1*4882a593Smuzhiyun /* SPDX-License-Identifier: GPL-2.0-or-later */
2*4882a593Smuzhiyun /* -*- mode: c; c-basic-offset: 8; -*-
3*4882a593Smuzhiyun * vim: noexpandtab sw=8 ts=8 sts=0:
4*4882a593Smuzhiyun *
5*4882a593Smuzhiyun * journal.h
6*4882a593Smuzhiyun *
7*4882a593Smuzhiyun * Defines journalling api and structures.
8*4882a593Smuzhiyun *
9*4882a593Smuzhiyun * Copyright (C) 2003, 2005 Oracle. All rights reserved.
10*4882a593Smuzhiyun */
11*4882a593Smuzhiyun
12*4882a593Smuzhiyun #ifndef OCFS2_JOURNAL_H
13*4882a593Smuzhiyun #define OCFS2_JOURNAL_H
14*4882a593Smuzhiyun
15*4882a593Smuzhiyun #include <linux/fs.h>
16*4882a593Smuzhiyun #include <linux/jbd2.h>
17*4882a593Smuzhiyun
18*4882a593Smuzhiyun enum ocfs2_journal_state {
19*4882a593Smuzhiyun OCFS2_JOURNAL_FREE = 0,
20*4882a593Smuzhiyun OCFS2_JOURNAL_LOADED,
21*4882a593Smuzhiyun OCFS2_JOURNAL_IN_SHUTDOWN,
22*4882a593Smuzhiyun };
23*4882a593Smuzhiyun
24*4882a593Smuzhiyun struct ocfs2_super;
25*4882a593Smuzhiyun struct ocfs2_dinode;
26*4882a593Smuzhiyun
27*4882a593Smuzhiyun /*
28*4882a593Smuzhiyun * The recovery_list is a simple linked list of node numbers to recover.
29*4882a593Smuzhiyun * It is protected by the recovery_lock.
30*4882a593Smuzhiyun */
31*4882a593Smuzhiyun
32*4882a593Smuzhiyun struct ocfs2_recovery_map {
33*4882a593Smuzhiyun unsigned int rm_used;
34*4882a593Smuzhiyun unsigned int *rm_entries;
35*4882a593Smuzhiyun };
36*4882a593Smuzhiyun
37*4882a593Smuzhiyun
38*4882a593Smuzhiyun struct ocfs2_journal {
39*4882a593Smuzhiyun enum ocfs2_journal_state j_state; /* Journals current state */
40*4882a593Smuzhiyun
41*4882a593Smuzhiyun journal_t *j_journal; /* The kernels journal type */
42*4882a593Smuzhiyun struct inode *j_inode; /* Kernel inode pointing to
43*4882a593Smuzhiyun * this journal */
44*4882a593Smuzhiyun struct ocfs2_super *j_osb; /* pointer to the super
45*4882a593Smuzhiyun * block for the node
46*4882a593Smuzhiyun * we're currently
47*4882a593Smuzhiyun * running on -- not
48*4882a593Smuzhiyun * necessarily the super
49*4882a593Smuzhiyun * block from the node
50*4882a593Smuzhiyun * which we usually run
51*4882a593Smuzhiyun * from (recovery,
52*4882a593Smuzhiyun * etc) */
53*4882a593Smuzhiyun struct buffer_head *j_bh; /* Journal disk inode block */
54*4882a593Smuzhiyun atomic_t j_num_trans; /* Number of transactions
55*4882a593Smuzhiyun * currently in the system. */
56*4882a593Smuzhiyun spinlock_t j_lock;
57*4882a593Smuzhiyun unsigned long j_trans_id;
58*4882a593Smuzhiyun struct rw_semaphore j_trans_barrier;
59*4882a593Smuzhiyun wait_queue_head_t j_checkpointed;
60*4882a593Smuzhiyun
61*4882a593Smuzhiyun /* both fields protected by j_lock*/
62*4882a593Smuzhiyun struct list_head j_la_cleanups;
63*4882a593Smuzhiyun struct work_struct j_recovery_work;
64*4882a593Smuzhiyun };
65*4882a593Smuzhiyun
66*4882a593Smuzhiyun extern spinlock_t trans_inc_lock;
67*4882a593Smuzhiyun
68*4882a593Smuzhiyun /* wrap j_trans_id so we never have it equal to zero. */
ocfs2_inc_trans_id(struct ocfs2_journal * j)69*4882a593Smuzhiyun static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
70*4882a593Smuzhiyun {
71*4882a593Smuzhiyun unsigned long old_id;
72*4882a593Smuzhiyun spin_lock(&trans_inc_lock);
73*4882a593Smuzhiyun old_id = j->j_trans_id++;
74*4882a593Smuzhiyun if (unlikely(!j->j_trans_id))
75*4882a593Smuzhiyun j->j_trans_id = 1;
76*4882a593Smuzhiyun spin_unlock(&trans_inc_lock);
77*4882a593Smuzhiyun return old_id;
78*4882a593Smuzhiyun }
79*4882a593Smuzhiyun
ocfs2_set_ci_lock_trans(struct ocfs2_journal * journal,struct ocfs2_caching_info * ci)80*4882a593Smuzhiyun static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
81*4882a593Smuzhiyun struct ocfs2_caching_info *ci)
82*4882a593Smuzhiyun {
83*4882a593Smuzhiyun spin_lock(&trans_inc_lock);
84*4882a593Smuzhiyun ci->ci_last_trans = journal->j_trans_id;
85*4882a593Smuzhiyun spin_unlock(&trans_inc_lock);
86*4882a593Smuzhiyun }
87*4882a593Smuzhiyun
88*4882a593Smuzhiyun /* Used to figure out whether it's safe to drop a metadata lock on an
