1 /* 2 * linux/include/linux/jbd2.h 3 * 4 * Written by Stephen C. Tweedie <[email protected]> 5 * 6 * Copyright 1998-2000 Red Hat, Inc --- All Rights Reserved 7 * 8 * This file is part of the Linux kernel and is made available under 9 * the terms of the GNU General Public License, version 2, or at your 10 * option, any later version, incorporated herein by reference. 11 * 12 * Definitions for transaction data structures for the buffer cache 13 * filesystem journaling support. 14 */ 15 16 #ifndef _LINUX_JBD2_H 17 #define _LINUX_JBD2_H 18 19 /* Allow this file to be included directly into e2fsprogs */ 20 #ifndef __KERNEL__ 21 #include "jfs_compat.h" 22 #define JBD2_DEBUG 23 #define jfs_debug jbd_debug 24 #else 25 26 #include <linux/types.h> 27 #include <linux/buffer_head.h> 28 #include <linux/journal-head.h> 29 #include <linux/stddef.h> 30 #include <linux/bit_spinlock.h> 31 #include <linux/mutex.h> 32 #include <linux/timer.h> 33 #endif 34 35 #define journal_oom_retry 1 36 37 /* 38 * Define JBD2_PARANIOD_IOFAIL to cause a kernel BUG() if ext4 finds 39 * certain classes of error which can occur due to failed IOs. Under 40 * normal use we want ext4 to continue after such errors, because 41 * hardware _can_ fail, but for debugging purposes when running tests on 42 * known-good hardware we may want to trap these errors. 43 */ 44 #undef JBD2_PARANOID_IOFAIL 45 46 /* 47 * The default maximum commit age, in seconds. 48 */ 49 #define JBD2_DEFAULT_MAX_COMMIT_AGE 5 50 51 #ifdef CONFIG_JBD2_DEBUG 52 /* 53 * Define JBD2_EXPENSIVE_CHECKING to enable more expensive internal 54 * consistency checks. By default we don't do this unless 55 * CONFIG_JBD2_DEBUG is on. 56 */ 57 #define JBD2_EXPENSIVE_CHECKING 58 extern u8 jbd2_journal_enable_debug; 59 60 #define jbd_debug(n, f, a...) \ 61 do { \ 62 if ((n) <= jbd2_journal_enable_debug) { \ 63 printk (KERN_DEBUG "(%s, %d): %s: ", \ 64 __FILE__, __LINE__, __func__); \ 65 printk (f, ## a); \ 66 } \ 67 } while (0) 68 #else 69 #define jbd_debug(f, a...) /**/ 70 #endif 71 72 static inline void *jbd2_alloc(size_t size, gfp_t flags) 73 { 74 return (void *)__get_free_pages(flags, get_order(size)); 75 } 76 77 static inline void jbd2_free(void *ptr, size_t size) 78 { 79 free_pages((unsigned long)ptr, get_order(size)); 80 }; 81 82 #define JBD2_MIN_JOURNAL_BLOCKS 1024 83 84 #ifdef __KERNEL__ 85 86 /** 87 * typedef handle_t - The handle_t type represents a single atomic update being performed by some process. 88 * 89 * All filesystem modifications made by the process go 90 * through this handle. Recursive operations (such as quota operations) 91 * are gathered into a single update. 92 * 93 * The buffer credits field is used to account for journaled buffers 94 * being modified by the running process. To ensure that there is 95 * enough log space for all outstanding operations, we need to limit the 96 * number of outstanding buffers possible at any time. When the 97 * operation completes, any buffer credits not used are credited back to 98 * the transaction, so that at all times we know how many buffers the 99 * outstanding updates on a transaction might possibly touch. 100 * 101 * This is an opaque datatype. 102 **/ 103 typedef struct handle_s handle_t; /* Atomic operation type */ 104 105 106 /** 107 * typedef journal_t - The journal_t maintains all of the journaling state information for a single filesystem. 108 * 109 * journal_t is linked to from the fs superblock structure. 110 * 111 * We use the journal_t to keep track of all outstanding transaction 112 * activity on the filesystem, and to manage the state of the log 113 * writing process. 114 * 115 * This is an opaque datatype. 116 **/ 117 typedef struct journal_s journal_t; /* Journal control structure */ 118 #endif 119 120 /* 121 * Internal structures used by the logging mechanism: 122 */ 123 124 #define JBD2_MAGIC_NUMBER 0xc03b3998U /* The first 4 bytes of /dev/random! */ 125 126 /* 127 * On-disk structures 128 */ 129 130 /* 131 * Descriptor block types: 132 */ 133 134 #define JBD2_DESCRIPTOR_BLOCK 1 135 #define JBD2_COMMIT_BLOCK 2 136 #define JBD2_SUPERBLOCK_V1 3 137 #define JBD2_SUPERBLOCK_V2 4 138 #define JBD2_REVOKE_BLOCK 5 139 140 /* 141 * Standard header for all descriptor blocks: 142 */ 143 typedef struct journal_header_s 144 { 145 __be32 h_magic; 146 __be32 h_blocktype; 147 __be32 h_sequence; 148 } journal_header_t; 149 150 /* 151 * Checksum types. 152 */ 153 #define JBD2_CRC32_CHKSUM 1 154 #define JBD2_MD5_CHKSUM 2 155 #define JBD2_SHA1_CHKSUM 3 156 157 #define JBD2_CRC32_CHKSUM_SIZE 4 158 159 #define JBD2_CHECKSUM_BYTES (32 / sizeof(u32)) 160 /* 161 * Commit block header for storing transactional checksums: 162 */ 163 struct commit_header { 164 __be32 h_magic; 165 __be32 h_blocktype; 166 __be32 h_sequence; 167 unsigned char h_chksum_type; 168 unsigned char h_chksum_size; 169 unsigned char h_padding[2]; 170 __be32 h_chksum[JBD2_CHECKSUM_BYTES]; 171 __be64 h_commit_sec; 172 __be32 h_commit_nsec; 173 }; 174 175 /* 176 * The block tag: used to describe a single buffer in the journal. 