blkdev_put(bdev); 4881 return NULL; 4882 } 4883 4884 static int ext4_load_journal(struct super_block *sb, 4885 struct ext4_super_block *es, 4886 unsigned long journal_devnum) 4887 { 4888 journal_t *journal; 4889 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum); 4890 dev_t journal_dev; 4891 int err = 0; 4892 int really_read_only; 4893 4894 BUG_ON(!ext4_has_feature_journal(sb)); 4895 4896 if (journal_devnum && 4897 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 4898 ext4_msg(sb, KERN_INFO, "external journal device major/minor " 4899 "numbers have changed"); 4900 journal_dev = new_decode_dev(journal_devnum); 4901 } else 4902 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev)); 4903 4904 really_read_only = bdev_read_only(sb->s_bdev); 4905 4906 /* 4907 * Are we loading a blank journal or performing recovery after a 4908 * crash? For recovery, we need to check in advance whether we 4909 * can get read-write access to the device. 4910 */ 4911 if (ext4_has_feature_journal_needs_recovery(sb)) { 4912 if (sb_rdonly(sb)) { 4913 ext4_msg(sb, KERN_INFO, "INFO: recovery " 4914 "required on readonly filesystem"); 4915 if (really_read_only) { 4916 ext4_msg(sb, KERN_ERR, "write access " 4917 "unavailable, cannot proceed " 4918 "(try mounting with noload)"); 4919 return -EROFS; 4920 } 4921 ext4_msg(sb, KERN_INFO, "write access will " 4922 "be enabled during recovery"); 4923 } 4924 } 4925 4926 if (journal_inum && journal_dev) { 4927 ext4_msg(sb, KERN_ERR, "filesystem has both journal " 4928 "and inode journals!"); 4929 return -EINVAL; 4930 } 4931 4932 if (journal_inum) { 4933 if (!(journal = ext4_get_journal(sb, journal_inum))) 4934 return -EINVAL; 4935 } else { 4936 if (!(journal = ext4_get_dev_journal(sb, journal_dev))) 4937 return -EINVAL; 4938 } 4939 4940 if (!(journal->j_flags & JBD2_BARRIER)) 4941 ext4_msg(sb, KERN_INFO, "barriers disabled"); 4942 4943 if (!ext4_has_feature_journal_needs_recovery(sb)) 4944 err = jbd2_journal_wipe(journal, !really_read_only); 4945 if (!err) { 4946 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL); 4947 if (save) 4948 memcpy(save, ((char *) es) + 4949 EXT4_S_ERR_START, EXT4_S_ERR_LEN); 4950 err = jbd2_journal_load(journal); 4951 if (save) 4952 memcpy(((char *) es) + EXT4_S_ERR_START, 4953 save, EXT4_S_ERR_LEN); 4954 kfree(save); 4955 } 4956 4957 if (err) { 4958 ext4_msg(sb, KERN_ERR, "error loading journal"); 4959 jbd2_journal_destroy(journal); 4960 return err; 4961 } 4962 4963 EXT4_SB(sb)->s_journal = journal; 4964 ext4_clear_journal_err(sb, es); 4965 4966 if (!really_read_only && journal_devnum && 4967 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 4968 es->s_journal_dev = cpu_to_le32(journal_devnum); 4969 4970 /* Make sure we flush the recovery flag to disk. */ 4971 ext4_commit_super(sb, 1); 4972 } 4973 4974 return 0; 4975 } 4976 4977 static int ext4_commit_super(struct super_block *sb, int sync) 4978 { 4979 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 4980 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh; 4981 int error = 0; 4982 4983 if (!sbh || block_device_ejected(sb)) 4984 return error; 4985 4986 /* 4987 * The superblock bh should be mapped, but it might not be if the 4988 * device was hot-removed. Not much we can do but fail the I/O. 4989 */ 4990 if (!buffer_mapped(sbh)) 4991 return error; 4992 4993 /* 4994 * If the file system is mounted read-only, don't update the 4995 * superblock write time. This avoids updating the superblock 4996 * write time when we are mounting the root file system 4997 * read/only but we need to replay the journal; at that point, 4998 * for people who are east of GMT and who make their clock 4999 * tick in localtime for Windows bug-for-bug compatibility, 5000 * the clock is set in the future, and this will cause e2fsck 5001 * to complain and force a full file system check. 5002 */ 5003 if (!(sb->s_flags & SB_RDONLY)) 5004 ext4_update_tstamp(es, s_wtime); 5005 if (sb->s_bdev->bd_part) 5006 es->s_kbytes_written = 5007 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 5008 ((part_stat_read(sb->s_bdev->bd_part, 5009 sectors[STAT_WRITE]) - 5010 EXT4_SB(sb)->s_sectors_written_start) >> 1)); 5011 else 5012 es->s_kbytes_written = 5013 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written); 5014 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter)) 5015 ext4_free_blocks_count_set(es, 5016 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive( 5017 &EXT4_SB(sb)->s_freeclusters_counter))); 5018 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter)) 5019 es->s_free_inodes_count = 5020 cpu_to_le32(percpu_counter_sum_positive( 5021 &EXT4_SB(sb)->s_freeinodes_counter)); 5022 BUFFER_TRACE(sbh, "marking dirty"); 5023 ext4_superblock_csum_set(sb); 5024 if (sync) 5025 lock_buffer(sbh); 5026 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) { 5027 /* 5028 * Oh, dear. A previous attempt to write the 5029 * superblock failed. This could happen because the 5030 * USB device was yanked out. Or it could happen to 5031 * be a transient write error and maybe the block will 5032 * be remapped. Nothing we can do but to retry the 5033 * write and hope for the best. 5034 */ 5035 ext4_msg(sb, KERN_ERR, "previous I/O error to " 5036 "superblock detected"); 5037 clear_buffer_write_io_error(sbh); 5038 set_buffer_uptodate(sbh); 5039 } 5040 mark_buffer_dirty(sbh); 5041 if (sync) { 5042 unlock_buffer(sbh); 5043 error = __sync_dirty_buffer(sbh, 5044 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0)); 5045 if (buffer_write_io_error(sbh)) { 5046 ext4_msg(sb, KERN_ERR, "I/O error while writing " 5047 "superblock"); 5048 clear_buffer_write_io_error(sbh); 5049 set_buffer_uptodate(sbh); 5050 } 5051 } 5052 return error; 5053 } 5054 5055 /* 5056 * Have we just finished recovery? If so, and if we are mounting (or 5057 * remounting) the filesystem readonly, then we will end up with a 5058 * consistent fs on disk. Record that fact. 5059 */ 5060 static void ext4_mark_recovery_complete(struct super_block *sb, 5061 struct ext4_super_block *es) 5062 { 5063 journal_t *journal = EXT4_SB(sb)->s_journal; 5064 5065 if (!ext4_has_feature_journal(sb)) { 5066 BUG_ON(journal != NULL); 5067 return; 5068 } 5069 jbd2_journal_lock_updates(journal); 5070 if (jbd2_journal_flush(journal) < 0) 5071 goto out; 5072 5073 if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) { 5074 ext4_clear_feature_journal_needs_recovery(sb); 5075 ext4_commit_super(sb, 1); 5076 } 5077 5078 out: 5079 jbd2_journal_unlock_updates(journal); 5080 } 5081 5082 /* 5083 * If we are mounting (or read-write remounting) a filesystem whose journal 5084 * has recorded an error from a previous lifetime, move that error to the 5085 * main filesystem now. 5086 */ 5087 static void ext4_clear_journal_err(struct super_block *sb, 5088 struct ext4_super_block *es) 5089 { 5090 journal_t *journal; 5091 int j_errno; 5092 const char *errstr; 5093 5094 BUG_ON(!ext4_has_feature_journal(sb)); 5095 5096 journal = EXT4_SB(sb)->s_journal; 5097 5098 /* 5099 * Now check for any error status which may have been recorded in the 5100 * journal by a prior ext4_error() or ext4_abort() 5101 */ 5102 5103 j_errno = jbd2_journal_errno(journal); 5104 if (j_errno) { 5105 char nbuf[16]; 5106 5107 errstr = ext4_decode_error(sb, j_errno, nbuf); 5108 ext4_warning(sb, "Filesystem error recorded " 5109 "from previous mount: %s", errstr); 5110 ext4_warning(sb, "Marking fs in need of filesystem check."); 5111 5112 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS; 5113 es->s_state |= cpu_to_le16(EXT4_ERROR_FS); 5114 ext4_commit_super(sb, 1); 5115 5116 jbd2_journal_clear_err(journal); 5117 jbd2_journal_update_sb_errno(journal); 5118 } 5119 } 5120 5121 /* 5122 * Force the running and committing transactions to commit, 5123 * and wait on the commit. 5124 */ 5125 int ext4_force_commit(struct super_block *sb) 5126 { 5127 journal_t *journal; 5128 5129 if (sb_rdonly(sb)) 5130 return 0; 5131 5132 journal = EXT4_SB(sb)->s_journal; 5133 return ext4_journal_force_commit(journal); 5134 } 5135 5136 static int ext4_sync_fs(struct super_block *sb, int wait) 5137 { 5138 int ret = 0; 5139 tid_t target; 5140 bool needs_barrier = false; 5141 struct ext4_sb_info *sbi = EXT4_SB(sb); 5142 5143 if (unlikely(ext4_forced_shutdown(sbi))) 5144 return 0; 5145 5146 trace_ext4_sync_fs(sb, wait); 5147 flush_workqueue(sbi->rsv_conversion_wq); 5148 /* 5149 * Writeback quota in non-journalled quota case - journalled quota has 5150 * no dirty dquots 5151 */ 5152 dquot_writeback_dquots(sb, -1); 5153 /* 5154 * Data writeback is possible w/o journal transaction, so barrier must 5155 * being sent at the end of the function. But we can skip it if 5156 * transaction_commit will do it for us. 5157 */ 5158 if (sbi->s_journal) { 5159 target = jbd2_get_latest_transaction(sbi->s_journal); 5160 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER && 5161 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target)) 5162 needs_barrier = true; 5163 5164 if (jbd2_journal_start_commit(sbi->s_journal, &target)) { 5165 if (wait) 5166 ret = jbd2_log_wait_commit(sbi->s_journal, 5167 target); 5168 } 5169 } else if (wait && test_opt(sb, BARRIER)) 5170 needs_barrier = true; 5171 if (needs_barrier) { 5172 int err; 5173 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL); 5174 if (!ret) 5175 ret = err; 5176 } 5177 5178 return ret; 5179 } 5180 5181 /* 5182 * LVM calls this function before a (read-only) snapshot is created. This 5183 * gives us a chance to flush the journal completely and mark the fs clean. 5184 * 5185 * Note that only this function cannot bring a filesystem to be in a clean 5186 * state independently. It relies on upper layer to stop all data & metadata 5187 * modifications. 