1 /* SPDX-License-Identifier: GPL-2.0-or-later */
5 * Defines journalling api and structures.
7 * Copyright (C) 2003, 2005 Oracle. All rights reserved.
10 #ifndef OCFS2_JOURNAL_H
11 #define OCFS2_JOURNAL_H
14 #include <linux/jbd2.h>
16 enum ocfs2_journal_state {
17 OCFS2_JOURNAL_FREE = 0,
19 OCFS2_JOURNAL_IN_SHUTDOWN,
26 * The recovery_list is a simple linked list of node numbers to recover.
27 * It is protected by the recovery_lock.
30 struct ocfs2_recovery_map {
32 unsigned int *rm_entries;
36 struct ocfs2_journal {
37 enum ocfs2_journal_state j_state; /* Journals current state */
39 journal_t *j_journal; /* The kernels journal type */
40 struct inode *j_inode; /* Kernel inode pointing to
42 struct ocfs2_super *j_osb; /* pointer to the super
46 * necessarily the super
48 * which we usually run
51 struct buffer_head *j_bh; /* Journal disk inode block */
52 atomic_t j_num_trans; /* Number of transactions
53 * currently in the system. */
55 unsigned long j_trans_id;
56 struct rw_semaphore j_trans_barrier;
57 wait_queue_head_t j_checkpointed;
59 /* both fields protected by j_lock*/
60 struct list_head j_la_cleanups;
61 struct work_struct j_recovery_work;
64 extern spinlock_t trans_inc_lock;
66 /* wrap j_trans_id so we never have it equal to zero. */
67 static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
70 spin_lock(&trans_inc_lock);
71 old_id = j->j_trans_id++;
72 if (unlikely(!j->j_trans_id))
74 spin_unlock(&trans_inc_lock);
78 static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
79 struct ocfs2_caching_info *ci)
81 spin_lock(&trans_inc_lock);
82 ci->ci_last_trans = journal->j_trans_id;
83 spin_unlock(&trans_inc_lock);
86 /* Used to figure out whether it's safe to drop a metadata lock on an
87 * cached object. Returns true if all the object's changes have been
88 * checkpointed to disk. You should be holding the spinlock on the
89 * metadata lock while calling this to be sure that nobody can take
90 * the lock and put it on another transaction. */
91 static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
94 struct ocfs2_journal *journal =
95 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
97 spin_lock(&trans_inc_lock);
98 ret = time_after(journal->j_trans_id, ci->ci_last_trans);
99 spin_unlock(&trans_inc_lock);
103 /* convenience function to check if an object backed by struct
104 * ocfs2_caching_info is still new (has never hit disk) Will do you a
105 * favor and set created_trans = 0 when you've
106 * been checkpointed. returns '1' if the ci is still new. */
107 static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
110 struct ocfs2_journal *journal =
111 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
113 spin_lock(&trans_inc_lock);
114 ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
116 ci->ci_created_trans = 0;
117 spin_unlock(&trans_inc_lock);
121 /* Wrapper for inodes so we can check system files */
122 static inline int ocfs2_inode_is_new(struct inode *inode)
124 /* System files are never "new" as they're written out by
125 * mkfs. This helps us early during mount, before we have the
126 * journal open and j_trans_id could be junk. */
127 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
130 return ocfs2_ci_is_new(INODE_CACHE(inode));
133 static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
134 struct ocfs2_caching_info *ci)
136 spin_lock(&trans_inc_lock);
137 ci->ci_created_trans = osb->journal->j_trans_id;
138 spin_unlock(&trans_inc_lock);
141 /* Exported only for the journal struct init code in super.c. Do not call. */
142 void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
143 void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
144 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
146 void ocfs2_complete_recovery(struct work_struct *work);
147 void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
149 int ocfs2_recovery_init(struct ocfs2_super *osb);
150 void ocfs2_recovery_exit(struct ocfs2_super *osb);
152 int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
155 * Initialize, Load, Shutdown, Wipe a journal.