177 * t_blocknr_high is only used if INCOMPAT_64BIT is set, so this 178 * raw struct shouldn't be used for pointer math or sizeof() - use 179 * journal_tag_bytes(journal) instead to compute this. 180 */ 181 typedef struct journal_block_tag_s 182 { 183 __be32 t_blocknr; /* The on-disk block number */ 184 __be32 t_flags; /* See below */ 185 __be32 t_blocknr_high; /* most-significant high 32bits. */ 186 } journal_block_tag_t; 187 188 #define JBD2_TAG_SIZE32 (offsetof(journal_block_tag_t, t_blocknr_high)) 189 #define JBD2_TAG_SIZE64 (sizeof(journal_block_tag_t)) 190 191 /* 192 * The revoke descriptor: used on disk to describe a series of blocks to 193 * be revoked from the log 194 */ 195 typedef struct jbd2_journal_revoke_header_s 196 { 197 journal_header_t r_header; 198 __be32 r_count; /* Count of bytes used in the block */ 199 } jbd2_journal_revoke_header_t; 200 201 202 /* Definitions for the journal tag flags word: */ 203 #define JBD2_FLAG_ESCAPE 1 /* on-disk block is escaped */ 204 #define JBD2_FLAG_SAME_UUID 2 /* block has same uuid as previous */ 205 #define JBD2_FLAG_DELETED 4 /* block deleted by this transaction */ 206 #define JBD2_FLAG_LAST_TAG 8 /* last tag in this descriptor block */ 207 208 209 /* 210 * The journal superblock. All fields are in big-endian byte order. 211 */ 212 typedef struct journal_superblock_s 213 { 214 /* 0x0000 */ 215 journal_header_t s_header; 216 217 /* 0x000C */ 218 /* Static information describing the journal */ 219 __be32 s_blocksize; /* journal device blocksize */ 220 __be32 s_maxlen; /* total blocks in journal file */ 221 __be32 s_first; /* first block of log information */ 222 223 /* 0x0018 */ 224 /* Dynamic information describing the current state of the log */ 225 __be32 s_sequence; /* first commit ID expected in log */ 226 __be32 s_start; /* blocknr of start of log */ 227 228 /* 0x0020 */ 229 /* Error value, as set by jbd2_journal_abort(). */ 230 __be32 s_errno; 231 232 /* 0x0024 */ 233 /* Remaining fields are only valid in a version-2 superblock */ 234 __be32 s_feature_compat; /* compatible feature set */ 235 __be32 s_feature_incompat; /* incompatible feature set */ 236 __be32 s_feature_ro_compat; /* readonly-compatible feature set */ 237 /* 0x0030 */ 238 __u8 s_uuid[16]; /* 128-bit uuid for journal */ 239 240 /* 0x0040 */ 241 __be32 s_nr_users; /* Nr of filesystems sharing log */ 242 243 __be32 s_dynsuper; /* Blocknr of dynamic superblock copy*/ 244 245 /* 0x0048 */ 246 __be32 s_max_transaction; /* Limit of journal blocks per trans.*/ 247 __be32 s_max_trans_data; /* Limit of data blocks per trans. */ 248 249 /* 0x0050 */ 250 __u32 s_padding[44]; 251 252 /* 0x0100 */ 253 __u8 s_users[16*48]; /* ids of all fs'es sharing the log */ 254 /* 0x0400 */ 255 } journal_superblock_t; 256 257 #define JBD2_HAS_COMPAT_FEATURE(j,mask) \ 258 ((j)->j_format_version >= 2 && \ 259 ((j)->j_superblock->s_feature_compat & cpu_to_be32((mask)))) 260 #define JBD2_HAS_RO_COMPAT_FEATURE(j,mask) \ 261 ((j)->j_format_version >= 2 && \ 262 ((j)->j_superblock->s_feature_ro_compat & cpu_to_be32((mask)))) 263 #define JBD2_HAS_INCOMPAT_FEATURE(j,mask) \ 264 ((j)->j_format_version >= 2 && \ 265 ((j)->j_superblock->s_feature_incompat & cpu_to_be32((mask)))) 266 267 #define JBD2_FEATURE_COMPAT_CHECKSUM 0x00000001 268 269 #define JBD2_FEATURE_INCOMPAT_REVOKE 0x00000001 270 #define JBD2_FEATURE_INCOMPAT_64BIT 0x00000002 271 #define JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT 0x00000004 272 273 /* Features known to this kernel version: */ 274 #define JBD2_KNOWN_COMPAT_FEATURES JBD2_FEATURE_COMPAT_CHECKSUM 275 #define JBD2_KNOWN_ROCOMPAT_FEATURES 0 276 #define JBD2_KNOWN_INCOMPAT_FEATURES (JBD2_FEATURE_INCOMPAT_REVOKE | \ 277 JBD2_FEATURE_INCOMPAT_64BIT | \ 278 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) 279 280 #ifdef __KERNEL__ 281 282 #include <linux/fs.h> 283 #include <linux/sched.h> 284 285 #define J_ASSERT(assert) BUG_ON(!(assert)) 286 287 #if defined(CONFIG_BUFFER_DEBUG) 288 void buffer_assertion_failure(struct buffer_head *bh); 289 #define J_ASSERT_BH(bh, expr) \ 290 do { \ 291 if (!(expr)) \ 292 buffer_assertion_failure(bh); \ 293 J_ASSERT(expr); \ 294 } while (0) 295 #define J_ASSERT_JH(jh, expr) J_ASSERT_BH(jh2bh(jh), expr) 296 #else 297 #define J_ASSERT_BH(bh, expr) J_ASSERT(expr) 298 #define J_ASSERT_JH(jh, expr) J_ASSERT(expr) 299 #endif 300 301 #if defined(JBD2_PARANOID_IOFAIL) 302 #define J_EXPECT(expr, why...) J_ASSERT(expr) 303 #define J_EXPECT_BH(bh, expr, why...) J_ASSERT_BH(bh, expr) 304 #define J_EXPECT_JH(jh, expr, why...) J_ASSERT_JH(jh, expr) 305 #else 306 #define __journal_expect(expr, why...) \ 307 ({ \ 308 int val = (expr); \ 309 if (!val) { \ 310 printk(KERN_ERR \ 311 "EXT3-fs unexpected failure: %s;\n",# expr); \ 312 printk(KERN_ERR why "\n"); \ 313 } \ 314 val; \ 315 }) 316 #define J_EXPECT(expr, why...) __journal_expect(expr, ## why) 317 #define J_EXPECT_BH(bh, expr, why...) __journal_expect(expr, ## why) 318 #define J_EXPECT_JH(jh, expr, why...) __journal_expect(expr, ## why) 319 #endif 320 321 enum jbd_state_bits { 322 BH_JBD /* Has an attached ext3 journal_head */ 323 = BH_PrivateStart, 324 BH_JWrite, /* Being written to log (@@@ DEBUGGING) */ 325 BH_Freed, /* Has been freed (truncated) */ 326 BH_Revoked, /* Has been revoked from the log */ 327 BH_RevokeValid, /* Revoked flag is valid */ 328 BH_JBDDirty, /* Is dirty but journaled */ 329 BH_State, /* Pins most journal_head state */ 330 BH_JournalHead, /* Pins bh->b_private and jh->b_bh */ 331 BH_Unshadow, /* Dummy bit, for BJ_Shadow wakeup filtering */ 332 BH_JBDPrivateStart, /* First bit available