5188 */ 5189 static int ext4_freeze(struct super_block *sb) 5190 { 5191 int error = 0; 5192 journal_t *journal; 5193 5194 if (sb_rdonly(sb)) 5195 return 0; 5196 5197 journal = EXT4_SB(sb)->s_journal; 5198 5199 if (journal) { 5200 /* Now we set up the journal barrier. */ 5201 jbd2_journal_lock_updates(journal); 5202 5203 /* 5204 * Don't clear the needs_recovery flag if we failed to 5205 * flush the journal. 5206 */ 5207 error = jbd2_journal_flush(journal); 5208 if (error < 0) 5209 goto out; 5210 5211 /* Journal blocked and flushed, clear needs_recovery flag. */ 5212 ext4_clear_feature_journal_needs_recovery(sb); 5213 } 5214 5215 error = ext4_commit_super(sb, 1); 5216 out: 5217 if (journal) 5218 /* we rely on upper layer to stop further updates */ 5219 jbd2_journal_unlock_updates(journal); 5220 return error; 5221 } 5222 5223 /* 5224 * Called by LVM after the snapshot is done. We need to reset the RECOVER 5225 * flag here, even though the filesystem is not technically dirty yet. 5226 */ 5227 static int ext4_unfreeze(struct super_block *sb) 5228 { 5229 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb))) 5230 return 0; 5231 5232 if (EXT4_SB(sb)->s_journal) { 5233 /* Reset the needs_recovery flag before the fs is unlocked. */ 5234 ext4_set_feature_journal_needs_recovery(sb); 5235 } 5236 5237 ext4_commit_super(sb, 1); 5238 return 0; 5239 } 5240 5241 /* 5242 * Structure to save mount options for ext4_remount's benefit 5243 */ 5244 struct ext4_mount_options { 5245 unsigned long s_mount_opt; 5246 unsigned long s_mount_opt2; 5247 kuid_t s_resuid; 5248 kgid_t s_resgid; 5249 unsigned long s_commit_interval; 5250 u32 s_min_batch_time, s_max_batch_time; 5251 #ifdef CONFIG_QUOTA 5252 int s_jquota_fmt; 5253 char *s_qf_names[EXT4_MAXQUOTAS]; 5254 #endif 5255 }; 5256 5257 static int ext4_remount(struct super_block *sb, int *flags, char *data) 5258 { 5259 struct ext4_super_block *es; 5260 struct ext4_sb_info *sbi = EXT4_SB(sb); 5261 unsigned long old_sb_flags; 5262 struct ext4_mount_options old_opts; 5263 int enable_quota = 0; 5264 ext4_group_t g; 5265 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 5266 int err = 0; 5267 #ifdef CONFIG_QUOTA 5268 int i, j; 5269 char *to_free[EXT4_MAXQUOTAS]; 5270 #endif 5271 char *orig_data = kstrdup(data, GFP_KERNEL); 5272 5273 if (data && !orig_data) 5274 return -ENOMEM; 5275 5276 /* Store the original options */ 5277 old_sb_flags = sb->s_flags; 5278 old_opts.s_mount_opt = sbi->s_mount_opt; 5279 old_opts.s_mount_opt2 = sbi->s_mount_opt2; 5280 old_opts.s_resuid = sbi->s_resuid; 5281 old_opts.s_resgid = sbi->s_resgid; 5282 old_opts.s_commit_interval = sbi->s_commit_interval; 5283 old_opts.s_min_batch_time = sbi->s_min_batch_time; 5284 old_opts.s_max_batch_time = sbi->s_max_batch_time; 5285 #ifdef CONFIG_QUOTA 5286 old_opts.s_jquota_fmt = sbi->s_jquota_fmt; 5287 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5288 if (sbi->s_qf_names[i]) { 5289 char *qf_name = get_qf_name(sb, sbi, i); 5290 5291 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL); 5292 if (!old_opts.s_qf_names[i]) { 5293 for (j = 0; j < i; j++) 5294 kfree(old_opts.s_qf_names[j]); 5295 kfree(orig_data); 5296 return -ENOMEM; 5297 } 5298 } else 5299 old_opts.s_qf_names[i] = NULL; 5300 #endif 5301 if (sbi->s_journal && sbi->s_journal->j_task->io_context) 5302 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio; 5303 5304 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) { 5305 err = -EINVAL; 5306 goto restore_opts; 5307 } 5308 5309 ext4_clamp_want_extra_isize(sb); 5310 5311 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^ 5312 test_opt(sb, JOURNAL_CHECKSUM)) { 5313 ext4_msg(sb, KERN_ERR, "changing journal_checksum " 5314 "during remount not supported; ignoring"); 5315 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM; 5316 } 5317 5318 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) { 5319 if (test_opt2(sb, EXPLICIT_DELALLOC)) { 5320 ext4_msg(sb, KERN_ERR, "can't mount with " 5321 "both data=journal and delalloc"); 5322 err = -EINVAL; 5323 goto restore_opts; 5324 } 5325 if (test_opt(sb, DIOREAD_NOLOCK)) { 5326 ext4_msg(sb, KERN_ERR, "can't mount with " 5327 "both data=journal and dioread_nolock"); 5328 err = -EINVAL; 5329 goto restore_opts; 5330 } 5331 if (test_opt(sb, DAX)) { 5332 ext4_msg(sb, KERN_ERR, "can't mount with " 5333 "both data=journal and dax"); 5334 err = -EINVAL; 5335 goto restore_opts; 5336 } 5337 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) { 5338 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) { 5339 ext4_msg(sb, KERN_ERR, "can't mount with " 5340 "journal_async_commit in data=ordered mode"); 5341 err = -EINVAL; 5342 goto restore_opts; 5343 } 5344 } 5345 5346 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) { 5347 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount"); 5348 err = -EINVAL; 5349 goto restore_opts; 5350 } 5351 5352 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) { 5353 ext4_msg(sb, KERN_WARNING, "warning: refusing change of " 5354 "dax flag with busy inodes while remounting"); 5355 sbi->s_mount_opt ^= EXT4_MOUNT_DAX; 5356 } 5357 5358 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) 