blob: 23103d68423c58faa60bfb38177d87f40311775e [file] [log] [blame]
Dave Chinner0b61f8a2018-06-05 19:42:14 -07001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Nathan Scott7b718762005-11-02 14:58:39 +11003 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 */
6#ifndef __XFS_LOG_PRIV_H__
7#define __XFS_LOG_PRIV_H__
8
9struct xfs_buf;
Mark Tinguelyad223e62012-06-14 09:22:15 -050010struct xlog;
Nathan Scotta844f452005-11-02 14:38:42 +110011struct xlog_ticket;
Linus Torvalds1da177e2005-04-16 15:20:36 -070012struct xfs_mount;
13
14/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 * get client id from packed copy.
16 *
17 * this hack is here because the xlog_pack code copies four bytes
18 * of xlog_op_header containing the fields oh_clientid, oh_flags
19 * and oh_res2 into the packed copy.
20 *
21 * later on this four byte chunk is treated as an int and the
22 * client id is pulled out.
23 *
24 * this has endian issues, of course.
25 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +100026static inline uint xlog_get_client_id(__be32 i)
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100027{
Christoph Hellwigb53e6752007-10-12 10:59:34 +100028 return be32_to_cpu(i) >> 24;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100029}
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Linus Torvalds1da177e2005-04-16 15:20:36 -070031/*
32 * In core log state
33 */
Christoph Hellwig1858bb02019-10-14 10:36:43 -070034enum xlog_iclog_state {
35 XLOG_STATE_ACTIVE, /* Current IC log being written to */
36 XLOG_STATE_WANT_SYNC, /* Want to sync this iclog; no more writes */
37 XLOG_STATE_SYNCING, /* This IC log is syncing */
38 XLOG_STATE_DONE_SYNC, /* Done syncing to disk */
Christoph Hellwig1858bb02019-10-14 10:36:43 -070039 XLOG_STATE_CALLBACK, /* Callback functions now */
40 XLOG_STATE_DIRTY, /* Dirty IC log, not ready for ACTIVE status */
Christoph Hellwig1858bb02019-10-14 10:36:43 -070041};
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Dave Chinner956f6da2021-06-18 11:57:05 -070043#define XLOG_STATE_STRINGS \
44 { XLOG_STATE_ACTIVE, "XLOG_STATE_ACTIVE" }, \
45 { XLOG_STATE_WANT_SYNC, "XLOG_STATE_WANT_SYNC" }, \
46 { XLOG_STATE_SYNCING, "XLOG_STATE_SYNCING" }, \
47 { XLOG_STATE_DONE_SYNC, "XLOG_STATE_DONE_SYNC" }, \
48 { XLOG_STATE_CALLBACK, "XLOG_STATE_CALLBACK" }, \
Dave Chinner5112e2062021-08-10 17:59:01 -070049 { XLOG_STATE_DIRTY, "XLOG_STATE_DIRTY" }
Dave Chinner956f6da2021-06-18 11:57:05 -070050
Dave Chinnerb2ae3a92021-07-27 16:23:50 -070051/*
52 * In core log flags
53 */
54#define XLOG_ICL_NEED_FLUSH (1 << 0) /* iclog needs REQ_PREFLUSH */
55#define XLOG_ICL_NEED_FUA (1 << 1) /* iclog needs REQ_FUA */
56
57#define XLOG_ICL_STRINGS \
58 { XLOG_ICL_NEED_FLUSH, "XLOG_ICL_NEED_FLUSH" }, \
59 { XLOG_ICL_NEED_FUA, "XLOG_ICL_NEED_FUA" }
60
Dave Chinner956f6da2021-06-18 11:57:05 -070061
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/*
Dave Chinner70e42f22020-03-25 18:18:22 -070063 * Log ticket flags
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 */
Dave Chinner70e42f22020-03-25 18:18:22 -070065#define XLOG_TIC_PERM_RESERV 0x1 /* permanent reservation */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000066
67#define XLOG_TIC_FLAGS \
Dave Chinner10547942010-12-21 12:02:25 +110068 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000069
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/*
71 * Below are states for covering allocation transactions.
72 * By covering, we mean changing the h_tail_lsn in the last on-disk
73 * log write such that no allocation transactions will be re-done during
74 * recovery after a system crash. Recovery starts at the last on-disk
75 * log write.
