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, 2014, 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>
31 #define DEBUG_SUBSYSTEM S_LQUOTA
33 #include "qsd_internal.h"
36 * Completion function invoked when the global quota lock enqueue has completed
38 static void qsd_reint_completion(const struct lu_env *env,
39 struct qsd_qtype_info *qqi,
40 struct quota_body *req_qbody,
41 struct quota_body *rep_qbody,
42 struct lustre_handle *lockh,
43 struct lquota_lvb *lvb,
46 struct qsd_instance *qsd = qqi->qqi_qsd;
47 __u64 *slv_ver = (__u64 *)arg;
51 CDEBUG_LIMIT(rc != -EAGAIN ? D_ERROR : D_QUOTA,
52 "%s: failed to enqueue global quota lock, glb fid:"
53 DFID", rc:%d\n", qsd->qsd_svname,
54 PFID(&req_qbody->qb_fid), rc);
58 CDEBUG(D_QUOTA, "%s: global quota lock successfully acquired, glb "
59 "fid:"DFID", glb ver:"LPU64", slv fid:"DFID", slv ver:"LPU64"\n",
60 qsd->qsd_svname, PFID(&req_qbody->qb_fid),
61 lvb->lvb_glb_ver, PFID(&rep_qbody->qb_slv_fid),
62 rep_qbody->qb_slv_ver);
64 *slv_ver = rep_qbody->qb_slv_ver;
65 memcpy(&qqi->qqi_slv_fid, &rep_qbody->qb_slv_fid,
66 sizeof(struct lu_fid));
67 lustre_handle_copy(&qqi->qqi_lockh, lockh);
71 static int qsd_reint_qid(const struct lu_env *env, struct qsd_qtype_info *qqi,
72 bool global, union lquota_id *qid, void *rec)
74 struct lquota_entry *lqe;
78 lqe = lqe_locate(env, qqi->qqi_site, qid);
82 LQUOTA_DEBUG(lqe, "reintegrating entry");
84 rc = qsd_update_lqe(env, lqe, global, rec);
88 rc = qsd_update_index(env, qqi, qid, global, 0, rec);
94 static int qsd_reint_entries(const struct lu_env *env,
95 struct qsd_qtype_info *qqi,
96 struct idx_info *ii, bool global,
98 unsigned int npages, bool need_swab)
100 struct qsd_thread_info *qti = qsd_info(env);
101 struct qsd_instance *qsd = qqi->qqi_qsd;
102 union lquota_id *qid = &qti->qti_id;
107 CDEBUG(D_QUOTA, "%s: processing %d pages for %s index\n",
108 qsd->qsd_svname, npages, global ? "global" : "slave");
110 /* sanity check on the record size */
111 if ((global && ii->ii_recsize != sizeof(struct lquota_glb_rec)) ||
112 (!global && ii->ii_recsize != sizeof(struct lquota_slv_rec))) {
113 CERROR("%s: invalid record size (%d) for %s index\n",
114 qsd->qsd_svname, ii->ii_recsize,
115 global ? "global" : "slave");
119 size = ii->ii_recsize + ii->ii_keysize;
121 for (i = 0; i < npages; i++) {
122 union lu_page *lip = kmap(pages[i]);
124 for (j = 0; j < LU_PAGE_COUNT; j++) {
127 lustre_swab_lip_header(&lip->lp_idx);
129 if (lip->lp_idx.lip_magic != LIP_MAGIC) {
130 CERROR("%s: invalid magic (%x != %x) for page "
131 "%d/%d while transferring %s index\n",
132 qsd->qsd_svname, lip->lp_idx.lip_magic,
133 LIP_MAGIC, i + 1, npages,
134 global ? "global" : "slave");
135 GOTO(out, rc = -EINVAL);
138 CDEBUG(D_QUOTA, "%s: processing page %d/%d with %d "
139 "entries for %s index\n", qsd->qsd_svname, i + 1,
