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/jbd.h>
32 enum ocfs2_journal_state {
33 OCFS2_JOURNAL_FREE = 0,
35 OCFS2_JOURNAL_IN_SHUTDOWN,
40 struct ocfs2_journal_handle;
42 struct ocfs2_journal {
43 enum ocfs2_journal_state j_state; /* Journals current state */
45 journal_t *j_journal; /* The kernels journal type */
46 struct inode *j_inode; /* Kernel inode pointing to
48 struct ocfs2_super *j_osb; /* pointer to the super
52 * necessarily the super
54 * which we usually run
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;
65 struct list_head j_la_cleanups;
66 struct work_struct j_recovery_work;
69 extern spinlock_t trans_inc_lock;
71 /* wrap j_trans_id so we never have it equal to zero. */
72 static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
75 spin_lock(&trans_inc_lock);
76 old_id = j->j_trans_id++;
77 if (unlikely(!j->j_trans_id))
79 spin_unlock(&trans_inc_lock);
83 static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal *journal,
86 spin_lock(&trans_inc_lock);
87 OCFS2_I(inode)->ip_last_trans = journal->j_trans_id;
88 spin_unlock(&trans_inc_lock);
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. */
96 static inline int ocfs2_inode_fully_checkpointed(struct inode *inode)
99 struct ocfs2_journal *journal = OCFS2_SB(inode->i_sb)->journal;
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);
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. */
110 static inline int ocfs2_inode_is_new(struct inode *inode)
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)
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));
123 OCFS2_I(inode)->ip_created_trans = 0;
124 spin_unlock(&trans_inc_lock);
128 static inline void ocfs2_inode_set_new(struct ocfs2_super *osb,
131 spin_lock(&trans_inc_lock);
132 OCFS2_I(inode)->ip_created_trans = osb->journal->j_trans_id;
133 spin_unlock(&trans_inc_lock);
136 extern kmem_cache_t *ocfs2_lock_cache;
138 struct ocfs2_journal_lock {
139 struct inode *jl_inode;
140 struct list_head jl_lock_list;
143 struct ocfs2_journal_handle {
144 handle_t *k_handle; /* kernel handle. */
145 struct ocfs2_journal *journal;
146 u32 flags; /* see flags below. */
148 /* The following two fields are for ocfs2_handle_add_lock */
150 struct list_head locks; /* A bunch of locks to
151 * release on commit. This
152 * should be a list_head */
154 struct list_head inode_list;
157 #define OCFS2_HANDLE_STARTED 1
158 /* should we sync-commit this handle? */
159 #define OCFS2_HANDLE_SYNC 2
160 static inline int ocfs2_handle_started(struct ocfs2_journal_handle *handle)
162 return handle->flags & OCFS2_HANDLE_STARTED;
165 static inline void ocfs2_handle_set_sync(struct ocfs2_journal_handle *handle, int sync)
168 handle->flags |= OCFS2_HANDLE_SYNC;
170 handle->flags &= ~OCFS2_HANDLE_SYNC;
173 /* Exported only for the journal struct init code in super.c. Do not call. */
174 void ocfs2_complete_recovery(void *data);
178 * Initialize, Load, Shutdown, Wipe a journal.
180 * ocfs2_journal_init - Initialize journal structures in the OSB.
181 * ocfs2_journal_load - Load the given journal off disk. Replay it if
182 * there's transactions still in there.
183 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
184 * uncommitted, uncheckpointed transactions.
185 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
186 * zero out each block.
