xref: /openbmc/linux/fs/ocfs2/journal.h (revision 6ea9a2b8)
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(&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 
78 static 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. */
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(&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. */
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(&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 */
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(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(&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. */
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 /*
154  *  Journal Control:
155  *  Initialize, Load, Shutdown, Wipe a journal.
156  *
157  *  ocfs2_journal_init     - Initialize journal structures in the OSB.
158  *  ocfs2_journal_load     - Load the given journal off disk. Replay it if
159  *                          there's transactions still in there.
160  *  ocfs2_journal_shutdown - Shutdown a journal, this will flush all
161  *                          uncommitted, uncheckpointed transactions.
162  *  ocfs2_journal_wipe     - Wipe transactions from a journal. Optionally
163  *                          zero out each block.
164  *  ocfs2_recovery_thread  - Perform recovery on a node. osb is our own osb.
165  *  ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
166  *                          event on.
167  *  ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
168  */
169 void   ocfs2_set_journal_params(struct ocfs2_super *osb);
170 int    ocfs2_journal_init(struct ocfs2_journal *journal,
171 			  int *dirty);
172 void   ocfs2_journal_shutdown(struct ocfs2_super *osb);
173 int    ocfs2_journal_wipe(struct ocfs2_journal *journal,
174 			  int full);
175 int    ocfs2_journal_load(struct ocfs2_journal *journal, int local,
176 			  int replayed);
177 int    ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
178 void   ocfs2_recovery_thread(struct ocfs2_super *osb,
179 			     int node_num);
180 int    ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
181 void   ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
182 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
183 
184 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
185 {
186 	wake_up(&osb->checkpoint_event);
187 }
188 
189 static inline void ocfs2_checkpoint_inode(struct inode *inode)
190 {
191 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
192 
193 	if (ocfs2_mount_local(osb))
194 		return;
195 
196 	if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
197 		/* WARNING: This only kicks off a single
198 		 * checkpoint. If someone races you and adds more
199 		 * metadata to the journal, you won't know, and will
200 		 * wind up waiting *a lot* longer than necessary. Right
201 		 * now we only use this in clear_inode so that's
202 		 * OK. */
203 		ocfs2_start_checkpoint(osb);
204 
205 		wait_event(osb->journal->j_checkpointed,
206 			   ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
207 	}
208 }
209 
210 /*
211  *  Transaction Handling:
212  *  Manage the lifetime of a transaction handle.
213  *
214  *  ocfs2_start_trans      - Begin a transaction. Give it an upper estimate of
215  *                          the number of blocks that will be changed during
216  *                          this handle.
217  *  ocfs2_commit_trans - Complete a handle. It might return -EIO if
218  *                       the journal was aborted. The majority of paths don't
219  *                       check the return value as an error there comes too
220  *                       late to do anything (and will be picked up in a
221  *                       later transaction).
222  *  ocfs2_extend_trans     - Extend a handle by nblocks credits. This may
223  *                          commit the handle to disk in the process, but will
224  *                          not release any locks taken during the transaction.
225  *  ocfs2_journal_access* - Notify the handle that we want to journal this
226  *                          buffer. Will have to call ocfs2_journal_dirty once
227  *                          we've actually dirtied it. Type is one of . or .
228  *                          Always call the specific flavor of
229  *                          ocfs2_journal_access_*() unless you intend to
230  *                          manage the checksum by hand.
231  *  ocfs2_journal_dirty    - Mark a journalled buffer as having dirty data.
232  *  ocfs2_jbd2_inode_add_write  - Mark an inode with range so that its data goes
233  *                                out before the current handle commits.
234  */
235 
236 /* You must always start_trans with a number of buffs > 0, but it's
237  * perfectly legal to go through an entire transaction without having
238  * dirtied any buffers. */
239 handle_t		    *ocfs2_start_trans(struct ocfs2_super *osb,
240 					       int max_buffs);
241 int			     ocfs2_commit_trans(struct ocfs2_super *osb,
242 						handle_t *handle);
243 int			     ocfs2_extend_trans(handle_t *handle, int nblocks);
244 int			     ocfs2_allocate_extend_trans(handle_t *handle,
245 						int thresh);
246 
247 /*
248  * Define an arbitrary limit for the amount of data we will anticipate
249  * writing to any given transaction.  For unbounded transactions such as
250  * fallocate(2) we can write more than this, but we always
251  * start off at the maximum transaction size and grow the transaction
252  * optimistically as we go.