76 *
77 * These states are used to insert dummy log entries to cover
78 * space allocation transactions which can undo non-transactional changes
79 * after a crash. Writes to a file with space
80 * already allocated do not result in any transactions. Allocations
81 * might include space beyond the EOF. So if we just push the EOF a
82 * little, the last transaction for the file could contain the wrong
83 * size. If there is no file system activity, after an allocation
84 * transaction, and the system crashes, the allocation transaction
85 * will get replayed and the file will be truncated. This could
86 * be hours/days/... after the allocation occurred.
87 *
88 * The fix for this is to do two dummy transactions when the
89 * system is idle. We need two dummy transaction because the h_tail_lsn
90 * in the log record header needs to point beyond the last possible
91 * non-dummy transaction. The first dummy changes the h_tail_lsn to
92 * the first transaction before the dummy. The second dummy causes
93 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
94 *
95 * These dummy transactions get committed when everything
96 * is idle (after there has been some activity).
97 *
98 * There are 5 states used to control this.
99 *
100 * IDLE -- no logging has been done on the file system or
101 * we are done covering previous transactions.
102 * NEED -- logging has occurred and we need a dummy transaction
103 * when the log becomes idle.
104 * DONE -- we were in the NEED state and have committed a dummy
105 * transaction.
106 * NEED2 -- we detected that a dummy transaction has gone to the
107 * on disk log with no other transactions.
108 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
109 *
110 * There are two places where we switch states:
111 *
112 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
113 * We commit the dummy transaction and switch to DONE or DONE2,
114 * respectively. In all other states, we don't do anything.
115 *
116 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
117 *
118 * No matter what state we are in, if this isn't the dummy
119 * transaction going out, the next state is NEED.
120 * So, if we aren't in the DONE or DONE2 states, the next state
121 * is NEED. We can't be finishing a write of the dummy record
122 * unless it was committed and the state switched to DONE or DONE2.
123 *
124 * If we are in the DONE state and this was a write of the
125 * dummy transaction, we move to NEED2.
126 *
127 * If we are in the DONE2 state and this was a write of the
128 * dummy transaction, we move to IDLE.
129 *
130 *
131 * Writing only one dummy transaction can get appended to
132 * one file space allocation. When this happens, the log recovery
133 * code replays the space allocation and a file could be truncated.
134 * This is why we have the NEED2 and DONE2 states before going idle.
135 */
136
137#define XLOG_STATE_COVER_IDLE 0
138#define XLOG_STATE_COVER_NEED 1
139#define XLOG_STATE_COVER_DONE 2
140#define XLOG_STATE_COVER_NEED2 3
141#define XLOG_STATE_COVER_DONE2 4
142
143#define XLOG_COVER_OPS 5
144
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000145/* Ticket reservation region accounting */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000146#define XLOG_TIC_LEN_MAX 15
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000147
148/*
149 * Reservation region
150 * As would be stored in xfs_log_iovec but without the i_addr which
151 * we don't care about.
152 */
153typedef struct xlog_res {
Tim Shimmin12598452006-01-11 21:02:47 +1100154 uint r_len; /* region length :4 */
155 uint r_type; /* region's transaction type :4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000156} xlog_res_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158typedef struct xlog_ticket {
Dave Chinner10547942010-12-21 12:02:25 +1100159 struct list_head t_queue; /* reserve/write queue */
Christoph Hellwig14a7235f2012-02-20 02:31:24 +0000160 struct task_struct *t_task; /* task that owns this ticket */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000161 xlog_tid_t t_tid; /* transaction identifier : 4 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100162 atomic_t t_ref; /* ticket reference count : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000163 int t_curr_res; /* current reservation in bytes : 4 */
164 int t_unit_res; /* unit reservation in bytes : 4 */
165 char t_ocnt; /* original count : 1 */
166 char t_cnt; /* current count : 1 */
167 char t_clientid; /* who does this belong to; : 1 */
168 char t_flags; /* properties of reservation : 1 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000169
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000170 /* reservation array fields */
171 uint t_res_num; /* num in array : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000172 uint t_res_num_ophdrs; /* num op hdrs : 4 */
173 uint t_res_arr_sum; /* array sum : 4 */
174 uint t_res_o_flow; /* sum overflow : 4 */
Tim Shimmin12598452006-01-11 21:02:47 +1100175 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176} xlog_ticket_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178/*
179 * - A log record header is 512 bytes. There is plenty of room to grow the
180 * xlog_rec_header_t into the reserved space.
181 * - ic_data follows, so a write to disk can start at the beginning of
182 * the iclog.
David Chinner12017fa2008-08-13 16:34:31 +1000183 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 * - ic_next is the pointer to the next iclog in the ring.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 * - ic_log is a pointer back to the global log structure.