140 npages, lip->lp_idx.lip_nr,
141 global ? "global" : "slave");
143 for (k = 0; k < lip->lp_idx.lip_nr; k++) {
146 entry = lip->lp_idx.lip_entries + k * size;
147 memcpy(qid, entry, ii->ii_keysize); /* key */
148 entry += ii->ii_keysize; /* value */
154 __swab64s(&qid->qid_uid);
155 /* quota records only include 64-bit
157 while (offset < ii->ii_recsize) {
160 offset += sizeof(__u64);
164 rc = qsd_reint_qid(env, qqi, global, qid,
179 static int qsd_reint_index(const struct lu_env *env, struct qsd_qtype_info *qqi,
182 struct qsd_thread_info *qti = qsd_info(env);
183 struct qsd_instance *qsd = qqi->qqi_qsd;
184 struct idx_info *ii = &qti->qti_ii;
186 struct page **pages = NULL;
187 unsigned int npages, pg_cnt;
188 __u64 start_hash = 0, ver = 0;
189 bool need_swab = false;
193 fid = global ? &qqi->qqi_fid : &qqi->qqi_slv_fid;
195 /* let's do a 1MB bulk */
196 npages = min_t(unsigned int, OFD_MAX_BRW_SIZE, 1 << 20);
197 npages /= PAGE_CACHE_SIZE;
199 /* allocate pages for bulk index read */
200 OBD_ALLOC(pages, npages * sizeof(*pages));
202 GOTO(out, rc = -ENOMEM);
203 for (i = 0; i < npages; i++) {
204 pages[i] = alloc_page(GFP_IOFS);
205 if (pages[i] == NULL)
206 GOTO(out, rc = -ENOMEM);
210 /* initialize index_info request with FID of global index */
211 memset(ii, 0, sizeof(*ii));
212 memcpy(&ii->ii_fid, fid, sizeof(*fid));
213 ii->ii_magic = IDX_INFO_MAGIC;
214 ii->ii_flags = II_FL_NOHASH;
215 ii->ii_count = npages * LU_PAGE_COUNT;
216 ii->ii_hash_start = start_hash;
218 /* send bulk request to quota master to read global index */
219 rc = qsd_fetch_index(env, qsd->qsd_exp, ii, npages, pages, &need_swab);
221 CWARN("%s: failed to fetch index for "DFID". %d\n",
222 qsd->qsd_svname, PFID(fid), rc);
226 /* various sanity checks */
227 if (ii->ii_magic != IDX_INFO_MAGIC) {
228 CERROR("%s: invalid magic in index transfer %x != %x\n",
229 qsd->qsd_svname, ii->ii_magic, IDX_INFO_MAGIC);
230 GOTO(out, rc = -EPROTO);
232 if ((ii->ii_flags & II_FL_VARKEY) != 0)
233 CWARN("%s: II_FL_VARKEY is set on index transfer for fid "DFID
234 ", it shouldn't be\n", qsd->qsd_svname, PFID(fid));
235 if ((ii->ii_flags & II_FL_NONUNQ) != 0)
236 CWARN("%s: II_FL_NONUNQ is set on index transfer for fid "DFID
237 ", it shouldn't be\n", qsd->qsd_svname, PFID(fid));
238 if (ii->ii_keysize != sizeof(__u64)) {
239 CERROR("%s: invalid key size reported on index transfer for "
240 "fid "DFID", %u != %u\n", qsd->qsd_svname, PFID(fid),
241 ii->ii_keysize, (int)sizeof(__u64));
242 GOTO(out, rc = -EPROTO);
244 if (ii->ii_version == 0 && ii->ii_count != 0)
245 CWARN("%s: index version for fid "DFID" is 0, but index isn't "
246 "empty (%d)\n", qsd->qsd_svname, PFID(fid), ii->ii_count);
248 CDEBUG(D_QUOTA, "%s: reintegration process for fid "DFID" successfully "
249 "fetched %s index, count = %d\n", qsd->qsd_svname,
