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);
153 int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
156 * Initialize, Load, Shutdown, Wipe a journal.
158 * ocfs2_journal_init - Initialize journal structures in the OSB.
159 * ocfs2_journal_load - Load the given journal off disk. Replay it if
160 * there's transactions still in there.
161 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
162 * uncommitted, uncheckpointed transactions.
163 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
164 * zero out each block.
165 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
166 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
168 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
170 void ocfs2_set_journal_params(struct ocfs2_super *osb);
171 int ocfs2_journal_init(struct ocfs2_journal *journal,
173 void ocfs2_journal_shutdown(struct ocfs2_super *osb);
174 int ocfs2_journal_wipe(struct ocfs2_journal *journal,
176 int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
178 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
179 void ocfs2_recovery_thread(struct ocfs2_super *osb,
181 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
182 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
183 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
185 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
187 atomic_set(&osb->needs_checkpoint, 1);
188 wake_up(&osb->checkpoint_event);
191 static inline void ocfs2_checkpoint_inode(struct inode *inode)
193 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
195 if (ocfs2_mount_local(osb))
198 if (!ocfs2_inode_fully_checkpointed(inode)) {
199 /* WARNING: This only kicks off a single
200 * checkpoint. If someone races you and adds more
201 * metadata to the journal, you won't know, and will
202 * wind up waiting *alot* longer than necessary. Right
203 * now we only use this in clear_inode so that's
205 ocfs2_start_checkpoint(osb);
207 wait_event(osb->journal->j_checkpointed,
208 ocfs2_inode_fully_checkpointed(inode));
213 * Transaction Handling:
214 * Manage the lifetime of a transaction handle.
216 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
217 * the number of blocks that will be changed during
219 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
220 * the journal was aborted. The majority of paths don't
221 * check the return value as an error there comes too
222 * late to do anything (and will be picked up in a
223 * later transaction).
224 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
225 * commit the handle to disk in the process, but will
226 * not release any locks taken during the transaction.
227 * ocfs2_journal_access* - Notify the handle that we want to journal this
228 * buffer. Will have to call ocfs2_journal_dirty once
229 * we've actually dirtied it. Type is one of . or .
230 * Always call the specific flavor of
231 * ocfs2_journal_access_*() unless you intend to
232 * manage the checksum by hand.
233 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
234 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
235 * the current handle commits.
238 /* You must always start_trans with a number of buffs > 0, but it's
239 * perfectly legal to go through an entire transaction without having
240 * dirtied any buffers. */
241 handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
243 int ocfs2_commit_trans(struct ocfs2_super *osb,
245 int ocfs2_extend_trans(handle_t *handle, int nblocks);
248 * Create access is for when we get a newly created buffer and we're
249 * not gonna read it off disk, but rather fill it ourselves. Right
250 * now, we don't do anything special with this (it turns into a write
251 * request), but this is a good placeholder in case we do...
253 * Write access is for when we read a block off disk and are going to
254 * modify it. This way the journalling layer knows it may need to make
255 * a copy of that block (if it's part of another, uncommitted
256 * transaction) before we do so.
258 #define OCFS2_JOURNAL_ACCESS_CREATE 0
259 #define OCFS2_JOURNAL_ACCESS_WRITE 1
260 #define OCFS2_JOURNAL_ACCESS_UNDO 2
264 int ocfs2_journal_access_di(handle_t *handle, struct inode *inode,
265 struct buffer_head *bh, int type);
266 /* ocfs2_extent_block */
267 int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode,
268 struct buffer_head *bh, int type);
269 /* ocfs2_group_desc */
270 int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode,
271 struct buffer_head *bh, int type);
272 /* ocfs2_xattr_block */
273 int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode,
274 struct buffer_head *bh, int type);
276 int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode,
277 struct buffer_head *bh, int type);
279 int ocfs2_journal_access_db(handle_t *handle, struct inode *inode,
280 struct buffer_head *bh, int type);
281 /* ocfs2_dx_root_block */
282 int ocfs2_journal_access_dr(handle_t *handle, struct inode *inode,
283 struct buffer_head *bh, int type);
285 int ocfs2_journal_access_dl(handle_t *handle, struct inode *inode,
286 struct buffer_head *bh, int type);
287 /* Anything that has no ecc */
288 int ocfs2_journal_access(handle_t *handle, struct inode *inode,
289 struct buffer_head *bh, int type);
292 * A word about the journal_access/journal_dirty "dance". It is
293 * entirely legal to journal_access a buffer more than once (as long
294 * as the access type is the same -- I'm not sure what will happen if
295 * access type is different but this should never happen anyway) It is
296 * also legal to journal_dirty a buffer more than once. In fact, you
297 * can even journal_access a buffer after you've done a
298 * journal_access/journal_dirty pair. The only thing you cannot do
299 * however, is journal_dirty a buffer which you haven't yet passed to
300 * journal_access at least once.
302 * That said, 99% of the time this doesn't matter and this is what the
306 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
308 * ocfs2_journal_dirty(handle, bh);
310 int ocfs2_journal_dirty(handle_t *handle,
311 struct buffer_head *bh);
315 * Convenience macros to calculate number of credits needed.
317 * For convenience sake, I have a set of macros here which calculate
318 * the *maximum* number of sectors which will be changed for various
322 /* simple file updates like chmod, etc. */
323 #define OCFS2_INODE_UPDATE_CREDITS 1
325 /* extended attribute block update */
326 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
328 /* global quotafile inode update, data block */
329 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
332 * The two writes below can accidentally see global info dirty due
333 * to set_info() quotactl so make them prepared for the writes.
