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, 2017, 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 * Quota Slave Driver (QSD) management.
34 * The quota slave feature is implemented under the form of a library called
35 * QSD. Each OSD device should create a QSD instance via qsd_init() which will
36 * be used to manage quota enforcement for this device. This implies:
37 * - completing the reintegration procedure with the quota master (aka QMT, see
38 * qmt_dev.c) to retrieve the latest quota settings and space distribution.
39 * - managing quota locks in order to be notified of configuration changes.
40 * - acquiring space from the QMT when quota space for a given user/group is
41 * close to exhaustion.
42 * - allocating quota space to service threads for local request processing.
44 * Once the QSD instance created, the OSD device should invoke qsd_start()
45 * when recovery is completed. This notifies the QSD that we are about to
46 * process new requests on which quota should be strictly enforced.
47 * Then, qsd_op_begin/end can be used to reserve/release/pre-acquire quota space
48 * for/after each operation until shutdown where the QSD instance should be
49 * freed via qsd_fini().
52 #define DEBUG_SUBSYSTEM S_LQUOTA
54 #include <obd_class.h>
55 #include "qsd_internal.h"
57 struct kmem_cache *upd_kmem;
59 struct lu_kmem_descr qsd_caches[] = {
61 .ckd_cache = &upd_kmem,
62 .ckd_name = "upd_kmem",
63 .ckd_size = sizeof(struct qsd_upd_rec)
70 /* define qsd thread key */
71 LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
72 LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
73 LU_KEY_INIT_GENERIC(qsd);
75 /* some procfs helpers */
76 static int qsd_state_seq_show(struct seq_file *m, void *data)
78 struct qsd_instance *qsd = m->private;
83 memset(enabled, 0, sizeof(enabled));
84 if (qsd_type_enabled(qsd, USRQUOTA))
86 if (qsd_type_enabled(qsd, GRPQUOTA))
88 if (qsd_type_enabled(qsd, PRJQUOTA))
89 strncat(enabled, "p", 1);
90 if (strlen(enabled) == 0)
91 strcat(enabled, "none");
93 /* TODO: further pool ID should be removed or
94 * replaced with pool Name */
95 seq_printf(m, "target name: %s\n"
99 "conn to master: %s\n",
101 qsd->qsd_is_md ? "md" : "dt", enabled,
102 qsd->qsd_exp_valid ? "setup" : "not setup yet");
104 if (qsd->qsd_prepared) {
105 memset(enabled, 0, sizeof(enabled));
106 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
107 strcat(enabled, "u");
108 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
109 strcat(enabled, "g");
110 if (qsd->qsd_type_array[PRJQUOTA]->qqi_acct_obj != NULL)
111 strncat(enabled, "p", 1);
112 if (strlen(enabled) == 0)
113 strcat(enabled, "none");
114 seq_printf(m, "space acct: %s\n"
115 "user uptodate: glb[%d],slv[%d],reint[%d]\n"
116 "group uptodate: glb[%d],slv[%d],reint[%d]\n"
117 "project uptodate: glb[%d],slv[%d],reint[%d]\n",
119 qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
120 qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
121 qsd->qsd_type_array[USRQUOTA]->qqi_reint,
122 qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
123 qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
124 qsd->qsd_type_array[GRPQUOTA]->qqi_reint,
125 qsd->qsd_type_array[PRJQUOTA]->qqi_glb_uptodate,
126 qsd->qsd_type_array[PRJQUOTA]->qqi_slv_uptodate,
127 qsd->qsd_type_array[PRJQUOTA]->qqi_reint);
131 LPROC_SEQ_FOPS_RO(qsd_state);
133 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
135 struct qsd_instance *qsd = m->private;
138 LASSERT(qsd != NULL);
140 memset(enabled, 0, sizeof(enabled));
141 if (qsd_type_enabled(qsd, USRQUOTA))
