blob: db82be2532ed5ef333c8a58a14e19cfbd3bbd6aa [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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/jbd.h>
31
Mark Fashehccd979b2005-12-15 14:31:24 -080032enum ocfs2_journal_state {
33 OCFS2_JOURNAL_FREE = 0,
34 OCFS2_JOURNAL_LOADED,
35 OCFS2_JOURNAL_IN_SHUTDOWN,
36};
37
38struct ocfs2_super;
39struct ocfs2_dinode;
Mark Fashehccd979b2005-12-15 14:31:24 -080040
41struct 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
68extern spinlock_t trans_inc_lock;
69
70/* wrap j_trans_id so we never have it equal to zero. */
71static 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
82static 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. */
95static 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. */
109static 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
127static 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
Mark Fashehccd979b2005-12-15 14:31:24 -0800135/* Exported only for the journal struct init code in super.c. Do not call. */
David Howellsc4028952006-11-22 14:57:56 +0000136void ocfs2_complete_recovery(struct work_struct *work);
Joel Becker553abd02008-02-01 12:03:57 -0800137void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
138
139int ocfs2_recovery_init(struct ocfs2_super *osb);
140void ocfs2_recovery_exit(struct ocfs2_super *osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800141
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 */
158void ocfs2_set_journal_params(struct ocfs2_super *osb);
159int ocfs2_journal_init(struct ocfs2_journal *journal,
160 int *dirty);
161void ocfs2_journal_shutdown(struct ocfs2_super *osb);
162int ocfs2_journal_wipe(struct ocfs2_journal *journal,
163 int full);
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800164int ocfs2_journal_load(struct ocfs2_journal *journal, int local);
Mark Fashehccd979b2005-12-15 14:31:24 -0800165int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
166void ocfs2_recovery_thread(struct ocfs2_super *osb,
167 int node_num);
168int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
169void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
170
171static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
172{
173 atomic_set(&osb->needs_checkpoint, 1);
174 wake_up(&osb->checkpoint_event);
175}
176
177static inline void ocfs2_checkpoint_inode(struct inode *inode)
178{
179 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
180
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800181 if (ocfs2_mount_local(osb))
182 return;
183
Mark Fashehccd979b2005-12-15 14:31:24 -0800184 if (!ocfs2_inode_fully_checkpointed(inode)) {
185 /* WARNING: This only kicks off a single
186 * checkpoint. If someone races you and adds more
187 * metadata to the journal, you won't know, and will
188 * wind up waiting *alot* longer than necessary. Right
189 * now we only use this in clear_inode so that's
190 * OK. */
191 ocfs2_start_checkpoint(osb);
192
193 wait_event(osb->journal->j_checkpointed,
194 ocfs2_inode_fully_checkpointed(inode));
195 }
196}
197
198/*
199 * Transaction Handling:
200 * Manage the lifetime of a transaction handle.
201 *
Mark Fashehccd979b2005-12-15 14:31:24 -0800202 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
203 * the number of blocks that will be changed during
204 * this handle.
Mark Fasheh1fabe142006-10-09 18:11:45 -0700205 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
206 * the journal was aborted. The majority of paths don't
207 * check the return value as an error there comes too
208 * late to do anything (and will be picked up in a
209 * later transaction).
Mark Fashehccd979b2005-12-15 14:31:24 -0800210 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
211 * commit the handle to disk in the process, but will
212 * not release any locks taken during the transaction.
213 * ocfs2_journal_access - Notify the handle that we want to journal this
214 * buffer. Will have to call ocfs2_journal_dirty once
215 * we've actually dirtied it. Type is one of . or .
216 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
217 * ocfs2_journal_dirty_data - Indicate that a data buffer should go out before
218 * the current handle commits.
