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/jbd2.h>
31
32 enum ocfs2_journal_state {
33 OCFS2_JOURNAL_FREE = 0,
34 OCFS2_JOURNAL_LOADED,
35 OCFS2_JOURNAL_IN_SHUTDOWN,
36 };
37
38 struct ocfs2_super;
39 struct ocfs2_dinode;
40
41 struct ocfs2_journal {
42 enum ocfs2_journal_state j_state; /* Journals current state */
43
44 journal_t *j_journal; /* The kernels journal type */
45 struct inode *j_inode; /* Kernel inode pointing to
46 * this journal */
47 struct ocfs2_super *j_osb; /* pointer to the super
48 * block for the node
49 * we're currently
50 * running on -- not
51 * necessarily the super
52 * block from the node
53 * which we usually run
54 * from (recovery,
55 * etc) */
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;
62
63 spinlock_t j_lock;
64 struct list_head j_la_cleanups;
65 struct work_struct j_recovery_work;
66 };
67
68 extern spinlock_t trans_inc_lock;
69
70 /* wrap j_trans_id so we never have it equal to zero. */
ocfs2_inc_trans_id(struct ocfs2_journal * j)71 static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
72 {
73 unsigned long old_id;
74 spin_lock(&trans_inc_lock);
75 old_id = j->j_trans_id++;
76 if (unlikely(!j->j_trans_id))
77 j->j_trans_id = 1;
78 spin_unlock(&trans_inc_lock);
79 return old_id;
80 }
81
ocfs2_set_inode_lock_trans(struct ocfs2_journal * journal,struct inode * inode)82 static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal *journal,
83 struct inode *inode)
84 {
85 spin_lock(&trans_inc_lock);
86 OCFS2_I(inode)->ip_last_trans = journal->j_trans_id;
87 spin_unlock(&trans_inc_lock);
88 }
89
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. */
ocfs2_inode_fully_checkpointed(struct inode * inode)95 static inline int ocfs2_inode_fully_checkpointed(struct inode *inode)
96 {
97 int ret;
98 struct ocfs2_journal *journal = OCFS2_SB(inode->i_sb)->journal;
99
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);
103 return ret;
104 }
105
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. */
ocfs2_inode_is_new(struct inode * inode)109 static inline int ocfs2_inode_is_new(struct inode *inode)
110 {
111 int ret;
112
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)
117 return 0;
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));
121 if (!ret)
122 OCFS2_I(inode)->ip_created_trans = 0;
123 spin_unlock(&trans_inc_lock);
124 return ret;
125 }
126
ocfs2_inode_set_new(struct ocfs2_super * osb,struct inode * inode)127 static inline void ocfs2_inode_set_new(struct ocfs2_super *osb,
128 struct inode *inode)
129 {
130 spin_lock(&trans_inc_lock);
131 OCFS2_I(inode)->ip_created_trans = osb->journal->j_trans_id;
132 spin_unlock(&trans_inc_lock);
133 }
134
135 /* Exported only for the journal struct init code in super.c. Do not call. */
136 void ocfs2_complete_recovery(struct work_struct *work);
137 void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
138
139 int ocfs2_recovery_init(struct ocfs2_super *osb);
140 void ocfs2_recovery_exit(struct ocfs2_super *osb);
141
142 /*
143 * Journal Control:
144 * Initialize, Load, Shutdown, Wipe a journal.
145 *
146 * ocfs2_journal_init - Initialize journal structures in the OSB.
147 * ocfs2_journal_load - Load the given journal off disk. Replay it if
148 * there's transactions still in there.
149 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
150 * uncommitted, uncheckpointed transactions.
151 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
152 * zero out each block.
153 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
154 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
155 * event on.
