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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 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/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 struct list_head ll_idle_capas = LIST_HEAD_INIT(ll_idle_capas);
60 static struct ptlrpc_thread ll_capa_thread;
61 static struct list_head *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 atomic_t ll_capa_debug = 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 int ll_update_capa(struct obd_capa *ocapa, struct lustre_capa *capa);
74 static inline void update_capa_timer(struct obd_capa *ocapa, cfs_time_t expiry)
76 if (cfs_time_before(expiry, ll_capa_timer.expires) ||
77 !timer_pending(&ll_capa_timer)) {
78 mod_timer(&ll_capa_timer, expiry);
79 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
80 "ll_capa_timer update: %lu/%lu by", expiry, jiffies);
84 static inline cfs_time_t capa_renewal_time(struct obd_capa *ocapa)
86 return cfs_time_sub(ocapa->c_expiry,
87 cfs_time_seconds(ocapa->c_capa.lc_timeout) / 2);
90 static inline int capa_is_to_expire(struct obd_capa *ocapa)
92 return cfs_time_beforeq(capa_renewal_time(ocapa), cfs_time_current());
95 static inline int have_expired_capa(void)
97 struct obd_capa *ocapa = NULL;
100 /* if ll_capa_list has client capa to expire or ll_idle_capas has
101 * expired capa, return 1.
103 spin_lock(&capa_lock);
104 if (!list_empty(ll_capa_list)) {
105 ocapa = list_entry(ll_capa_list->next, struct obd_capa,
107 expired = capa_is_to_expire(ocapa);
109 update_capa_timer(ocapa, capa_renewal_time(ocapa));
110 } else if (!list_empty(&ll_idle_capas)) {
111 ocapa = list_entry(ll_idle_capas.next, struct obd_capa,
113 expired = capa_is_expired(ocapa);
115 update_capa_timer(ocapa, ocapa->c_expiry);
117 spin_unlock(&capa_lock);
120 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "expired");
124 static void sort_add_capa(struct obd_capa *ocapa, struct list_head *head)
126 struct obd_capa *tmp;
127 struct list_head *before = NULL;
129 /* TODO: client capa is sorted by expiry, this could be optimized */
130 list_for_each_entry_reverse(tmp, head, c_list) {
131 if (cfs_time_aftereq(ocapa->c_expiry, tmp->c_expiry)) {
132 before = &tmp->c_list;
137 LASSERT(&ocapa->c_list != before);
138 list_add(&ocapa->c_list, before ?: head);
141 static inline int obd_capa_open_count(struct obd_capa *oc)
143 struct ll_inode_info *lli = ll_i2info(oc->u.cli.inode);
144 return atomic_read(&lli->lli_open_count);
147 static void ll_delete_capa(struct obd_capa *ocapa)
149 struct ll_inode_info *lli = ll_i2info(ocapa->u.cli.inode);
151 if (capa_for_mds(&ocapa->c_capa)) {
152 LASSERT(lli->lli_mds_capa == ocapa);
153 lli->lli_mds_capa = NULL;
154 } else if (capa_for_oss(&ocapa->c_capa)) {
155 list_del_init(&ocapa->u.cli.lli_list);
158 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "free client");
159 list_del_init(&ocapa->c_list);
160 capa_count[CAPA_SITE_CLIENT]--;
161 /* release the ref when alloc */
165 /* three places where client capa is deleted:
166 * 1. capa_thread_main(), main place to delete expired capa.
167 * 2. ll_clear_inode_capas() in ll_clear_inode().
168 * 3. ll_truncate_free_capa() delete truncate capa explicitly in ll_setattr_ost().