for private use by FS */ 333 }; 334 335 BUFFER_FNS(JBD, jbd) 336 BUFFER_FNS(JWrite, jwrite) 337 BUFFER_FNS(JBDDirty, jbddirty) 338 TAS_BUFFER_FNS(JBDDirty, jbddirty) 339 BUFFER_FNS(Revoked, revoked) 340 TAS_BUFFER_FNS(Revoked, revoked) 341 BUFFER_FNS(RevokeValid, revokevalid) 342 TAS_BUFFER_FNS(RevokeValid, revokevalid) 343 BUFFER_FNS(Freed, freed) 344 345 static inline struct buffer_head *jh2bh(struct journal_head *jh) 346 { 347 return jh->b_bh; 348 } 349 350 static inline struct journal_head *bh2jh(struct buffer_head *bh) 351 { 352 return bh->b_private; 353 } 354 355 static inline void jbd_lock_bh_state(struct buffer_head *bh) 356 { 357 bit_spin_lock(BH_State, &bh->b_state); 358 } 359 360 static inline int jbd_trylock_bh_state(struct buffer_head *bh) 361 { 362 return bit_spin_trylock(BH_State, &bh->b_state); 363 } 364 365 static inline int jbd_is_locked_bh_state(struct buffer_head *bh) 366 { 367 return bit_spin_is_locked(BH_State, &bh->b_state); 368 } 369 370 static inline void jbd_unlock_bh_state(struct buffer_head *bh) 371 { 372 bit_spin_unlock(BH_State, &bh->b_state); 373 } 374 375 static inline void jbd_lock_bh_journal_head(struct buffer_head *bh) 376 { 377 bit_spin_lock(BH_JournalHead, &bh->b_state); 378 } 379 380 static inline void jbd_unlock_bh_journal_head(struct buffer_head *bh) 381 { 382 bit_spin_unlock(BH_JournalHead, &bh->b_state); 383 } 384 385 /* Flags in jbd_inode->i_flags */ 386 #define __JI_COMMIT_RUNNING 0 387 /* Commit of the inode data in progress. We use this flag to protect us from 388 * concurrent deletion of inode. We cannot use reference to inode for this 389 * since we cannot afford doing last iput() on behalf of kjournald 390 */ 391 #define JI_COMMIT_RUNNING (1 << __JI_COMMIT_RUNNING) 392 393 /** 394 * struct jbd_inode is the structure linking inodes in ordered mode 395 * present in a transaction so that we can sync them during commit. 396 */ 397 struct jbd2_inode { 398 /* Which transaction does this inode belong to? Either the running 399 * transaction or the committing one. [j_list_lock] */ 400 transaction_t *i_transaction; 401 402 /* Pointer to the running transaction modifying inode's data in case 403 * there is already a committing transaction touching it. [j_list_lock] */ 404 transaction_t *i_next_transaction; 405 406 /* List of inodes in the i_transaction [j_list_lock] */ 407 struct list_head i_list; 408 409 /* VFS inode this inode belongs to [constant during the lifetime 410 * of the structure] */ 411 struct inode *i_vfs_inode; 412 413 /* Flags of inode [j_list_lock] */ 414 unsigned int i_flags; 415 }; 416 417 struct jbd2_revoke_table_s; 418 419 /** 420 * struct handle_s - The handle_s type is the concrete type associated with 421 * handle_t. 422 * @h_transaction: Which compound transaction is this update a part of? 423 * @h_buffer_credits: Number of remaining buffers we are allowed to dirty. 424 * @h_ref: Reference count on this handle 425 * @h_err: Field for caller's use to track errors through large fs operations 426 * @h_sync: flag for sync-on-close 427 * @h_jdata: flag to force data journaling 428 * @h_aborted: flag indicating fatal error on handle 429 **/ 430 431 /* Docbook can't yet cope with the bit fields, but will leave the documentation 432 * in so it can be fixed later. 433 */ 434 435 struct handle_s 436 { 437 /* Which compound transaction is this update a part of? */ 438 transaction_t *h_transaction; 439 440 /* Number of remaining buffers we are allowed to dirty: */ 441 int h_buffer_credits; 442 443 /* Reference count on this handle */ 444 int h_ref; 445 446 /* Field for caller's use to track errors through large fs */ 447 /* operations */ 448 int h_err; 449 450 /* Flags [no locking] */ 451 unsigned int h_sync: 1; /* sync-on-close */ 452 unsigned int h_jdata: 1; /* force data journaling */ 453 unsigned int h_aborted: 1; /* fatal error on handle */ 454 455 #ifdef CONFIG_DEBUG_LOCK_ALLOC 456 struct lockdep_map h_lockdep_map; 457 #endif 458 }; 459 460 461 /* 462 * Some stats for checkpoint phase 463 */ 464 struct transaction_chp_stats_s { 465 unsigned long cs_chp_time; 466 unsigned long cs_forced_to_close; 467 unsigned long cs_written; 468 unsigned long cs_dropped; 469 }; 470 471 /* The transaction_t type is the guts of the journaling mechanism. It 472 * tracks a compound transaction through its various states: 473 * 474 * RUNNING: accepting new updates 475 * LOCKED: Updates still running but we don't accept new ones 476 * RUNDOWN: Updates are tidying up but have finished requesting 477 * new buffers to modify (state not used for now) 478 * FLUSH: All updates complete, but we are still writing to disk 479 * COMMIT: All data on disk, writing commit record 480 * FINISHED: We still have to keep the transaction for checkpointing. 481 * 482 * The transaction keeps track of all of the buffers modified by a 483 * running transaction, and all of the buffers committed but not yet 484 * flushed to home for finished transactions. 