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 struct ocfs2_journal_handle {
137 handle_t *k_handle; /* kernel handle. */
138 struct ocfs2_journal *journal;
141 /* Exported only for the journal struct init code in super.c. Do not call. */
142 void ocfs2_complete_recovery(void *data);
146 * Initialize, Load, Shutdown, Wipe a journal.
148 * ocfs2_journal_init - Initialize journal structures in the OSB.
149 * ocfs2_journal_load - Load the given journal off disk. Replay it if
150 * there's transactions still in there.
151 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
152 * uncommitted, uncheckpointed transactions.
153 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
154 * zero out each block.
155 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
156 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
158 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
160 void ocfs2_set_journal_params(struct ocfs2_super *osb);
161 int ocfs2_journal_init(struct ocfs2_journal *journal,
163 void ocfs2_journal_shutdown(struct ocfs2_super *osb);
164 int ocfs2_journal_wipe(struct ocfs2_journal *journal,
166 int ocfs2_journal_load(struct ocfs2_journal *journal);
167 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
168 void ocfs2_recovery_thread(struct ocfs2_super *osb,
170 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
171 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
173 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
175 atomic_set(&osb->needs_checkpoint, 1);
176 wake_up(&osb->checkpoint_event);
179 static inline void ocfs2_checkpoint_inode(struct inode *inode)
181 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
183 if (!ocfs2_inode_fully_checkpointed(inode)) {
184 /* WARNING: This only kicks off a single
185 * checkpoint. If someone races you and adds more
186 * metadata to the journal, you won't know, and will
187 * wind up waiting *alot* longer than necessary. Right
188 * now we only use this in clear_inode so that's
190 ocfs2_start_checkpoint(osb);
192 wait_event(osb->journal->j_checkpointed,
193 ocfs2_inode_fully_checkpointed(inode));
198 * Transaction Handling:
199 * Manage the lifetime of a transaction handle.
201 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
202 * the number of blocks that will be changed during
204 * ocfs2_commit_trans - Complete a handle.
205 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
206 * commit the handle to disk in the process, but will
207 * not release any locks taken during the transaction.
208 * ocfs2_journal_access - Notify the handle that we want to journal this
209 * buffer. Will have to call ocfs2_journal_dirty once
210 * we've actually dirtied it. Type is one of . or .
211 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
212 * ocfs2_journal_dirty_data - Indicate that a data buffer should go out before
213 * the current handle commits.
216 /* You must always start_trans with a number of buffs > 0, but it's
217 * perfectly legal to go through an entire transaction without having
218 * dirtied any buffers. */
219 struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb,
220 struct ocfs2_journal_handle *handle,
222 void ocfs2_commit_trans(struct ocfs2_journal_handle *handle);
223 int ocfs2_extend_trans(handle_t *handle, int nblocks);
226 * Create access is for when we get a newly created buffer and we're
227 * not gonna read it off disk, but rather fill it ourselves. Right
228 * now, we don't do anything special with this (it turns into a write
229 * request), but this is a good placeholder in case we do...
231 * Write access is for when we read a block off disk and are going to
232 * modify it. This way the journalling layer knows it may need to make
233 * a copy of that block (if it's part of another, uncommitted
234 * transaction) before we do so.
236 #define OCFS2_JOURNAL_ACCESS_CREATE 0
237 #define OCFS2_JOURNAL_ACCESS_WRITE 1
238 #define OCFS2_JOURNAL_ACCESS_UNDO 2
240 int ocfs2_journal_access(struct ocfs2_journal_handle *handle,
242 struct buffer_head *bh,
245 * A word about the journal_access/journal_dirty "dance". It is
246 * entirely legal to journal_access a buffer more than once (as long
247 * as the access type is the same -- I'm not sure what will happen if
248 * access type is different but this should never happen anyway) It is
249 * also legal to journal_dirty a buffer more than once. In fact, you
250 * can even journal_access a buffer after you've done a
251 * journal_access/journal_dirty pair. The only thing you cannot do
252 * however, is journal_dirty a buffer which you haven't yet passed to
253 * journal_access at least once.
