1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Defines journalling api and structures.
8 * Copyright (C) 2003, 2005 Oracle. All rights reserved.
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.
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.
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.
26 #ifndef OCFS2_JOURNAL_H
27 #define OCFS2_JOURNAL_H
30 #include <linux/jbd2.h>
32 enum ocfs2_journal_state
{
33 OCFS2_JOURNAL_FREE
= 0,
35 OCFS2_JOURNAL_IN_SHUTDOWN
,
42 * The recovery_list is a simple linked list of node numbers to recover.
43 * It is protected by the recovery_lock.
46 struct ocfs2_recovery_map
{
48 unsigned int *rm_entries
;
52 struct ocfs2_journal
{
53 enum ocfs2_journal_state j_state
; /* Journals current state */
55 journal_t
*j_journal
; /* The kernels journal type */
56 struct inode
*j_inode
; /* Kernel inode pointing to
58 struct ocfs2_super
*j_osb
; /* pointer to the super
62 * necessarily the super
64 * which we usually run
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 unsigned long j_trans_id
;
71 struct rw_semaphore j_trans_barrier
;
72 wait_queue_head_t j_checkpointed
;
75 struct list_head j_la_cleanups
;
76 struct work_struct j_recovery_work
;
79 extern spinlock_t trans_inc_lock
;
81 /* wrap j_trans_id so we never have it equal to zero. */
82 static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal
*j
)
85 spin_lock(&trans_inc_lock
);
86 old_id
= j
->j_trans_id
++;
87 if (unlikely(!j
->j_trans_id
))
89 spin_unlock(&trans_inc_lock
);
93 static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal
*journal
,
96 spin_lock(&trans_inc_lock
);
97 OCFS2_I(inode
)->ip_last_trans
= journal
->j_trans_id
;
98 spin_unlock(&trans_inc_lock
);
101 /* Used to figure out whether it's safe to drop a metadata lock on an
102 * inode. Returns true if all the inodes changes have been
103 * checkpointed to disk. You should be holding the spinlock on the
104 * metadata lock while calling this to be sure that nobody can take
105 * the lock and put it on another transaction. */
106 static inline int ocfs2_inode_fully_checkpointed(struct inode
*inode
)
109 struct ocfs2_journal
*journal
= OCFS2_SB(inode
->i_sb
)->journal
;
111 spin_lock(&trans_inc_lock
);
112 ret
= time_after(journal
->j_trans_id
, OCFS2_I(inode
)->ip_last_trans
);
113 spin_unlock(&trans_inc_lock
);
117 /* convenience function to check if an inode is still new (has never
118 * hit disk) Will do you a favor and set created_trans = 0 when you've
119 * been checkpointed. returns '1' if the inode is still new. */
120 static inline int ocfs2_inode_is_new(struct inode
*inode
)
124 /* System files are never "new" as they're written out by
125 * mkfs. This helps us early during mount, before we have the
126 * journal open and j_trans_id could be junk. */
127 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
129 spin_lock(&trans_inc_lock
);
130 ret
= !(time_after(OCFS2_SB(inode
->i_sb
)->journal
->j_trans_id
,
131 OCFS2_I(inode
)->ip_created_trans
));
133 OCFS2_I(inode
)->ip_created_trans
= 0;
134 spin_unlock(&trans_inc_lock
);
138 static inline void ocfs2_inode_set_new(struct ocfs2_super
*osb
,
141 spin_lock(&trans_inc_lock
);
142 OCFS2_I(inode
)->ip_created_trans
= osb
->journal
->j_trans_id
;
143 spin_unlock(&trans_inc_lock
);
146 /* Exported only for the journal struct init code in super.c. Do not call. */
147 int ocfs2_orphan_scan_init(struct ocfs2_super
*osb
);
148 void ocfs2_orphan_scan_stop(struct ocfs2_super
*osb
);
149 void ocfs2_orphan_scan_exit(struct ocfs2_super
*osb
);
151 void ocfs2_complete_recovery(struct work_struct
*work
);
152 void ocfs2_wait_for_recovery(struct ocfs2_super
*osb
);
154 int ocfs2_recovery_init(struct ocfs2_super
*osb
);
155 void ocfs2_recovery_exit(struct ocfs2_super
*osb
);
157 int ocfs2_compute_replay_slots(struct ocfs2_super
*osb
);
160 * Initialize, Load, Shutdown, Wipe a journal.