Christoph Hellwig79b54d92019-06-28 19:27:25 -0700186 * - ic_size is the full size of the log buffer, minus the cycle headers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 * - ic_offset is the current number of bytes written to in this iclog.
188 * - ic_refcnt is bumped when someone is writing to the log.
189 * - ic_state is the state of the iclog.
David Chinner114d23a2008-04-10 12:18:39 +1000190 *
191 * Because of cacheline contention on large machines, we need to separate
192 * various resources onto different cachelines. To start with, make the
193 * structure cacheline aligned. The following fields can be contended on
194 * by independent processes:
195 *
Christoph Hellwig89ae3792019-06-28 19:27:34 -0700196 * - ic_callbacks
David Chinner114d23a2008-04-10 12:18:39 +1000197 * - ic_refcnt
198 * - fields protected by the global l_icloglock
199 *
200 * so we need to ensure that these fields are located in separate cachelines.
201 * We'll put all the read-only and l_icloglock fields in the first cacheline,
202 * and move everything else out to subsequent cachelines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 */
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100204typedef struct xlog_in_core {
Dave Chinnereb40a872010-12-21 12:09:01 +1100205 wait_queue_head_t ic_force_wait;
206 wait_queue_head_t ic_write_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 struct xlog_in_core *ic_next;
208 struct xlog_in_core *ic_prev;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500209 struct xlog *ic_log;
Christoph Hellwig79b54d92019-06-28 19:27:25 -0700210 u32 ic_size;
Christoph Hellwig79b54d92019-06-28 19:27:25 -0700211 u32 ic_offset;
Christoph Hellwig1858bb02019-10-14 10:36:43 -0700212 enum xlog_iclog_state ic_state;
Dave Chinnereef983f2021-06-18 08:21:51 -0700213 unsigned int ic_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 char *ic_datap; /* pointer to iclog data */
Christoph Hellwig89ae3792019-06-28 19:27:34 -0700215 struct list_head ic_callbacks;
David Chinner114d23a2008-04-10 12:18:39 +1000216
217 /* reference counts need their own cacheline */
218 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100219 xlog_in_core_2_t *ic_data;
220#define ic_header ic_data->hic_header
Christoph Hellwig366fc4b2019-06-28 19:27:21 -0700221#ifdef DEBUG
222 bool ic_fail_crc : 1;
223#endif
Christoph Hellwig79b54d92019-06-28 19:27:25 -0700224 struct semaphore ic_sema;
225 struct work_struct ic_end_io_work;
226 struct bio ic_bio;
227 struct bio_vec ic_bvec[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228} xlog_in_core_t;
229
230/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000231 * The CIL context is used to aggregate per-transaction details as well be
232 * passed to the iclog for checkpoint post-commit processing. After being
233 * passed to the iclog, another context needs to be allocated for tracking the
234 * next set of transactions to be aggregated into a checkpoint.
235 */
236struct xfs_cil;
237
238struct xfs_cil_ctx {
239 struct xfs_cil *cil;
Dave Chinner5f9b4b02021-06-18 08:21:52 -0700240 xfs_csn_t sequence; /* chkpt sequence # */
Dave Chinner71e330b2010-05-21 14:37:18 +1000241 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
242 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
Dave Chinnercaa80092021-08-10 18:00:43 -0700243 struct xlog_in_core *commit_iclog;
Dave Chinner71e330b2010-05-21 14:37:18 +1000244 struct xlog_ticket *ticket; /* chkpt ticket */
245 int nvecs; /* number of regions */
246 int space_used; /* aggregate size of regions */
247 struct list_head busy_extents; /* busy extents in chkpt */
248 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
Christoph Hellwig89ae3792019-06-28 19:27:34 -0700249 struct list_head iclog_entry;
Dave Chinner71e330b2010-05-21 14:37:18 +1000250 struct list_head committing; /* ctx committing list */
Christoph Hellwig4560e782017-02-07 14:07:58 -0800251 struct work_struct discard_endio_work;
Dave Chinner39823d02021-08-10 18:00:45 -0700252 struct work_struct push_work;
Dave Chinner71e330b2010-05-21 14:37:18 +1000253};
254
255/*
256 * Committed Item List structure
257 *
258 * This structure is used to track log items that have been committed but not
259 * yet written into the log. It is used only when the delayed logging mount
260 * option is enabled.
261 *
262 * This structure tracks the list of committing checkpoint contexts so
263 * we can avoid the problem of having to hold out new transactions during a
264 * flush until we have a the commit record LSN of the checkpoint. We can
265 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
266 * sequence match and extract the commit LSN directly from there. If the
267 * checkpoint is still in the process of committing, we can block waiting for
268 * the commit LSN to be determined as well. This should make synchronous
269 * operations almost as efficient as the old logging methods.
