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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/mgs/mgs_nids.c
33 * NID table management for lustre.
35 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_MGS
39 #define D_MGS D_CONFIG
41 #include <linux/kthread.h>
42 #include <linux/pagemap.h>
45 #include <obd_class.h>
46 #include <lustre_disk.h>
48 #include "mgs_internal.h"
50 static time64_t ir_timeout;
52 static int nidtbl_is_sane(struct mgs_nidtbl *tbl)
54 struct mgs_nidtbl_target *tgt;
57 LASSERT(mutex_is_locked(&tbl->mn_lock));
58 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
59 if (!tgt->mnt_version)
62 if (version >= tgt->mnt_version)
65 version = tgt->mnt_version;
71 * Fetch nidtbl entries whose version are not less than @version
72 * nidtbl entries will be packed in @pages by @unit_size units - entries
73 * shouldn't cross unit boundaries.
75 static int mgs_nidtbl_read(struct obd_export *exp, struct mgs_nidtbl *tbl,
76 struct mgs_config_res *res, u8 nid_size,
77 struct page **pages, int nrpages,
78 int units_total, int unit_size)
80 struct mgs_nidtbl_target *tgt;
81 struct mgs_nidtbl_entry *entry;
82 struct mgs_nidtbl_entry *last_in_unit = NULL;
83 struct mgs_target_info *mti;
84 __u64 version = res->mcr_offset;
87 int bytes_in_unit = 0;
88 int units_in_page = 0;
94 /* make sure unit_size is power 2 */
95 LASSERT((unit_size & (unit_size - 1)) == 0);
96 LASSERT(nrpages << PAGE_SHIFT >= units_total * unit_size);
98 mutex_lock(&tbl->mn_lock);
99 LASSERT(nidtbl_is_sane(tbl));
101 /* no more entries ? */
102 if (version > tbl->mn_version) {
103 version = tbl->mn_version;
108 * iterate over all targets to compose a bitmap by the type of llog.
109 * If the llog is for MDTs, llog entries for OSTs will be returned;
110 * otherwise, it's for clients, then llog entries for both OSTs and
111 * MDTs will be returned.
113 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
114 int entry_len = sizeof(*entry);
117 if (tgt->mnt_version < version)
120 /* write target recover information */
123 entry_len += mti->mti_nid_count * sizeof(lnet_nid_t);
125 entry_len += mti->mti_nid_count * nid_size;
127 if (entry_len > unit_size) {
128 CWARN("nidtbl: too large entry: entry length %d, unit size: %d\n",
129 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 && bytes_in_unit) {
141 last_in_unit->mne_length += bytes_in_unit;
143 buf += bytes_in_unit;
146 LASSERT((rc & (unit_size - 1)) == 0);
148 if (units_in_page == 0) {
149 /* allocate a new page */
150 pages[index] = alloc_page(GFP_KERNEL);
156 /* destroy previous map */
158 kunmap(pages[index - 1]);
160 /* reassign buffer */
161 buf = kmap(pages[index]);
164 units_in_page = PAGE_SIZE / unit_size;
165 LASSERT(units_in_page > 0);
168 /* allocate an unit */
169 LASSERT(((long)buf & (unit_size - 1)) == 0);
170 bytes_in_unit = unit_size;
176 entry = (struct mgs_nidtbl_entry *)buf;
177 entry->mne_version = tgt->mnt_version;
178 entry->mne_instance = mti->mti_instance;
179 entry->mne_index = mti->mti_stripe_index;
180 entry->mne_length = entry_len;
181 entry->mne_type = tgt->mnt_type;
183 entry->mne_nid_size = nid_size;
184 entry->mne_nid_type = 1;
186 entry->mne_nid_size = sizeof(lnet_nid_t);
187 entry->mne_nid_type = 0;
189 entry->mne_nid_count = 0;
190 /* We have been sent the newer larger NID format but the
191 * current nidtbl doesn't support it. So filter the NIDs
192 * sent to reject any real larger size NIDS.