Mark Fashehccd979b2005-12-15 14:31:24 -0800219 */
220
221/* You must always start_trans with a number of buffs > 0, but it's
222 * perfectly legal to go through an entire transaction without having
223 * dirtied any buffers. */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700224handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -0800225 int max_buffs);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700226int ocfs2_commit_trans(struct ocfs2_super *osb,
227 handle_t *handle);
Mark Fasheh1fc58142006-10-05 14:15:36 -0700228int ocfs2_extend_trans(handle_t *handle, int nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800229
230/*
231 * Create access is for when we get a newly created buffer and we're
232 * not gonna read it off disk, but rather fill it ourselves. Right
233 * now, we don't do anything special with this (it turns into a write
234 * request), but this is a good placeholder in case we do...
235 *
236 * Write access is for when we read a block off disk and are going to
237 * modify it. This way the journalling layer knows it may need to make
238 * a copy of that block (if it's part of another, uncommitted
239 * transaction) before we do so.
240 */
241#define OCFS2_JOURNAL_ACCESS_CREATE 0
242#define OCFS2_JOURNAL_ACCESS_WRITE 1
243#define OCFS2_JOURNAL_ACCESS_UNDO 2
244
Mark Fasheh1fabe142006-10-09 18:11:45 -0700245int ocfs2_journal_access(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800246 struct inode *inode,
247 struct buffer_head *bh,
248 int type);
249/*
250 * A word about the journal_access/journal_dirty "dance". It is
251 * entirely legal to journal_access a buffer more than once (as long
252 * as the access type is the same -- I'm not sure what will happen if
253 * access type is different but this should never happen anyway) It is
254 * also legal to journal_dirty a buffer more than once. In fact, you
255 * can even journal_access a buffer after you've done a
256 * journal_access/journal_dirty pair. The only thing you cannot do
257 * however, is journal_dirty a buffer which you haven't yet passed to
258 * journal_access at least once.
259 *
260 * That said, 99% of the time this doesn't matter and this is what the
261 * path looks like:
262 *
263 * <read a bh>
264 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
265 * <modify the bh>
266 * ocfs2_journal_dirty(handle, bh);
267 */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700268int ocfs2_journal_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800269 struct buffer_head *bh);
270int ocfs2_journal_dirty_data(handle_t *handle,
271 struct buffer_head *bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800272
273/*
274 * Credit Macros:
275 * Convenience macros to calculate number of credits needed.
276 *
277 * For convenience sake, I have a set of macros here which calculate
278 * the *maximum* number of sectors which will be changed for various
279 * metadata updates.
280 */
281
282/* simple file updates like chmod, etc. */
283#define OCFS2_INODE_UPDATE_CREDITS 1
284
Tao Mad6590722007-12-18 15:47:03 +0800285/* group extend. inode update and last group update. */
286#define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
287
Tao Ma7909f2b2007-12-18 15:47:25 +0800288/* group add. inode update and the new group update. */
289#define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
290
Mark Fashehccd979b2005-12-15 14:31:24 -0800291/* get one bit out of a suballocator: dinode + group descriptor +
292 * prev. group desc. if we relink. */
293#define OCFS2_SUBALLOC_ALLOC (3)
294
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700295#define OCFS2_INLINE_TO_EXTENTS_CREDITS (OCFS2_SUBALLOC_ALLOC \
296 + OCFS2_INODE_UPDATE_CREDITS)
297
Mark Fashehccd979b2005-12-15 14:31:24 -0800298/* dinode + group descriptor update. We don't relink on free yet. */
299#define OCFS2_SUBALLOC_FREE (2)
300
301#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
302#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
303 + OCFS2_TRUNCATE_LOG_UPDATE)
304
Mark Fasheh063c4562007-07-03 13:34:11 -0700305#define OCFS2_REMOVE_EXTENT_CREDITS (OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS)
306
Mark Fashehccd979b2005-12-15 14:31:24 -0800307/* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