270 */
271struct xfs_cil {
Mark Tinguelyad223e62012-06-14 09:22:15 -0500272 struct xlog *xc_log;
Dave Chinner71e330b2010-05-21 14:37:18 +1000273 struct list_head xc_cil;
274 spinlock_t xc_cil_lock;
Dave Chinner33c0dd72021-08-10 18:00:45 -0700275 struct workqueue_struct *xc_push_wq;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000276
277 struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000278 struct xfs_cil_ctx *xc_ctx;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000279
280 spinlock_t xc_push_lock ____cacheline_aligned_in_smp;
Dave Chinner5f9b4b02021-06-18 08:21:52 -0700281 xfs_csn_t xc_push_seq;
Dave Chinner0020a192021-08-10 18:00:44 -0700282 bool xc_push_commit_stable;
Dave Chinner71e330b2010-05-21 14:37:18 +1000283 struct list_head xc_committing;
Dave Chinnereb40a872010-12-21 12:09:01 +1100284 wait_queue_head_t xc_commit_wait;
Dave Chinner68a74dc2021-08-10 18:00:44 -0700285 wait_queue_head_t xc_start_wait;
Dave Chinner5f9b4b02021-06-18 08:21:52 -0700286 xfs_csn_t xc_current_sequence;
Dave Chinnerc7f87f32020-06-16 08:57:43 -0700287 wait_queue_head_t xc_push_wait; /* background push throttle */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000288} ____cacheline_aligned_in_smp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000289
290/*
Dave Chinner80168672010-09-24 18:13:44 +1000291 * The amount of log space we allow the CIL to aggregate is difficult to size.
292 * Whatever we choose, we have to make sure we can get a reservation for the
293 * log space effectively, that it is large enough to capture sufficient
294 * relogging to reduce log buffer IO significantly, but it is not too large for
295 * the log or induces too much latency when writing out through the iclogs. We
296 * track both space consumed and the number of vectors in the checkpoint
297 * context, so we need to decide which to use for limiting.
Dave Chinnerdf806152010-05-17 15:52:13 +1000298 *
299 * Every log buffer we write out during a push needs a header reserved, which
300 * is at least one sector and more for v2 logs. Hence we need a reservation of
301 * at least 512 bytes per 32k of log space just for the LR headers. That means
302 * 16KB of reservation per megabyte of delayed logging space we will consume,
303 * plus various headers. The number of headers will vary based on the num of
304 * io vectors, so limiting on a specific number of vectors is going to result
305 * in transactions of varying size. IOWs, it is more consistent to track and
306 * limit space consumed in the log rather than by the number of objects being
307 * logged in order to prevent checkpoint ticket overruns.
308 *
309 * Further, use of static reservations through the log grant mechanism is
310 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
311 * grant) and a significant deadlock potential because regranting write space
312 * can block on log pushes. Hence if we have to regrant log space during a log
313 * push, we can deadlock.
314 *
315 * However, we can avoid this by use of a dynamic "reservation stealing"
316 * technique during transaction commit whereby unused reservation space in the
317 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
318 * space needed by the checkpoint transaction. This means that we never need to
319 * specifically reserve space for the CIL checkpoint transaction, nor do we
320 * need to regrant space once the checkpoint completes. This also means the
321 * checkpoint transaction ticket is specific to the checkpoint context, rather
322 * than the CIL itself.
323 *
Dave Chinner80168672010-09-24 18:13:44 +1000324 * With dynamic reservations, we can effectively make up arbitrary limits for
325 * the checkpoint size so long as they don't violate any other size rules.
326 * Recovery imposes a rule that no transaction exceed half the log, so we are
327 * limited by that. Furthermore, the log transaction reservation subsystem
328 * tries to keep 25% of the log free, so we need to keep below that limit or we
329 * risk running out of free log space to start any new transactions.
330 *
Dave Chinner108a4232020-03-24 20:10:26 -0700331 * In order to keep background CIL push efficient, we only need to ensure the
332 * CIL is large enough to maintain sufficient in-memory relogging to avoid
333 * repeated physical writes of frequently modified metadata. If we allow the CIL
334 * to grow to a substantial fraction of the log, then we may be pinning hundreds
335 * of megabytes of metadata in memory until the CIL flushes. This can cause
336 * issues when we are running low on memory - pinned memory cannot be reclaimed,
337 * and the CIL consumes a lot of memory. Hence we need to set an upper physical
338 * size limit for the CIL that limits the maximum amount of memory pinned by the
339 * CIL but does not limit performance by reducing relogging efficiency
340 * significantly.
