4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2012, 2013, Intel Corporation.
25 * Use is subject to license terms.
27 * Author: Johann Lombardi <johann.lombardi@intel.com>
28 * Author: Niu Yawei <yawei.niu@intel.com>
32 # define EXPORT_SYMTAB
35 #define DEBUG_SUBSYSTEM S_LQUOTA
37 #include "qsd_internal.h"
39 extern cfs_mem_cache_t *upd_kmem;
42 * Allocate and fill an qsd_upd_rec structure to be processed by the writeback
45 * \param qqi - is the qsd_qtype_info structure relevant to the update
46 * \param lqe - is the lquota entry subject to the update
47 * \param qid - is the identifier subject to the update
48 * \param rec - is the record storing the new quota settings
49 * \param ver - is the version associated with the update
50 * \param global - is a boolean set to true if this is an update of the global
51 * index and false for a slave index.
53 static struct qsd_upd_rec *qsd_upd_alloc(struct qsd_qtype_info *qqi,
54 struct lquota_entry *lqe,
56 union lquota_rec *rec, __u64 ver,
59 struct qsd_upd_rec *upd;
61 OBD_SLAB_ALLOC_PTR_GFP(upd, upd_kmem, CFS_ALLOC_IO);
63 CERROR("Failed to allocate upd");
68 CFS_INIT_LIST_HEAD(&upd->qur_link);
76 upd->qur_global = global;
81 static void qsd_upd_free(struct qsd_upd_rec *upd)
84 lqe_putref(upd->qur_lqe);
85 OBD_SLAB_FREE_PTR(upd, upd_kmem);
88 /* must hold the qsd_lock */
89 static void qsd_upd_add(struct qsd_instance *qsd, struct qsd_upd_rec *upd)
91 if (!qsd->qsd_stopping) {
92 list_add_tail(&upd->qur_link, &qsd->qsd_upd_list);
93 /* wake up the upd thread */
94 cfs_waitq_signal(&qsd->qsd_upd_thread.t_ctl_waitq);
96 CWARN("%s: discard deferred update.\n", qsd->qsd_svname);
98 LQUOTA_WARN(upd->qur_lqe, "discard deferred update.");
103 /* must hold the qsd_lock */
104 static void qsd_add_deferred(cfs_list_t *list, struct qsd_upd_rec *upd)
106 struct qsd_upd_rec *tmp, *n;
108 /* Sort the updates in ascending order */
109 cfs_list_for_each_entry_safe_reverse(tmp, n, list, qur_link) {
111 /* There could be some legacy records which have duplicated
112 * version. Imagine following scenario: slave received global
113 * glimpse and queued a record in the deferred list, then
114 * master crash and rollback to an ealier version, then the
115 * version of queued record will be conflicting with later
116 * updates. We should just delete the legacy record in such
118 if (upd->qur_ver == tmp->qur_ver) {
119 LASSERT(tmp->qur_lqe);
120 LQUOTA_ERROR(tmp->qur_lqe, "Found a conflict record "
121 "with ver:"LPU64"", tmp->qur_ver);
122 cfs_list_del_init(&tmp->qur_link);
126 if (upd->qur_ver < tmp->qur_ver) {
129 cfs_list_add_tail(&upd->qur_link, &tmp->qur_link);
133 cfs_list_add(&upd->qur_link, list);
136 /* must hold the qsd_lock */
137 static void qsd_kickoff_deferred(struct qsd_qtype_info *qqi, cfs_list_t *list,
140 struct qsd_upd_rec *upd, *tmp;
143 /* Get the first update record in the list, which has the smallest
144 * version, discard all records with versions smaller than the current
146 cfs_list_for_each_entry_safe(upd, tmp, list, qur_link) {
147 if (upd->qur_ver <= ver) {
148 /* drop this update */
149 cfs_list_del_init(&upd->qur_link);
150 CDEBUG(D_QUOTA, "%s: skipping deferred update ver:"
151 LPU64"/"LPU64", global:%d, qid:"LPU64"\n",
152 qqi->qqi_qsd->qsd_svname, upd->qur_ver, ver,
153 upd->qur_global, upd->qur_qid.qid_uid);
160 /* No remaining deferred update */
161 if (cfs_list_empty(list))
164 CDEBUG(D_QUOTA, "%s: found deferred update record. "
165 "version:"LPU64"/"LPU64", global:%d, qid:"LPU64"\n",
166 qqi->qqi_qsd->qsd_svname, upd->qur_ver, ver,
167 upd->qur_global, upd->qur_qid.qid_uid);
169 LASSERTF(upd->qur_ver > ver, "lur_ver:"LPU64", cur_ver:"LPU64"\n",
172 /* Kick off the deferred udpate */
173 if (upd->qur_ver == ver + 1) {
174 list_del_init(&upd->qur_link);
175 qsd_upd_add(qqi->qqi_qsd, upd);
180 /* Bump version of global or slave index copy
182 * \param qqi - qsd_qtype_info
183 * \param ver - version to be bumped to
184 * \param global - global or slave index copy?
