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 INODE_CACHE(inode)->ci_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,
113 INODE_CACHE(inode)->ci_last_trans);
114 spin_unlock(&trans_inc_lock);
118 /* convenience function to check if an inode is still new (has never
119 * hit disk) Will do you a favor and set created_trans = 0 when you've
120 * been checkpointed. returns '1' if the inode is still new. */
121 static inline int ocfs2_inode_is_new(struct inode *inode)
125 /* System files are never "new" as they're written out by
126 * mkfs. This helps us early during mount, before we have the
127 * journal open and j_trans_id could be junk. */
128 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
130 spin_lock(&trans_inc_lock);
131 ret = !(time_after(OCFS2_SB(inode->i_sb)->journal->j_trans_id,
132 INODE_CACHE(inode)->ci_created_trans));
134 INODE_CACHE(inode)->ci_created_trans = 0;
135 spin_unlock(&trans_inc_lock);
139 static inline void ocfs2_inode_set_new(struct ocfs2_super *osb,
142 spin_lock(&trans_inc_lock);
143 INODE_CACHE(inode)->ci_created_trans = osb->journal->j_trans_id;
144 spin_unlock(&trans_inc_lock);
147 /* Exported only for the journal struct init code in super.c. Do not call. */
148 void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
149 void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
150 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
151 void ocfs2_orphan_scan_exit(struct ocfs2_super *osb);
153 void ocfs2_complete_recovery(struct work_struct *work);
154 void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
156 int ocfs2_recovery_init(struct ocfs2_super *osb);
157 void ocfs2_recovery_exit(struct ocfs2_super *osb);
159 int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
162 * Initialize, Load, Shutdown, Wipe a journal.
164 * ocfs2_journal_init - Initialize journal structures in the OSB.
165 * ocfs2_journal_load - Load the given journal off disk. Replay it if
166 * there's transactions still in there.
167 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
168 * uncommitted, uncheckpointed transactions.
169 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
170 * zero out each block.
171 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
172 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
174 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
176 void ocfs2_set_journal_params(struct ocfs2_super *osb);
177 int ocfs2_journal_init(struct ocfs2_journal *journal,
179 void ocfs2_journal_shutdown(struct ocfs2_super *osb);
180 int ocfs2_journal_wipe(struct ocfs2_journal *journal,
182 int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
184 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
185 void ocfs2_recovery_thread(struct ocfs2_super *osb,
187 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
188 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
189 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
191 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
193 atomic_set(&osb->needs_checkpoint, 1);
194 wake_up(&osb->checkpoint_event);
197 static inline void ocfs2_checkpoint_inode(struct inode *inode)
199 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
201 if (ocfs2_mount_local(osb))
204 if (!ocfs2_inode_fully_checkpointed(inode)) {
205 /* WARNING: This only kicks off a single
206 * checkpoint. If someone races you and adds more
207 * metadata to the journal, you won't know, and will
208 * wind up waiting *alot* longer than necessary. Right
209 * now we only use this in clear_inode so that's
211 ocfs2_start_checkpoint(osb);
213 wait_event(osb->journal->j_checkpointed,
214 ocfs2_inode_fully_checkpointed(inode));
219 * Transaction Handling:
220 * Manage the lifetime of a transaction handle.
222 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
223 * the number of blocks that will be changed during
225 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
226 * the journal was aborted. The majority of paths don't
227 * check the return value as an error there comes too
228 * late to do anything (and will be picked up in a
229 * later transaction).
230 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
231 * commit the handle to disk in the process, but will
232 * not release any locks taken during the transaction.
233 * ocfs2_journal_access* - Notify the handle that we want to journal this
234 * buffer. Will have to call ocfs2_journal_dirty once
235 * we've actually dirtied it. Type is one of . or .
236 * Always call the specific flavor of
237 * ocfs2_journal_access_*() unless you intend to
238 * manage the checksum by hand.
239 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
240 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
241 * the current handle commits.
244 /* You must always start_trans with a number of buffs > 0, but it's
245 * perfectly legal to go through an entire transaction without having
246 * dirtied any buffers. */
247 handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
249 int ocfs2_commit_trans(struct ocfs2_super *osb,
251 int ocfs2_extend_trans(handle_t *handle, int nblocks);
254 * Create access is for when we get a newly created buffer and we're
255 * not gonna read it off disk, but rather fill it ourselves. Right
256 * now, we don't do anything special with this (it turns into a write
257 * request), but this is a good placeholder in case we do...
