blob: 2f3a6acdac452e590e67e1235e3225c8975aeb8e [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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
Mark Fashehccd979b2005-12-15 14:31:24 -080032enum ocfs2_journal_state {
33 OCFS2_JOURNAL_FREE = 0,
34 OCFS2_JOURNAL_LOADED,
35 OCFS2_JOURNAL_IN_SHUTDOWN,
36};
37
38struct ocfs2_super;
39struct ocfs2_dinode;
40struct ocfs2_journal_handle;
41
42struct ocfs2_journal {
43 enum ocfs2_journal_state j_state; /* Journals current state */
44
45 journal_t *j_journal; /* The kernels journal type */
46 struct inode *j_inode; /* Kernel inode pointing to
47 * this journal */
48 struct ocfs2_super *j_osb; /* pointer to the super
49 * block for the node
50 * we're currently
51 * running on -- not
52 * necessarily the super
53 * block from the node
54 * which we usually run
55 * from (recovery,
56 * etc) */
57 struct buffer_head *j_bh; /* Journal disk inode block */
58 atomic_t j_num_trans; /* Number of transactions
59 * currently in the system. */
60 unsigned long j_trans_id;
61 struct rw_semaphore j_trans_barrier;
62 wait_queue_head_t j_checkpointed;
63
64 spinlock_t j_lock;
65 struct list_head j_la_cleanups;
66 struct work_struct j_recovery_work;
67};
68
69extern spinlock_t trans_inc_lock;
70
71/* wrap j_trans_id so we never have it equal to zero. */
72static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
73{
74 unsigned long old_id;
75 spin_lock(&trans_inc_lock);
76 old_id = j->j_trans_id++;
77 if (unlikely(!j->j_trans_id))
78 j->j_trans_id = 1;
79 spin_unlock(&trans_inc_lock);
80 return old_id;
81}
82
83static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal *journal,
84 struct inode *inode)
85{
86 spin_lock(&trans_inc_lock);
87 OCFS2_I(inode)->ip_last_trans = journal->j_trans_id;
88 spin_unlock(&trans_inc_lock);
89}
90
91/* Used to figure out whether it's safe to drop a metadata lock on an
92 * inode. Returns true if all the inodes changes have been
93 * checkpointed to disk. You should be holding the spinlock on the
94 * metadata lock while calling this to be sure that nobody can take
95 * the lock and put it on another transaction. */
96static inline int ocfs2_inode_fully_checkpointed(struct inode *inode)
97{
98 int ret;
99 struct ocfs2_journal *journal = OCFS2_SB(inode->i_sb)->journal;
100
101 spin_lock(&trans_inc_lock);
102 ret = time_after(journal->j_trans_id, OCFS2_I(inode)->ip_last_trans);
103 spin_unlock(&trans_inc_lock);
104 return ret;
105}
106
107/* convenience function to check if an inode is still new (has never
108 * hit disk) Will do you a favor and set created_trans = 0 when you've
109 * been checkpointed. returns '1' if the inode is still new. */
110static inline int ocfs2_inode_is_new(struct inode *inode)
111{
112 int ret;
113
114 /* System files are never "new" as they're written out by
115 * mkfs. This helps us early during mount, before we have the
116 * journal open and j_trans_id could be junk. */
117 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
118 return 0;
119 spin_lock(&trans_inc_lock);
120 ret = !(time_after(OCFS2_SB(inode->i_sb)->journal->j_trans_id,
121 OCFS2_I(inode)->ip_created_trans));
122 if (!ret)
123 OCFS2_I(inode)->ip_created_trans = 0;
124 spin_unlock(&trans_inc_lock);
125 return ret;
126}
127
128static inline void ocfs2_inode_set_new(struct ocfs2_super *osb,
129 struct inode *inode)
130{
131 spin_lock(&trans_inc_lock);
132 OCFS2_I(inode)->ip_created_trans = osb->journal->j_trans_id;
133 spin_unlock(&trans_inc_lock);
134}
135
136extern kmem_cache_t *ocfs2_lock_cache;
137
138struct ocfs2_journal_lock {
139 struct inode *jl_inode;
140 struct list_head jl_lock_list;
141};
142
143struct ocfs2_journal_handle {
144 handle_t *k_handle; /* kernel handle. */
145 struct ocfs2_journal *journal;
146 u32 flags; /* see flags below. */
147 int max_buffs; /* Buffs reserved by this handle */
148
149 /* The following two fields are for ocfs2_handle_add_lock */
150 int num_locks;
151 struct list_head locks; /* A bunch of locks to
152 * release on commit. This
153 * should be a list_head */
154
155 struct list_head inode_list;
156};
157
158#define OCFS2_HANDLE_STARTED 1
159/* should we sync-commit this handle? */
160#define OCFS2_HANDLE_SYNC 2
161static inline int ocfs2_handle_started(struct ocfs2_journal_handle *handle)
162{
163 return handle->flags & OCFS2_HANDLE_STARTED;
164}
165
166static inline void ocfs2_handle_set_sync(struct ocfs2_journal_handle *handle, int sync)
167{
168 if (sync)
169 handle->flags |= OCFS2_HANDLE_SYNC;
170 else
171 handle->flags &= ~OCFS2_HANDLE_SYNC;
172}
173
174/* Exported only for the journal struct init code in super.c. Do not call. */
175void ocfs2_complete_recovery(void *data);
176
177/*
178 * Journal Control:
179 * Initialize, Load, Shutdown, Wipe a journal.
