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, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mgs/mgs_nids.c
38 * NID table management for lustre.
40 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_MGS
44 #define D_MGS D_CONFIG
47 #include <linux/pagemap.h>
51 #include <obd_class.h>
52 #include <lustre_disk.h>
54 #include "mgs_internal.h"
56 static unsigned int ir_timeout;
58 static int nidtbl_is_sane(struct mgs_nidtbl *tbl)
60 struct mgs_nidtbl_target *tgt;
63 LASSERT(mutex_is_locked(&tbl->mn_lock));
64 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
65 if (!tgt->mnt_version)
68 if (version >= tgt->mnt_version)
71 version = tgt->mnt_version;
77 * Fetch nidtbl entries whose version are not less than @version
78 * nidtbl entries will be packed in @pages by @unit_size units - entries
79 * shouldn't cross unit boundaries.
81 static int mgs_nidtbl_read(struct obd_export *exp, struct mgs_nidtbl *tbl,
82 struct mgs_config_res *res, struct page **pages,
83 int nrpages, int units_total, int unit_size)
85 struct mgs_nidtbl_target *tgt;
86 struct mgs_nidtbl_entry *entry;
87 struct mgs_nidtbl_entry *last_in_unit = NULL;
88 struct mgs_target_info *mti;
89 __u64 version = res->mcr_offset;
92 int bytes_in_unit = 0;
93 int units_in_page = 0;
98 /* make sure unit_size is power 2 */
99 LASSERT((unit_size & (unit_size - 1)) == 0);
100 LASSERT(nrpages << PAGE_CACHE_SHIFT >= units_total * unit_size);
102 mutex_lock(&tbl->mn_lock);
103 LASSERT(nidtbl_is_sane(tbl));
105 /* no more entries ? */
106 if (version > tbl->mn_version) {
107 version = tbl->mn_version;
111 /* iterate over all targets to compose a bitmap by the type of llog.
112 * If the llog is for MDTs, llog entries for OSTs will be returned;
113 * otherwise, it's for clients, then llog entries for both OSTs and
114 * MDTs will be returned.
116 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
117 int entry_len = sizeof(*entry);
119 if (tgt->mnt_version < version)
122 /* write target recover information */
124 LASSERT(mti->mti_nid_count < MTI_NIDS_MAX);
125 entry_len += mti->mti_nid_count * sizeof(lnet_nid_t);
127 if (entry_len > unit_size) {
128 CWARN("nidtbl: too large entry: entry length %d,"
129 "unit size: %d\n", entry_len, unit_size);
130 GOTO(out, rc = -EOVERFLOW);
133 if (bytes_in_unit < entry_len) {
134 if (units_total == 0) {
139 /* check if we need to consume remaining bytes. */
140 if (last_in_unit != NULL && bytes_in_unit) {
141 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0)
142 /* May need to swab back to update the length.*/
143 if (exp->exp_need_mne_swab)
144 lustre_swab_mgs_nidtbl_entry(last_in_unit);
146 last_in_unit->mne_length += bytes_in_unit;
147 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0)
148 if (exp->exp_need_mne_swab)
149 lustre_swab_mgs_nidtbl_entry(last_in_unit);
152 buf += bytes_in_unit;
155 LASSERT((rc & (unit_size - 1)) == 0);
157 if (units_in_page == 0) {
158 /* allocate a new page */
159 pages[index] = alloc_page(GFP_IOFS);
160 if (pages[index] == NULL) {
165 /* destroy previous map */
167 kunmap(pages[index - 1]);
169 /* reassign buffer */
170 buf = kmap(pages[index]);
173 units_in_page = PAGE_CACHE_SIZE / unit_size;
174 LASSERT(units_in_page > 0);
177 /* allocate an unit */
178 LASSERT(((long)buf & (unit_size - 1)) == 0);
