Loading...
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * journal.h
4 *
5 * Defines journalling api and structures.
6 *
7 * Copyright (C) 2003, 2005 Oracle. All rights reserved.
8 */
9
10#ifndef OCFS2_JOURNAL_H
11#define OCFS2_JOURNAL_H
12
13#include <linux/fs.h>
14#include <linux/jbd2.h>
15
16enum ocfs2_journal_state {
17 OCFS2_JOURNAL_FREE = 0,
18 OCFS2_JOURNAL_LOADED,
19 OCFS2_JOURNAL_IN_SHUTDOWN,
20};
21
22struct ocfs2_super;
23struct ocfs2_dinode;
24
25/*
26 * The recovery_list is a simple linked list of node numbers to recover.
27 * It is protected by the recovery_lock.
28 */
29
30struct ocfs2_recovery_map {
31 unsigned int rm_used;
32 unsigned int rm_entries[];
33};
34
35
36struct ocfs2_journal {
37 enum ocfs2_journal_state j_state; /* Journals current state */
38
39 journal_t *j_journal; /* The kernels journal type */
40 struct inode *j_inode; /* Kernel inode pointing to
41 * this journal */
42 struct ocfs2_super *j_osb; /* pointer to the super
43 * block for the node
44 * we're currently
45 * running on -- not
46 * necessarily the super
47 * block from the node
48 * which we usually run
49 * from (recovery,
50 * etc) */
51 struct buffer_head *j_bh; /* Journal disk inode block */
52 atomic_t j_num_trans; /* Number of transactions
53 * currently in the system. */
54 spinlock_t j_lock;
55 unsigned long j_trans_id;
56 struct rw_semaphore j_trans_barrier;
57 wait_queue_head_t j_checkpointed;
58
59 /* both fields protected by j_lock*/
60 struct list_head j_la_cleanups;
61 struct work_struct j_recovery_work;
62};
63
64extern spinlock_t trans_inc_lock;
65
66/* wrap j_trans_id so we never have it equal to zero. */
67static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
68{
69 unsigned long old_id;
70 spin_lock(&trans_inc_lock);
71 old_id = j->j_trans_id++;
72 if (unlikely(!j->j_trans_id))
73 j->j_trans_id = 1;
74 spin_unlock(&trans_inc_lock);
75 return old_id;
76}
77
78static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
79 struct ocfs2_caching_info *ci)
80{
81 spin_lock(&trans_inc_lock);
82 ci->ci_last_trans = journal->j_trans_id;
83 spin_unlock(&trans_inc_lock);
84}
85
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. */
91static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
92{
93 int ret;
94 struct ocfs2_journal *journal =
95 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
96
97 spin_lock(&trans_inc_lock);
98 ret = time_after(journal->j_trans_id, ci->ci_last_trans);
99 spin_unlock(&trans_inc_lock);
100 return ret;
101}
102
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. */
107static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
108{
109 int ret;
110 struct ocfs2_journal *journal =
111 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
112
113 spin_lock(&trans_inc_lock);
114 ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
115 if (!ret)
116 ci->ci_created_trans = 0;
117 spin_unlock(&trans_inc_lock);
118 return ret;
119}
120
121/* Wrapper for inodes so we can check system files */
122static inline int ocfs2_inode_is_new(struct inode *inode)
123{
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)
128 return 0;
129
130 return ocfs2_ci_is_new(INODE_CACHE(inode));
131}
132
133static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
134 struct ocfs2_caching_info *ci)
135{
136 spin_lock(&trans_inc_lock);
137 ci->ci_created_trans = osb->journal->j_trans_id;
138 spin_unlock(&trans_inc_lock);
139}
140
141/* Exported only for the journal struct init code in super.c. Do not call. */
142void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
143void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
144void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
145
146void ocfs2_complete_recovery(struct work_struct *work);
147void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
148
149int ocfs2_recovery_init(struct ocfs2_super *osb);
150void ocfs2_recovery_exit(struct ocfs2_super *osb);
151
152int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
153void ocfs2_free_replay_slots(struct ocfs2_super *osb);
154/*
155 * Journal Control:
156 * Initialize, Load, Shutdown, Wipe a journal.
157 *
158 * ocfs2_journal_alloc - Initialize skeleton for journal structure.
159 * ocfs2_journal_init - Initialize journal structures in the OSB.
160 * ocfs2_journal_load - Load the given journal off disk. Replay it if
161 * there's transactions still in there.
162 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
163 * uncommitted, uncheckpointed transactions.
164 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
165 * zero out each block.
166 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
167 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
168 * event on.