170 static int capa_thread_main(void *unused)
172 struct obd_capa *ocapa, *tmp, *next;
173 struct inode *inode = NULL;
174 struct l_wait_info lwi = { 0 };
178 thread_set_flags(&ll_capa_thread, SVC_RUNNING);
179 wake_up(&ll_capa_thread.t_ctl_waitq);
182 l_wait_event(ll_capa_thread.t_ctl_waitq,
183 !thread_is_running(&ll_capa_thread) ||
187 if (!thread_is_running(&ll_capa_thread))
192 spin_lock(&capa_lock);
193 list_for_each_entry_safe(ocapa, tmp, ll_capa_list, c_list) {
196 LASSERT(ocapa->c_capa.lc_opc != CAPA_OPC_OSS_TRUNC);
198 if (!capa_is_to_expire(ocapa)) {
203 list_del_init(&ocapa->c_list);
205 /* for MDS capability, only renew those which belong to
206 * dir, or its inode is opened, or client holds LOOKUP
209 /* ibits may be changed by ll_have_md_lock() so we have
210 * to set it each time */
211 ibits = MDS_INODELOCK_LOOKUP;
212 if (capa_for_mds(&ocapa->c_capa) &&
213 !S_ISDIR(ocapa->u.cli.inode->i_mode) &&
214 obd_capa_open_count(ocapa) == 0 &&
215 !ll_have_md_lock(ocapa->u.cli.inode,
216 &ibits, LCK_MINMODE)) {
217 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
219 sort_add_capa(ocapa, &ll_idle_capas);
223 /* for OSS capability, only renew those whose inode is
226 if (capa_for_oss(&ocapa->c_capa) &&
227 obd_capa_open_count(ocapa) == 0) {
228 /* oss capa with open count == 0 won't renew,
229 * move to idle list */
230 sort_add_capa(ocapa, &ll_idle_capas);
234 /* NB iput() is in ll_update_capa() */
235 inode = igrab(ocapa->u.cli.inode);
237 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
244 spin_unlock(&capa_lock);
245 rc = md_renew_capa(ll_i2mdexp(inode), ocapa,
247 spin_lock(&capa_lock);
249 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
250 "renew failed: %d", rc);
251 ll_capa_renewal_failed++;
256 update_capa_timer(next, capa_renewal_time(next));
258 list_for_each_entry_safe(ocapa, tmp, &ll_idle_capas,
260 if (!capa_is_expired(ocapa)) {
262 update_capa_timer(ocapa,
267 if (atomic_read(&ocapa->c_refc) > 1) {
268 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
269 "expired(c_refc %d), don't release",
270 atomic_read(&ocapa->c_refc));
271 /* don't try to renew any more */
272 list_del_init(&ocapa->c_list);
276 /* expired capa is released. */
277 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "release expired");
278 ll_delete_capa(ocapa);
281 spin_unlock(&capa_lock);
284 thread_set_flags(&ll_capa_thread, SVC_STOPPED);
285 wake_up(&ll_capa_thread.t_ctl_waitq);
289 void ll_capa_timer_callback(unsigned long unused)
291 wake_up(&ll_capa_thread.t_ctl_waitq);
294 int ll_capa_thread_start(void)
296 struct task_struct *task;
299 init_waitqueue_head(&ll_capa_thread.t_ctl_waitq);
301 task = kthread_run(capa_thread_main, NULL, "ll_capa");
303 CERROR("cannot start expired capa thread: rc %ld\n",
305 RETURN(PTR_ERR(task));
307 wait_event(ll_capa_thread.t_ctl_waitq,
308 thread_is_running(&ll_capa_thread));
313 void ll_capa_thread_stop(void)
315 thread_set_flags(&ll_capa_thread, SVC_STOPPING);
316 wake_up(&ll_capa_thread.t_ctl_waitq);
317 wait_event(ll_capa_thread.t_ctl_waitq,
318 thread_is_stopped(&ll_capa_thread));
321 struct obd_capa *ll_osscapa_get(struct inode *inode, __u64 opc)
323 struct ll_inode_info *lli = ll_i2info(inode);
324 struct obd_capa *ocapa;
329 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_OSS_CAPA) == 0)
332 LASSERT(opc == CAPA_OPC_OSS_WRITE || opc == CAPA_OPC_OSS_RW ||
333 opc == CAPA_OPC_OSS_TRUNC);
335 spin_lock(&capa_lock);
336 list_for_each_entry(ocapa, &lli->lli_oss_capas, u.cli.lli_list) {
337 if (capa_is_expired(ocapa))
