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/jbd.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(void *data
);
140 * Initialize, Load, Shutdown, Wipe a journal.
142 * ocfs2_journal_init - Initialize journal structures in the OSB.
143 * ocfs2_journal_load - Load the given journal off disk. Replay it if
144 * there's transactions still in there.
145 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
146 * uncommitted, uncheckpointed transactions.
147 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
148 * zero out each block.
149 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
150 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
152 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
154 void ocfs2_set_journal_params(struct ocfs2_super
*osb
);
155 int ocfs2_journal_init(struct ocfs2_journal
*journal
,
157 void ocfs2_journal_shutdown(struct ocfs2_super
*osb
);
158 int ocfs2_journal_wipe(struct ocfs2_journal
*journal
,
160 int ocfs2_journal_load(struct ocfs2_journal
*journal
);
161 int ocfs2_check_journals_nolocks(struct ocfs2_super
*osb
);
162 void ocfs2_recovery_thread(struct ocfs2_super
*osb
,
164 int ocfs2_mark_dead_nodes(struct ocfs2_super
*osb
);
165 void ocfs2_complete_mount_recovery(struct ocfs2_super
*osb
);
167 static inline void ocfs2_start_checkpoint(struct ocfs2_super
*osb
)
169 atomic_set(&osb
->needs_checkpoint
, 1);
170 wake_up(&osb
->checkpoint_event
);
173 static inline void ocfs2_checkpoint_inode(struct inode
*inode
)
175 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
177 if (!ocfs2_inode_fully_checkpointed(inode
)) {
178 /* WARNING: This only kicks off a single
179 * checkpoint. If someone races you and adds more
180 * metadata to the journal, you won't know, and will
181 * wind up waiting *alot* longer than necessary. Right
182 * now we only use this in clear_inode so that's
184 ocfs2_start_checkpoint(osb
);
186 wait_event(osb
->journal
->j_checkpointed
,
187 ocfs2_inode_fully_checkpointed(inode
));
192 * Transaction Handling:
193 * Manage the lifetime of a transaction handle.
195 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
196 * the number of blocks that will be changed during
198 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
199 * the journal was aborted. The majority of paths don't
200 * check the return value as an error there comes too
201 * late to do anything (and will be picked up in a
202 * later transaction).
203 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
204 * commit the handle to disk in the process, but will
205 * not release any locks taken during the transaction.
206 * ocfs2_journal_access - Notify the handle that we want to journal this
207 * buffer. Will have to call ocfs2_journal_dirty once
208 * we've actually dirtied it. Type is one of . or .
209 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
210 * ocfs2_journal_dirty_data - Indicate that a data buffer should go out before
211 * the current handle commits.
214 /* You must always start_trans with a number of buffs > 0, but it's
215 * perfectly legal to go through an entire transaction without having
216 * dirtied any buffers. */
217 handle_t
*ocfs2_start_trans(struct ocfs2_super
*osb
,
219 int ocfs2_commit_trans(struct ocfs2_super
*osb
,
221 int ocfs2_extend_trans(handle_t
*handle
, int nblocks
);
224 * Create access is for when we get a newly created buffer and we're
225 * not gonna read it off disk, but rather fill it ourselves. Right
226 * now, we don't do anything special with this (it turns into a write
227 * request), but this is a good placeholder in case we do...
229 * Write access is for when we read a block off disk and are going to
230 * modify it. This way the journalling layer knows it may need to make
231 * a copy of that block (if it's part of another, uncommitted
232 * transaction) before we do so.
234 #define OCFS2_JOURNAL_ACCESS_CREATE 0
235 #define OCFS2_JOURNAL_ACCESS_WRITE 1
236 #define OCFS2_JOURNAL_ACCESS_UNDO 2
238 int ocfs2_journal_access(handle_t
*handle
,
240 struct buffer_head
*bh
,
243 * A word about the journal_access/journal_dirty "dance". It is
244 * entirely legal to journal_access a buffer more than once (as long
245 * as the access type is the same -- I'm not sure what will happen if
246 * access type is different but this should never happen anyway) It is
247 * also legal to journal_dirty a buffer more than once. In fact, you
248 * can even journal_access a buffer after you've done a
249 * journal_access/journal_dirty pair. The only thing you cannot do
250 * however, is journal_dirty a buffer which you haven't yet passed to
251 * journal_access at least once.
253 * That said, 99% of the time this doesn't matter and this is what the
257 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
259 * ocfs2_journal_dirty(handle, bh);
261 int ocfs2_journal_dirty(handle_t
*handle
,
262 struct buffer_head
*bh
);
263 int ocfs2_journal_dirty_data(handle_t
*handle
,
264 struct buffer_head
*bh
);
268 * Convenience macros to calculate number of credits needed.