169 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
170 */
171void ocfs2_set_journal_params(struct ocfs2_super *osb);
172int ocfs2_journal_alloc(struct ocfs2_super *osb);
173int ocfs2_journal_init(struct ocfs2_super *osb, int *dirty);
174void ocfs2_journal_shutdown(struct ocfs2_super *osb);
175int ocfs2_journal_wipe(struct ocfs2_journal *journal,
176 int full);
177int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
178 int replayed);
179int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
180void ocfs2_recovery_thread(struct ocfs2_super *osb,
181 int node_num);
182int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
183void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
184void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
185
186static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
187{
188 wake_up(&osb->checkpoint_event);
189}
190
191static inline void ocfs2_checkpoint_inode(struct inode *inode)
192{
193 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
194
195 if (ocfs2_mount_local(osb))
196 return;
197
198 if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
199 /* WARNING: This only kicks off a single
200 * checkpoint. If someone races you and adds more
201 * metadata to the journal, you won't know, and will
202 * wind up waiting *a lot* longer than necessary. Right
203 * now we only use this in clear_inode so that's
204 * OK. */
205 ocfs2_start_checkpoint(osb);
206
207 wait_event(osb->journal->j_checkpointed,
208 ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
209 }
210}
211
212/*
213 * Transaction Handling:
214 * Manage the lifetime of a transaction handle.
215 *
216 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
217 * the number of blocks that will be changed during
218 * this handle.
219 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
220 * the journal was aborted. The majority of paths don't
221 * check the return value as an error there comes too
222 * late to do anything (and will be picked up in a
223 * later transaction).
224 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
225 * commit the handle to disk in the process, but will
226 * not release any locks taken during the transaction.
227 * ocfs2_journal_access* - Notify the handle that we want to journal this
228 * buffer. Will have to call ocfs2_journal_dirty once
229 * we've actually dirtied it. Type is one of . or .
230 * Always call the specific flavor of
231 * ocfs2_journal_access_*() unless you intend to
232 * manage the checksum by hand.
233 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
234 * ocfs2_jbd2_inode_add_write - Mark an inode with range so that its data goes
235 * out before the current handle commits.
236 */
237
238/* You must always start_trans with a number of buffs > 0, but it's
239 * perfectly legal to go through an entire transaction without having
240 * dirtied any buffers. */
241handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
242 int max_buffs);
243int ocfs2_commit_trans(struct ocfs2_super *osb,
244 handle_t *handle);
245int ocfs2_extend_trans(handle_t *handle, int nblocks);
246int ocfs2_assure_trans_credits(handle_t *handle,
247 int nblocks);
248int ocfs2_allocate_extend_trans(handle_t *handle,
249 int thresh);
250
251/*
252 * Define an arbitrary limit for the amount of data we will anticipate
253 * writing to any given transaction. For unbounded transactions such as
254 * fallocate(2) we can write more than this, but we always
255 * start off at the maximum transaction size and grow the transaction
256 * optimistically as we go.
257 */
258#define OCFS2_MAX_TRANS_DATA 64U
259
260/*
261 * Create access is for when we get a newly created buffer and we're
262 * not gonna read it off disk, but rather fill it ourselves. Right
263 * now, we don't do anything special with this (it turns into a write
264 * request), but this is a good placeholder in case we do...
265 *
266 * Write access is for when we read a block off disk and are going to
267 * modify it. This way the journalling layer knows it may need to make
268 * a copy of that block (if it's part of another, uncommitted
269 * transaction) before we do so.
270 */
271#define OCFS2_JOURNAL_ACCESS_CREATE 0
272#define OCFS2_JOURNAL_ACCESS_WRITE 1
273#define OCFS2_JOURNAL_ACCESS_UNDO 2
274
275
276/* ocfs2_inode */
277int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
278 struct buffer_head *bh, int type);
279/* ocfs2_extent_block */
280int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
281 struct buffer_head *bh, int type);
282/* ocfs2_refcount_block */
283int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
284 struct buffer_head *bh, int type);
285/* ocfs2_group_desc */
286int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
287 struct buffer_head *bh, int type);
288/* ocfs2_xattr_block */
289int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
290 struct buffer_head *bh, int type);
291/* quota blocks */
292int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
293 struct buffer_head *bh, int type);
294/* dirblock */
295int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
296 struct buffer_head *bh, int type);
297/* ocfs2_dx_root_block */
298int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
299 struct buffer_head *bh, int type);
300/* ocfs2_dx_leaf */
301int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
302 struct buffer_head *bh, int type);
303/* Anything that has no ecc */
304int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
305 struct buffer_head *bh, int type);
306
307/*
308 * A word about the journal_access/journal_dirty "dance". It is
309 * entirely legal to journal_access a buffer more than once (as long
310 * as the access type is the same -- I'm not sure what will happen if
311 * access type is different but this should never happen anyway) It is
312 * also legal to journal_dirty a buffer more than once. In fact, you
313 * can even journal_access a buffer after you've done a
314 * journal_access/journal_dirty pair. The only thing you cannot do
315 * however, is journal_dirty a buffer which you haven't yet passed to
316 * journal_access at least once.