255 * That said, 99% of the time this doesn't matter and this is what the
259 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
261 * ocfs2_journal_dirty(handle, bh);
263 int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle,
264 struct buffer_head *bh);
265 int ocfs2_journal_dirty_data(handle_t *handle,
266 struct buffer_head *bh);
270 * Convenience macros to calculate number of credits needed.
272 * For convenience sake, I have a set of macros here which calculate
273 * the *maximum* number of sectors which will be changed for various
277 /* simple file updates like chmod, etc. */
278 #define OCFS2_INODE_UPDATE_CREDITS 1
280 /* get one bit out of a suballocator: dinode + group descriptor +
281 * prev. group desc. if we relink. */
282 #define OCFS2_SUBALLOC_ALLOC (3)
284 /* dinode + group descriptor update. We don't relink on free yet. */
285 #define OCFS2_SUBALLOC_FREE (2)
287 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
288 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
289 + OCFS2_TRUNCATE_LOG_UPDATE)
291 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
292 * bitmap block for the new bit) */
293 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
295 /* parent fe, parent block, new file entry, inode alloc fe, inode alloc
296 * group descriptor + mkdir/symlink blocks */
297 #define OCFS2_MKNOD_CREDITS (3 + OCFS2_SUBALLOC_ALLOC \
298 + OCFS2_DIR_LINK_ADDITIONAL_CREDITS)
300 /* local alloc metadata change + main bitmap updates */
301 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
302 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
304 /* used when we don't need an allocation change for a dir extend. One
305 * for the dinode, one for the new block. */
306 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
308 /* file update (nlink, etc) + dir entry block */
309 #define OCFS2_LINK_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
311 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
313 #define OCFS2_UNLINK_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 1 \
314 + OCFS2_LINK_CREDITS)
316 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
317 * inode alloc group descriptor */
318 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
320 /* dinode update, old dir dinode update, new dir dinode update, old
321 * dir dir entry, new dir dir entry, dir entry update for renaming
322 * directory + target unlink */
323 #define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3 \
324 + OCFS2_UNLINK_CREDITS)
326 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
327 struct ocfs2_dinode *fe,
330 int bitmap_blocks, sysfile_bitmap_blocks, dinode_blocks;
332 /* bitmap dinode, group desc. + relinked group. */
333 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
335 /* we might need to shift tree depth so lets assume an
336 * absolute worst case of complete fragmentation. Even with
337 * that, we only need one update for the dinode, and then
338 * however many metadata chunks needed * a remaining suballoc
340 sysfile_bitmap_blocks = 1 +
341 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(fe);
343 /* this does not include *new* metadata blocks, which are
344 * accounted for in sysfile_bitmap_blocks. fe +
345 * prev. last_eb_blk + blocks along edge of tree.
346 * calc_symlink_credits passes because we just need 1
347 * credit for the dinode there. */
348 dinode_blocks = 1 + 1 + le16_to_cpu(fe->id2.i_list.l_tree_depth);
350 return bitmap_blocks + sysfile_bitmap_blocks + dinode_blocks;
353 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
355 int blocks = OCFS2_MKNOD_CREDITS;
357 /* links can be longer than one block so we may update many
358 * within our single allocated extent. */
359 blocks += ocfs2_clusters_to_blocks(sb, 1);
364 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
368 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
369 /* parent inode update + new block group header + bitmap inode update
370 + bitmap blocks affected */
371 blocks = 1 + 1 + 1 + bitmap_blocks;
375 static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
376 unsigned int clusters_to_del,
377 struct ocfs2_dinode *fe,
378 struct ocfs2_extent_list *last_el)
380 /* for dinode + all headers in this pass + update to next leaf */
381 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
382 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
383 int credits = 1 + tree_depth + 1;
389 /* We may be deleting metadata blocks, so metadata alloc dinode +
390 one desc. block for each possible delete. */
391 if (tree_depth && next_free == 1 &&
392 le32_to_cpu(last_el->l_recs[i].e_clusters) == clusters_to_del)
393 credits += 1 + tree_depth;
395 /* update to the truncate log. */
396 credits += OCFS2_TRUNCATE_LOG_UPDATE;
401 #endif /* OCFS2_JOURNAL_H */