485 */ 486 487 /* 488 * Lock ranking: 489 * 490 * j_list_lock 491 * ->jbd_lock_bh_journal_head() (This is "innermost") 492 * 493 * j_state_lock 494 * ->jbd_lock_bh_state() 495 * 496 * jbd_lock_bh_state() 497 * ->j_list_lock 498 * 499 * j_state_lock 500 * ->t_handle_lock 501 * 502 * j_state_lock 503 * ->j_list_lock (journal_unmap_buffer) 504 * 505 */ 506 507 struct transaction_s 508 { 509 /* Pointer to the journal for this transaction. [no locking] */ 510 journal_t *t_journal; 511 512 /* Sequence number for this transaction [no locking] */ 513 tid_t t_tid; 514 515 /* 516 * Transaction's current state 517 * [no locking - only kjournald2 alters this] 518 * [j_list_lock] guards transition of a transaction into T_FINISHED 519 * state and subsequent call of __jbd2_journal_drop_transaction() 520 * FIXME: needs barriers 521 * KLUDGE: [use j_state_lock] 522 */ 523 enum { 524 T_RUNNING, 525 T_LOCKED, 526 T_RUNDOWN, 527 T_FLUSH, 528 T_COMMIT, 529 T_FINISHED 530 } t_state; 531 532 /* 533 * Where in the log does this transaction's commit start? [no locking] 534 */ 535 unsigned long t_log_start; 536 537 /* Number of buffers on the t_buffers list [j_list_lock] */ 538 int t_nr_buffers; 539 540 /* 541 * Doubly-linked circular list of all buffers reserved but not yet 542 * modified by this transaction [j_list_lock] 543 */ 544 struct journal_head *t_reserved_list; 545 546 /* 547 * Doubly-linked circular list of all metadata buffers owned by this 548 * transaction [j_list_lock] 549 */ 550 struct journal_head *t_buffers; 551 552 /* 553 * Doubly-linked circular list of all forget buffers (superseded 554 * buffers which we can un-checkpoint once this transaction commits) 555 * [j_list_lock] 556 */ 557 struct journal_head *t_forget; 558 559 /* 560 * Doubly-linked circular list of all buffers still to be flushed before 561 * this transaction can be checkpointed. [j_list_lock] 562 */ 563 struct journal_head *t_checkpoint_list; 564 565 /* 566 * Doubly-linked circular list of all buffers submitted for IO while 567 * checkpointing. [j_list_lock] 568 */ 569 struct journal_head *t_checkpoint_io_list; 570 571 /* 572 * Doubly-linked circular list of temporary buffers currently undergoing 573 * IO in the log [j_list_lock] 574 */ 575 struct journal_head *t_iobuf_list; 576 577 /* 578 * Doubly-linked circular list of metadata buffers being shadowed by log 579 * IO. The IO buffers on the iobuf list and the shadow buffers on this 580 * list match each other one for one at all times. [j_list_lock] 581 */ 582 struct journal_head *t_shadow_list; 583 584 /* 585 * Doubly-linked circular list of control buffers being written to the 586 * log. [j_list_lock] 587 */ 588 struct journal_head *t_log_list; 589 590 /* 591 * List of inodes whose data we've modified in data=ordered mode. 592 * [j_list_lock] 593 */ 594 struct list_head t_inode_list; 595 596 /* 597 * Protects info related to handles 598 */ 599 spinlock_t t_handle_lock; 600 601 /* 602 * Longest time some handle had to wait for running transaction 603 */ 604 unsigned long t_max_wait; 605 606 /* 607 * When transaction started 608 */ 609 unsigned long t_start; 610 611 /* 612 * Checkpointing stats [j_checkpoint_sem] 613 */ 614 struct transaction_chp_stats_s t_chp_stats; 615 616 /* 617 * Number of outstanding updates running on this transaction 618 * [t_handle_lock] 619 */ 620 int t_updates; 621 622 /* 623 * Number of buffers reserved for use by all handles in this transaction 624 * handle but not yet modified. [t_handle_lock] 625 */ 626 int t_outstanding_credits; 627 628 /* 629 * Forward and backward links for the circular list of all transactions 630 * awaiting checkpoint. [j_list_lock] 631 */ 632 transaction_t *t_cpnext, *t_cpprev; 633 634 /* 635 * When will the transaction expire (become due for commit), in jiffies? 636 * [no locking] 637 */ 638 unsigned long t_expires; 639 640 /* 641 * How many handles used this transaction? [t_handle_lock] 642 */ 643 int t_handle_count; 644 645 /* 646 * For use by the filesystem to store fs-specific data 647 * structures associated with the transaction 648 */ 649 struct list_head t_private_list; 650 }; 651 652 struct transaction_run_stats_s { 653 unsigned long rs_wait; 654 unsigned long rs_running; 655 unsigned long rs_locked; 656 unsigned long rs_flushing; 657 unsigned long rs_logging; 658 659 unsigned long rs_handle_count; 660 unsigned long rs_blocks; 661 unsigned long rs_blocks_logged; 662 }; 663 664 struct transaction_stats_s { 665 int ts_type; 666 unsigned long ts_tid; 667 union { 668 struct transaction_run_stats_s run; 669 struct transaction_chp_stats_s chp; 670 } u; 671 }; 672 673 #define JBD2_STATS_RUN 1 674 #define JBD2_STATS_CHECKPOINT 2 675 676 static inline unsigned long 677 jbd2_time_diff(unsigned long start, unsigned long end) 678 { 679 if (end >= start) 680 return end - start; 681 682 return end + (MAX_JIFFY_OFFSET - start); 683 } 684 685 /** 686 * struct journal_s - The journal_s type is the concrete type associated with 687 * journal_t. 688 * @j_flags: General journaling state flags 689 * @j_errno: Is there an outstanding uncleared error on the journal (from a 690 * prior abort)? 691 * @j_sb_buffer: First part of superblock buffer 692 * @j_superblock: Second part of superblock buffer 693 * @j_format_version: Version of the superblock format 694 * @j_state_lock: Protect the various scalars in the journal 695 * @j_barrier_count: Number of processes waiting to create a barrier lock 696 * @j_barrier: The barrier lock itself 697 * @j_running_transaction: The current running transaction.. 