1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_capa.c
38 * Author: Lai Siyao <lsy@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <linux/version.h>
45 #include <asm/uaccess.h>
46 #include <linux/file.h>
47 #include <linux/kmod.h>
49 #include <lustre_lite.h>
50 #include "llite_internal.h"
52 /* for obd_capa.c_list, client capa might stay in three places:
55 * 3. stand alone: just allocated.
58 /* capas for oss writeback and those failed to renew */
59 static CFS_LIST_HEAD(ll_idle_capas);
60 static struct ptlrpc_thread ll_capa_thread;
61 static cfs_list_t *ll_capa_list = &capa_list[CAPA_SITE_CLIENT];
63 /* llite capa renewal timer */
64 struct timer_list ll_capa_timer;
65 /* for debug: indicate whether capa on llite is enabled or not */
66 static cfs_atomic_t ll_capa_debug = CFS_ATOMIC_INIT(0);
67 static unsigned long long ll_capa_renewed = 0;
68 static unsigned long long ll_capa_renewal_noent = 0;
69 static unsigned long long ll_capa_renewal_failed = 0;
70 static unsigned long long ll_capa_renewal_retries = 0;
72 static inline void update_capa_timer(struct obd_capa *ocapa, cfs_time_t expiry)
74 if (cfs_time_before(expiry, ll_capa_timer.expires) ||
75 !timer_pending(&ll_capa_timer)) {
76 mod_timer(&ll_capa_timer, expiry);
77 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
78 "ll_capa_timer update: %lu/%lu by", expiry, jiffies);
82 static inline cfs_time_t capa_renewal_time(struct obd_capa *ocapa)
84 return cfs_time_sub(ocapa->c_expiry,
85 cfs_time_seconds(ocapa->c_capa.lc_timeout) / 2);
88 static inline int capa_is_to_expire(struct obd_capa *ocapa)
90 return cfs_time_beforeq(capa_renewal_time(ocapa), cfs_time_current());
93 static inline int have_expired_capa(void)
95 struct obd_capa *ocapa = NULL;
98 /* if ll_capa_list has client capa to expire or ll_idle_capas has
99 * expired capa, return 1.
101 cfs_spin_lock(&capa_lock);
102 if (!cfs_list_empty(ll_capa_list)) {
103 ocapa = cfs_list_entry(ll_capa_list->next, struct obd_capa,
105 expired = capa_is_to_expire(ocapa);
107 update_capa_timer(ocapa, capa_renewal_time(ocapa));
108 } else if (!cfs_list_empty(&ll_idle_capas)) {
109 ocapa = cfs_list_entry(ll_idle_capas.next, struct obd_capa,
111 expired = capa_is_expired(ocapa);
113 update_capa_timer(ocapa, ocapa->c_expiry);
115 cfs_spin_unlock(&capa_lock);
118 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "expired");
122 static inline int ll_capa_check_stop(void)
124 return (ll_capa_thread.t_flags & SVC_STOPPING) ? 1: 0;
127 static void sort_add_capa(struct obd_capa *ocapa, cfs_list_t *head)
129 struct obd_capa *tmp;
130 cfs_list_t *before = NULL;
132 /* TODO: client capa is sorted by expiry, this could be optimized */
133 cfs_list_for_each_entry_reverse(tmp, head, c_list) {
134 if (cfs_time_aftereq(ocapa->c_expiry, tmp->c_expiry)) {
135 before = &tmp->c_list;
140 LASSERT(&ocapa->c_list != before);
141 cfs_list_add(&ocapa->c_list, before ?: head);
144 static inline int obd_capa_open_count(struct obd_capa *oc)
146 struct ll_inode_info *lli = ll_i2info(oc->u.cli.inode);
147 return cfs_atomic_read(&lli->lli_open_count);
150 static void ll_delete_capa(struct obd_capa *ocapa)
152 struct ll_inode_info *lli = ll_i2info(ocapa->u.cli.inode);
154 if (capa_for_mds(&ocapa->c_capa)) {
155 LASSERT(lli->lli_mds_capa == ocapa);
156 lli->lli_mds_capa = NULL;
157 } else if (capa_for_oss(&ocapa->c_capa)) {
158 cfs_list_del_init(&ocapa->u.cli.lli_list);
161 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "free client");
162 cfs_list_del_init(&ocapa->c_list);
163 capa_count[CAPA_SITE_CLIENT]--;
164 /* release the ref when alloc */
168 /* three places where client capa is deleted:
169 * 1. capa_thread_main(), main place to delete expired capa.