341 *
342 * As such, the CIL push threshold ends up being the smaller of two thresholds:
343 * - a threshold large enough that it allows CIL to be pushed and progress to be
344 * made without excessive blocking of incoming transaction commits. This is
345 * defined to be 12.5% of the log space - half the 25% push threshold of the
346 * AIL.
347 * - small enough that it doesn't pin excessive amounts of memory but maintains
348 * close to peak relogging efficiency. This is defined to be 16x the iclog
349 * buffer window (32MB) as measurements have shown this to be roughly the
350 * point of diminishing performance increases under highly concurrent
351 * modification workloads.
Dave Chinner0e7ab7e2020-03-24 20:10:27 -0700352 *
353 * To prevent the CIL from overflowing upper commit size bounds, we introduce a
354 * new threshold at which we block committing transactions until the background
355 * CIL commit commences and switches to a new context. While this is not a hard
356 * limit, it forces the process committing a transaction to the CIL to block and
357 * yeild the CPU, giving the CIL push work a chance to be scheduled and start
358 * work. This prevents a process running lots of transactions from overfilling
359 * the CIL because it is not yielding the CPU. We set the blocking limit at
360 * twice the background push space threshold so we keep in line with the AIL
361 * push thresholds.
362 *
363 * Note: this is not a -hard- limit as blocking is applied after the transaction
364 * is inserted into the CIL and the push has been triggered. It is largely a
365 * throttling mechanism that allows the CIL push to be scheduled and run. A hard
366 * limit will be difficult to implement without introducing global serialisation
367 * in the CIL commit fast path, and it's not at all clear that we actually need
368 * such hard limits given the ~7 years we've run without a hard limit before
369 * finding the first situation where a checkpoint size overflow actually
370 * occurred. Hence the simple throttle, and an ASSERT check to tell us that
371 * we've overrun the max size.
Dave Chinnerdf806152010-05-17 15:52:13 +1000372 */
Dave Chinner108a4232020-03-24 20:10:26 -0700373#define XLOG_CIL_SPACE_LIMIT(log) \
374 min_t(int, (log)->l_logsize >> 3, BBTOB(XLOG_TOTAL_REC_SHIFT(log)) << 4)
Dave Chinnerdf806152010-05-17 15:52:13 +1000375
Dave Chinner0e7ab7e2020-03-24 20:10:27 -0700376#define XLOG_CIL_BLOCKING_SPACE_LIMIT(log) \
377 (XLOG_CIL_SPACE_LIMIT(log) * 2)
378
Dave Chinnerdf806152010-05-17 15:52:13 +1000379/*
Christoph Hellwig28496962012-02-20 02:31:25 +0000380 * ticket grant locks, queues and accounting have their own cachlines
381 * as these are quite hot and can be operated on concurrently.
382 */
383struct xlog_grant_head {
384 spinlock_t lock ____cacheline_aligned_in_smp;
385 struct list_head waiters;
386 atomic64_t grant;
387};
388
389/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * The reservation head lsn is not made up of a cycle number and block number.
391 * Instead, it uses a cycle number and byte number. Logs don't expect to
392 * overflow 31 bits worth of byte offset, so using a byte number will mean
393 * that round off problems won't occur when releasing partial reservations.