698 * @j_committing_transaction: the transaction we are pushing to disk 699 * @j_checkpoint_transactions: a linked circular list of all transactions 700 * waiting for checkpointing 701 * @j_wait_transaction_locked: Wait queue for waiting for a locked transaction 702 * to start committing, or for a barrier lock to be released 703 * @j_wait_logspace: Wait queue for waiting for checkpointing to complete 704 * @j_wait_done_commit: Wait queue for waiting for commit to complete 705 * @j_wait_checkpoint: Wait queue to trigger checkpointing 706 * @j_wait_commit: Wait queue to trigger commit 707 * @j_wait_updates: Wait queue to wait for updates to complete 708 * @j_checkpoint_mutex: Mutex for locking against concurrent checkpoints 709 * @j_head: Journal head - identifies the first unused block in the journal 710 * @j_tail: Journal tail - identifies the oldest still-used block in the 711 * journal. 712 * @j_free: Journal free - how many free blocks are there in the journal? 713 * @j_first: The block number of the first usable block 714 * @j_last: The block number one beyond the last usable block 715 * @j_dev: Device where we store the journal 716 * @j_blocksize: blocksize for the location where we store the journal. 717 * @j_blk_offset: starting block offset for into the device where we store the 718 * journal 719 * @j_fs_dev: Device which holds the client fs. For internal journal this will 720 * be equal to j_dev 721 * @j_maxlen: Total maximum capacity of the journal region on disk. 722 * @j_list_lock: Protects the buffer lists and internal buffer state. 723 * @j_inode: Optional inode where we store the journal. If present, all journal 724 * block numbers are mapped into this inode via bmap(). 725 * @j_tail_sequence: Sequence number of the oldest transaction in the log 726 * @j_transaction_sequence: Sequence number of the next transaction to grant 727 * @j_commit_sequence: Sequence number of the most recently committed 728 * transaction 729 * @j_commit_request: Sequence number of the most recent transaction wanting 730 * commit 731 * @j_uuid: Uuid of client object. 732 * @j_task: Pointer to the current commit thread for this journal 733 * @j_max_transaction_buffers: Maximum number of metadata buffers to allow in a 734 * single compound commit transaction 735 * @j_commit_interval: What is the maximum transaction lifetime before we begin 736 * a commit? 737 * @j_commit_timer: The timer used to wakeup the commit thread 738 * @j_revoke_lock: Protect the revoke table 739 * @j_revoke: The revoke table - maintains the list of revoked blocks in the 740 * current transaction. 741 * @j_revoke_table: alternate revoke tables for j_revoke 742 * @j_wbuf: array of buffer_heads for jbd2_journal_commit_transaction 743 * @j_wbufsize: maximum number of buffer_heads allowed in j_wbuf, the 744 * number that will fit in j_blocksize 745 * @j_last_sync_writer: most recent pid which did a synchronous write 746 * @j_history: Buffer storing the transactions statistics history 747 * @j_history_max: Maximum number of transactions in the statistics history 748 * @j_history_cur: Current number of transactions in the statistics history 749 * @j_history_lock: Protect the transactions statistics history 750 * @j_proc_entry: procfs entry for the jbd statistics directory 751 * @j_stats: Overall statistics 752 * @j_private: An opaque pointer to fs-private information. 753 */ 754 755 struct journal_s 756 { 757 /* General journaling state flags [j_state_lock] */ 758 unsigned long j_flags; 759 760 /* 761 * Is there an outstanding uncleared error on the journal (from a prior 762 * abort)? [j_state_lock] 763 */ 764 int j_errno; 765 766 /* The superblock buffer */ 767 struct buffer_head *j_sb_buffer; 768 journal_superblock_t *j_superblock; 769 770 /* Version of the superblock format */ 771 int j_format_version; 772 773 /* 774 * Protect the various scalars in the journal 775 */ 776 spinlock_t j_state_lock; 777 778 /* 779 * Number of processes waiting to create a barrier lock [j_state_lock] 780 */ 781 int j_barrier_count; 782 783 /* The barrier lock itself */ 784 struct mutex j_barrier; 785 786 /* 787 * Transactions: The current running transaction... 788 * [j_state_lock] [caller holding open handle] 789 */ 790 transaction_t *j_running_transaction; 791 792 /* 793 * the transaction we are pushing to disk 794 * [j_state_lock] [caller holding open handle] 795 */ 796 transaction_t *j_committing_transaction; 797 798 /* 799 * ... and a linked circular list of all transactions waiting for 800 * checkpointing. [j_list_lock] 801 */ 802 transaction_t *j_checkpoint_transactions; 803 804 /* 805 * Wait queue for waiting for a locked transaction to start committing, 806 * or for a barrier lock to be released 807 */ 808 wait_queue_head_t j_wait_transaction_locked; 809 810 /* Wait queue for waiting for checkpointing to complete */ 811 wait_queue_head_t j_wait_logspace; 812 813 /* Wait queue for waiting for commit to complete */ 814 wait_queue_head_t j_wait_done_commit; 815 816 /* Wait queue to trigger checkpointing */ 817 wait_queue_head_t j_wait_checkpoint; 818 819 /* Wait queue to trigger commit */ 820 wait_queue_head_t j_wait_commit; 821 822 /* Wait queue to wait for updates to complete */ 823 wait_queue_head_t j_wait_updates; 824 825 /* Semaphore for locking against concurrent checkpoints */ 826 struct mutex j_checkpoint_mutex; 827 828 /* 829 * Journal head: identifies the first unused block in the journal. 