156 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
157 */
158 void ocfs2_set_journal_params(struct ocfs2_super *osb);
159 int ocfs2_journal_init(struct ocfs2_journal *journal,
160 int *dirty);
161 void ocfs2_journal_shutdown(struct ocfs2_super *osb);
162 int ocfs2_journal_wipe(struct ocfs2_journal *journal,
163 int full);
164 int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
165 int replayed);
166 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
167 void ocfs2_recovery_thread(struct ocfs2_super *osb,
168 int node_num);
169 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
170 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
171 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
172
ocfs2_start_checkpoint(struct ocfs2_super * osb)173 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
174 {
175 atomic_set(&osb->needs_checkpoint, 1);
176 wake_up(&osb->checkpoint_event);
177 }
178
ocfs2_checkpoint_inode(struct inode * inode)179 static inline void ocfs2_checkpoint_inode(struct inode *inode)
180 {
181 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
182
183 if (ocfs2_mount_local(osb))
184 return;
185
186 if (!ocfs2_inode_fully_checkpointed(inode)) {
187 /* WARNING: This only kicks off a single
188 * checkpoint. If someone races you and adds more
189 * metadata to the journal, you won't know, and will
190 * wind up waiting *alot* longer than necessary. Right
191 * now we only use this in clear_inode so that's
192 * OK. */
193 ocfs2_start_checkpoint(osb);
194
195 wait_event(osb->journal->j_checkpointed,
196 ocfs2_inode_fully_checkpointed(inode));
197 }
198 }
199
200 /*
201 * Transaction Handling:
202 * Manage the lifetime of a transaction handle.
203 *
204 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
205 * the number of blocks that will be changed during
206 * this handle.
207 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
208 * the journal was aborted. The majority of paths don't
209 * check the return value as an error there comes too
210 * late to do anything (and will be picked up in a
211 * later transaction).
212 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
213 * commit the handle to disk in the process, but will
214 * not release any locks taken during the transaction.
215 * ocfs2_journal_access* - Notify the handle that we want to journal this
216 * buffer. Will have to call ocfs2_journal_dirty once
217 * we've actually dirtied it. Type is one of . or .
218 * Always call the specific flavor of
219 * ocfs2_journal_access_*() unless you intend to
220 * manage the checksum by hand.
221 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
222 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
223 * the current handle commits.
224 */
225
226 /* You must always start_trans with a number of buffs > 0, but it's
227 * perfectly legal to go through an entire transaction without having
228 * dirtied any buffers. */
229 handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
230 int max_buffs);
231 int ocfs2_commit_trans(struct ocfs2_super *osb,
232 handle_t *handle);
233 int ocfs2_extend_trans(handle_t *handle, int nblocks);
234
235 /*
236 * Create access is for when we get a newly created buffer and we're
237 * not gonna read it off disk, but rather fill it ourselves. Right
238 * now, we don't do anything special with this (it turns into a write
239 * request), but this is a good placeholder in case we do...
240 *
241 * Write access is for when we read a block off disk and are going to
242 * modify it. This way the journalling layer knows it may need to make
243 * a copy of that block (if it's part of another, uncommitted
244 * transaction) before we do so.
245 */
246 #define OCFS2_JOURNAL_ACCESS_CREATE 0
247 #define OCFS2_JOURNAL_ACCESS_WRITE 1
248 #define OCFS2_JOURNAL_ACCESS_UNDO 2
249
250
251 /* ocfs2_inode */
252 int ocfs2_journal_access_di(handle_t *handle, struct inode *inode,
253 struct buffer_head *bh, int type);
254 /* ocfs2_extent_block */
255 int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode,
256 struct buffer_head *bh, int type);
257 /* ocfs2_group_desc */
258 int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode,
259 struct buffer_head *bh, int type);
260 /* ocfs2_xattr_block */
261 int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode,
262 struct buffer_head *bh, int type);
263 /* quota blocks */
264 int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode,
265 struct buffer_head *bh, int type);
266 /* dirblock */
267 int ocfs2_journal_access_db(handle_t *handle, struct inode *inode,
268 struct buffer_head *bh, int type);
269 /* Anything that has no ecc */
270 int ocfs2_journal_access(handle_t *handle, struct inode *inode,
271 struct buffer_head *bh, int type);
272
273 /*
274 * A word about the journal_access/journal_dirty "dance". It is
275 * entirely legal to journal_access a buffer more than once (as long
276 * as the access type is the same -- I'm not sure what will happen if
277 * access type is different but this should never happen anyway) It is
278 * also legal to journal_dirty a buffer more than once. In fact, you
279 * can even journal_access a buffer after you've done a
280 * journal_access/journal_dirty pair. The only thing you cannot do
281 * however, is journal_dirty a buffer which you haven't yet passed to
282 * journal_access at least once.