394 */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500395struct xlog {
David Chinner4679b2d2008-04-10 12:18:54 +1000396 /* The following fields don't need locking */
397 struct xfs_mount *l_mp; /* mount point */
David Chinnera9c21c12008-10-30 17:39:35 +1100398 struct xfs_ail *l_ailp; /* AIL log is working with */
Dave Chinner71e330b2010-05-21 14:37:18 +1000399 struct xfs_cil *l_cilp; /* CIL log is working with */
David Chinner4679b2d2008-04-10 12:18:54 +1000400 struct xfs_buftarg *l_targ; /* buftarg of log */
Christoph Hellwig1058d0f2019-06-28 19:27:25 -0700401 struct workqueue_struct *l_ioend_workqueue; /* for I/O completions */
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100402 struct delayed_work l_work; /* background flush work */
Dave Chinnere1d06e52021-08-10 17:59:02 -0700403 long l_opstate; /* operational state */
David Chinner4679b2d2008-04-10 12:18:54 +1000404 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000405 struct list_head *l_buf_cancel_table;
David Chinner4679b2d2008-04-10 12:18:54 +1000406 int l_iclog_hsize; /* size of iclog header */
407 int l_iclog_heads; /* # of iclog header sectors */
Alex Elder48389ef2010-04-20 17:10:21 +1000408 uint l_sectBBsize; /* sector size in BBs (2^n) */
David Chinner4679b2d2008-04-10 12:18:54 +1000409 int l_iclog_size; /* size of log in bytes */
David Chinner4679b2d2008-04-10 12:18:54 +1000410 int l_iclog_bufs; /* number of iclog buffers */
411 xfs_daddr_t l_logBBstart; /* start block of log */
412 int l_logsize; /* size of log in bytes */
413 int l_logBBsize; /* size of log in BB chunks */
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 /* The following block of fields are changed while holding icloglock */
Dave Chinnereb40a872010-12-21 12:09:01 +1100416 wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp;
Matthew Wilcoxd748c622008-05-19 16:34:27 +1000417 /* waiting for iclog flush */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 int l_covered_state;/* state of "covering disk
419 * log entries" */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 xlog_in_core_t *l_iclog; /* head log queue */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000421 spinlock_t l_icloglock; /* grab to change iclog state */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 int l_curr_cycle; /* Cycle number of log writes */
423 int l_prev_cycle; /* Cycle number before last
424 * block increment */
425 int l_curr_block; /* current logical log block */
426 int l_prev_block; /* previous logical log block */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Dave Chinner84f3c682010-12-03 22:11:29 +1100428 /*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100429 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
430 * read without needing to hold specific locks. To avoid operations
431 * contending with other hot objects, place each of them on a separate
432 * cacheline.
Dave Chinner84f3c682010-12-03 22:11:29 +1100433 */
434 /* lsn of last LR on disk */
435 atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100436 /* lsn of 1st LR with unflushed * buffers */
437 atomic64_t l_tail_lsn ____cacheline_aligned_in_smp;
Dave Chinner84f3c682010-12-03 22:11:29 +1100438
Christoph Hellwig28496962012-02-20 02:31:25 +0000439 struct xlog_grant_head l_reserve_head;
440 struct xlog_grant_head l_write_head;
Dave Chinner3f16b982010-12-21 12:29:01 +1100441
Brian Fosterbaff4e42014-07-15 08:07:29 +1000442 struct xfs_kobj l_kobj;
443
David Chinner4679b2d2008-04-10 12:18:54 +1000444 /* The following field are used for debugging; need to hold icloglock */
445#ifdef DEBUG
Christoph Hellwig5809d5e2015-06-22 09:44:47 +1000446 void *l_iclog_bak[XLOG_MAX_ICLOGS];
David Chinner4679b2d2008-04-10 12:18:54 +1000447#endif
Brian Foster12818d22016-09-26 08:22:16 +1000448 /* log recovery lsn tracking (for buffer submission */
449 xfs_lsn_t l_recovery_lsn;
Dave Chinnera6a65fe2021-06-18 08:21:48 -0700450
451 uint32_t l_iclog_roundoff;/* padding roundoff */
Darrick J. Wong2b73a2c2021-08-08 08:27:12 -0700452
453 /* Users of log incompat features should take a read lock. */
454 struct rw_semaphore l_incompat_users;
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500455};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000457#define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
Darrick J. Wongc8ce5402017-06-16 11:00:05 -0700458 ((log)->l_buf_cancel_table + ((uint64_t)blkno % XLOG_BC_TABLE_SIZE))
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000459