317 *
318 * That said, 99% of the time this doesn't matter and this is what the
319 * path looks like:
320 *
321 * <read a bh>
322 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
323 * <modify the bh>
324 * ocfs2_journal_dirty(handle, bh);
325 */
326void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
327
328/*
329 * Credit Macros:
330 * Convenience macros to calculate number of credits needed.
331 *
332 * For convenience sake, I have a set of macros here which calculate
333 * the *maximum* number of sectors which will be changed for various
334 * metadata updates.
335 */
336
337/* simple file updates like chmod, etc. */
338#define OCFS2_INODE_UPDATE_CREDITS 1
339
340/* extended attribute block update */
341#define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
342
343/* Update of a single quota block */
344#define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
345
346/* global quotafile inode update, data block */
347#define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
348 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
349
350#define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
351/*
352 * The two writes below can accidentally see global info dirty due
353 * to set_info() quotactl so make them prepared for the writes.
354 */
355/* quota data block, global info */
356/* Write to local quota file */
357#define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
358 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
359
360/* global quota data block, local quota data block, global quota inode,
361 * global quota info */
362#define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
363 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
364
365static inline int ocfs2_quota_trans_credits(struct super_block *sb)
366{
367 int credits = 0;
368
369 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
370 credits += OCFS2_QWRITE_CREDITS;
371 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
372 credits += OCFS2_QWRITE_CREDITS;
373 return credits;
374}
375
376/* group extend. inode update and last group update. */
377#define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
378
379/* group add. inode update and the new group update. */
380#define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
381
382/* get one bit out of a suballocator: dinode + group descriptor +
383 * prev. group desc. if we relink. */
384#define OCFS2_SUBALLOC_ALLOC (3)
385
386static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
387{
388 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
389 ocfs2_quota_trans_credits(sb);
390}
391
392/* dinode + group descriptor update. We don't relink on free yet. */
393#define OCFS2_SUBALLOC_FREE (2)
394
395#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
396#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
397 + OCFS2_TRUNCATE_LOG_UPDATE)
398
399static inline int ocfs2_remove_extent_credits(struct super_block *sb)
400{
401 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
402 ocfs2_quota_trans_credits(sb);
403}
404
405/* data block for new dir/symlink, allocation of directory block, dx_root
406 * update for free list */
407#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
408
409static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
410{
411 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
412 * worth of blocks for initial extent. */
413 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
414 ocfs2_clusters_to_blocks(sb, 1);
415}
416
417/* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
418 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
419 * blocks + quota update */
420static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
421 int xattr_credits)
422{
423 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
424
425 if (is_dir)
426 dir_credits += ocfs2_add_dir_index_credits(sb);
427
428 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
429 ocfs2_quota_trans_credits(sb);
430}
431
432/* local alloc metadata change + main bitmap updates */
433#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
434 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
435
436/* used when we don't need an allocation change for a dir extend. One
437 * for the dinode, one for the new block. */
438#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
439
440/* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
441 * update on dir + index leaf + dx root update for free list +
442 * previous dirblock update in the free list */
443static inline int ocfs2_link_credits(struct super_block *sb)
444{
445 return 2 * OCFS2_INODE_UPDATE_CREDITS + 4 +
446 ocfs2_quota_trans_credits(sb);
447}
448
449/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
450 * dir inode link + dir inode index leaf + dir index root */
451static inline int ocfs2_unlink_credits(struct super_block *sb)
452{
453 /* The quota update from ocfs2_link_credits is unused here... */
454 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
455}
456
457/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
458 * inode alloc group descriptor + orphan dir index root +
459 * orphan dir index leaf */
460#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
461
462/* dinode + orphan dir dinode + extent tree leaf block + orphan dir entry +
463 * orphan dir index root + orphan dir index leaf */
464#define OCFS2_INODE_ADD_TO_ORPHAN_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 4)
465#define OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS OCFS2_INODE_ADD_TO_ORPHAN_CREDITS
466
467/* dinode update, old dir dinode update, new dir dinode update, old
468 * dir dir entry, new dir dir entry, dir entry update for renaming
469 * directory + target unlink + 3 x dir index leaves */
470static inline int ocfs2_rename_credits(struct super_block *sb)
471{
472 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
473}
474
475/* global bitmap dinode, group desc., relinked group,
476 * suballocator dinode, group desc., relinked group,
477 * dinode, xattr block */
478#define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
479 + OCFS2_INODE_UPDATE_CREDITS \
480 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
481
482/* inode update, removal of dx root block from allocator */
483#define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
484 OCFS2_SUBALLOC_FREE)
485
486static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
487{
488 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
489
490 credits += ocfs2_clusters_to_blocks(sb, 1);
491 credits += ocfs2_quota_trans_credits(sb);
492
493 return credits;
494}
495
496/* inode update, new refcount block and its allocation credits. */
497#define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
498 + OCFS2_SUBALLOC_ALLOC)
499
500/* inode and the refcount block update. */
501#define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
502
503/*
504 * inode and the refcount block update.