162 * ocfs2_journal_init - Initialize journal structures in the OSB.
163 * ocfs2_journal_load - Load the given journal off disk. Replay it if
164 * there's transactions still in there.
165 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
166 * uncommitted, uncheckpointed transactions.
167 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
168 * zero out each block.
169 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
170 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
172 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
174 void ocfs2_set_journal_params(struct ocfs2_super
*osb
);
175 int ocfs2_journal_init(struct ocfs2_journal
*journal
,
177 void ocfs2_journal_shutdown(struct ocfs2_super
*osb
);
178 int ocfs2_journal_wipe(struct ocfs2_journal
*journal
,
180 int ocfs2_journal_load(struct ocfs2_journal
*journal
, int local
,
182 int ocfs2_check_journals_nolocks(struct ocfs2_super
*osb
);
183 void ocfs2_recovery_thread(struct ocfs2_super
*osb
,
185 int ocfs2_mark_dead_nodes(struct ocfs2_super
*osb
);
186 void ocfs2_complete_mount_recovery(struct ocfs2_super
*osb
);
187 void ocfs2_complete_quota_recovery(struct ocfs2_super
*osb
);
189 static inline void ocfs2_start_checkpoint(struct ocfs2_super
*osb
)
191 atomic_set(&osb
->needs_checkpoint
, 1);
192 wake_up(&osb
->checkpoint_event
);
195 static inline void ocfs2_checkpoint_inode(struct inode
*inode
)
197 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
199 if (ocfs2_mount_local(osb
))
202 if (!ocfs2_inode_fully_checkpointed(inode
)) {
203 /* WARNING: This only kicks off a single
204 * checkpoint. If someone races you and adds more
205 * metadata to the journal, you won't know, and will
206 * wind up waiting *alot* longer than necessary. Right
207 * now we only use this in clear_inode so that's
209 ocfs2_start_checkpoint(osb
);
211 wait_event(osb
->journal
->j_checkpointed
,
212 ocfs2_inode_fully_checkpointed(inode
));
217 * Transaction Handling:
218 * Manage the lifetime of a transaction handle.
220 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
221 * the number of blocks that will be changed during
223 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
224 * the journal was aborted. The majority of paths don't
225 * check the return value as an error there comes too
226 * late to do anything (and will be picked up in a
227 * later transaction).
228 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
229 * commit the handle to disk in the process, but will
230 * not release any locks taken during the transaction.
231 * ocfs2_journal_access* - Notify the handle that we want to journal this
232 * buffer. Will have to call ocfs2_journal_dirty once
233 * we've actually dirtied it. Type is one of . or .
234 * Always call the specific flavor of
235 * ocfs2_journal_access_*() unless you intend to
236 * manage the checksum by hand.
237 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
238 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
239 * the current handle commits.
242 /* You must always start_trans with a number of buffs > 0, but it's
243 * perfectly legal to go through an entire transaction without having
244 * dirtied any buffers. */
245 handle_t
*ocfs2_start_trans(struct ocfs2_super
*osb
,
247 int ocfs2_commit_trans(struct ocfs2_super
*osb
,
249 int ocfs2_extend_trans(handle_t
*handle
, int nblocks
);
252 * Create access is for when we get a newly created buffer and we're
253 * not gonna read it off disk, but rather fill it ourselves. Right
254 * now, we don't do anything special with this (it turns into a write
255 * request), but this is a good placeholder in case we do...