194 if (target_supports_large_nid(mti)) {
195 for (i = 0; i < mti->mti_nid_count; i++) {
199 err = libcfs_strnid(&nid, mti->mti_nidlist[i]);
204 if (!nid_is_nid4(&nid))
207 entry->u.nids[entry->mne_nid_count] =
208 lnet_nid_to_nid4(&nid);
210 /* If the mgs_target_info NIDs are
211 * struct lnet_nid that have been
212 * expanded in size we still can
213 * use the nid if it fits in what
214 * the client supports.
216 if (NID_BYTES(&nid) > nid_size)
219 entry->u.nidlist[entry->mne_nid_count] =
222 entry->mne_nid_count++;
226 for (i = 0; i < mti->mti_nid_count; i++)
227 lnet_nid4_to_nid(mti->mti_nids[i],
228 &entry->u.nidlist[i]);
230 memcpy(entry->u.nids, mti->mti_nids,
231 mti->mti_nid_count * sizeof(lnet_nid_t));
233 entry->mne_nid_count = mti->mti_nid_count;
236 version = tgt->mnt_version;
240 bytes_in_unit -= entry_len;
241 last_in_unit = entry;
243 CDEBUG(D_MGS, "fsname %s, entry size %d, pages %d/%d/%d/%d.\n",
244 tbl->mn_fsdb->fsdb_name, entry_len,
245 bytes_in_unit, index, nrpages, units_total);
248 kunmap(pages[index - 1]);
250 LASSERT(version <= tbl->mn_version);
251 res->mcr_size = tbl->mn_version;
252 res->mcr_offset = nobuf ? version : tbl->mn_version;
253 mutex_unlock(&tbl->mn_lock);
254 LASSERT(ergo(version == 1, rc == 0)); /* get the log first time */
256 CDEBUG(D_MGS, "Read IR logs %s return with %d, version %llu\n",
257 tbl->mn_fsdb->fsdb_name, rc, version);
261 static int nidtbl_update_version(const struct lu_env *env,
262 struct mgs_device *mgs,
263 struct mgs_nidtbl *tbl)
265 struct dt_object *fsdb;
268 struct lu_buf buf = {
270 .lb_len = sizeof(version)
277 if (mgs->mgs_bottom->dd_rdonly)
280 LASSERT(mutex_is_locked(&tbl->mn_lock));
282 fsdb = local_file_find_or_create(env, mgs->mgs_los, mgs->mgs_nidtbl_dir,
283 tbl->mn_fsdb->fsdb_name,
284 S_IFREG | S_IRUGO | S_IWUSR);
286 RETURN(PTR_ERR(fsdb));
288 th = dt_trans_create(env, mgs->mgs_bottom);
290 GOTO(out_put, rc = PTR_ERR(th));
292 th->th_sync = 1; /* update table synchronously */
293 rc = dt_declare_record_write(env, fsdb, &buf, off, th);
297 rc = dt_trans_start_local(env, mgs->mgs_bottom, th);
301 version = cpu_to_le64(tbl->mn_version);
302 rc = dt_record_write(env, fsdb, &buf, &off, th);
305 dt_trans_stop(env, mgs->mgs_bottom, th);
307 dt_object_put(env, fsdb);
311 #define MGS_NIDTBL_VERSION_INIT 2
313 static int nidtbl_read_version(const struct lu_env *env,
314 struct mgs_device *mgs, struct mgs_nidtbl *tbl,
317 struct dt_object *fsdb;
320 struct lu_buf buf = {
322 .lb_len = sizeof(tmpver)
329 LASSERT(mutex_is_locked(&tbl->mn_lock));
331 LASSERT(mgs->mgs_nidtbl_dir);
332 rc = dt_lookup_dir(env, mgs->mgs_nidtbl_dir, tbl->mn_fsdb->fsdb_name,
335 *version = MGS_NIDTBL_VERSION_INIT;
341 fsdb = dt_locate_at(env, mgs->mgs_bottom, &fid,
342 &mgs->mgs_dt_dev.dd_lu_dev, NULL);
344 RETURN(PTR_ERR(fsdb));
346 rc = dt_read(env, fsdb, &buf, &off);