170 * 2. ll_clear_inode_capas() in ll_clear_inode().
171 * 3. ll_truncate_free_capa() delete truncate capa explicitly in ll_truncate().
173 static int capa_thread_main(void *unused)
175 struct obd_capa *ocapa, *tmp, *next;
176 struct inode *inode = NULL;
177 struct l_wait_info lwi = { 0 };
181 cfs_daemonize("ll_capa");
183 ll_capa_thread.t_flags = SVC_RUNNING;
184 cfs_waitq_signal(&ll_capa_thread.t_ctl_waitq);
187 l_wait_event(ll_capa_thread.t_ctl_waitq,
188 (ll_capa_check_stop() || have_expired_capa()),
191 if (ll_capa_check_stop())
196 cfs_spin_lock(&capa_lock);
197 cfs_list_for_each_entry_safe(ocapa, tmp, ll_capa_list, c_list) {
200 LASSERT(ocapa->c_capa.lc_opc != CAPA_OPC_OSS_TRUNC);
202 if (!capa_is_to_expire(ocapa)) {
207 cfs_list_del_init(&ocapa->c_list);
209 /* for MDS capability, only renew those which belong to
210 * dir, or its inode is opened, or client holds LOOKUP
213 /* ibits may be changed by ll_have_md_lock() so we have
214 * to set it each time */
215 ibits = MDS_INODELOCK_LOOKUP;
216 if (capa_for_mds(&ocapa->c_capa) &&
217 !S_ISDIR(ocapa->u.cli.inode->i_mode) &&
218 obd_capa_open_count(ocapa) == 0 &&
219 !ll_have_md_lock(ocapa->u.cli.inode,
220 &ibits, LCK_MINMODE)) {
221 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
223 sort_add_capa(ocapa, &ll_idle_capas);
227 /* for OSS capability, only renew those whose inode is
230 if (capa_for_oss(&ocapa->c_capa) &&
231 obd_capa_open_count(ocapa) == 0) {
232 /* oss capa with open count == 0 won't renew,
233 * move to idle list */
234 sort_add_capa(ocapa, &ll_idle_capas);
238 /* NB iput() is in ll_update_capa() */
239 inode = igrab(ocapa->u.cli.inode);
241 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
248 cfs_spin_unlock(&capa_lock);
249 rc = md_renew_capa(ll_i2mdexp(inode), ocapa,
251 cfs_spin_lock(&capa_lock);
253 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
254 "renew failed: %d", rc);
255 ll_capa_renewal_failed++;
260 update_capa_timer(next, capa_renewal_time(next));
262 cfs_list_for_each_entry_safe(ocapa, tmp, &ll_idle_capas,
264 if (!capa_is_expired(ocapa)) {
266 update_capa_timer(ocapa,
271 if (cfs_atomic_read(&ocapa->c_refc) > 1) {
272 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
273 "expired(c_refc %d), don't release",
274 cfs_atomic_read(&ocapa->c_refc));
275 /* don't try to renew any more */
276 cfs_list_del_init(&ocapa->c_list);
280 /* expired capa is released. */
281 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "release expired");
282 ll_delete_capa(ocapa);
285 cfs_spin_unlock(&capa_lock);
288 ll_capa_thread.t_flags = SVC_STOPPED;
289 cfs_waitq_signal(&ll_capa_thread.t_ctl_waitq);
293 void ll_capa_timer_callback(unsigned long unused)
295 cfs_waitq_signal(&ll_capa_thread.t_ctl_waitq);
