aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/journal.h
diff options
context:
space:
mode:
authorMark Fasheh <mark.fasheh@oracle.com>2005-12-15 17:31:24 -0500
committerJoel Becker <joel.becker@oracle.com>2006-01-03 14:45:47 -0500
commitccd979bdbce9fba8412beb3f1de68a9d0171b12c (patch)
treec50ed941849ce06ccadd4ce27599b3ef9fdbe2ae /fs/ocfs2/journal.h
parent8df08c89c668e1bd922a053fdb5ba1fadbecbb38 (diff)
[PATCH] OCFS2: The Second Oracle Cluster Filesystem
The OCFS2 file system module. Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Diffstat (limited to 'fs/ocfs2/journal.h')
-rw-r--r--fs/ocfs2/journal.h457
1 files changed, 457 insertions, 0 deletions
diff --git a/fs/ocfs2/journal.h b/fs/ocfs2/journal.h
new file mode 100644
index 000000000000..7d0a816184fa
--- /dev/null
+++ b/fs/ocfs2/journal.h
@@ -0,0 +1,457 @@
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * journal.h
5 *
6 * Defines journalling api and structures.
7 *
8 * Copyright (C) 2003, 2005 Oracle. All rights reserved.
9 *
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.
14 *
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.
19 *
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.
24 */
25
26#ifndef OCFS2_JOURNAL_H
27#define OCFS2_JOURNAL_H
28
29#include <linux/fs.h>
30#include <linux/jbd.h>
31
32#define OCFS2_CHECKPOINT_INTERVAL (8 * HZ)
33
34enum ocfs2_journal_state {
35 OCFS2_JOURNAL_FREE = 0,
36 OCFS2_JOURNAL_LOADED,
37 OCFS2_JOURNAL_IN_SHUTDOWN,
38};
39
40struct ocfs2_super;
41struct ocfs2_dinode;
42struct ocfs2_journal_handle;
43
44struct ocfs2_journal {
45 enum ocfs2_journal_state j_state; /* Journals current state */
46
47 journal_t *j_journal; /* The kernels journal type */
48 struct inode *j_inode; /* Kernel inode pointing to
49 * this journal */
50 struct ocfs2_super *j_osb; /* pointer to the super
51 * block for the node
52 * we're currently
53 * running on -- not
54 * necessarily the super
55 * block from the node
56 * which we usually run
57 * from (recovery,
58 * etc) */
59 struct buffer_head *j_bh; /* Journal disk inode block */
60 atomic_t j_num_trans; /* Number of transactions
61 * currently in the system. */
62 unsigned long j_trans_id;
63 struct rw_semaphore j_trans_barrier;
64 wait_queue_head_t j_checkpointed;
65
66 spinlock_t j_lock;
67 struct list_head j_la_cleanups;
68 struct work_struct j_recovery_work;
69};
70
71extern spinlock_t trans_inc_lock;
72
73/* wrap j_trans_id so we never have it equal to zero. */
74static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
75{
76 unsigned long old_id;
77 spin_lock(&trans_inc_lock);
78 old_id = j->j_trans_id++;
79 if (unlikely(!j->j_trans_id))
80 j->j_trans_id = 1;
81 spin_unlock(&trans_inc_lock);
82 return old_id;
83}
84
85static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal *journal,
86 struct inode *inode)
87{
88 spin_lock(&trans_inc_lock);
89 OCFS2_I(inode)->ip_last_trans = journal->j_trans_id;
90 spin_unlock(&trans_inc_lock);
91}
92
93/* Used to figure out whether it's safe to drop a metadata lock on an
94 * inode. Returns true if all the inodes changes have been
95 * checkpointed to disk. You should be holding the spinlock on the
96 * metadata lock while calling this to be sure that nobody can take
97 * the lock and put it on another transaction. */
98static inline int ocfs2_inode_fully_checkpointed(struct inode *inode)
99{
100 int ret;
101 struct ocfs2_journal *journal = OCFS2_SB(inode->i_sb)->journal;
102
103 spin_lock(&trans_inc_lock);
104 ret = time_after(journal->j_trans_id, OCFS2_I(inode)->ip_last_trans);
105 spin_unlock(&trans_inc_lock);
106 return ret;
107}
108
109/* convenience function to check if an inode is still new (has never
110 * hit disk) Will do you a favor and set created_trans = 0 when you've
111 * been checkpointed. returns '1' if the inode is still new. */
112static inline int ocfs2_inode_is_new(struct inode *inode)
113{
114 int ret;
115
116 /* System files are never "new" as they're written out by