186 void qsd_bump_version(struct qsd_qtype_info *qqi, __u64 ver, bool global)
191 idx_ver = global ? &qqi->qqi_glb_ver : &qqi->qqi_slv_ver;
192 list = global ? &qqi->qqi_deferred_glb : &qqi->qqi_deferred_slv;
194 write_lock(&qqi->qqi_qsd->qsd_lock);
197 qqi->qqi_glb_uptodate = 1;
199 qqi->qqi_slv_uptodate = 1;
200 qsd_kickoff_deferred(qqi, list, ver);
201 write_unlock(&qqi->qqi_qsd->qsd_lock);
205 * Schedule a commit of a lquota entry
207 * \param qqi - qsd_qtype_info
208 * \param lqe - lquota_entry
209 * \param qid - quota id
210 * \param rec - global or slave record to be updated to disk
211 * \param ver - new index file version
212 * \param global- true: master record; false: slave record
214 void qsd_upd_schedule(struct qsd_qtype_info *qqi, struct lquota_entry *lqe,
215 union lquota_id *qid, union lquota_rec *rec, __u64 ver,
218 struct qsd_upd_rec *upd;
219 struct qsd_instance *qsd = qqi->qqi_qsd;
223 CDEBUG(D_QUOTA, "%s: schedule update. global:%s, version:"LPU64"\n",
224 qsd->qsd_svname, global ? "true" : "false", ver);
226 upd = qsd_upd_alloc(qqi, lqe, qid, rec, ver, global);
230 /* If we don't want update index version, no need to sort the
231 * records in version order, just schedule the updates instantly. */
233 write_lock(&qsd->qsd_lock);
234 qsd_upd_add(qsd, upd);
235 write_unlock(&qsd->qsd_lock);
239 write_lock(&qsd->qsd_lock);
241 cur_ver = global ? qqi->qqi_glb_ver : qqi->qqi_slv_ver;
243 if (ver <= cur_ver) {
245 /* legitimate race between glimpse AST and
247 CDEBUG(D_QUOTA, "%s: discarding glb update from glimpse"
248 " ver:"LPU64" local ver:"LPU64"\n",
249 qsd->qsd_svname, ver, cur_ver);
251 CERROR("%s: discard slv update, ver:"LPU64" local ver:"
252 LPU64"\n", qsd->qsd_svname, ver, cur_ver);
254 } else if ((ver == cur_ver + 1) && qqi->qqi_glb_uptodate &&
255 qqi->qqi_slv_uptodate) {
256 /* In order update, and reintegration has been done. */
257 qsd_upd_add(qsd, upd);
259 /* Out of order update (the one with smaller version hasn't
260 * reached slave or hasn't been flushed to disk yet), or
261 * the reintegration is in progress. Defer the update. */
262 cfs_list_t *list = global ? &qqi->qqi_deferred_glb :
263 &qqi->qqi_deferred_slv;
264 qsd_add_deferred(list, upd);
267 write_unlock(&qsd->qsd_lock);
272 static int qsd_process_upd(const struct lu_env *env, struct qsd_upd_rec *upd)
274 struct lquota_entry *lqe = upd->qur_lqe;
275 struct qsd_qtype_info *qqi = upd->qur_qqi;
280 lqe = lqe_locate(env, qqi->qqi_site, &upd->qur_qid);
282 GOTO(out, rc = PTR_ERR(lqe));
285 /* The in-memory lqe update for slave index copy isn't deferred,
286 * we shouldn't touch it here. */
287 if (upd->qur_global) {
288 rc = qsd_update_lqe(env, lqe, upd->qur_global, &upd->qur_rec);
292 qsd_refresh_usage(env, lqe);
293 /* Report usage asynchronously */
294 rc = qsd_adjust(env, lqe);
296 LQUOTA_ERROR(lqe, "failed to report usage, rc:%d", rc);