142 strncat(enabled, "u", sizeof(enabled) - strlen(enabled));
143 if (qsd_type_enabled(qsd, GRPQUOTA))
144 strncat(enabled, "g", sizeof(enabled) - strlen(enabled));
145 if (qsd_type_enabled(qsd, PRJQUOTA))
146 strncat(enabled, "p", sizeof(enabled) - strlen(enabled));
147 if (strlen(enabled) == 0)
148 strncat(enabled, "none", sizeof(enabled) - strlen(enabled));
150 seq_printf(m, "%s\n", enabled);
154 static ssize_t qsd_enabled_seq_write(struct file *file,
155 const char __user *buffer,
156 size_t count, loff_t *off)
158 struct seq_file *m = file->private_data;
159 struct qsd_instance *qsd = m->private;
160 char fsname[LUSTRE_MAXFSNAME + 1];
168 if (copy_from_user(valstr, buffer, count))
169 GOTO(out, count = -EFAULT);
171 valstr[sizeof(valstr) - 1] = 0;
172 if (strchr(valstr, 'u'))
173 enabled |= BIT(USRQUOTA);
174 if (strchr(valstr, 'g'))
175 enabled |= BIT(GRPQUOTA);
176 if (strchr(valstr, 'p'))
177 enabled |= BIT(PRJQUOTA);
179 if (enabled == 0 && strcmp(valstr, "none"))
180 GOTO(out, count = -EINVAL);
183 pool = LQUOTA_RES_MD;
185 pool = LQUOTA_RES_DT;
187 if (server_name2fsname(qsd->qsd_svname, fsname, NULL))
188 GOTO(out, count = -EINVAL);
190 rc = qsd_config(valstr, fsname, pool);
196 LPROC_SEQ_FOPS(qsd_enabled);
198 /* force reintegration procedure to be executed.
199 * Used for test/debugging purpose */
201 lprocfs_force_reint_seq_write(struct file *file, const char __user *buffer,
202 size_t count, loff_t *off)
204 struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
207 LASSERT(qsd != NULL);
209 write_lock(&qsd->qsd_lock);
210 if (qsd->qsd_stopping) {
211 /* don't mess up with shutdown procedure, it is already
212 * complicated enough */
214 } else if (!qsd->qsd_prepared) {
217 /* mark all indexes as stale */
218 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
219 qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
220 qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
223 write_unlock(&qsd->qsd_lock);
228 /* kick off reintegration */
229 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
230 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
234 return rc == 0 ? count : rc;
236 LPROC_SEQ_FOPS_WR_ONLY(qsd, force_reint);
238 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
240 struct qsd_instance *qsd = m->private;
241 LASSERT(qsd != NULL);
243 seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
248 qsd_timeout_seq_write(struct file *file, const char __user *buffer,
249 size_t count, loff_t *off)
251 struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
255 LASSERT(qsd != NULL);
256 rc = kstrtoll_from_user(buffer, count, 0, &timeout);
263 qsd->qsd_timeout = timeout;
266 LPROC_SEQ_FOPS(qsd_timeout);
268 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
270 .fops = &qsd_state_fops },
272 .fops = &qsd_enabled_fops },
273 { .name = "force_reint",
274 .fops = &qsd_force_reint_fops },
276 .fops = &qsd_timeout_fops },
281 * Callback function invoked by the OSP layer when the connection to the master
284 * \param data - is a pointer to the qsd_instance
286 * \retval - 0 on success, appropriate error on failure
288 static int qsd_conn_callback(void *data)
290 struct qsd_instance *qsd = (struct qsd_instance *)data;
294 /* qsd_exp should now be valid */
295 LASSERT(qsd->qsd_exp);
297 qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
299 write_lock(&qsd->qsd_lock);
300 /* notify that qsd_exp is now valid */
301 qsd->qsd_exp_valid = true;
302 write_unlock(&qsd->qsd_lock);
304 /* Now that the connection to master is setup, we can initiate the
305 * reintegration procedure for quota types which are enabled.
