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 last_in_unit->mne_length += bytes_in_unit;
141 buf += bytes_in_unit;
144 LASSERT((rc & (unit_size - 1)) == 0);
146 if (units_in_page == 0) {
147 /* allocate a new page */
148 pages[index] = alloc_page(GFP_IOFS);
149 if (pages[index] == NULL) {
154 /* destroy previous map */
156 kunmap(pages[index - 1]);
158 /* reassign buffer */
159 buf = kmap(pages[index]);
162 units_in_page = PAGE_CACHE_SIZE / unit_size;
163 LASSERT(units_in_page > 0);
166 /* allocate an unit */
167 LASSERT(((long)buf & (unit_size - 1)) == 0);
168 bytes_in_unit = unit_size;
174 entry = (struct mgs_nidtbl_entry *)buf;
175 entry->mne_version = tgt->mnt_version;
176 entry->mne_instance = mti->mti_instance;
177 entry->mne_index = mti->mti_stripe_index;
178 entry->mne_length = entry_len;
179 entry->mne_type = tgt->mnt_type;
180 entry->mne_nid_type = 0;
181 entry->mne_nid_size = sizeof(lnet_nid_t);
182 entry->mne_nid_count = mti->mti_nid_count;
183 memcpy(entry->u.nids, mti->mti_nids,
184 mti->mti_nid_count * sizeof(lnet_nid_t));
186 version = tgt->mnt_version;
190 bytes_in_unit -= entry_len;
191 last_in_unit = entry;
193 CDEBUG(D_MGS, "fsname %s, entry size %d, pages %d/%d/%d/%d.\n",
194 tbl->mn_fsdb->fsdb_name, entry_len,
195 bytes_in_unit, index, nrpages, units_total);
198 kunmap(pages[index - 1]);
200 LASSERT(version <= tbl->mn_version);
201 res->mcr_size = tbl->mn_version;
202 res->mcr_offset = nobuf ? version : tbl->mn_version;
203 mutex_unlock(&tbl->mn_lock);
204 LASSERT(ergo(version == 1, rc == 0)); /* get the log first time */
206 CDEBUG(D_MGS, "Read IR logs %s return with %d, version %llu\n",
207 tbl->mn_fsdb->fsdb_name, rc, version);
211 static int nidtbl_update_version(const struct lu_env *env,
212 struct mgs_device *mgs,
213 struct mgs_nidtbl *tbl)
215 struct dt_object *fsdb;
218 struct lu_buf buf = {
220 .lb_len = sizeof(version)
226 LASSERT(mutex_is_locked(&tbl->mn_lock));
228 fsdb = local_file_find_or_create(env, mgs->mgs_los, mgs->mgs_nidtbl_dir,
229 tbl->mn_fsdb->fsdb_name,
230 S_IFREG | S_IRUGO | S_IWUSR);
232 RETURN(PTR_ERR(fsdb));
234 th = dt_trans_create(env, mgs->mgs_bottom);
236 GOTO(out_put, rc = PTR_ERR(th));
238 th->th_sync = 1; /* update table synchronously */
239 rc = dt_declare_record_write(env, fsdb, &buf, off, th);
243 rc = dt_trans_start_local(env, mgs->mgs_bottom, th);
247 version = cpu_to_le64(tbl->mn_version);
248 rc = dt_record_write(env, fsdb, &buf, &off, th);
251 dt_trans_stop(env, mgs->mgs_bottom, th);
253 lu_object_put(env, &fsdb->do_lu);
257 #define MGS_NIDTBL_VERSION_INIT 2
259 static int nidtbl_read_version(const struct lu_env *env,
260 struct mgs_device *mgs, struct mgs_nidtbl *tbl,
263 struct dt_object *fsdb;
266 struct lu_buf buf = {
268 .lb_len = sizeof(tmpver)
274 LASSERT(mutex_is_locked(&tbl->mn_lock));
276 LASSERT(mgs->mgs_nidtbl_dir);
277 rc = dt_lookup_dir(env, mgs->mgs_nidtbl_dir, tbl->mn_fsdb->fsdb_name,
280 *version = MGS_NIDTBL_VERSION_INIT;
286 fsdb = dt_locate_at(env, mgs->mgs_bottom, &fid,
287 &mgs->mgs_dt_dev.dd_lu_dev, NULL);
289 RETURN(PTR_ERR(fsdb));
291 rc = dt_read(env, fsdb, &buf, &off);
292 if (rc == buf.lb_len) {
293 *version = le64_to_cpu(tmpver);
295 } else if (rc == 0) {
296 *version = MGS_NIDTBL_VERSION_INIT;
298 CERROR("%s: read version file %s error %d\n",
299 mgs->mgs_obd->obd_name, tbl->mn_fsdb->fsdb_name, rc);
301 lu_object_put(env, &fsdb->do_lu);