180 *
181 * ocfs2_journal_init - Initialize journal structures in the OSB.
182 * ocfs2_journal_load - Load the given journal off disk. Replay it if
183 * there's transactions still in there.
184 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
185 * uncommitted, uncheckpointed transactions.
186 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
187 * zero out each block.
188 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
189 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
190 * event on.
191 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
192 */
193void ocfs2_set_journal_params(struct ocfs2_super *osb);
194int ocfs2_journal_init(struct ocfs2_journal *journal,
195 int *dirty);
196void ocfs2_journal_shutdown(struct ocfs2_super *osb);
197int ocfs2_journal_wipe(struct ocfs2_journal *journal,
198 int full);
199int ocfs2_journal_load(struct ocfs2_journal *journal);
200int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
201void ocfs2_recovery_thread(struct ocfs2_super *osb,
202 int node_num);
203int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
204void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
205
206static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
207{
208 atomic_set(&osb->needs_checkpoint, 1);
209 wake_up(&osb->checkpoint_event);
210}
211
212static inline void ocfs2_checkpoint_inode(struct inode *inode)
213{
214 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
215
216 if (!ocfs2_inode_fully_checkpointed(inode)) {
217 /* WARNING: This only kicks off a single
218 * checkpoint. If someone races you and adds more
219 * metadata to the journal, you won't know, and will
220 * wind up waiting *alot* longer than necessary. Right
221 * now we only use this in clear_inode so that's
222 * OK. */
223 ocfs2_start_checkpoint(osb);
224
225 wait_event(osb->journal->j_checkpointed,
226 ocfs2_inode_fully_checkpointed(inode));
227 }
228}
229
230/*
231 * Transaction Handling:
232 * Manage the lifetime of a transaction handle.
233 *
234 * ocfs2_alloc_handle - Only allocate a handle so we can start putting
235 * cluster locks on it. To actually change blocks,
236 * call ocfs2_start_trans with the handle returned
237 * from this function. You may call ocfs2_commit_trans
238 * at any time in the lifetime of a handle.
239 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
240 * the number of blocks that will be changed during
241 * this handle.
242 * ocfs2_commit_trans - Complete a handle.
243 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
244 * commit the handle to disk in the process, but will
245 * not release any locks taken during the transaction.
246 * ocfs2_journal_access - Notify the handle that we want to journal this
247 * buffer. Will have to call ocfs2_journal_dirty once
248 * we've actually dirtied it. Type is one of . or .
249 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
250 * ocfs2_journal_dirty_data - Indicate that a data buffer should go out before
251 * the current handle commits.
252 * ocfs2_handle_add_lock - Sometimes we need to delay lock release
253 * until after a transaction has been completed. Use
254 * ocfs2_handle_add_lock to indicate that a lock needs
255 * to be released at the end of that handle. Locks
256 * will be released in the order that they are added.
257 * ocfs2_handle_add_inode - Add a locked inode to a transaction.
258 */
259
260/* You must always start_trans with a number of buffs > 0, but it's
261 * perfectly legal to go through an entire transaction without having
262 * dirtied any buffers. */
263struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb);
264struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb,
265 struct ocfs2_journal_handle *handle,
266 int max_buffs);
267void ocfs2_commit_trans(struct ocfs2_journal_handle *handle);
268int ocfs2_extend_trans(struct ocfs2_journal_handle *handle,
269 int nblocks);
270
271/*
272 * Create access is for when we get a newly created buffer and we're
273 * not gonna read it off disk, but rather fill it ourselves. Right
274 * now, we don't do anything special with this (it turns into a write
275 * request), but this is a good placeholder in case we do...
276 *
277 * Write access is for when we read a block off disk and are going to
278 * modify it. This way the journalling layer knows it may need to make
279 * a copy of that block (if it's part of another, uncommitted
280 * transaction) before we do so.
281 */
282#define OCFS2_JOURNAL_ACCESS_CREATE 0
283#define OCFS2_JOURNAL_ACCESS_WRITE 1
284#define OCFS2_JOURNAL_ACCESS_UNDO 2
285
286int ocfs2_journal_access(struct ocfs2_journal_handle *handle,
287 struct inode *inode,
288 struct buffer_head *bh,
289 int type);
290/*
291 * A word about the journal_access/journal_dirty "dance". It is
292 * entirely legal to journal_access a buffer more than once (as long
293 * as the access type is the same -- I'm not sure what will happen if
294 * access type is different but this should never happen anyway) It is
295 * also legal to journal_dirty a buffer more than once. In fact, you
296 * can even journal_access a buffer after you've done a
297 * journal_access/journal_dirty pair. The only thing you cannot do
298 * however, is journal_dirty a buffer which you haven't yet passed to
299 * journal_access at least once.
