blob: b5baaa8e710f234b08878c111d84ffe55f52df7d [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>
Mark Fasheh53ef99c2008-11-18 16:53:43 -080030#include <linux/jbd2.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080031
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
Sunil Mushran96a6c642008-12-16 15:49:16 -080041/*
42 * The recovery_list is a simple linked list of node numbers to recover.
43 * It is protected by the recovery_lock.
44 */
45
46struct ocfs2_recovery_map {
47 unsigned int rm_used;
48 unsigned int *rm_entries;
49};
50
51
Mark Fashehccd979b2005-12-15 14:31:24 -080052struct ocfs2_journal {
53 enum ocfs2_journal_state j_state; /* Journals current state */
54
55 journal_t *j_journal; /* The kernels journal type */
56 struct inode *j_inode; /* Kernel inode pointing to
57 * this journal */
58 struct ocfs2_super *j_osb; /* pointer to the super
59 * block for the node
60 * we're currently
61 * running on -- not
62 * necessarily the super
63 * block from the node
64 * which we usually run
65 * from (recovery,
66 * etc) */
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;
73
74 spinlock_t j_lock;
75 struct list_head j_la_cleanups;
76 struct work_struct j_recovery_work;
77};
78
79extern spinlock_t trans_inc_lock;
80
81/* wrap j_trans_id so we never have it equal to zero. */
82static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
83{
84 unsigned long old_id;
85 spin_lock(&trans_inc_lock);
86 old_id = j->j_trans_id++;
87 if (unlikely(!j->j_trans_id))
88 j->j_trans_id = 1;
89 spin_unlock(&trans_inc_lock);
90 return old_id;
91}
92
Joel Becker0cf2f762009-02-12 16:41:25 -080093static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
94 struct ocfs2_caching_info *ci)
Mark Fashehccd979b2005-12-15 14:31:24 -080095{
96 spin_lock(&trans_inc_lock);
Joel Becker0cf2f762009-02-12 16:41:25 -080097 ci->ci_last_trans = journal->j_trans_id;
Mark Fashehccd979b2005-12-15 14:31:24 -080098 spin_unlock(&trans_inc_lock);
99}
100
101/* Used to figure out whether it's safe to drop a metadata lock on an
Joel Becker0cf2f762009-02-12 16:41:25 -0800102 * cached object. Returns true if all the object's changes have been
Mark Fashehccd979b2005-12-15 14:31:24 -0800103 * 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. */
Joel Becker0cf2f762009-02-12 16:41:25 -0800106static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
Mark Fashehccd979b2005-12-15 14:31:24 -0800107{
108 int ret;
Joel Becker0cf2f762009-02-12 16:41:25 -0800109 struct ocfs2_journal *journal =
110 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -0800111
112 spin_lock(&trans_inc_lock);
Joel Becker0cf2f762009-02-12 16:41:25 -0800113 ret = time_after(journal->j_trans_id, ci->ci_last_trans);
Mark Fashehccd979b2005-12-15 14:31:24 -0800114 spin_unlock(&trans_inc_lock);
115 return ret;
116}
117
Joel Becker0cf2f762009-02-12 16:41:25 -0800118/* convenience function to check if an object backed by struct
119 * ocfs2_caching_info is still new (has never hit disk) Will do you a
120 * favor and set created_trans = 0 when you've
121 * been checkpointed. returns '1' if the ci is still new. */
122static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
Mark Fashehccd979b2005-12-15 14:31:24 -0800123{
124 int ret;
Joel Becker0cf2f762009-02-12 16:41:25 -0800125 struct ocfs2_journal *journal =
126 OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -0800127
Joel Becker0cf2f762009-02-12 16:41:25 -0800128 spin_lock(&trans_inc_lock);
129 ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
130 if (!ret)
131 ci->ci_created_trans = 0;
132 spin_unlock(&trans_inc_lock);
133 return ret;
134}
135
136/* Wrapper for inodes so we can check system files */
137static inline int ocfs2_inode_is_new(struct inode *inode)
138{
Mark Fashehccd979b2005-12-15 14:31:24 -0800139 /* System files are never "new" as they're written out by
140 * mkfs. This helps us early during mount, before we have the
141 * journal open and j_trans_id could be junk. */
142 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
143 return 0;
Joel Becker0cf2f762009-02-12 16:41:25 -0800144
145 return ocfs2_ci_is_new(INODE_CACHE(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800146}
147
Joel Becker0cf2f762009-02-12 16:41:25 -0800148static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