253  */
254 #define OCFS2_MAX_TRANS_DATA	64U
255 
256 /*
257  * Create access is for when we get a newly created buffer and we're
258  * not gonna read it off disk, but rather fill it ourselves.  Right
259  * now, we don't do anything special with this (it turns into a write
260  * request), but this is a good placeholder in case we do...
261  *
262  * Write access is for when we read a block off disk and are going to
263  * modify it. This way the journalling layer knows it may need to make
264  * a copy of that block (if it's part of another, uncommitted
265  * transaction) before we do so.
266  */
267 #define OCFS2_JOURNAL_ACCESS_CREATE 0
268 #define OCFS2_JOURNAL_ACCESS_WRITE  1
269 #define OCFS2_JOURNAL_ACCESS_UNDO   2
270 
271 
272 /* ocfs2_inode */
273 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
274 			    struct buffer_head *bh, int type);
275 /* ocfs2_extent_block */
276 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
277 			    struct buffer_head *bh, int type);
278 /* ocfs2_refcount_block */
279 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
280 			    struct buffer_head *bh, int type);
281 /* ocfs2_group_desc */
282 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
283 			    struct buffer_head *bh, int type);
284 /* ocfs2_xattr_block */
285 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
286 			    struct buffer_head *bh, int type);
287 /* quota blocks */
288 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
289 			    struct buffer_head *bh, int type);
290 /* dirblock */
291 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
292 			    struct buffer_head *bh, int type);
293 /* ocfs2_dx_root_block */
294 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
295 			    struct buffer_head *bh, int type);
296 /* ocfs2_dx_leaf */
297 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
298 			    struct buffer_head *bh, int type);
299 /* Anything that has no ecc */
300 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
301 			 struct buffer_head *bh, int type);
302 
303 /*
304  * A word about the journal_access/journal_dirty "dance". It is
305  * entirely legal to journal_access a buffer more than once (as long
306  * as the access type is the same -- I'm not sure what will happen if
307  * access type is different but this should never happen anyway) It is
308  * also legal to journal_dirty a buffer more than once. In fact, you
309  * can even journal_access a buffer after you've done a
310  * journal_access/journal_dirty pair. The only thing you cannot do
311  * however, is journal_dirty a buffer which you haven't yet passed to
312  * journal_access at least once.
313  *
314  * That said, 99% of the time this doesn't matter and this is what the
315  * path looks like:
316  *
317  *	<read a bh>
318  *	ocfs2_journal_access(handle, bh,	OCFS2_JOURNAL_ACCESS_WRITE);
319  *	<modify the bh>
320  * 	ocfs2_journal_dirty(handle, bh);
321  */
322 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
323 
324 /*
325  *  Credit Macros:
326  *  Convenience macros to calculate number of credits needed.
327  *
328  *  For convenience sake, I have a set of macros here which calculate
329  *  the *maximum* number of sectors which will be changed for various
330  *  metadata updates.
331  */
332 
333 /* simple file updates like chmod, etc. */
334 #define OCFS2_INODE_UPDATE_CREDITS 1
335 
336 /* extended attribute block update */
337 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
338 
339 /* Update of a single quota block */
340 #define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
341 
342 /* global quotafile inode update, data block */
343 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
344 				   OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
345 
346 #define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
347 /*
348  * The two writes below can accidentally see global info dirty due
349  * to set_info() quotactl so make them prepared for the writes.
