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
46 #include <linux/pagemap.h>
49 #include <obd_class.h>
50 #include <lustre_disk.h>
52 #include "mgs_internal.h"
54 static unsigned int ir_timeout;
56 static int nidtbl_is_sane(struct mgs_nidtbl *tbl)
58 struct mgs_nidtbl_target *tgt;
61 LASSERT(mutex_is_locked(&tbl->mn_lock));
62 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
63 if (!tgt->mnt_version)
66 if (version >= tgt->mnt_version)
69 version = tgt->mnt_version;
75 * Fetch nidtbl entries whose version are not less than @version
76 * nidtbl entries will be packed in @pages by @unit_size units - entries
77 * shouldn't cross unit boundaries.
79 static int mgs_nidtbl_read(struct obd_export *exp, struct mgs_nidtbl *tbl,
80 struct mgs_config_res *res, struct page **pages,
81 int nrpages, int units_total, int unit_size)
83 struct mgs_nidtbl_target *tgt;
84 struct mgs_nidtbl_entry *entry;
85 struct mgs_nidtbl_entry *last_in_unit = NULL;
86 struct mgs_target_info *mti;
87 __u64 version = res->mcr_offset;
90 int bytes_in_unit = 0;
91 int units_in_page = 0;
96 /* make sure unit_size is power 2 */
97 LASSERT((unit_size & (unit_size - 1)) == 0);
98 LASSERT(nrpages << PAGE_CACHE_SHIFT >= units_total * unit_size);
100 mutex_lock(&tbl->mn_lock);
101 LASSERT(nidtbl_is_sane(tbl));
103 /* no more entries ? */
104 if (version > tbl->mn_version) {
105 version = tbl->mn_version;
109 /* iterate over all targets to compose a bitmap by the type of llog.
110 * If the llog is for MDTs, llog entries for OSTs will be returned;
111 * otherwise, it's for clients, then llog entries for both OSTs and
112 * MDTs will be returned.
114 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
115 int entry_len = sizeof(*entry);
117 if (tgt->mnt_version < version)
120 /* write target recover information */
122 LASSERT(mti->mti_nid_count < MTI_NIDS_MAX);
123 entry_len += mti->mti_nid_count * sizeof(lnet_nid_t);
125 if (entry_len > unit_size) {
126 CWARN("nidtbl: too large entry: entry length %d,"
127 "unit size: %d\n", entry_len, unit_size);
128 GOTO(out, rc = -EOVERFLOW);
131 if (bytes_in_unit < entry_len) {
132 if (units_total == 0) {
137 /* check if we need to consume remaining bytes. */
138 if (last_in_unit != NULL && bytes_in_unit) {
139 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 53, 0)
140 /* May need to swab back to update the length.*/
141 if (exp->exp_need_mne_swab)
142 lustre_swab_mgs_nidtbl_entry(last_in_unit);
144 last_in_unit->mne_length += bytes_in_unit;
145 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 53, 0)
146 if (exp->exp_need_mne_swab)
147 lustre_swab_mgs_nidtbl_entry(last_in_unit);
150 buf += bytes_in_unit;
153 LASSERT((rc & (unit_size - 1)) == 0);
155 if (units_in_page == 0) {
156 /* allocate a new page */
157 pages[index] = alloc_page(GFP_IOFS);
158 if (pages[index] == NULL) {
163 /* destroy previous map */
165 kunmap(pages[index - 1]);
167 /* reassign buffer */
168 buf = kmap(pages[index]);
171 units_in_page = PAGE_CACHE_SIZE / unit_size;
172 LASSERT(units_in_page > 0);
175 /* allocate an unit */
176 LASSERT(((long)buf & (unit_size - 1)) == 0);
177 bytes_in_unit = unit_size;
183 entry = (struct mgs_nidtbl_entry *)buf;
184 entry->mne_version = tgt->mnt_version;
185 entry->mne_instance = mti->mti_instance;