149 struct ocfs2_caching_info *ci)
Mark Fashehccd979b2005-12-15 14:31:24 -0800150{
151 spin_lock(&trans_inc_lock);
Joel Becker0cf2f762009-02-12 16:41:25 -0800152 ci->ci_created_trans = osb->journal->j_trans_id;
Mark Fashehccd979b2005-12-15 14:31:24 -0800153 spin_unlock(&trans_inc_lock);
154}
155
Mark Fashehccd979b2005-12-15 14:31:24 -0800156/* Exported only for the journal struct init code in super.c. Do not call. */
Sunil Mushrandf152c22009-06-22 11:40:07 -0700157void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
Jeff Mahoney8b712cd2009-07-07 17:22:12 -0400158void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
Srinivas Eeda83273932009-06-03 17:02:55 -0700159void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
160void ocfs2_orphan_scan_exit(struct ocfs2_super *osb);
161
David Howellsc4028952006-11-22 14:57:56 +0000162void ocfs2_complete_recovery(struct work_struct *work);
Joel Becker553abd02008-02-01 12:03:57 -0800163void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
164
165int ocfs2_recovery_init(struct ocfs2_super *osb);
166void ocfs2_recovery_exit(struct ocfs2_super *osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800167
Srinivas Eeda9140db02009-03-06 14:21:46 -0800168int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800169/*
170 * Journal Control:
171 * Initialize, Load, Shutdown, Wipe a journal.
172 *
173 * ocfs2_journal_init - Initialize journal structures in the OSB.
174 * ocfs2_journal_load - Load the given journal off disk. Replay it if
175 * there's transactions still in there.
176 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
177 * uncommitted, uncheckpointed transactions.
178 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
179 * zero out each block.
180 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
181 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
182 * event on.
183 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
184 */
185void ocfs2_set_journal_params(struct ocfs2_super *osb);
186int ocfs2_journal_init(struct ocfs2_journal *journal,
187 int *dirty);
188void ocfs2_journal_shutdown(struct ocfs2_super *osb);
189int ocfs2_journal_wipe(struct ocfs2_journal *journal,
190 int full);
Sunil Mushran539d8262008-07-14 17:31:10 -0700191int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
192 int replayed);
Mark Fashehccd979b2005-12-15 14:31:24 -0800193int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
194void ocfs2_recovery_thread(struct ocfs2_super *osb,
195 int node_num);
196int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
197void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
Jan Kara22053632008-10-20 23:50:38 +0200198void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800199
200static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
201{
202 atomic_set(&osb->needs_checkpoint, 1);
203 wake_up(&osb->checkpoint_event);
204}
205
206static inline void ocfs2_checkpoint_inode(struct inode *inode)
207{
208 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
209
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800210 if (ocfs2_mount_local(osb))
211 return;
212
Joel Becker0cf2f762009-02-12 16:41:25 -0800213 if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800214 /* WARNING: This only kicks off a single
215 * checkpoint. If someone races you and adds more
216 * metadata to the journal, you won't know, and will
217 * wind up waiting *alot* longer than necessary. Right
218 * now we only use this in clear_inode so that's
219 * OK. */
220 ocfs2_start_checkpoint(osb);
221
222 wait_event(osb->journal->j_checkpointed,
Joel Becker0cf2f762009-02-12 16:41:25 -0800223 ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
Mark Fashehccd979b2005-12-15 14:31:24 -0800224 }
225}
226
227/*
228 * Transaction Handling:
229 * Manage the lifetime of a transaction handle.
230 *
Mark Fashehccd979b2005-12-15 14:31:24 -0800231 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
232 * the number of blocks that will be changed during
233 * this handle.
Mark Fasheh1fabe142006-10-09 18:11:45 -0700234 * ocfs2_commit_trans - Complete a handle. It might return -EIO if
235 * the journal was aborted. The majority of paths don't
236 * check the return value as an error there comes too
237 * late to do anything (and will be picked up in a
238 * later transaction).
