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 <linux/kthread.h>
34 #include "qsd_internal.h"
37 * Completion function invoked when the global quota lock enqueue has completed
39 static void qsd_reint_completion(const struct lu_env *env,
40 struct qsd_qtype_info *qqi,
41 struct quota_body *req_qbody,
42 struct quota_body *rep_qbody,
43 struct lustre_handle *lockh,
44 struct lquota_lvb *lvb,
47 struct qsd_instance *qsd = qqi->qqi_qsd;
48 __u64 *slv_ver = (__u64 *)arg;
52 CDEBUG_LIMIT(rc != -EAGAIN ? D_ERROR : D_QUOTA,
53 "%s: failed to enqueue global quota lock, glb fid:"
54 DFID", rc:%d\n", qsd->qsd_svname,
55 PFID(&req_qbody->qb_fid), rc);
59 CDEBUG(D_QUOTA, "%s: global quota lock successfully acquired, glb "
60 "fid:"DFID", glb ver:"LPU64", slv fid:"DFID", slv ver:"LPU64"\n",
61 qsd->qsd_svname, PFID(&req_qbody->qb_fid),
62 lvb->lvb_glb_ver, PFID(&rep_qbody->qb_slv_fid),
63 rep_qbody->qb_slv_ver);
65 *slv_ver = rep_qbody->qb_slv_ver;
66 memcpy(&qqi->qqi_slv_fid, &rep_qbody->qb_slv_fid,
67 sizeof(struct lu_fid));
68 lustre_handle_copy(&qqi->qqi_lockh, lockh);
72 static int qsd_reint_qid(const struct lu_env *env, struct qsd_qtype_info *qqi,
73 bool global, union lquota_id *qid, void *rec)
75 struct lquota_entry *lqe;
79 lqe = lqe_locate(env, qqi->qqi_site, qid);
83 LQUOTA_DEBUG(lqe, "reintegrating entry");
85 rc = qsd_update_lqe(env, lqe, global, rec);
89 rc = qsd_update_index(env, qqi, qid, global, 0, rec);
95 static int qsd_reint_entries(const struct lu_env *env,
96 struct qsd_qtype_info *qqi,
97 struct idx_info *ii, bool global,
99 unsigned int npages, bool need_swab)
101 struct qsd_thread_info *qti = qsd_info(env);
102 struct qsd_instance *qsd = qqi->qqi_qsd;
103 union lquota_id *qid = &qti->qti_id;
108 CDEBUG(D_QUOTA, "%s: processing %d pages for %s index\n",
109 qsd->qsd_svname, npages, global ? "global" : "slave");
111 /* sanity check on the record size */
112 if ((global && ii->ii_recsize != sizeof(struct lquota_glb_rec)) ||
113 (!global && ii->ii_recsize != sizeof(struct lquota_slv_rec))) {
114 CERROR("%s: invalid record size (%d) for %s index\n",
115 qsd->qsd_svname, ii->ii_recsize,
116 global ? "global" : "slave");
120 size = ii->ii_recsize + ii->ii_keysize;
122 for (i = 0; i < npages; i++) {
123 union lu_page *lip = kmap(pages[i]);
125 for (j = 0; j < LU_PAGE_COUNT; j++) {
128 lustre_swab_lip_header(&lip->lp_idx);
130 if (lip->lp_idx.lip_magic != LIP_MAGIC) {
131 CERROR("%s: invalid magic (%x != %x) for page "
132 "%d/%d while transferring %s index\n",
133 qsd->qsd_svname, lip->lp_idx.lip_magic,
134 LIP_MAGIC, i + 1, npages,
135 global ? "global" : "slave");
136 GOTO(out, rc = -EINVAL);
139 CDEBUG(D_QUOTA, "%s: processing page %d/%d with %d "
140 "entries for %s index\n", qsd->qsd_svname, i + 1,
141 npages, lip->lp_idx.lip_nr,
142 global ? "global" : "slave");