187 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
188 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
190 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
192 void ocfs2_set_journal_params(struct ocfs2_super *osb);
193 int ocfs2_journal_init(struct ocfs2_journal *journal,
195 void ocfs2_journal_shutdown(struct ocfs2_super *osb);
196 int ocfs2_journal_wipe(struct ocfs2_journal *journal,
198 int ocfs2_journal_load(struct ocfs2_journal *journal);
199 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
200 void ocfs2_recovery_thread(struct ocfs2_super *osb,
202 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
203 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
205 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
207 atomic_set(&osb->needs_checkpoint, 1);
208 wake_up(&osb->checkpoint_event);
211 static inline void ocfs2_checkpoint_inode(struct inode *inode)
213 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
215 if (!ocfs2_inode_fully_checkpointed(inode)) {
216 /* WARNING: This only kicks off a single
217 * checkpoint. If someone races you and adds more
218 * metadata to the journal, you won't know, and will
219 * wind up waiting *alot* longer than necessary. Right
220 * now we only use this in clear_inode so that's
222 ocfs2_start_checkpoint(osb);
224 wait_event(osb->journal->j_checkpointed,
225 ocfs2_inode_fully_checkpointed(inode));
230 * Transaction Handling:
231 * Manage the lifetime of a transaction handle.
233 * ocfs2_alloc_handle - Only allocate a handle so we can start putting
234 * cluster locks on it. To actually change blocks,
235 * call ocfs2_start_trans with the handle returned
236 * from this function. You may call ocfs2_commit_trans
237 * at any time in the lifetime of a handle.
238 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
239 * the number of blocks that will be changed during
241 * ocfs2_commit_trans - Complete a handle.
242 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
243 * commit the handle to disk in the process, but will
244 * not release any locks taken during the transaction.
245 * ocfs2_journal_access - Notify the handle that we want to journal this
246 * buffer. Will have to call ocfs2_journal_dirty once
247 * we've actually dirtied it. Type is one of . or .
248 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
249 * ocfs2_journal_dirty_data - Indicate that a data buffer should go out before
250 * the current handle commits.
251 * ocfs2_handle_add_lock - Sometimes we need to delay lock release
252 * until after a transaction has been completed. Use
253 * ocfs2_handle_add_lock to indicate that a lock needs
254 * to be released at the end of that handle. Locks
255 * will be released in the order that they are added.
256 * ocfs2_handle_add_inode - Add a locked inode to a transaction.
259 /* You must always start_trans with a number of buffs > 0, but it's
260 * perfectly legal to go through an entire transaction without having
261 * dirtied any buffers. */
262 struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb);
263 struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb,
264 struct ocfs2_journal_handle *handle,
266 void ocfs2_commit_trans(struct ocfs2_journal_handle *handle);
267 int ocfs2_extend_trans(handle_t *handle, int nblocks);
270 * Create access is for when we get a newly created buffer and we're
271 * not gonna read it off disk, but rather fill it ourselves. Right
272 * now, we don't do anything special with this (it turns into a write
273 * request), but this is a good placeholder in case we do...
275 * Write access is for when we read a block off disk and are going to
276 * modify it. This way the journalling layer knows it may need to make
277 * a copy of that block (if it's part of another, uncommitted
278 * transaction) before we do so.
280 #define OCFS2_JOURNAL_ACCESS_CREATE 0
281 #define OCFS2_JOURNAL_ACCESS_WRITE 1
282 #define OCFS2_JOURNAL_ACCESS_UNDO 2
284 int ocfs2_journal_access(struct ocfs2_journal_handle *handle,
286 struct buffer_head *bh,
289 * A word about the journal_access/journal_dirty "dance". It is
290 * entirely legal to journal_access a buffer more than once (as long
291 * as the access type is the same -- I'm not sure what will happen if
292 * access type is different but this should never happen anyway) It is
293 * also legal to journal_dirty a buffer more than once. In fact, you
294 * can even journal_access a buffer after you've done a
295 * journal_access/journal_dirty pair. The only thing you cannot do
296 * however, is journal_dirty a buffer which you haven't yet passed to
297 * journal_access at least once.
299 * That said, 99% of the time this doesn't matter and this is what the
303 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
305 * ocfs2_journal_dirty(handle, bh);
307 int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle,
308 struct buffer_head *bh);
309 int ocfs2_journal_dirty_data(handle_t *handle,
310 struct buffer_head *bh);
311 int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle,
312 struct inode *inode);
314 * Use this to protect from other processes reading buffer state while
317 void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle,
318 struct inode *inode);
322 * Convenience macros to calculate number of credits needed.