270 * For convenience sake, I have a set of macros here which calculate
271 * the *maximum* number of sectors which will be changed for various
275 /* simple file updates like chmod, etc. */
276 #define OCFS2_INODE_UPDATE_CREDITS 1
278 /* get one bit out of a suballocator: dinode + group descriptor +
279 * prev. group desc. if we relink. */
280 #define OCFS2_SUBALLOC_ALLOC (3)
282 /* dinode + group descriptor update. We don't relink on free yet. */
283 #define OCFS2_SUBALLOC_FREE (2)
285 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
286 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
287 + OCFS2_TRUNCATE_LOG_UPDATE)
289 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
290 * bitmap block for the new bit) */
291 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
293 /* parent fe, parent block, new file entry, inode alloc fe, inode alloc
294 * group descriptor + mkdir/symlink blocks */
295 #define OCFS2_MKNOD_CREDITS (3 + OCFS2_SUBALLOC_ALLOC \
296 + OCFS2_DIR_LINK_ADDITIONAL_CREDITS)
298 /* local alloc metadata change + main bitmap updates */
299 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
300 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
302 /* used when we don't need an allocation change for a dir extend. One
303 * for the dinode, one for the new block. */
304 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
306 /* file update (nlink, etc) + dir entry block */
307 #define OCFS2_LINK_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
309 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
311 #define OCFS2_UNLINK_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 1 \
312 + OCFS2_LINK_CREDITS)
314 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
315 * inode alloc group descriptor */
316 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
318 /* dinode update, old dir dinode update, new dir dinode update, old
319 * dir dir entry, new dir dir entry, dir entry update for renaming
320 * directory + target unlink */
321 #define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3 \
322 + OCFS2_UNLINK_CREDITS)
324 static inline int ocfs2_calc_extend_credits(struct super_block
*sb
,
325 struct ocfs2_dinode
*fe
,
328 int bitmap_blocks
, sysfile_bitmap_blocks
, dinode_blocks
;
330 /* bitmap dinode, group desc. + relinked group. */
331 bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
;
333 /* we might need to shift tree depth so lets assume an
334 * absolute worst case of complete fragmentation. Even with
335 * that, we only need one update for the dinode, and then
336 * however many metadata chunks needed * a remaining suballoc
338 sysfile_bitmap_blocks
= 1 +
339 (OCFS2_SUBALLOC_ALLOC
- 1) * ocfs2_extend_meta_needed(fe
);
341 /* this does not include *new* metadata blocks, which are
342 * accounted for in sysfile_bitmap_blocks. fe +
343 * prev. last_eb_blk + blocks along edge of tree.
344 * calc_symlink_credits passes because we just need 1
345 * credit for the dinode there. */
346 dinode_blocks
= 1 + 1 + le16_to_cpu(fe
->id2
.i_list
.l_tree_depth
);
348 return bitmap_blocks
+ sysfile_bitmap_blocks
+ dinode_blocks
;
351 static inline int ocfs2_calc_symlink_credits(struct super_block
*sb
)
353 int blocks
= OCFS2_MKNOD_CREDITS
;
355 /* links can be longer than one block so we may update many
356 * within our single allocated extent. */
357 blocks
+= ocfs2_clusters_to_blocks(sb
, 1);
362 static inline int ocfs2_calc_group_alloc_credits(struct super_block
*sb
,
366 int bitmap_blocks
= OCFS2_SUBALLOC_ALLOC
+ 1;
367 /* parent inode update + new block group header + bitmap inode update
368 + bitmap blocks affected */
369 blocks
= 1 + 1 + 1 + bitmap_blocks
;
373 static inline int ocfs2_calc_tree_trunc_credits(struct super_block
*sb
,
374 unsigned int clusters_to_del
,
375 struct ocfs2_dinode
*fe
,
376 struct ocfs2_extent_list
*last_el
)
378 /* for dinode + all headers in this pass + update to next leaf */
379 u16 next_free
= le16_to_cpu(last_el
->l_next_free_rec
);
380 u16 tree_depth
= le16_to_cpu(fe
->id2
.i_list
.l_tree_depth
);
381 int credits
= 1 + tree_depth
+ 1;
387 /* We may be deleting metadata blocks, so metadata alloc dinode +
388 one desc. block for each possible delete. */
389 if (tree_depth
&& next_free
== 1 &&
390 le32_to_cpu(last_el
->l_recs
[i
].e_clusters
) == clusters_to_del
)
391 credits
+= 1 + tree_depth
;
393 /* update to the truncate log. */
394 credits
+= OCFS2_TRUNCATE_LOG_UPDATE
;
399 #endif /* OCFS2_JOURNAL_H */