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 void ocfs2_complete_recovery(struct work_struct
*work
);
148 void ocfs2_wait_for_recovery(struct ocfs2_super
*osb
);
150 int ocfs2_recovery_init(struct ocfs2_super
*osb
);
151 void ocfs2_recovery_exit(struct ocfs2_super
*osb
);
155 * Initialize, Load, Shutdown, Wipe a journal.
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
167 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
169 void ocfs2_set_journal_params(struct ocfs2_super
*osb
);
170 int ocfs2_journal_init(struct ocfs2_journal
*journal
,
172 void ocfs2_journal_shutdown(struct ocfs2_super
*osb
);
173 int ocfs2_journal_wipe(struct ocfs2_journal
*journal
,
175 int ocfs2_journal_load(struct ocfs2_journal
*journal
, int local
,
177 int ocfs2_check_journals_nolocks(struct ocfs2_super
*osb
);
178 void ocfs2_recovery_thread(struct ocfs2_super
*osb
,
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
);
184 static inline void ocfs2_start_checkpoint(struct ocfs2_super
*osb
)
186 atomic_set(&osb
->needs_checkpoint
, 1);
187 wake_up(&osb
->checkpoint_event
);
190 static inline void ocfs2_checkpoint_inode(struct inode
*inode
)
192 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
194 if (ocfs2_mount_local(osb
))
197 if (!ocfs2_inode_fully_checkpointed(inode
)) {
198 /* WARNING: This only kicks off a single
199 * checkpoint. If someone races you and adds more
200 * metadata to the journal, you won't know, and will
201 * wind up waiting *alot* longer than necessary. Right
202 * now we only use this in clear_inode so that's
204 ocfs2_start_checkpoint(osb
);
206 wait_event(osb
->journal
->j_checkpointed
,
207 ocfs2_inode_fully_checkpointed(inode
));
212 * Transaction Handling:
213 * Manage the lifetime of a transaction handle.
215 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
216 * the number of blocks that will be changed during
218 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
219 * the journal was aborted. The majority of paths don't
220 * check the return value as an error there comes too
221 * late to do anything (and will be picked up in a
222 * later transaction).
223 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
224 * commit the handle to disk in the process, but will
225 * not release any locks taken during the transaction.
226 * ocfs2_journal_access* - Notify the handle that we want to journal this
227 * buffer. Will have to call ocfs2_journal_dirty once
228 * we've actually dirtied it. Type is one of . or .
229 * Always call the specific flavor of
230 * ocfs2_journal_access_*() unless you intend to
231 * manage the checksum by hand.
232 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
233 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
234 * the current handle commits.
237 /* You must always start_trans with a number of buffs > 0, but it's
238 * perfectly legal to go through an entire transaction without having
239 * dirtied any buffers. */
240 handle_t
*ocfs2_start_trans(struct ocfs2_super
*osb
,
242 int ocfs2_commit_trans(struct ocfs2_super
*osb
,
244 int ocfs2_extend_trans(handle_t
*handle
, int nblocks
);
247 * Create access is for when we get a newly created buffer and we're
248 * not gonna read it off disk, but rather fill it ourselves. Right
249 * now, we don't do anything special with this (it turns into a write
250 * request), but this is a good placeholder in case we do...
252 * Write access is for when we read a block off disk and are going to
253 * modify it. This way the journalling layer knows it may need to make
254 * a copy of that block (if it's part of another, uncommitted
255 * transaction) before we do so.
257 #define OCFS2_JOURNAL_ACCESS_CREATE 0
258 #define OCFS2_JOURNAL_ACCESS_WRITE 1
259 #define OCFS2_JOURNAL_ACCESS_UNDO 2
263 int ocfs2_journal_access_di(handle_t
*handle
, struct inode
*inode
,
264 struct buffer_head
*bh
, int type
);
265 /* ocfs2_extent_block */
266 int ocfs2_journal_access_eb(handle_t
*handle
, struct inode
*inode
,
267 struct buffer_head
*bh
, int type
);
268 /* ocfs2_group_desc */
269 int ocfs2_journal_access_gd(handle_t
*handle
, struct inode
*inode
,
270 struct buffer_head
*bh
, int type
);
271 /* ocfs2_xattr_block */
272 int ocfs2_journal_access_xb(handle_t
*handle
, struct inode
*inode
,
273 struct buffer_head
*bh
, int type
);
275 int ocfs2_journal_access_dq(handle_t
*handle
, struct inode
*inode
,
276 struct buffer_head
*bh
, int type
);
278 int ocfs2_journal_access_db(handle_t
*handle
, struct inode
*inode
,
279 struct buffer_head
*bh
, int type
);
280 /* Anything that has no ecc */
281 int ocfs2_journal_access(handle_t
*handle
, struct inode
*inode
,
282 struct buffer_head
*bh
, int type
);
285 * A word about the journal_access/journal_dirty "dance". It is
286 * entirely legal to journal_access a buffer more than once (as long
287 * as the access type is the same -- I'm not sure what will happen if
288 * access type is different but this should never happen anyway) It is
289 * also legal to journal_dirty a buffer more than once. In fact, you
290 * can even journal_access a buffer after you've done a
291 * journal_access/journal_dirty pair. The only thing you cannot do
292 * however, is journal_dirty a buffer which you haven't yet passed to
293 * journal_access at least once.