257 * Write access is for when we read a block off disk and are going to
258 * modify it. This way the journalling layer knows it may need to make
259 * a copy of that block (if it's part of another, uncommitted
260 * transaction) before we do so.
262 #define OCFS2_JOURNAL_ACCESS_CREATE 0
263 #define OCFS2_JOURNAL_ACCESS_WRITE 1
264 #define OCFS2_JOURNAL_ACCESS_UNDO 2
268 int ocfs2_journal_access_di(handle_t
*handle
, struct inode
*inode
,
269 struct buffer_head
*bh
, int type
);
270 /* ocfs2_extent_block */
271 int ocfs2_journal_access_eb(handle_t
*handle
, struct inode
*inode
,
272 struct buffer_head
*bh
, int type
);
273 /* ocfs2_group_desc */
274 int ocfs2_journal_access_gd(handle_t
*handle
, struct inode
*inode
,
275 struct buffer_head
*bh
, int type
);
276 /* ocfs2_xattr_block */
277 int ocfs2_journal_access_xb(handle_t
*handle
, struct inode
*inode
,
278 struct buffer_head
*bh
, int type
);
280 int ocfs2_journal_access_dq(handle_t
*handle
, struct inode
*inode
,
281 struct buffer_head
*bh
, int type
);
283 int ocfs2_journal_access_db(handle_t
*handle
, struct inode
*inode
,
284 struct buffer_head
*bh
, int type
);
285 /* ocfs2_dx_root_block */
286 int ocfs2_journal_access_dr(handle_t
*handle
, struct inode
*inode
,
287 struct buffer_head
*bh
, int type
);
289 int ocfs2_journal_access_dl(handle_t
*handle
, struct inode
*inode
,
290 struct buffer_head
*bh
, int type
);
291 /* Anything that has no ecc */
292 int ocfs2_journal_access(handle_t
*handle
, struct inode
*inode
,
293 struct buffer_head
*bh
, int type
);
296 * A word about the journal_access/journal_dirty "dance". It is
297 * entirely legal to journal_access a buffer more than once (as long
298 * as the access type is the same -- I'm not sure what will happen if
299 * access type is different but this should never happen anyway) It is
300 * also legal to journal_dirty a buffer more than once. In fact, you
301 * can even journal_access a buffer after you've done a
302 * journal_access/journal_dirty pair. The only thing you cannot do
303 * however, is journal_dirty a buffer which you haven't yet passed to
304 * journal_access at least once.
306 * That said, 99% of the time this doesn't matter and this is what the
310 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
312 * ocfs2_journal_dirty(handle, bh);
314 int ocfs2_journal_dirty(handle_t
*handle
,
315 struct buffer_head
*bh
);
319 * Convenience macros to calculate number of credits needed.
321 * For convenience sake, I have a set of macros here which calculate
322 * the *maximum* number of sectors which will be changed for various
326 /* simple file updates like chmod, etc. */
327 #define OCFS2_INODE_UPDATE_CREDITS 1
329 /* extended attribute block update */
330 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
332 /* global quotafile inode update, data block */
333 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
336 * The two writes below can accidentally see global info dirty due
337 * to set_info() quotactl so make them prepared for the writes.