259 * Write access is for when we read a block off disk and are going to
260 * modify it. This way the journalling layer knows it may need to make
261 * a copy of that block (if it's part of another, uncommitted
262 * transaction) before we do so.
264 #define OCFS2_JOURNAL_ACCESS_CREATE 0
265 #define OCFS2_JOURNAL_ACCESS_WRITE 1
266 #define OCFS2_JOURNAL_ACCESS_UNDO 2
270 int ocfs2_journal_access_di(handle_t *handle, struct inode *inode,
271 struct buffer_head *bh, int type);
272 /* ocfs2_extent_block */
273 int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode,
274 struct buffer_head *bh, int type);
275 /* ocfs2_group_desc */
276 int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode,
277 struct buffer_head *bh, int type);
278 /* ocfs2_xattr_block */
279 int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode,
280 struct buffer_head *bh, int type);
282 int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode,
283 struct buffer_head *bh, int type);
285 int ocfs2_journal_access_db(handle_t *handle, struct inode *inode,
286 struct buffer_head *bh, int type);
287 /* ocfs2_dx_root_block */
288 int ocfs2_journal_access_dr(handle_t *handle, struct inode *inode,
289 struct buffer_head *bh, int type);
291 int ocfs2_journal_access_dl(handle_t *handle, struct inode *inode,
292 struct buffer_head *bh, int type);
293 /* Anything that has no ecc */
294 int ocfs2_journal_access(handle_t *handle, struct inode *inode,
295 struct buffer_head *bh, int type);
298 * A word about the journal_access/journal_dirty "dance". It is
299 * entirely legal to journal_access a buffer more than once (as long
300 * as the access type is the same -- I'm not sure what will happen if
301 * access type is different but this should never happen anyway) It is
302 * also legal to journal_dirty a buffer more than once. In fact, you
303 * can even journal_access a buffer after you've done a
304 * journal_access/journal_dirty pair. The only thing you cannot do
305 * however, is journal_dirty a buffer which you haven't yet passed to
306 * journal_access at least once.
308 * That said, 99% of the time this doesn't matter and this is what the
312 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
314 * ocfs2_journal_dirty(handle, bh);
316 int ocfs2_journal_dirty(handle_t *handle,
317 struct buffer_head *bh);
321 * Convenience macros to calculate number of credits needed.
323 * For convenience sake, I have a set of macros here which calculate
324 * the *maximum* number of sectors which will be changed for various
328 /* simple file updates like chmod, etc. */
329 #define OCFS2_INODE_UPDATE_CREDITS 1
331 /* extended attribute block update */
332 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
334 /* Update of a single quota block */
335 #define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
337 /* global quotafile inode update, data block */
338 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
339 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
341 #define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
343 * The two writes below can accidentally see global info dirty due
344 * to set_info() quotactl so make them prepared for the writes.
