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
,
41 struct ocfs2_journal
{
42 enum ocfs2_journal_state j_state
; /* Journals current state */
44 journal_t
*j_journal
; /* The kernels journal type */
45 struct inode
*j_inode
; /* Kernel inode pointing to
47 struct ocfs2_super
*j_osb
; /* pointer to the super
51 * necessarily the super
53 * which we usually run
56 struct buffer_head
*j_bh
; /* Journal disk inode block */
57 atomic_t j_num_trans
; /* Number of transactions
58 * currently in the system. */
59 unsigned long j_trans_id
;
60 struct rw_semaphore j_trans_barrier
;
61 wait_queue_head_t j_checkpointed
;
64 struct list_head j_la_cleanups
;
65 struct work_struct j_recovery_work
;
68 extern spinlock_t trans_inc_lock
;
70 /* wrap j_trans_id so we never have it equal to zero. */
71 static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal
*j
)
74 spin_lock(&trans_inc_lock
);
75 old_id
= j
->j_trans_id
++;
76 if (unlikely(!j
->j_trans_id
))
78 spin_unlock(&trans_inc_lock
);
82 static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal
*journal
,
85 spin_lock(&trans_inc_lock
);
86 OCFS2_I(inode
)->ip_last_trans
= journal
->j_trans_id
;
87 spin_unlock(&trans_inc_lock
);
90 /* Used to figure out whether it's safe to drop a metadata lock on an
91 * inode. Returns true if all the inodes changes have been
92 * checkpointed to disk. You should be holding the spinlock on the
93 * metadata lock while calling this to be sure that nobody can take
94 * the lock and put it on another transaction. */
95 static inline int ocfs2_inode_fully_checkpointed(struct inode
*inode
)
98 struct ocfs2_journal
*journal
= OCFS2_SB(inode
->i_sb
)->journal
;
100 spin_lock(&trans_inc_lock
);
101 ret
= time_after(journal
->j_trans_id
, OCFS2_I(inode
)->ip_last_trans
);
102 spin_unlock(&trans_inc_lock
);
106 /* convenience function to check if an inode is still new (has never
107 * hit disk) Will do you a favor and set created_trans = 0 when you've
108 * been checkpointed. returns '1' if the inode is still new. */
109 static inline int ocfs2_inode_is_new(struct inode
*inode
)
113 /* System files are never "new" as they're written out by
114 * mkfs. This helps us early during mount, before we have the
115 * journal open and j_trans_id could be junk. */
116 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
118 spin_lock(&trans_inc_lock
);
119 ret
= !(time_after(OCFS2_SB(inode
->i_sb
)->journal
->j_trans_id
,
120 OCFS2_I(inode
)->ip_created_trans
));
122 OCFS2_I(inode
)->ip_created_trans
= 0;
123 spin_unlock(&trans_inc_lock
);
127 static inline void ocfs2_inode_set_new(struct ocfs2_super
*osb
,
130 spin_lock(&trans_inc_lock
);
131 OCFS2_I(inode
)->ip_created_trans
= osb
->journal
->j_trans_id
;
132 spin_unlock(&trans_inc_lock
);
135 /* Exported only for the journal struct init code in super.c. Do not call. */
136 void ocfs2_complete_recovery(struct work_struct
*work
);
137 void ocfs2_wait_for_recovery(struct ocfs2_super
*osb
);
139 int ocfs2_recovery_init(struct ocfs2_super
*osb
);
140 void ocfs2_recovery_exit(struct ocfs2_super
*osb
);
144 * Initialize, Load, Shutdown, Wipe a journal.
146 * ocfs2_journal_init - Initialize journal structures in the OSB.
147 * ocfs2_journal_load - Load the given journal off disk. Replay it if
148 * there's transactions still in there.
149 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
150 * uncommitted, uncheckpointed transactions.
151 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
152 * zero out each block.