339 if ((opc & CAPA_OPC_OSS_WRITE) &&
340 capa_opc_supported(&ocapa->c_capa, CAPA_OPC_OSS_WRITE)) {
343 } else if ((opc & CAPA_OPC_OSS_READ) &&
344 capa_opc_supported(&ocapa->c_capa,
345 CAPA_OPC_OSS_READ)) {
348 } else if ((opc & CAPA_OPC_OSS_TRUNC) &&
349 capa_opc_supported(&ocapa->c_capa, opc)) {
356 LASSERT(lu_fid_eq(capa_fid(&ocapa->c_capa),
357 ll_inode2fid(inode)));
358 LASSERT(ocapa->c_site == CAPA_SITE_CLIENT);
362 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "found client");
366 if (atomic_read(&ll_capa_debug)) {
367 CERROR("no capability for "DFID" opc "LPX64"\n",
368 PFID(&lli->lli_fid), opc);
369 atomic_set(&ll_capa_debug, 0);
372 spin_unlock(&capa_lock);
376 EXPORT_SYMBOL(ll_osscapa_get);
378 struct obd_capa *ll_mdscapa_get(struct inode *inode)
380 struct ll_inode_info *lli = ll_i2info(inode);
381 struct obd_capa *ocapa;
384 LASSERT(inode != NULL);
386 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_MDS_CAPA) == 0)
389 spin_lock(&capa_lock);
390 ocapa = capa_get(lli->lli_mds_capa);
391 spin_unlock(&capa_lock);
392 if (!ocapa && atomic_read(&ll_capa_debug)) {
393 CERROR("no mds capability for "DFID"\n", PFID(&lli->lli_fid));
394 atomic_set(&ll_capa_debug, 0);
400 static struct obd_capa *do_add_mds_capa(struct inode *inode,
401 struct obd_capa *ocapa)
403 struct ll_inode_info *lli = ll_i2info(inode);
404 struct obd_capa *old = lli->lli_mds_capa;
405 struct lustre_capa *capa = &ocapa->c_capa;
408 ocapa->u.cli.inode = inode;
409 lli->lli_mds_capa = ocapa;
410 capa_count[CAPA_SITE_CLIENT]++;
412 DEBUG_CAPA(D_SEC, capa, "add MDS");
414 spin_lock(&old->c_lock);
416 spin_unlock(&old->c_lock);
418 DEBUG_CAPA(D_SEC, capa, "update MDS");
426 static struct obd_capa *do_lookup_oss_capa(struct inode *inode, int opc)
428 struct ll_inode_info *lli = ll_i2info(inode);
429 struct obd_capa *ocapa;
431 /* inside capa_lock */
432 list_for_each_entry(ocapa, &lli->lli_oss_capas, u.cli.lli_list) {
433 if ((capa_opc(&ocapa->c_capa) & opc) != opc)
436 LASSERT(lu_fid_eq(capa_fid(&ocapa->c_capa),
437 ll_inode2fid(inode)));
438 LASSERT(ocapa->c_site == CAPA_SITE_CLIENT);
440 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "found client");
447 static inline void inode_add_oss_capa(struct inode *inode,
448 struct obd_capa *ocapa)
450 struct ll_inode_info *lli = ll_i2info(inode);
451 struct obd_capa *tmp;
452 struct list_head *next = NULL;
454 /* capa is sorted in lli_oss_capas so lookup can always find the
456 list_for_each_entry(tmp, &lli->lli_oss_capas, u.cli.lli_list) {
457 if (cfs_time_after(ocapa->c_expiry, tmp->c_expiry)) {
458 next = &tmp->u.cli.lli_list;
462 LASSERT(&ocapa->u.cli.lli_list != next);
463 list_move_tail(&ocapa->u.cli.lli_list, next ?: &lli->lli_oss_capas);
466 static struct obd_capa *do_add_oss_capa(struct inode *inode,
467 struct obd_capa *ocapa)
469 struct obd_capa *old;
470 struct lustre_capa *capa = &ocapa->c_capa;
472 LASSERTF(S_ISREG(inode->i_mode),
473 "inode has oss capa, but not regular file, mode: %d\n",
476 /* FIXME: can't replace it so easily with fine-grained opc */
477 old = do_lookup_oss_capa(inode, capa_opc(capa) & CAPA_OPC_OSS_ONLY);
479 ocapa->u.cli.inode = inode;
480 INIT_LIST_HEAD(&ocapa->u.cli.lli_list);
481 capa_count[CAPA_SITE_CLIENT]++;
483 DEBUG_CAPA(D_SEC, capa, "add OSS");
485 spin_lock(&old->c_lock);
487 spin_unlock(&old->c_lock);
489 DEBUG_CAPA(D_SEC, capa, "update OSS");
495 inode_add_oss_capa(inode, ocapa);
499 struct obd_capa *ll_add_capa(struct inode *inode, struct obd_capa *ocapa)
501 spin_lock(&capa_lock);