179 bytes_in_unit = unit_size;
185 entry = (struct mgs_nidtbl_entry *)buf;
186 entry->mne_version = tgt->mnt_version;
187 entry->mne_instance = mti->mti_instance;
188 entry->mne_index = mti->mti_stripe_index;
189 entry->mne_length = entry_len;
190 entry->mne_type = tgt->mnt_type;
191 entry->mne_nid_type = 0;
192 entry->mne_nid_size = sizeof(lnet_nid_t);
193 entry->mne_nid_count = mti->mti_nid_count;
194 memcpy(entry->u.nids, mti->mti_nids,
195 mti->mti_nid_count * sizeof(lnet_nid_t));
197 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0)
198 /* For LU-1644, swab entry for 2.2 clients. */
199 if (exp->exp_need_mne_swab)
200 lustre_swab_mgs_nidtbl_entry(entry);
203 version = tgt->mnt_version;
207 bytes_in_unit -= entry_len;
208 last_in_unit = entry;
210 CDEBUG(D_MGS, "fsname %s, entry size %d, pages %d/%d/%d/%d.\n",
211 tbl->mn_fsdb->fsdb_name, entry_len,
212 bytes_in_unit, index, nrpages, units_total);
215 kunmap(pages[index - 1]);
217 LASSERT(version <= tbl->mn_version);
218 res->mcr_size = tbl->mn_version;
219 res->mcr_offset = nobuf ? version : tbl->mn_version;
220 mutex_unlock(&tbl->mn_lock);
221 LASSERT(ergo(version == 1, rc == 0)); /* get the log first time */
223 CDEBUG(D_MGS, "Read IR logs %s return with %d, version %llu\n",
224 tbl->mn_fsdb->fsdb_name, rc, version);
228 static int nidtbl_update_version(const struct lu_env *env,
229 struct mgs_device *mgs,
230 struct mgs_nidtbl *tbl)
232 struct dt_object *fsdb;
235 struct lu_buf buf = {
237 .lb_len = sizeof(version)
243 LASSERT(mutex_is_locked(&tbl->mn_lock));
245 fsdb = local_file_find_or_create(env, mgs->mgs_los, mgs->mgs_nidtbl_dir,
246 tbl->mn_fsdb->fsdb_name,
247 S_IFREG | S_IRUGO | S_IWUSR);
249 RETURN(PTR_ERR(fsdb));
251 th = dt_trans_create(env, mgs->mgs_bottom);
253 GOTO(out_put, rc = PTR_ERR(th));
255 th->th_sync = 1; /* update table synchronously */
256 rc = dt_declare_record_write(env, fsdb, &buf, off, th);
260 rc = dt_trans_start_local(env, mgs->mgs_bottom, th);
264 version = cpu_to_le64(tbl->mn_version);
265 rc = dt_record_write(env, fsdb, &buf, &off, th);
268 dt_trans_stop(env, mgs->mgs_bottom, th);
270 lu_object_put(env, &fsdb->do_lu);
274 #define MGS_NIDTBL_VERSION_INIT 2
276 static int nidtbl_read_version(const struct lu_env *env,
277 struct mgs_device *mgs, struct mgs_nidtbl *tbl,
280 struct dt_object *fsdb;
283 struct lu_buf buf = {
285 .lb_len = sizeof(tmpver)
291 LASSERT(mutex_is_locked(&tbl->mn_lock));
293 LASSERT(mgs->mgs_nidtbl_dir);
294 rc = dt_lookup_dir(env, mgs->mgs_nidtbl_dir, tbl->mn_fsdb->fsdb_name,
297 *version = MGS_NIDTBL_VERSION_INIT;
303 fsdb = dt_locate_at(env, mgs->mgs_bottom, &fid,
304 &mgs->mgs_dt_dev.dd_lu_dev, NULL);
306 RETURN(PTR_ERR(fsdb));
308 rc = dt_read(env, fsdb, &buf, &off);
309 if (rc == buf.lb_len) {
310 *version = le64_to_cpu(tmpver);
312 } else if (rc == 0) {
313 *version = MGS_NIDTBL_VERSION_INIT;
315 CERROR("%s: read version file %s error %d\n",
316 mgs->mgs_obd->obd_name, tbl->mn_fsdb->fsdb_name, rc);
318 lu_object_put(env, &fsdb->do_lu);
322 static int mgs_nidtbl_write(const struct lu_env *env, struct fs_db *fsdb,
323 struct mgs_target_info *mti)
325 struct mgs_nidtbl *tbl;
326 struct mgs_nidtbl_target *tgt;
328 int type = mti->mti_flags & LDD_F_SV_TYPE_MASK;
332 type &= ~LDD_F_SV_TYPE_MGS;
335 tbl = &fsdb->fsdb_nidtbl;
