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) 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/llite/llite_capa.c
38 * Author: Lai Siyao <lsy@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <linux/uaccess.h>
45 #include <linux/file.h>
46 #include <linux/kmod.h>
48 #include "../include/lustre_lite.h"
49 #include "llite_internal.h"
51 /* for obd_capa.c_list, client capa might stay in three places:
54 * 3. stand alone: just allocated.
57 /* capas for oss writeback and those failed to renew */
58 static LIST_HEAD(ll_idle_capas
);
59 static struct ptlrpc_thread ll_capa_thread
;
60 static struct list_head
*ll_capa_list
= &capa_list
[CAPA_SITE_CLIENT
];
62 /* llite capa renewal timer */
63 struct timer_list ll_capa_timer
;
64 /* for debug: indicate whether capa on llite is enabled or not */
65 static atomic_t ll_capa_debug
= ATOMIC_INIT(0);
66 static unsigned long long ll_capa_renewed
;
67 static unsigned long long ll_capa_renewal_noent
;
68 static unsigned long long ll_capa_renewal_failed
;
69 static unsigned long long ll_capa_renewal_retries
;
71 static int ll_update_capa(struct obd_capa
*ocapa
, struct lustre_capa
*capa
);
73 static inline void update_capa_timer(struct obd_capa
*ocapa
, unsigned long expiry
)
75 if (time_before(expiry
, ll_capa_timer
.expires
) ||
76 !timer_pending(&ll_capa_timer
)) {
77 mod_timer(&ll_capa_timer
, expiry
);
78 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
,
79 "ll_capa_timer update: %lu/%lu by", expiry
, jiffies
);
83 static inline unsigned long capa_renewal_time(struct obd_capa
*ocapa
)
85 return cfs_time_sub(ocapa
->c_expiry
,
86 cfs_time_seconds(ocapa
->c_capa
.lc_timeout
) / 2);
89 static inline int capa_is_to_expire(struct obd_capa
*ocapa
)
91 return time_before_eq(capa_renewal_time(ocapa
), cfs_time_current());
94 static inline int have_expired_capa(void)
96 struct obd_capa
*ocapa
= NULL
;
99 /* if ll_capa_list has client capa to expire or ll_idle_capas has
100 * expired capa, return 1.
102 spin_lock(&capa_lock
);
103 if (!list_empty(ll_capa_list
)) {
104 ocapa
= list_entry(ll_capa_list
->next
, struct obd_capa
,
106 expired
= capa_is_to_expire(ocapa
);
108 update_capa_timer(ocapa
, capa_renewal_time(ocapa
));
109 } else if (!list_empty(&ll_idle_capas
)) {
110 ocapa
= list_entry(ll_idle_capas
.next
, struct obd_capa
,
112 expired
= capa_is_expired(ocapa
);
114 update_capa_timer(ocapa
, ocapa
->c_expiry
);
116 spin_unlock(&capa_lock
);
119 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
, "expired");
123 static void sort_add_capa(struct obd_capa
*ocapa
, struct list_head
*head
)
125 struct obd_capa
*tmp
;
126 struct list_head
*before
= NULL
;
128 /* TODO: client capa is sorted by expiry, this could be optimized */
129 list_for_each_entry_reverse(tmp
, head
, c_list
) {
130 if (cfs_time_aftereq(ocapa
->c_expiry
, tmp
->c_expiry
)) {
131 before
= &tmp
->c_list
;
136 LASSERT(&ocapa
->c_list
!= before
);
137 list_add(&ocapa
->c_list