324 * For convenience sake, I have a set of macros here which calculate
325 * the *maximum* number of sectors which will be changed for various
329 /* simple file updates like chmod, etc. */
330 #define OCFS2_INODE_UPDATE_CREDITS 1
332 /* get one bit out of a suballocator: dinode + group descriptor +
333 * prev. group desc. if we relink. */
334 #define OCFS2_SUBALLOC_ALLOC (3)
336 /* dinode + group descriptor update. We don't relink on free yet. */
337 #define OCFS2_SUBALLOC_FREE (2)
339 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
340 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
341 + OCFS2_TRUNCATE_LOG_UPDATE)
343 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
344 * bitmap block for the new bit) */
345 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
347 /* parent fe, parent block, new file entry, inode alloc fe, inode alloc
348 * group descriptor + mkdir/symlink blocks */
349 #define OCFS2_MKNOD_CREDITS (3 + OCFS2_SUBALLOC_ALLOC \
350 + OCFS2_DIR_LINK_ADDITIONAL_CREDITS)
352 /* local alloc metadata change + main bitmap updates */
353 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
354 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
356 /* used when we don't need an allocation change for a dir extend. One
357 * for the dinode, one for the new block. */
358 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
360 /* file update (nlink, etc) + dir entry block */
361 #define OCFS2_LINK_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
363 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
365 #define OCFS2_UNLINK_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 1 \
366 + OCFS2_LINK_CREDITS)
368 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
369 * inode alloc group descriptor */
370 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
372 /* dinode update, old dir dinode update, new dir dinode update, old
373 * dir dir entry, new dir dir entry, dir entry update for renaming
374 * directory + target unlink */
375 #define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3 \
376 + OCFS2_UNLINK_CREDITS)
378 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
379 struct ocfs2_dinode *fe,
382 int bitmap_blocks, sysfile_bitmap_blocks, dinode_blocks;
384 /* bitmap dinode, group desc. + relinked group. */
385 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
387 /* we might need to shift tree depth so lets assume an
388 * absolute worst case of complete fragmentation. Even with
389 * that, we only need one update for the dinode, and then
390 * however many metadata chunks needed * a remaining suballoc
392 sysfile_bitmap_blocks = 1 +
393 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(fe);
395 /* this does not include *new* metadata blocks, which are
396 * accounted for in sysfile_bitmap_blocks. fe +
397 * prev. last_eb_blk + blocks along edge of tree.
398 * calc_symlink_credits passes because we just need 1
399 * credit for the dinode there. */
400 dinode_blocks = 1 + 1 + le16_to_cpu(fe->id2.i_list.l_tree_depth);
402 return bitmap_blocks + sysfile_bitmap_blocks + dinode_blocks;
405 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
407 int blocks = OCFS2_MKNOD_CREDITS;
409 /* links can be longer than one block so we may update many
410 * within our single allocated extent. */
411 blocks += ocfs2_clusters_to_blocks(sb, 1);
416 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
420 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
421 /* parent inode update + new block group header + bitmap inode update
422 + bitmap blocks affected */
423 blocks = 1 + 1 + 1 + bitmap_blocks;
427 static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
428 unsigned int clusters_to_del,
429 struct ocfs2_dinode *fe,
430 struct ocfs2_extent_list *last_el)
432 /* for dinode + all headers in this pass + update to next leaf */
433 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
434 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
435 int credits = 1 + tree_depth + 1;
441 /* We may be deleting metadata blocks, so metadata alloc dinode +
442 one desc. block for each possible delete. */
443 if (tree_depth && next_free == 1 &&
444 le32_to_cpu(last_el->l_recs[i].e_clusters) == clusters_to_del)
445 credits += 1 + tree_depth;
447 /* update to the truncate log. */
448 credits += OCFS2_TRUNCATE_LOG_UPDATE;
453 #endif /* OCFS2_JOURNAL_H */