336 mutex_lock(&tbl->mn_lock);
337 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
338 struct mgs_target_info *info = &tgt->mnt_mti;
340 if (type == tgt->mnt_type &&
341 mti->mti_stripe_index == info->mti_stripe_index) {
349 GOTO(out, rc = -ENOMEM);
351 INIT_LIST_HEAD(&tgt->mnt_list);
353 tgt->mnt_version = 0; /* 0 means invalid */
354 tgt->mnt_type = type;
356 ++tbl->mn_nr_targets;
359 tgt->mnt_version = ++tbl->mn_version;
362 list_move_tail(&tgt->mnt_list, &tbl->mn_targets);
364 rc = nidtbl_update_version(env, fsdb->fsdb_mgs, tbl);
368 mutex_unlock(&tbl->mn_lock);
370 CERROR("Write NID table version for file system %s error %d\n",
371 fsdb->fsdb_name, rc);
375 static void mgs_nidtbl_fini_fs(struct fs_db *fsdb)
377 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
378 struct list_head head = LIST_HEAD_INIT(head);
380 mutex_lock(&tbl->mn_lock);
381 tbl->mn_nr_targets = 0;
382 list_splice_init(&tbl->mn_targets, &head);
383 mutex_unlock(&tbl->mn_lock);
385 while (!list_empty(&head)) {
386 struct mgs_nidtbl_target *tgt;
387 tgt = list_entry(head.next, struct mgs_nidtbl_target, mnt_list);
388 list_del(&tgt->mnt_list);
393 static int mgs_nidtbl_init_fs(const struct lu_env *env, struct fs_db *fsdb)
395 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
398 INIT_LIST_HEAD(&tbl->mn_targets);
399 mutex_init(&tbl->mn_lock);
400 tbl->mn_nr_targets = 0;
402 mutex_lock(&tbl->mn_lock);
403 rc = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl, &tbl->mn_version);
404 mutex_unlock(&tbl->mn_lock);
406 CERROR("%s: IR: failed to read current version, rc = %d\n",
407 fsdb->fsdb_mgs->mgs_obd->obd_name, rc);
409 CDEBUG(D_MGS, "IR: current version is %llu\n",
415 /* --------- Imperative Recovery relies on nidtbl stuff ------- */
416 void mgs_ir_notify_complete(struct fs_db *fsdb)
419 cfs_duration_t delta;
421 atomic_set(&fsdb->fsdb_notify_phase, 0);
424 fsdb->fsdb_notify_count++;
425 delta = cfs_time_sub(cfs_time_current(), fsdb->fsdb_notify_start);
426 fsdb->fsdb_notify_total += delta;
427 if (delta > fsdb->fsdb_notify_max)
428 fsdb->fsdb_notify_max = delta;
430 cfs_duration_usec(delta, &tv);
431 CDEBUG(D_MGS, "Revoke recover lock of %s completed after %ld.%06lds\n",
432 fsdb->fsdb_name, tv.tv_sec, tv.tv_usec);
435 static int mgs_ir_notify(void *arg)
437 struct fs_db *fsdb = arg;
438 struct ldlm_res_id resid;
440 char name[sizeof(fsdb->fsdb_name) + 20];
442 LASSERTF(sizeof(name) < 32, "name is too large to be in stack.\n");
443 sprintf(name, "mgs_%s_notify", fsdb->fsdb_name);
445 complete(&fsdb->fsdb_notify_comp);
447 set_user_nice(current, -2);
449 mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER);
451 struct l_wait_info lwi = { 0 };
453 l_wait_event(fsdb->fsdb_notify_waitq,
454 fsdb->fsdb_notify_stop ||
455 atomic_read(&fsdb->fsdb_notify_phase),
457 if (fsdb->fsdb_notify_stop)
460 CDEBUG(D_MGS, "%s woken up, phase is %d\n",
461 name, atomic_read(&fsdb->fsdb_notify_phase));
463 fsdb->fsdb_notify_start = cfs_time_current();
464 mgs_revoke_lock(fsdb->fsdb_mgs, fsdb, CONFIG_T_RECOVER);
467 complete(&fsdb->fsdb_notify_comp);
471 int mgs_ir_init_fs(const struct lu_env *env, struct mgs_device *mgs,
474 struct task_struct *task;
477 ir_timeout = OBD_IR_MGS_TIMEOUT;
479 fsdb->fsdb_ir_state = IR_FULL;
480 if (cfs_time_before(cfs_time_current_sec(),