117 * mkfs. This helps us early during mount, before we have the
118 * journal open and j_trans_id could be junk. */
119 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
120 return 0;
121 spin_lock(&trans_inc_lock);
122 ret = !(time_after(OCFS2_SB(inode->i_sb)->journal->j_trans_id,
123 OCFS2_I(inode)->ip_created_trans));
124 if (!ret)
125 OCFS2_I(inode)->ip_created_trans = 0;
126 spin_unlock(&trans_inc_lock);
127 return ret;
128}
129
130static inline void ocfs2_inode_set_new(struct ocfs2_super *osb,
131 struct inode *inode)
132{
133 spin_lock(&trans_inc_lock);
134 OCFS2_I(inode)->ip_created_trans = osb->journal->j_trans_id;
135 spin_unlock(&trans_inc_lock);
136}
137
138extern kmem_cache_t *ocfs2_lock_cache;
139
140struct ocfs2_journal_lock {
141 struct inode *jl_inode;
142 struct list_head jl_lock_list;
143};
144
145struct ocfs2_journal_handle {
146 handle_t *k_handle; /* kernel handle. */
147 struct ocfs2_journal *journal;
148 u32 flags; /* see flags below. */
149 int max_buffs; /* Buffs reserved by this handle */
150
151 /* The following two fields are for ocfs2_handle_add_lock */
152 int num_locks;
153 struct list_head locks; /* A bunch of locks to
154 * release on commit. This
155 * should be a list_head */
156
157 struct list_head inode_list;
158};
159
160#define OCFS2_HANDLE_STARTED 1
161/* should we sync-commit this handle? */
162#define OCFS2_HANDLE_SYNC 2
163static inline int ocfs2_handle_started(struct ocfs2_journal_handle *handle)
164{
165 return handle->flags & OCFS2_HANDLE_STARTED;
166}
167
168static inline void ocfs2_handle_set_sync(struct ocfs2_journal_handle *handle, int sync)
169{
170 if (sync)
171 handle->flags |= OCFS2_HANDLE_SYNC;
172 else
173 handle->flags &= ~OCFS2_HANDLE_SYNC;
174}
175
176/* Exported only for the journal struct init code in super.c. Do not call. */
177void ocfs2_complete_recovery(void *data);
178
179/*
180 * Journal Control:
181 * Initialize, Load, Shutdown, Wipe a journal.
182 *
183 * ocfs2_journal_init - Initialize journal structures in the OSB.
184 * ocfs2_journal_load - Load the given journal off disk. Replay it if
185 * there's transactions still in there.
186 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
187 * uncommitted, uncheckpointed transactions.
188 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
189 * zero out each block.
190 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
191 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
192 * event on.
193 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
194 */
195void ocfs2_set_journal_params(struct ocfs2_super *osb);
196int ocfs2_journal_init(struct ocfs2_journal *journal,
197 int *dirty);
198void ocfs2_journal_shutdown(struct ocfs2_super *osb);
199int ocfs2_journal_wipe(struct ocfs2_journal *journal,
200 int full);
201int ocfs2_journal_load(struct ocfs2_journal *journal);
202int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
203void ocfs2_recovery_thread(struct ocfs2_super *osb,
204 int node_num);
205int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
206void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
207
208static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
209{
210 atomic_set(&osb->needs_checkpoint, 1);
211 wake_up(&osb->checkpoint_event);
212}
213
214static inline void ocfs2_checkpoint_inode(struct inode *inode)
215{
216 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
217
218 if (!ocfs2_inode_fully_checkpointed(inode)) {
219 /* WARNING: This only kicks off a single
220 * checkpoint. If someone races you and adds more
221 * metadata to the journal, you won't know, and will
222 * wind up waiting *alot* longer than necessary. Right
223 * now we only use this in clear_inode so that's
224 * OK. */
225 ocfs2_start_checkpoint(osb);
226
227 wait_event(osb->journal->j_checkpointed,
228 ocfs2_inode_fully_checkpointed(inode));
229 }
230}
231
232/*
233 * Transaction Handling:
234 * Manage the lifetime of a transaction handle.