89*4882a593Smuzhiyun * cached object. Returns true if all the object's changes have been
90*4882a593Smuzhiyun * checkpointed to disk. You should be holding the spinlock on the
91*4882a593Smuzhiyun * metadata lock while calling this to be sure that nobody can take
92*4882a593Smuzhiyun * the lock and put it on another transaction. */
ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info * ci)93*4882a593Smuzhiyun static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
94*4882a593Smuzhiyun {
95*4882a593Smuzhiyun int ret;
96*4882a593Smuzhiyun struct ocfs2_journal *journal =
97*4882a593Smuzhiyun OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
98*4882a593Smuzhiyun
99*4882a593Smuzhiyun spin_lock(&trans_inc_lock);
100*4882a593Smuzhiyun ret = time_after(journal->j_trans_id, ci->ci_last_trans);
101*4882a593Smuzhiyun spin_unlock(&trans_inc_lock);
102*4882a593Smuzhiyun return ret;
103*4882a593Smuzhiyun }
104*4882a593Smuzhiyun
105*4882a593Smuzhiyun /* convenience function to check if an object backed by struct
106*4882a593Smuzhiyun * ocfs2_caching_info is still new (has never hit disk) Will do you a
107*4882a593Smuzhiyun * favor and set created_trans = 0 when you've
108*4882a593Smuzhiyun * been checkpointed. returns '1' if the ci is still new. */
ocfs2_ci_is_new(struct ocfs2_caching_info * ci)109*4882a593Smuzhiyun static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
110*4882a593Smuzhiyun {
111*4882a593Smuzhiyun int ret;
112*4882a593Smuzhiyun struct ocfs2_journal *journal =
113*4882a593Smuzhiyun OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
114*4882a593Smuzhiyun
115*4882a593Smuzhiyun spin_lock(&trans_inc_lock);
116*4882a593Smuzhiyun ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
117*4882a593Smuzhiyun if (!ret)
118*4882a593Smuzhiyun ci->ci_created_trans = 0;
119*4882a593Smuzhiyun spin_unlock(&trans_inc_lock);
120*4882a593Smuzhiyun return ret;
121*4882a593Smuzhiyun }
122*4882a593Smuzhiyun
123*4882a593Smuzhiyun /* Wrapper for inodes so we can check system files */
ocfs2_inode_is_new(struct inode * inode)124*4882a593Smuzhiyun static inline int ocfs2_inode_is_new(struct inode *inode)
125*4882a593Smuzhiyun {
126*4882a593Smuzhiyun /* System files are never "new" as they're written out by
127*4882a593Smuzhiyun * mkfs. This helps us early during mount, before we have the
128*4882a593Smuzhiyun * journal open and j_trans_id could be junk. */
129*4882a593Smuzhiyun if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
130*4882a593Smuzhiyun return 0;
131*4882a593Smuzhiyun
132*4882a593Smuzhiyun return ocfs2_ci_is_new(INODE_CACHE(inode));
133*4882a593Smuzhiyun }
134*4882a593Smuzhiyun
ocfs2_ci_set_new(struct ocfs2_super * osb,struct ocfs2_caching_info * ci)135*4882a593Smuzhiyun static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
136*4882a593Smuzhiyun struct ocfs2_caching_info *ci)
137*4882a593Smuzhiyun {
138*4882a593Smuzhiyun spin_lock(&trans_inc_lock);
139*4882a593Smuzhiyun ci->ci_created_trans = osb->journal->j_trans_id;
140*4882a593Smuzhiyun spin_unlock(&trans_inc_lock);
141*4882a593Smuzhiyun }
142*4882a593Smuzhiyun
143*4882a593Smuzhiyun /* Exported only for the journal struct init code in super.c. Do not call. */
144*4882a593Smuzhiyun void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
145*4882a593Smuzhiyun void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
146*4882a593Smuzhiyun void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
147*4882a593Smuzhiyun
148*4882a593Smuzhiyun void ocfs2_complete_recovery(struct work_struct *work);
149*4882a593Smuzhiyun void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
150*4882a593Smuzhiyun
151*4882a593Smuzhiyun int ocfs2_recovery_init(struct ocfs2_super *osb);
152*4882a593Smuzhiyun void ocfs2_recovery_exit(struct ocfs2_super *osb);
153*4882a593Smuzhiyun
154*4882a593Smuzhiyun int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
155*4882a593Smuzhiyun /*
156*4882a593Smuzhiyun * Journal Control:
157*4882a593Smuzhiyun * Initialize, Load, Shutdown, Wipe a journal.