339 /* quota data block, global info */
340 /* Write to local quota file */
341 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + 1)
343 /* global quota data block, local quota data block, global quota inode,
344 * global quota info */
345 #define OCFS2_QSYNC_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 3)
347 static inline int ocfs2_quota_trans_credits(struct super_block
*sb
)
351 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb
, OCFS2_FEATURE_RO_COMPAT_USRQUOTA
))
352 credits
+= OCFS2_QWRITE_CREDITS
;
353 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb
, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA
))
354 credits
+= OCFS2_QWRITE_CREDITS
;
358 /* Number of credits needed for removing quota structure from file */
359 int ocfs2_calc_qdel_credits(struct super_block
*sb
, int type
);
360 /* Number of credits needed for initialization of new quota structure */
361 int ocfs2_calc_qinit_credits(struct super_block
*sb
, int type
);
363 /* group extend. inode update and last group update. */
364 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
366 /* group add. inode update and the new group update. */
367 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
369 /* get one bit out of a suballocator: dinode + group descriptor +
370 * prev. group desc. if we relink. */
371 #define OCFS2_SUBALLOC_ALLOC (3)
373 static inline int ocfs2_inline_to_extents_credits(struct super_block
*sb
)
375 return OCFS2_SUBALLOC_ALLOC
+ OCFS2_INODE_UPDATE_CREDITS
+
376 ocfs2_quota_trans_credits(sb
);
379 /* dinode + group descriptor update. We don't relink on free yet. */
380 #define OCFS2_SUBALLOC_FREE (2)
382 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
383 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
384 + OCFS2_TRUNCATE_LOG_UPDATE)
386 static inline int ocfs2_remove_extent_credits(struct super_block
*sb
)
388 return OCFS2_TRUNCATE_LOG_UPDATE
+ OCFS2_INODE_UPDATE_CREDITS
+
389 ocfs2_quota_trans_credits(sb
);
392 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
393 * bitmap block for the new bit) dx_root update for free list */
394 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2 + 1)
396 static inline int ocfs2_add_dir_index_credits(struct super_block
*sb
)
398 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
399 * worth of blocks for initial extent. */
400 return 1 + 2 * OCFS2_SUBALLOC_ALLOC
+
401 ocfs2_clusters_to_blocks(sb
, 1);
404 /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
405 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
406 * blocks + quota update */
407 static inline int ocfs2_mknod_credits(struct super_block
*sb
, int is_dir
,
410 int dir_credits
= OCFS2_DIR_LINK_ADDITIONAL_CREDITS
;
413 dir_credits
+= ocfs2_add_dir_index_credits(sb
);
415 return 4 + OCFS2_SUBALLOC_ALLOC
+ dir_credits
+ xattr_credits
+
416 ocfs2_quota_trans_credits(sb
);
419 /* local alloc metadata change + main bitmap updates */
420 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
421 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
423 /* used when we don't need an allocation change for a dir extend. One
424 * for the dinode, one for the new block. */
425 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
427 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
428 * update on dir + index leaf + dx root update for free list */
429 static inline int ocfs2_link_credits(struct super_block
*sb
)
431 return 2*OCFS2_INODE_UPDATE_CREDITS
+ 3 +
432 ocfs2_quota_trans_credits(sb
);
435 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
436 * dir inode link + dir inode index leaf + dir index root */
437 static inline int ocfs2_unlink_credits(struct super_block
*sb
)
439 /* The quota update from ocfs2_link_credits is unused here... */
440 return 2 * OCFS2_INODE_UPDATE_CREDITS
+ 3 + ocfs2_link_credits(sb
);
443 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
444 * inode alloc group descriptor + orphan dir index root +
445 * orphan dir index leaf */
446 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
448 /* dinode update, old dir dinode update, new dir dinode update, old
449 * dir dir entry, new dir dir entry, dir entry update for renaming
450 * directory + target unlink + 3 x dir index leaves */
451 static inline int ocfs2_rename_credits(struct super_block
*sb
)
453 return 3 * OCFS2_INODE_UPDATE_CREDITS
+ 6 + ocfs2_unlink_credits(sb
);
456 /* global bitmap dinode, group desc., relinked group,
457 * suballocator dinode, group desc., relinked group,
458 * dinode, xattr block */
459 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
460 + OCFS2_INODE_UPDATE_CREDITS \
461 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
463 /* inode update, removal of dx root block from allocator */
464 #define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
467 static inline int ocfs2_calc_dxi_expand_credits(struct super_block
*sb
)
469 int credits
= 1 + OCFS2_SUBALLOC_ALLOC
;
471 credits
+= ocfs2_clusters_to_blocks(sb
, 1);
472 credits
+= ocfs2_quota_trans_credits(sb
);
478 * Please note that the caller must make sure that root_el is the root
479 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
480 * the result may be wrong.