5359 ext4_abort(sb, "Abort forced by user"); 5360 5361 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | 5362 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); 5363 5364 es = sbi->s_es; 5365 5366 if (sbi->s_journal) { 5367 ext4_init_journal_params(sb, sbi->s_journal); 5368 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 5369 } 5370 5371 if (*flags & SB_LAZYTIME) 5372 sb->s_flags |= SB_LAZYTIME; 5373 5374 if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) { 5375 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) { 5376 err = -EROFS; 5377 goto restore_opts; 5378 } 5379 5380 if (*flags & SB_RDONLY) { 5381 err = sync_filesystem(sb); 5382 if (err < 0) 5383 goto restore_opts; 5384 err = dquot_suspend(sb, -1); 5385 if (err < 0) 5386 goto restore_opts; 5387 5388 /* 5389 * First of all, the unconditional stuff we have to do 5390 * to disable replay of the journal when we next remount 5391 */ 5392 sb->s_flags |= SB_RDONLY; 5393 5394 /* 5395 * OK, test if we are remounting a valid rw partition 5396 * readonly, and if so set the rdonly flag and then 5397 * mark the partition as valid again. 5398 */ 5399 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) && 5400 (sbi->s_mount_state & EXT4_VALID_FS)) 5401 es->s_state = cpu_to_le16(sbi->s_mount_state); 5402 5403 if (sbi->s_journal) 5404 ext4_mark_recovery_complete(sb, es); 5405 if (sbi->s_mmp_tsk) 5406 kthread_stop(sbi->s_mmp_tsk); 5407 } else { 5408 /* Make sure we can mount this feature set readwrite */ 5409 if (ext4_has_feature_readonly(sb) || 5410 !ext4_feature_set_ok(sb, 0)) { 5411 err = -EROFS; 5412 goto restore_opts; 5413 } 5414 /* 5415 * Make sure the group descriptor checksums 5416 * are sane. If they aren't, refuse to remount r/w. 5417 */ 5418 for (g = 0; g < sbi->s_groups_count; g++) { 5419 struct ext4_group_desc *gdp = 5420 ext4_get_group_desc(sb, g, NULL); 5421 5422 if (!ext4_group_desc_csum_verify(sb, g, gdp)) { 5423 ext4_msg(sb, KERN_ERR, 5424 "ext4_remount: Checksum for group %u failed (%u!=%u)", 5425 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)), 5426 le16_to_cpu(gdp->bg_checksum)); 5427 err = -EFSBADCRC; 5428 goto restore_opts; 5429 } 5430 } 5431 5432 /* 5433 * If we have an unprocessed orphan list hanging 5434 * around from a previously readonly bdev mount, 5435 * require a full umount/remount for now. 5436 */ 5437 if (es->s_last_orphan) { 5438 ext4_msg(sb, KERN_WARNING, "Couldn't " 5439 "remount RDWR because of unprocessed " 5440 "orphan inode list. Please " 5441 "umount/remount instead"); 5442 err = -EINVAL; 5443 goto restore_opts; 5444 } 5445 5446 /* 5447 * Mounting a RDONLY partition read-write, so reread 5448 * and store the current valid flag. (It may have 5449 * been changed by e2fsck since we originally mounted 5450 * the partition.) 5451 */ 5452 if (sbi->s_journal) 5453 ext4_clear_journal_err(sb, es); 5454 sbi->s_mount_state = le16_to_cpu(es->s_state); 5455 5456 err = ext4_setup_super(sb, es, 0); 5457 if (err) 5458 goto restore_opts; 5459 5460 sb->s_flags &= ~SB_RDONLY; 5461 if (ext4_has_feature_mmp(sb)) 5462 if (ext4_multi_mount_protect(sb, 5463 le64_to_cpu(es->s_mmp_block))) { 5464 err = -EROFS; 5465 goto restore_opts; 5466 } 5467 enable_quota = 1; 5468 } 5469 } 5470 5471 /* 5472 * Reinitialize lazy itable initialization thread based on 5473 * current settings 5474 */ 5475 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE)) 5476 ext4_unregister_li_request(sb); 5477 else { 5478 ext4_group_t first_not_zeroed; 5479 first_not_zeroed = ext4_has_uninit_itable(sb); 5480 ext4_register_li_request(sb, first_not_zeroed); 5481 } 5482 5483 ext4_setup_system_zone(sb); 5484 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) { 5485 err = ext4_commit_super(sb, 1); 5486 if (err) 5487 goto restore_opts; 5488 } 5489 5490 #ifdef CONFIG_QUOTA 5491 /* Release old quota file names */ 5492 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5493 kfree(old_opts.s_qf_names[i]); 5494 if (enable_quota) { 5495 if (sb_any_quota_suspended(sb)) 5496 dquot_resume(sb, -1); 5497 else if (ext4_has_feature_quota(sb)) { 5498 err = ext4_enable_quotas(sb); 5499 if (err) 5500 goto restore_opts; 5501 } 5502 } 5503 #endif 5504 5505 *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME); 5506 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data); 5507 kfree(orig_data); 5508 return 0; 5509 5510 restore_opts: 5511 sb->s_flags = old_sb_flags; 5512 sbi->s_mount_opt = old_opts.s_mount_opt; 5513 sbi->s_mount_opt2 = old_opts.s_mount_opt2; 5514 sbi->s_resuid = old_opts.s_resuid; 5515 sbi->s_resgid = old_opts.s_resgid; 5516 sbi->s_commit_interval = old_opts.s_commit_interval; 5517 sbi->s_min_batch_time = old_opts.s_min_batch_time; 5518 sbi->s_max_batch_time = old_opts.s_max_batch_time; 5519 #ifdef CONFIG_QUOTA 5520 sbi->s_jquota_fmt = old_opts.s_jquota_fmt; 5521 for (i = 0; i < EXT4_MAXQUOTAS; i++) { 5522 to_free[i] = get_qf_name(sb, sbi, i); 5523 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]); 5524 } 5525 synchronize_rcu(); 5526 for (i = 0; i < EXT4_MAXQUOTAS; i++) 5527 kfree(to_free[i]); 5528 #endif 5529 kfree(orig_data); 5530 return err; 5531 } 5532 5533 #ifdef CONFIG_QUOTA 5534 static int ext4_statfs_project(struct super_block *sb, 5535 kprojid_t projid, struct kstatfs *buf) 5536 { 5537 struct kqid qid; 5538 struct dquot *dquot; 5539 u64 limit; 5540 u64 curblock; 5541 5542 qid = make_kqid_projid(projid); 5543 dquot = dqget(sb, qid); 5544 if (IS_ERR(dquot)) 5545 return PTR_ERR(dquot); 5546 spin_lock(&dquot->dq_dqb_lock); 5547 5548 limit = (dquot->dq_dqb.dqb_bsoftlimit ? 