157 * ocfs2_journal_init - Initialize journal structures in the OSB.
158 * ocfs2_journal_load - Load the given journal off disk. Replay it if
159 * there's transactions still in there.
160 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
161 * uncommitted, uncheckpointed transactions.
162 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
163 * zero out each block.
164 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
165 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
167 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
169 void ocfs2_set_journal_params(struct ocfs2_super *osb);
170 int ocfs2_journal_init(struct ocfs2_super *osb, int *dirty);
171 void ocfs2_journal_shutdown(struct ocfs2_super *osb);
172 int ocfs2_journal_wipe(struct ocfs2_journal *journal,
174 int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
176 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
177 void ocfs2_recovery_thread(struct ocfs2_super *osb,
179 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
180 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
181 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
183 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
185 wake_up(&osb->checkpoint_event);
188 static inline void ocfs2_checkpoint_inode(struct inode *inode)
190 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
192 if (ocfs2_mount_local(osb))
195 if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
196 /* WARNING: This only kicks off a single
197 * checkpoint. If someone races you and adds more
198 * metadata to the journal, you won't know, and will
199 * wind up waiting *a lot* longer than necessary. Right
200 * now we only use this in clear_inode so that's
202 ocfs2_start_checkpoint(osb);
204 wait_event(osb->journal->j_checkpointed,
205 ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
210 * Transaction Handling:
211 * Manage the lifetime of a transaction handle.
213 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
214 * the number of blocks that will be changed during
216 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
217 * the journal was aborted. The majority of paths don't
218 * check the return value as an error there comes too
219 * late to do anything (and will be picked up in a
220 * later transaction).
221 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
222 * commit the handle to disk in the process, but will
223 * not release any locks taken during the transaction.
224 * ocfs2_journal_access* - Notify the handle that we want to journal this
225 * buffer. Will have to call ocfs2_journal_dirty once
226 * we've actually dirtied it. Type is one of . or .
227 * Always call the specific flavor of
228 * ocfs2_journal_access_*() unless you intend to
229 * manage the checksum by hand.
230 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
231 * ocfs2_jbd2_inode_add_write - Mark an inode with range so that its data goes
232 * out before the current handle commits.
235 /* You must always start_trans with a number of buffs > 0, but it's
236 * perfectly legal to go through an entire transaction without having
237 * dirtied any buffers. */
238 handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
240 int ocfs2_commit_trans(struct ocfs2_super *osb,
242 int ocfs2_extend_trans(handle_t *handle, int nblocks);
243 int ocfs2_allocate_extend_trans(handle_t *handle,
247 * Define an arbitrary limit for the amount of data we will anticipate
248 * writing to any given transaction. For unbounded transactions such as
249 * fallocate(2) we can write more than this, but we always
250 * start off at the maximum transaction size and grow the transaction
251 * optimistically as we go.
253 #define OCFS2_MAX_TRANS_DATA 64U
256 * Create access is for when we get a newly created buffer and we're
257 * not gonna read it off disk, but rather fill it ourselves. Right
258 * now, we don't do anything special with this (it turns into a write
259 * request), but this is a good placeholder in case we do...
261 * Write access is for when we read a block off disk and are going to
262 * modify it. This way the journalling layer knows it may need to make
263 * a copy of that block (if it's part of another, uncommitted
264 * transaction) before we do so.