306 * It is worth noting that, if the qsd_instance hasn't been started
307 * already, then we can only complete the first two steps of the
308 * reintegration procedure (i.e. global lock enqueue and slave
309 * index transfer) since the space usage reconciliation (i.e.
310 * step 3) will have to wait for qsd_start() to be called */
311 for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
312 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
313 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
320 * Release qsd_qtype_info structure which contains data associated with a
321 * given quota type. This releases the accounting objects.
322 * It's called on OSD cleanup when the qsd instance is released.
324 * \param env - is the environment passed by the caller
325 * \param qsd - is the qsd instance managing the qsd_qtype_info structure
327 * \param qtype - is the quota type to be shutdown
329 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
332 struct qsd_qtype_info *qqi;
336 if (qsd->qsd_type_array[qtype] == NULL)
338 qqi = qsd->qsd_type_array[qtype];
339 qsd->qsd_type_array[qtype] = NULL;
341 /* all deferred work lists should be empty */
342 LASSERT(list_empty(&qqi->qqi_deferred_glb));
343 LASSERT(list_empty(&qqi->qqi_deferred_slv));
345 /* shutdown lquota site */
346 if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
347 lquota_site_free(env, qqi->qqi_site);
348 qqi->qqi_site = NULL;
351 /* The qqi may still be holding by global locks which are being
352 * canceled asynchronously (LU-4365), see the following steps:
354 * - On server umount, we try to clear all quota locks first by
355 * disconnecting LWP (which will invalidate import and cleanup
356 * all locks on it), however, if quota reint process is holding
357 * the global lock for reintegration at that time, global lock
358 * will fail to be cleared on LWP disconnection.
360 * - Umount process goes on and stops reint process, the global
361 * lock will be dropped on reint process exit, however, the lock
362 * cancel in done in asynchronous way, so the
363 * qsd_glb_blocking_ast() might haven't been called yet when we
366 while (atomic_read(&qqi->qqi_ref) > 1) {
367 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
368 atomic_read(&qqi->qqi_ref), repeat);
370 set_current_state(TASK_INTERRUPTIBLE);
371 schedule_timeout(cfs_time_seconds(1));
374 /* by now, all qqi users should have gone away */
375 LASSERT(atomic_read(&qqi->qqi_ref) == 1);
376 lu_ref_fini(&qqi->qqi_reference);
378 /* release accounting object */
379 if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
380 dt_object_put(env, qqi->qqi_acct_obj);
381 qqi->qqi_acct_obj = NULL;
384 /* release slv index */
385 if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
386 dt_object_put(env, qqi->qqi_slv_obj);
387 qqi->qqi_slv_obj = NULL;
388 qqi->qqi_slv_ver = 0;
391 /* release global index */
392 if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
393 dt_object_put(env, qqi->qqi_glb_obj);
394 qqi->qqi_glb_obj = NULL;
395 qqi->qqi_glb_ver = 0;
402 static const char *qtype2acct_name(int qtype)
404 static char unknown[24];
412 return "acct_project";
415 snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
419 static const char *qtype2glb_name(int qtype)
421 static char unknown[24];
427 return "limit_group";
429 return "limit_project";
432 snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
437 * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
438 * This opens the accounting object and initializes the proc file.
439 * It's called on OSD start when the qsd_prepare() is invoked on the qsd
442 * \param env - the environment passed by the caller
443 * \param qsd - is the qsd instance which will be in charge of the new
444 * qsd_qtype_info instance.