5549 dquot->dq_dqb.dqb_bsoftlimit : 5550 dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits; 5551 if (limit && buf->f_blocks > limit) { 5552 curblock = (dquot->dq_dqb.dqb_curspace + 5553 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits; 5554 buf->f_blocks = limit; 5555 buf->f_bfree = buf->f_bavail = 5556 (buf->f_blocks > curblock) ? 5557 (buf->f_blocks - curblock) : 0; 5558 } 5559 5560 limit = dquot->dq_dqb.dqb_isoftlimit ? 5561 dquot->dq_dqb.dqb_isoftlimit : 5562 dquot->dq_dqb.dqb_ihardlimit; 5563 if (limit && buf->f_files > limit) { 5564 buf->f_files = limit; 5565 buf->f_ffree = 5566 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ? 5567 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0; 5568 } 5569 5570 spin_unlock(&dquot->dq_dqb_lock); 5571 dqput(dquot); 5572 return 0; 5573 } 5574 #endif 5575 5576 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf) 5577 { 5578 struct super_block *sb = dentry->d_sb; 5579 struct ext4_sb_info *sbi = EXT4_SB(sb); 5580 struct ext4_super_block *es = sbi->s_es; 5581 ext4_fsblk_t overhead = 0, resv_blocks; 5582 u64 fsid; 5583 s64 bfree; 5584 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters)); 5585 5586 if (!test_opt(sb, MINIX_DF)) 5587 overhead = sbi->s_overhead; 5588 5589 buf->f_type = EXT4_SUPER_MAGIC; 5590 buf->f_bsize = sb->s_blocksize; 5591 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead); 5592 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) - 5593 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter); 5594 /* prevent underflow in case that few free space is available */ 5595 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0)); 5596 buf->f_bavail = buf->f_bfree - 5597 (ext4_r_blocks_count(es) + resv_blocks); 5598 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks)) 5599 buf->f_bavail = 0; 5600 buf->f_files = le32_to_cpu(es->s_inodes_count); 5601 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter); 5602 buf->f_namelen = EXT4_NAME_LEN; 5603 fsid = le64_to_cpup((void *)es->s_uuid) ^ 5604 le64_to_cpup((void *)es->s_uuid + sizeof(u64)); 5605 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL; 5606 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL; 5607 5608 #ifdef CONFIG_QUOTA 5609 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) && 5610 sb_has_quota_limits_enabled(sb, PRJQUOTA)) 5611 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf); 5612 #endif 5613 return 0; 5614 } 5615 5616 5617 #ifdef CONFIG_QUOTA 5618 5619 /* 5620 * Helper functions so that transaction is started before we acquire dqio_sem 5621 * to keep correct lock ordering of transaction > dqio_sem 5622 */ 5623 static inline struct inode *dquot_to_inode(struct dquot *dquot) 5624 { 5625 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type]; 5626 } 5627 5628 static int ext4_write_dquot(struct dquot *dquot) 5629 { 5630 int ret, err; 5631 handle_t *handle; 5632 struct inode *inode; 5633 5634 inode = dquot_to_inode(dquot); 5635 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 5636 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb)); 5637 if (IS_ERR(handle)) 5638 return PTR_ERR(handle); 5639 ret = dquot_commit(dquot); 5640 err = ext4_journal_stop(handle); 5641 if (!ret) 5642 ret = err; 5643 return ret; 5644 } 5645 5646 static int ext4_acquire_dquot(struct dquot *dquot) 5647 { 5648 int ret, err; 5649 handle_t *handle; 5650 5651 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 5652 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb)); 5653 if (IS_ERR(handle)) 5654 return PTR_ERR(handle); 5655 ret = dquot_acquire(dquot); 5656 err = ext4_journal_stop(handle); 5657 if (!ret) 5658 ret = err; 5659 return ret; 5660 } 5661 5662 static int ext4_release_dquot(struct dquot *dquot) 5663 { 5664 int ret, err; 5665 handle_t *handle; 5666 5667 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA, 5668 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb)); 5669 if (IS_ERR(handle)) { 5670 /* Release dquot anyway to avoid endless cycle in dqput() */ 5671 dquot_release(dquot); 5672 return PTR_ERR(handle); 5673 } 5674 ret = dquot_release(dquot); 5675 err = ext4_journal_stop(handle); 5676 if (!ret) 5677 ret = err; 5678 return ret; 5679 } 5680 5681 static int ext4_mark_dquot_dirty(struct dquot *dquot) 5682 { 5683 struct super_block *sb = dquot->dq_sb; 5684 struct ext4_sb_info *sbi = EXT4_SB(sb); 5685 5686 /* Are we journaling quotas? */ 5687 if (ext4_has_feature_quota(sb) || 5688 sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) { 5689 dquot_mark_dquot_dirty(dquot); 5690 return ext4_write_dquot(dquot); 5691 } else { 5692 return dquot_mark_dquot_dirty(dquot); 5693 } 5694 } 5695 5696 static int ext4_write_info(struct super_block *sb, int type) 5697 { 5698 int ret, err; 5699 handle_t *handle; 5700 5701 /* Data block + inode block */ 5702 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2); 5703 if (IS_ERR(handle)) 5704 return PTR_ERR(handle); 5705 ret = dquot_commit_info(sb, type); 5706 err = ext4_journal_stop(handle); 5707 if (!ret) 5708 ret = err; 5709 return ret; 5710 } 5711 5712 /* 5713 * Turn on quotas during mount time - we need to find 5714 * the quota file and such... 5715 */ 5716 static int ext4_quota_on_mount(struct super_block *sb, int type) 5717 { 5718 return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type), 5719 EXT4_SB(sb)->s_jquota_fmt, type); 5720 } 5721 5722 static void lockdep_set_quota_inode(struct inode *inode, int subclass) 5723 { 5724 struct ext4_inode_info *ei = EXT4_I(inode); 5725 5726 /* The first argument of lockdep_set_subclass has to be 5727 * *exactly* the same as the argument to init_rwsem() --- in 5728 * this case, in init_once() --- or lockdep gets unhappy 5729 * because the name of the lock is set using the 5730 * stringification of the argument to init_rwsem(). 5731 */ 5732 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */ 5733 lockdep_set_subclass(&ei->i_data_sem, subclass); 5734 } 5735 5736 /* 5737 * Standard function to be called on quota_on 5738 */ 5739 static int ext4_quota_on(struct super_block *sb, int type, int format_id, 5740 const struct path *path) 5741 { 5742 int err; 5743 5744 if (!test_opt(sb, QUOTA)) 5745 return -EINVAL; 5746 5747 /* Quotafile not on the same filesystem? */ 5748 if (path->dentry->d_sb != sb) 5749 return -EXDEV; 5750 /* Journaling quota? */ 5751 if (EXT4_SB(sb)->s_qf_names[type]) { 5752 /* Quotafile not in fs root? */ 5753 if (path->dentry->d_parent != sb->s_root) 5754 ext4_msg(sb, KERN_WARNING, 5755 "Quota file not on filesystem root. " 5756 "Journaled quota will not work"); 5757 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY; 5758 } else { 5759 /* 5760 * Clear the flag just in case mount options changed since 5761 * last time. 5762 */ 5763 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY; 5764 } 5765 5766 /* 5767 * When we journal data on quota file, we have to flush journal to see 5768 * all updates to the file when we bypass pagecache... 5769 */ 5770 if (EXT4_SB(sb)->s_journal && 5771 ext4_should_journal_data(d_inode(path->dentry))) { 5772 /* 5773 * We don't need to lock updates but journal_flush() could 5774 * otherwise be livelocked... 5775 */ 5776 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 5777 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 5778 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 5779 if (err) 5780 return err; 5781 } 5782 5783 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA); 5784 err = dquot_quota_on(sb, type, format_id, path); 5785 if (err) { 5786 lockdep_set_quota_inode(path->dentry->d_inode, 5787 I_DATA_SEM_NORMAL); 5788 } else { 5789 struct inode *inode = d_inode(path->dentry); 5790 handle_t *handle; 5791 5792 /* 5793 * Set inode flags to prevent userspace from messing with quota 5794 * files. If this fails, we return success anyway since quotas 5795 * are already enabled and this is not a hard failure. 5796 */ 5797 inode_lock(inode); 5798 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 5799 if (IS_ERR(handle)) 5800 goto unlock_inode; 5801 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL; 5802 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE, 5803 S_NOATIME | S_IMMUTABLE); 5804 ext4_mark_inode_dirty(handle, inode); 5805 ext4_journal_stop(handle); 5806 unlock_inode: 5807 inode_unlock(inode); 5808 } 5809 return err; 5810 } 5811 5812 static int ext4_quota_enable(struct super_block *sb, int type, int format_id, 5813 unsigned int flags) 5814 { 5815 int err; 5816 struct inode *qf_inode; 5817 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 5818 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 5819 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 5820 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 5821 }; 5822 5823 BUG_ON(!ext4_has_feature_quota(sb)); 5824 5825 if (!qf_inums[type]) 5826 return -EPERM; 5827 5828 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL); 5829 if (IS_ERR(qf_inode)) { 5830 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]); 5831 return PTR_ERR(qf_inode); 5832 } 5833 5834 /* Don't account quota for quota files to avoid recursion */ 5835 qf_inode->i_flags |= S_NOQUOTA; 5836 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA); 5837 err = dquot_load_quota_inode(qf_inode, type, format_id, flags); 5838 if (err) 5839 