505 * It doesn't include the credits for sub alloc change.
506 * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
507 */
508#define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
509
510/* 2 metadata alloc, 2 new blocks and root refcount block */
511#define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
512
513/*
514 * Please note that the caller must make sure that root_el is the root
515 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
516 * the result may be wrong.
517 */
518static inline int ocfs2_calc_extend_credits(struct super_block *sb,
519 struct ocfs2_extent_list *root_el)
520{
521 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
522
523 /* bitmap dinode, group desc. + relinked group. */
524 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
525
526 /* we might need to shift tree depth so lets assume an
527 * absolute worst case of complete fragmentation. Even with
528 * that, we only need one update for the dinode, and then
529 * however many metadata chunks needed * a remaining suballoc
530 * alloc. */
531 sysfile_bitmap_blocks = 1 +
532 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
533
534 /* this does not include *new* metadata blocks, which are
535 * accounted for in sysfile_bitmap_blocks. root_el +
536 * prev. last_eb_blk + blocks along edge of tree.
537 * calc_symlink_credits passes because we just need 1
538 * credit for the dinode there. */
539 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
540
541 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
542 ocfs2_quota_trans_credits(sb);
543}
544
545static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
546{
547 int blocks = ocfs2_mknod_credits(sb, 0, 0);
548
549 /* links can be longer than one block so we may update many
550 * within our single allocated extent. */
551 blocks += ocfs2_clusters_to_blocks(sb, 1);
552
553 return blocks + ocfs2_quota_trans_credits(sb);
554}
555
556static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
557 unsigned int cpg)
558{
559 int blocks;
560 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
561 /* parent inode update + new block group header + bitmap inode update
562 + bitmap blocks affected */
563 blocks = 1 + 1 + 1 + bitmap_blocks;
564 return blocks;
565}
566
567/*
568 * Allocating a discontiguous block group requires the credits from
569 * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
570 * the group descriptor's extent list. The caller already has started
571 * the transaction with ocfs2_calc_group_alloc_credits(). They extend
572 * it with these credits.
573 */
574static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
575{
576 return ocfs2_extent_recs_per_gd(sb);
577}
578
579static inline int ocfs2_jbd2_inode_add_write(handle_t *handle, struct inode *inode,
580 loff_t start_byte, loff_t length)
581{
582 return jbd2_journal_inode_ranged_write(handle,
583 &OCFS2_I(inode)->ip_jinode,
584 start_byte, length);
585}
586
587static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
588 loff_t new_size)
589{
590 return jbd2_journal_begin_ordered_truncate(
591 OCFS2_SB(inode->i_sb)->journal->j_journal,
592 &OCFS2_I(inode)->ip_jinode,
593 new_size);
594}
595
596static inline void ocfs2_update_inode_fsync_trans(handle_t *handle,
597 struct inode *inode,
598 int datasync)
599{
600 struct ocfs2_inode_info *oi = OCFS2_I(inode);
601
602 if (!is_handle_aborted(handle)) {
603 oi->i_sync_tid = handle->h_transaction->t_tid;
604 if (datasync)
605 oi->i_datasync_tid = handle->h_transaction->t_tid;
606 }
607}
608
609#endif /* OCFS2_JOURNAL_H */
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * journal.h
5 *
6 * Defines journalling api and structures.
7 *
8 * Copyright (C) 2003, 2005 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#ifndef OCFS2_JOURNAL_H
27#define OCFS2_JOURNAL_H
28
29#include <linux/fs.h>
30#include <linux/jbd2.h>
31
32enum ocfs2_journal_state {
33 OCFS2_JOURNAL_FREE = 0,
34 OCFS2_JOURNAL_LOADED,
35 OCFS2_JOURNAL_IN_SHUTDOWN,
36};
37
38struct ocfs2_super;
39struct ocfs2_dinode;
40
41/*
42 * The recovery_list is a simple linked list of node numbers to recover.