298 int ll_capa_thread_start(void)
303 cfs_waitq_init(&ll_capa_thread.t_ctl_waitq);
305 rc = cfs_create_thread(capa_thread_main, NULL, 0);
307 CERROR("cannot start expired capa thread: rc %d\n", rc);
310 cfs_wait_event(ll_capa_thread.t_ctl_waitq,
311 ll_capa_thread.t_flags & SVC_RUNNING);
316 void ll_capa_thread_stop(void)
318 ll_capa_thread.t_flags = SVC_STOPPING;
319 cfs_waitq_signal(&ll_capa_thread.t_ctl_waitq);
320 cfs_wait_event(ll_capa_thread.t_ctl_waitq,
321 ll_capa_thread.t_flags & SVC_STOPPED);
324 struct obd_capa *ll_osscapa_get(struct inode *inode, __u64 opc)
326 struct ll_inode_info *lli = ll_i2info(inode);
327 struct obd_capa *ocapa;
332 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_OSS_CAPA) == 0)
335 LASSERT(opc == CAPA_OPC_OSS_WRITE || opc == CAPA_OPC_OSS_RW ||
336 opc == CAPA_OPC_OSS_TRUNC);
338 cfs_spin_lock(&capa_lock);
339 cfs_list_for_each_entry(ocapa, &lli->lli_oss_capas, u.cli.lli_list) {
340 if (capa_is_expired(ocapa))
342 if ((opc & CAPA_OPC_OSS_WRITE) &&
343 capa_opc_supported(&ocapa->c_capa, CAPA_OPC_OSS_WRITE)) {
346 } else if ((opc & CAPA_OPC_OSS_READ) &&
347 capa_opc_supported(&ocapa->c_capa,
348 CAPA_OPC_OSS_READ)) {
351 } else if ((opc & CAPA_OPC_OSS_TRUNC) &&
352 capa_opc_supported(&ocapa->c_capa, opc)) {
359 LASSERT(lu_fid_eq(capa_fid(&ocapa->c_capa),
360 ll_inode2fid(inode)));
361 LASSERT(ocapa->c_site == CAPA_SITE_CLIENT);
365 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "found client");
369 if (cfs_atomic_read(&ll_capa_debug)) {
370 CERROR("no capability for "DFID" opc "LPX64"\n",
371 PFID(&lli->lli_fid), opc);
372 cfs_atomic_set(&ll_capa_debug, 0);
375 cfs_spin_unlock(&capa_lock);
379 EXPORT_SYMBOL(ll_osscapa_get);
381 struct obd_capa *ll_mdscapa_get(struct inode *inode)
383 struct ll_inode_info *lli = ll_i2info(inode);
384 struct obd_capa *ocapa;
387 LASSERT(inode != NULL);
389 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_MDS_CAPA) == 0)
392 cfs_spin_lock(&capa_lock);
393 ocapa = capa_get(lli->lli_mds_capa);
394 cfs_spin_unlock(&capa_lock);
395 if (!ocapa && cfs_atomic_read(&ll_capa_debug)) {
396 CERROR("no mds capability for "DFID"\n", PFID(&lli->lli_fid));
397 cfs_atomic_set(&ll_capa_debug, 0);
403 static struct obd_capa *do_add_mds_capa(struct inode *inode,
404 struct obd_capa *ocapa)
406 struct ll_inode_info *lli = ll_i2info(inode);
407 struct obd_capa *old = lli->lli_mds_capa;
408 struct lustre_capa *capa = &ocapa->c_capa;
411 ocapa->u.cli.inode = inode;
412 lli->lli_mds_capa = ocapa;
413 capa_count[CAPA_SITE_CLIENT]++;
415 DEBUG_CAPA(D_SEC, capa, "add MDS");
417 cfs_spin_lock(&old->c_lock);
419 cfs_spin_unlock(&old->c_lock);
421 DEBUG_CAPA(D_SEC, capa, "update MDS");
429 static struct obd_capa *do_lookup_oss_capa(struct inode *inode, int opc)