283 *
284 * That said, 99% of the time this doesn't matter and this is what the
285 * path looks like:
286 *
287 * <read a bh>
288 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
289 * <modify the bh>
290 * ocfs2_journal_dirty(handle, bh);
291 */
292 int ocfs2_journal_dirty(handle_t *handle,
293 struct buffer_head *bh);
294
295 /*
296 * Credit Macros:
297 * Convenience macros to calculate number of credits needed.
298 *
299 * For convenience sake, I have a set of macros here which calculate
300 * the *maximum* number of sectors which will be changed for various
301 * metadata updates.
302 */
303
304 /* simple file updates like chmod, etc. */
305 #define OCFS2_INODE_UPDATE_CREDITS 1
306
307 /* extended attribute block update */
308 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
309
310 /* global quotafile inode update, data block */
311 #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
312
313 /*
314 * The two writes below can accidentally see global info dirty due
315 * to set_info() quotactl so make them prepared for the writes.
316 */
317 /* quota data block, global info */
318 /* Write to local quota file */
319 #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + 1)
320
321 /* global quota data block, local quota data block, global quota inode,
322 * global quota info */
323 #define OCFS2_QSYNC_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 3)
324
ocfs2_quota_trans_credits(struct super_block * sb)325 static inline int ocfs2_quota_trans_credits(struct super_block *sb)
326 {
327 int credits = 0;
328
329 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
330 credits += OCFS2_QWRITE_CREDITS;
331 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
332 credits += OCFS2_QWRITE_CREDITS;
333 return credits;
334 }
335
336 /* Number of credits needed for removing quota structure from file */
337 int ocfs2_calc_qdel_credits(struct super_block *sb, int type);
338 /* Number of credits needed for initialization of new quota structure */
339 int ocfs2_calc_qinit_credits(struct super_block *sb, int type);
340
341 /* group extend. inode update and last group update. */
342 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
343
344 /* group add. inode update and the new group update. */
345 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
346
347 /* get one bit out of a suballocator: dinode + group descriptor +
348 * prev. group desc. if we relink. */
349 #define OCFS2_SUBALLOC_ALLOC (3)
350
ocfs2_inline_to_extents_credits(struct super_block * sb)351 static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
352 {
353 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
354 ocfs2_quota_trans_credits(sb);
355 }
356
357 /* dinode + group descriptor update. We don't relink on free yet. */
358 #define OCFS2_SUBALLOC_FREE (2)
359
360 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
361 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
362 + OCFS2_TRUNCATE_LOG_UPDATE)
363
ocfs2_remove_extent_credits(struct super_block * sb)364 static inline int ocfs2_remove_extent_credits(struct super_block *sb)
365 {
366 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
367 ocfs2_quota_trans_credits(sb);
368 }
369
370 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
371 * bitmap block for the new bit) */
372 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
373
374 /* parent fe, parent block, new file entry, inode alloc fe, inode alloc
375 * group descriptor + mkdir/symlink blocks + quota update */
ocfs2_mknod_credits(struct super_block * sb)376 static inline int ocfs2_mknod_credits(struct super_block *sb)
377 {
378 return 3 + OCFS2_SUBALLOC_ALLOC + OCFS2_DIR_LINK_ADDITIONAL_CREDITS +
379 ocfs2_quota_trans_credits(sb);
380 }
381
382 /* local alloc metadata change + main bitmap updates */
383 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
384 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
385
386 /* used when we don't need an allocation change for a dir extend. One
387 * for the dinode, one for the new block. */
388 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
389
390 /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
391 * update on dir */
ocfs2_link_credits(struct super_block * sb)392 static inline int ocfs2_link_credits(struct super_block *sb)
393 {
394 return 2*OCFS2_INODE_UPDATE_CREDITS + 1 +
395 ocfs2_quota_trans_credits(sb);
396 }
397
398 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
399 * dir inode link */
ocfs2_unlink_credits(struct super_block * sb)400 static inline int ocfs2_unlink_credits(struct super_block *sb)
401 {
402 /* The quota update from ocfs2_link_credits is unused here... */
403 return 2 * OCFS2_INODE_UPDATE_CREDITS + 1 + ocfs2_link_credits(sb);
404 }
405
406 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
407 * inode alloc group descriptor */
408 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
409
410 /* dinode update, old dir dinode update, new dir dinode update, old
411 * dir dir entry, new dir dir entry, dir entry update for renaming
412 * directory + target unlink */
ocfs2_rename_credits(struct super_block * sb)413 static inline int ocfs2_rename_credits(struct super_block *sb)
414 {
415 return 3 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_unlink_credits(sb);
416 }
417
418 /* global bitmap dinode, group desc., relinked group,
419 * suballocator dinode, group desc., relinked group,
420 * dinode, xattr block */
421 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
422 + OCFS2_INODE_UPDATE_CREDITS \
423 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
424
425 /*
426 * Please note that the caller must make sure that root_el is the root
427 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
428 * the result may be wrong.