43 * It is protected by the recovery_lock.
44 */
45
46struct ocfs2_recovery_map {
47 unsigned int rm_used;
48 unsigned int *rm_entries;
49};
50
51
52struct ocfs2_journal {
53 enum ocfs2_journal_state j_state; /* Journals current state */
54
55 journal_t *j_journal; /* The kernels journal type */
56 struct inode *j_inode; /* Kernel inode pointing to
57 * this journal */
58 struct ocfs2_super *j_osb; /* pointer to the super
59 * block for the node
60 * we're currently
61 * running on -- not
62 * necessarily the super
63 * block from the node
64 * which we usually run
65 * from (recovery,
66 * etc) */
67 struct buffer_head *j_bh; /* Journal disk inode block */
68 atomic_t j_num_trans; /* Number of transactions
69 * currently in the system. */
70 spinlock_t j_lock;
71 unsigned long j_trans_id;
72 struct rw_semaphore j_trans_barrier;
73 wait_queue_head_t j_checkpointed;
74
75 /* both fields protected by j_lock*/
76 struct list_head j_la_cleanups;
77 struct work_struct j_recovery_work;
78};
79
80extern spinlock_t trans_inc_lock;
81
82/* wrap j_trans_id so we never have it equal to zero. */
83static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
84{
85 unsigned long old_id;
86 spin_lock(&trans_inc_lock);
87 old_id = j->j_trans_id++;
88 if (unlikely(!j->j_trans_id))
89 j->j_trans_id = 1;
90 spin_unlock(&trans_inc_lock);
91 return old_id;
92}
93
94static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
95 struct ocfs2_caching_info *ci)
96{
97 spin_lock(&trans_inc_lock);
98 ci->ci_last_trans = journal->j_trans_id;
99 spin_unlock(&trans_inc_lock);
100}
101
102/* Used to figure out whether it's safe to drop a metadata lock on an
103 * cached object. Returns true if all the object's changes have been
104 * checkpointed to disk. You should be holding the spinlock on the
105 * metadata lock while calling this to be sure that nobody can take
106 * the lock and put it on another transaction. */
107static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
108{
109 int ret;
110 struct ocfs2_journal *journal =
111 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
112
113 spin_lock(&trans_inc_lock);
114 ret = time_after(journal->j_trans_id, ci->ci_last_trans);
115 spin_unlock(&trans_inc_lock);
116 return ret;
117}
118
119/* convenience function to check if an object backed by struct
120 * ocfs2_caching_info is still new (has never hit disk) Will do you a
121 * favor and set created_trans = 0 when you've
122 * been checkpointed. returns '1' if the ci is still new. */
123static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
124{
125 int ret;
126 struct ocfs2_journal *journal =
127 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
128
129 spin_lock(&trans_inc_lock);
130 ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
131 if (!ret)
132 ci->ci_created_trans = 0;
133 spin_unlock(&trans_inc_lock);
134 return ret;
135}
136
137/* Wrapper for inodes so we can check system files */
138static inline int ocfs2_inode_is_new(struct inode *inode)
139{
140 /* System files are never "new" as they're written out by
141 * mkfs. This helps us early during mount, before we have the
142 * journal open and j_trans_id could be junk. */
143 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
144 return 0;
145
146 return ocfs2_ci_is_new(INODE_CACHE(inode));
147}
148
149static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
150 struct ocfs2_caching_info *ci)
151{
152 spin_lock(&trans_inc_lock);
153 ci->ci_created_trans = osb->journal->j_trans_id;
154 spin_unlock(&trans_inc_lock);
155}
156
157/* Exported only for the journal struct init code in super.c. Do not call. */
158void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
159void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
160void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
161void ocfs2_orphan_scan_exit(struct ocfs2_super *osb);
162
163void ocfs2_complete_recovery(struct work_struct *work);
164void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
165
166int ocfs2_recovery_init(struct ocfs2_super *osb);
167void ocfs2_recovery_exit(struct ocfs2_super *osb);
168
169int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
170/*
171 * Journal Control:
172 * Initialize, Load, Shutdown, Wipe a journal.
173 *
174 * ocfs2_journal_init - Initialize journal structures in the OSB.
175 * ocfs2_journal_load - Load the given journal off disk. Replay it if
176 * there's transactions still in there.
177 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
178 * uncommitted, uncheckpointed transactions.