186 entry->mne_index = mti->mti_stripe_index;
187 entry->mne_length = entry_len;
188 entry->mne_type = tgt->mnt_type;
189 entry->mne_nid_type = 0;
190 entry->mne_nid_size = sizeof(lnet_nid_t);
191 entry->mne_nid_count = mti->mti_nid_count;
192 memcpy(entry->u.nids, mti->mti_nids,
193 mti->mti_nid_count * sizeof(lnet_nid_t));
195 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 53, 0)
196 /* For LU-1644, swab entry for 2.2 clients. */
197 if (exp->exp_need_mne_swab)
198 lustre_swab_mgs_nidtbl_entry(entry);
201 version = tgt->mnt_version;
205 bytes_in_unit -= entry_len;
206 last_in_unit = entry;
208 CDEBUG(D_MGS, "fsname %s, entry size %d, pages %d/%d/%d/%d.\n",
209 tbl->mn_fsdb->fsdb_name, entry_len,
210 bytes_in_unit, index, nrpages, units_total);
213 kunmap(pages[index - 1]);
215 LASSERT(version <= tbl->mn_version);
216 res->mcr_size = tbl->mn_version;
217 res->mcr_offset = nobuf ? version : tbl->mn_version;
218 mutex_unlock(&tbl->mn_lock);
219 LASSERT(ergo(version == 1, rc == 0)); /* get the log first time */
221 CDEBUG(D_MGS, "Read IR logs %s return with %d, version %llu\n",
222 tbl->mn_fsdb->fsdb_name, rc, version);
226 static int nidtbl_update_version(const struct lu_env *env,
227 struct mgs_device *mgs,
228 struct mgs_nidtbl *tbl)
230 struct dt_object *fsdb;
233 struct lu_buf buf = {
235 .lb_len = sizeof(version)
241 LASSERT(mutex_is_locked(&tbl->mn_lock));
243 fsdb = local_file_find_or_create(env, mgs->mgs_los, mgs->mgs_nidtbl_dir,
244 tbl->mn_fsdb->fsdb_name,
245 S_IFREG | S_IRUGO | S_IWUSR);
247 RETURN(PTR_ERR(fsdb));
249 th = dt_trans_create(env, mgs->mgs_bottom);
251 GOTO(out_put, rc = PTR_ERR(th));
253 th->th_sync = 1; /* update table synchronously */
254 rc = dt_declare_record_write(env, fsdb, &buf, off, th);
258 rc = dt_trans_start_local(env, mgs->mgs_bottom, th);
262 version = cpu_to_le64(tbl->mn_version);
263 rc = dt_record_write(env, fsdb, &buf, &off, th);
266 dt_trans_stop(env, mgs->mgs_bottom, th);
268 lu_object_put(env, &fsdb->do_lu);
272 #define MGS_NIDTBL_VERSION_INIT 2
274 static int nidtbl_read_version(const struct lu_env *env,
275 struct mgs_device *mgs, struct mgs_nidtbl *tbl,
278 struct dt_object *fsdb;
281 struct lu_buf buf = {
283 .lb_len = sizeof(tmpver)
289 LASSERT(mutex_is_locked(&tbl->mn_lock));
291 LASSERT(mgs->mgs_nidtbl_dir);
292 rc = dt_lookup_dir(env, mgs->mgs_nidtbl_dir, tbl->mn_fsdb->fsdb_name,
295 *version = MGS_NIDTBL_VERSION_INIT;
301 fsdb = dt_locate_at(env, mgs->mgs_bottom, &fid,
302 &mgs->mgs_dt_dev.dd_lu_dev, NULL);
304 RETURN(PTR_ERR(fsdb));
306 rc = dt_read(env, fsdb, &buf, &off);
307 if (rc == buf.lb_len) {
308 *version = le64_to_cpu(tmpver);
310 } else if (rc == 0) {
311 *version = MGS_NIDTBL_VERSION_INIT;
313 CERROR("%s: read version file %s error %d\n",
314 mgs->mgs_obd->obd_name, tbl->mn_fsdb->fsdb_name, rc);
316 lu_object_put(env, &fsdb->do_lu);
320 static int mgs_nidtbl_write(const struct lu_env *env, struct fs_db *fsdb,
321 struct mgs_target_info *mti)
323 struct mgs_nidtbl *tbl;
324 struct mgs_nidtbl_target *tgt;
326 int type = mti->mti_flags & LDD_F_SV_TYPE_MASK;
330 type &= ~LDD_F_SV_TYPE_MGS;
333 tbl = &fsdb->fsdb_nidtbl;
334 mutex_lock(&tbl->mn_lock);
335 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
336 struct mgs_target_info *info = &tgt->mnt_mti;