144 for (k = 0; k < lip->lp_idx.lip_nr; k++) {
147 entry = lip->lp_idx.lip_entries + k * size;
148 memcpy(qid, entry, ii->ii_keysize); /* key */
149 entry += ii->ii_keysize; /* value */
155 __swab64s(&qid->qid_uid);
156 /* quota records only include 64-bit
158 while (offset < ii->ii_recsize) {
161 offset += sizeof(__u64);
165 rc = qsd_reint_qid(env, qqi, global, qid,
180 static int qsd_reint_index(const struct lu_env *env, struct qsd_qtype_info *qqi,
183 struct qsd_thread_info *qti = qsd_info(env);
184 struct qsd_instance *qsd = qqi->qqi_qsd;
185 struct idx_info *ii = &qti->qti_ii;
187 struct page **pages = NULL;
188 unsigned int npages, pg_cnt;
189 __u64 start_hash = 0, ver = 0;
190 bool need_swab = false;
194 fid = global ? &qqi->qqi_fid : &qqi->qqi_slv_fid;
196 /* let's do a 1MB bulk */
197 npages = min_t(unsigned int, OFD_MAX_BRW_SIZE, 1 << 20);
198 npages /= PAGE_CACHE_SIZE;
200 /* allocate pages for bulk index read */
201 OBD_ALLOC(pages, npages * sizeof(*pages));
203 GOTO(out, rc = -ENOMEM);
204 for (i = 0; i < npages; i++) {
205 pages[i] = alloc_page(GFP_IOFS);
206 if (pages[i] == NULL)
207 GOTO(out, rc = -ENOMEM);
211 /* initialize index_info request with FID of global index */
212 memset(ii, 0, sizeof(*ii));
213 memcpy(&ii->ii_fid, fid, sizeof(*fid));
214 ii->ii_magic = IDX_INFO_MAGIC;
215 ii->ii_flags = II_FL_NOHASH;
216 ii->ii_count = npages * LU_PAGE_COUNT;
217 ii->ii_hash_start = start_hash;
219 /* send bulk request to quota master to read global index */
220 rc = qsd_fetch_index(env, qsd->qsd_exp, ii, npages, pages, &need_swab);
222 CWARN("%s: failed to fetch index for "DFID". %d\n",
223 qsd->qsd_svname, PFID(fid), rc);
227 /* various sanity checks */
228 if (ii->ii_magic != IDX_INFO_MAGIC) {
229 CERROR("%s: invalid magic in index transfer %x != %x\n",
230 qsd->qsd_svname, ii->ii_magic, IDX_INFO_MAGIC);
231 GOTO(out, rc = -EPROTO);
233 if ((ii->ii_flags & II_FL_VARKEY) != 0)
234 CWARN("%s: II_FL_VARKEY is set on index transfer for fid "DFID
235 ", it shouldn't be\n", qsd->qsd_svname, PFID(fid));
236 if ((ii->ii_flags & II_FL_NONUNQ) != 0)
237 CWARN("%s: II_FL_NONUNQ is set on index transfer for fid "DFID
238 ", it shouldn't be\n", qsd->qsd_svname, PFID(fid));
239 if (ii->ii_keysize != sizeof(__u64)) {
240 CERROR("%s: invalid key size reported on index transfer for "
241 "fid "DFID", %u != %u\n", qsd->qsd_svname, PFID(fid),
242 ii->ii_keysize, (int)sizeof(__u64));
243 GOTO(out, rc = -EPROTO);
245 if (ii->ii_version == 0 && ii->ii_count != 0)
246 CWARN("%s: index version for fid "DFID" is 0, but index isn't "
247 "empty (%d)\n", qsd->qsd_svname, PFID(fid), ii->ii_count);
249 CDEBUG(D_QUOTA, "%s: reintegration process for fid "DFID" successfully "
250 "fetched %s index, count = %d\n", qsd->qsd_svname,
251 PFID(fid), global ? "global" : "slave", ii->ii_count);