347 if (rc == buf.lb_len) {
348 *version = le64_to_cpu(tmpver);
350 } else if (rc == 0) {
351 *version = MGS_NIDTBL_VERSION_INIT;
353 CERROR("%s: read version file %s error %d\n",
354 mgs->mgs_obd->obd_name, tbl->mn_fsdb->fsdb_name, rc);
356 dt_object_put(env, fsdb);
360 static int mgs_nidtbl_write(const struct lu_env *env, struct fs_db *fsdb,
361 struct mgs_target_info *mti)
363 struct mgs_nidtbl *tbl;
364 struct mgs_nidtbl_target *tgt;
366 int type = mti->mti_flags & LDD_F_SV_TYPE_MASK;
371 type &= ~LDD_F_SV_TYPE_MGS;
374 tbl = &fsdb->fsdb_nidtbl;
375 mutex_lock(&tbl->mn_lock);
376 list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
377 struct mgs_target_info *info = &tgt->mnt_mti;
379 if (type == tgt->mnt_type &&
380 mti->mti_stripe_index == info->mti_stripe_index) {
386 if (target_supports_large_nid(mti))
387 mti_len = mti->mti_nid_count * LNET_NIDSTR_SIZE;
389 size_t len = offsetof(struct mgs_nidtbl_target,
390 mnt_mti.mti_nidlist);
392 OBD_ALLOC(tgt, len + mti_len);
394 GOTO(out, rc = -ENOMEM);
396 INIT_LIST_HEAD(&tgt->mnt_list);
398 tgt->mnt_version = 0; /* 0 means invalid */
399 tgt->mnt_type = type;
401 ++tbl->mn_nr_targets;
404 tgt->mnt_version = ++tbl->mn_version;
406 if (target_supports_large_nid(mti))
407 memcpy(tgt->mnt_mti.mti_nidlist, mti->mti_nidlist, mti_len);
409 list_move_tail(&tgt->mnt_list, &tbl->mn_targets);
411 rc = nidtbl_update_version(env, fsdb->fsdb_mgs, tbl);
415 mutex_unlock(&tbl->mn_lock);
417 CERROR("Write NID table version for file system %s error %d\n",
418 fsdb->fsdb_name, rc);
422 static void mgs_nidtbl_fini_fs(struct fs_db *fsdb)
424 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
425 size_t len = offsetof(struct mgs_nidtbl_target,
426 mnt_mti.mti_nidlist);
429 mutex_lock(&tbl->mn_lock);
430 tbl->mn_nr_targets = 0;
431 list_splice_init(&tbl->mn_targets, &head);
432 mutex_unlock(&tbl->mn_lock);
434 while (!list_empty(&head)) {
435 struct mgs_nidtbl_target *tgt;
438 tgt = list_first_entry(&head, struct mgs_nidtbl_target,
440 if (target_supports_large_nid(&tgt->mnt_mti))
441 mti_len += tgt->mnt_mti.mti_nid_count * LNET_NIDSTR_SIZE;
442 list_del(&tgt->mnt_list);
443 OBD_FREE(tgt, len + mti_len);
447 static int mgs_nidtbl_init_fs(const struct lu_env *env, struct fs_db *fsdb)
449 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
452 INIT_LIST_HEAD(&tbl->mn_targets);
453 mutex_init(&tbl->mn_lock);
454 tbl->mn_nr_targets = 0;
456 mutex_lock(&tbl->mn_lock);
457 rc = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl, &tbl->mn_version);
458 mutex_unlock(&tbl->mn_lock);
460 CERROR("%s: IR: failed to read current version, rc = %d\n",
461 fsdb->fsdb_mgs->mgs_obd->obd_name, rc);
463 CDEBUG(D_MGS, "IR: current version is %llu\n",
469 /* --------- Imperative Recovery relies on nidtbl stuff ------- */
470 void mgs_ir_notify_complete(struct fs_db *fsdb)