179 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
180 * zero out each block.
181 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
182 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
183 * event on.
184 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
185 */
186void ocfs2_set_journal_params(struct ocfs2_super *osb);
187int ocfs2_journal_init(struct ocfs2_journal *journal,
188 int *dirty);
189void ocfs2_journal_shutdown(struct ocfs2_super *osb);
190int ocfs2_journal_wipe(struct ocfs2_journal *journal,
191 int full);
192int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
193 int replayed);
194int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
195void ocfs2_recovery_thread(struct ocfs2_super *osb,
196 int node_num);
197int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
198void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
199void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
200
201static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
202{
203 wake_up(&osb->checkpoint_event);
204}
205
206static inline void ocfs2_checkpoint_inode(struct inode *inode)
207{
208 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
209
210 if (ocfs2_mount_local(osb))
211 return;
212
213 if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
214 /* WARNING: This only kicks off a single
215 * checkpoint. If someone races you and adds more
216 * metadata to the journal, you won't know, and will
217 * wind up waiting *a lot* longer than necessary. Right
218 * now we only use this in clear_inode so that's
219 * OK. */
220 ocfs2_start_checkpoint(osb);
221
222 wait_event(osb->journal->j_checkpointed,
223 ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
224 }
225}
226
227/*
228 * Transaction Handling:
229 * Manage the lifetime of a transaction handle.
230 *
231 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
232 * the number of blocks that will be changed during
233 * this handle.
234 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
235 * the journal was aborted. The majority of paths don't
236 * check the return value as an error there comes too
237 * late to do anything (and will be picked up in a
238 * later transaction).
239 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
240 * commit the handle to disk in the process, but will
241 * not release any locks taken during the transaction.
242 * ocfs2_journal_access* - Notify the handle that we want to journal this
243 * buffer. Will have to call ocfs2_journal_dirty once
244 * we've actually dirtied it. Type is one of . or .
245 * Always call the specific flavor of
246 * ocfs2_journal_access_*() unless you intend to
247 * manage the checksum by hand.
248 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
249 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
250 * the current handle commits.
251 */
252
253/* You must always start_trans with a number of buffs > 0, but it's
254 * perfectly legal to go through an entire transaction without having
255 * dirtied any buffers. */
256handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
257 int max_buffs);
258int ocfs2_commit_trans(struct ocfs2_super *osb,
259 handle_t *handle);
260int ocfs2_extend_trans(handle_t *handle, int nblocks);
261int ocfs2_allocate_extend_trans(handle_t *handle,
262 int thresh);
263
264/*
265 * Define an arbitrary limit for the amount of data we will anticipate
266 * writing to any given transaction. For unbounded transactions such as
267 * fallocate(2) we can write more than this, but we always
268 * start off at the maximum transaction size and grow the transaction
269 * optimistically as we go.
270 */
271#define OCFS2_MAX_TRANS_DATA 64U
272
273/*
274 * Create access is for when we get a newly created buffer and we're
275 * not gonna read it off disk, but rather fill it ourselves. Right
276 * now, we don't do anything special with this (it turns into a write
277 * request), but this is a good placeholder in case we do...
278 *
279 * Write access is for when we read a block off disk and are going to
280 * modify it. This way the journalling layer knows it may need to make
281 * a copy of that block (if it's part of another, uncommitted
282 * transaction) before we do so.
283 */
284#define OCFS2_JOURNAL_ACCESS_CREATE 0
285#define OCFS2_JOURNAL_ACCESS_WRITE 1
286#define OCFS2_JOURNAL_ACCESS_UNDO 2
287
288
289/* ocfs2_inode */
290int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
291 struct buffer_head *bh, int type);
292/* ocfs2_extent_block */
293int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
294 struct buffer_head *bh, int type);
295/* ocfs2_refcount_block */
296int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
297 struct buffer_head *bh, int type);
298/* ocfs2_group_desc */
299int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
300 struct buffer_head *bh, int type);
301/* ocfs2_xattr_block */
302int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
303 struct buffer_head *bh, int type);
304/* quota blocks */
305int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
306 struct buffer_head *bh, int type);
307/* dirblock */
308int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
309 struct buffer_head *bh, int type);
310/* ocfs2_dx_root_block */
311int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
312 struct buffer_head *bh, int type);
313/* ocfs2_dx_leaf */
314int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
315 struct buffer_head *bh, int type);
316/* Anything that has no ecc */
317int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
318 struct buffer_head *bh, int type);
319
320/*
321 * A word about the journal_access/journal_dirty "dance". It is
322 * entirely legal to journal_access a buffer more than once (as long
323 * as the access type is the same -- I'm not sure what will happen if
324 * access type is different but this should never happen anyway) It is
325 * also legal to journal_dirty a buffer more than once. In fact, you
326 * can even journal_access a buffer after you've done a
327 * journal_access/journal_dirty pair. The only thing you cannot do
328 * however, is journal_dirty a buffer which you haven't yet passed to
329 * journal_access at least once.
