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 | |
16 | enum ocfs2_journal_state { |
17 | OCFS2_JOURNAL_FREE = 0, |
18 | OCFS2_JOURNAL_LOADED, |
19 | OCFS2_JOURNAL_IN_SHUTDOWN, |
20 | }; |
21 | |
22 | struct ocfs2_super; |
23 | struct 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 | |
30 | struct ocfs2_recovery_map { |
31 | unsigned int rm_used; |
32 | unsigned int rm_entries[]; |
33 | }; |
34 | |
35 | |
36 | struct 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 | |
64 | extern spinlock_t trans_inc_lock; |
65 | |
66 | /* wrap j_trans_id so we never have it equal to zero. */ |
67 | static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j) |
68 | { |
69 | unsigned long old_id; |
70 | spin_lock(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(lock: &trans_inc_lock); |
75 | return old_id; |
76 | } |
77 | |
78 | static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal, |
79 | struct ocfs2_caching_info *ci) |
80 | { |
81 | spin_lock(lock: &trans_inc_lock); |
82 | ci->ci_last_trans = journal->j_trans_id; |
83 | spin_unlock(lock: &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. */ |
91 | static 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(lock: &trans_inc_lock); |
98 | ret = time_after(journal->j_trans_id, ci->ci_last_trans); |
99 | spin_unlock(lock: &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. */ |
107 | static 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(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(lock: &trans_inc_lock); |
118 | return ret; |
119 | } |
120 | |
121 | /* Wrapper for inodes so we can check system files */ |
122 | static 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(ci: INODE_CACHE(inode)); |
131 | } |
132 | |
133 | static inline void ocfs2_ci_set_new(struct ocfs2_super *osb, |
134 | struct ocfs2_caching_info *ci) |
135 | { |
136 | spin_lock(lock: &trans_inc_lock); |
137 | ci->ci_created_trans = osb->journal->j_trans_id; |
138 | spin_unlock(lock: &trans_inc_lock); |
139 | } |
140 | |
141 | /* Exported only for the journal struct init code in super.c. Do not call. */ |
142 | void ocfs2_orphan_scan_init(struct ocfs2_super *osb); |
143 | void ocfs2_orphan_scan_start(struct ocfs2_super *osb); |
144 | void ocfs2_orphan_scan_stop(struct ocfs2_super *osb); |
145 | |
146 | void ocfs2_complete_recovery(struct work_struct *work); |
147 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb); |
148 | |
149 | int ocfs2_recovery_init(struct ocfs2_super *osb); |
150 | void ocfs2_recovery_exit(struct ocfs2_super *osb); |
151 | |
152 | int ocfs2_compute_replay_slots(struct ocfs2_super *osb); |
153 | void 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 | */ |
171 | void ocfs2_set_journal_params(struct ocfs2_super *osb); |
172 | int ocfs2_journal_alloc(struct ocfs2_super *osb); |
173 | int ocfs2_journal_init(struct ocfs2_super *osb, int *dirty); |
174 | void ocfs2_journal_shutdown(struct ocfs2_super *osb); |
175 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, |
176 | int full); |
177 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, |
178 | int replayed); |
179 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb); |
180 | void ocfs2_recovery_thread(struct ocfs2_super *osb, |
181 | int node_num); |
182 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb); |
183 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb); |
184 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb); |
185 | |
186 | static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb) |
187 | { |
188 | wake_up(&osb->checkpoint_event); |
189 | } |
190 | |
191 | static 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(ci: 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. */ |
241 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, |
242 | int max_buffs); |
243 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
244 | handle_t *handle); |
245 | int ocfs2_extend_trans(handle_t *handle, int nblocks); |
246 | int ocfs2_allocate_extend_trans(handle_t *handle, |
247 | int thresh); |
248 | |
249 | /* |
250 | * Define an arbitrary limit for the amount of data we will anticipate |
251 | * writing to any given transaction. For unbounded transactions such as |
252 | * fallocate(2) we can write more than this, but we always |
253 | * start off at the maximum transaction size and grow the transaction |
254 | * optimistically as we go. |
255 | */ |
256 | #define OCFS2_MAX_TRANS_DATA 64U |
257 | |
258 | /* |
259 | * Create access is for when we get a newly created buffer and we're |
260 | * not gonna read it off disk, but rather fill it ourselves. Right |
261 | * now, we don't do anything special with this (it turns into a write |
262 | * request), but this is a good placeholder in case we do... |
263 | * |
264 | * Write access is for when we read a block off disk and are going to |
265 | * modify it. This way the journalling layer knows it may need to make |
266 | * a copy of that block (if it's part of another, uncommitted |
267 | * transaction) before we do so. |
268 | */ |
269 | #define OCFS2_JOURNAL_ACCESS_CREATE 0 |
270 | #define OCFS2_JOURNAL_ACCESS_WRITE 1 |
271 | #define OCFS2_JOURNAL_ACCESS_UNDO 2 |