305 static int mgs_nidtbl_write(const struct lu_env *env, struct fs_db *fsdb,
306 struct mgs_target_info *mti)
308 struct mgs_nidtbl *tbl;
309 struct mgs_nidtbl_target *tgt;
311 int type = mti->mti_flags & LDD_F_SV_TYPE_MASK;
315 type &= ~LDD_F_SV_TYPE_MGS;
318 tbl = &fsdb->fsdb_nidtbl;
319 mutex_lock(&tbl->mn_lock);
320 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
321 struct mgs_target_info *info = &tgt->mnt_mti;
323 if (type == tgt->mnt_type &&
324 mti->mti_stripe_index == info->mti_stripe_index) {
332 GOTO(out, rc = -ENOMEM);
334 INIT_LIST_HEAD(&tgt->mnt_list);
336 tgt->mnt_version = 0; /* 0 means invalid */
337 tgt->mnt_type = type;
339 ++tbl->mn_nr_targets;
342 tgt->mnt_version = ++tbl->mn_version;
345 list_move_tail(&tgt->mnt_list, &tbl->mn_targets);
347 rc = nidtbl_update_version(env, fsdb->fsdb_mgs, tbl);
351 mutex_unlock(&tbl->mn_lock);
353 CERROR("Write NID table version for file system %s error %d\n",
354 fsdb->fsdb_name, rc);
358 static void mgs_nidtbl_fini_fs(struct fs_db *fsdb)
360 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
361 struct list_head head = LIST_HEAD_INIT(head);
363 mutex_lock(&tbl->mn_lock);
364 tbl->mn_nr_targets = 0;
365 list_splice_init(&tbl->mn_targets, &head);
366 mutex_unlock(&tbl->mn_lock);
368 while (!list_empty(&head)) {
369 struct mgs_nidtbl_target *tgt;
370 tgt = list_entry(head.next, struct mgs_nidtbl_target, mnt_list);
371 list_del(&tgt->mnt_list);
376 static int mgs_nidtbl_init_fs(const struct lu_env *env, struct fs_db *fsdb)
378 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
381 INIT_LIST_HEAD(&tbl->mn_targets);
382 mutex_init(&tbl->mn_lock);
383 tbl->mn_nr_targets = 0;
385 mutex_lock(&tbl->mn_lock);
386 rc = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl, &tbl->mn_version);
387 mutex_unlock(&tbl->mn_lock);
389 CERROR("%s: IR: failed to read current version, rc = %d\n",
390 fsdb->fsdb_mgs->mgs_obd->obd_name, rc);
392 CDEBUG(D_MGS, "IR: current version is %llu\n",
398 /* --------- Imperative Recovery relies on nidtbl stuff ------- */
399 void mgs_ir_notify_complete(struct fs_db *fsdb)
402 cfs_duration_t delta;
404 atomic_set(&fsdb->fsdb_notify_phase, 0);
407 fsdb->fsdb_notify_count++;
408 delta = cfs_time_sub(cfs_time_current(), fsdb->fsdb_notify_start);
409 fsdb->fsdb_notify_total += delta;
410 if (delta > fsdb->fsdb_notify_max)
411 fsdb->fsdb_notify_max = delta;
413 cfs_duration_usec(delta, &tv);
414 CDEBUG(D_MGS, "Revoke recover lock of %s completed after %ld.%06lds\n",
415 fsdb->fsdb_name, tv.tv_sec, tv.tv_usec);
418 static int mgs_ir_notify(void *arg)
420 struct fs_db *fsdb = arg;
421 struct ldlm_res_id resid;
423 char name[sizeof(fsdb->fsdb_name) + 20];
425 LASSERTF(sizeof(name) < 32, "name is too large to be in stack.\n");
426 sprintf(name, "mgs_%s_notify", fsdb->fsdb_name);
428 complete(&fsdb->fsdb_notify_comp);
430 set_user_nice(current, -2);
432 mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER);
434 struct l_wait_info lwi = { 0 };
436 l_wait_event(fsdb->fsdb_notify_waitq,
437 fsdb->fsdb_notify_stop ||
438 atomic_read(&fsdb->fsdb_notify_phase),
440 if (fsdb->fsdb_notify_stop)
443 CDEBUG(D_MGS, "%s woken up, phase is %d\n",
444 name, atomic_read(&fsdb->fsdb_notify_phase));
446 fsdb->fsdb_notify_start = cfs_time_current();
447 mgs_revoke_lock(fsdb->fsdb_mgs, fsdb, CONFIG_T_RECOVER);
450 complete(&fsdb->fsdb_notify_comp);