481 mgs->mgs_start_time + ir_timeout))
482 fsdb->fsdb_ir_state = IR_STARTUP;
483 fsdb->fsdb_nonir_clients = 0;
484 INIT_LIST_HEAD(&fsdb->fsdb_clients);
486 /* start notify thread */
487 fsdb->fsdb_mgs = mgs;
488 atomic_set(&fsdb->fsdb_notify_phase, 0);
489 init_waitqueue_head(&fsdb->fsdb_notify_waitq);
490 init_completion(&fsdb->fsdb_notify_comp);
492 task = kthread_run(mgs_ir_notify, fsdb,
493 "mgs_%s_notify", fsdb->fsdb_name);
495 wait_for_completion(&fsdb->fsdb_notify_comp);
497 CERROR("Start notify thread error %ld\n", PTR_ERR(task));
499 mgs_nidtbl_init_fs(env, fsdb);
503 void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb)
505 if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
508 mgs_fsc_cleanup_by_fsdb(fsdb);
510 mgs_nidtbl_fini_fs(fsdb);
512 LASSERT(list_empty(&fsdb->fsdb_clients));
514 fsdb->fsdb_notify_stop = 1;
515 wake_up(&fsdb->fsdb_notify_waitq);
516 wait_for_completion(&fsdb->fsdb_notify_comp);
519 /* caller must have held fsdb_mutex */
520 static inline void ir_state_graduate(struct fs_db *fsdb)
522 if (fsdb->fsdb_ir_state == IR_STARTUP) {
523 if (cfs_time_before(fsdb->fsdb_mgs->mgs_start_time + ir_timeout,
524 cfs_time_current_sec())) {
525 fsdb->fsdb_ir_state = IR_FULL;
526 if (fsdb->fsdb_nonir_clients)
527 fsdb->fsdb_ir_state = IR_PARTIAL;
532 int mgs_ir_update(const struct lu_env *env, struct mgs_device *mgs,
533 struct mgs_target_info *mti)
539 if (mti->mti_instance == 0)
542 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
546 rc = mgs_nidtbl_write(env, fsdb, mti);
551 mutex_lock(&fsdb->fsdb_mutex);
552 ir_state_graduate(fsdb);
553 switch (fsdb->fsdb_ir_state) {
555 mti->mti_flags |= LDD_F_IR_CAPABLE;
565 mutex_unlock(&fsdb->fsdb_mutex);
567 LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
569 CDEBUG(D_MGS, "Try to revoke recover lock of %s\n",
571 atomic_inc(&fsdb->fsdb_notify_phase);
572 wake_up(&fsdb->fsdb_notify_waitq);
577 /* NID table can be cached by two entities: Clients and MDTs */
583 static int delogname(char *logname, char *fsname, int *typ)
589 ptr = strrchr(logname, '-');
593 /* decouple file system name. The llog name may be:
594 * - "prefix-fsname", prefix is "cliir" or "mdtir"
596 if (strncmp(ptr, "-mdtir", 6) == 0)
598 else if (strncmp(ptr, "-cliir", 6) == 0)
607 memcpy(fsname, logname, len);
614 int mgs_get_ir_logs(struct ptlrpc_request *req)
616 struct lu_env *env = req->rq_svc_thread->t_env;
617 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
619 struct mgs_config_body *body;
620 struct mgs_config_res *res;
621 struct ptlrpc_bulk_desc *desc;
622 struct l_wait_info lwi;
632 struct page **pages = NULL;
635 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
639 if (body->mcb_type != CONFIG_T_RECOVER)
642 rc = delogname(body->mcb_name, fsname, &type);
646 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
650 bufsize = body->mcb_units << body->mcb_bits;
651 nrpages = (bufsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
652 if (nrpages > PTLRPC_MAX_BRW_PAGES)
655 CDEBUG(D_MGS, "Reading IR log %s bufsize %ld.\n",
656 body->mcb_name, bufsize);
658 OBD_ALLOC(pages, sizeof(*pages) * nrpages);
662 res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
664 GOTO(out, rc = -EINVAL);
666 res->mcr_offset = body->mcb_offset;
667 unit_size = min_t(int, 1 << body->mcb_bits, PAGE_CACHE_SIZE);