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL); 5840 iput(qf_inode); 5841 5842 return err; 5843 } 5844 5845 /* Enable usage tracking for all quota types. */ 5846 static int ext4_enable_quotas(struct super_block *sb) 5847 { 5848 int type, err = 0; 5849 unsigned long qf_inums[EXT4_MAXQUOTAS] = { 5850 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum), 5851 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum), 5852 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum) 5853 }; 5854 bool quota_mopt[EXT4_MAXQUOTAS] = { 5855 test_opt(sb, USRQUOTA), 5856 test_opt(sb, GRPQUOTA), 5857 test_opt(sb, PRJQUOTA), 5858 }; 5859 5860 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; 5861 for (type = 0; type < EXT4_MAXQUOTAS; type++) { 5862 if (qf_inums[type]) { 5863 err = ext4_quota_enable(sb, type, QFMT_VFS_V1, 5864 DQUOT_USAGE_ENABLED | 5865 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); 5866 if (err) { 5867 ext4_warning(sb, 5868 "Failed to enable quota tracking " 5869 "(type=%d, err=%d). Please run " 5870 "e2fsck to fix.", type, err); 5871 for (type--; type >= 0; type--) 5872 dquot_quota_off(sb, type); 5873 5874 return err; 5875 } 5876 } 5877 } 5878 return 0; 5879 } 5880 5881 static int ext4_quota_off(struct super_block *sb, int type) 5882 { 5883 struct inode *inode = sb_dqopt(sb)->files[type]; 5884 handle_t *handle; 5885 int err; 5886 5887 /* Force all delayed allocation blocks to be allocated. 5888 * Caller already holds s_umount sem */ 5889 if (test_opt(sb, DELALLOC)) 5890 sync_filesystem(sb); 5891 5892 if (!inode || !igrab(inode)) 5893 goto out; 5894 5895 err = dquot_quota_off(sb, type); 5896 if (err || ext4_has_feature_quota(sb)) 5897 goto out_put; 5898 5899 inode_lock(inode); 5900 /* 5901 * Update modification times of quota files when userspace can 5902 * start looking at them. If we fail, we return success anyway since 5903 * this is not a hard failure and quotas are already disabled. 5904 */ 5905 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1); 5906 if (IS_ERR(handle)) 5907 goto out_unlock; 5908 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL); 5909 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE); 5910 inode->i_mtime = inode->i_ctime = current_time(inode); 5911 ext4_mark_inode_dirty(handle, inode); 5912 ext4_journal_stop(handle); 5913 out_unlock: 5914 inode_unlock(inode); 5915 out_put: 5916 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL); 5917 iput(inode); 5918 return err; 5919 out: 5920 return dquot_quota_off(sb, type); 5921 } 5922 5923 /* Read data from quotafile - avoid pagecache and such because we cannot afford 5924 * acquiring the locks... As quota files are never truncated and quota code 5925 * itself serializes the operations (and no one else should touch the files) 5926 * we don't have to be afraid of races */ 5927 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data, 5928 size_t len, loff_t off) 5929 { 5930 struct inode *inode = sb_dqopt(sb)->files[type]; 5931 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 5932 int offset = off & (sb->s_blocksize - 1); 5933 int tocopy; 5934 size_t toread; 5935 struct buffer_head *bh; 5936 loff_t i_size = i_size_read(inode); 5937 5938 if (off > i_size) 5939 return 0; 5940 if (off+len > i_size) 5941 len = i_size-off; 5942 toread = len; 5943 while (toread > 0) { 5944 tocopy = sb->s_blocksize - offset < toread ? 5945 sb->s_blocksize - offset : toread; 5946 bh = ext4_bread(NULL, inode, blk, 0); 5947 if (IS_ERR(bh)) 5948 return PTR_ERR(bh); 5949 if (!bh) /* A hole? */ 5950 memset(data, 0, tocopy); 5951 else 5952 memcpy(data, bh->b_data+offset, tocopy); 5953 brelse(bh); 5954 offset = 0; 5955 toread -= tocopy; 5956 data += tocopy; 5957 blk++; 5958 } 5959 return len; 5960 } 5961 5962 /* Write to quotafile (we know the transaction is already started and has 5963 * enough credits) */ 5964 static ssize_t ext4_quota_write(struct super_block *sb, int type, 5965 const char *data, size_t len, loff_t off) 5966 { 5967 struct inode *inode = sb_dqopt(sb)->files[type]; 5968 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb); 5969 int err, offset = off & (sb->s_blocksize - 1); 5970 int retries = 0; 5971 struct buffer_head *bh; 5972 handle_t *handle = journal_current_handle(); 5973 5974 if (EXT4_SB(sb)->s_journal && !handle) { 5975 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 5976 " cancelled because transaction is not started", 5977 (unsigned long long)off, (unsigned long long)len); 5978 return -EIO; 5979 } 5980 /* 5981 * Since we account only one data block in transaction credits, 5982 * then it is impossible to cross a block boundary. 