295 * That said, 99% of the time this doesn't matter and this is what the
299 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
301 * ocfs2_journal_dirty(handle, bh);
303 int ocfs2_journal_dirty(handle_t
*handle
,
304 struct buffer_head
*bh
);
308 * Convenience macros to calculate number of credits needed.
310 * For convenience sake, I have a set of macros here which calculate
311 * the *maximum* number of sectors which will be changed for various
315 /* simple file updates like chmod, etc. */
316 #define OCFS2_INODE_UPDATE_CREDITS 1
318 /* extended attribute block update */
319 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
321 /* global quotafile inode update, data block */
322 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
325 * The two writes below can accidentally see global info dirty due
326 * to set_info() quotactl so make them prepared for the writes.
328 /* quota data block, global info */
329 /* Write to local quota file */
330 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + 1)
332 /* global quota data block, local quota data block, global quota inode,
333 * global quota info */
334 #define OCFS2_QSYNC_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 3)
336 static inline int ocfs2_quota_trans_credits(struct super_block
*sb
)
340 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb
, OCFS2_FEATURE_RO_COMPAT_USRQUOTA
))
341 credits
+= OCFS2_QWRITE_CREDITS
;
342 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb
, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA
))
343 credits
+= OCFS2_QWRITE_CREDITS
;
347 /* Number of credits needed for removing quota structure from file */
348 int ocfs2_calc_qdel_credits(struct super_block
*sb
, int type
);
349 /* Number of credits needed for initialization of new quota structure */
350 int ocfs2_calc_qinit_credits(struct super_block
*sb
, int type
);
352 /* group extend. inode update and last group update. */
353 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
355 /* group add. inode update and the new group update. */
356 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
358 /* get one bit out of a suballocator: dinode + group descriptor +
359 * prev. group desc. if we relink. */
360 #define OCFS2_SUBALLOC_ALLOC (3)
362 static inline int ocfs2_inline_to_extents_credits(struct super_block
*sb
)
364 return OCFS2_SUBALLOC_ALLOC
+ OCFS2_INODE_UPDATE_CREDITS
+
365 ocfs2_quota_trans_credits(sb
);
368 /* dinode + group descriptor update. We don't relink on free yet. */
369 #define OCFS2_SUBALLOC_FREE (2)
371 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
372 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
373 + OCFS2_TRUNCATE_LOG_UPDATE)
375 static inline int ocfs2_remove_extent_credits(struct super_block
*sb
)
377 return OCFS2_TRUNCATE_LOG_UPDATE
+ OCFS2_INODE_UPDATE_CREDITS
+
378 ocfs2_quota_trans_credits(sb
);
381 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
382 * bitmap block for the new bit) */
383 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
385 /* parent fe, parent block, new file entry, inode alloc fe, inode alloc
386 * group descriptor + mkdir/symlink blocks + quota update */
387 static inline int ocfs2_mknod_credits(struct super_block
*sb
)
389 return 3 + OCFS2_SUBALLOC_ALLOC
+ OCFS2_DIR_LINK_ADDITIONAL_CREDITS
+
390 ocfs2_quota_trans_credits(sb
);
393 /* local alloc metadata change + main bitmap updates */
394 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
395 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
397 /* used when we don't need an allocation change for a dir extend. One
398 * for the dinode, one for the new block. */
399 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
401 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
403 static inline int ocfs2_link_credits(struct super_block
*sb
)
405 return 2*OCFS2_INODE_UPDATE_CREDITS
+ 1 +
406 ocfs2_quota_trans_credits(sb
);
409 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
411 static inline int ocfs2_unlink_credits(struct super_block
*sb
)
413 /* The quota update from ocfs2_link_credits is unused here... */
414 return 2 * OCFS2_INODE_UPDATE_CREDITS
+ 1 + ocfs2_link_credits(sb
);
417 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
418 * inode alloc group descriptor */
419 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
421 /* dinode update, old dir dinode update, new dir dinode update, old
422 * dir dir entry, new dir dir entry, dir entry update for renaming
423 * directory + target unlink */
424 static inline int ocfs2_rename_credits(struct super_block
*sb
)
426 return 3 * OCFS2_INODE_UPDATE_CREDITS
+ 3 + ocfs2_unlink_credits(sb
);
429 /* global bitmap dinode, group desc., relinked group,
430 * suballocator dinode, group desc., relinked group,
431 * dinode, xattr block */
432 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
433 + OCFS2_INODE_UPDATE_CREDITS \
434 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
437 * Please note that the caller must make sure that root_el is the root
438 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
439 * the result may be wrong.