308 * bitmap block for the new bit) */
309#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)
310
311/* parent fe, parent block, new file entry, inode alloc fe, inode alloc
312 * group descriptor + mkdir/symlink blocks */
313#define OCFS2_MKNOD_CREDITS (3 + OCFS2_SUBALLOC_ALLOC \
314 + OCFS2_DIR_LINK_ADDITIONAL_CREDITS)
315
316/* local alloc metadata change + main bitmap updates */
317#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
318 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
319
320/* used when we don't need an allocation change for a dir extend. One
321 * for the dinode, one for the new block. */
322#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
323
Mark Fashehe051fda2007-02-01 11:40:16 -0800324/* file update (nlink, etc) + directory mtime/ctime + dir entry block */
325#define OCFS2_LINK_CREDITS (2*OCFS2_INODE_UPDATE_CREDITS + 1)
Mark Fashehccd979b2005-12-15 14:31:24 -0800326
327/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
328 * dir inode link */
329#define OCFS2_UNLINK_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 1 \
330 + OCFS2_LINK_CREDITS)
331
332/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
333 * inode alloc group descriptor */
334#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1)
335
336/* dinode update, old dir dinode update, new dir dinode update, old
337 * dir dir entry, new dir dir entry, dir entry update for renaming
338 * directory + target unlink */
339#define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3 \
340 + OCFS2_UNLINK_CREDITS)
341
342static inline int ocfs2_calc_extend_credits(struct super_block *sb,
343 struct ocfs2_dinode *fe,
344 u32 bits_wanted)
345{
346 int bitmap_blocks, sysfile_bitmap_blocks, dinode_blocks;
347
348 /* bitmap dinode, group desc. + relinked group. */
349 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
350
351 /* we might need to shift tree depth so lets assume an
352 * absolute worst case of complete fragmentation. Even with
353 * that, we only need one update for the dinode, and then
354 * however many metadata chunks needed * a remaining suballoc
355 * alloc. */
356 sysfile_bitmap_blocks = 1 +
357 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(fe);
358
359 /* this does not include *new* metadata blocks, which are
360 * accounted for in sysfile_bitmap_blocks. fe +
361 * prev. last_eb_blk + blocks along edge of tree.
362 * calc_symlink_credits passes because we just need 1
363 * credit for the dinode there. */
364 dinode_blocks = 1 + 1 + le16_to_cpu(fe->id2.i_list.l_tree_depth);
365
366 return bitmap_blocks + sysfile_bitmap_blocks + dinode_blocks;
367}
368
369static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
370{
371 int blocks = OCFS2_MKNOD_CREDITS;
372
373 /* links can be longer than one block so we may update many
374 * within our single allocated extent. */
375 blocks += ocfs2_clusters_to_blocks(sb, 1);
376
377 return blocks;
378}
379
380static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
381 unsigned int cpg)
382{
383 int blocks;
384 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
385 /* parent inode update + new block group header + bitmap inode update
386 + bitmap blocks affected */
387 blocks = 1 + 1 + 1 + bitmap_blocks;
388 return blocks;
389}
390
391static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
392 unsigned int clusters_to_del,
393 struct ocfs2_dinode *fe,
394 struct ocfs2_extent_list *last_el)
395{
396 /* for dinode + all headers in this pass + update to next leaf */
397 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
398 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
399 int credits = 1 + tree_depth + 1;
400 int i;
401
402 i = next_free - 1;
403 BUG_ON(i < 0);
404
405 /* We may be deleting metadata blocks, so metadata alloc dinode +
406 one desc. block for each possible delete. */
407 if (tree_depth && next_free == 1 &&
Mark Fashehe48edee2007-03-07 16:46:57 -0800408 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
Mark Fashehccd979b2005-12-15 14:31:24 -0800409 credits += 1 + tree_depth;
410
411 /* update to the truncate log. */
412 credits += OCFS2_TRUNCATE_LOG_UPDATE;
413
414 return credits;
415}
416
417#endif /* OCFS2_JOURNAL_H */