668 bytes = mgs_nidtbl_read(req->rq_export, &fsdb->fsdb_nidtbl, res,
669 pages, nrpages, bufsize / unit_size, unit_size);
671 GOTO(out, rc = bytes);
673 /* start bulk transfer */
674 page_count = (bytes + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
675 LASSERT(page_count <= nrpages);
676 desc = ptlrpc_prep_bulk_exp(req, page_count, 1,
677 BULK_PUT_SOURCE, MGS_BULK_PORTAL);
679 GOTO(out, rc = -ENOMEM);
681 for (i = 0; i < page_count && bytes > 0; i++) {
682 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0,
683 min_t(int, bytes, PAGE_CACHE_SIZE));
684 bytes -= PAGE_CACHE_SIZE;
687 rc = target_bulk_io(req->rq_export, desc, &lwi);
688 ptlrpc_free_bulk_pin(desc);
691 for (i = 0; i < nrpages; i++) {
692 if (pages[i] == NULL)
694 __free_page(pages[i]);
696 OBD_FREE(pages, sizeof(*pages) * nrpages);
700 static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf)
702 const char *strings[] = IR_STRINGS;
706 for (i = 0; i < ARRAY_SIZE(strings); i++) {
707 if (strcmp(strings[i], buf) == 0) {
715 CDEBUG(D_MGS, "change fsr state of %s from %s to %s\n",
716 fsdb->fsdb_name, strings[fsdb->fsdb_ir_state], strings[state]);
717 mutex_lock(&fsdb->fsdb_mutex);
718 if (state == IR_FULL && fsdb->fsdb_nonir_clients)
720 fsdb->fsdb_ir_state = state;
721 mutex_unlock(&fsdb->fsdb_mutex);
726 static int lprocfs_ir_set_timeout(struct fs_db *fsdb, const char *buf)
731 static int lprocfs_ir_clear_stats(struct fs_db *fsdb, const char *buf)
736 fsdb->fsdb_notify_total = 0;
737 fsdb->fsdb_notify_max = 0;
738 fsdb->fsdb_notify_count = 0;
742 static struct lproc_ir_cmd {
745 int (*handler)(struct fs_db *, const char *);
747 { "state=", 6, lprocfs_ir_set_state },
748 { "timeout=", 8, lprocfs_ir_set_timeout },
749 { "0", 1, lprocfs_ir_clear_stats }
752 int lprocfs_wr_ir_state(struct file *file, const char __user *buffer,
753 size_t count, void *data)
755 struct fs_db *fsdb = data;
760 if (count == 0 || count >= PAGE_CACHE_SIZE)
763 OBD_ALLOC(kbuf, count + 1);
767 if (copy_from_user(kbuf, buffer, count)) {
768 OBD_FREE(kbuf, count + 1);
772 kbuf[count] = 0; /* buffer is supposed to end with 0 */
773 if (kbuf[count - 1] == '\n')
777 /* fsname=<file system name> must be the 1st entry */
778 while (ptr != NULL) {
782 tmpptr = strchr(ptr, ';');
787 for (i = 0; i < ARRAY_SIZE(ir_cmds); i++) {
788 struct lproc_ir_cmd *cmd;
792 cmdlen = cmd->namelen;
793 if (strncmp(cmd->name, ptr, cmdlen) == 0) {
795 rc = cmd->handler(fsdb, ptr);
805 CERROR("Unable to process command: %s(%d)\n", ptr, rc);
806 OBD_FREE(kbuf, count + 1);
810 int lprocfs_rd_ir_state(struct seq_file *seq, void *data)
812 struct fs_db *fsdb = data;
813 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
814 const char *ir_strings[] = IR_STRINGS;
815 struct timeval tv_max;
818 /* mgs_live_seq_show() already holds fsdb_mutex. */
819 ir_state_graduate(fsdb);
821 seq_printf(seq, "\nimperative_recovery_state:\n");
824 " nonir_clients: %d\n"
825 " nidtbl_version: %lld\n",
826 ir_strings[fsdb->fsdb_ir_state], fsdb->fsdb_nonir_clients,
829 cfs_duration_usec(fsdb->fsdb_notify_total, &tv);
830 cfs_duration_usec(fsdb->fsdb_notify_max, &tv_max);
832 seq_printf(seq, " notify_duration_total: %lu.%06lu\n"
833 " notify_duation_max: %lu.%06lu\n"
834 " notify_count: %u\n",
835 tv.tv_sec, tv.tv_usec,
836 tv_max.tv_sec, tv_max.tv_usec,