431 struct ll_inode_info *lli = ll_i2info(inode);
432 struct obd_capa *ocapa;
434 /* inside capa_lock */
435 cfs_list_for_each_entry(ocapa, &lli->lli_oss_capas, u.cli.lli_list) {
436 if ((capa_opc(&ocapa->c_capa) & opc) != opc)
439 LASSERT(lu_fid_eq(capa_fid(&ocapa->c_capa),
440 ll_inode2fid(inode)));
441 LASSERT(ocapa->c_site == CAPA_SITE_CLIENT);
443 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "found client");
450 static inline void inode_add_oss_capa(struct inode *inode,
451 struct obd_capa *ocapa)
453 struct ll_inode_info *lli = ll_i2info(inode);
454 struct obd_capa *tmp;
455 cfs_list_t *next = NULL;
457 /* capa is sorted in lli_oss_capas so lookup can always find the
459 cfs_list_for_each_entry(tmp, &lli->lli_oss_capas, u.cli.lli_list) {
460 if (cfs_time_after(ocapa->c_expiry, tmp->c_expiry)) {
461 next = &tmp->u.cli.lli_list;
465 LASSERT(&ocapa->u.cli.lli_list != next);
466 cfs_list_move_tail(&ocapa->u.cli.lli_list, next ?: &lli->lli_oss_capas);
469 static struct obd_capa *do_add_oss_capa(struct inode *inode,
470 struct obd_capa *ocapa)
472 struct obd_capa *old;
473 struct lustre_capa *capa = &ocapa->c_capa;
475 LASSERTF(S_ISREG(inode->i_mode),
476 "inode has oss capa, but not regular file, mode: %d\n",
479 /* FIXME: can't replace it so easily with fine-grained opc */
480 old = do_lookup_oss_capa(inode, capa_opc(capa) & CAPA_OPC_OSS_ONLY);
482 ocapa->u.cli.inode = inode;
483 CFS_INIT_LIST_HEAD(&ocapa->u.cli.lli_list);
484 capa_count[CAPA_SITE_CLIENT]++;
486 DEBUG_CAPA(D_SEC, capa, "add OSS");
488 cfs_spin_lock(&old->c_lock);
490 cfs_spin_unlock(&old->c_lock);
492 DEBUG_CAPA(D_SEC, capa, "update OSS");
498 inode_add_oss_capa(inode, ocapa);
502 struct obd_capa *ll_add_capa(struct inode *inode, struct obd_capa *ocapa)
504 cfs_spin_lock(&capa_lock);
505 ocapa = capa_for_mds(&ocapa->c_capa) ? do_add_mds_capa(inode, ocapa) :
506 do_add_oss_capa(inode, ocapa);
508 /* truncate capa won't renew */
509 if (ocapa->c_capa.lc_opc != CAPA_OPC_OSS_TRUNC) {
510 set_capa_expiry(ocapa);
511 cfs_list_del_init(&ocapa->c_list);
512 sort_add_capa(ocapa, ll_capa_list);
514 update_capa_timer(ocapa, capa_renewal_time(ocapa));
517 cfs_spin_unlock(&capa_lock);
519 cfs_atomic_set(&ll_capa_debug, 1);
523 static inline void delay_capa_renew(struct obd_capa *oc, cfs_time_t delay)
525 /* NB: set a fake expiry for this capa to prevent it renew too soon */
526 oc->c_expiry = cfs_time_add(oc->c_expiry, cfs_time_seconds(delay));
529 int ll_update_capa(struct obd_capa *ocapa, struct lustre_capa *capa)
531 struct inode *inode = ocapa->u.cli.inode;
540 cfs_spin_lock(&capa_lock);
542 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
543 "renewal canceled because object removed");
544 ll_capa_renewal_noent++;