Mark Fashehccd979b2005-12-15 14:31:24 -0800239 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
240 * commit the handle to disk in the process, but will
241 * not release any locks taken during the transaction.
Joel Becker50655ae2008-09-11 15:53:07 -0700242 * ocfs2_journal_access* - Notify the handle that we want to journal this
Mark Fashehccd979b2005-12-15 14:31:24 -0800243 * buffer. Will have to call ocfs2_journal_dirty once
244 * we've actually dirtied it. Type is one of . or .
Joel Becker50655ae2008-09-11 15:53:07 -0700245 * Always call the specific flavor of
246 * ocfs2_journal_access_*() unless you intend to
247 * manage the checksum by hand.
Mark Fashehccd979b2005-12-15 14:31:24 -0800248 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
Joel Becker2b4e30f2008-09-03 20:03:41 -0700249 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
250 * the current handle commits.
Mark Fashehccd979b2005-12-15 14:31:24 -0800251 */
252
253/* You must always start_trans with a number of buffs > 0, but it's
254 * perfectly legal to go through an entire transaction without having
255 * dirtied any buffers. */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700256handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -0800257 int max_buffs);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700258int ocfs2_commit_trans(struct ocfs2_super *osb,
259 handle_t *handle);
Mark Fasheh1fc58142006-10-05 14:15:36 -0700260int ocfs2_extend_trans(handle_t *handle, int nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800261
262/*
263 * Create access is for when we get a newly created buffer and we're
264 * not gonna read it off disk, but rather fill it ourselves. Right
265 * now, we don't do anything special with this (it turns into a write
266 * request), but this is a good placeholder in case we do...
267 *
268 * Write access is for when we read a block off disk and are going to
269 * modify it. This way the journalling layer knows it may need to make
270 * a copy of that block (if it's part of another, uncommitted
271 * transaction) before we do so.
272 */
273#define OCFS2_JOURNAL_ACCESS_CREATE 0
274#define OCFS2_JOURNAL_ACCESS_WRITE 1
275#define OCFS2_JOURNAL_ACCESS_UNDO 2
276
Joel Becker13723d02008-10-17 19:25:01 -0700277
Joel Becker50655ae2008-09-11 15:53:07 -0700278/* ocfs2_inode */
Joel Becker0cf2f762009-02-12 16:41:25 -0800279int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker13723d02008-10-17 19:25:01 -0700280 struct buffer_head *bh, int type);
Joel Becker50655ae2008-09-11 15:53:07 -0700281/* ocfs2_extent_block */
Joel Becker0cf2f762009-02-12 16:41:25 -0800282int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700283 struct buffer_head *bh, int type);
Tao Ma93c97082009-08-18 11:19:20 +0800284/* ocfs2_refcount_block */
285int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
286 struct buffer_head *bh, int type);
Joel Becker50655ae2008-09-11 15:53:07 -0700287/* ocfs2_group_desc */
Joel Becker0cf2f762009-02-12 16:41:25 -0800288int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700289 struct buffer_head *bh, int type);
290/* ocfs2_xattr_block */
Joel Becker0cf2f762009-02-12 16:41:25 -0800291int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700292 struct buffer_head *bh, int type);
293/* quota blocks */
Joel Becker0cf2f762009-02-12 16:41:25 -0800294int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700295 struct buffer_head *bh, int type);
296/* dirblock */
Joel Becker0cf2f762009-02-12 16:41:25 -0800297int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700298 struct buffer_head *bh, int type);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800299/* ocfs2_dx_root_block */
Joel Becker0cf2f762009-02-12 16:41:25 -0800300int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800301 struct buffer_head *bh, int type);
302/* ocfs2_dx_leaf */
Joel Becker0cf2f762009-02-12 16:41:25 -0800303int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800304 struct buffer_head *bh, int type);
Joel Becker50655ae2008-09-11 15:53:07 -0700305/* Anything that has no ecc */
Joel Becker0cf2f762009-02-12 16:41:25 -0800306int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700307 struct buffer_head *bh, int type);
308
Mark Fashehccd979b2005-12-15 14:31:24 -0800309/*
310 * A word about the journal_access/journal_dirty "dance". It is
311 * entirely legal to journal_access a buffer more than once (as long
312 * as the access type is the same -- I'm not sure what will happen if
313 * access type is different but this should never happen anyway) It is
314 * also legal to journal_dirty a buffer more than once. In fact, you
315 * can even journal_access a buffer after you've done a
316 * journal_access/journal_dirty pair. The only thing you cannot do
317 * however, is journal_dirty a buffer which you haven't yet passed to
318 * journal_access at least once.