, before
?: head
);
140 static inline int obd_capa_open_count(struct obd_capa
*oc
)
142 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 };
177 thread_set_flags(&ll_capa_thread
, SVC_RUNNING
);
178 wake_up(&ll_capa_thread
.t_ctl_waitq
);
181 l_wait_event(ll_capa_thread
.t_ctl_waitq
,
182 !thread_is_running(&ll_capa_thread
) ||
186 if (!thread_is_running(&ll_capa_thread
))
191 spin_lock(&capa_lock
);
192 list_for_each_entry_safe(ocapa
, tmp
, ll_capa_list
, c_list
) {
195 LASSERT(ocapa
->c_capa
.lc_opc
!= CAPA_OPC_OSS_TRUNC
);
197 if (!capa_is_to_expire(ocapa
)) {
202 list_del_init(&ocapa
->c_list
);
204 /* for MDS capability, only renew those which belong to
205 * dir, or its inode is opened, or client holds LOOKUP
208 /* ibits may be changed by ll_have_md_lock() so we have
209 * to set it each time */
210 ibits
= MDS_INODELOCK_LOOKUP
;
211 if (capa_for_mds(&ocapa
->c_capa
) &&
212 !S_ISDIR(ocapa
->u
.cli
.inode
->i_mode
) &&
213 obd_capa_open_count(ocapa
) == 0 &&
214 !ll_have_md_lock(ocapa
->u
.cli
.inode
,
215 &ibits
, LCK_MINMODE
)) {
216 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
,
218 sort_add_capa(ocapa
, &ll_idle_capas
);
222 /* for OSS capability, only renew those whose inode is
225 if (capa_for_oss(&ocapa
->c_capa
) &&
226 obd_capa_open_count(ocapa
) == 0) {
227 /* oss capa with open count == 0 won't renew,
228 * move to idle list */
229 sort_add_capa(ocapa
, &ll_idle_capas
);
233 /* NB iput() is in ll_update_capa() */
234 inode
= igrab(ocapa
->u
.cli
.inode
);
236 DEBUG_CAPA(D_ERROR
, &ocapa
->c_capa
,
243 spin_unlock(&capa_lock
);
244 rc
= md_renew_capa(ll_i2mdexp(inode
), ocapa
,
246 spin_lock(&capa_lock
);
248 DEBUG_CAPA(D_ERROR
, &ocapa
->c_capa
,
249 "renew failed: %d", rc
);
250 ll_capa_renewal_failed
++;
255 update_capa_timer(next
, capa_renewal_time(next
));
257 list_for_each_entry_safe(ocapa
, tmp
, &ll_idle_capas
,
259 if (!capa_is_expired(ocapa
)) {
261 update_capa_timer(ocapa
,
266 if (atomic_read(&ocapa
->c_refc
) > 1) {
267 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
,
268 "expired(c_refc %d), don't release",
269 atomic_read(&ocapa
->c_refc
));
270 /* don't try to renew any more */
271 list_del_init(&ocapa
->c_list
);
275 /* expired capa is released. */
276 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
, "release expired");
277 ll_delete_capa(ocapa
);
280 spin_unlock(&capa_lock
);
283 thread_set_flags(&ll_capa_thread
, SVC_STOPPED
);
284 wake_up(&ll_capa_thread
.t_ctl_waitq
);
288 void ll_capa_timer_callback(unsigned long unused
)
290 wake_up(&ll_capa_thread
.t_ctl_waitq
);
293 int ll_capa_thread_start(void)
295 struct task_struct
*task
;
297 init_waitqueue_head(&ll_capa_thread
.t_ctl_waitq
);
299 task
= kthread_run(capa_thread_main
, NULL
, "ll_capa");
301 CERROR("cannot start expired capa thread: rc %ld\n",
303 return PTR_ERR(task
);
305 wait_event(ll_capa_thread
.t_ctl_waitq
,
306 thread_is_running(&ll_capa_thread
));