Dave Chinnere1d06e52021-08-10 17:59:02 -0700460/*
461 * Bits for operational state
462 */
463#define XLOG_ACTIVE_RECOVERY 0 /* in the middle of recovery */
464#define XLOG_RECOVERY_NEEDED 1 /* log was recovered */
465#define XLOG_IO_ERROR 2 /* log hit an I/O error, and being
466 shutdown */
467#define XLOG_TAIL_WARN 3 /* log tail verify warning issued */
468
469static inline bool
470xlog_recovery_needed(struct xlog *log)
471{
472 return test_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate);
473}
474
475static inline bool
476xlog_in_recovery(struct xlog *log)
477{
478 return test_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);
479}
480
Dave Chinner2039a272021-08-10 17:59:01 -0700481static inline bool
482xlog_is_shutdown(struct xlog *log)
483{
Dave Chinnere1d06e52021-08-10 17:59:02 -0700484 return test_bit(XLOG_IO_ERROR, &log->l_opstate);
Dave Chinner2039a272021-08-10 17:59:01 -0700485}
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487/* common routines */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500488extern int
489xlog_recover(
490 struct xlog *log);
491extern int
492xlog_recover_finish(
493 struct xlog *log);
Hariprasad Kelama7a92502019-07-03 07:34:18 -0700494extern void
Brian Fosterf0b2efa2015-08-19 09:58:36 +1000495xlog_recover_cancel(struct xlog *);
Christoph Hellwig0e446be2012-11-12 22:54:24 +1100496
Dave Chinnerf9668a02012-11-28 13:01:03 +1100497extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
Christoph Hellwig0e446be2012-11-12 22:54:24 +1100498 char *dp, int size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Darrick J. Wong182696f2021-10-12 11:09:23 -0700500extern struct kmem_cache *xfs_log_ticket_cache;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500501struct xlog_ticket *
502xlog_ticket_alloc(
503 struct xlog *log,
504 int unit_bytes,
505 int count,
506 char client,
Carlos Maiolinoca4f2582020-07-22 09:23:17 -0700507 bool permanent);
David Chinnereb01c9c2008-04-10 12:18:46 +1000508
Christoph Hellwige6b1f272010-03-23 11:47:38 +1100509static inline void
510xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
511{
512 *ptr += bytes;
513 *len -= bytes;
514 *off += bytes;
515}
516
Dave Chinner71e330b2010-05-21 14:37:18 +1000517void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
Brian Fosterd4ca1d52017-06-14 21:29:50 -0700518void xlog_print_trans(struct xfs_trans *);
Dave Chinnerc45aba42021-08-10 18:00:42 -0700519int xlog_write(struct xlog *log, struct xfs_cil_ctx *ctx,
520 struct xfs_log_vec *log_vector, struct xlog_ticket *tic,
Dave Chinnercaa80092021-08-10 18:00:43 -0700521 uint optype);
Christoph Hellwig8b41e3f2020-03-25 18:18:23 -0700522void xfs_log_ticket_ungrant(struct xlog *log, struct xlog_ticket *ticket);
523void xfs_log_ticket_regrant(struct xlog *log, struct xlog_ticket *ticket);
Dave Chinner71e330b2010-05-21 14:37:18 +1000524
Dave Chinner0020a192021-08-10 18:00:44 -0700525void xlog_state_switch_iclogs(struct xlog *log, struct xlog_in_core *iclog,
526 int eventual_size);
Dave Chinner0dc8f7f2021-07-27 16:23:48 -0700527int xlog_state_release_iclog(struct xlog *log, struct xlog_in_core *iclog,
528 xfs_lsn_t log_tail_lsn);
Dave Chinnereef983f2021-06-18 08:21:51 -0700529
Dave Chinner71e330b2010-05-21 14:37:18 +1000530/*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100531 * When we crack an atomic LSN, we sample it first so that the value will not
532 * change while we are cracking it into the component values. This means we
533 * will always get consistent component values to work from. This should always
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300534 * be used to sample and crack LSNs that are stored and updated in atomic
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100535 * variables.
536 */
537static inline void
538xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
539{
540 xfs_lsn_t val = atomic64_read(lsn);
541
542 *cycle = CYCLE_LSN(val);
543 *block = BLOCK_LSN(val);
544}
545
546/*
547 * Calculate and assign a value to an atomic LSN variable from component pieces.
548 */
549static inline void
550xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
551{
552 atomic64_set(lsn, xlog_assign_lsn(cycle, block));
553}
554
555/*
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100556 * When we crack the grant head, we sample it first so that the value will not
Dave Chinnera69ed032010-12-21 12:08:20 +1100557 * change while we are cracking it into the component values. This means we
558 * will always get consistent component values to work from.