158*4882a593Smuzhiyun *
159*4882a593Smuzhiyun * ocfs2_journal_init - Initialize journal structures in the OSB.
160*4882a593Smuzhiyun * ocfs2_journal_load - Load the given journal off disk. Replay it if
161*4882a593Smuzhiyun * there's transactions still in there.
162*4882a593Smuzhiyun * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
163*4882a593Smuzhiyun * uncommitted, uncheckpointed transactions.
164*4882a593Smuzhiyun * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
165*4882a593Smuzhiyun * zero out each block.
166*4882a593Smuzhiyun * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
167*4882a593Smuzhiyun * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
168*4882a593Smuzhiyun * event on.
169*4882a593Smuzhiyun * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
170*4882a593Smuzhiyun */
171*4882a593Smuzhiyun void ocfs2_set_journal_params(struct ocfs2_super *osb);
172*4882a593Smuzhiyun int ocfs2_journal_init(struct ocfs2_journal *journal,
173*4882a593Smuzhiyun int *dirty);
174*4882a593Smuzhiyun void ocfs2_journal_shutdown(struct ocfs2_super *osb);
175*4882a593Smuzhiyun int ocfs2_journal_wipe(struct ocfs2_journal *journal,
176*4882a593Smuzhiyun int full);
177*4882a593Smuzhiyun int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
178*4882a593Smuzhiyun int replayed);
179*4882a593Smuzhiyun int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
180*4882a593Smuzhiyun void ocfs2_recovery_thread(struct ocfs2_super *osb,
181*4882a593Smuzhiyun int node_num);
182*4882a593Smuzhiyun int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
183*4882a593Smuzhiyun void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
184*4882a593Smuzhiyun void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
185*4882a593Smuzhiyun
ocfs2_start_checkpoint(struct ocfs2_super * osb)186*4882a593Smuzhiyun static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
187*4882a593Smuzhiyun {
188*4882a593Smuzhiyun wake_up(&osb->checkpoint_event);
189*4882a593Smuzhiyun }
190*4882a593Smuzhiyun
ocfs2_checkpoint_inode(struct inode * inode)191*4882a593Smuzhiyun static inline void ocfs2_checkpoint_inode(struct inode *inode)
192*4882a593Smuzhiyun {
193*4882a593Smuzhiyun struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
194*4882a593Smuzhiyun
195*4882a593Smuzhiyun if (ocfs2_mount_local(osb))
196*4882a593Smuzhiyun return;
197*4882a593Smuzhiyun
198*4882a593Smuzhiyun if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
199*4882a593Smuzhiyun /* WARNING: This only kicks off a single
200*4882a593Smuzhiyun * checkpoint. If someone races you and adds more
201*4882a593Smuzhiyun * metadata to the journal, you won't know, and will
202*4882a593Smuzhiyun * wind up waiting *a lot* longer than necessary. Right
203*4882a593Smuzhiyun * now we only use this in clear_inode so that's
204*4882a593Smuzhiyun * OK. */
205*4882a593Smuzhiyun ocfs2_start_checkpoint(osb);
206*4882a593Smuzhiyun
207*4882a593Smuzhiyun wait_event(osb->journal->j_checkpointed,
208*4882a593Smuzhiyun ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
209*4882a593Smuzhiyun }
210*4882a593Smuzhiyun }
211*4882a593Smuzhiyun
212*4882a593Smuzhiyun /*
213*4882a593Smuzhiyun * Transaction Handling:
214*4882a593Smuzhiyun * Manage the lifetime of a transaction handle.
215*4882a593Smuzhiyun *
216*4882a593Smuzhiyun * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
217*4882a593Smuzhiyun * the number of blocks that will be changed during
218*4882a593Smuzhiyun * this handle.
219*4882a593Smuzhiyun * ocfs2_commit_trans - Complete a handle. It might return -EIO if
220*4882a593Smuzhiyun * the journal was aborted. The majority of paths don't
221*4882a593Smuzhiyun * check the return value as an error there comes too
222*4882a593Smuzhiyun * late to do anything (and will be picked up in a
223*4882a593Smuzhiyun * later transaction).
224*4882a593Smuzhiyun * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
225*4882a593Smuzhiyun * commit the handle to disk in the process, but will
226*4882a593Smuzhiyun * not release any locks taken during the transaction.
227*4882a593Smuzhiyun * ocfs2_journal_access* - Notify the handle that we want to journal this
228*4882a593Smuzhiyun * buffer. Will have to call ocfs2_journal_dirty once
229*4882a593Smuzhiyun * we've actually dirtied it. Type is one of . or .
