Loading...
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 atomic_set(&osb->needs_checkpoint, 1);
204 wake_up(&osb->checkpoint_event);
205}
206
207static inline void ocfs2_checkpoint_inode(struct inode *inode)
208{
209 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
210
211 if (ocfs2_mount_local(osb))
212 return;
213
214 if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
215 /* WARNING: This only kicks off a single
216 * checkpoint. If someone races you and adds more
217 * metadata to the journal, you won't know, and will
218 * wind up waiting *a lot* longer than necessary. Right
219 * now we only use this in clear_inode so that's
220 * OK. */
221 ocfs2_start_checkpoint(osb);
222
223 wait_event(osb->journal->j_checkpointed,
224 ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
225 }
226}
227
228/*
229 * Transaction Handling:
230 * Manage the lifetime of a transaction handle.
231 *
232 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
233 * the number of blocks that will be changed during
234 * this handle.
235 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
236 * the journal was aborted. The majority of paths don't
237 * check the return value as an error there comes too
238 * late to do anything (and will be picked up in a
239 * later transaction).
240 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
241 * commit the handle to disk in the process, but will
242 * not release any locks taken during the transaction.
243 * ocfs2_journal_access* - Notify the handle that we want to journal this
244 * buffer. Will have to call ocfs2_journal_dirty once
245 * we've actually dirtied it. Type is one of . or .
246 * Always call the specific flavor of
247 * ocfs2_journal_access_*() unless you intend to
248 * manage the checksum by hand.
249 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
250 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
251 * the current handle commits.
252 */
253
254/* You must always start_trans with a number of buffs > 0, but it's
255 * perfectly legal to go through an entire transaction without having
256 * dirtied any buffers. */
257handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
258 int max_buffs);
259int ocfs2_commit_trans(struct ocfs2_super *osb,
260 handle_t *handle);
261int ocfs2_extend_trans(handle_t *handle, int nblocks);
262
263/*
264 * Create access is for when we get a newly created buffer and we're
265 * not gonna read it off disk, but rather fill it ourselves. Right
266 * now, we don't do anything special with this (it turns into a write
267 * request), but this is a good placeholder in case we do...
268 *
269 * Write access is for when we read a block off disk and are going to
270 * modify it. This way the journalling layer knows it may need to make
271 * a copy of that block (if it's part of another, uncommitted
272 * transaction) before we do so.
273 */
274#define OCFS2_JOURNAL_ACCESS_CREATE 0
275#define OCFS2_JOURNAL_ACCESS_WRITE 1
276#define OCFS2_JOURNAL_ACCESS_UNDO 2
277
278
279/* ocfs2_inode */
280int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
281 struct buffer_head *bh, int type);
282/* ocfs2_extent_block */
283int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
284 struct buffer_head *bh, int type);
285/* ocfs2_refcount_block */
286int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
287 struct buffer_head *bh, int type);
288/* ocfs2_group_desc */
289int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
290 struct buffer_head *bh, int type);
291/* ocfs2_xattr_block */
292int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
293 struct buffer_head *bh, int type);
294/* quota blocks */
295int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
296 struct buffer_head *bh, int type);
297/* dirblock */
298int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
299 struct buffer_head *bh, int type);
300/* ocfs2_dx_root_block */
301int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
302 struct buffer_head *bh, int type);
303/* ocfs2_dx_leaf */
304int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
305 struct buffer_head *bh, int type);
306/* Anything that has no ecc */
307int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
308 struct buffer_head *bh, int type);
309
310/*
311 * A word about the journal_access/journal_dirty "dance". It is
312 * entirely legal to journal_access a buffer more than once (as long
313 * as the access type is the same -- I'm not sure what will happen if
314 * access type is different but this should never happen anyway) It is
315 * also legal to journal_dirty a buffer more than once. In fact, you
316 * can even journal_access a buffer after you've done a
317 * journal_access/journal_dirty pair. The only thing you cannot do
318 * however, is journal_dirty a buffer which you haven't yet passed to
319 * journal_access at least once.