153 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
154 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
156 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
158 void ocfs2_set_journal_params(struct ocfs2_super
*osb
);
159 int ocfs2_journal_init(struct ocfs2_journal
*journal
,
161 void ocfs2_journal_shutdown(struct ocfs2_super
*osb
);
162 int ocfs2_journal_wipe(struct ocfs2_journal
*journal
,
164 int ocfs2_journal_load(struct ocfs2_journal
*journal
, int local
,
166 int ocfs2_check_journals_nolocks(struct ocfs2_super
*osb
);
167 void ocfs2_recovery_thread(struct ocfs2_super
*osb
,
169 int ocfs2_mark_dead_nodes(struct ocfs2_super
*osb
);
170 void ocfs2_complete_mount_recovery(struct ocfs2_super
*osb
);
172 static inline void ocfs2_start_checkpoint(struct ocfs2_super
*osb
)
174 atomic_set(&osb
->needs_checkpoint
, 1);
175 wake_up(&osb
->checkpoint_event
);
178 static inline void ocfs2_checkpoint_inode(struct inode
*inode
)
180 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
182 if (ocfs2_mount_local(osb
))
185 if (!ocfs2_inode_fully_checkpointed(inode
)) {
186 /* WARNING: This only kicks off a single
187 * checkpoint. If someone races you and adds more
188 * metadata to the journal, you won't know, and will
189 * wind up waiting *alot* longer than necessary. Right
190 * now we only use this in clear_inode so that's
192 ocfs2_start_checkpoint(osb
);
194 wait_event(osb
->journal
->j_checkpointed
,
195 ocfs2_inode_fully_checkpointed(inode
));
200 * Transaction Handling:
201 * Manage the lifetime of a transaction handle.
203 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
204 * the number of blocks that will be changed during
206 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
207 * the journal was aborted. The majority of paths don't
208 * check the return value as an error there comes too
209 * late to do anything (and will be picked up in a
210 * later transaction).
211 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
212 * commit the handle to disk in the process, but will
213 * not release any locks taken during the transaction.
214 * ocfs2_journal_access - Notify the handle that we want to journal this
215 * buffer. Will have to call ocfs2_journal_dirty once
216 * we've actually dirtied it. Type is one of . or .
217 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
218 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
219 * the current handle commits.
222 /* You must always start_trans with a number of buffs > 0, but it's
223 * perfectly legal to go through an entire transaction without having
224 * dirtied any buffers. */
225 handle_t
*ocfs2_start_trans(struct ocfs2_super
*osb
,
227 int ocfs2_commit_trans(struct ocfs2_super
*osb
,
229 int ocfs2_extend_trans(handle_t
*handle
, int nblocks
);
232 * Create access is for when we get a newly created buffer and we're
233 * not gonna read it off disk, but rather fill it ourselves. Right
234 * now, we don't do anything special with this (it turns into a write
235 * request), but this is a good placeholder in case we do...
237 * Write access is for when we read a block off disk and are going to
238 * modify it. This way the journalling layer knows it may need to make
239 * a copy of that block (if it's part of another, uncommitted
240 * transaction) before we do so.
242 #define OCFS2_JOURNAL_ACCESS_CREATE 0
243 #define OCFS2_JOURNAL_ACCESS_WRITE 1
244 #define OCFS2_JOURNAL_ACCESS_UNDO 2
246 int ocfs2_journal_access(handle_t
*handle
,
248 struct buffer_head
*bh
,
251 * A word about the journal_access/journal_dirty "dance". It is
252 * entirely legal to journal_access a buffer more than once (as long
253 * as the access type is the same -- I'm not sure what will happen if
254 * access type is different but this should never happen anyway) It is
255 * also legal to journal_dirty a buffer more than once. In fact, you
256 * can even journal_access a buffer after you've done a
257 * journal_access/journal_dirty pair. The only thing you cannot do
258 * however, is journal_dirty a buffer which you haven't yet passed to
259 * journal_access at least once.