338 if (type == tgt->mnt_type &&
339 mti->mti_stripe_index == info->mti_stripe_index) {
347 GOTO(out, rc = -ENOMEM);
349 INIT_LIST_HEAD(&tgt->mnt_list);
351 tgt->mnt_version = 0; /* 0 means invalid */
352 tgt->mnt_type = type;
354 ++tbl->mn_nr_targets;
357 tgt->mnt_version = ++tbl->mn_version;
360 list_move_tail(&tgt->mnt_list, &tbl->mn_targets);
362 rc = nidtbl_update_version(env, fsdb->fsdb_mgs, tbl);
366 mutex_unlock(&tbl->mn_lock);
368 CERROR("Write NID table version for file system %s error %d\n",
369 fsdb->fsdb_name, rc);
373 static void mgs_nidtbl_fini_fs(struct fs_db *fsdb)
375 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
376 struct list_head head = LIST_HEAD_INIT(head);
378 mutex_lock(&tbl->mn_lock);
379 tbl->mn_nr_targets = 0;
380 list_splice_init(&tbl->mn_targets, &head);
381 mutex_unlock(&tbl->mn_lock);
383 while (!list_empty(&head)) {
384 struct mgs_nidtbl_target *tgt;
385 tgt = list_entry(head.next, struct mgs_nidtbl_target, mnt_list);
386 list_del(&tgt->mnt_list);
391 static int mgs_nidtbl_init_fs(const struct lu_env *env, struct fs_db *fsdb)
393 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
396 INIT_LIST_HEAD(&tbl->mn_targets);
397 mutex_init(&tbl->mn_lock);
398 tbl->mn_nr_targets = 0;
400 mutex_lock(&tbl->mn_lock);
401 rc = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl, &tbl->mn_version);
402 mutex_unlock(&tbl->mn_lock);
404 CERROR("%s: IR: failed to read current version, rc = %d\n",
405 fsdb->fsdb_mgs->mgs_obd->obd_name, rc);
407 CDEBUG(D_MGS, "IR: current version is %llu\n",
413 /* --------- Imperative Recovery relies on nidtbl stuff ------- */
414 void mgs_ir_notify_complete(struct fs_db *fsdb)
417 cfs_duration_t delta;
419 atomic_set(&fsdb->fsdb_notify_phase, 0);
422 fsdb->fsdb_notify_count++;
423 delta = cfs_time_sub(cfs_time_current(), fsdb->fsdb_notify_start);
424 fsdb->fsdb_notify_total += delta;
425 if (delta > fsdb->fsdb_notify_max)
426 fsdb->fsdb_notify_max = delta;
428 cfs_duration_usec(delta, &tv);
429 CDEBUG(D_MGS, "Revoke recover lock of %s completed after %ld.%06lds\n",
430 fsdb->fsdb_name, tv.tv_sec, tv.tv_usec);
433 static int mgs_ir_notify(void *arg)
435 struct fs_db *fsdb = arg;
436 struct ldlm_res_id resid;
438 char name[sizeof(fsdb->fsdb_name) + 20];
440 LASSERTF(sizeof(name) < 32, "name is too large to be in stack.\n");
441 sprintf(name, "mgs_%s_notify", fsdb->fsdb_name);
443 complete(&fsdb->fsdb_notify_comp);
445 set_user_nice(current, -2);
447 mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER);
449 struct l_wait_info lwi = { 0 };
451 l_wait_event(fsdb->fsdb_notify_waitq,
452 fsdb->fsdb_notify_stop ||
453 atomic_read(&fsdb->fsdb_notify_phase),
455 if (fsdb->fsdb_notify_stop)
458 CDEBUG(D_MGS, "%s woken up, phase is %d\n",
459 name, atomic_read(&fsdb->fsdb_notify_phase));
461 fsdb->fsdb_notify_start = cfs_time_current();
462 mgs_revoke_lock(fsdb->fsdb_mgs, fsdb, CONFIG_T_RECOVER);
465 complete(&fsdb->fsdb_notify_comp);
469 int mgs_ir_init_fs(const struct lu_env *env, struct mgs_device *mgs,
472 struct task_struct *task;
475 ir_timeout = OBD_IR_MGS_TIMEOUT;
477 fsdb->fsdb_ir_state = IR_FULL;
478 if (cfs_time_before(cfs_time_current_sec(),
479 mgs->mgs_start_time + ir_timeout))
480 fsdb->fsdb_ir_state = IR_STARTUP;
481 fsdb->fsdb_nonir_clients = 0;