272 | |
273 | |
274 | /* ocfs2_inode */ |
275 | int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, |
276 | struct buffer_head *bh, int type); |
277 | /* ocfs2_extent_block */ |
278 | int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, |
279 | struct buffer_head *bh, int type); |
280 | /* ocfs2_refcount_block */ |
281 | int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, |
282 | struct buffer_head *bh, int type); |
283 | /* ocfs2_group_desc */ |
284 | int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, |
285 | struct buffer_head *bh, int type); |
286 | /* ocfs2_xattr_block */ |
287 | int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, |
288 | struct buffer_head *bh, int type); |
289 | /* quota blocks */ |
290 | int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, |
291 | struct buffer_head *bh, int type); |
292 | /* dirblock */ |
293 | int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, |
294 | struct buffer_head *bh, int type); |
295 | /* ocfs2_dx_root_block */ |
296 | int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, |
297 | struct buffer_head *bh, int type); |
298 | /* ocfs2_dx_leaf */ |
299 | int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, |
300 | struct buffer_head *bh, int type); |
301 | /* Anything that has no ecc */ |
302 | int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, |
303 | struct buffer_head *bh, int type); |
304 | |
305 | /* |
306 | * A word about the journal_access/journal_dirty "dance". It is |
307 | * entirely legal to journal_access a buffer more than once (as long |
308 | * as the access type is the same -- I'm not sure what will happen if |
309 | * access type is different but this should never happen anyway) It is |
310 | * also legal to journal_dirty a buffer more than once. In fact, you |
311 | * can even journal_access a buffer after you've done a |
312 | * journal_access/journal_dirty pair. The only thing you cannot do |
313 | * however, is journal_dirty a buffer which you haven't yet passed to |
314 | * journal_access at least once. |
315 | * |
316 | * That said, 99% of the time this doesn't matter and this is what the |
317 | * path looks like: |
318 | * |
319 | * <read a bh> |
320 | * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE); |
321 | * <modify the bh> |
322 | * ocfs2_journal_dirty(handle, bh); |
323 | */ |
324 | void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh); |
325 | |
326 | /* |
327 | * Credit Macros: |
328 | * Convenience macros to calculate number of credits needed. |
329 | * |
330 | * For convenience sake, I have a set of macros here which calculate |
331 | * the *maximum* number of sectors which will be changed for various |
332 | * metadata updates. |
333 | */ |
334 | |
335 | /* simple file updates like chmod, etc. */ |
336 | #define OCFS2_INODE_UPDATE_CREDITS 1 |
337 | |
338 | /* extended attribute block update */ |
339 | #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1 |
340 | |
341 | /* Update of a single quota block */ |
342 | #define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1 |
343 | |
344 | /* global quotafile inode update, data block */ |
345 | #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \ |
346 | OCFS2_QUOTA_BLOCK_UPDATE_CREDITS) |
347 | |
348 | #define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS |
349 | /* |
350 | * The two writes below can accidentally see global info dirty due |
351 | * to set_info() quotactl so make them prepared for the writes. |
352 | */ |
353 | /* quota data block, global info */ |
354 | /* Write to local quota file */ |
355 | #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \ |
356 | OCFS2_QUOTA_BLOCK_UPDATE_CREDITS) |
357 | |
358 | /* global quota data block, local quota data block, global quota inode, |
359 | * global quota info */ |
360 | #define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \ |
361 | 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS) |
362 | |
363 | static inline int ocfs2_quota_trans_credits(struct super_block *sb) |
364 | { |
365 | int credits = 0; |
366 | |
367 | if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) |
368 | credits += OCFS2_QWRITE_CREDITS; |
369 | if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) |
370 | credits += OCFS2_QWRITE_CREDITS; |
371 | return credits; |
372 | } |
373 | |
374 | /* group extend. inode update and last group update. */ |
375 | #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1) |
376 | |
377 | /* group add. inode update and the new group update. */ |
378 | #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1) |
379 | |
380 | /* get one bit out of a suballocator: dinode + group descriptor + |
381 | * prev. group desc. if we relink. */ |
382 | #define OCFS2_SUBALLOC_ALLOC (3) |
383 | |
384 | static inline int ocfs2_inline_to_extents_credits(struct super_block *sb) |
385 | { |
386 | return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS + |
387 | ocfs2_quota_trans_credits(sb); |
388 | } |
389 | |
390 | /* dinode + group descriptor update. We don't relink on free yet. */ |
391 | #define OCFS2_SUBALLOC_FREE (2) |
392 | |
393 | #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS |
394 | #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \ |
395 | + OCFS2_TRUNCATE_LOG_UPDATE) |
396 | |
397 | static inline int ocfs2_remove_extent_credits(struct super_block *sb) |
398 | { |
399 | return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS + |
400 | ocfs2_quota_trans_credits(sb); |
401 | } |
402 | |
403 | /* data block for new dir/symlink, allocation of directory block, dx_root |
404 | * update for free list */ |
405 | #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1) |
406 | |
407 | static inline int ocfs2_add_dir_index_credits(struct super_block *sb) |
408 | { |
409 | /* 1 block for index, 2 allocs (data, metadata), 1 clusters |
410 | * worth of blocks for initial extent. */ |
411 | return 1 + 2 * OCFS2_SUBALLOC_ALLOC + |
412 | ocfs2_clusters_to_blocks(sb, clusters: 1); |
413 | } |
414 | |
415 | /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode |
416 | * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr |
417 | * blocks + quota update */ |
418 | static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir, |
419 | int xattr_credits) |
420 | { |
421 | int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS; |
422 | |
423 | if (is_dir) |
424 | dir_credits += ocfs2_add_dir_index_credits(sb); |
425 | |
426 | return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits + |
427 | ocfs2_quota_trans_credits(sb); |
428 | } |
429 | |
430 | /* local alloc metadata change + main bitmap updates */ |
431 | #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \ |
432 | + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE) |
433 | |
434 | /* used when we don't need an allocation change for a dir extend. One |
435 | * for the dinode, one for the new block. */ |
436 | #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2) |
437 | |
438 | /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota |
439 | * update on dir + index leaf + dx root update for free list + |
440 | * previous dirblock update in the free list */ |
441 | static inline int ocfs2_link_credits(struct super_block *sb) |
442 | { |
443 | return 2 * OCFS2_INODE_UPDATE_CREDITS + 4 + |
444 | ocfs2_quota_trans_credits(sb); |
445 | } |
446 | |
447 | /* inode + dir inode (if we unlink a dir), + dir entry block + orphan |
448 | * dir inode link + dir inode index leaf + dir index root */ |
449 | static inline int ocfs2_unlink_credits(struct super_block *sb) |
450 | { |
451 | /* The quota update from ocfs2_link_credits is unused here... */ |
452 | return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb); |
453 | } |
454 | |
455 | /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry + |
456 | * inode alloc group descriptor + orphan dir index root + |
457 | * orphan dir index leaf */ |
458 | #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4) |
459 | |
460 | /* dinode + orphan dir dinode + extent tree leaf block + orphan dir entry + |
461 | * orphan dir index root + orphan dir index leaf */ |
462 | #define OCFS2_INODE_ADD_TO_ORPHAN_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 4) |
463 | #define OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS OCFS2_INODE_ADD_TO_ORPHAN_CREDITS |
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 */ |
468 | static 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 | |
484 | static 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, clusters: 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 | */ |
516 | static inline int ocfs2_calc_extend_credits(struct super_block *sb, |
517 | struct ocfs2_extent_list *root_el) |
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 | |
543 | static inline int ocfs2_calc_symlink_credits(struct super_block *sb) |
544 | { |
545 | int blocks = ocfs2_mknod_credits(sb, is_dir: 0, xattr_credits: 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, clusters: 1); |
550 | |
551 | return blocks + ocfs2_quota_trans_credits(sb); |
552 | } |
553 | |
554 | static 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 | */ |
572 | static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb) |
573 | { |
574 | return ocfs2_extent_recs_per_gd(sb); |
575 | } |
576 | |
577 | static inline int ocfs2_jbd2_inode_add_write(handle_t *handle, struct inode *inode, |
578 | loff_t start_byte, loff_t length) |
579 | { |
580 | return jbd2_journal_inode_ranged_write(handle, |
581 | inode: &OCFS2_I(inode)->ip_jinode, |
582 | start_byte, length); |
583 | } |
584 | |
585 | static inline int ocfs2_begin_ordered_truncate(struct inode *inode, |
586 | loff_t new_size) |
587 | { |
588 | return jbd2_journal_begin_ordered_truncate( |
589 | OCFS2_SB(inode->i_sb)->journal->j_journal, |
590 | inode: &OCFS2_I(inode)->ip_jinode, |
591 | new_size); |
592 | } |
593 | |
594 | static inline void ocfs2_update_inode_fsync_trans(handle_t *handle, |
595 | struct inode *inode, |
596 | int datasync) |
597 | { |
598 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
599 | |
600 | if (!is_handle_aborted(handle)) { |
601 | oi->i_sync_tid = handle->h_transaction->t_tid; |
602 | if (datasync) |
603 | oi->i_datasync_tid = handle->h_transaction->t_tid; |
604 | } |
605 | } |
606 | |
607 | #endif /* OCFS2_JOURNAL_H */ |
608 | |