320 *
321 * That said, 99% of the time this doesn't matter and this is what the
322 * path looks like:
323 *
324 * <read a bh>
325 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
326 * <modify the bh>
327 * ocfs2_journal_dirty(handle, bh);
328 */
329void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
330
331/*
332 * Credit Macros:
333 * Convenience macros to calculate number of credits needed.
334 *
335 * For convenience sake, I have a set of macros here which calculate
336 * the *maximum* number of sectors which will be changed for various
337 * metadata updates.
338 */
339
340/* simple file updates like chmod, etc. */
341#define OCFS2_INODE_UPDATE_CREDITS 1
342
343/* extended attribute block update */
344#define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
345
346/* Update of a single quota block */
347#define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
348
349/* global quotafile inode update, data block */
350#define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
351 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
352
353#define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
354/*
355 * The two writes below can accidentally see global info dirty due
356 * to set_info() quotactl so make them prepared for the writes.
357 */
358/* quota data block, global info */
359/* Write to local quota file */
360#define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
361 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
362
363/* global quota data block, local quota data block, global quota inode,
364 * global quota info */
365#define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
366 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
367
368static inline int ocfs2_quota_trans_credits(struct super_block *sb)
369{
370 int credits = 0;
371
372 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
373 credits += OCFS2_QWRITE_CREDITS;
374 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
375 credits += OCFS2_QWRITE_CREDITS;
376 return credits;
377}
378
379/* group extend. inode update and last group update. */
380#define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
381
382/* group add. inode update and the new group update. */
383#define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
384
385/* get one bit out of a suballocator: dinode + group descriptor +
386 * prev. group desc. if we relink. */
387#define OCFS2_SUBALLOC_ALLOC (3)
388
389static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
390{
391 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
392 ocfs2_quota_trans_credits(sb);
393}
394
395/* dinode + group descriptor update. We don't relink on free yet. */
396#define OCFS2_SUBALLOC_FREE (2)
397
398#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
399#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
400 + OCFS2_TRUNCATE_LOG_UPDATE)
401
402static inline int ocfs2_remove_extent_credits(struct super_block *sb)
403{
404 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
405 ocfs2_quota_trans_credits(sb);
406}
407
408/* data block for new dir/symlink, allocation of directory block, dx_root
409 * update for free list */
410#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
411
412static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
413{
414 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
415 * worth of blocks for initial extent. */
416 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
417 ocfs2_clusters_to_blocks(sb, 1);
418}
419
420/* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
421 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
422 * blocks + quota update */
423static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
424 int xattr_credits)
425{
426 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
427
428 if (is_dir)
429 dir_credits += ocfs2_add_dir_index_credits(sb);
430
431 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
432 ocfs2_quota_trans_credits(sb);
433}
434
435/* local alloc metadata change + main bitmap updates */
436#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
437 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
438
439/* used when we don't need an allocation change for a dir extend. One
440 * for the dinode, one for the new block. */
441#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
442
443/* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
444 * update on dir + index leaf + dx root update for free list +
445 * previous dirblock update in the free list */
446static inline int ocfs2_link_credits(struct super_block *sb)
447{
448 return 2*OCFS2_INODE_UPDATE_CREDITS + 4 +
449 ocfs2_quota_trans_credits(sb);
450}
451
452/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
453 * dir inode link + dir inode index leaf + dir index root */
454static inline int ocfs2_unlink_credits(struct super_block *sb)
455{
456 /* The quota update from ocfs2_link_credits is unused here... */
457 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
458}
459
460/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
461 * inode alloc group descriptor + orphan dir index root +
462 * orphan dir index leaf */
463#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
464
465/* dinode update, old dir dinode update, new dir dinode update, old
466 * dir dir entry, new dir dir entry, dir entry update for renaming
467 * directory + target unlink + 3 x dir index leaves */
468static inline int ocfs2_rename_credits(struct super_block *sb)
469{
470 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
471}
472
473/* global bitmap dinode, group desc., relinked group,
474 * suballocator dinode, group desc., relinked group,
475 * dinode, xattr block */
476#define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
477 + OCFS2_INODE_UPDATE_CREDITS \
478 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
479
480/* inode update, removal of dx root block from allocator */
481#define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
482 OCFS2_SUBALLOC_FREE)
483
484static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
485{
486 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
487
488 credits += ocfs2_clusters_to_blocks(sb, 1);
489 credits += ocfs2_quota_trans_credits(sb);
490
491 return credits;
492}
493
494/* inode update, new refcount block and its allocation credits. */
495#define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
496 + OCFS2_SUBALLOC_ALLOC)
497
498/* inode and the refcount block update. */
499#define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
500
501/*
502 * inode and the refcount block update.