445 * \param qtype - is quota type to set up
447 * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
448 * appropriate error on failure
450 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
453 struct qsd_qtype_info *qqi;
455 struct obd_uuid uuid;
458 LASSERT(qsd->qsd_type_array[qtype] == NULL);
460 /* allocate structure for this quota type */
464 qsd->qsd_type_array[qtype] = qqi;
465 atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
467 /* set backpointer and other parameters */
469 qqi->qqi_qtype = qtype;
470 lu_ref_init(&qqi->qqi_reference);
471 qqi->qqi_glb_uptodate = false;
472 qqi->qqi_slv_uptodate = false;
473 qqi->qqi_reint = false;
474 init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
475 thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
476 INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
477 INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
478 lquota_generate_fid(&qqi->qqi_fid, QSD_RES_TYPE(qsd), qtype);
480 /* open accounting object */
481 LASSERT(qqi->qqi_acct_obj == NULL);
482 qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
483 if (IS_ERR(qqi->qqi_acct_obj)) {
484 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
485 qsd->qsd_svname, qtype_name(qtype),
486 PTR_ERR(qqi->qqi_acct_obj));
487 qqi->qqi_acct_obj = NULL;
488 qqi->qqi_acct_failed = true;
491 /* open global index copy */
492 LASSERT(qqi->qqi_glb_obj == NULL);
493 qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
495 &qqi->qqi_fid, true);
496 if (IS_ERR(qqi->qqi_glb_obj)) {
497 CERROR("%s: can't open global index copy "DFID" %ld\n",
498 qsd->qsd_svname, PFID(&qqi->qqi_fid),
499 PTR_ERR(qqi->qqi_glb_obj));
500 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
502 qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
504 /* open slave index copy */
505 LASSERT(qqi->qqi_slv_obj == NULL);
506 obd_str2uuid(&uuid, qsd->qsd_svname);
507 qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
509 &qqi->qqi_fid, &uuid,
511 if (IS_ERR(qqi->qqi_slv_obj)) {
512 CERROR("%s: can't open slave index copy "DFID" %ld\n",
513 qsd->qsd_svname, PFID(&qqi->qqi_fid),
514 PTR_ERR(qqi->qqi_slv_obj));
515 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
517 qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
520 qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
521 if (IS_ERR(qqi->qqi_site)) {
522 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
523 PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
524 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
527 /* register proc entry for accounting & global index copy objects */
528 rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
529 0444, &lprocfs_quota_seq_fops,
532 CERROR("%s: can't add procfs entry for accounting file %d\n",
533 qsd->qsd_svname, rc);
537 rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
538 0444, &lprocfs_quota_seq_fops,
541 CERROR("%s: can't add procfs entry for global index copy %d\n",
542 qsd->qsd_svname, rc);
548 qsd_qtype_fini(env, qsd, qtype);
553 * Release a qsd_instance. Companion of qsd_init(). This releases all data
554 * structures associated with the quota slave (on-disk objects, lquota entry
556 * This function should be called when the OSD is shutting down.
558 * \param env - is the environment passed by the caller
559 * \param qsd - is the qsd instance to shutdown
561 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
566 if (unlikely(qsd == NULL))
569 CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
570 write_lock(&qsd->qsd_lock);
571 qsd->qsd_stopping = true;
572 write_unlock(&qsd->qsd_lock);
574 /* remove qsd proc entry */
575 if (qsd->qsd_proc != NULL) {
576 lprocfs_remove(&qsd->qsd_proc);
577 qsd->qsd_proc = NULL;
580 /* stop the writeback thread */
581 qsd_stop_upd_thread(qsd);
583 /* shutdown the reintegration threads */
584 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
585 if (qsd->qsd_type_array[qtype] == NULL)
587 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
590 if (qsd->qsd_ns != NULL) {
594 /* release per-filesystem information */
595 if (qsd->qsd_fsinfo != NULL) {
596 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
597 /* remove from the list of fsinfo */
598 list_del_init(&qsd->qsd_link);
599 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
600 qsd_put_fsinfo(qsd->qsd_fsinfo);
601 qsd->qsd_fsinfo = NULL;
604 /* free per-quota type data */
605 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
606 qsd_qtype_fini(env, qsd, qtype);
608 /* deregister connection to the quota master */
609 qsd->qsd_exp_valid = false;
610 lustre_deregister_lwp_item(&qsd->qsd_exp);
612 /* release quota root directory */
613 if (qsd->qsd_root != NULL) {
614 dt_object_put(env, qsd->qsd_root);
615 qsd->qsd_root = NULL;
618 /* release reference on dt_device */
619 if (qsd->qsd_dev != NULL) {
620 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
621 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
625 CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
629 EXPORT_SYMBOL(qsd_fini);
632 * Create a new qsd_instance to be associated with backend osd device
633 * identified by \dev.