230*4882a593Smuzhiyun * Always call the specific flavor of
231*4882a593Smuzhiyun * ocfs2_journal_access_*() unless you intend to
232*4882a593Smuzhiyun * manage the checksum by hand.
233*4882a593Smuzhiyun * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
234*4882a593Smuzhiyun * ocfs2_jbd2_inode_add_write - Mark an inode with range so that its data goes
235*4882a593Smuzhiyun * out before the current handle commits.
236*4882a593Smuzhiyun */
237*4882a593Smuzhiyun
238*4882a593Smuzhiyun /* You must always start_trans with a number of buffs > 0, but it's
239*4882a593Smuzhiyun * perfectly legal to go through an entire transaction without having
240*4882a593Smuzhiyun * dirtied any buffers. */
241*4882a593Smuzhiyun handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
242*4882a593Smuzhiyun int max_buffs);
243*4882a593Smuzhiyun int ocfs2_commit_trans(struct ocfs2_super *osb,
244*4882a593Smuzhiyun handle_t *handle);
245*4882a593Smuzhiyun int ocfs2_extend_trans(handle_t *handle, int nblocks);
246*4882a593Smuzhiyun int ocfs2_allocate_extend_trans(handle_t *handle,
247*4882a593Smuzhiyun int thresh);
248*4882a593Smuzhiyun
249*4882a593Smuzhiyun /*
250*4882a593Smuzhiyun * Define an arbitrary limit for the amount of data we will anticipate
251*4882a593Smuzhiyun * writing to any given transaction. For unbounded transactions such as
252*4882a593Smuzhiyun * fallocate(2) we can write more than this, but we always
253*4882a593Smuzhiyun * start off at the maximum transaction size and grow the transaction
254*4882a593Smuzhiyun * optimistically as we go.
255*4882a593Smuzhiyun */
256*4882a593Smuzhiyun #define OCFS2_MAX_TRANS_DATA 64U
257*4882a593Smuzhiyun
258*4882a593Smuzhiyun /*
259*4882a593Smuzhiyun * Create access is for when we get a newly created buffer and we're
260*4882a593Smuzhiyun * not gonna read it off disk, but rather fill it ourselves. Right
261*4882a593Smuzhiyun * now, we don't do anything special with this (it turns into a write
262*4882a593Smuzhiyun * request), but this is a good placeholder in case we do...
263*4882a593Smuzhiyun *
264*4882a593Smuzhiyun * Write access is for when we read a block off disk and are going to
265*4882a593Smuzhiyun * modify it. This way the journalling layer knows it may need to make
266*4882a593Smuzhiyun * a copy of that block (if it's part of another, uncommitted
267*4882a593Smuzhiyun * transaction) before we do so.
268*4882a593Smuzhiyun */
269*4882a593Smuzhiyun #define OCFS2_JOURNAL_ACCESS_CREATE 0
270*4882a593Smuzhiyun #define OCFS2_JOURNAL_ACCESS_WRITE 1
271*4882a593Smuzhiyun #define OCFS2_JOURNAL_ACCESS_UNDO 2
272*4882a593Smuzhiyun
273*4882a593Smuzhiyun
274*4882a593Smuzhiyun /* ocfs2_inode */
275*4882a593Smuzhiyun int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
276*4882a593Smuzhiyun struct buffer_head *bh, int type);
277*4882a593Smuzhiyun /* ocfs2_extent_block */
278*4882a593Smuzhiyun int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
279*4882a593Smuzhiyun struct buffer_head *bh, int type);
280*4882a593Smuzhiyun /* ocfs2_refcount_block */
281*4882a593Smuzhiyun int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
282*4882a593Smuzhiyun struct buffer_head *bh, int type);
283*4882a593Smuzhiyun /* ocfs2_group_desc */
284*4882a593Smuzhiyun int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
285*4882a593Smuzhiyun struct buffer_head *bh, int type);
286*4882a593Smuzhiyun /* ocfs2_xattr_block */
287*4882a593Smuzhiyun int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
288*4882a593Smuzhiyun struct buffer_head *bh, int type);
289*4882a593Smuzhiyun /* quota blocks */
290*4882a593Smuzhiyun int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
291*4882a593Smuzhiyun struct buffer_head *bh, int type);
292*4882a593Smuzhiyun /* dirblock */
293*4882a593Smuzhiyun int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
294*4882a593Smuzhiyun struct buffer_head *bh, int type);
295*4882a593Smuzhiyun /* ocfs2_dx_root_block */
296*4882a593Smuzhiyun int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
297*4882a593Smuzhiyun struct buffer_head *bh, int type);
298*4882a593Smuzhiyun /* ocfs2_dx_leaf */
299*4882a593Smuzhiyun int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
300*4882a593Smuzhiyun struct buffer_head *bh, int type);
301*4882a593Smuzhiyun /* Anything that has no ecc */
302*4882a593Smuzhiyun int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
303*4882a593Smuzhiyun struct buffer_head *bh, int type);
304*4882a593Smuzhiyun
305*4882a593Smuzhiyun /*
306*4882a593Smuzhiyun * A word about the journal_access/journal_dirty "dance". It is
307*4882a593Smuzhiyun * entirely legal to journal_access a buffer more than once (as long
308*4882a593Smuzhiyun * as the access type is the same -- I'm not sure what will happen if
309*4882a593Smuzhiyun * access type is different but this should never happen anyway) It is
310*4882a593Smuzhiyun * also legal to journal_dirty a buffer more than once. In fact, you
311*4882a593Smuzhiyun * can even journal_access a buffer after you've done a
312*4882a593Smuzhiyun * journal_access/journal_dirty pair. The only thing you cannot do
313*4882a593Smuzhiyun * however, is journal_dirty a buffer which you haven't yet passed to
314*4882a593Smuzhiyun * journal_access at least once.