261 * That said, 99% of the time this doesn't matter and this is what the
265 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
267 * ocfs2_journal_dirty(handle, bh);
269 int ocfs2_journal_dirty(handle_t
*handle
,
270 struct buffer_head
*bh
);
274 * Convenience macros to calculate number of credits needed.
276 * For convenience sake, I have a set of macros here which calculate
277 * the *maximum* number of sectors which will be changed for various
281 /* simple file updates like chmod, etc. */
282 #define OCFS2_INODE_UPDATE_CREDITS 1
284 /* extended attribute block update */
285 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
287 /* global quotafile inode update, data block */
288 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
291 * The two writes below can accidentally see global info dirty due
292 * to set_info() quotactl so make them prepared for the writes.
294 /* quota data block, global info */
295 /* Write to local quota file */
296 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + 1)
298 /* global quota data block, local quota data block, global quota inode,
299 * global quota info */
300 #define OCFS2_QSYNC_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 3)
302 static inline int ocfs2_quota_trans_credits(struct super_block
*sb
)
306 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb
, OCFS2_FEATURE_RO_COMPAT_USRQUOTA
))
307 credits
+= OCFS2_QWRITE_CREDITS
;
308 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb
, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA
))
309 credits
+= OCFS2_QWRITE_CREDITS
;
313 /* Number of credits needed for removing quota structure from file */
314 int ocfs2_calc_qdel_credits(struct super_block
*sb
, int type
);
315 /* Number of credits needed for initialization of new quota structure */
316 int ocfs2_calc_qinit_credits(struct super_block
*sb
, int type
);
318 /* group extend. inode update and last group update. */
319 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
321 /* group add. inode update and the new group update. */
322 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
324 /* get one bit out of a suballocator: dinode + group descriptor +
325 * prev. group desc. if we relink. */
326 #define OCFS2_SUBALLOC_ALLOC (3)
328 static inline int ocfs2_inline_to_extents_credits(struct super_block
*sb
)
330 return OCFS2_SUBALLOC_ALLOC
+ OCFS2_INODE_UPDATE_CREDITS
+
331 ocfs2_quota_trans_credits(sb
);
334 /* dinode + group descriptor update. We don't relink on free yet. */
335 #define OCFS2_SUBALLOC_FREE (2)
337 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
338 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
339 + OCFS2_TRUNCATE_LOG_UPDATE)
341 static inline int ocfs2_remove_extent_credits(struct super_block
*sb
)
343 return OCFS2_TRUNCATE_LOG_UPDATE
+ OCFS2_INODE_UPDATE_CREDITS
+
344 ocfs2_quota_trans_credits(sb
);
347 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
348 * bitmap block for the new bit) */
349 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
351 /* parent fe, parent block, new file entry, inode alloc fe, inode alloc
352 * group descriptor + mkdir/symlink blocks + quota update */
353 static inline int ocfs2_mknod_credits(struct super_block
*sb
)
355 return 3 + OCFS2_SUBALLOC_ALLOC
+ OCFS2_DIR_LINK_ADDITIONAL_CREDITS
+
356 ocfs2_quota_trans_credits(sb
);
359 /* local alloc metadata change + main bitmap updates */
360 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
361 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
363 /* used when we don't need an allocation change for a dir extend. One
364 * for the dinode, one for the new block. */
365 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
367 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
369 static inline int ocfs2_link_credits(struct super_block
*sb
)
371 return 2*OCFS2_INODE_UPDATE_CREDITS
+ 1 +
372 ocfs2_quota_trans_credits(sb
);
375 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
377 static inline int ocfs2_unlink_credits(struct super_block
*sb
)
379 /* The quota update from ocfs2_link_credits is unused here... */
380 return 2 * OCFS2_INODE_UPDATE_CREDITS
+ 1 + ocfs2_link_credits(sb
);
383 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
384 * inode alloc group descriptor */
385 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
387 /* dinode update, old dir dinode update, new dir dinode update, old
388 * dir dir entry, new dir dir entry, dir entry update for renaming
389 * directory + target unlink */
390 static inline int ocfs2_rename_credits(struct super_block
*sb
)
392 return 3 * OCFS2_INODE_UPDATE_CREDITS
+ 3 + ocfs2_unlink_credits(sb
);
395 /* global bitmap dinode, group desc., relinked group,
396 * suballocator dinode, group desc., relinked group,
397 * dinode, xattr block */
398 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
399 + OCFS2_INODE_UPDATE_CREDITS \
400 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
403 * Please note that the caller must make sure that root_el is the root
404 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
405 * the result may be wrong.