830 * [j_state_lock] 831 */ 832 unsigned long j_head; 833 834 /* 835 * Journal tail: identifies the oldest still-used block in the journal. 836 * [j_state_lock] 837 */ 838 unsigned long j_tail; 839 840 /* 841 * Journal free: how many free blocks are there in the journal? 842 * [j_state_lock] 843 */ 844 unsigned long j_free; 845 846 /* 847 * Journal start and end: the block numbers of the first usable block 848 * and one beyond the last usable block in the journal. [j_state_lock] 849 */ 850 unsigned long j_first; 851 unsigned long j_last; 852 853 /* 854 * Device, blocksize and starting block offset for the location where we 855 * store the journal. 856 */ 857 struct block_device *j_dev; 858 int j_blocksize; 859 unsigned long long j_blk_offset; 860 char j_devname[BDEVNAME_SIZE+24]; 861 862 /* 863 * Device which holds the client fs. For internal journal this will be 864 * equal to j_dev. 865 */ 866 struct block_device *j_fs_dev; 867 868 /* Total maximum capacity of the journal region on disk. */ 869 unsigned int j_maxlen; 870 871 /* 872 * Protects the buffer lists and internal buffer state. 873 */ 874 spinlock_t j_list_lock; 875 876 /* Optional inode where we store the journal. If present, all */ 877 /* journal block numbers are mapped into this inode via */ 878 /* bmap(). */ 879 struct inode *j_inode; 880 881 /* 882 * Sequence number of the oldest transaction in the log [j_state_lock] 883 */ 884 tid_t j_tail_sequence; 885 886 /* 887 * Sequence number of the next transaction to grant [j_state_lock] 888 */ 889 tid_t j_transaction_sequence; 890 891 /* 892 * Sequence number of the most recently committed transaction 893 * [j_state_lock]. 894 */ 895 tid_t j_commit_sequence; 896 897 /* 898 * Sequence number of the most recent transaction wanting commit 899 * [j_state_lock] 900 */ 901 tid_t j_commit_request; 902 903 /* 904 * Journal uuid: identifies the object (filesystem, LVM volume etc) 905 * backed by this journal. This will eventually be replaced by an array 906 * of uuids, allowing us to index multiple devices within a single 907 * journal and to perform atomic updates across them. 908 */ 909 __u8 j_uuid[16]; 910 911 /* Pointer to the current commit thread for this journal */ 912 struct task_struct *j_task; 913 914 /* 915 * Maximum number of metadata buffers to allow in a single compound 916 * commit transaction 917 */ 918 int j_max_transaction_buffers; 919 920 /* 921 * What is the maximum transaction lifetime before we begin a commit? 922 */ 923 unsigned long j_commit_interval; 924 925 /* The timer used to wakeup the commit thread: */ 926 struct timer_list j_commit_timer; 927 928 /* 929 * The revoke table: maintains the list of revoked blocks in the 930 * current transaction. [j_revoke_lock] 931 */ 932 spinlock_t j_revoke_lock; 933 struct jbd2_revoke_table_s *j_revoke; 934 struct jbd2_revoke_table_s *j_revoke_table[2]; 935 936 /* 937 * array of bhs for jbd2_journal_commit_transaction 938 */ 939 struct buffer_head **j_wbuf; 940 int j_wbufsize; 941 942 pid_t j_last_sync_writer; 943 944 /* This function is called when a transaction is closed */ 945 void (*j_commit_callback)(journal_t *, 946 transaction_t *); 947 948 /* 949 * Journal statistics 950 */ 951 struct transaction_stats_s *j_history; 952 int j_history_max; 953 int j_history_cur; 954 /* 955 * Protect the transactions statistics history 956 */ 957 spinlock_t j_history_lock; 958 struct proc_dir_entry *j_proc_entry; 959 struct transaction_stats_s j_stats; 960 961 /* Failed journal commit ID */ 962 unsigned int j_failed_commit; 963 964 /* 965 * An opaque pointer to fs-private information. ext3 puts its 966 * superblock pointer here 967 */ 968 void *j_private; 969 }; 970 971 /* 972 * Journal flag definitions 973 */ 974 #define JBD2_UNMOUNT 0x001 /* Journal thread is being destroyed */ 975 #define JBD2_ABORT 0x002 /* Journaling has been aborted for errors. */ 976 #define JBD2_ACK_ERR 0x004 /* The errno in the sb has been acked */ 977 #define JBD2_FLUSHED 0x008 /* The journal superblock has been flushed */ 978 #define JBD2_LOADED 0x010 /* The journal superblock has been loaded */ 979 #define JBD2_BARRIER 0x020 /* Use IDE barriers */ 980 #define JBD2_ABORT_ON_SYNCDATA_ERR 0x040 /* Abort the journal on file 981 * data write error in ordered 982 * mode */ 983 984 /* 985 * Function declarations for the journaling transaction and buffer 986 * management 987 */ 988 989 /* Filing buffers */ 990 extern void jbd2_journal_unfile_buffer(journal_t *, struct journal_head *); 991 extern void __jbd2_journal_unfile_buffer(struct journal_head *); 992 extern void __jbd2_journal_refile_buffer(struct journal_head *); 993 extern void jbd2_journal_refile_buffer(journal_t *, struct journal_head *); 994 extern void __jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int); 995 extern void __journal_free_buffer(struct journal_head *bh); 996 extern void jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int); 997 extern void __journal_clean_data_list(transaction_t *transaction); 998 999 /* Log buffer allocation */ 1000 extern struct journal_head * jbd2_journal_get_descriptor_buffer(journal_t *); 1001 int jbd2_journal_next_log_block(journal_t *, unsigned long long *); 1002 1003 /* Commit management */ 1004 extern void jbd2_journal_commit_transaction(journal_t *); 1005 1006 /* Checkpoint list management */ 1007 int __jbd2_journal_clean_checkpoint_list(journal_t *journal); 1008 int __jbd2_journal_remove_checkpoint(struct journal_head *); 1009 void __jbd2_journal_insert_checkpoint(struct journal_head *, transaction_t *); 1010 1011 1012 /* 1013 * Triggers 1014 */ 1015 1016 struct jbd2_buffer_trigger_type { 1017 /* 1018 * Fired just before a buffer is written to the journal. 