335 /* quota data block, global info */
336 /* Write to local quota file */
337 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + 1)
339 /* global quota data block, local quota data block, global quota inode,
340 * global quota info */
341 #define OCFS2_QSYNC_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 3)
343 static inline int ocfs2_quota_trans_credits(struct super_block *sb)
347 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
348 credits += OCFS2_QWRITE_CREDITS;
349 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
350 credits += OCFS2_QWRITE_CREDITS;
354 /* Number of credits needed for removing quota structure from file */
355 int ocfs2_calc_qdel_credits(struct super_block *sb, int type);
356 /* Number of credits needed for initialization of new quota structure */
357 int ocfs2_calc_qinit_credits(struct super_block *sb, int type);
359 /* group extend. inode update and last group update. */
360 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
362 /* group add. inode update and the new group update. */
363 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
365 /* get one bit out of a suballocator: dinode + group descriptor +
366 * prev. group desc. if we relink. */
367 #define OCFS2_SUBALLOC_ALLOC (3)
369 static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
371 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
372 ocfs2_quota_trans_credits(sb);
375 /* dinode + group descriptor update. We don't relink on free yet. */
376 #define OCFS2_SUBALLOC_FREE (2)
378 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
379 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
380 + OCFS2_TRUNCATE_LOG_UPDATE)
382 static inline int ocfs2_remove_extent_credits(struct super_block *sb)
384 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
385 ocfs2_quota_trans_credits(sb);
388 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
389 * bitmap block for the new bit) dx_root update for free list */
390 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2 + 1)
392 static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
394 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
395 * worth of blocks for initial extent. */
396 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
397 ocfs2_clusters_to_blocks(sb, 1);
400 /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
401 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
402 * blocks + quota update */
403 static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
406 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
409 dir_credits += ocfs2_add_dir_index_credits(sb);
411 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
412 ocfs2_quota_trans_credits(sb);
415 /* local alloc metadata change + main bitmap updates */
416 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
417 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
419 /* used when we don't need an allocation change for a dir extend. One
420 * for the dinode, one for the new block. */
421 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
423 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
424 * update on dir + index leaf + dx root update for free list */
425 static inline int ocfs2_link_credits(struct super_block *sb)
427 return 2*OCFS2_INODE_UPDATE_CREDITS + 3 +
428 ocfs2_quota_trans_credits(sb);
431 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
432 * dir inode link + dir inode index leaf + dir index root */
433 static inline int ocfs2_unlink_credits(struct super_block *sb)
435 /* The quota update from ocfs2_link_credits is unused here... */
436 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
439 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
440 * inode alloc group descriptor + orphan dir index root +
441 * orphan dir index leaf */
442 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
444 /* dinode update, old dir dinode update, new dir dinode update, old
445 * dir dir entry, new dir dir entry, dir entry update for renaming
446 * directory + target unlink + 3 x dir index leaves */
447 static inline int ocfs2_rename_credits(struct super_block *sb)
449 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
452 /* global bitmap dinode, group desc., relinked group,
453 * suballocator dinode, group desc., relinked group,
454 * dinode, xattr block */
455 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
456 + OCFS2_INODE_UPDATE_CREDITS \
457 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
459 /* inode update, removal of dx root block from allocator */
460 #define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
463 static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
465 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
467 credits += ocfs2_clusters_to_blocks(sb, 1);
468 credits += ocfs2_quota_trans_credits(sb);
474 * Please note that the caller must make sure that root_el is the root
475 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
476 * the result may be wrong.
478 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
479 struct ocfs2_extent_list *root_el,
482 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
484 /* bitmap dinode, group desc. + relinked group. */
485 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
487 /* we might need to shift tree depth so lets assume an
488 * absolute worst case of complete fragmentation. Even with
489 * that, we only need one update for the dinode, and then
490 * however many metadata chunks needed * a remaining suballoc
492 sysfile_bitmap_blocks = 1 +
493 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
495 /* this does not include *new* metadata blocks, which are
496 * accounted for in sysfile_bitmap_blocks. root_el +
497 * prev. last_eb_blk + blocks along edge of tree.
498 * calc_symlink_credits passes because we just need 1
499 * credit for the dinode there. */
500 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
502 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
503 ocfs2_quota_trans_credits(sb);
506 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
508 int blocks = ocfs2_mknod_credits(sb, 0, 0);
510 /* links can be longer than one block so we may update many
511 * within our single allocated extent. */
512 blocks += ocfs2_clusters_to_blocks(sb, 1);
514 return blocks + ocfs2_quota_trans_credits(sb);
517 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
521 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
522 /* parent inode update + new block group header + bitmap inode update
523 + bitmap blocks affected */
524 blocks = 1 + 1 + 1 + bitmap_blocks;
528 static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
529 unsigned int clusters_to_del,
530 struct ocfs2_dinode *fe,
531 struct ocfs2_extent_list *last_el)
533 /* for dinode + all headers in this pass + update to next leaf */
534 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
535 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
536 int credits = 1 + tree_depth + 1;
542 /* We may be deleting metadata blocks, so metadata alloc dinode +
543 one desc. block for each possible delete. */
544 if (tree_depth && next_free == 1 &&
545 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
546 credits += 1 + tree_depth;
548 /* update to the truncate log. */
549 credits += OCFS2_TRUNCATE_LOG_UPDATE;
551 credits += ocfs2_quota_trans_credits(sb);
556 static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
558 return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
561 static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
564 return jbd2_journal_begin_ordered_truncate(
565 OCFS2_SB(inode->i_sb)->journal->j_journal,
566 &OCFS2_I(inode)->ip_jinode,
570 #endif /* OCFS2_JOURNAL_H */