546 ll_capa_renewal_failed++;
548 /* failed capa won't be renewed any longer, but if -EIO,
549 * client might be doing recovery, retry in 2 min. */
550 if (rc == -EIO && !capa_is_expired(ocapa)) {
551 delay_capa_renew(ocapa, 120);
552 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
553 "renewal failed: -EIO, "
555 ll_capa_renewal_retries++;
558 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
559 "renewal failed(rc: %d) for", rc);
563 cfs_list_del_init(&ocapa->c_list);
564 sort_add_capa(ocapa, &ll_idle_capas);
565 cfs_spin_unlock(&capa_lock);
572 cfs_spin_lock(&ocapa->c_lock);
573 LASSERT(!memcmp(&ocapa->c_capa, capa,
574 offsetof(struct lustre_capa, lc_opc)));
575 ocapa->c_capa = *capa;
576 set_capa_expiry(ocapa);
577 cfs_spin_unlock(&ocapa->c_lock);
579 cfs_spin_lock(&capa_lock);
580 if (capa_for_oss(capa))
581 inode_add_oss_capa(inode, ocapa);
582 DEBUG_CAPA(D_SEC, capa, "renew");
585 cfs_list_del_init(&ocapa->c_list);
586 sort_add_capa(ocapa, ll_capa_list);
587 update_capa_timer(ocapa, capa_renewal_time(ocapa));
588 cfs_spin_unlock(&capa_lock);
595 void ll_capa_open(struct inode *inode)
597 struct ll_inode_info *lli = ll_i2info(inode);
599 if ((ll_i2sbi(inode)->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
603 if (!S_ISREG(inode->i_mode))
606 cfs_atomic_inc(&lli->lli_open_count);
609 void ll_capa_close(struct inode *inode)
611 struct ll_inode_info *lli = ll_i2info(inode);
613 if ((ll_i2sbi(inode)->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
617 if (!S_ISREG(inode->i_mode))
620 cfs_atomic_dec(&lli->lli_open_count);
623 /* delete CAPA_OPC_OSS_TRUNC only */
624 void ll_truncate_free_capa(struct obd_capa *ocapa)
629 LASSERT(ocapa->c_capa.lc_opc & CAPA_OPC_OSS_TRUNC);
630 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "free truncate");
632 /* release ref when find */
634 if (likely(ocapa->c_capa.lc_opc == CAPA_OPC_OSS_TRUNC)) {
635 cfs_spin_lock(&capa_lock);
636 ll_delete_capa(ocapa);
637 cfs_spin_unlock(&capa_lock);
641 void ll_clear_inode_capas(struct inode *inode)
643 struct ll_inode_info *lli = ll_i2info(inode);
644 struct obd_capa *ocapa, *tmp;
646 cfs_spin_lock(&capa_lock);
647 ocapa = lli->lli_mds_capa;
649 ll_delete_capa(ocapa);
651 cfs_list_for_each_entry_safe(ocapa, tmp, &lli->lli_oss_capas,
653 ll_delete_capa(ocapa);
654 cfs_spin_unlock(&capa_lock);
657 void ll_print_capa_stat(struct ll_sb_info *sbi)
659 if (sbi->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
660 LCONSOLE_INFO("Fid capabilities renewed: %llu\n"
661 "Fid capabilities renewal ENOENT: %llu\n"
662 "Fid capabilities failed to renew: %llu\n"
663 "Fid capabilities renewal retries: %llu\n",
664 ll_capa_renewed, ll_capa_renewal_noent,
665 ll_capa_renewal_failed, ll_capa_renewal_retries);