407 static inline int ocfs2_calc_extend_credits(struct super_block
*sb
,
408 struct ocfs2_extent_list
*root_el
,
411 int bitmap_blocks
, sysfile_bitmap_blocks
, extent_blocks
;
413 /* bitmap dinode, group desc. + relinked group. */
414 bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
;
416 /* we might need to shift tree depth so lets assume an
417 * absolute worst case of complete fragmentation. Even with
418 * that, we only need one update for the dinode, and then
419 * however many metadata chunks needed * a remaining suballoc
421 sysfile_bitmap_blocks
= 1 +
422 (OCFS2_SUBALLOC_ALLOC
- 1) * ocfs2_extend_meta_needed(root_el
);
424 /* this does not include *new* metadata blocks, which are
425 * accounted for in sysfile_bitmap_blocks. root_el +
426 * prev. last_eb_blk + blocks along edge of tree.
427 * calc_symlink_credits passes because we just need 1
428 * credit for the dinode there. */
429 extent_blocks
= 1 + 1 + le16_to_cpu(root_el
->l_tree_depth
);
431 return bitmap_blocks
+ sysfile_bitmap_blocks
+ extent_blocks
+
432 ocfs2_quota_trans_credits(sb
);
435 static inline int ocfs2_calc_symlink_credits(struct super_block
*sb
)
437 int blocks
= ocfs2_mknod_credits(sb
);
439 /* links can be longer than one block so we may update many
440 * within our single allocated extent. */
441 blocks
+= ocfs2_clusters_to_blocks(sb
, 1);
443 return blocks
+ ocfs2_quota_trans_credits(sb
);
446 static inline int ocfs2_calc_group_alloc_credits(struct super_block
*sb
,
450 int bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
+ 1;
451 /* parent inode update + new block group header + bitmap inode update
452 + bitmap blocks affected */
453 blocks
= 1 + 1 + 1 + bitmap_blocks
;
457 static inline int ocfs2_calc_tree_trunc_credits(struct super_block
*sb
,
458 unsigned int clusters_to_del
,
459 struct ocfs2_dinode
*fe
,
460 struct ocfs2_extent_list
*last_el
)
462 /* for dinode + all headers in this pass + update to next leaf */
463 u16 next_free
= le16_to_cpu(last_el
->l_next_free_rec
);
464 u16 tree_depth
= le16_to_cpu(fe
->id2
.i_list
.l_tree_depth
);
465 int credits
= 1 + tree_depth
+ 1;
471 /* We may be deleting metadata blocks, so metadata alloc dinode +
472 one desc. block for each possible delete. */
473 if (tree_depth
&& next_free
== 1 &&
474 ocfs2_rec_clusters(last_el
, &last_el
->l_recs
[i
]) == clusters_to_del
)
475 credits
+= 1 + tree_depth
;
477 /* update to the truncate log. */
478 credits
+= OCFS2_TRUNCATE_LOG_UPDATE
;
480 credits
+= ocfs2_quota_trans_credits(sb
);
485 static inline int ocfs2_jbd2_file_inode(handle_t
*handle
, struct inode
*inode
)
487 return jbd2_journal_file_inode(handle
, &OCFS2_I(inode
)->ip_jinode
);
490 static inline int ocfs2_begin_ordered_truncate(struct inode
*inode
,
493 return jbd2_journal_begin_ordered_truncate(&OCFS2_I(inode
)->ip_jinode
,
497 #endif /* OCFS2_JOURNAL_H */