315*4882a593Smuzhiyun *
316*4882a593Smuzhiyun * That said, 99% of the time this doesn't matter and this is what the
317*4882a593Smuzhiyun * path looks like:
318*4882a593Smuzhiyun *
319*4882a593Smuzhiyun * <read a bh>
320*4882a593Smuzhiyun * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
321*4882a593Smuzhiyun * <modify the bh>
322*4882a593Smuzhiyun * ocfs2_journal_dirty(handle, bh);
323*4882a593Smuzhiyun */
324*4882a593Smuzhiyun void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
325*4882a593Smuzhiyun
326*4882a593Smuzhiyun /*
327*4882a593Smuzhiyun * Credit Macros:
328*4882a593Smuzhiyun * Convenience macros to calculate number of credits needed.
329*4882a593Smuzhiyun *
330*4882a593Smuzhiyun * For convenience sake, I have a set of macros here which calculate
331*4882a593Smuzhiyun * the *maximum* number of sectors which will be changed for various
332*4882a593Smuzhiyun * metadata updates.
333*4882a593Smuzhiyun */
334*4882a593Smuzhiyun
335*4882a593Smuzhiyun /* simple file updates like chmod, etc. */
336*4882a593Smuzhiyun #define OCFS2_INODE_UPDATE_CREDITS 1
337*4882a593Smuzhiyun
338*4882a593Smuzhiyun /* extended attribute block update */
339*4882a593Smuzhiyun #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
340*4882a593Smuzhiyun
341*4882a593Smuzhiyun /* Update of a single quota block */
342*4882a593Smuzhiyun #define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
343*4882a593Smuzhiyun
344*4882a593Smuzhiyun /* global quotafile inode update, data block */
345*4882a593Smuzhiyun #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
346*4882a593Smuzhiyun OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
347*4882a593Smuzhiyun
348*4882a593Smuzhiyun #define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
349*4882a593Smuzhiyun /*
350*4882a593Smuzhiyun * The two writes below can accidentally see global info dirty due
351*4882a593Smuzhiyun * to set_info() quotactl so make them prepared for the writes.
352*4882a593Smuzhiyun */
353*4882a593Smuzhiyun /* quota data block, global info */
354*4882a593Smuzhiyun /* Write to local quota file */
355*4882a593Smuzhiyun #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
356*4882a593Smuzhiyun OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
357*4882a593Smuzhiyun
358*4882a593Smuzhiyun /* global quota data block, local quota data block, global quota inode,
359*4882a593Smuzhiyun * global quota info */
360*4882a593Smuzhiyun #define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
361*4882a593Smuzhiyun 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
362*4882a593Smuzhiyun
ocfs2_quota_trans_credits(struct super_block * sb)363*4882a593Smuzhiyun static inline int ocfs2_quota_trans_credits(struct super_block *sb)
364*4882a593Smuzhiyun {
365*4882a593Smuzhiyun int credits = 0;
366*4882a593Smuzhiyun
367*4882a593Smuzhiyun if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
368*4882a593Smuzhiyun credits += OCFS2_QWRITE_CREDITS;
369*4882a593Smuzhiyun if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
370*4882a593Smuzhiyun credits += OCFS2_QWRITE_CREDITS;
371*4882a593Smuzhiyun return credits;
372*4882a593Smuzhiyun }
373*4882a593Smuzhiyun
374*4882a593Smuzhiyun /* group extend. inode update and last group update. */
375*4882a593Smuzhiyun #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
376*4882a593Smuzhiyun
377*4882a593Smuzhiyun /* group add. inode update and the new group update. */
378*4882a593Smuzhiyun #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
379*4882a593Smuzhiyun
380*4882a593Smuzhiyun /* get one bit out of a suballocator: dinode + group descriptor +
381*4882a593Smuzhiyun * prev. group desc. if we relink. */
382*4882a593Smuzhiyun #define OCFS2_SUBALLOC_ALLOC (3)
383*4882a593Smuzhiyun
ocfs2_inline_to_extents_credits(struct super_block * sb)384*4882a593Smuzhiyun static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
385*4882a593Smuzhiyun {
386*4882a593Smuzhiyun return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
387*4882a593Smuzhiyun ocfs2_quota_trans_credits(sb);
388*4882a593Smuzhiyun }
389*4882a593Smuzhiyun
390*4882a593Smuzhiyun /* dinode + group descriptor update. We don't relink on free yet. */
391*4882a593Smuzhiyun #define OCFS2_SUBALLOC_FREE (2)
392*4882a593Smuzhiyun
393*4882a593Smuzhiyun #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