482 INIT_LIST_HEAD(&fsdb->fsdb_clients);
484 /* start notify thread */
485 fsdb->fsdb_mgs = mgs;
486 atomic_set(&fsdb->fsdb_notify_phase, 0);
487 init_waitqueue_head(&fsdb->fsdb_notify_waitq);
488 init_completion(&fsdb->fsdb_notify_comp);
490 task = kthread_run(mgs_ir_notify, fsdb,
491 "mgs_%s_notify", fsdb->fsdb_name);
493 wait_for_completion(&fsdb->fsdb_notify_comp);
495 CERROR("Start notify thread error %ld\n", PTR_ERR(task));
497 mgs_nidtbl_init_fs(env, fsdb);
501 void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb)
503 if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
506 mgs_fsc_cleanup_by_fsdb(fsdb);
508 mgs_nidtbl_fini_fs(fsdb);
510 LASSERT(list_empty(&fsdb->fsdb_clients));
512 fsdb->fsdb_notify_stop = 1;
513 wake_up(&fsdb->fsdb_notify_waitq);
514 wait_for_completion(&fsdb->fsdb_notify_comp);
517 /* caller must have held fsdb_mutex */
518 static inline void ir_state_graduate(struct fs_db *fsdb)
520 if (fsdb->fsdb_ir_state == IR_STARTUP) {
521 if (cfs_time_before(fsdb->fsdb_mgs->mgs_start_time + ir_timeout,
522 cfs_time_current_sec())) {
523 fsdb->fsdb_ir_state = IR_FULL;
524 if (fsdb->fsdb_nonir_clients)
525 fsdb->fsdb_ir_state = IR_PARTIAL;
530 int mgs_ir_update(const struct lu_env *env, struct mgs_device *mgs,
531 struct mgs_target_info *mti)
537 if (mti->mti_instance == 0)
540 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
544 rc = mgs_nidtbl_write(env, fsdb, mti);
549 mutex_lock(&fsdb->fsdb_mutex);
550 ir_state_graduate(fsdb);
551 switch (fsdb->fsdb_ir_state) {
553 mti->mti_flags |= LDD_F_IR_CAPABLE;
563 mutex_unlock(&fsdb->fsdb_mutex);
565 LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
567 CDEBUG(D_MGS, "Try to revoke recover lock of %s\n",
569 atomic_inc(&fsdb->fsdb_notify_phase);
570 wake_up(&fsdb->fsdb_notify_waitq);
575 /* NID table can be cached by two entities: Clients and MDTs */
581 static int delogname(char *logname, char *fsname, int *typ)
587 ptr = strrchr(logname, '-');
591 /* decouple file system name. The llog name may be:
592 * - "prefix-fsname", prefix is "cliir" or "mdtir"
594 if (strncmp(ptr, "-mdtir", 6) == 0)
596 else if (strncmp(ptr, "-cliir", 6) == 0)
605 memcpy(fsname, logname, len);
612 int mgs_get_ir_logs(struct ptlrpc_request *req)
614 struct lu_env *env = req->rq_svc_thread->t_env;
615 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
617 struct mgs_config_body *body;
618 struct mgs_config_res *res;
619 struct ptlrpc_bulk_desc *desc;
620 struct l_wait_info lwi;
630 struct page **pages = NULL;
633 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
637 if (body->mcb_type != CONFIG_T_RECOVER)
640 rc = delogname(body->mcb_name, fsname, &type);
644 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
648 bufsize = body->mcb_units << body->mcb_bits;
649 nrpages = (bufsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
650 if (nrpages > PTLRPC_MAX_BRW_PAGES)
653 CDEBUG(D_MGS, "Reading IR log %s bufsize %ld.\n",
654 body->mcb_name, bufsize);
656 OBD_ALLOC(pages, sizeof(*pages) * nrpages);
660 res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
662 GOTO(out, rc = -EINVAL);
664 res->mcr_offset = body->mcb_offset;
665 unit_size = min_t(int, 1 << body->mcb_bits, PAGE_CACHE_SIZE);
666 bytes = mgs_nidtbl_read(req->rq_export, &fsdb->fsdb_nidtbl, res,