472 struct timespec64 ts;
475 atomic_set(&fsdb->fsdb_notify_phase, 0);
478 fsdb->fsdb_notify_count++;
479 delta = ktime_sub(ktime_get(), fsdb->fsdb_notify_start);
480 fsdb->fsdb_notify_total = ktime_add(fsdb->fsdb_notify_total, delta);
481 if (ktime_after(delta, fsdb->fsdb_notify_max))
482 fsdb->fsdb_notify_max = delta;
484 ts = ktime_to_timespec64(fsdb->fsdb_notify_max);
485 CDEBUG(D_MGS, "Revoke recover lock of %s completed after %lld.%09lds\n",
486 fsdb->fsdb_name, (s64)ts.tv_sec, ts.tv_nsec);
489 static int mgs_ir_notify(void *arg)
491 struct fs_db *fsdb = arg;
492 struct ldlm_res_id resid;
493 char name[sizeof(fsdb->fsdb_name) + 16];
495 BUILD_BUG_ON(sizeof(name) >= 40); /* name is too large to be on stack */
497 snprintf(name, sizeof(name) - 1, "mgs_%s_notify", fsdb->fsdb_name);
498 complete(&fsdb->fsdb_notify_comp);
499 set_user_nice(current, -2);
500 mgc_fsname2resid(fsdb->fsdb_name, &resid, MGS_CFG_T_RECOVER);
502 wait_event_idle(fsdb->fsdb_notify_waitq,
503 fsdb->fsdb_notify_stop ||
504 atomic_read(&fsdb->fsdb_notify_phase));
506 if (fsdb->fsdb_notify_stop)
509 CDEBUG(D_MGS, "%s woken up, phase is %d\n",
510 name, atomic_read(&fsdb->fsdb_notify_phase));
512 fsdb->fsdb_notify_start = ktime_get();
513 mgs_revoke_lock(fsdb->fsdb_mgs, fsdb, MGS_CFG_T_RECOVER);
516 complete(&fsdb->fsdb_notify_comp);
520 int mgs_ir_init_fs(const struct lu_env *env, struct mgs_device *mgs,
523 struct task_struct *task;
526 ir_timeout = (time64_t)OBD_IR_MGS_TIMEOUT;
528 fsdb->fsdb_ir_state = IR_FULL;
529 if (mgs->mgs_start_time + ir_timeout > ktime_get_real_seconds())
530 fsdb->fsdb_ir_state = IR_STARTUP;
531 fsdb->fsdb_nonir_clients = 0;
532 /* start notify thread */
533 fsdb->fsdb_mgs = mgs;
534 task = kthread_run(mgs_ir_notify, fsdb,
535 "mgs_%s_notify", fsdb->fsdb_name);
537 wait_for_completion(&fsdb->fsdb_notify_comp);
539 CERROR("Start notify thread error %ld\n", PTR_ERR(task));
541 mgs_nidtbl_init_fs(env, fsdb);
545 void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb)
547 if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
550 mgs_fsc_cleanup_by_fsdb(fsdb);
552 mgs_nidtbl_fini_fs(fsdb);
554 LASSERT(list_empty(&fsdb->fsdb_clients));
556 fsdb->fsdb_notify_stop = 1;
557 wake_up(&fsdb->fsdb_notify_waitq);
558 wait_for_completion(&fsdb->fsdb_notify_comp);
561 /* caller must have held fsdb_mutex */
562 static inline void ir_state_graduate(struct fs_db *fsdb)
564 if (fsdb->fsdb_ir_state == IR_STARTUP) {
565 if (ktime_get_real_seconds() >
566 fsdb->fsdb_mgs->mgs_start_time + ir_timeout) {
567 fsdb->fsdb_ir_state = IR_FULL;
568 if (fsdb->fsdb_nonir_clients)
569 fsdb->fsdb_ir_state = IR_PARTIAL;
574 int mgs_ir_update(const struct lu_env *env, struct mgs_device *mgs,
575 struct mgs_target_info *mti)
581 if (mti->mti_instance == 0)
584 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