235 *
236 * ocfs2_alloc_handle - Only allocate a handle so we can start putting
237 * cluster locks on it. To actually change blocks,
238 * call ocfs2_start_trans with the handle returned
239 * from this function. You may call ocfs2_commit_trans
240 * at any time in the lifetime of a handle.
241 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
242 * the number of blocks that will be changed during
243 * this handle.
244 * ocfs2_commit_trans - Complete a handle.
245 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
246 * commit the handle to disk in the process, but will
247 * not release any locks taken during the transaction.
248 * ocfs2_journal_access - Notify the handle that we want to journal this
249 * buffer. Will have to call ocfs2_journal_dirty once
250 * we've actually dirtied it. Type is one of . or .
251 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
252 * ocfs2_journal_dirty_data - Indicate that a data buffer should go out before
253 * the current handle commits.
254 * ocfs2_handle_add_lock - Sometimes we need to delay lock release
255 * until after a transaction has been completed. Use
256 * ocfs2_handle_add_lock to indicate that a lock needs
257 * to be released at the end of that handle. Locks
258 * will be released in the order that they are added.
259 * ocfs2_handle_add_inode - Add a locked inode to a transaction.
260 */
261
262/* You must always start_trans with a number of buffs > 0, but it's
263 * perfectly legal to go through an entire transaction without having
264 * dirtied any buffers. */
265struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb);
266struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb,
267 struct ocfs2_journal_handle *handle,
268 int max_buffs);
269void ocfs2_commit_trans(struct ocfs2_journal_handle *handle);
270int ocfs2_extend_trans(struct ocfs2_journal_handle *handle,
271 int nblocks);
272
273/*
274 * Create access is for when we get a newly created buffer and we're
275 * not gonna read it off disk, but rather fill it ourselves. Right
276 * now, we don't do anything special with this (it turns into a write
277 * request), but this is a good placeholder in case we do...
278 *
279 * Write access is for when we read a block off disk and are going to
280 * modify it. This way the journalling layer knows it may need to make
281 * a copy of that block (if it's part of another, uncommitted
282 * transaction) before we do so.
283 */
284#define OCFS2_JOURNAL_ACCESS_CREATE 0
285#define OCFS2_JOURNAL_ACCESS_WRITE 1
286#define OCFS2_JOURNAL_ACCESS_UNDO 2
287
288int ocfs2_journal_access(struct ocfs2_journal_handle *handle,
289 struct inode *inode,
290 struct buffer_head *bh,
291 int type);
292/*
293 * A word about the journal_access/journal_dirty "dance". It is
294 * entirely legal to journal_access a buffer more than once (as long
295 * as the access type is the same -- I'm not sure what will happen if
296 * access type is different but this should never happen anyway) It is
297 * also legal to journal_dirty a buffer more than once. In fact, you
298 * can even journal_access a buffer after you've done a
299 * journal_access/journal_dirty pair. The only thing you cannot do
300 * however, is journal_dirty a buffer which you haven't yet passed to
301 * journal_access at least once.
302 *
303 * That said, 99% of the time this doesn't matter and this is what the
304 * path looks like:
305 *
306 * <read a bh>
307 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
308 * <modify the bh>
309 * ocfs2_journal_dirty(handle, bh);
310 */
311int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle,
312 struct buffer_head *bh);
313int ocfs2_journal_dirty_data(handle_t *handle,
314 struct buffer_head *bh);
315int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle,
316 struct inode *inode);
317/*
318 * Use this to protect from other processes reading buffer state while
319 * it's in flight.