667 pages, nrpages, bufsize / unit_size, unit_size);
669 GOTO(out, rc = bytes);
671 /* start bulk transfer */
672 page_count = (bytes + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
673 LASSERT(page_count <= nrpages);
674 desc = ptlrpc_prep_bulk_exp(req, page_count, 1,
675 BULK_PUT_SOURCE, MGS_BULK_PORTAL);
677 GOTO(out, rc = -ENOMEM);
679 for (i = 0; i < page_count && bytes > 0; i++) {
680 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0,
681 min_t(int, bytes, PAGE_CACHE_SIZE));
682 bytes -= PAGE_CACHE_SIZE;
685 rc = target_bulk_io(req->rq_export, desc, &lwi);
686 ptlrpc_free_bulk_pin(desc);
689 for (i = 0; i < nrpages; i++) {
690 if (pages[i] == NULL)
692 __free_page(pages[i]);
694 OBD_FREE(pages, sizeof(*pages) * nrpages);
698 static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf)
700 const char *strings[] = IR_STRINGS;
704 for (i = 0; i < ARRAY_SIZE(strings); i++) {
705 if (strcmp(strings[i], buf) == 0) {
713 CDEBUG(D_MGS, "change fsr state of %s from %s to %s\n",
714 fsdb->fsdb_name, strings[fsdb->fsdb_ir_state], strings[state]);
715 mutex_lock(&fsdb->fsdb_mutex);
716 if (state == IR_FULL && fsdb->fsdb_nonir_clients)
718 fsdb->fsdb_ir_state = state;
719 mutex_unlock(&fsdb->fsdb_mutex);
724 static int lprocfs_ir_set_timeout(struct fs_db *fsdb, const char *buf)
729 static int lprocfs_ir_clear_stats(struct fs_db *fsdb, const char *buf)
734 fsdb->fsdb_notify_total = 0;
735 fsdb->fsdb_notify_max = 0;
736 fsdb->fsdb_notify_count = 0;
740 static struct lproc_ir_cmd {
743 int (*handler)(struct fs_db *, const char *);
745 { "state=", 6, lprocfs_ir_set_state },
746 { "timeout=", 8, lprocfs_ir_set_timeout },
747 { "0", 1, lprocfs_ir_clear_stats }
750 int lprocfs_wr_ir_state(struct file *file, const char __user *buffer,
751 size_t count, void *data)
753 struct fs_db *fsdb = data;
758 if (count == 0 || count >= PAGE_CACHE_SIZE)
761 OBD_ALLOC(kbuf, count + 1);
765 if (copy_from_user(kbuf, buffer, count)) {
766 OBD_FREE(kbuf, count + 1);
770 kbuf[count] = 0; /* buffer is supposed to end with 0 */
771 if (kbuf[count - 1] == '\n')
775 /* fsname=<file system name> must be the 1st entry */
776 while (ptr != NULL) {
780 tmpptr = strchr(ptr, ';');
785 for (i = 0; i < ARRAY_SIZE(ir_cmds); i++) {
786 struct lproc_ir_cmd *cmd;
790 cmdlen = cmd->namelen;
791 if (strncmp(cmd->name, ptr, cmdlen) == 0) {
793 rc = cmd->handler(fsdb, ptr);
803 CERROR("Unable to process command: %s(%d)\n", ptr, rc);
804 OBD_FREE(kbuf, count + 1);
808 int lprocfs_rd_ir_state(struct seq_file *seq, void *data)
810 struct fs_db *fsdb = data;
811 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
812 const char *ir_strings[] = IR_STRINGS;
813 struct timeval tv_max;
816 /* mgs_live_seq_show() already holds fsdb_mutex. */
817 ir_state_graduate(fsdb);
819 seq_printf(seq, "\nimperative_recovery_state:\n");
822 " nonir_clients: %d\n"
823 " nidtbl_version: %lld\n",
824 ir_strings[fsdb->fsdb_ir_state], fsdb->fsdb_nonir_clients,
827 cfs_duration_usec(fsdb->fsdb_notify_total, &tv);
828 cfs_duration_usec(fsdb->fsdb_notify_max, &tv_max);
830 seq_printf(seq, " notify_duration_total: %lu.%06lu\n"
831 " notify_duation_max: %lu.%06lu\n"
832 " notify_count: %u\n",
833 tv.tv_sec, tv.tv_usec,
834 tv_max.tv_sec, tv_max.tv_usec,
835 fsdb->fsdb_notify_count);