350  */
351 /* quota data block, global info */
352 /* Write to local quota file */
353 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
354 			      OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
355 
356 /* global quota data block, local quota data block, global quota inode,
357  * global quota info */
358 #define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
359 			     2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
360 
361 static inline int ocfs2_quota_trans_credits(struct super_block *sb)
362 {
363 	int credits = 0;
364 
365 	if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
366 		credits += OCFS2_QWRITE_CREDITS;
367 	if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
368 		credits += OCFS2_QWRITE_CREDITS;
369 	return credits;
370 }
371 
372 /* group extend. inode update and last group update. */
373 #define OCFS2_GROUP_EXTEND_CREDITS	(OCFS2_INODE_UPDATE_CREDITS + 1)
374 
375 /* group add. inode update and the new group update. */
376 #define OCFS2_GROUP_ADD_CREDITS	(OCFS2_INODE_UPDATE_CREDITS + 1)
377 
378 /* get one bit out of a suballocator: dinode + group descriptor +
379  * prev. group desc. if we relink. */
380 #define OCFS2_SUBALLOC_ALLOC (3)
381 
382 static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
383 {
384 	return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
385 	       ocfs2_quota_trans_credits(sb);
386 }
387 
388 /* dinode + group descriptor update. We don't relink on free yet. */
389 #define OCFS2_SUBALLOC_FREE  (2)
390 
391 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
392 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE 		      \
393 					 + OCFS2_TRUNCATE_LOG_UPDATE)
394 
395 static inline int ocfs2_remove_extent_credits(struct super_block *sb)
396 {
397 	return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
398 	       ocfs2_quota_trans_credits(sb);
399 }
400 
401 /* data block for new dir/symlink, allocation of directory block, dx_root
402  * update for free list */
403 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
404 
405 static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
406 {
407 	/* 1 block for index, 2 allocs (data, metadata), 1 clusters
408 	 * worth of blocks for initial extent. */
409 	return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
410 		ocfs2_clusters_to_blocks(sb, 1);
411 }
412 
413 /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
414  * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
415  * blocks + quota update */
416 static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
417 				      int xattr_credits)
418 {
419 	int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
420 
421 	if (is_dir)
422 		dir_credits += ocfs2_add_dir_index_credits(sb);
423 
424 	return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
425 	       ocfs2_quota_trans_credits(sb);
426 }
427 
428 /* local alloc metadata change + main bitmap updates */
429 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS                 \
430 				  + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
431 
432 /* used when we don't need an allocation change for a dir extend. One
433  * for the dinode, one for the new block. */
434 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
435 
436 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
437  * update on dir + index leaf + dx root update for free list +
438  * previous dirblock update in the free list */
439 static inline int ocfs2_link_credits(struct super_block *sb)
440 {
441 	return 2 * OCFS2_INODE_UPDATE_CREDITS + 4 +
442 	       ocfs2_quota_trans_credits(sb);
443 }
444 
445 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
446  * dir inode link + dir inode index leaf + dir index root */
447 static inline int ocfs2_unlink_credits(struct super_block *sb)
448 {
449 	/* The quota update from ocfs2_link_credits is unused here... */
450 	return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
451 }
452 
453 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
454  * inode alloc group descriptor + orphan dir index root +
455  * orphan dir index leaf */
456 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
457 
458 /* dinode + orphan dir dinode + extent tree leaf block + orphan dir entry +
459  * orphan dir index root + orphan dir index leaf */
460 #define OCFS2_INODE_ADD_TO_ORPHAN_CREDITS  (2 * OCFS2_INODE_UPDATE_CREDITS + 4)
461 #define OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS  OCFS2_INODE_ADD_TO_ORPHAN_CREDITS
462 
463 /* dinode update, old dir dinode update, new dir dinode update, old
464  * dir dir entry, new dir dir entry, dir entry update for renaming
465  * directory + target unlink + 3 x dir index leaves */
466 static inline int ocfs2_rename_credits(struct super_block *sb)
467 {
468 	return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
469 }
470 
471 /* global bitmap dinode, group desc., relinked group,
472  * suballocator dinode, group desc., relinked group,
473  * dinode, xattr block */
474 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
475 					  + OCFS2_INODE_UPDATE_CREDITS \
476 					  + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
477 
478 /* inode update, removal of dx root block from allocator */
479 #define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS +	\
480 				      OCFS2_SUBALLOC_FREE)
481 
482 static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
483 {
484 	int credits = 1 + OCFS2_SUBALLOC_ALLOC;
485 
486 	credits += ocfs2_clusters_to_blocks(sb, 1);
487 	credits += ocfs2_quota_trans_credits(sb);
488 
489 	return credits;
490 }
491 
492 /* inode update, new refcount block and its allocation credits. */
493 #define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
494 					    + OCFS2_SUBALLOC_ALLOC)
495 
496 /* inode and the refcount block update. */
497 #define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
498 
499 /*
500  * inode and the refcount block update.