394*4882a593Smuzhiyun #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
395*4882a593Smuzhiyun + OCFS2_TRUNCATE_LOG_UPDATE)
396*4882a593Smuzhiyun
ocfs2_remove_extent_credits(struct super_block * sb)397*4882a593Smuzhiyun static inline int ocfs2_remove_extent_credits(struct super_block *sb)
398*4882a593Smuzhiyun {
399*4882a593Smuzhiyun return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
400*4882a593Smuzhiyun ocfs2_quota_trans_credits(sb);
401*4882a593Smuzhiyun }
402*4882a593Smuzhiyun
403*4882a593Smuzhiyun /* data block for new dir/symlink, allocation of directory block, dx_root
404*4882a593Smuzhiyun * update for free list */
405*4882a593Smuzhiyun #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
406*4882a593Smuzhiyun
ocfs2_add_dir_index_credits(struct super_block * sb)407*4882a593Smuzhiyun static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
408*4882a593Smuzhiyun {
409*4882a593Smuzhiyun /* 1 block for index, 2 allocs (data, metadata), 1 clusters
410*4882a593Smuzhiyun * worth of blocks for initial extent. */
411*4882a593Smuzhiyun return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
412*4882a593Smuzhiyun ocfs2_clusters_to_blocks(sb, 1);
413*4882a593Smuzhiyun }
414*4882a593Smuzhiyun
415*4882a593Smuzhiyun /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
416*4882a593Smuzhiyun * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
417*4882a593Smuzhiyun * blocks + quota update */
ocfs2_mknod_credits(struct super_block * sb,int is_dir,int xattr_credits)418*4882a593Smuzhiyun static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
419*4882a593Smuzhiyun int xattr_credits)
420*4882a593Smuzhiyun {
421*4882a593Smuzhiyun int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
422*4882a593Smuzhiyun
423*4882a593Smuzhiyun if (is_dir)
424*4882a593Smuzhiyun dir_credits += ocfs2_add_dir_index_credits(sb);
425*4882a593Smuzhiyun
426*4882a593Smuzhiyun return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
427*4882a593Smuzhiyun ocfs2_quota_trans_credits(sb);
428*4882a593Smuzhiyun }
429*4882a593Smuzhiyun
430*4882a593Smuzhiyun /* local alloc metadata change + main bitmap updates */
431*4882a593Smuzhiyun #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
432*4882a593Smuzhiyun + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
433*4882a593Smuzhiyun
434*4882a593Smuzhiyun /* used when we don't need an allocation change for a dir extend. One
435*4882a593Smuzhiyun * for the dinode, one for the new block. */
436*4882a593Smuzhiyun #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
437*4882a593Smuzhiyun
438*4882a593Smuzhiyun /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
439*4882a593Smuzhiyun * update on dir + index leaf + dx root update for free list +
440*4882a593Smuzhiyun * previous dirblock update in the free list */
ocfs2_link_credits(struct super_block * sb)441*4882a593Smuzhiyun static inline int ocfs2_link_credits(struct super_block *sb)
442*4882a593Smuzhiyun {
443*4882a593Smuzhiyun return 2 * OCFS2_INODE_UPDATE_CREDITS + 4 +
444*4882a593Smuzhiyun ocfs2_quota_trans_credits(sb);
445*4882a593Smuzhiyun }
446*4882a593Smuzhiyun
447*4882a593Smuzhiyun /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
448*4882a593Smuzhiyun * dir inode link + dir inode index leaf + dir index root */
ocfs2_unlink_credits(struct super_block * sb)449*4882a593Smuzhiyun static inline int ocfs2_unlink_credits(struct super_block *sb)
450*4882a593Smuzhiyun {
451*4882a593Smuzhiyun /* The quota update from ocfs2_link_credits is unused here... */
452*4882a593Smuzhiyun return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
453*4882a593Smuzhiyun }
454*4882a593Smuzhiyun
455*4882a593Smuzhiyun /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
456*4882a593Smuzhiyun * inode alloc group descriptor + orphan dir index root +
457*4882a593Smuzhiyun * orphan dir index leaf */
458*4882a593Smuzhiyun #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
459*4882a593Smuzhiyun
460*4882a593Smuzhiyun /* dinode + orphan dir dinode + extent tree leaf block + orphan dir entry +
461*4882a593Smuzhiyun * orphan dir index root + orphan dir index leaf */
462*4882a593Smuzhiyun #define OCFS2_INODE_ADD_TO_ORPHAN_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 4)