502 ocapa = capa_for_mds(&ocapa->c_capa) ? do_add_mds_capa(inode, ocapa) :
503 do_add_oss_capa(inode, ocapa);
505 /* truncate capa won't renew */
506 if (ocapa->c_capa.lc_opc != CAPA_OPC_OSS_TRUNC) {
507 set_capa_expiry(ocapa);
508 list_del_init(&ocapa->c_list);
509 sort_add_capa(ocapa, ll_capa_list);
511 update_capa_timer(ocapa, capa_renewal_time(ocapa));
514 spin_unlock(&capa_lock);
516 atomic_set(&ll_capa_debug, 1);
520 static inline void delay_capa_renew(struct obd_capa *oc, cfs_time_t delay)
522 /* NB: set a fake expiry for this capa to prevent it renew too soon */
523 oc->c_expiry = cfs_time_add(oc->c_expiry, cfs_time_seconds(delay));
526 static int ll_update_capa(struct obd_capa *ocapa, struct lustre_capa *capa)
528 struct inode *inode = ocapa->u.cli.inode;
537 spin_lock(&capa_lock);
539 DEBUG_CAPA(D_SEC, &ocapa->c_capa,
540 "renewal canceled because object removed");
541 ll_capa_renewal_noent++;
543 ll_capa_renewal_failed++;
545 /* failed capa won't be renewed any longer, but if -EIO,
546 * client might be doing recovery, retry in 2 min. */
547 if (rc == -EIO && !capa_is_expired(ocapa)) {
548 delay_capa_renew(ocapa, 120);
549 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
550 "renewal failed: -EIO, "
552 ll_capa_renewal_retries++;
555 DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
556 "renewal failed(rc: %d) for", rc);
560 list_del_init(&ocapa->c_list);
561 sort_add_capa(ocapa, &ll_idle_capas);
562 spin_unlock(&capa_lock);
569 spin_lock(&ocapa->c_lock);
570 LASSERT(!memcmp(&ocapa->c_capa, capa,
571 offsetof(struct lustre_capa, lc_opc)));
572 ocapa->c_capa = *capa;
573 set_capa_expiry(ocapa);
574 spin_unlock(&ocapa->c_lock);
576 spin_lock(&capa_lock);
577 if (capa_for_oss(capa))
578 inode_add_oss_capa(inode, ocapa);
579 DEBUG_CAPA(D_SEC, capa, "renew");
582 list_del_init(&ocapa->c_list);
583 sort_add_capa(ocapa, ll_capa_list);
584 update_capa_timer(ocapa, capa_renewal_time(ocapa));
585 spin_unlock(&capa_lock);
592 void ll_capa_open(struct inode *inode)
594 struct ll_inode_info *lli = ll_i2info(inode);
596 if ((ll_i2sbi(inode)->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
600 if (!S_ISREG(inode->i_mode))
603 atomic_inc(&lli->lli_open_count);
606 void ll_capa_close(struct inode *inode)
608 struct ll_inode_info *lli = ll_i2info(inode);
610 if ((ll_i2sbi(inode)->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
614 if (!S_ISREG(inode->i_mode))
617 atomic_dec(&lli->lli_open_count);
620 /* delete CAPA_OPC_OSS_TRUNC only */
621 void ll_truncate_free_capa(struct obd_capa *ocapa)
626 LASSERT(ocapa->c_capa.lc_opc & CAPA_OPC_OSS_TRUNC);
627 DEBUG_CAPA(D_SEC, &ocapa->c_capa, "free truncate");
629 /* release ref when find */
631 if (likely(ocapa->c_capa.lc_opc == CAPA_OPC_OSS_TRUNC)) {
632 spin_lock(&capa_lock);
633 ll_delete_capa(ocapa);
634 spin_unlock(&capa_lock);
638 void ll_clear_inode_capas(struct inode *inode)
640 struct ll_inode_info *lli = ll_i2info(inode);
641 struct obd_capa *ocapa, *tmp;
643 spin_lock(&capa_lock);
644 ocapa = lli->lli_mds_capa;
646 ll_delete_capa(ocapa);
648 list_for_each_entry_safe(ocapa, tmp, &lli->lli_oss_capas,
650 ll_delete_capa(ocapa);
651 spin_unlock(&capa_lock);
654 void ll_print_capa_stat(struct ll_sb_info *sbi)
656 if (sbi->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
657 LCONSOLE_INFO("Fid capabilities renewed: %llu\n"
658 "Fid capabilities renewal ENOENT: %llu\n"
659 "Fid capabilities failed to renew: %llu\n"
660 "Fid capabilities renewal retries: %llu\n",
661 ll_capa_renewed, ll_capa_renewal_noent,
662 ll_capa_renewal_failed, ll_capa_renewal_retries);