1019 * mapped_data is a mapped buffer that is the frozen data for 1020 * commit. 1021 */ 1022 void (*t_commit)(struct jbd2_buffer_trigger_type *type, 1023 struct buffer_head *bh, void *mapped_data, 1024 size_t size); 1025 1026 /* 1027 * Fired during journal abort for dirty buffers that will not be 1028 * committed. 1029 */ 1030 void (*t_abort)(struct jbd2_buffer_trigger_type *type, 1031 struct buffer_head *bh); 1032 }; 1033 1034 extern void jbd2_buffer_commit_trigger(struct journal_head *jh, 1035 void *mapped_data, 1036 struct jbd2_buffer_trigger_type *triggers); 1037 extern void jbd2_buffer_abort_trigger(struct journal_head *jh, 1038 struct jbd2_buffer_trigger_type *triggers); 1039 1040 /* Buffer IO */ 1041 extern int 1042 jbd2_journal_write_metadata_buffer(transaction_t *transaction, 1043 struct journal_head *jh_in, 1044 struct journal_head **jh_out, 1045 unsigned long long blocknr); 1046 1047 /* Transaction locking */ 1048 extern void __wait_on_journal (journal_t *); 1049 1050 /* 1051 * Journal locking. 1052 * 1053 * We need to lock the journal during transaction state changes so that nobody 1054 * ever tries to take a handle on the running transaction while we are in the 1055 * middle of moving it to the commit phase. j_state_lock does this. 1056 * 1057 * Note that the locking is completely interrupt unsafe. We never touch 1058 * journal structures from interrupts. 1059 */ 1060 1061 static inline handle_t *journal_current_handle(void) 1062 { 1063 return current->journal_info; 1064 } 1065 1066 /* The journaling code user interface: 1067 * 1068 * Create and destroy handles 1069 * Register buffer modifications against the current transaction. 1070 */ 1071 1072 extern handle_t *jbd2_journal_start(journal_t *, int nblocks); 1073 extern int jbd2_journal_restart (handle_t *, int nblocks); 1074 extern int jbd2_journal_extend (handle_t *, int nblocks); 1075 extern int jbd2_journal_get_write_access(handle_t *, struct buffer_head *); 1076 extern int jbd2_journal_get_create_access (handle_t *, struct buffer_head *); 1077 extern int jbd2_journal_get_undo_access(handle_t *, struct buffer_head *); 1078 void jbd2_journal_set_triggers(struct buffer_head *, 1079 struct jbd2_buffer_trigger_type *type); 1080 extern int jbd2_journal_dirty_metadata (handle_t *, struct buffer_head *); 1081 extern void jbd2_journal_release_buffer (handle_t *, struct buffer_head *); 1082 extern int jbd2_journal_forget (handle_t *, struct buffer_head *); 1083 extern void journal_sync_buffer (struct buffer_head *); 1084 extern void jbd2_journal_invalidatepage(journal_t *, 1085 struct page *, unsigned long); 1086 extern int jbd2_journal_try_to_free_buffers(journal_t *, struct page *, gfp_t); 1087 extern int jbd2_journal_stop(handle_t *); 1088 extern int jbd2_journal_flush (journal_t *); 1089 extern void jbd2_journal_lock_updates (journal_t *); 1090 extern void jbd2_journal_unlock_updates (journal_t *); 1091 1092 extern journal_t * jbd2_journal_init_dev(struct block_device *bdev, 1093 struct block_device *fs_dev, 1094 unsigned long long start, int len, int bsize); 1095 extern journal_t * jbd2_journal_init_inode (struct inode *); 1096 extern int jbd2_journal_update_format (journal_t *); 1097 extern int jbd2_journal_check_used_features 1098 (journal_t *, unsigned long, unsigned long, unsigned long); 1099 extern int jbd2_journal_check_available_features 1100 (journal_t *, unsigned long, unsigned long, unsigned long); 1101 extern int jbd2_journal_set_features 1102 (journal_t *, unsigned long, unsigned long, unsigned long); 1103 extern void jbd2_journal_clear_features 1104 (journal_t *, unsigned long, unsigned long, unsigned long); 1105 extern int jbd2_journal_create (journal_t *); 1106 extern int jbd2_journal_load (journal_t *journal); 1107 extern int jbd2_journal_destroy (journal_t *); 1108 extern int jbd2_journal_recover (journal_t *journal); 1109 extern int jbd2_journal_wipe (journal_t *, int); 1110 extern int jbd2_journal_skip_recovery (journal_t *); 1111 extern void jbd2_journal_update_superblock (journal_t *, int); 1112 extern void __jbd2_journal_abort_hard (journal_t *); 1113 extern void jbd2_journal_abort (journal_t *, int); 1114 extern int jbd2_journal_errno (journal_t *); 1115 extern void jbd2_journal_ack_err (journal_t *); 1116 extern int jbd2_journal_clear_err (journal_t *); 1117 extern int jbd2_journal_bmap(journal_t *, unsigned long, unsigned long long *); 1118 extern int jbd2_journal_force_commit(journal_t *); 1119 extern int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *inode); 1120 extern int jbd2_journal_begin_ordered_truncate(struct jbd2_inode *inode, loff_t new_size); 1121 extern void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode); 1122 extern void jbd2_journal_release_jbd_inode(journal_t *journal, struct jbd2_inode *jinode); 1123 1124 /* 1125 * journal_head management 1126 */ 1127 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh); 1128 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh); 1129 void