463*4882a593Smuzhiyun #define OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS OCFS2_INODE_ADD_TO_ORPHAN_CREDITS
464*4882a593Smuzhiyun
465*4882a593Smuzhiyun /* dinode update, old dir dinode update, new dir dinode update, old
466*4882a593Smuzhiyun * dir dir entry, new dir dir entry, dir entry update for renaming
467*4882a593Smuzhiyun * directory + target unlink + 3 x dir index leaves */
ocfs2_rename_credits(struct super_block * sb)468*4882a593Smuzhiyun static inline int ocfs2_rename_credits(struct super_block *sb)
469*4882a593Smuzhiyun {
470*4882a593Smuzhiyun return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
471*4882a593Smuzhiyun }
472*4882a593Smuzhiyun
473*4882a593Smuzhiyun /* global bitmap dinode, group desc., relinked group,
474*4882a593Smuzhiyun * suballocator dinode, group desc., relinked group,
475*4882a593Smuzhiyun * dinode, xattr block */
476*4882a593Smuzhiyun #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
477*4882a593Smuzhiyun + OCFS2_INODE_UPDATE_CREDITS \
478*4882a593Smuzhiyun + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
479*4882a593Smuzhiyun
480*4882a593Smuzhiyun /* inode update, removal of dx root block from allocator */
481*4882a593Smuzhiyun #define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
482*4882a593Smuzhiyun OCFS2_SUBALLOC_FREE)
483*4882a593Smuzhiyun
ocfs2_calc_dxi_expand_credits(struct super_block * sb)484*4882a593Smuzhiyun static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
485*4882a593Smuzhiyun {
486*4882a593Smuzhiyun int credits = 1 + OCFS2_SUBALLOC_ALLOC;
487*4882a593Smuzhiyun
488*4882a593Smuzhiyun credits += ocfs2_clusters_to_blocks(sb, 1);
489*4882a593Smuzhiyun credits += ocfs2_quota_trans_credits(sb);
490*4882a593Smuzhiyun
491*4882a593Smuzhiyun return credits;
492*4882a593Smuzhiyun }
493*4882a593Smuzhiyun
494*4882a593Smuzhiyun /* inode update, new refcount block and its allocation credits. */
495*4882a593Smuzhiyun #define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
496*4882a593Smuzhiyun + OCFS2_SUBALLOC_ALLOC)
497*4882a593Smuzhiyun
498*4882a593Smuzhiyun /* inode and the refcount block update. */
499*4882a593Smuzhiyun #define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
500*4882a593Smuzhiyun
501*4882a593Smuzhiyun /*
502*4882a593Smuzhiyun * inode and the refcount block update.
503*4882a593Smuzhiyun * It doesn't include the credits for sub alloc change.
504*4882a593Smuzhiyun * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
505*4882a593Smuzhiyun */
506*4882a593Smuzhiyun #define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
507*4882a593Smuzhiyun
508*4882a593Smuzhiyun /* 2 metadata alloc, 2 new blocks and root refcount block */
509*4882a593Smuzhiyun #define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
510*4882a593Smuzhiyun
511*4882a593Smuzhiyun /*
512*4882a593Smuzhiyun * Please note that the caller must make sure that root_el is the root
513*4882a593Smuzhiyun * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
514*4882a593Smuzhiyun * the result may be wrong.
515*4882a593Smuzhiyun */
ocfs2_calc_extend_credits(struct super_block * sb,struct ocfs2_extent_list * root_el)516*4882a593Smuzhiyun static inline int ocfs2_calc_extend_credits(struct super_block *sb,
517*4882a593Smuzhiyun struct ocfs2_extent_list *root_el)
518*4882a593Smuzhiyun {
519*4882a593Smuzhiyun int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
520*4882a593Smuzhiyun
521*4882a593Smuzhiyun /* bitmap dinode, group desc. + relinked group. */
522*4882a593Smuzhiyun bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
523*4882a593Smuzhiyun
524*4882a593Smuzhiyun /* we might need to shift tree depth so lets assume an
525*4882a593Smuzhiyun * absolute worst case of complete fragmentation. Even with
526*4882a593Smuzhiyun * that, we only need one update for the dinode, and then
527*4882a593Smuzhiyun * however many metadata chunks needed * a remaining suballoc
528*4882a593Smuzhiyun * alloc. */
529*4882a593Smuzhiyun sysfile_bitmap_blocks = 1 +
530*4882a593Smuzhiyun (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
531*4882a593Smuzhiyun
532*4882a593Smuzhiyun /* this does not include *new* metadata blocks, which are
533*4882a593Smuzhiyun * accounted for in sysfile_bitmap_blocks. root_el +
534*4882a593Smuzhiyun * prev. last_eb_blk + blocks along edge of tree.