503 * It doesn't include the credits for sub alloc change.
504 * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
505 */
506#define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
507
508/* 2 metadata alloc, 2 new blocks and root refcount block */
509#define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
510
511/*
512 * Please note that the caller must make sure that root_el is the root
513 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
514 * the result may be wrong.
515 */
516static inline int ocfs2_calc_extend_credits(struct super_block *sb,
517 struct ocfs2_extent_list *root_el,
518 u32 bits_wanted)
519{
520 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
521
522 /* bitmap dinode, group desc. + relinked group. */
523 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
524
525 /* we might need to shift tree depth so lets assume an
526 * absolute worst case of complete fragmentation. Even with
527 * that, we only need one update for the dinode, and then
528 * however many metadata chunks needed * a remaining suballoc
529 * alloc. */
530 sysfile_bitmap_blocks = 1 +
531 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
532
533 /* this does not include *new* metadata blocks, which are
534 * accounted for in sysfile_bitmap_blocks. root_el +
535 * prev. last_eb_blk + blocks along edge of tree.
536 * calc_symlink_credits passes because we just need 1
537 * credit for the dinode there. */
538 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
539
540 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
541 ocfs2_quota_trans_credits(sb);
542}
543
544static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
545{
546 int blocks = ocfs2_mknod_credits(sb, 0, 0);
547
548 /* links can be longer than one block so we may update many
549 * within our single allocated extent. */
550 blocks += ocfs2_clusters_to_blocks(sb, 1);
551
552 return blocks + ocfs2_quota_trans_credits(sb);
553}
554
555static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
556 unsigned int cpg)
557{
558 int blocks;
559 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
560 /* parent inode update + new block group header + bitmap inode update
561 + bitmap blocks affected */
562 blocks = 1 + 1 + 1 + bitmap_blocks;
563 return blocks;
564}
565
566/*
567 * Allocating a discontiguous block group requires the credits from
568 * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
569 * the group descriptor's extent list. The caller already has started
570 * the transaction with ocfs2_calc_group_alloc_credits(). They extend
571 * it with these credits.
572 */
573static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
574{
575 return ocfs2_extent_recs_per_gd(sb);
576}
577
578static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
579 unsigned int clusters_to_del,
580 struct ocfs2_dinode *fe,
581 struct ocfs2_extent_list *last_el)
582{
583 /* for dinode + all headers in this pass + update to next leaf */
584 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
585 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
586 int credits = 1 + tree_depth + 1;
587 int i;
588
589 i = next_free - 1;
590 BUG_ON(i < 0);
591
592 /* We may be deleting metadata blocks, so metadata alloc dinode +
593 one desc. block for each possible delete. */
594 if (tree_depth && next_free == 1 &&
595 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
596 credits += 1 + tree_depth;
597
598 /* update to the truncate log. */
599 credits += OCFS2_TRUNCATE_LOG_UPDATE;
600
601 credits += ocfs2_quota_trans_credits(sb);
602
603 return credits;
604}
605
606static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
607{
608 return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
609}
610
611static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
612 loff_t new_size)
613{
614 return jbd2_journal_begin_ordered_truncate(
615 OCFS2_SB(inode->i_sb)->journal->j_journal,
616 &OCFS2_I(inode)->ip_jinode,
617 new_size);
618}
619
620#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 atomic_set(&osb->needs_checkpoint, 1);
204 wake_up(&osb->checkpoint_event);
205}
206
207static inline void ocfs2_checkpoint_inode(struct inode *inode)
208{
209 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
210
211 if (ocfs2_mount_local(osb))
212 return;
213
214 if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
215 /* WARNING: This only kicks off a single
216 * checkpoint. If someone races you and adds more
217 * metadata to the journal, you won't know, and will
218 * wind up waiting *a lot* longer than necessary. Right
219 * now we only use this in clear_inode so that's
220 * OK. */
221 ocfs2_start_checkpoint(osb);
222
223 wait_event(osb->journal->j_checkpointed,
224 ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
225 }
226}
227
228/*
229 * Transaction Handling:
230 * Manage the lifetime of a transaction handle.