319 *
320 * That said, 99% of the time this doesn't matter and this is what the
321 * path looks like:
322 *
323 * <read a bh>
324 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
325 * <modify the bh>
326 * ocfs2_journal_dirty(handle, bh);
327 */
Joel Beckerec20cec2010-03-19 14:13:52 -0700328void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800329
330/*
331 * Credit Macros:
332 * Convenience macros to calculate number of credits needed.
333 *
334 * For convenience sake, I have a set of macros here which calculate
335 * the *maximum* number of sectors which will be changed for various
336 * metadata updates.
337 */
338
339/* simple file updates like chmod, etc. */
340#define OCFS2_INODE_UPDATE_CREDITS 1
341
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800342/* extended attribute block update */
343#define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
344
Jan Kara05849742009-07-22 13:17:21 +0200345/* Update of a single quota block */
346#define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
Jan Karaa90714c2008-10-09 19:38:40 +0200347
Jan Kara05849742009-07-22 13:17:21 +0200348/* global quotafile inode update, data block */
349#define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
350 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
351
352#define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
Jan Karaa90714c2008-10-09 19:38:40 +0200353/*
354 * The two writes below can accidentally see global info dirty due
355 * to set_info() quotactl so make them prepared for the writes.
356 */
357/* quota data block, global info */
358/* Write to local quota file */
Jan Kara05849742009-07-22 13:17:21 +0200359#define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
360 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
Jan Karaa90714c2008-10-09 19:38:40 +0200361
362/* global quota data block, local quota data block, global quota inode,
363 * global quota info */
Jan Kara05849742009-07-22 13:17:21 +0200364#define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
365 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
Jan Karaa90714c2008-10-09 19:38:40 +0200366
367static inline int ocfs2_quota_trans_credits(struct super_block *sb)
368{
369 int credits = 0;
370
371 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
372 credits += OCFS2_QWRITE_CREDITS;
373 if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
374 credits += OCFS2_QWRITE_CREDITS;
375 return credits;
376}
377
Tao Mad6590722007-12-18 15:47:03 +0800378/* group extend. inode update and last group update. */
379#define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
380
Tao Ma7909f2b2007-12-18 15:47:25 +0800381/* group add. inode update and the new group update. */
382#define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
383
Mark Fashehccd979b2005-12-15 14:31:24 -0800384/* get one bit out of a suballocator: dinode + group descriptor +
385 * prev. group desc. if we relink. */
386#define OCFS2_SUBALLOC_ALLOC (3)
387
Jan Karaa90714c2008-10-09 19:38:40 +0200388static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
389{
390 return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
391 ocfs2_quota_trans_credits(sb);
392}
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700393
Mark Fashehccd979b2005-12-15 14:31:24 -0800394/* dinode + group descriptor update. We don't relink on free yet. */
395#define OCFS2_SUBALLOC_FREE (2)
396
397#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
398#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
399 + OCFS2_TRUNCATE_LOG_UPDATE)
400
Jan Karaa90714c2008-10-09 19:38:40 +0200401static inline int ocfs2_remove_extent_credits(struct super_block *sb)
402{
403 return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
404 ocfs2_quota_trans_credits(sb);
405}
Mark Fasheh063c4562007-07-03 13:34:11 -0700406
Mark Fashehccd979b2005-12-15 14:31:24 -0800407/* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
Mark Fashehe7c17e42009-01-29 18:17:46 -0800408 * bitmap block for the new bit) dx_root update for free list */
409#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2 + 1)
Mark Fashehccd979b2005-12-15 14:31:24 -0800410
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800411static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
Jan Karaa90714c2008-10-09 19:38:40 +0200412{
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800413 /* 1 block for index, 2 allocs (data, metadata), 1 clusters
414 * worth of blocks for initial extent. */
415 return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
416 ocfs2_clusters_to_blocks(sb, 1);
417}
418
419/* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