250 PFID(fid), global ? "global" : "slave", ii->ii_count);
253 /* record version associated with the first bulk transfer */
254 ver = ii->ii_version;
256 pg_cnt = (ii->ii_count + (LU_PAGE_COUNT) - 1);
257 pg_cnt >>= PAGE_CACHE_SHIFT - LU_PAGE_SHIFT;
259 if (pg_cnt > npages) {
260 CERROR("%s: master returned more pages than expected, %u > %u"
261 "\n", qsd->qsd_svname, pg_cnt, npages);
265 rc = qsd_reint_entries(env, qqi, ii, global, pages, pg_cnt, need_swab);
269 if (ii->ii_hash_end != II_END_OFF) {
270 start_hash = ii->ii_hash_end;
275 for (i = 0; i < npages; i++)
276 if (pages[i] != NULL)
277 __free_page(pages[i]);
278 OBD_FREE(pages, npages * sizeof(*pages));
281 /* Update index version */
283 rc = qsd_write_version(env, qqi, ver, global);
285 CERROR("%s: write version "LPU64" to "DFID" failed. "
286 "%d\n", qsd->qsd_svname, ver, PFID(fid), rc);
292 static int qsd_reconciliation(const struct lu_env *env,
293 struct qsd_qtype_info *qqi)
295 struct qsd_thread_info *qti = qsd_info(env);
296 struct qsd_instance *qsd = qqi->qqi_qsd;
297 const struct dt_it_ops *iops;
300 struct lquota_entry *lqe;
301 union lquota_id *qid = &qti->qti_id;
305 LASSERT(qqi->qqi_glb_obj != NULL);
306 iops = &qqi->qqi_glb_obj->do_index_ops->dio_it;
308 it = iops->init(env, qqi->qqi_glb_obj, 0);
310 CWARN("%s: Initialize it for "DFID" failed. %ld\n",
311 qsd->qsd_svname, PFID(&qqi->qqi_fid), PTR_ERR(it));
315 rc = iops->load(env, it, 0);
317 CWARN("%s: Load first entry for "DFID" failed. %d\n",
318 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
320 } else if (rc == 0) {
321 rc = iops->next(env, it);
323 GOTO(out, rc = (rc < 0) ? rc : 0);
327 key = iops->key(env, it);
329 CWARN("%s: Error key for "DFID". %ld\n",
330 qsd->qsd_svname, PFID(&qqi->qqi_fid),
332 GOTO(out, rc = PTR_ERR(key));
335 /* skip the root user/group */
336 if (*((__u64 *)key) == 0)
339 qid->qid_uid = *((__u64 *)key);
341 lqe = lqe_locate(env, qqi->qqi_site, qid);
343 CWARN("%s: failed to locate lqe. "DFID", %ld\n",
344 qsd->qsd_svname, PFID(&qqi->qqi_fid),
346 GOTO(out, rc = PTR_ERR(lqe));
349 rc = qsd_refresh_usage(env, lqe);
351 CWARN("%s: failed to get usage. "DFID", %d\n",
352 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
357 rc = qsd_adjust(env, lqe);
360 CWARN("%s: failed to report quota. "DFID", %d\n",
361 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
365 rc = iops->next(env, it);
367 CWARN("%s: failed to parse index, ->next error:%d "DFID
368 "\n", qsd->qsd_svname, rc, PFID(&qqi->qqi_fid));
380 static int qsd_connected(struct qsd_instance *qsd)
384 read_lock(&qsd->qsd_lock);
385 connected = qsd->qsd_exp_valid ? 1 : 0;
386 read_unlock(&qsd->qsd_lock);
391 static int qsd_started(struct qsd_instance *qsd)
395 read_lock(&qsd->qsd_lock);
396 started = qsd->qsd_started ? 1 : 0;
397 read_unlock(&qsd->qsd_lock);
403 * Routine executed by the reintegration thread.