231 *
232 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
233 * the number of blocks that will be changed during
234 * this handle.
235 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
236 * the journal was aborted. The majority of paths don't
237 * check the return value as an error there comes too
238 * late to do anything (and will be picked up in a
239 * later transaction).
240 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
241 * commit the handle to disk in the process, but will
242 * not release any locks taken during the transaction.
243 * ocfs2_journal_access* - Notify the handle that we want to journal this
244 * buffer. Will have to call ocfs2_journal_dirty once
245 * we've actually dirtied it. Type is one of . or .
246 * Always call the specific flavor of
247 * ocfs2_journal_access_*() unless you intend to
248 * manage the checksum by hand.
249 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
250 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
251 * the current handle commits.
252 */
253
254/* You must always start_trans with a number of buffs > 0, but it's
255 * perfectly legal to go through an entire transaction without having
256 * dirtied any buffers. */
257handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
258 int max_buffs);
259int ocfs2_commit_trans(struct ocfs2_super *osb,
260 handle_t *handle);
261int ocfs2_extend_trans(handle_t *handle, int nblocks);
262
263/*
264 * Create access is for when we get a newly created buffer and we're
265 * not gonna read it off disk, but rather fill it ourselves. Right
266 * now, we don't do anything special with this (it turns into a write
267 * request), but this is a good placeholder in case we do...
268 *
269 * Write access is for when we read a block off disk and are going to
270 * modify it. This way the journalling layer knows it may need to make
271 * a copy of that block (if it's part of another, uncommitted
272 * transaction) before we do so.
273 */
274#define OCFS2_JOURNAL_ACCESS_CREATE 0
275#define OCFS2_JOURNAL_ACCESS_WRITE 1
276#define OCFS2_JOURNAL_ACCESS_UNDO 2
277
278
279/* ocfs2_inode */
280int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
281 struct buffer_head *bh, int type);
282/* ocfs2_extent_block */
283int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
284 struct buffer_head *bh, int type);
285/* ocfs2_refcount_block */
286int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
287 struct buffer_head *bh, int type);
288/* ocfs2_group_desc */
289int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
290 struct buffer_head *bh, int type);
291/* ocfs2_xattr_block */
292int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
293 struct buffer_head *bh, int type);
294/* quota blocks */
295int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
296 struct buffer_head *bh, int type);
297/* dirblock */
298int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
299 struct buffer_head *bh, int type);
300/* ocfs2_dx_root_block */
301int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
302 struct buffer_head *bh, int type);
303/* ocfs2_dx_leaf */
304int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
305 struct buffer_head *bh, int type);
306/* Anything that has no ecc */
307int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
308 struct buffer_head *bh, int type);
309
310/*
311 * A word about the journal_access/journal_dirty "dance". It is
312 * entirely legal to journal_access a buffer more than once (as long
313 * as the access type is the same -- I'm not sure what will happen if
314 * access type is different but this should never happen anyway) It is
315 * also legal to journal_dirty a buffer more than once. In fact, you
316 * can even journal_access a buffer after you've done a
317 * journal_access/journal_dirty pair. The only thing you cannot do
318 * however, is journal_dirty a buffer which you haven't yet passed to
319 * journal_access at least once.