635 * \param env - the environment passed by the caller
636 * \param svname - is the service name of the OSD device creating this instance
637 * \param dev - is the dt_device where to store quota index files
638 * \param osd_proc - is the procfs parent directory where to create procfs file
639 * related to this new qsd instance
641 * \retval - pointer to new qsd_instance associated with dev \dev on success,
642 * appropriate error on failure
644 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
645 struct dt_device *dev,
646 struct proc_dir_entry *osd_proc, bool is_md)
648 struct qsd_thread_info *qti = qsd_info(env);
649 struct qsd_instance *qsd;
653 /* only configure qsd for MDT & OST */
654 type = server_name2index(svname, &idx, NULL);
655 if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
658 /* allocate qsd instance */
661 RETURN(ERR_PTR(-ENOMEM));
663 /* generic initializations */
664 rwlock_init(&qsd->qsd_lock);
665 INIT_LIST_HEAD(&qsd->qsd_link);
666 thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
667 init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
668 INIT_LIST_HEAD(&qsd->qsd_upd_list);
669 spin_lock_init(&qsd->qsd_adjust_lock);
670 INIT_LIST_HEAD(&qsd->qsd_adjust_list);
671 qsd->qsd_prepared = false;
672 qsd->qsd_started = false;
673 qsd->qsd_is_md = is_md;
675 /* copy service name */
676 if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
677 >= sizeof(qsd->qsd_svname))
678 GOTO(out, rc = -E2BIG);
680 /* grab reference on osd device */
681 lu_device_get(&dev->dd_lu_dev);
682 lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
685 /* get fsname from svname */
686 rc = server_name2fsname(svname, qti->qti_buf, NULL);
688 CERROR("%s: fail to extract filesystem name\n", svname);
692 /* look up quota setting for the filesystem the target belongs to */
693 qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
694 if (qsd->qsd_fsinfo == NULL) {
695 CERROR("%s: failed to locate filesystem information\n", svname);
696 GOTO(out, rc = -EINVAL);
699 /* add in the list of lquota_fsinfo */
700 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
701 list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
702 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
704 /* register procfs directory */
706 qsd->qsd_proc = lprocfs_register(QSD_DIR_MD, osd_proc,
707 lprocfs_quota_qsd_vars, qsd);
709 qsd->qsd_proc = lprocfs_register(QSD_DIR_DT, osd_proc,
710 lprocfs_quota_qsd_vars, qsd);
712 if (type == LDD_F_SV_TYPE_MDT && qsd->qsd_is_md)
713 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_MD);
714 else if (type == LDD_F_SV_TYPE_OST && !qsd->qsd_is_md)
715 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_DT);
717 if (IS_ERR(qsd->qsd_proc)) {
718 rc = PTR_ERR(qsd->qsd_proc);
719 qsd->qsd_proc = NULL;
720 CERROR("%s: fail to create quota slave proc entry (%d)\n",
732 EXPORT_SYMBOL(qsd_init);
735 * Initialize on-disk structures in order to manage quota enforcement for
736 * the target associated with the qsd instance \qsd and starts the reintegration
737 * procedure for each quota type as soon as possible.
738 * The last step of the reintegration will be completed once qsd_start() is
739 * called, at which points the space reconciliation with the master will be
741 * This function must be called when the server stack is fully configured,
742 * typically when ->ldo_prepare is called across the stack.