588 rc = mgs_nidtbl_write(env, fsdb, mti);
593 mutex_lock(&fsdb->fsdb_mutex);
594 ir_state_graduate(fsdb);
595 switch (fsdb->fsdb_ir_state) {
597 mti->mti_flags |= LDD_F_IR_CAPABLE;
607 mutex_unlock(&fsdb->fsdb_mutex);
609 LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
611 CDEBUG(D_MGS, "Try to revoke recover lock of %s\n",
613 atomic_inc(&fsdb->fsdb_notify_phase);
614 wake_up(&fsdb->fsdb_notify_waitq);
618 mgs_put_fsdb(mgs, fsdb);
622 /* NID table can be cached by two entities: Clients and MDTs */
628 static int delogname(char *logname, char *fsname, int *typ)
634 ptr = strrchr(logname, '-');
639 * decouple file system name. The llog name may be:
640 * - "prefix-fsname", prefix is "cliir" or "mdtir"
642 if (strncmp(ptr, "-mdtir", 6) == 0)
644 else if (strncmp(ptr, "-cliir", 6) == 0)
653 memcpy(fsname, logname, len);
660 int mgs_get_ir_logs(struct ptlrpc_request *req)
662 struct lu_env *env = req->rq_svc_thread->t_env;
663 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
664 struct fs_db *fsdb = NULL;
665 struct mgs_config_body *body;
666 struct mgs_config_res *res;
667 struct ptlrpc_bulk_desc *desc;
677 struct page **pages = NULL;
681 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
685 if (body->mcb_type != MGS_CFG_T_RECOVER)
688 rc = delogname(body->mcb_name, fsname, &type);
692 bufsize = body->mcb_units << body->mcb_bits;
693 nrpages = (bufsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
694 if (nrpages > PTLRPC_MAX_BRW_PAGES)
697 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
701 CDEBUG(D_MGS, "Reading IR log %s bufsize %ld.\n",
702 body->mcb_name, bufsize);
704 OBD_ALLOC_PTR_ARRAY_LARGE(pages, nrpages);
706 GOTO(out, rc = -ENOMEM);
708 res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
710 GOTO(out, rc = -EINVAL);
712 res->mcr_offset = body->mcb_offset;
713 unit_size = min_t(int, 1 << body->mcb_bits, PAGE_SIZE);
714 bytes = mgs_nidtbl_read(req->rq_export, &fsdb->fsdb_nidtbl, res,
715 body->mcb_rec_nid_size, pages, nrpages,
716 bufsize / unit_size, unit_size);
718 GOTO(out, rc = bytes);
720 /* start bulk transfer */
721 page_count = (bytes + PAGE_SIZE - 1) >> PAGE_SHIFT;
722 LASSERT(page_count <= nrpages);
723 desc = ptlrpc_prep_bulk_exp(req, page_count, 1,
724 PTLRPC_BULK_PUT_SOURCE,
726 &ptlrpc_bulk_kiov_pin_ops);
728 GOTO(out, rc = -ENOMEM);
730 for (i = 0; i < page_count && bytes > 0; i++) {
731 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
737 rc = target_bulk_io(req->rq_export, desc);
738 ptlrpc_free_bulk(desc);
744 for (i = 0; i < nrpages; i++) {
748 __free_page(pages[i]);
751 OBD_FREE_PTR_ARRAY_LARGE(pages, nrpages);
755 mgs_put_fsdb(mgs, fsdb);
760 static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf)
762 const char *const strings[] = IR_STRINGS;
766 for (i = 0; i < ARRAY_SIZE(strings); i++) {
767 if (strcmp(strings[i], buf) == 0) {