254 /* record version associated with the first bulk transfer */
255 ver = ii->ii_version;
257 pg_cnt = (ii->ii_count + (LU_PAGE_COUNT) - 1);
258 pg_cnt >>= PAGE_CACHE_SHIFT - LU_PAGE_SHIFT;
260 if (pg_cnt > npages) {
261 CERROR("%s: master returned more pages than expected, %u > %u"
262 "\n", qsd->qsd_svname, pg_cnt, npages);
266 rc = qsd_reint_entries(env, qqi, ii, global, pages, pg_cnt, need_swab);
270 if (ii->ii_hash_end != II_END_OFF) {
271 start_hash = ii->ii_hash_end;
276 for (i = 0; i < npages; i++)
277 if (pages[i] != NULL)
278 __free_page(pages[i]);
279 OBD_FREE(pages, npages * sizeof(*pages));
282 /* Update index version */
284 rc = qsd_write_version(env, qqi, ver, global);
286 CERROR("%s: write version "LPU64" to "DFID" failed. "
287 "%d\n", qsd->qsd_svname, ver, PFID(fid), rc);
293 static int qsd_reconciliation(const struct lu_env *env,
294 struct qsd_qtype_info *qqi)
296 struct qsd_thread_info *qti = qsd_info(env);
297 struct qsd_instance *qsd = qqi->qqi_qsd;
298 const struct dt_it_ops *iops;
301 struct lquota_entry *lqe;
302 union lquota_id *qid = &qti->qti_id;
306 LASSERT(qqi->qqi_glb_obj != NULL);
307 iops = &qqi->qqi_glb_obj->do_index_ops->dio_it;
309 it = iops->init(env, qqi->qqi_glb_obj, 0);
311 CWARN("%s: Initialize it for "DFID" failed. %ld\n",
312 qsd->qsd_svname, PFID(&qqi->qqi_fid), PTR_ERR(it));
316 rc = iops->load(env, it, 0);
318 CWARN("%s: Load first entry for "DFID" failed. %d\n",
319 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
321 } else if (rc == 0) {
322 rc = iops->next(env, it);
324 GOTO(out, rc = (rc < 0) ? rc : 0);
328 key = iops->key(env, it);
330 CWARN("%s: Error key for "DFID". %ld\n",
331 qsd->qsd_svname, PFID(&qqi->qqi_fid),
333 GOTO(out, rc = PTR_ERR(key));
336 /* skip the root user/group */
337 if (*((__u64 *)key) == 0)
340 qid->qid_uid = *((__u64 *)key);
342 lqe = lqe_locate(env, qqi->qqi_site, qid);
344 CWARN("%s: failed to locate lqe. "DFID", %ld\n",
345 qsd->qsd_svname, PFID(&qqi->qqi_fid),
347 GOTO(out, rc = PTR_ERR(lqe));
350 rc = qsd_refresh_usage(env, lqe);
352 CWARN("%s: failed to get usage. "DFID", %d\n",
353 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
358 rc = qsd_adjust(env, lqe);
361 CWARN("%s: failed to report quota. "DFID", %d\n",
362 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
366 rc = iops->next(env, it);
368 CWARN("%s: failed to parse index, ->next error:%d "DFID
369 "\n", qsd->qsd_svname, rc, PFID(&qqi->qqi_fid));
381 static int qsd_connected(struct qsd_instance *qsd)
385 read_lock(&qsd->qsd_lock);
386 connected = qsd->qsd_exp_valid ? 1 : 0;
387 read_unlock(&qsd->qsd_lock);
392 static int qsd_started(struct qsd_instance *qsd)
396 read_lock(&qsd->qsd_lock);
397 started = qsd->qsd_started ? 1 : 0;
398 read_unlock(&qsd->qsd_lock);
404 * Routine executed by the reintegration thread.