299 rc = qsd_update_index(env, qqi, &upd->qur_qid, upd->qur_global,
300 upd->qur_ver, &upd->qur_rec);
302 if (lqe && !IS_ERR(lqe)) {
309 void qsd_adjust_schedule(struct lquota_entry *lqe, bool defer, bool cancel)
311 struct qsd_instance *qsd = lqe2qqi(lqe)->qqi_qsd;
315 spin_lock(&qsd->qsd_adjust_lock);
317 /* the lqe is being queued for the per-ID lock cancel, we should
318 * cancel the lock cancel and re-add it for quota adjust */
319 if (!cfs_list_empty(&lqe->lqe_link) &&
320 lqe->lqe_adjust_time == 0) {
321 cfs_list_del_init(&lqe->lqe_link);
325 if (cfs_list_empty(&lqe->lqe_link)) {
327 lqe->lqe_adjust_time = 0;
329 lqe->lqe_adjust_time = defer ?
330 cfs_time_shift_64(QSD_WB_INTERVAL) :
331 cfs_time_current_64();
332 /* lqe reference transfered to list */
334 cfs_list_add_tail(&lqe->lqe_link,
335 &qsd->qsd_adjust_list);
337 cfs_list_add(&lqe->lqe_link, &qsd->qsd_adjust_list);
340 spin_unlock(&qsd->qsd_adjust_lock);
343 cfs_waitq_signal(&qsd->qsd_upd_thread.t_ctl_waitq);
348 /* return true if there is pending writeback records or the pending
350 static bool qsd_job_pending(struct qsd_instance *qsd, cfs_list_t *upd,
353 bool job_pending = false;
356 LASSERT(cfs_list_empty(upd));
359 spin_lock(&qsd->qsd_adjust_lock);
360 if (!cfs_list_empty(&qsd->qsd_adjust_list)) {
361 struct lquota_entry *lqe;
362 lqe = cfs_list_entry(qsd->qsd_adjust_list.next,
363 struct lquota_entry, lqe_link);
364 if (cfs_time_beforeq_64(lqe->lqe_adjust_time,
365 cfs_time_current_64()))
368 spin_unlock(&qsd->qsd_adjust_lock);
370 write_lock(&qsd->qsd_lock);
371 if (!cfs_list_empty(&qsd->qsd_upd_list)) {
372 cfs_list_splice_init(&qsd->qsd_upd_list, upd);
376 if (qsd->qsd_acct_failed) {
377 /* don't bother kicking off reintegration if space accounting
378 * failed to be enabled */
379 write_unlock(&qsd->qsd_lock);
383 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
384 struct qsd_qtype_info *qqi = qsd->qsd_type_array[qtype];
386 if (!qsd_type_enabled(qsd, qtype))
389 if ((!qqi->qqi_glb_uptodate || !qqi->qqi_slv_uptodate) &&
391 /* global or slave index not up to date and reint
392 * thread not running */
396 write_unlock(&qsd->qsd_lock);
400 static int qsd_upd_thread(void *arg)
402 struct qsd_instance *qsd = (struct qsd_instance *)arg;
403 struct ptlrpc_thread *thread = &qsd->qsd_upd_thread;
404 struct l_wait_info lwi;
406 struct qsd_upd_rec *upd, *n;
407 char pname[MTI_NAME_MAXLEN];
411 struct lquota_entry *lqe, *tmp;
419 rc = lu_env_init(env, LCT_DT_THREAD);
421 CERROR("%s: Fail to init env.", qsd->qsd_svname);
426 snprintf(pname, MTI_NAME_MAXLEN, "lquota_wb_%s", qsd->qsd_svname);
427 cfs_daemonize(pname);
429 thread_set_flags(thread, SVC_RUNNING);
430 cfs_waitq_signal(&thread->t_ctl_waitq);
432 CFS_INIT_LIST_HEAD(&queue);
433 lwi = LWI_TIMEOUT(cfs_time_seconds(QSD_WB_INTERVAL), NULL, NULL);
435 l_wait_event(thread->t_ctl_waitq,
436 qsd_job_pending(qsd, &queue, &uptodate) ||