454 int mgs_ir_init_fs(const struct lu_env *env, struct mgs_device *mgs,
457 struct task_struct *task;
460 ir_timeout = OBD_IR_MGS_TIMEOUT;
462 fsdb->fsdb_ir_state = IR_FULL;
463 if (cfs_time_before(cfs_time_current_sec(),
464 mgs->mgs_start_time + ir_timeout))
465 fsdb->fsdb_ir_state = IR_STARTUP;
466 fsdb->fsdb_nonir_clients = 0;
467 INIT_LIST_HEAD(&fsdb->fsdb_clients);
469 /* start notify thread */
470 fsdb->fsdb_mgs = mgs;
471 atomic_set(&fsdb->fsdb_notify_phase, 0);
472 init_waitqueue_head(&fsdb->fsdb_notify_waitq);
473 init_completion(&fsdb->fsdb_notify_comp);
475 task = kthread_run(mgs_ir_notify, fsdb,
476 "mgs_%s_notify", fsdb->fsdb_name);
478 wait_for_completion(&fsdb->fsdb_notify_comp);
480 CERROR("Start notify thread error %ld\n", PTR_ERR(task));
482 mgs_nidtbl_init_fs(env, fsdb);
486 void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb)
488 if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
491 mgs_fsc_cleanup_by_fsdb(fsdb);
493 mgs_nidtbl_fini_fs(fsdb);
495 LASSERT(list_empty(&fsdb->fsdb_clients));
497 fsdb->fsdb_notify_stop = 1;
498 wake_up(&fsdb->fsdb_notify_waitq);
499 wait_for_completion(&fsdb->fsdb_notify_comp);
502 /* caller must have held fsdb_mutex */
503 static inline void ir_state_graduate(struct fs_db *fsdb)
505 if (fsdb->fsdb_ir_state == IR_STARTUP) {
506 if (cfs_time_before(fsdb->fsdb_mgs->mgs_start_time + ir_timeout,
507 cfs_time_current_sec())) {
508 fsdb->fsdb_ir_state = IR_FULL;
509 if (fsdb->fsdb_nonir_clients)
510 fsdb->fsdb_ir_state = IR_PARTIAL;
515 int mgs_ir_update(const struct lu_env *env, struct mgs_device *mgs,
516 struct mgs_target_info *mti)
522 if (mti->mti_instance == 0)
525 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
529 rc = mgs_nidtbl_write(env, fsdb, mti);
534 mutex_lock(&fsdb->fsdb_mutex);
535 ir_state_graduate(fsdb);
536 switch (fsdb->fsdb_ir_state) {
538 mti->mti_flags |= LDD_F_IR_CAPABLE;
548 mutex_unlock(&fsdb->fsdb_mutex);
550 LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
552 CDEBUG(D_MGS, "Try to revoke recover lock of %s\n",
554 atomic_inc(&fsdb->fsdb_notify_phase);
555 wake_up(&fsdb->fsdb_notify_waitq);
560 /* NID table can be cached by two entities: Clients and MDTs */
566 static int delogname(char *logname, char *fsname, int *typ)
572 ptr = strrchr(logname, '-');
576 /* decouple file system name. The llog name may be:
577 * - "prefix-fsname", prefix is "cliir" or "mdtir"
579 if (strncmp(ptr, "-mdtir", 6) == 0)
581 else if (strncmp(ptr, "-cliir", 6) == 0)
590 memcpy(fsname, logname, len);
597 int mgs_get_ir_logs(struct ptlrpc_request *req)
599 struct lu_env *env = req->rq_svc_thread->t_env;
600 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
602 struct mgs_config_body *body;
603 struct mgs_config_res *res;
604 struct ptlrpc_bulk_desc *desc;
605 struct l_wait_info lwi;
615 struct page **pages = NULL;
618 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
622 if (body->mcb_type != CONFIG_T_RECOVER)
625 rc = delogname(body->mcb_name, fsname, &type);
629 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
633 bufsize = body->mcb_units << body->mcb_bits;
634 nrpages = (bufsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
635 if (nrpages > PTLRPC_MAX_BRW_PAGES)
638 CDEBUG(D_MGS, "Reading IR log %s bufsize %ld.\n",
639 body->mcb_name, bufsize);
641 OBD_ALLOC(pages, sizeof(*pages) * nrpages);
645 res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