420 * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
421 * blocks + quota update */
422static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
423 int xattr_credits)
424{
425 int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
426
427 if (is_dir)
428 dir_credits += ocfs2_add_dir_index_credits(sb);
429
430 return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
Jan Karaa90714c2008-10-09 19:38:40 +0200431 ocfs2_quota_trans_credits(sb);
432}
Mark Fashehccd979b2005-12-15 14:31:24 -0800433
434/* local alloc metadata change + main bitmap updates */
435#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
436 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
437
438/* used when we don't need an allocation change for a dir extend. One
439 * for the dinode, one for the new block. */
440#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
441
Jan Karaa90714c2008-10-09 19:38:40 +0200442/* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
Mark Fashehe7c17e42009-01-29 18:17:46 -0800443 * update on dir + index leaf + dx root update for free list */
Jan Karaa90714c2008-10-09 19:38:40 +0200444static inline int ocfs2_link_credits(struct super_block *sb)
445{
Mark Fashehe7c17e42009-01-29 18:17:46 -0800446 return 2*OCFS2_INODE_UPDATE_CREDITS + 3 +
Jan Karaa90714c2008-10-09 19:38:40 +0200447 ocfs2_quota_trans_credits(sb);
448}
Mark Fashehccd979b2005-12-15 14:31:24 -0800449
450/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
Mark Fashehe7c17e42009-01-29 18:17:46 -0800451 * dir inode link + dir inode index leaf + dir index root */
Jan Karaa90714c2008-10-09 19:38:40 +0200452static inline int ocfs2_unlink_credits(struct super_block *sb)
453{
454 /* The quota update from ocfs2_link_credits is unused here... */
Mark Fashehe7c17e42009-01-29 18:17:46 -0800455 return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
Jan Karaa90714c2008-10-09 19:38:40 +0200456}
Mark Fashehccd979b2005-12-15 14:31:24 -0800457
458/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
Joel Beckerdfa13f32009-04-29 17:55:08 -0700459 * inode alloc group descriptor + orphan dir index root +
460 * orphan dir index leaf */
461#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
Mark Fashehccd979b2005-12-15 14:31:24 -0800462
463/* dinode update, old dir dinode update, new dir dinode update, old
464 * dir dir entry, new dir dir entry, dir entry update for renaming
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800465 * directory + target unlink + 3 x dir index leaves */
Jan Karaa90714c2008-10-09 19:38:40 +0200466static inline int ocfs2_rename_credits(struct super_block *sb)
467{
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800468 return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
Jan Karaa90714c2008-10-09 19:38:40 +0200469}
Mark Fashehccd979b2005-12-15 14:31:24 -0800470
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800471/* global bitmap dinode, group desc., relinked group,
472 * suballocator dinode, group desc., relinked group,
473 * dinode, xattr block */
474#define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
475 + OCFS2_INODE_UPDATE_CREDITS \
476 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
477
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800478/* inode update, removal of dx root block from allocator */
479#define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
480 OCFS2_SUBALLOC_FREE)
481
Mark Fasheh4ed8a6b2008-11-24 17:02:08 -0800482static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
483{
484 int credits = 1 + OCFS2_SUBALLOC_ALLOC;
485
486 credits += ocfs2_clusters_to_blocks(sb, 1);
487 credits += ocfs2_quota_trans_credits(sb);
488
489 return credits;
490}
491
Tao Ma8bf396d2009-08-24 11:12:02 +0800492/* inode update, new refcount block and its allocation credits. */
493#define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
494 + OCFS2_SUBALLOC_ALLOC)
495
496/* inode and the refcount block update. */
497#define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
498
499/*
500 * inode and the refcount block update.
501 * It doesn't include the credits for sub alloc change.
502 * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
503 */
504#define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
505
Tao Mabcbbb242009-08-18 11:29:12 +0800506/* 2 metadata alloc, 2 new blocks and root refcount block */
507#define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
508
Tao Ma811f9332008-08-18 17:38:43 +0800509/*
510 * Please note that the caller must make sure that root_el is the root
511 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
512 * the result may be wrong.