320 */
321void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle,
322 struct inode *inode);
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/* get one bit out of a suballocator: dinode + group descriptor +
337 * prev. group desc. if we relink. */
338#define OCFS2_SUBALLOC_ALLOC (3)
339
340/* dinode + group descriptor update. We don't relink on free yet. */
341#define OCFS2_SUBALLOC_FREE (2)
342
343#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
344#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
345 + OCFS2_TRUNCATE_LOG_UPDATE)
346
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)
350
351/* parent fe, parent block, new file entry, inode alloc fe, inode alloc
352 * group descriptor + mkdir/symlink blocks */
353#define OCFS2_MKNOD_CREDITS (3 + OCFS2_SUBALLOC_ALLOC \
354 + OCFS2_DIR_LINK_ADDITIONAL_CREDITS)
355
356/* local alloc metadata change + main bitmap updates */
357#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
358 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
359
360/* used when we don't need an allocation change for a dir extend. One
361 * for the dinode, one for the new block. */
362#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
363
364/* file update (nlink, etc) + dir entry block */
365#define OCFS2_LINK_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
366
367/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
368 * dir inode link */
369#define OCFS2_UNLINK_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 1 \
370 + OCFS2_LINK_CREDITS)
371
372/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
373 * inode alloc group descriptor */
374#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
375
376/* dinode update, old dir dinode update, new dir dinode update, old
377 * dir dir entry, new dir dir entry, dir entry update for renaming
378 * directory + target unlink */
379#define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3 \
380 + OCFS2_UNLINK_CREDITS)
381
382static inline int ocfs2_calc_extend_credits(struct super_block *sb,
383 struct ocfs2_dinode *fe,
384 u32 bits_wanted)
385{
386 int bitmap_blocks, sysfile_bitmap_blocks, dinode_blocks;
387
388 /* bitmap dinode, group desc. + relinked group. */
389 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
390
391 /* we might need to shift tree depth so lets assume an
392 * absolute worst case of complete fragmentation. Even with
393 * that, we only need one update for the dinode, and then
394 * however many metadata chunks needed * a remaining suballoc
395 * alloc. */
396 sysfile_bitmap_blocks = 1 +
397 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(fe);
398
399 /* this does not include *new* metadata blocks, which are
400 * accounted for in sysfile_bitmap_blocks. fe +
401 * prev. last_eb_blk + blocks along edge of tree.
402 * calc_symlink_credits passes because we just need 1
403 * credit for the dinode there. */
404 dinode_blocks = 1 + 1 + le16_to_cpu(fe->id2.i_list.l_tree_depth);
405
406 return bitmap_blocks + sysfile_bitmap_blocks + dinode_blocks;
407}
408
409static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
410{
411 int blocks = OCFS2_MKNOD_CREDITS;
412
413 /* links can be longer than one block so we may update many
414 * within our single allocated extent. */
415 blocks += ocfs2_clusters_to_blocks(sb, 1);
416
417 return blocks;
418}
419
420static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
421 unsigned int cpg)
422{
423 int blocks;
424 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
425 /* parent inode update + new block group header + bitmap inode update
426 + bitmap blocks affected */
427 blocks = 1 + 1 + 1 + bitmap_blocks;
428 return blocks;
429}
430
431static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
432 unsigned int clusters_to_del,
433 struct ocfs2_dinode *fe,
434 struct ocfs2_extent_list *last_el)
435{
436 /* for dinode + all headers in this pass + update to next leaf */
437 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
438 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
439 int credits = 1 + tree_depth + 1;
440 int i;
441
442 i = next_free - 1;
443 BUG_ON(i < 0);
444
445 /* We may be deleting metadata blocks, so metadata alloc dinode +
446 one desc. block for each possible delete. */
447 if (tree_depth && next_free == 1 &&
448 le32_to_cpu(last_el->l_recs[i].e_clusters) == clusters_to_del)
449 credits += 1 + tree_depth;
450
451 /* update to the truncate log. */
452 credits += OCFS2_TRUNCATE_LOG_UPDATE;
453
454 return credits;
455}
456
457#endif /* OCFS2_JOURNAL_H */