jbd2_journal_remove_journal_head(struct buffer_head *bh); 1130 void jbd2_journal_put_journal_head(struct journal_head *jh); 1131 1132 /* 1133 * handle management 1134 */ 1135 extern struct kmem_cache *jbd2_handle_cache; 1136 1137 static inline handle_t *jbd2_alloc_handle(gfp_t gfp_flags) 1138 { 1139 return kmem_cache_alloc(jbd2_handle_cache, gfp_flags); 1140 } 1141 1142 static inline void jbd2_free_handle(handle_t *handle) 1143 { 1144 kmem_cache_free(jbd2_handle_cache, handle); 1145 } 1146 1147 /* Primary revoke support */ 1148 #define JOURNAL_REVOKE_DEFAULT_HASH 256 1149 extern int jbd2_journal_init_revoke(journal_t *, int); 1150 extern void jbd2_journal_destroy_revoke_caches(void); 1151 extern int jbd2_journal_init_revoke_caches(void); 1152 1153 extern void jbd2_journal_destroy_revoke(journal_t *); 1154 extern int jbd2_journal_revoke (handle_t *, unsigned long long, struct buffer_head *); 1155 extern int jbd2_journal_cancel_revoke(handle_t *, struct journal_head *); 1156 extern void jbd2_journal_write_revoke_records(journal_t *, transaction_t *); 1157 1158 /* Recovery revoke support */ 1159 extern int jbd2_journal_set_revoke(journal_t *, unsigned long long, tid_t); 1160 extern int jbd2_journal_test_revoke(journal_t *, unsigned long long, tid_t); 1161 extern void jbd2_journal_clear_revoke(journal_t *); 1162 extern void jbd2_journal_switch_revoke_table(journal_t *journal); 1163 1164 /* 1165 * The log thread user interface: 1166 * 1167 * Request space in the current transaction, and force transaction commit 1168 * transitions on demand. 1169 */ 1170 1171 int __jbd2_log_space_left(journal_t *); /* Called with journal locked */ 1172 int jbd2_log_start_commit(journal_t *journal, tid_t tid); 1173 int __jbd2_log_start_commit(journal_t *journal, tid_t tid); 1174 int jbd2_journal_start_commit(journal_t *journal, tid_t *tid); 1175 int jbd2_journal_force_commit_nested(journal_t *journal); 1176 int jbd2_log_wait_commit(journal_t *journal, tid_t tid); 1177 int jbd2_log_do_checkpoint(journal_t *journal); 1178 1179 void __jbd2_log_wait_for_space(journal_t *journal); 1180 extern void __jbd2_journal_drop_transaction(journal_t *, transaction_t *); 1181 extern int jbd2_cleanup_journal_tail(journal_t *); 1182 1183 /* Debugging code only: */ 1184 1185 #define jbd_ENOSYS() \ 1186 do { \ 1187 printk (KERN_ERR "JBD unimplemented function %s\n", __func__); \ 1188 current->state = TASK_UNINTERRUPTIBLE; \ 1189 schedule(); \ 1190 } while (1) 1191 1192 /* 1193 * is_journal_abort 1194 * 1195 * Simple test wrapper function to test the JBD2_ABORT state flag. This 1196 * bit, when set, indicates that we have had a fatal error somewhere, 1197 * either inside the journaling layer or indicated to us by the client 1198 * (eg. ext3), and that we and should not commit any further 1199 * transactions. 1200 */ 1201 1202 static inline int is_journal_aborted(journal_t *journal) 1203 { 1204 return journal->j_flags & JBD2_ABORT; 1205 } 1206 1207 static inline int is_handle_aborted(handle_t *handle) 1208 { 1209 if (handle->h_aborted) 1210 return 1; 1211 return is_journal_aborted(handle->h_transaction->t_journal); 1212 } 1213 1214 static inline void jbd2_journal_abort_handle(handle_t *handle) 1215 { 1216 handle->h_aborted = 1; 1217 } 1218 1219 #endif /* __KERNEL__ */ 1220 1221 /* Comparison functions for transaction IDs: perform comparisons using 1222 * modulo arithmetic so that they work over sequence number wraps. */ 1223 1224 static inline int tid_gt(tid_t x, tid_t y) 1225 { 1226 int difference = (x - y); 1227 return (difference > 0); 1228 } 1229 1230 static inline int tid_geq(tid_t x, tid_t y) 1231 { 1232 int difference = (x - y); 1233 return (difference >= 0); 1234 } 1235 1236 extern int jbd2_journal_blocks_per_page(struct inode *inode); 1237 extern size_t journal_tag_bytes(journal_t *journal); 1238 1239 /* 1240 * Return the minimum number of blocks which must be free in the journal 1241 * before a new transaction may be started. Must be called under j_state_lock. 1242 */ 1243 static inline int jbd_space_needed(journal_t *journal) 1244 { 1245 int nblocks = journal->j_max_transaction_buffers; 1246 if (journal->j_committing_transaction) 1247 nblocks += journal->j_committing_transaction-> 1248 t_outstanding_credits; 1249 return nblocks; 1250 } 1251 1252 /* 1253 * Definitions which augment the buffer_head layer 1254 */ 1255 1256 /* journaling buffer types */ 1257 #define BJ_None 0 /* Not journaled */ 1258 #define BJ_Metadata 1 /* Normal journaled metadata */ 1259 #define BJ_Forget 2 /* Buffer superseded by this transaction */ 1260 #define BJ_IO 3 /* Buffer is for temporary IO use */ 1261 #define BJ_Shadow 4 /* Buffer contents being shadowed to the log */ 1262 #define BJ_LogCtl 5 /* Buffer contains log descriptors */ 1263 #define BJ_Reserved 6 /* Buffer is reserved for access by journal */ 1264 #define BJ_Types 7 1265 1266 extern int jbd_blocks_per_page(struct inode *inode); 1267 1268 #ifdef __KERNEL__ 1269 1270 #define buffer_trace_init(bh) do {} while (0) 1271 #define print_buffer_fields(bh) do {} while (0) 1272 #define print_buffer_trace(bh) do {} while (0) 1273 #define BUFFER_TRACE(bh, info) do {} while (0) 1274 #define BUFFER_TRACE2(bh, bh2, info) do {} while (0) 1275 #define JBUFFER_TRACE(jh, info) do {} while (0) 1276 1277 #endif /* __KERNEL__ */ 1278 1279 #endif /* _LINUX_JBD2_H */ 1280