346 /* quota data block, global info */
347 /* Write to local quota file */
348 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
349 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
351 /* global quota data block, local quota data block, global quota inode,
352 * global quota info */
353 #define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
354 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
356 static inline int ocfs2_quota_trans_credits(struct super_block *sb)
360 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
361 credits += OCFS2_QWRITE_CREDITS;
362 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
363 credits += OCFS2_QWRITE_CREDITS;
367 /* group extend. inode update and last group update. */
368 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
370 /* group add. inode update and the new group update. */
371 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
373 /* get one bit out of a suballocator: dinode + group descriptor +
374 * prev. group desc. if we relink. */
375 #define OCFS2_SUBALLOC_ALLOC (3)
377 static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
379 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
380 ocfs2_quota_trans_credits(sb);
383 /* dinode + group descriptor update. We don't relink on free yet. */
384 #define OCFS2_SUBALLOC_FREE (2)
386 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
387 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
388 + OCFS2_TRUNCATE_LOG_UPDATE)
390 static inline int ocfs2_remove_extent_credits(struct super_block *sb)
392 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
393 ocfs2_quota_trans_credits(sb);
396 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
397 * bitmap block for the new bit) dx_root update for free list */
398 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2 + 1)
400 static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
402 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
403 * worth of blocks for initial extent. */
404 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
405 ocfs2_clusters_to_blocks(sb, 1);
408 /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
409 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
410 * blocks + quota update */
411 static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
414 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
417 dir_credits += ocfs2_add_dir_index_credits(sb);
419 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
420 ocfs2_quota_trans_credits(sb);
423 /* local alloc metadata change + main bitmap updates */
424 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
425 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
427 /* used when we don't need an allocation change for a dir extend. One
428 * for the dinode, one for the new block. */
429 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
431 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
432 * update on dir + index leaf + dx root update for free list */
433 static inline int ocfs2_link_credits(struct super_block *sb)
435 return 2*OCFS2_INODE_UPDATE_CREDITS + 3 +
436 ocfs2_quota_trans_credits(sb);
439 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
440 * dir inode link + dir inode index leaf + dir index root */
441 static inline int ocfs2_unlink_credits(struct super_block *sb)
443 /* The quota update from ocfs2_link_credits is unused here... */
444 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
447 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
448 * inode alloc group descriptor + orphan dir index root +
449 * orphan dir index leaf */
450 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
452 /* dinode update, old dir dinode update, new dir dinode update, old
453 * dir dir entry, new dir dir entry, dir entry update for renaming
454 * directory + target unlink + 3 x dir index leaves */
455 static inline int ocfs2_rename_credits(struct super_block *sb)
457 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
460 /* global bitmap dinode, group desc., relinked group,
461 * suballocator dinode, group desc., relinked group,
462 * dinode, xattr block */
463 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
464 + OCFS2_INODE_UPDATE_CREDITS \
465 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
467 /* inode update, removal of dx root block from allocator */
468 #define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
471 static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
473 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
475 credits += ocfs2_clusters_to_blocks(sb, 1);
476 credits += ocfs2_quota_trans_credits(sb);
482 * Please note that the caller must make sure that root_el is the root
483 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
484 * the result may be wrong.
486 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
487 struct ocfs2_extent_list *root_el,
490 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
492 /* bitmap dinode, group desc. + relinked group. */
493 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
495 /* we might need to shift tree depth so lets assume an
496 * absolute worst case of complete fragmentation. Even with
497 * that, we only need one update for the dinode, and then
498 * however many metadata chunks needed * a remaining suballoc
500 sysfile_bitmap_blocks = 1 +
501 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
503 /* this does not include *new* metadata blocks, which are
504 * accounted for in sysfile_bitmap_blocks. root_el +
505 * prev. last_eb_blk + blocks along edge of tree.
506 * calc_symlink_credits passes because we just need 1
507 * credit for the dinode there. */
508 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
510 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
511 ocfs2_quota_trans_credits(sb);
514 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
516 int blocks = ocfs2_mknod_credits(sb, 0, 0);
518 /* links can be longer than one block so we may update many
519 * within our single allocated extent. */
520 blocks += ocfs2_clusters_to_blocks(sb, 1);
522 return blocks + ocfs2_quota_trans_credits(sb);
525 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
529 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
530 /* parent inode update + new block group header + bitmap inode update
531 + bitmap blocks affected */
532 blocks = 1 + 1 + 1 + bitmap_blocks;
536 static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
537 unsigned int clusters_to_del,
538 struct ocfs2_dinode *fe,
539 struct ocfs2_extent_list *last_el)
541 /* for dinode + all headers in this pass + update to next leaf */
542 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
543 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
544 int credits = 1 + tree_depth + 1;
550 /* We may be deleting metadata blocks, so metadata alloc dinode +
551 one desc. block for each possible delete. */
552 if (tree_depth && next_free == 1 &&
553 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
554 credits += 1 + tree_depth;
556 /* update to the truncate log. */
557 credits += OCFS2_TRUNCATE_LOG_UPDATE;
559 credits += ocfs2_quota_trans_credits(sb);
564 static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
566 return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
569 static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
572 return jbd2_journal_begin_ordered_truncate(
573 OCFS2_SB(inode->i_sb)->journal->j_journal,
574 &OCFS2_I(inode)->ip_jinode,
578 #endif /* OCFS2_JOURNAL_H */