405 static int qsd_reint_main(void *args)
408 struct qsd_thread_info *qti;
409 struct qsd_qtype_info *qqi = (struct qsd_qtype_info *)args;
410 struct qsd_instance *qsd = qqi->qqi_qsd;
411 struct ptlrpc_thread *thread = &qqi->qqi_reint_thread;
412 struct l_wait_info lwi = { 0 };
416 CDEBUG(D_QUOTA, "%s: Starting reintegration thread for "DFID"\n",
417 qsd->qsd_svname, PFID(&qqi->qqi_fid));
420 lu_ref_add(&qqi->qqi_reference, "reint_thread", thread);
422 thread_set_flags(thread, SVC_RUNNING);
423 wake_up(&thread->t_ctl_waitq);
427 GOTO(out, rc = -ENOMEM);
429 /* initialize environment */
430 rc = lu_env_init(env, LCT_DT_THREAD);
435 /* wait for the connection to master established */
436 l_wait_event(thread->t_ctl_waitq,
437 qsd_connected(qsd) || !thread_is_running(thread), &lwi);
439 /* Step 1: enqueue global index lock */
440 if (!thread_is_running(thread))
441 GOTO(out_env_init, rc = 0);
443 LASSERT(qsd->qsd_exp != NULL);
444 LASSERT(qqi->qqi_glb_uptodate == 0 || qqi->qqi_slv_uptodate == 0);
446 memset(&qti->qti_lvb, 0, sizeof(qti->qti_lvb));
448 read_lock(&qsd->qsd_lock);
449 /* check whether we already own a global quota lock for this type */
450 if (lustre_handle_is_used(&qqi->qqi_lockh) &&
451 ldlm_lock_addref_try(&qqi->qqi_lockh, qsd_glb_einfo.ei_mode) == 0) {
452 read_unlock(&qsd->qsd_lock);
453 /* force refresh of global & slave index copy */
454 qti->qti_lvb.lvb_glb_ver = ~0ULL;
455 qti->qti_slv_ver = ~0ULL;
457 /* no valid lock found, let's enqueue a new one */
458 read_unlock(&qsd->qsd_lock);
460 memset(&qti->qti_body, 0, sizeof(qti->qti_body));
461 memcpy(&qti->qti_body.qb_fid, &qqi->qqi_fid,
462 sizeof(qqi->qqi_fid));
464 rc = qsd_intent_lock(env, qsd->qsd_exp, &qti->qti_body, true,
465 IT_QUOTA_CONN, qsd_reint_completion, qqi,
466 &qti->qti_lvb, (void *)&qti->qti_slv_ver);
468 GOTO(out_env_init, rc);
470 CDEBUG(D_QUOTA, "%s: glb_ver:"LPU64"/"LPU64",slv_ver:"LPU64"/"
471 LPU64"\n", qsd->qsd_svname,
472 qti->qti_lvb.lvb_glb_ver, qqi->qqi_glb_ver,
473 qti->qti_slv_ver, qqi->qqi_slv_ver);
476 /* Step 2: reintegrate global index */
477 if (!thread_is_running(thread))
478 GOTO(out_lock, rc = 0);
480 OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REINT, 10);
482 if (qqi->qqi_glb_ver != qti->qti_lvb.lvb_glb_ver) {
483 rc = qsd_reint_index(env, qqi, true);
485 CWARN("%s: reint global for "DFID" failed. %d\n",
486 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
490 qsd_bump_version(qqi, qqi->qqi_glb_ver, true);
493 /* Step 3: reintegrate slave index */
494 if (!thread_is_running(thread))
495 GOTO(out_lock, rc = 0);
497 if (qqi->qqi_slv_ver != qti->qti_slv_ver) {
498 rc = qsd_reint_index(env, qqi, false);
500 CWARN("%s: reintegration for "DFID" failed with %d\n",
501 qsd->qsd_svname, PFID(&qqi->qqi_slv_fid), rc);
505 qsd_bump_version(qqi, qqi->qqi_slv_ver, false);
508 /* wait for the qsd instance started (target recovery done) */
509 l_wait_event(thread->t_ctl_waitq,
510 qsd_started(qsd) || !thread_is_running(thread), &lwi);
512 if (!thread_is_running(thread))
513 GOTO(out_lock, rc = 0);
515 /* Step 4: start reconciliation for each enforced ID */
516 rc = qsd_reconciliation(env, qqi);
518 CWARN("%s: reconciliation for "DFID" failed with %d\n",
519 qsd->qsd_svname, PFID(&qqi->qqi_slv_fid), rc);
523 ldlm_lock_decref(&qqi->qqi_lockh, qsd_glb_einfo.ei_mode);
529 write_lock(&qsd->qsd_lock);
531 write_unlock(&qsd->qsd_lock);