837 fsdb->fsdb_notify_count);
842 int lprocfs_ir_timeout_seq_show(struct seq_file *m, void *data)
844 return lprocfs_uint_seq_show(m, &ir_timeout);
847 ssize_t lprocfs_ir_timeout_seq_write(struct file *file,
848 const char __user *buffer,
849 size_t count, loff_t *off)
851 return lprocfs_wr_uint(file, buffer, count, &ir_timeout);
854 /* --------------- Handle non IR support clients --------------- */
855 /* attach a lustre file system to an export */
856 int mgs_fsc_attach(const struct lu_env *env, struct obd_export *exp,
859 struct mgs_export_data *data = &exp->u.eu_mgs_data;
860 struct mgs_device *mgs = exp2mgs_dev(exp);
862 struct mgs_fsc *fsc = NULL;
863 struct mgs_fsc *new_fsc = NULL;
868 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
872 /* allocate a new fsc in case we need it in spinlock. */
873 OBD_ALLOC_PTR(new_fsc);
877 INIT_LIST_HEAD(&new_fsc->mfc_export_list);
878 INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list);
879 new_fsc->mfc_fsdb = fsdb;
880 new_fsc->mfc_export = class_export_get(exp);
881 new_fsc->mfc_ir_capable = !!(exp_connect_flags(exp) &
882 OBD_CONNECT_IMP_RECOV);
885 mutex_lock(&fsdb->fsdb_mutex);
887 /* tend to find it in export list because this list is shorter. */
888 spin_lock(&data->med_lock);
889 list_for_each_entry(fsc, &data->med_clients, mfc_export_list) {
890 if (strcmp(fsname, fsc->mfc_fsdb->fsdb_name) == 0) {
899 /* add it into export list. */
900 list_add(&fsc->mfc_export_list, &data->med_clients);
902 /* add into fsdb list. */
903 list_add(&fsc->mfc_fsdb_list, &fsdb->fsdb_clients);
904 if (!fsc->mfc_ir_capable) {
905 ++fsdb->fsdb_nonir_clients;
906 if (fsdb->fsdb_ir_state == IR_FULL)
907 fsdb->fsdb_ir_state = IR_PARTIAL;
911 spin_unlock(&data->med_lock);
912 mutex_unlock(&fsdb->fsdb_mutex);
915 class_export_put(new_fsc->mfc_export);
916 OBD_FREE_PTR(new_fsc);
921 void mgs_fsc_cleanup(struct obd_export *exp)
923 struct mgs_export_data *data = &exp->u.eu_mgs_data;
924 struct mgs_fsc *fsc, *tmp;
925 struct list_head head = LIST_HEAD_INIT(head);
927 spin_lock(&data->med_lock);
928 list_splice_init(&data->med_clients, &head);
929 spin_unlock(&data->med_lock);
931 list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) {
932 struct fs_db *fsdb = fsc->mfc_fsdb;
934 LASSERT(fsc->mfc_export == exp);
936 mutex_lock(&fsdb->fsdb_mutex);
937 list_del_init(&fsc->mfc_fsdb_list);
938 if (fsc->mfc_ir_capable == 0) {
939 --fsdb->fsdb_nonir_clients;
940 LASSERT(fsdb->fsdb_ir_state != IR_FULL);
941 if (fsdb->fsdb_nonir_clients == 0 &&
942 fsdb->fsdb_ir_state == IR_PARTIAL)
943 fsdb->fsdb_ir_state = IR_FULL;
945 mutex_unlock(&fsdb->fsdb_mutex);
946 list_del_init(&fsc->mfc_export_list);
947 class_export_put(fsc->mfc_export);
952 /* must be called with fsdb->fsdb_mutex held */
953 void mgs_fsc_cleanup_by_fsdb(struct fs_db *fsdb)
955 struct mgs_fsc *fsc, *tmp;
957 list_for_each_entry_safe(fsc, tmp, &fsdb->fsdb_clients,
959 struct mgs_export_data *data = &fsc->mfc_export->u.eu_mgs_data;
961 LASSERT(fsdb == fsc->mfc_fsdb);
962 list_del_init(&fsc->mfc_fsdb_list);
964 spin_lock(&data->med_lock);
965 list_del_init(&fsc->mfc_export_list);
966 spin_unlock(&data->med_lock);
967 class_export_put(fsc->mfc_export);
971 fsdb->fsdb_nonir_clients = 0;
972 if (fsdb->fsdb_ir_state == IR_PARTIAL)
973 fsdb->fsdb_ir_state = IR_FULL;