535*4882a593Smuzhiyun * calc_symlink_credits passes because we just need 1
536*4882a593Smuzhiyun * credit for the dinode there. */
537*4882a593Smuzhiyun extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
538*4882a593Smuzhiyun
539*4882a593Smuzhiyun return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
540*4882a593Smuzhiyun ocfs2_quota_trans_credits(sb);
541*4882a593Smuzhiyun }
542*4882a593Smuzhiyun
ocfs2_calc_symlink_credits(struct super_block * sb)543*4882a593Smuzhiyun static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
544*4882a593Smuzhiyun {
545*4882a593Smuzhiyun int blocks = ocfs2_mknod_credits(sb, 0, 0);
546*4882a593Smuzhiyun
547*4882a593Smuzhiyun /* links can be longer than one block so we may update many
548*4882a593Smuzhiyun * within our single allocated extent. */
549*4882a593Smuzhiyun blocks += ocfs2_clusters_to_blocks(sb, 1);
550*4882a593Smuzhiyun
551*4882a593Smuzhiyun return blocks + ocfs2_quota_trans_credits(sb);
552*4882a593Smuzhiyun }
553*4882a593Smuzhiyun
ocfs2_calc_group_alloc_credits(struct super_block * sb,unsigned int cpg)554*4882a593Smuzhiyun static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
555*4882a593Smuzhiyun unsigned int cpg)
556*4882a593Smuzhiyun {
557*4882a593Smuzhiyun int blocks;
558*4882a593Smuzhiyun int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
559*4882a593Smuzhiyun /* parent inode update + new block group header + bitmap inode update
560*4882a593Smuzhiyun + bitmap blocks affected */
561*4882a593Smuzhiyun blocks = 1 + 1 + 1 + bitmap_blocks;
562*4882a593Smuzhiyun return blocks;
563*4882a593Smuzhiyun }
564*4882a593Smuzhiyun
565*4882a593Smuzhiyun /*
566*4882a593Smuzhiyun * Allocating a discontiguous block group requires the credits from
567*4882a593Smuzhiyun * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
568*4882a593Smuzhiyun * the group descriptor's extent list. The caller already has started
569*4882a593Smuzhiyun * the transaction with ocfs2_calc_group_alloc_credits(). They extend
570*4882a593Smuzhiyun * it with these credits.
571*4882a593Smuzhiyun */
ocfs2_calc_bg_discontig_credits(struct super_block * sb)572*4882a593Smuzhiyun static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
573*4882a593Smuzhiyun {
574*4882a593Smuzhiyun return ocfs2_extent_recs_per_gd(sb);
575*4882a593Smuzhiyun }
576*4882a593Smuzhiyun
ocfs2_jbd2_inode_add_write(handle_t * handle,struct inode * inode,loff_t start_byte,loff_t length)577*4882a593Smuzhiyun static inline int ocfs2_jbd2_inode_add_write(handle_t *handle, struct inode *inode,
578*4882a593Smuzhiyun loff_t start_byte, loff_t length)
579*4882a593Smuzhiyun {
580*4882a593Smuzhiyun return jbd2_journal_inode_ranged_write(handle,
581*4882a593Smuzhiyun &OCFS2_I(inode)->ip_jinode,
582*4882a593Smuzhiyun start_byte, length);
583*4882a593Smuzhiyun }
584*4882a593Smuzhiyun
ocfs2_begin_ordered_truncate(struct inode * inode,loff_t new_size)585*4882a593Smuzhiyun static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
586*4882a593Smuzhiyun loff_t new_size)
587*4882a593Smuzhiyun {
588*4882a593Smuzhiyun return jbd2_journal_begin_ordered_truncate(
589*4882a593Smuzhiyun OCFS2_SB(inode->i_sb)->journal->j_journal,
590*4882a593Smuzhiyun &OCFS2_I(inode)->ip_jinode,
591*4882a593Smuzhiyun new_size);
592*4882a593Smuzhiyun }
593*4882a593Smuzhiyun
ocfs2_update_inode_fsync_trans(handle_t * handle,struct inode * inode,int datasync)594*4882a593Smuzhiyun static inline void ocfs2_update_inode_fsync_trans(handle_t *handle,
595*4882a593Smuzhiyun struct inode *inode,
596*4882a593Smuzhiyun int datasync)
597*4882a593Smuzhiyun {
598*4882a593Smuzhiyun struct ocfs2_inode_info *oi = OCFS2_I(inode);
599*4882a593Smuzhiyun
600*4882a593Smuzhiyun if (!is_handle_aborted(handle)) {
601*4882a593Smuzhiyun oi->i_sync_tid = handle->h_transaction->t_tid;
602*4882a593Smuzhiyun if (datasync)
603*4882a593Smuzhiyun oi->i_datasync_tid = handle->h_transaction->t_tid;
604*4882a593Smuzhiyun }
605*4882a593Smuzhiyun }
606*4882a593Smuzhiyun
607*4882a593Smuzhiyun #endif /* OCFS2_JOURNAL_H */
608