501  * It doesn't include the credits for sub alloc change.
502  * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
503  */
504 #define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
505 
506 /* 2 metadata alloc, 2 new blocks and root refcount block */
507 #define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
508 
509 /*
510  * Please note that the caller must make sure that root_el is the root
511  * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
512  * the result may be wrong.
513  */
514 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
515 					    struct ocfs2_extent_list *root_el)
516 {
517 	int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
518 
519 	/* bitmap dinode, group desc. + relinked group. */
520 	bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
521 
522 	/* we might need to shift tree depth so lets assume an
523 	 * absolute worst case of complete fragmentation.  Even with
524 	 * that, we only need one update for the dinode, and then
525 	 * however many metadata chunks needed * a remaining suballoc
526 	 * alloc. */
527 	sysfile_bitmap_blocks = 1 +
528 		(OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
529 
530 	/* this does not include *new* metadata blocks, which are
531 	 * accounted for in sysfile_bitmap_blocks. root_el +
532 	 * prev. last_eb_blk + blocks along edge of tree.
533 	 * calc_symlink_credits passes because we just need 1
534 	 * credit for the dinode there. */
535 	extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
536 
537 	return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
538 	       ocfs2_quota_trans_credits(sb);
539 }
540 
541 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
542 {
543 	int blocks = ocfs2_mknod_credits(sb, 0, 0);
544 
545 	/* links can be longer than one block so we may update many
546 	 * within our single allocated extent. */
547 	blocks += ocfs2_clusters_to_blocks(sb, 1);
548 
549 	return blocks + ocfs2_quota_trans_credits(sb);
550 }
551 
552 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
553 						 unsigned int cpg)
554 {
555 	int blocks;
556 	int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
557 	/* parent inode update + new block group header + bitmap inode update
558 	   + bitmap blocks affected */
559 	blocks = 1 + 1 + 1 + bitmap_blocks;
560 	return blocks;
561 }
562 
563 /*
564  * Allocating a discontiguous block group requires the credits from
565  * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
566  * the group descriptor's extent list.  The caller already has started
567  * the transaction with ocfs2_calc_group_alloc_credits().  They extend
568  * it with these credits.
569  */
570 static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
571 {
572 	return ocfs2_extent_recs_per_gd(sb);
573 }
574 
575 static inline int ocfs2_jbd2_inode_add_write(handle_t *handle, struct inode *inode,
576 					     loff_t start_byte, loff_t length)
577 {
578 	return jbd2_journal_inode_ranged_write(handle,
579 					       &OCFS2_I(inode)->ip_jinode,
580 					       start_byte, length);
581 }
582 
583 static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
584 					       loff_t new_size)
585 {
586 	return jbd2_journal_begin_ordered_truncate(
587 				OCFS2_SB(inode->i_sb)->journal->j_journal,
588 				&OCFS2_I(inode)->ip_jinode,
589 				new_size);
590 }
591 
592 static inline void ocfs2_update_inode_fsync_trans(handle_t *handle,
593 						  struct inode *inode,
594 						  int datasync)
595 {
596 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
597 
598 	if (!is_handle_aborted(handle)) {
599 		oi->i_sync_tid = handle->h_transaction->t_tid;
600 		if (datasync)
601 			oi->i_datasync_tid = handle->h_transaction->t_tid;
602 	}
603 }
604 
605 #endif /* OCFS2_JOURNAL_H */
606