513 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800514static inline int ocfs2_calc_extend_credits(struct super_block *sb,
Tao Ma811f9332008-08-18 17:38:43 +0800515 struct ocfs2_extent_list *root_el,
Mark Fashehccd979b2005-12-15 14:31:24 -0800516 u32 bits_wanted)
517{
Tao Ma811f9332008-08-18 17:38:43 +0800518 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800519
520 /* bitmap dinode, group desc. + relinked group. */
521 bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
522
523 /* we might need to shift tree depth so lets assume an
524 * absolute worst case of complete fragmentation. Even with
525 * that, we only need one update for the dinode, and then
526 * however many metadata chunks needed * a remaining suballoc
527 * alloc. */
528 sysfile_bitmap_blocks = 1 +
Tao Ma811f9332008-08-18 17:38:43 +0800529 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
Mark Fashehccd979b2005-12-15 14:31:24 -0800530
531 /* this does not include *new* metadata blocks, which are
Tao Ma811f9332008-08-18 17:38:43 +0800532 * accounted for in sysfile_bitmap_blocks. root_el +
Mark Fashehccd979b2005-12-15 14:31:24 -0800533 * prev. last_eb_blk + blocks along edge of tree.
534 * calc_symlink_credits passes because we just need 1
535 * credit for the dinode there. */
Tao Ma811f9332008-08-18 17:38:43 +0800536 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -0800537
Jan Karaa90714c2008-10-09 19:38:40 +0200538 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
539 ocfs2_quota_trans_credits(sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800540}
541
542static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
543{
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800544 int blocks = ocfs2_mknod_credits(sb, 0, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800545
546 /* links can be longer than one block so we may update many
547 * within our single allocated extent. */
548 blocks += ocfs2_clusters_to_blocks(sb, 1);
549
Jan Karaa90714c2008-10-09 19:38:40 +0200550 return blocks + ocfs2_quota_trans_credits(sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800551}
552
553static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
554 unsigned int cpg)
555{
556 int blocks;
557 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
558 /* parent inode update + new block group header + bitmap inode update
559 + bitmap blocks affected */
560 blocks = 1 + 1 + 1 + bitmap_blocks;
561 return blocks;
562}
563
Joel Becker8b06bc52010-03-26 10:09:29 +0800564/*
565 * Allocating a discontiguous block group requires the credits from
566 * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
567 * the group descriptor's extent list. The caller already has started
568 * the transaction with ocfs2_calc_group_alloc_credits(). They extend
569 * it with these credits.
570 */
571static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
572{
573 return ocfs2_extent_recs_per_gd(sb);
574}
575
Mark Fashehccd979b2005-12-15 14:31:24 -0800576static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
577 unsigned int clusters_to_del,
578 struct ocfs2_dinode *fe,
579 struct ocfs2_extent_list *last_el)
580{
581 /* for dinode + all headers in this pass + update to next leaf */
582 u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
583 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
584 int credits = 1 + tree_depth + 1;
585 int i;
586
587 i = next_free - 1;
588 BUG_ON(i < 0);
589
590 /* We may be deleting metadata blocks, so metadata alloc dinode +
591 one desc. block for each possible delete. */
592 if (tree_depth && next_free == 1 &&
Mark Fashehe48edee2007-03-07 16:46:57 -0800593 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
Mark Fashehccd979b2005-12-15 14:31:24 -0800594 credits += 1 + tree_depth;
595
596 /* update to the truncate log. */
597 credits += OCFS2_TRUNCATE_LOG_UPDATE;
598
Jan Karaa90714c2008-10-09 19:38:40 +0200599 credits += ocfs2_quota_trans_credits(sb);
600
Mark Fashehccd979b2005-12-15 14:31:24 -0800601 return credits;
602}
603
Joel Becker2b4e30f2008-09-03 20:03:41 -0700604static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
605{
606 return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
607}
608
609static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
610 loff_t new_size)
611{
Jan Kara7f5aa212009-02-10 11:15:34 -0500612 return jbd2_journal_begin_ordered_truncate(
613 OCFS2_SB(inode->i_sb)->journal->j_journal,
614 &OCFS2_I(inode)->ip_jinode,
615 new_size);
Joel Becker2b4e30f2008-09-03 20:03:41 -0700616}
617
Mark Fashehccd979b2005-12-15 14:31:24 -0800618#endif /* OCFS2_JOURNAL_H */