647 GOTO(out, rc = -EINVAL);
649 res->mcr_offset = body->mcb_offset;
650 unit_size = min_t(int, 1 << body->mcb_bits, PAGE_CACHE_SIZE);
651 bytes = mgs_nidtbl_read(req->rq_export, &fsdb->fsdb_nidtbl, res,
652 pages, nrpages, bufsize / unit_size, unit_size);
654 GOTO(out, rc = bytes);
656 /* start bulk transfer */
657 page_count = (bytes + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
658 LASSERT(page_count <= nrpages);
659 desc = ptlrpc_prep_bulk_exp(req, page_count, 1,
660 BULK_PUT_SOURCE, MGS_BULK_PORTAL);
662 GOTO(out, rc = -ENOMEM);
664 for (i = 0; i < page_count && bytes > 0; i++) {
665 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0,
666 min_t(int, bytes, PAGE_CACHE_SIZE));
667 bytes -= PAGE_CACHE_SIZE;
670 rc = target_bulk_io(req->rq_export, desc, &lwi);
671 ptlrpc_free_bulk_pin(desc);
674 for (i = 0; i < nrpages; i++) {
675 if (pages[i] == NULL)
677 __free_page(pages[i]);
679 OBD_FREE(pages, sizeof(*pages) * nrpages);
683 static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf)
685 const char *strings[] = IR_STRINGS;
689 for (i = 0; i < ARRAY_SIZE(strings); i++) {
690 if (strcmp(strings[i], buf) == 0) {
698 CDEBUG(D_MGS, "change fsr state of %s from %s to %s\n",
699 fsdb->fsdb_name, strings[fsdb->fsdb_ir_state], strings[state]);
700 mutex_lock(&fsdb->fsdb_mutex);
701 if (state == IR_FULL && fsdb->fsdb_nonir_clients)
703 fsdb->fsdb_ir_state = state;
704 mutex_unlock(&fsdb->fsdb_mutex);
709 static int lprocfs_ir_set_timeout(struct fs_db *fsdb, const char *buf)
714 static int lprocfs_ir_clear_stats(struct fs_db *fsdb, const char *buf)
719 fsdb->fsdb_notify_total = 0;
720 fsdb->fsdb_notify_max = 0;
721 fsdb->fsdb_notify_count = 0;
725 static struct lproc_ir_cmd {
728 int (*handler)(struct fs_db *, const char *);
730 { "state=", 6, lprocfs_ir_set_state },
731 { "timeout=", 8, lprocfs_ir_set_timeout },
732 { "0", 1, lprocfs_ir_clear_stats }
735 int lprocfs_wr_ir_state(struct file *file, const char __user *buffer,
736 size_t count, void *data)
738 struct fs_db *fsdb = data;
743 if (count == 0 || count >= PAGE_CACHE_SIZE)
746 OBD_ALLOC(kbuf, count + 1);
750 if (copy_from_user(kbuf, buffer, count)) {
751 OBD_FREE(kbuf, count + 1);
755 kbuf[count] = 0; /* buffer is supposed to end with 0 */
756 if (kbuf[count - 1] == '\n')
760 /* fsname=<file system name> must be the 1st entry */
761 while (ptr != NULL) {
765 tmpptr = strchr(ptr, ';');
770 for (i = 0; i < ARRAY_SIZE(ir_cmds); i++) {
771 struct lproc_ir_cmd *cmd;
775 cmdlen = cmd->namelen;
776 if (strncmp(cmd->name, ptr, cmdlen) == 0) {
778 rc = cmd->handler(fsdb, ptr);
788 CERROR("Unable to process command: %s(%d)\n", ptr, rc);
789 OBD_FREE(kbuf, count + 1);
793 int lprocfs_rd_ir_state(struct seq_file *seq, void *data)
795 struct fs_db *fsdb = data;
796 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
797 const char *ir_strings[] = IR_STRINGS;
798 struct timeval tv_max;
801 /* mgs_live_seq_show() already holds fsdb_mutex. */
802 ir_state_graduate(fsdb);
804 seq_printf(seq, "\nimperative_recovery_state:\n");
807 " nonir_clients: %d\n"
808 " nidtbl_version: %lld\n",
809 ir_strings[fsdb->fsdb_ir_state], fsdb->fsdb_nonir_clients,
812 cfs_duration_usec(fsdb->fsdb_notify_total, &tv);
813 cfs_duration_usec(fsdb->fsdb_notify_max, &tv_max);
815 seq_printf(seq, " notify_duration_total: %lu.%06lu\n"
816 " notify_duation_max: %lu.%06lu\n"