266 #define OCFS2_JOURNAL_ACCESS_CREATE 0
267 #define OCFS2_JOURNAL_ACCESS_WRITE 1
268 #define OCFS2_JOURNAL_ACCESS_UNDO 2
272 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
273 struct buffer_head *bh, int type);
274 /* ocfs2_extent_block */
275 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
276 struct buffer_head *bh, int type);
277 /* ocfs2_refcount_block */
278 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
279 struct buffer_head *bh, int type);
280 /* ocfs2_group_desc */
281 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
282 struct buffer_head *bh, int type);
283 /* ocfs2_xattr_block */
284 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
285 struct buffer_head *bh, int type);
287 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
288 struct buffer_head *bh, int type);
290 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
291 struct buffer_head *bh, int type);
292 /* ocfs2_dx_root_block */
293 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
294 struct buffer_head *bh, int type);
296 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
297 struct buffer_head *bh, int type);
298 /* Anything that has no ecc */
299 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
300 struct buffer_head *bh, int type);
303 * A word about the journal_access/journal_dirty "dance". It is
304 * entirely legal to journal_access a buffer more than once (as long
305 * as the access type is the same -- I'm not sure what will happen if
306 * access type is different but this should never happen anyway) It is
307 * also legal to journal_dirty a buffer more than once. In fact, you
308 * can even journal_access a buffer after you've done a
309 * journal_access/journal_dirty pair. The only thing you cannot do
310 * however, is journal_dirty a buffer which you haven't yet passed to
311 * journal_access at least once.
313 * That said, 99% of the time this doesn't matter and this is what the
317 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
319 * ocfs2_journal_dirty(handle, bh);
321 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
325 * Convenience macros to calculate number of credits needed.
327 * For convenience sake, I have a set of macros here which calculate
328 * the *maximum* number of sectors which will be changed for various
332 /* simple file updates like chmod, etc. */
333 #define OCFS2_INODE_UPDATE_CREDITS 1
335 /* extended attribute block update */
336 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
338 /* Update of a single quota block */
339 #define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
341 /* global quotafile inode update, data block */
342 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
343 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
345 #define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
347 * The two writes below can accidentally see global info dirty due
348 * to set_info() quotactl so make them prepared for the writes.
350 /* quota data block, global info */
351 /* Write to local quota file */
352 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
353 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
355 /* global quota data block, local quota data block, global quota inode,
356 * global quota info */
357 #define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
358 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
360 static inline int ocfs2_quota_trans_credits(struct super_block *sb)
364 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
365 credits += OCFS2_QWRITE_CREDITS;
366 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
367 credits += OCFS2_QWRITE_CREDITS;
371 /* group extend. inode update and last group update. */
372 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
374 /* group add. inode update and the new group update. */
375 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
377 /* get one bit out of a suballocator: dinode + group descriptor +
378 * prev. group desc. if we relink. */
379 #define OCFS2_SUBALLOC_ALLOC (3)
381 static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
383 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
384 ocfs2_quota_trans_credits(sb);
387 /* dinode + group descriptor update. We don't relink on free yet. */
388 #define OCFS2_SUBALLOC_FREE (2)
390 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
391 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
392 + OCFS2_TRUNCATE_LOG_UPDATE)
394 static inline int ocfs2_remove_extent_credits(struct super_block *sb)
396 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
397 ocfs2_quota_trans_credits(sb);
400 /* data block for new dir/symlink, allocation of directory block, dx_root
401 * update for free list */
402 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
404 static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
406 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
407 * worth of blocks for initial extent. */
408 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
409 ocfs2_clusters_to_blocks(sb, 1);
412 /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
413 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
414 * blocks + quota update */
415 static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
418 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
421 dir_credits += ocfs2_add_dir_index_credits(sb);
423 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
424 ocfs2_quota_trans_credits(sb);
427 /* local alloc metadata change + main bitmap updates */
428 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
429 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
431 /* used when we don't need an allocation change for a dir extend. One