482 static inline int ocfs2_calc_extend_credits(struct super_block
*sb
,
483 struct ocfs2_extent_list
*root_el
,
486 int bitmap_blocks
, sysfile_bitmap_blocks
, extent_blocks
;
488 /* bitmap dinode, group desc. + relinked group. */
489 bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
;
491 /* we might need to shift tree depth so lets assume an
492 * absolute worst case of complete fragmentation. Even with
493 * that, we only need one update for the dinode, and then
494 * however many metadata chunks needed * a remaining suballoc
496 sysfile_bitmap_blocks
= 1 +
497 (OCFS2_SUBALLOC_ALLOC
- 1) * ocfs2_extend_meta_needed(root_el
);
499 /* this does not include *new* metadata blocks, which are
500 * accounted for in sysfile_bitmap_blocks. root_el +
501 * prev. last_eb_blk + blocks along edge of tree.
502 * calc_symlink_credits passes because we just need 1
503 * credit for the dinode there. */
504 extent_blocks
= 1 + 1 + le16_to_cpu(root_el
->l_tree_depth
);
506 return bitmap_blocks
+ sysfile_bitmap_blocks
+ extent_blocks
+
507 ocfs2_quota_trans_credits(sb
);
510 static inline int ocfs2_calc_symlink_credits(struct super_block
*sb
)
512 int blocks
= ocfs2_mknod_credits(sb
, 0, 0);
514 /* links can be longer than one block so we may update many
515 * within our single allocated extent. */
516 blocks
+= ocfs2_clusters_to_blocks(sb
, 1);
518 return blocks
+ ocfs2_quota_trans_credits(sb
);
521 static inline int ocfs2_calc_group_alloc_credits(struct super_block
*sb
,
525 int bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
+ 1;
526 /* parent inode update + new block group header + bitmap inode update
527 + bitmap blocks affected */
528 blocks
= 1 + 1 + 1 + bitmap_blocks
;
532 static inline int ocfs2_calc_tree_trunc_credits(struct super_block
*sb
,
533 unsigned int clusters_to_del
,
534 struct ocfs2_dinode
*fe
,
535 struct ocfs2_extent_list
*last_el
)
537 /* for dinode + all headers in this pass + update to next leaf */
538 u16 next_free
= le16_to_cpu(last_el
->l_next_free_rec
);
539 u16 tree_depth
= le16_to_cpu(fe
->id2
.i_list
.l_tree_depth
);
540 int credits
= 1 + tree_depth
+ 1;
546 /* We may be deleting metadata blocks, so metadata alloc dinode +
547 one desc. block for each possible delete. */
548 if (tree_depth
&& next_free
== 1 &&
549 ocfs2_rec_clusters(last_el
, &last_el
->l_recs
[i
]) == clusters_to_del
)
550 credits
+= 1 + tree_depth
;
552 /* update to the truncate log. */
553 credits
+= OCFS2_TRUNCATE_LOG_UPDATE
;
555 credits
+= ocfs2_quota_trans_credits(sb
);
560 static inline int ocfs2_jbd2_file_inode(handle_t
*handle
, struct inode
*inode
)
562 return jbd2_journal_file_inode(handle
, &OCFS2_I(inode
)->ip_jinode
);
565 static inline int ocfs2_begin_ordered_truncate(struct inode
*inode
,
568 return jbd2_journal_begin_ordered_truncate(
569 OCFS2_SB(inode
->i_sb
)->journal
->j_journal
,
570 &OCFS2_I(inode
)->ip_jinode
,
574 #endif /* OCFS2_JOURNAL_H */