406 static int qsd_reint_main(void *args)
409 struct qsd_thread_info *qti;
410 struct qsd_qtype_info *qqi = (struct qsd_qtype_info *)args;
411 struct qsd_instance *qsd = qqi->qqi_qsd;
412 struct ptlrpc_thread *thread = &qqi->qqi_reint_thread;
413 struct l_wait_info lwi = { 0 };
417 CDEBUG(D_QUOTA, "%s: Starting reintegration thread for "DFID"\n",
418 qsd->qsd_svname, PFID(&qqi->qqi_fid));
421 lu_ref_add(&qqi->qqi_reference, "reint_thread", thread);
423 thread_set_flags(thread, SVC_RUNNING);
424 wake_up(&thread->t_ctl_waitq);
428 GOTO(out, rc = -ENOMEM);
430 /* initialize environment */
431 rc = lu_env_init(env, LCT_DT_THREAD);
436 /* wait for the connection to master established */
437 l_wait_event(thread->t_ctl_waitq,
438 qsd_connected(qsd) || !thread_is_running(thread), &lwi);
440 /* Step 1: enqueue global index lock */
441 if (!thread_is_running(thread))
442 GOTO(out_env_init, rc = 0);
444 LASSERT(qsd->qsd_exp != NULL);
445 LASSERT(qqi->qqi_glb_uptodate == 0 || qqi->qqi_slv_uptodate == 0);
447 memset(&qti->qti_lvb, 0, sizeof(qti->qti_lvb));
449 read_lock(&qsd->qsd_lock);
450 /* check whether we already own a global quota lock for this type */
451 if (lustre_handle_is_used(&qqi->qqi_lockh) &&
452 ldlm_lock_addref_try(&qqi->qqi_lockh, qsd_glb_einfo.ei_mode) == 0) {
453 read_unlock(&qsd->qsd_lock);
454 /* force refresh of global & slave index copy */
455 qti->qti_lvb.lvb_glb_ver = ~0ULL;
456 qti->qti_slv_ver = ~0ULL;
458 /* no valid lock found, let's enqueue a new one */
459 read_unlock(&qsd->qsd_lock);
461 memset(&qti->qti_body, 0, sizeof(qti->qti_body));
462 memcpy(&qti->qti_body.qb_fid, &qqi->qqi_fid,
463 sizeof(qqi->qqi_fid));
465 rc = qsd_intent_lock(env, qsd->qsd_exp, &qti->qti_body, true,
466 IT_QUOTA_CONN, qsd_reint_completion, qqi,
467 &qti->qti_lvb, (void *)&qti->qti_slv_ver);
469 GOTO(out_env_init, rc);
471 CDEBUG(D_QUOTA, "%s: glb_ver:"LPU64"/"LPU64",slv_ver:"LPU64"/"
472 LPU64"\n", qsd->qsd_svname,
473 qti->qti_lvb.lvb_glb_ver, qqi->qqi_glb_ver,
474 qti->qti_slv_ver, qqi->qqi_slv_ver);
477 /* Step 2: reintegrate global index */
478 if (!thread_is_running(thread))
479 GOTO(out_lock, rc = 0);
481 OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REINT, 10);
483 if (qqi->qqi_glb_ver != qti->qti_lvb.lvb_glb_ver) {
484 rc = qsd_reint_index(env, qqi, true);
486 CWARN("%s: reint global for "DFID" failed. %d\n",
487 qsd->qsd_svname, PFID(&qqi->qqi_fid), rc);
491 qsd_bump_version(qqi, qqi->qqi_glb_ver, true);
494 /* Step 3: reintegrate slave index */
495 if (!thread_is_running(thread))
496 GOTO(out_lock, rc = 0);
498 if (qqi->qqi_slv_ver != qti->qti_slv_ver) {
499 rc = qsd_reint_index(env, qqi, false);
501 CWARN("%s: reintegration for "DFID" failed with %d\n",
502 qsd->qsd_svname, PFID(&qqi->qqi_slv_fid), rc);
506 qsd_bump_version(qqi, qqi->qqi_slv_ver, false);
509 /* wait for the qsd instance started (target recovery done) */
510 l_wait_event(thread->t_ctl_waitq,
511 qsd_started(qsd) || !thread_is_running(thread), &lwi);
513 if (!thread_is_running(thread))
514 GOTO(out_lock, rc = 0);
516 /* Step 4: start reconciliation for each enforced ID */
517 rc = qsd_reconciliation(env, qqi);
519 CWARN("%s: reconciliation for "DFID" failed with %d\n",
520 qsd->qsd_svname, PFID(&qqi->qqi_slv_fid), rc);
524 ldlm_lock_decref(&qqi->qqi_lockh, qsd_glb_einfo.ei_mode);
530 write_lock(&qsd->qsd_lock);
532 write_unlock(&qsd->qsd_lock);