429 */
ocfs2_calc_extend_credits(struct super_block * sb,struct ocfs2_extent_list * root_el,u32 bits_wanted)430 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
431 struct ocfs2_extent_list *root_el,
432 u32 bits_wanted)
433 {
434 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
435
436 /* bitmap dinode, group desc. + relinked group. */
437 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
438
439 /* we might need to shift tree depth so lets assume an
440 * absolute worst case of complete fragmentation. Even with
441 * that, we only need one update for the dinode, and then
442 * however many metadata chunks needed * a remaining suballoc
443 * alloc. */
444 sysfile_bitmap_blocks = 1 +
445 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
446
447 /* this does not include *new* metadata blocks, which are
448 * accounted for in sysfile_bitmap_blocks. root_el +
449 * prev. last_eb_blk + blocks along edge of tree.
450 * calc_symlink_credits passes because we just need 1
451 * credit for the dinode there. */
452 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
453
454 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
455 ocfs2_quota_trans_credits(sb);
456 }
457
ocfs2_calc_symlink_credits(struct super_block * sb)458 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
459 {
460 int blocks = ocfs2_mknod_credits(sb);
461
462 /* links can be longer than one block so we may update many
463 * within our single allocated extent. */
464 blocks += ocfs2_clusters_to_blocks(sb, 1);
465
466 return blocks + ocfs2_quota_trans_credits(sb);
467 }
468
ocfs2_calc_group_alloc_credits(struct super_block * sb,unsigned int cpg)469 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
470 unsigned int cpg)
471 {
472 int blocks;
473 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
474 /* parent inode update + new block group header + bitmap inode update
475 + bitmap blocks affected */
476 blocks = 1 + 1 + 1 + bitmap_blocks;
477 return blocks;
478 }
479
ocfs2_calc_tree_trunc_credits(struct super_block * sb,unsigned int clusters_to_del,struct ocfs2_dinode * fe,struct ocfs2_extent_list * last_el)480 static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
481 unsigned int clusters_to_del,
482 struct ocfs2_dinode *fe,
483 struct ocfs2_extent_list *last_el)
484 {
485 /* for dinode + all headers in this pass + update to next leaf */
486 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
487 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
488 int credits = 1 + tree_depth + 1;
489 int i;
490
491 i = next_free - 1;
492 BUG_ON(i < 0);
493
494 /* We may be deleting metadata blocks, so metadata alloc dinode +
495 one desc. block for each possible delete. */
496 if (tree_depth && next_free == 1 &&
497 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
498 credits += 1 + tree_depth;
499
500 /* update to the truncate log. */
501 credits += OCFS2_TRUNCATE_LOG_UPDATE;
502
503 credits += ocfs2_quota_trans_credits(sb);
504
505 return credits;
506 }
507
ocfs2_jbd2_file_inode(handle_t * handle,struct inode * inode)508 static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
509 {
510 return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
511 }
512
ocfs2_begin_ordered_truncate(struct inode * inode,loff_t new_size)513 static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
514 loff_t new_size)
515 {
516 return jbd2_journal_begin_ordered_truncate(
517 OCFS2_SB(inode->i_sb)->journal->j_journal,
518 &OCFS2_I(inode)->ip_jinode,
519 new_size);
520 }
521
522 #endif /* OCFS2_JOURNAL_H */
523