330 *
331 * That said, 99% of the time this doesn't matter and this is what the
332 * path looks like:
333 *
334 * <read a bh>
335 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
336 * <modify the bh>
337 * ocfs2_journal_dirty(handle, bh);
338 */
339void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
340
341/*
342 * Credit Macros:
343 * Convenience macros to calculate number of credits needed.
344 *
345 * For convenience sake, I have a set of macros here which calculate
346 * the *maximum* number of sectors which will be changed for various
347 * metadata updates.
348 */
349
350/* simple file updates like chmod, etc. */
351#define OCFS2_INODE_UPDATE_CREDITS 1
352
353/* extended attribute block update */
354#define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
355
356/* Update of a single quota block */
357#define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
358
359/* global quotafile inode update, data block */
360#define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
361 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
362
363#define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
364/*
365 * The two writes below can accidentally see global info dirty due
366 * to set_info() quotactl so make them prepared for the writes.
367 */
368/* quota data block, global info */
369/* Write to local quota file */
370#define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
371 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
372
373/* global quota data block, local quota data block, global quota inode,
374 * global quota info */
375#define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
376 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
377
378static inline int ocfs2_quota_trans_credits(struct super_block *sb)
379{
380 int credits = 0;
381
382 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
383 credits += OCFS2_QWRITE_CREDITS;
384 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
385 credits += OCFS2_QWRITE_CREDITS;
386 return credits;
387}
388
389/* group extend. inode update and last group update. */
390#define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
391
392/* group add. inode update and the new group update. */
393#define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
394
395/* get one bit out of a suballocator: dinode + group descriptor +
396 * prev. group desc. if we relink. */
397#define OCFS2_SUBALLOC_ALLOC (3)
398
399static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
400{
401 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
402 ocfs2_quota_trans_credits(sb);
403}
404
405/* dinode + group descriptor update. We don't relink on free yet. */
406#define OCFS2_SUBALLOC_FREE (2)
407
408#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
409#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
410 + OCFS2_TRUNCATE_LOG_UPDATE)
411
412static inline int ocfs2_remove_extent_credits(struct super_block *sb)
413{
414 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
415 ocfs2_quota_trans_credits(sb);
416}
417
418/* data block for new dir/symlink, allocation of directory block, dx_root
419 * update for free list */
420#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
421
422static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
423{
424 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
425 * worth of blocks for initial extent. */
426 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
427 ocfs2_clusters_to_blocks(sb, 1);
428}
429
430/* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
431 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
432 * blocks + quota update */
433static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
434 int xattr_credits)
435{
436 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
437
438 if (is_dir)
439 dir_credits += ocfs2_add_dir_index_credits(sb);
440
441 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
442 ocfs2_quota_trans_credits(sb);
443}
444
445/* local alloc metadata change + main bitmap updates */
446#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
447 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
448
449/* used when we don't need an allocation change for a dir extend. One
450 * for the dinode, one for the new block. */
451#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
452
453/* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
454 * update on dir + index leaf + dx root update for free list +
455 * previous dirblock update in the free list */
456static inline int ocfs2_link_credits(struct super_block *sb)
457{
458 return 2*OCFS2_INODE_UPDATE_CREDITS + 4 +
459 ocfs2_quota_trans_credits(sb);
460}
461
462/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
463 * dir inode link + dir inode index leaf + dir index root */
464static inline int ocfs2_unlink_credits(struct super_block *sb)
465{
466 /* The quota update from ocfs2_link_credits is unused here... */
467 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
468}
469
470/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
471 * inode alloc group descriptor + orphan dir index root +
472 * orphan dir index leaf */
473#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
474
475/* dinode + orphan dir dinode + extent tree leaf block + orphan dir entry +
476 * orphan dir index root + orphan dir index leaf */
477#define OCFS2_INODE_ADD_TO_ORPHAN_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 4)
478#define OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS OCFS2_INODE_ADD_TO_ORPHAN_CREDITS
479
480/* dinode update, old dir dinode update, new dir dinode update, old
481 * dir dir entry, new dir dir entry, dir entry update for renaming
482 * directory + target unlink + 3 x dir index leaves */
483static inline int ocfs2_rename_credits(struct super_block *sb)
484{
485 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
486}
487
488/* global bitmap dinode, group desc., relinked group,
489 * suballocator dinode, group desc., relinked group,
490 * dinode, xattr block */
491#define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
492 + OCFS2_INODE_UPDATE_CREDITS \
493 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
494
495/* inode update, removal of dx root block from allocator */
496#define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
497 OCFS2_SUBALLOC_FREE)
498
499static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
500{
501 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
502
503 credits += ocfs2_clusters_to_blocks(sb, 1);
504 credits += ocfs2_quota_trans_credits(sb);
505
506 return credits;
507}
508
509/* inode update, new refcount block and its allocation credits. */
510#define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
511 + OCFS2_SUBALLOC_ALLOC)
512
513/* inode and the refcount block update. */
514#define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
515
516/*
517 * inode and the refcount block update.