5983 */ 5984 if (sb->s_blocksize - offset < len) { 5985 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)" 5986 " cancelled because not block aligned", 5987 (unsigned long long)off, (unsigned long long)len); 5988 return -EIO; 5989 } 5990 5991 do { 5992 bh = ext4_bread(handle, inode, blk, 5993 EXT4_GET_BLOCKS_CREATE | 5994 EXT4_GET_BLOCKS_METADATA_NOFAIL); 5995 } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) && 5996 ext4_should_retry_alloc(inode->i_sb, &retries)); 5997 if (IS_ERR(bh)) 5998 return PTR_ERR(bh); 5999 if (!bh) 6000 goto out; 6001 BUFFER_TRACE(bh, "get write access"); 6002 err = ext4_journal_get_write_access(handle, bh); 6003 if (err) { 6004 brelse(bh); 6005 return err; 6006 } 6007 lock_buffer(bh); 6008 memcpy(bh->b_data+offset, data, len); 6009 flush_dcache_page(bh->b_page); 6010 unlock_buffer(bh); 6011 err = ext4_handle_dirty_metadata(handle, NULL, bh); 6012 brelse(bh); 6013 out: 6014 if (inode->i_size < off + len) { 6015 i_size_write(inode, off + len); 6016 EXT4_I(inode)->i_disksize = inode->i_size; 6017 ext4_mark_inode_dirty(handle, inode); 6018 } 6019 return len; 6020 } 6021 #endif 6022 6023 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags, 6024 const char *dev_name, void *data) 6025 { 6026 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super); 6027 } 6028 6029 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2) 6030 static inline void register_as_ext2(void) 6031 { 6032 int err = register_filesystem(&ext2_fs_type); 6033 if (err) 6034 printk(KERN_WARNING 6035 "EXT4-fs: Unable to register as ext2 (%d)\n", err); 6036 } 6037 6038 static inline void unregister_as_ext2(void) 6039 { 6040 unregister_filesystem(&ext2_fs_type); 6041 } 6042 6043 static inline int ext2_feature_set_ok(struct super_block *sb) 6044 { 6045 if (ext4_has_unknown_ext2_incompat_features(sb)) 6046 return 0; 6047 if (sb_rdonly(sb)) 6048 return 1; 6049 if (ext4_has_unknown_ext2_ro_compat_features(sb)) 6050 return 0; 6051 return 1; 6052 } 6053 #else 6054 static inline void register_as_ext2(void) { } 6055 static inline void unregister_as_ext2(void) { } 6056 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; } 6057 #endif 6058 6059 static inline void register_as_ext3(void) 6060 { 6061 int err = register_filesystem(&ext3_fs_type); 6062 if (err) 6063 printk(KERN_WARNING 6064 "EXT4-fs: Unable to register as ext3 (%d)\n", err); 6065 } 6066 6067 static inline void unregister_as_ext3(void) 6068 { 6069 unregister_filesystem(&ext3_fs_type); 6070 } 6071 6072 static inline int ext3_feature_set_ok(struct super_block *sb) 6073 { 6074 if (ext4_has_unknown_ext3_incompat_features(sb)) 6075 return 0; 6076 if (!ext4_has_feature_journal(sb)) 6077 return 0; 6078 if (sb_rdonly(sb)) 6079 return 1; 6080 if (ext4_has_unknown_ext3_ro_compat_features(sb)) 6081 return 0; 6082 return 1; 6083 } 6084 6085 static struct file_system_type ext4_fs_type = { 6086 .owner = THIS_MODULE, 6087 .name = "ext4", 6088 .mount = ext4_mount, 6089 .kill_sb = kill_block_super, 6090 .fs_flags = FS_REQUIRES_DEV, 6091 }; 6092 MODULE_ALIAS_FS("ext4"); 6093 6094 /* Shared across all ext4 file systems */ 6095 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 6096 6097 static int __init ext4_init_fs(void) 6098 { 6099 int i, err; 6100 6101 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64); 6102 ext4_li_info = NULL; 6103 mutex_init(&ext4_li_mtx); 6104 6105 /* Build-time check for flags consistency */ 6106 ext4_check_flag_values(); 6107 6108 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) 6109 init_waitqueue_head(&ext4__ioend_wq[i]); 6110 6111 err = ext4_init_es(); 6112 if (err) 6113 return err; 6114 6115 err = ext4_init_pending(); 6116 if (err) 6117 goto out7; 6118 6119 err = ext4_init_post_read_processing(); 6120 if (err) 6121 goto out6; 6122 6123 err = ext4_init_pageio(); 6124 if (err) 6125 goto out5; 6126 6127 err = ext4_init_system_zone(); 6128 if (err) 6129 goto out4; 6130 6131 err = ext4_init_sysfs(); 6132 if (err) 6133 goto out3; 6134 6135 err = ext4_init_mballoc(); 6136 if (err) 6137 goto out2; 6138 err = init_inodecache(); 6139 if (err) 6140 goto out1; 6141 register_as_ext3(); 6142 register_as_ext2(); 6143 err = register_filesystem(&ext4_fs_type); 6144 if (err) 6145 goto out; 6146 6147 return 0; 6148 out: 6149 unregister_as_ext2(); 6150 unregister_as_ext3(); 6151 destroy_inodecache(); 6152 out1: 6153 ext4_exit_mballoc(); 6154 out2: 6155 ext4_exit_sysfs(); 6156 out3: 6157 ext4_exit_system_zone(); 6158 out4: 6159 ext4_exit_pageio(); 6160 out5: 6161 ext4_exit_post_read_processing(); 6162 out6: 6163 ext4_exit_pending(); 6164 out7: 6165 ext4_exit_es(); 6166 6167 return err; 6168 } 6169 6170 static void __exit ext4_exit_fs(void) 6171 { 6172 ext4_destroy_lazyinit_thread(); 6173 unregister_as_ext2(); 6174 unregister_as_ext3(); 6175 unregister_filesystem(&ext4_fs_type); 6176 destroy_inodecache(); 6177 ext4_exit_mballoc(); 6178 ext4_exit_sysfs(); 6179 ext4_exit_system_zone(); 6180 ext4_exit_pageio(); 6181 ext4_exit_post_read_processing(); 6182 ext4_exit_es(); 6183 ext4_exit_pending(); 6184 } 6185 6186 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 6187 MODULE_DESCRIPTION("Fourth Extended Filesystem"); 6188 MODULE_LICENSE("GPL"); 6189 MODULE_SOFTDEP("pre: crc32c"); 6190 module_init(ext4_init_fs) 6191 module_exit(ext4_exit_fs) 6192