534 lu_ref_del(&qqi->qqi_reference, "reint_thread", thread);
536 thread_set_flags(thread, SVC_STOPPED);
537 wake_up(&thread->t_ctl_waitq);
541 void qsd_stop_reint_thread(struct qsd_qtype_info *qqi)
543 struct ptlrpc_thread *thread = &qqi->qqi_reint_thread;
544 struct l_wait_info lwi = { 0 };
546 if (!thread_is_stopped(thread)) {
547 thread_set_flags(thread, SVC_STOPPING);
548 wake_up(&thread->t_ctl_waitq);
550 l_wait_event(thread->t_ctl_waitq,
551 thread_is_stopped(thread), &lwi);
555 static int qsd_entry_iter_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
556 struct hlist_node *hnode, void *data)
558 struct lquota_entry *lqe;
559 int *pending = (int *)data;
561 lqe = hlist_entry(hnode, struct lquota_entry, lqe_hash);
562 LASSERT(atomic_read(&lqe->lqe_ref) > 0);
565 *pending += lqe->lqe_pending_req;
566 lqe_read_unlock(lqe);
571 static bool qsd_pending_updates(struct qsd_qtype_info *qqi)
573 struct qsd_instance *qsd = qqi->qqi_qsd;
574 struct qsd_upd_rec *upd;
575 struct lquota_entry *lqe, *n;
577 bool updates = false;
580 /* any pending quota adjust? */
581 spin_lock(&qsd->qsd_adjust_lock);
582 list_for_each_entry_safe(lqe, n, &qsd->qsd_adjust_list, lqe_link) {
583 if (lqe2qqi(lqe) == qqi) {
584 list_del_init(&lqe->lqe_link);
588 spin_unlock(&qsd->qsd_adjust_lock);
590 /* any pending updates? */
591 read_lock(&qsd->qsd_lock);
592 list_for_each_entry(upd, &qsd->qsd_upd_list, qur_link) {
593 if (upd->qur_qqi == qqi) {
594 read_unlock(&qsd->qsd_lock);
595 CDEBUG(D_QUOTA, "%s: pending %s updates for type:%d.\n",
597 upd->qur_global ? "global" : "slave",
599 GOTO(out, updates = true);
602 read_unlock(&qsd->qsd_lock);
604 /* any pending quota request? */
605 cfs_hash_for_each_safe(qqi->qqi_site->lqs_hash, qsd_entry_iter_cb,
608 CDEBUG(D_QUOTA, "%s: pending dqacq for type:%d.\n",
609 qsd->qsd_svname, qqi->qqi_qtype);
615 CERROR("%s: Delaying reintegration for qtype:%d until pending "
616 "updates are flushed.\n",
617 qsd->qsd_svname, qqi->qqi_qtype);
621 int qsd_start_reint_thread(struct qsd_qtype_info *qqi)
623 struct ptlrpc_thread *thread = &qqi->qqi_reint_thread;
624 struct qsd_instance *qsd = qqi->qqi_qsd;
625 struct l_wait_info lwi = { 0 };
626 struct task_struct *task;
631 /* don't bother to do reintegration when quota isn't enabled */
632 if (!qsd_type_enabled(qsd, qqi->qqi_qtype))
635 if (qsd->qsd_acct_failed)
636 /* no space accounting support, can't enable enforcement */
639 /* check if the reintegration has already started or finished */
640 write_lock(&qsd->qsd_lock);
642 if ((qqi->qqi_glb_uptodate && qqi->qqi_slv_uptodate) ||
643 qqi->qqi_reint || qsd->qsd_stopping) {
644 write_unlock(&qsd->qsd_lock);
649 write_unlock(&qsd->qsd_lock);
651 /* there could be some unfinished global or index entry updates
652 * (very unlikely), to avoid them messing up with the reint
653 * procedure, we just return and try to re-start reint later. */
654 if (qsd_pending_updates(qqi)) {
655 write_lock(&qsd->qsd_lock);
657 write_unlock(&qsd->qsd_lock);
661 OBD_ALLOC(name, MTI_NAME_MAXLEN);
665 snprintf(name, MTI_NAME_MAXLEN, "qsd_reint_%d.%s",
666 qqi->qqi_qtype, qsd->qsd_svname);
668 task = kthread_run(qsd_reint_main, qqi, name);
669 OBD_FREE(name, MTI_NAME_MAXLEN);
673 thread_set_flags(thread, SVC_STOPPED);
674 write_lock(&qsd->qsd_lock);
676 write_unlock(&qsd->qsd_lock);
680 l_wait_event(thread->t_ctl_waitq,
681 thread_is_running(thread) || thread_is_stopped(thread),