535 lu_ref_del(&qqi->qqi_reference, "reint_thread", thread);
537 thread_set_flags(thread, SVC_STOPPED);
538 wake_up(&thread->t_ctl_waitq);
542 void qsd_stop_reint_thread(struct qsd_qtype_info *qqi)
544 struct ptlrpc_thread *thread = &qqi->qqi_reint_thread;
545 struct l_wait_info lwi = { 0 };
547 if (!thread_is_stopped(thread)) {
548 thread_set_flags(thread, SVC_STOPPING);
549 wake_up(&thread->t_ctl_waitq);
551 l_wait_event(thread->t_ctl_waitq,
552 thread_is_stopped(thread), &lwi);
556 static int qsd_entry_iter_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
557 struct hlist_node *hnode, void *data)
559 struct lquota_entry *lqe;
560 int *pending = (int *)data;
562 lqe = hlist_entry(hnode, struct lquota_entry, lqe_hash);
563 LASSERT(atomic_read(&lqe->lqe_ref) > 0);
566 *pending += lqe->lqe_pending_req;
567 lqe_read_unlock(lqe);
572 static bool qsd_pending_updates(struct qsd_qtype_info *qqi)
574 struct qsd_instance *qsd = qqi->qqi_qsd;
575 struct qsd_upd_rec *upd;
576 struct lquota_entry *lqe, *n;
578 bool updates = false;
581 /* any pending quota adjust? */
582 spin_lock(&qsd->qsd_adjust_lock);
583 list_for_each_entry_safe(lqe, n, &qsd->qsd_adjust_list, lqe_link) {
584 if (lqe2qqi(lqe) == qqi) {
585 list_del_init(&lqe->lqe_link);
589 spin_unlock(&qsd->qsd_adjust_lock);
591 /* any pending updates? */
592 read_lock(&qsd->qsd_lock);
593 list_for_each_entry(upd, &qsd->qsd_upd_list, qur_link) {
594 if (upd->qur_qqi == qqi) {
595 read_unlock(&qsd->qsd_lock);
596 CDEBUG(D_QUOTA, "%s: pending %s updates for type:%d.\n",
598 upd->qur_global ? "global" : "slave",
600 GOTO(out, updates = true);
603 read_unlock(&qsd->qsd_lock);
605 /* any pending quota request? */
606 cfs_hash_for_each_safe(qqi->qqi_site->lqs_hash, qsd_entry_iter_cb,
609 CDEBUG(D_QUOTA, "%s: pending dqacq for type:%d.\n",
610 qsd->qsd_svname, qqi->qqi_qtype);
616 CERROR("%s: Delaying reintegration for qtype:%d until pending "
617 "updates are flushed.\n",
618 qsd->qsd_svname, qqi->qqi_qtype);
622 int qsd_start_reint_thread(struct qsd_qtype_info *qqi)
624 struct ptlrpc_thread *thread = &qqi->qqi_reint_thread;
625 struct qsd_instance *qsd = qqi->qqi_qsd;
626 struct l_wait_info lwi = { 0 };
627 struct task_struct *task;
632 /* don't bother to do reintegration when quota isn't enabled */
633 if (!qsd_type_enabled(qsd, qqi->qqi_qtype))
636 if (qsd->qsd_acct_failed)
637 /* no space accounting support, can't enable enforcement */
640 /* check if the reintegration has already started or finished */
641 write_lock(&qsd->qsd_lock);
643 if ((qqi->qqi_glb_uptodate && qqi->qqi_slv_uptodate) ||
644 qqi->qqi_reint || qsd->qsd_stopping) {
645 write_unlock(&qsd->qsd_lock);
650 write_unlock(&qsd->qsd_lock);
652 /* there could be some unfinished global or index entry updates
653 * (very unlikely), to avoid them messing up with the reint
654 * procedure, we just return and try to re-start reint later. */
655 if (qsd_pending_updates(qqi)) {
656 write_lock(&qsd->qsd_lock);
658 write_unlock(&qsd->qsd_lock);
662 OBD_ALLOC(name, MTI_NAME_MAXLEN);
666 snprintf(name, MTI_NAME_MAXLEN, "qsd_reint_%d.%s",
667 qqi->qqi_qtype, qsd->qsd_svname);
669 task = kthread_run(qsd_reint_main, qqi, name);
670 OBD_FREE(name, MTI_NAME_MAXLEN);
674 thread_set_flags(thread, SVC_STOPPED);
675 write_lock(&qsd->qsd_lock);
677 write_unlock(&qsd->qsd_lock);
681 l_wait_event(thread->t_ctl_waitq,
682 thread_is_running(thread) || thread_is_stopped(thread),