817 " notify_count: %u\n",
818 tv.tv_sec, tv.tv_usec,
819 tv_max.tv_sec, tv_max.tv_usec,
820 fsdb->fsdb_notify_count);
825 int lprocfs_ir_timeout_seq_show(struct seq_file *m, void *data)
827 return lprocfs_uint_seq_show(m, &ir_timeout);
830 ssize_t lprocfs_ir_timeout_seq_write(struct file *file,
831 const char __user *buffer,
832 size_t count, loff_t *off)
834 return lprocfs_wr_uint(file, buffer, count, &ir_timeout);
837 /* --------------- Handle non IR support clients --------------- */
838 /* attach a lustre file system to an export */
839 int mgs_fsc_attach(const struct lu_env *env, struct obd_export *exp,
842 struct mgs_export_data *data = &exp->u.eu_mgs_data;
843 struct mgs_device *mgs = exp2mgs_dev(exp);
845 struct mgs_fsc *fsc = NULL;
846 struct mgs_fsc *new_fsc = NULL;
851 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
855 /* allocate a new fsc in case we need it in spinlock. */
856 OBD_ALLOC_PTR(new_fsc);
860 INIT_LIST_HEAD(&new_fsc->mfc_export_list);
861 INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list);
862 new_fsc->mfc_fsdb = fsdb;
863 new_fsc->mfc_export = class_export_get(exp);
864 new_fsc->mfc_ir_capable = !!(exp_connect_flags(exp) &
865 OBD_CONNECT_IMP_RECOV);
868 mutex_lock(&fsdb->fsdb_mutex);
870 /* tend to find it in export list because this list is shorter. */
871 spin_lock(&data->med_lock);
872 list_for_each_entry(fsc, &data->med_clients, mfc_export_list) {
873 if (strcmp(fsname, fsc->mfc_fsdb->fsdb_name) == 0) {
882 /* add it into export list. */
883 list_add(&fsc->mfc_export_list, &data->med_clients);
885 /* add into fsdb list. */
886 list_add(&fsc->mfc_fsdb_list, &fsdb->fsdb_clients);
887 if (!fsc->mfc_ir_capable) {
888 ++fsdb->fsdb_nonir_clients;
889 if (fsdb->fsdb_ir_state == IR_FULL)
890 fsdb->fsdb_ir_state = IR_PARTIAL;
894 spin_unlock(&data->med_lock);
895 mutex_unlock(&fsdb->fsdb_mutex);
898 class_export_put(new_fsc->mfc_export);
899 OBD_FREE_PTR(new_fsc);
904 void mgs_fsc_cleanup(struct obd_export *exp)
906 struct mgs_export_data *data = &exp->u.eu_mgs_data;
907 struct mgs_fsc *fsc, *tmp;
908 struct list_head head = LIST_HEAD_INIT(head);
910 spin_lock(&data->med_lock);
911 list_splice_init(&data->med_clients, &head);
912 spin_unlock(&data->med_lock);
914 list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) {
915 struct fs_db *fsdb = fsc->mfc_fsdb;
917 LASSERT(fsc->mfc_export == exp);
919 mutex_lock(&fsdb->fsdb_mutex);
920 list_del_init(&fsc->mfc_fsdb_list);
921 if (fsc->mfc_ir_capable == 0) {
922 --fsdb->fsdb_nonir_clients;
923 LASSERT(fsdb->fsdb_ir_state != IR_FULL);
924 if (fsdb->fsdb_nonir_clients == 0 &&
925 fsdb->fsdb_ir_state == IR_PARTIAL)
926 fsdb->fsdb_ir_state = IR_FULL;
928 mutex_unlock(&fsdb->fsdb_mutex);
929 list_del_init(&fsc->mfc_export_list);
930 class_export_put(fsc->mfc_export);
935 /* must be called with fsdb->fsdb_mutex held */
936 void mgs_fsc_cleanup_by_fsdb(struct fs_db *fsdb)
938 struct mgs_fsc *fsc, *tmp;
940 list_for_each_entry_safe(fsc, tmp, &fsdb->fsdb_clients,
942 struct mgs_export_data *data = &fsc->mfc_export->u.eu_mgs_data;
944 LASSERT(fsdb == fsc->mfc_fsdb);
945 list_del_init(&fsc->mfc_fsdb_list);
947 spin_lock(&data->med_lock);
948 list_del_init(&fsc->mfc_export_list);
949 spin_unlock(&data->med_lock);
950 class_export_put(fsc->mfc_export);
954 fsdb->fsdb_nonir_clients = 0;
955 if (fsdb->fsdb_ir_state == IR_PARTIAL)
956 fsdb->fsdb_ir_state = IR_FULL;