775 CDEBUG(D_MGS, "change fsr state of %s from %s to %s\n",
776 fsdb->fsdb_name, strings[fsdb->fsdb_ir_state], strings[state]);
777 mutex_lock(&fsdb->fsdb_mutex);
778 if (state == IR_FULL && fsdb->fsdb_nonir_clients)
780 fsdb->fsdb_ir_state = state;
781 mutex_unlock(&fsdb->fsdb_mutex);
786 static int lprocfs_ir_set_timeout(struct fs_db *fsdb, const char *buf)
791 static int lprocfs_ir_clear_stats(struct fs_db *fsdb, const char *buf)
796 fsdb->fsdb_notify_total = ktime_set(0, 0);
797 fsdb->fsdb_notify_max = ktime_set(0, 0);
798 fsdb->fsdb_notify_count = 0;
802 static struct lproc_ir_cmd {
805 int (*handler)(struct fs_db *, const char *);
807 { "state=", 6, lprocfs_ir_set_state },
808 { "timeout=", 8, lprocfs_ir_set_timeout },
809 { "0", 1, lprocfs_ir_clear_stats }
812 int lprocfs_wr_ir_state(struct file *file, const char __user *buffer,
813 size_t count, void *data)
815 struct fs_db *fsdb = data;
820 if (count == 0 || count >= PAGE_SIZE)
823 OBD_ALLOC(kbuf, count + 1);
827 if (copy_from_user(kbuf, buffer, count)) {
828 OBD_FREE(kbuf, count + 1);
832 kbuf[count] = 0; /* buffer is supposed to end with 0 */
833 if (kbuf[count - 1] == '\n')
837 /* fsname=<file system name> must be the 1st entry */
842 tmpptr = strchr(ptr, ';');
847 for (i = 0; i < ARRAY_SIZE(ir_cmds); i++) {
848 struct lproc_ir_cmd *cmd;
852 cmdlen = cmd->namelen;
853 if (strncmp(cmd->name, ptr, cmdlen) == 0) {
855 rc = cmd->handler(fsdb, ptr);
865 CERROR("Unable to process command: %s(%d)\n", ptr, rc);
866 OBD_FREE(kbuf, count + 1);
870 int lprocfs_rd_ir_state(struct seq_file *seq, void *data)
872 struct fs_db *fsdb = data;
873 struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
874 const char *const ir_strings[] = IR_STRINGS;
875 struct timespec64 ts_max;
876 struct timespec64 ts;
878 /* mgs_live_seq_show() already holds fsdb_mutex. */
879 ir_state_graduate(fsdb);
881 seq_printf(seq, "\nimperative_recovery_state:\n");
884 " nonir_clients: %d\n"
885 " nidtbl_version: %lld\n",
886 ir_strings[fsdb->fsdb_ir_state], fsdb->fsdb_nonir_clients,
889 ts = ktime_to_timespec64(fsdb->fsdb_notify_total);
890 ts_max = ktime_to_timespec64(fsdb->fsdb_notify_max);
892 seq_printf(seq, " notify_duration_total: %lld.%09ld\n"
893 " notify_duation_max: %lld.%09ld\n"
894 " notify_count: %u\n",
895 (s64)ts.tv_sec, ts.tv_nsec,
896 (s64)ts_max.tv_sec, ts_max.tv_nsec,
897 fsdb->fsdb_notify_count);
902 int lprocfs_ir_timeout_seq_show(struct seq_file *m, void *data)
904 seq_printf(m, "%lld\n", ir_timeout);
908 ssize_t lprocfs_ir_timeout_seq_write(struct file *file,
909 const char __user *buffer,
910 size_t count, loff_t *off)
912 return kstrtoll_from_user(buffer, count, 0, &ir_timeout);
915 /* --------------- Handle non IR support clients --------------- */
916 /* attach a lustre file system to an export */
917 int mgs_fsc_attach(const struct lu_env *env, struct obd_export *exp,