441 static inline int ocfs2_calc_extend_credits(struct super_block
*sb
,
442 struct ocfs2_extent_list
*root_el
,
445 int bitmap_blocks
, sysfile_bitmap_blocks
, extent_blocks
;
447 /* bitmap dinode, group desc. + relinked group. */
448 bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
;
450 /* we might need to shift tree depth so lets assume an
451 * absolute worst case of complete fragmentation. Even with
452 * that, we only need one update for the dinode, and then
453 * however many metadata chunks needed * a remaining suballoc
455 sysfile_bitmap_blocks
= 1 +
456 (OCFS2_SUBALLOC_ALLOC
- 1) * ocfs2_extend_meta_needed(root_el
);
458 /* this does not include *new* metadata blocks, which are
459 * accounted for in sysfile_bitmap_blocks. root_el +
460 * prev. last_eb_blk + blocks along edge of tree.
461 * calc_symlink_credits passes because we just need 1
462 * credit for the dinode there. */
463 extent_blocks
= 1 + 1 + le16_to_cpu(root_el
->l_tree_depth
);
465 return bitmap_blocks
+ sysfile_bitmap_blocks
+ extent_blocks
+
466 ocfs2_quota_trans_credits(sb
);
469 static inline int ocfs2_calc_symlink_credits(struct super_block
*sb
)
471 int blocks
= ocfs2_mknod_credits(sb
);
473 /* links can be longer than one block so we may update many
474 * within our single allocated extent. */
475 blocks
+= ocfs2_clusters_to_blocks(sb
, 1);
477 return blocks
+ ocfs2_quota_trans_credits(sb
);
480 static inline int ocfs2_calc_group_alloc_credits(struct super_block
*sb
,
484 int bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
+ 1;
485 /* parent inode update + new block group header + bitmap inode update
486 + bitmap blocks affected */
487 blocks
= 1 + 1 + 1 + bitmap_blocks
;
491 static inline int ocfs2_calc_tree_trunc_credits(struct super_block
*sb
,
492 unsigned int clusters_to_del
,
493 struct ocfs2_dinode
*fe
,
494 struct ocfs2_extent_list
*last_el
)
496 /* for dinode + all headers in this pass + update to next leaf */
497 u16 next_free
= le16_to_cpu(last_el
->l_next_free_rec
);
498 u16 tree_depth
= le16_to_cpu(fe
->id2
.i_list
.l_tree_depth
);
499 int credits
= 1 + tree_depth
+ 1;
505 /* We may be deleting metadata blocks, so metadata alloc dinode +
506 one desc. block for each possible delete. */
507 if (tree_depth
&& next_free
== 1 &&
508 ocfs2_rec_clusters(last_el
, &last_el
->l_recs
[i
]) == clusters_to_del
)
509 credits
+= 1 + tree_depth
;
511 /* update to the truncate log. */
512 credits
+= OCFS2_TRUNCATE_LOG_UPDATE
;
514 credits
+= ocfs2_quota_trans_credits(sb
);
519 static inline int ocfs2_jbd2_file_inode(handle_t
*handle
, struct inode
*inode
)
521 return jbd2_journal_file_inode(handle
, &OCFS2_I(inode
)->ip_jinode
);
524 static inline int ocfs2_begin_ordered_truncate(struct inode
*inode
,
527 return jbd2_journal_begin_ordered_truncate(
528 OCFS2_SB(inode
->i_sb
)->journal
->j_journal
,
529 &OCFS2_I(inode
)->ip_jinode
,
533 #endif /* OCFS2_JOURNAL_H */