559 */
560static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100561xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100562{
Dave Chinnera69ed032010-12-21 12:08:20 +1100563 *cycle = val >> 32;
564 *space = val & 0xffffffff;
565}
566
567static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100568xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
569{
570 xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
571}
572
573static inline int64_t
574xlog_assign_grant_head_val(int cycle, int space)
575{
576 return ((int64_t)cycle << 32) | space;
577}
578
579static inline void
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100580xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100581{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100582 atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
Dave Chinnera69ed032010-12-21 12:08:20 +1100583}
584
585/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000586 * Committed Item List interfaces
587 */
Dave Chinner2c6e24c2013-10-15 09:17:49 +1100588int xlog_cil_init(struct xlog *log);
589void xlog_cil_init_post_recovery(struct xlog *log);
590void xlog_cil_destroy(struct xlog *log);
591bool xlog_cil_empty(struct xlog *log);
Dave Chinner5f9b4b02021-06-18 08:21:52 -0700592void xlog_cil_commit(struct xlog *log, struct xfs_trans *tp,
593 xfs_csn_t *commit_seq, bool regrant);
Dave Chinnerc45aba42021-08-10 18:00:42 -0700594void xlog_cil_set_ctx_write_state(struct xfs_cil_ctx *ctx,
595 struct xlog_in_core *iclog);
596
Dave Chinner71e330b2010-05-21 14:37:18 +1000597
Dave Chinnera44f13e2010-08-24 11:40:03 +1000598/*
599 * CIL force routines
600 */
Dave Chinner0020a192021-08-10 18:00:44 -0700601void xlog_cil_flush(struct xlog *log);
Dave Chinner5f9b4b02021-06-18 08:21:52 -0700602xfs_lsn_t xlog_cil_force_seq(struct xlog *log, xfs_csn_t sequence);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000603
604static inline void
Mark Tinguelyad223e62012-06-14 09:22:15 -0500605xlog_cil_force(struct xlog *log)
Dave Chinnera44f13e2010-08-24 11:40:03 +1000606{
Dave Chinner5f9b4b02021-06-18 08:21:52 -0700607 xlog_cil_force_seq(log, log->l_cilp->xc_current_sequence);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000608}
Dave Chinner71e330b2010-05-21 14:37:18 +1000609
Tim Shimmin955e47a2006-09-28 11:04:16 +1000610/*
Dave Chinnereb40a872010-12-21 12:09:01 +1100611 * Wrapper function for waiting on a wait queue serialised against wakeups
612 * by a spinlock. This matches the semantics of all the wait queues used in the
613 * log code.
614 */
Darrick J. Wongf7559792019-11-06 08:41:20 -0800615static inline void
616xlog_wait(
617 struct wait_queue_head *wq,
618 struct spinlock *lock)
619 __releases(lock)
Dave Chinnereb40a872010-12-21 12:09:01 +1100620{
621 DECLARE_WAITQUEUE(wait, current);
622
623 add_wait_queue_exclusive(wq, &wait);
624 __set_current_state(TASK_UNINTERRUPTIBLE);
625 spin_unlock(lock);
626 schedule();
627 remove_wait_queue(wq, &wait);
628}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Dave Chinnera79b28c2021-06-18 08:21:48 -0700630int xlog_wait_on_iclog(struct xlog_in_core *iclog);
631
Brian Fostera45086e2015-10-12 15:59:25 +1100632/*
633 * The LSN is valid so long as it is behind the current LSN. If it isn't, this
634 * means that the next log record that includes this metadata could have a
635 * smaller LSN. In turn, this means that the modification in the log would not
636 * replay.
637 */
638static inline bool
639xlog_valid_lsn(
640 struct xlog *log,
641 xfs_lsn_t lsn)
642{
643 int cur_cycle;
644 int cur_block;
645 bool valid = true;
646
647 /*
648 * First, sample the current lsn without locking to avoid added
649 * contention from metadata I/O. The current cycle and block are updated
650 * (in xlog_state_switch_iclogs()) and read here in a particular order
651 * to avoid false negatives (e.g., thinking the metadata LSN is valid
652 * when it is not).
653 *
654 * The current block is always rewound before the cycle is bumped in
655 * xlog_state_switch_iclogs() to ensure the current LSN is never seen in
656 * a transiently forward state. Instead, we can see the LSN in a
657 * transiently behind state if we happen to race with a cycle wrap.
658 */
Mark Rutland6aa7de02017-10-23 14:07:29 -0700659 cur_cycle = READ_ONCE(log->l_curr_cycle);
Brian Fostera45086e2015-10-12 15:59:25 +1100660 smp_rmb();
Mark Rutland6aa7de02017-10-23 14:07:29 -0700661 cur_block = READ_ONCE(log->l_curr_block);
Brian Fostera45086e2015-10-12 15:59:25 +1100662
663 if ((CYCLE_LSN(lsn) > cur_cycle) ||
664 (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) {
665 /*
666 * If the metadata LSN appears invalid, it's possible the check
667 * above raced with a wrap to the next log cycle. Grab the lock
668 * to check for sure.
669 */
670 spin_lock(&log->l_icloglock);
671 cur_cycle = log->l_curr_cycle;
672 cur_block = log->l_curr_block;
673 spin_unlock(&log->l_icloglock);
674
675 if ((CYCLE_LSN(lsn) > cur_cycle) ||
676 (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block))
677 valid = false;
678 }
679
680 return valid;
681}
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683#endif /* __XFS_LOG_PRIV_H__ */