920 struct mgs_export_data *data = &exp->u.eu_mgs_data;
921 struct mgs_device *mgs = exp2mgs_dev(exp);
922 struct fs_db *fsdb = NULL;
923 struct mgs_fsc *fsc = NULL;
924 struct mgs_fsc *new_fsc = NULL;
930 rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
934 /* allocate a new fsc in case we need it in spinlock. */
935 OBD_ALLOC_PTR(new_fsc);
937 GOTO(out, rc = -ENOMEM);
939 INIT_LIST_HEAD(&new_fsc->mfc_export_list);
940 INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list);
941 new_fsc->mfc_fsdb = fsdb;
942 new_fsc->mfc_export = class_export_get(exp);
943 new_fsc->mfc_ir_capable = !!(exp_connect_flags(exp) &
944 OBD_CONNECT_IMP_RECOV);
947 mutex_lock(&fsdb->fsdb_mutex);
949 /* tend to find it in export list because this list is shorter. */
950 spin_lock(&data->med_lock);
951 list_for_each_entry(fsc, &data->med_clients, mfc_export_list) {
952 if (strcmp(fsname, fsc->mfc_fsdb->fsdb_name) == 0) {
961 /* add it into export list. */
962 list_add(&fsc->mfc_export_list, &data->med_clients);
964 /* add into fsdb list. */
965 list_add(&fsc->mfc_fsdb_list, &fsdb->fsdb_clients);
966 if (!fsc->mfc_ir_capable) {
967 ++fsdb->fsdb_nonir_clients;
968 if (fsdb->fsdb_ir_state == IR_FULL)
969 fsdb->fsdb_ir_state = IR_PARTIAL;
973 spin_unlock(&data->med_lock);
974 mutex_unlock(&fsdb->fsdb_mutex);
977 class_export_put(new_fsc->mfc_export);
978 OBD_FREE_PTR(new_fsc);
982 mgs_put_fsdb(mgs, fsdb);
986 void mgs_fsc_cleanup(struct obd_export *exp)
988 struct mgs_export_data *data = &exp->u.eu_mgs_data;
989 struct mgs_fsc *fsc, *tmp;
992 spin_lock(&data->med_lock);
993 list_splice_init(&data->med_clients, &head);
994 spin_unlock(&data->med_lock);
996 list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) {
997 struct fs_db *fsdb = fsc->mfc_fsdb;
999 LASSERT(fsc->mfc_export == exp);
1001 mutex_lock(&fsdb->fsdb_mutex);
1002 list_del_init(&fsc->mfc_fsdb_list);
1003 if (fsc->mfc_ir_capable == 0) {
1004 --fsdb->fsdb_nonir_clients;
1005 LASSERT(fsdb->fsdb_ir_state != IR_FULL);
1006 if (fsdb->fsdb_nonir_clients == 0 &&
1007 fsdb->fsdb_ir_state == IR_PARTIAL)
1008 fsdb->fsdb_ir_state = IR_FULL;
1010 mutex_unlock(&fsdb->fsdb_mutex);
1011 list_del_init(&fsc->mfc_export_list);
1012 class_export_put(fsc->mfc_export);
1017 /* must be called with fsdb->fsdb_mutex held */
1018 void mgs_fsc_cleanup_by_fsdb(struct fs_db *fsdb)
1020 struct mgs_fsc *fsc, *tmp;
1022 list_for_each_entry_safe(fsc, tmp, &fsdb->fsdb_clients,
1024 struct mgs_export_data *data = &fsc->mfc_export->u.eu_mgs_data;
1026 LASSERT(fsdb == fsc->mfc_fsdb);
1027 list_del_init(&fsc->mfc_fsdb_list);
1029 spin_lock(&data->med_lock);
1030 list_del_init(&fsc->mfc_export_list);
1031 spin_unlock(&data->med_lock);
1032 class_export_put(fsc->mfc_export);
1036 fsdb->fsdb_nonir_clients = 0;
1037 if (fsdb->fsdb_ir_state == IR_PARTIAL)
1038 fsdb->fsdb_ir_state = IR_FULL;