320 *
321 * That said, 99% of the time this doesn't matter and this is what the
322 * path looks like:
323 *
324 * <read a bh>
325 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
326 * <modify the bh>
327 * ocfs2_journal_dirty(handle, bh);
328 */
329void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
330
331/*
332 * Credit Macros:
333 * Convenience macros to calculate number of credits needed.
334 *
335 * For convenience sake, I have a set of macros here which calculate
336 * the *maximum* number of sectors which will be changed for various
337 * metadata updates.
338 */
339
340/* simple file updates like chmod, etc. */
341#define OCFS2_INODE_UPDATE_CREDITS 1
342
343/* extended attribute block update */
344#define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
345
346/* Update of a single quota block */
347#define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
348
349/* global quotafile inode update, data block */
350#define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
351 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
352
353#define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
354/*
355 * The two writes below can accidentally see global info dirty due
356 * to set_info() quotactl so make them prepared for the writes.
357 */
358/* quota data block, global info */
359/* Write to local quota file */
360#define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
361 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
362
363/* global quota data block, local quota data block, global quota inode,
364 * global quota info */
365#define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
366 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
367
368static inline int ocfs2_quota_trans_credits(struct super_block *sb)
369{
370 int credits = 0;
371
372 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
373 credits += OCFS2_QWRITE_CREDITS;
374 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
375 credits += OCFS2_QWRITE_CREDITS;
376 return credits;
377}
378
379/* group extend. inode update and last group update. */
380#define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
381
382/* group add. inode update and the new group update. */
383#define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
384
385/* get one bit out of a suballocator: dinode + group descriptor +
386 * prev. group desc. if we relink. */
387#define OCFS2_SUBALLOC_ALLOC (3)
388
389static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
390{
391 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
392 ocfs2_quota_trans_credits(sb);
393}
394
395/* dinode + group descriptor update. We don't relink on free yet. */
396#define OCFS2_SUBALLOC_FREE (2)
397
398#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
399#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
400 + OCFS2_TRUNCATE_LOG_UPDATE)
401
402static inline int ocfs2_remove_extent_credits(struct super_block *sb)
403{
404 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
405 ocfs2_quota_trans_credits(sb);
406}
407
408/* data block for new dir/symlink, allocation of directory block, dx_root
409 * update for free list */
410#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
411
412static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
413{
414 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
415 * worth of blocks for initial extent. */
416 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
417 ocfs2_clusters_to_blocks(sb, 1);
418}
419
420/* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
421 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
422 * blocks + quota update */
423static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
424 int xattr_credits)
425{
426 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
427
428 if (is_dir)
429 dir_credits += ocfs2_add_dir_index_credits(sb);
430
431 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
432 ocfs2_quota_trans_credits(sb);
433}
434
435/* local alloc metadata change + main bitmap updates */
436#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
437 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
438
439/* used when we don't need an allocation change for a dir extend. One
440 * for the dinode, one for the new block. */
441#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
442
443/* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
444 * update on dir + index leaf + dx root update for free list */
445static inline int ocfs2_link_credits(struct super_block *sb)
446{
447 return 2*OCFS2_INODE_UPDATE_CREDITS + 3 +
448 ocfs2_quota_trans_credits(sb);
449}
450
451/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
452 * dir inode link + dir inode index leaf + dir index root */
453static inline int ocfs2_unlink_credits(struct super_block *sb)
454{
455 /* The quota update from ocfs2_link_credits is unused here... */
456 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
457}
458
459/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
460 * inode alloc group descriptor + orphan dir index root +
461 * orphan dir index leaf */
462#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
463
464/* dinode update, old dir dinode update, new dir dinode update, old
465 * dir dir entry, new dir dir entry, dir entry update for renaming
466 * directory + target unlink + 3 x dir index leaves */
467static inline int ocfs2_rename_credits(struct super_block *sb)
468{
469 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
470}
471
472/* global bitmap dinode, group desc., relinked group,
473 * suballocator dinode, group desc., relinked group,
474 * dinode, xattr block */
475#define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
476 + OCFS2_INODE_UPDATE_CREDITS \
477 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
478
479/* inode update, removal of dx root block from allocator */
480#define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
481 OCFS2_SUBALLOC_FREE)
482
483static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
484{
485 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
486
487 credits += ocfs2_clusters_to_blocks(sb, 1);
488 credits += ocfs2_quota_trans_credits(sb);
489
490 return credits;
491}
492
493/* inode update, new refcount block and its allocation credits. */
494#define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
495 + OCFS2_SUBALLOC_ALLOC)
496
497/* inode and the refcount block update. */
498#define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
499
500/*
501 * inode and the refcount block update.