432 * for the dinode, one for the new block. */
433 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
435 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
436 * update on dir + index leaf + dx root update for free list +
437 * previous dirblock update in the free list */
438 static inline int ocfs2_link_credits(struct super_block *sb)
440 return 2 * OCFS2_INODE_UPDATE_CREDITS + 4 +
441 ocfs2_quota_trans_credits(sb);
444 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
445 * dir inode link + dir inode index leaf + dir index root */
446 static inline int ocfs2_unlink_credits(struct super_block *sb)
448 /* The quota update from ocfs2_link_credits is unused here... */
449 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
452 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
453 * inode alloc group descriptor + orphan dir index root +
454 * orphan dir index leaf */
455 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
457 /* dinode + orphan dir dinode + extent tree leaf block + orphan dir entry +
458 * orphan dir index root + orphan dir index leaf */
459 #define OCFS2_INODE_ADD_TO_ORPHAN_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 4)
460 #define OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS OCFS2_INODE_ADD_TO_ORPHAN_CREDITS
462 /* dinode update, old dir dinode update, new dir dinode update, old
463 * dir dir entry, new dir dir entry, dir entry update for renaming
464 * directory + target unlink + 3 x dir index leaves */
465 static inline int ocfs2_rename_credits(struct super_block *sb)
467 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
470 /* global bitmap dinode, group desc., relinked group,
471 * suballocator dinode, group desc., relinked group,
472 * dinode, xattr block */
473 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
474 + OCFS2_INODE_UPDATE_CREDITS \
475 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
477 /* inode update, removal of dx root block from allocator */
478 #define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
481 static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
483 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
485 credits += ocfs2_clusters_to_blocks(sb, 1);
486 credits += ocfs2_quota_trans_credits(sb);
491 /* inode update, new refcount block and its allocation credits. */
492 #define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
493 + OCFS2_SUBALLOC_ALLOC)
495 /* inode and the refcount block update. */
496 #define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
499 * inode and the refcount block update.
500 * It doesn't include the credits for sub alloc change.
501 * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
503 #define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
505 /* 2 metadata alloc, 2 new blocks and root refcount block */
506 #define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
509 * Please note that the caller must make sure that root_el is the root
510 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
511 * the result may be wrong.
513 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
514 struct ocfs2_extent_list *root_el)
516 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
518 /* bitmap dinode, group desc. + relinked group. */
519 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
521 /* we might need to shift tree depth so lets assume an
522 * absolute worst case of complete fragmentation. Even with
523 * that, we only need one update for the dinode, and then
524 * however many metadata chunks needed * a remaining suballoc
526 sysfile_bitmap_blocks = 1 +
527 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
529 /* this does not include *new* metadata blocks, which are
530 * accounted for in sysfile_bitmap_blocks. root_el +
531 * prev. last_eb_blk + blocks along edge of tree.
532 * calc_symlink_credits passes because we just need 1
533 * credit for the dinode there. */
534 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
536 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
537 ocfs2_quota_trans_credits(sb);
540 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
542 int blocks = ocfs2_mknod_credits(sb, 0, 0);
544 /* links can be longer than one block so we may update many
545 * within our single allocated extent. */
546 blocks += ocfs2_clusters_to_blocks(sb, 1);
548 return blocks + ocfs2_quota_trans_credits(sb);
551 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
555 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
556 /* parent inode update + new block group header + bitmap inode update
557 + bitmap blocks affected */
558 blocks = 1 + 1 + 1 + bitmap_blocks;
563 * Allocating a discontiguous block group requires the credits from
564 * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
565 * the group descriptor's extent list. The caller already has started
566 * the transaction with ocfs2_calc_group_alloc_credits(). They extend
567 * it with these credits.
569 static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
571 return ocfs2_extent_recs_per_gd(sb);
574 static inline int ocfs2_jbd2_inode_add_write(handle_t *handle, struct inode *inode,
575 loff_t start_byte, loff_t length)
577 return jbd2_journal_inode_ranged_write(handle,
578 &OCFS2_I(inode)->ip_jinode,
582 static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
585 return jbd2_journal_begin_ordered_truncate(
586 OCFS2_SB(inode->i_sb)->journal->j_journal,
587 &OCFS2_I(inode)->ip_jinode,
591 static inline void ocfs2_update_inode_fsync_trans(handle_t *handle,
595 struct ocfs2_inode_info *oi = OCFS2_I(inode);
597 if (!is_handle_aborted(handle)) {
598 oi->i_sync_tid = handle->h_transaction->t_tid;
600 oi->i_datasync_tid = handle->h_transaction->t_tid;
604 #endif /* OCFS2_JOURNAL_H */