840 int lprocfs_ir_timeout_seq_show(struct seq_file *m, void *data)
842 return lprocfs_uint_seq_show(m, &ir_timeout);
845 ssize_t lprocfs_ir_timeout_seq_write(struct file *file,
846 const char __user *buffer,
847 size_t count, loff_t *off)
849 return lprocfs_wr_uint(file, buffer, count, &ir_timeout);
852 /* --------------- Handle non IR support clients --------------- */
853 /* attach a lustre file system to an export */
854 int mgs_fsc_attach(const struct lu_env *env, struct obd_export *exp,
857 struct mgs_export_data *data = &exp->u.eu_mgs_data;
858 struct mgs_device *mgs = exp2mgs_dev(exp);
860 struct mgs_fsc *fsc = NULL;
861 struct mgs_fsc *new_fsc = NULL;
866 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
870 /* allocate a new fsc in case we need it in spinlock. */
871 OBD_ALLOC_PTR(new_fsc);
875 INIT_LIST_HEAD(&new_fsc->mfc_export_list);
876 INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list);
877 new_fsc->mfc_fsdb = fsdb;
878 new_fsc->mfc_export = class_export_get(exp);
879 new_fsc->mfc_ir_capable = !!(exp_connect_flags(exp) &
880 OBD_CONNECT_IMP_RECOV);
883 mutex_lock(&fsdb->fsdb_mutex);
885 /* tend to find it in export list because this list is shorter. */
886 spin_lock(&data->med_lock);
887 list_for_each_entry(fsc, &data->med_clients, mfc_export_list) {
888 if (strcmp(fsname, fsc->mfc_fsdb->fsdb_name) == 0) {
897 /* add it into export list. */
898 list_add(&fsc->mfc_export_list, &data->med_clients);
900 /* add into fsdb list. */
901 list_add(&fsc->mfc_fsdb_list, &fsdb->fsdb_clients);
902 if (!fsc->mfc_ir_capable) {
903 ++fsdb->fsdb_nonir_clients;
904 if (fsdb->fsdb_ir_state == IR_FULL)
905 fsdb->fsdb_ir_state = IR_PARTIAL;
909 spin_unlock(&data->med_lock);
910 mutex_unlock(&fsdb->fsdb_mutex);
913 class_export_put(new_fsc->mfc_export);
914 OBD_FREE_PTR(new_fsc);
919 void mgs_fsc_cleanup(struct obd_export *exp)
921 struct mgs_export_data *data = &exp->u.eu_mgs_data;
922 struct mgs_fsc *fsc, *tmp;
923 struct list_head head = LIST_HEAD_INIT(head);
925 spin_lock(&data->med_lock);
926 list_splice_init(&data->med_clients, &head);
927 spin_unlock(&data->med_lock);
929 list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) {
930 struct fs_db *fsdb = fsc->mfc_fsdb;
932 LASSERT(fsc->mfc_export == exp);
934 mutex_lock(&fsdb->fsdb_mutex);
935 list_del_init(&fsc->mfc_fsdb_list);
936 if (fsc->mfc_ir_capable == 0) {
937 --fsdb->fsdb_nonir_clients;
938 LASSERT(fsdb->fsdb_ir_state != IR_FULL);
939 if (fsdb->fsdb_nonir_clients == 0 &&
940 fsdb->fsdb_ir_state == IR_PARTIAL)
941 fsdb->fsdb_ir_state = IR_FULL;
943 mutex_unlock(&fsdb->fsdb_mutex);
944 list_del_init(&fsc->mfc_export_list);
945 class_export_put(fsc->mfc_export);
950 /* must be called with fsdb->fsdb_mutex held */
951 void mgs_fsc_cleanup_by_fsdb(struct fs_db *fsdb)
953 struct mgs_fsc *fsc, *tmp;
955 list_for_each_entry_safe(fsc, tmp, &fsdb->fsdb_clients,
957 struct mgs_export_data *data = &fsc->mfc_export->u.eu_mgs_data;
959 LASSERT(fsdb == fsc->mfc_fsdb);
960 list_del_init(&fsc->mfc_fsdb_list);
962 spin_lock(&data->med_lock);
963 list_del_init(&fsc->mfc_export_list);
964 spin_unlock(&data->med_lock);
965 class_export_put(fsc->mfc_export);
969 fsdb->fsdb_nonir_clients = 0;
970 if (fsdb->fsdb_ir_state == IR_PARTIAL)
971 fsdb->fsdb_ir_state = IR_FULL;