518 * It doesn't include the credits for sub alloc change.
519 * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
520 */
521#define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
522
523/* 2 metadata alloc, 2 new blocks and root refcount block */
524#define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
525
526/*
527 * Please note that the caller must make sure that root_el is the root
528 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
529 * the result may be wrong.
530 */
531static inline int ocfs2_calc_extend_credits(struct super_block *sb,
532 struct ocfs2_extent_list *root_el)
533{
534 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
535
536 /* bitmap dinode, group desc. + relinked group. */
537 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
538
539 /* we might need to shift tree depth so lets assume an
540 * absolute worst case of complete fragmentation. Even with
541 * that, we only need one update for the dinode, and then
542 * however many metadata chunks needed * a remaining suballoc
543 * alloc. */
544 sysfile_bitmap_blocks = 1 +
545 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
546
547 /* this does not include *new* metadata blocks, which are
548 * accounted for in sysfile_bitmap_blocks. root_el +
549 * prev. last_eb_blk + blocks along edge of tree.
550 * calc_symlink_credits passes because we just need 1
551 * credit for the dinode there. */
552 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
553
554 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
555 ocfs2_quota_trans_credits(sb);
556}
557
558static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
559{
560 int blocks = ocfs2_mknod_credits(sb, 0, 0);
561
562 /* links can be longer than one block so we may update many
563 * within our single allocated extent. */
564 blocks += ocfs2_clusters_to_blocks(sb, 1);
565
566 return blocks + ocfs2_quota_trans_credits(sb);
567}
568
569static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
570 unsigned int cpg)
571{
572 int blocks;
573 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
574 /* parent inode update + new block group header + bitmap inode update
575 + bitmap blocks affected */
576 blocks = 1 + 1 + 1 + bitmap_blocks;
577 return blocks;
578}
579
580/*
581 * Allocating a discontiguous block group requires the credits from
582 * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
583 * the group descriptor's extent list. The caller already has started
584 * the transaction with ocfs2_calc_group_alloc_credits(). They extend
585 * it with these credits.
586 */
587static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
588{
589 return ocfs2_extent_recs_per_gd(sb);
590}
591
592static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
593 unsigned int clusters_to_del,
594 struct ocfs2_dinode *fe,
595 struct ocfs2_extent_list *last_el)
596{
597 /* for dinode + all headers in this pass + update to next leaf */
598 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
599 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
600 int credits = 1 + tree_depth + 1;
601 int i;
602
603 i = next_free - 1;
604 BUG_ON(i < 0);
605
606 /* We may be deleting metadata blocks, so metadata alloc dinode +
607 one desc. block for each possible delete. */
608 if (tree_depth && next_free == 1 &&
609 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
610 credits += 1 + tree_depth;
611
612 /* update to the truncate log. */
613 credits += OCFS2_TRUNCATE_LOG_UPDATE;
614
615 credits += ocfs2_quota_trans_credits(sb);
616
617 return credits;
618}
619
620static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
621{
622 return jbd2_journal_inode_add_write(handle, &OCFS2_I(inode)->ip_jinode);
623}
624
625static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
626 loff_t new_size)
627{
628 return jbd2_journal_begin_ordered_truncate(
629 OCFS2_SB(inode->i_sb)->journal->j_journal,
630 &OCFS2_I(inode)->ip_jinode,
631 new_size);
632}
633
634static inline void ocfs2_update_inode_fsync_trans(handle_t *handle,
635 struct inode *inode,
636 int datasync)
637{
638 struct ocfs2_inode_info *oi = OCFS2_I(inode);
639
640 oi->i_sync_tid = handle->h_transaction->t_tid;
641 if (datasync)
642 oi->i_datasync_tid = handle->h_transaction->t_tid;
643}
644
645#endif /* OCFS2_JOURNAL_H */