502 * It doesn't include the credits for sub alloc change.
503 * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
504 */
505#define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
506
507/* 2 metadata alloc, 2 new blocks and root refcount block */
508#define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
509
510/*
511 * Please note that the caller must make sure that root_el is the root
512 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
513 * the result may be wrong.
514 */
515static inline int ocfs2_calc_extend_credits(struct super_block *sb,
516 struct ocfs2_extent_list *root_el,
517 u32 bits_wanted)
518{
519 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
520
521 /* bitmap dinode, group desc. + relinked group. */
522 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
523
524 /* we might need to shift tree depth so lets assume an
525 * absolute worst case of complete fragmentation. Even with
526 * that, we only need one update for the dinode, and then
527 * however many metadata chunks needed * a remaining suballoc
528 * alloc. */
529 sysfile_bitmap_blocks = 1 +
530 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
531
532 /* this does not include *new* metadata blocks, which are
533 * accounted for in sysfile_bitmap_blocks. root_el +
534 * prev. last_eb_blk + blocks along edge of tree.
535 * calc_symlink_credits passes because we just need 1
536 * credit for the dinode there. */
537 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
538
539 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
540 ocfs2_quota_trans_credits(sb);
541}
542
543static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
544{
545 int blocks = ocfs2_mknod_credits(sb, 0, 0);
546
547 /* links can be longer than one block so we may update many
548 * within our single allocated extent. */
549 blocks += ocfs2_clusters_to_blocks(sb, 1);
550
551 return blocks + ocfs2_quota_trans_credits(sb);
552}
553
554static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
555 unsigned int cpg)
556{
557 int blocks;
558 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
559 /* parent inode update + new block group header + bitmap inode update
560 + bitmap blocks affected */
561 blocks = 1 + 1 + 1 + bitmap_blocks;
562 return blocks;
563}
564
565/*
566 * Allocating a discontiguous block group requires the credits from
567 * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
568 * the group descriptor's extent list. The caller already has started
569 * the transaction with ocfs2_calc_group_alloc_credits(). They extend
570 * it with these credits.
571 */
572static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
573{
574 return ocfs2_extent_recs_per_gd(sb);
575}
576
577static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
578 unsigned int clusters_to_del,
579 struct ocfs2_dinode *fe,
580 struct ocfs2_extent_list *last_el)
581{
582 /* for dinode + all headers in this pass + update to next leaf */
583 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
584 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
585 int credits = 1 + tree_depth + 1;
586 int i;
587
588 i = next_free - 1;
589 BUG_ON(i < 0);
590
591 /* We may be deleting metadata blocks, so metadata alloc dinode +
592 one desc. block for each possible delete. */
593 if (tree_depth && next_free == 1 &&
594 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
595 credits += 1 + tree_depth;
596
597 /* update to the truncate log. */
598 credits += OCFS2_TRUNCATE_LOG_UPDATE;
599
600 credits += ocfs2_quota_trans_credits(sb);
601
602 return credits;
603}
604
605static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
606{
607 return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
608}
609
610static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
611 loff_t new_size)
612{
613 return jbd2_journal_begin_ordered_truncate(
614 OCFS2_SB(inode->i_sb)->journal->j_journal,
615 &OCFS2_I(inode)->ip_jinode,
616 new_size);
617}
618
619#endif /* OCFS2_JOURNAL_H */