437 !thread_is_running(thread), &lwi);
439 cfs_list_for_each_entry_safe(upd, n, &queue, qur_link) {
440 cfs_list_del_init(&upd->qur_link);
441 qsd_process_upd(env, upd);
445 spin_lock(&qsd->qsd_adjust_lock);
446 cur_time = cfs_time_current_64();
447 cfs_list_for_each_entry_safe(lqe, tmp, &qsd->qsd_adjust_list,
449 /* deferred items are sorted by time */
450 if (!cfs_time_beforeq_64(lqe->lqe_adjust_time,
454 cfs_list_del_init(&lqe->lqe_link);
455 spin_unlock(&qsd->qsd_adjust_lock);
457 if (thread_is_running(thread) && uptodate) {
458 qsd_refresh_usage(env, lqe);
459 if (lqe->lqe_adjust_time == 0)
460 qsd_id_lock_cancel(env, lqe);
462 qsd_adjust(env, lqe);
466 spin_lock(&qsd->qsd_adjust_lock);
468 spin_unlock(&qsd->qsd_adjust_lock);
470 if (!thread_is_running(thread))
476 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
477 qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
481 thread_set_flags(thread, SVC_STOPPED);
482 cfs_waitq_signal(&thread->t_ctl_waitq);
486 int qsd_start_upd_thread(struct qsd_instance *qsd)
488 struct ptlrpc_thread *thread = &qsd->qsd_upd_thread;
489 struct l_wait_info lwi = { 0 };
493 rc = cfs_create_thread(qsd_upd_thread, (void *)qsd, 0);
495 CERROR("Fail to start quota update thread. rc: %d\n", rc);
496 thread_set_flags(thread, SVC_STOPPED);
500 l_wait_event(thread->t_ctl_waitq,
501 thread_is_running(thread) || thread_is_stopped(thread),
506 static void qsd_cleanup_deferred(struct qsd_instance *qsd)
510 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
511 struct qsd_upd_rec *upd, *tmp;
512 struct qsd_qtype_info *qqi = qsd->qsd_type_array[qtype];
517 write_lock(&qsd->qsd_lock);
518 cfs_list_for_each_entry_safe(upd, tmp, &qqi->qqi_deferred_glb,
520 CWARN("%s: Free global deferred upd: ID:"LPU64", "
521 "ver:"LPU64"/"LPU64"\n", qsd->qsd_svname,
522 upd->qur_qid.qid_uid, upd->qur_ver,
524 list_del_init(&upd->qur_link);
527 cfs_list_for_each_entry_safe(upd, tmp, &qqi->qqi_deferred_slv,
529 CWARN("%s: Free slave deferred upd: ID:"LPU64", "
530 "ver:"LPU64"/"LPU64"\n", qsd->qsd_svname,
531 upd->qur_qid.qid_uid, upd->qur_ver,
533 list_del_init(&upd->qur_link);
536 write_unlock(&qsd->qsd_lock);
540 static void qsd_cleanup_adjust(struct qsd_instance *qsd)
542 struct lquota_entry *lqe;
544 spin_lock(&qsd->qsd_adjust_lock);
545 while (!cfs_list_empty(&qsd->qsd_adjust_list)) {
546 lqe = cfs_list_entry(qsd->qsd_adjust_list.next,
547 struct lquota_entry, lqe_link);
548 cfs_list_del_init(&lqe->lqe_link);
551 spin_unlock(&qsd->qsd_adjust_lock);
554 void qsd_stop_upd_thread(struct qsd_instance *qsd)
556 struct ptlrpc_thread *thread = &qsd->qsd_upd_thread;
557 struct l_wait_info lwi = { 0 };
559 if (!thread_is_stopped(thread)) {
560 thread_set_flags(thread, SVC_STOPPING);
561 cfs_waitq_signal(&thread->t_ctl_waitq);
563 l_wait_event(thread->t_ctl_waitq, thread_is_stopped(thread),
566 qsd_cleanup_deferred(qsd);
567 qsd_cleanup_adjust(qsd);