744 * \param env - the environment passed by the caller
745 * \param qsd - is qsd_instance to prepare
747 * \retval - 0 on success, appropriate error on failure
749 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
751 struct qsd_thread_info *qti = qsd_info(env);
755 if (unlikely(qsd == NULL))
758 read_lock(&qsd->qsd_lock);
759 if (qsd->qsd_prepared) {
760 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
763 read_unlock(&qsd->qsd_lock);
767 /* Record whether this qsd instance is managing quota enforcement for a
768 * MDT (i.e. inode quota) or OST (block quota) */
770 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
772 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
774 /* look-up on-disk directory for the quota slave */
775 qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
777 if (IS_ERR(qsd->qsd_root)) {
778 rc = PTR_ERR(qsd->qsd_root);
779 qsd->qsd_root = NULL;
780 CERROR("%s: failed to create quota slave root dir (%d)\n",
781 qsd->qsd_svname, rc);
785 /* initialize per-quota type data */
786 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
787 rc = qsd_qtype_init(env, qsd, qtype);
792 /* pools successfully setup, mark the qsd as prepared */
793 write_lock(&qsd->qsd_lock);
794 qsd->qsd_prepared = true;
795 write_unlock(&qsd->qsd_lock);
797 if (qsd->qsd_dev->dd_rdonly)
800 /* start reintegration thread for each type, if required */
801 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
802 struct qsd_qtype_info *qqi = qsd->qsd_type_array[qtype];
804 if (qsd_type_enabled(qsd, qtype) &&
805 qqi->qqi_acct_failed) {
806 LCONSOLE_ERROR("%s: can't enable quota enforcement "
807 "since space accounting isn't functional"
808 ". Please run tunefs.lustre --quota on "
809 "an unmounted filesystem if not done "
810 "already\n", qsd->qsd_svname);
814 rc = qsd_start_reint_thread(qqi);
816 CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
817 qsd->qsd_svname, qtype_name(qtype), rc);
822 /* start writeback thread */
823 rc = qsd_start_upd_thread(qsd);
825 CERROR("%s: failed to start writeback thread (%d)\n",
826 qsd->qsd_svname, rc);
830 /* generate osp name */
831 rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
834 CERROR("%s: failed to generate ospname (%d)\n",
835 qsd->qsd_svname, rc);
839 /* the connection callback will start the reintegration
840 * procedure if quota is enabled */
841 rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
842 qsd_conn_callback, (void *)qsd);
844 CERROR("%s: fail to get connection to master (%d)\n",
845 qsd->qsd_svname, rc);
851 EXPORT_SYMBOL(qsd_prepare);
854 * Start a qsd instance. This will complete the last step of the reintegration
855 * procedure as soon as possible (provided that the master is reachable).
856 * This should be called when recovery has been completed and quota should now
857 * be enforced on every operations.
859 * \param env - the environment passed by the caller
860 * \param qsd - is the qsd instance associated with the osd device to start
862 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
867 if (unlikely(qsd == NULL))
870 write_lock(&qsd->qsd_lock);
871 if (!qsd->qsd_prepared) {
872 CERROR("%s: can't start qsd instance since it wasn't properly "
873 "initialized\n", qsd->qsd_svname);
875 } else if (qsd->qsd_started) {
876 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
879 /* notify that the qsd_instance is now started */
880 qsd->qsd_started = true;
882 write_unlock(&qsd->qsd_lock);
887 /* Trigger the 3rd step of reintegration: If usage > granted, acquire
888 * up to usage; If usage < granted, release down to usage. */
889 for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
890 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
891 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
896 EXPORT_SYMBOL(qsd_start);
898 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
901 * Global initialization performed at module load time
903 int qsd_glb_init(void)
907 rc = lu_kmem_init(qsd_caches);
911 qsd_key_init_generic(&qsd_thread_key, NULL);
912 lu_context_key_register(&qsd_thread_key);
913 lustre_register_quota_process_config(qsd_process_config);
919 * Companion of qsd_glb_init() called at module unload time
921 void qsd_glb_fini(void)
923 lustre_register_quota_process_config(NULL);
924 lu_kmem_fini(qsd_caches);
925 lu_context_key_degister(&qsd_thread_key);