300 *
301 * That said, 99% of the time this doesn't matter and this is what the
302 * path looks like:
303 *
304 * <read a bh>
305 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
306 * <modify the bh>
307 * ocfs2_journal_dirty(handle, bh);
308 */
309int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle,
310 struct buffer_head *bh);
311int ocfs2_journal_dirty_data(handle_t *handle,
312 struct buffer_head *bh);
313int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle,
314 struct inode *inode);
315/*
316 * Use this to protect from other processes reading buffer state while
317 * it's in flight.
318 */
319void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle,
320 struct inode *inode);
321
322/*
323 * Credit Macros:
324 * Convenience macros to calculate number of credits needed.
325 *
326 * For convenience sake, I have a set of macros here which calculate
327 * the *maximum* number of sectors which will be changed for various
328 * metadata updates.
329 */
330
331/* simple file updates like chmod, etc. */
332#define OCFS2_INODE_UPDATE_CREDITS 1
333
334/* get one bit out of a suballocator: dinode + group descriptor +
335 * prev. group desc. if we relink. */
336#define OCFS2_SUBALLOC_ALLOC (3)
337
338/* dinode + group descriptor update. We don't relink on free yet. */
339#define OCFS2_SUBALLOC_FREE (2)
340
341#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
342#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
343 + OCFS2_TRUNCATE_LOG_UPDATE)
344
345/* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
346 * bitmap block for the new bit) */
347#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
348
349/* parent fe, parent block, new file entry, inode alloc fe, inode alloc
350 * group descriptor + mkdir/symlink blocks */
351#define OCFS2_MKNOD_CREDITS (3 + OCFS2_SUBALLOC_ALLOC \
352 + OCFS2_DIR_LINK_ADDITIONAL_CREDITS)
353
354/* local alloc metadata change + main bitmap updates */
355#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
356 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
357
358/* used when we don't need an allocation change for a dir extend. One
359 * for the dinode, one for the new block. */
360#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
361
362/* file update (nlink, etc) + dir entry block */
363#define OCFS2_LINK_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
364
365/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
366 * dir inode link */
367#define OCFS2_UNLINK_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 1 \
368 + OCFS2_LINK_CREDITS)
369
370/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
371 * inode alloc group descriptor */
372#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
373
374/* dinode update, old dir dinode update, new dir dinode update, old
375 * dir dir entry, new dir dir entry, dir entry update for renaming
376 * directory + target unlink */
377#define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3 \
378 + OCFS2_UNLINK_CREDITS)
379
380static inline int ocfs2_calc_extend_credits(struct super_block *sb,
381 struct ocfs2_dinode *fe,
382 u32 bits_wanted)
383{
384 int bitmap_blocks, sysfile_bitmap_blocks, dinode_blocks;
385
386 /* bitmap dinode, group desc. + relinked group. */
387 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
388
389 /* we might need to shift tree depth so lets assume an
390 * absolute worst case of complete fragmentation. Even with
391 * that, we only need one update for the dinode, and then
392 * however many metadata chunks needed * a remaining suballoc
393 * alloc. */
394 sysfile_bitmap_blocks = 1 +
395 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(fe);
396
397 /* this does not include *new* metadata blocks, which are
398 * accounted for in sysfile_bitmap_blocks. fe +
399 * prev. last_eb_blk + blocks along edge of tree.
400 * calc_symlink_credits passes because we just need 1
401 * credit for the dinode there. */
402 dinode_blocks = 1 + 1 + le16_to_cpu(fe->id2.i_list.l_tree_depth);
403
404 return bitmap_blocks + sysfile_bitmap_blocks + dinode_blocks;
405}
406
407static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
408{
409 int blocks = OCFS2_MKNOD_CREDITS;
410
411 /* links can be longer than one block so we may update many
412 * within our single allocated extent. */
413 blocks += ocfs2_clusters_to_blocks(sb, 1);
414
415 return blocks;
416}
417
418static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
419 unsigned int cpg)
420{
421 int blocks;
422 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
423 /* parent inode update + new block group header + bitmap inode update
424 + bitmap blocks affected */
425 blocks = 1 + 1 + 1 + bitmap_blocks;
426 return blocks;
427}
428
429static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
430 unsigned int clusters_to_del,
431 struct ocfs2_dinode *fe,
432 struct ocfs2_extent_list *last_el)
433{
434 /* for dinode + all headers in this pass + update to next leaf */
435 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
436 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
437 int credits = 1 + tree_depth + 1;
438 int i;
439
440 i = next_free - 1;
441 BUG_ON(i < 0);
442
443 /* We may be deleting metadata blocks, so metadata alloc dinode +
444 one desc. block for each possible delete. */
445 if (tree_depth && next_free == 1 &&
446 le32_to_cpu(last_el->l_recs[i].e_clusters) == clusters_to_del)
447 credits += 1 + tree_depth;
448
449 /* update to the truncate log. */
450 credits += OCFS2_TRUNCATE_LOG_UPDATE;
451
452 return credits;
453}
454
455#endif /* OCFS2_JOURNAL_H */