311 void ll_capa_thread_stop(void)
313 thread_set_flags(&ll_capa_thread
, SVC_STOPPING
);
314 wake_up(&ll_capa_thread
.t_ctl_waitq
);
315 wait_event(ll_capa_thread
.t_ctl_waitq
,
316 thread_is_stopped(&ll_capa_thread
));
319 struct obd_capa
*ll_osscapa_get(struct inode
*inode
, __u64 opc
)
321 struct ll_inode_info
*lli
= ll_i2info(inode
);
322 struct obd_capa
*ocapa
;
325 if ((ll_i2sbi(inode
)->ll_flags
& LL_SBI_OSS_CAPA
) == 0)
328 LASSERT(opc
== CAPA_OPC_OSS_WRITE
|| opc
== CAPA_OPC_OSS_RW
||
329 opc
== CAPA_OPC_OSS_TRUNC
);
331 spin_lock(&capa_lock
);
332 list_for_each_entry(ocapa
, &lli
->lli_oss_capas
, u
.cli
.lli_list
) {
333 if (capa_is_expired(ocapa
))
335 if ((opc
& CAPA_OPC_OSS_WRITE
) &&
336 capa_opc_supported(&ocapa
->c_capa
, CAPA_OPC_OSS_WRITE
)) {
339 } else if ((opc
& CAPA_OPC_OSS_READ
) &&
340 capa_opc_supported(&ocapa
->c_capa
,
341 CAPA_OPC_OSS_READ
)) {
344 } else if ((opc
& CAPA_OPC_OSS_TRUNC
) &&
345 capa_opc_supported(&ocapa
->c_capa
, opc
)) {
352 LASSERT(lu_fid_eq(capa_fid(&ocapa
->c_capa
),
353 ll_inode2fid(inode
)));
354 LASSERT(ocapa
->c_site
== CAPA_SITE_CLIENT
);
358 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
, "found client");
362 if (atomic_read(&ll_capa_debug
)) {
363 CERROR("no capability for "DFID
" opc %#llx\n",
364 PFID(&lli
->lli_fid
), opc
);
365 atomic_set(&ll_capa_debug
, 0);
368 spin_unlock(&capa_lock
);
372 EXPORT_SYMBOL(ll_osscapa_get
);
374 struct obd_capa
*ll_mdscapa_get(struct inode
*inode
)
376 struct ll_inode_info
*lli
= ll_i2info(inode
);
377 struct obd_capa
*ocapa
;
379 LASSERT(inode
!= NULL
);
381 if ((ll_i2sbi(inode
)->ll_flags
& LL_SBI_MDS_CAPA
) == 0)
384 spin_lock(&capa_lock
);
385 ocapa
= capa_get(lli
->lli_mds_capa
);
386 spin_unlock(&capa_lock
);
387 if (!ocapa
&& atomic_read(&ll_capa_debug
)) {
388 CERROR("no mds capability for "DFID
"\n", PFID(&lli
->lli_fid
));
389 atomic_set(&ll_capa_debug
, 0);
395 static struct obd_capa
*do_add_mds_capa(struct inode
*inode
,
396 struct obd_capa
*ocapa
)
398 struct ll_inode_info
*lli
= ll_i2info(inode
);
399 struct obd_capa
*old
= lli
->lli_mds_capa
;
400 struct lustre_capa
*capa
= &ocapa
->c_capa
;
403 ocapa
->u
.cli
.inode
= inode
;
404 lli
->lli_mds_capa
= ocapa
;
405 capa_count
[CAPA_SITE_CLIENT
]++;
407 DEBUG_CAPA(D_SEC
, capa
, "add MDS");
409 spin_lock(&old
->c_lock
);
411 spin_unlock(&old
->c_lock
);
413 DEBUG_CAPA(D_SEC
, capa
, "update MDS");
421 static struct obd_capa
*do_lookup_oss_capa(struct inode
*inode
, int opc
)
423 struct ll_inode_info
*lli
= ll_i2info(inode
);
424 struct obd_capa
*ocapa
;
426 /* inside capa_lock */
427 list_for_each_entry(ocapa
, &lli
->lli_oss_capas
, u
.cli
.lli_list
) {
428 if ((capa_opc(&ocapa
->c_capa
) & opc
) != opc
)
431 LASSERT(lu_fid_eq(capa_fid(&ocapa
->c_capa
),
432 ll_inode2fid(inode
)));
433 LASSERT(ocapa
->c_site
== CAPA_SITE_CLIENT
);
435 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
, "found client");
442 static inline void inode_add_oss_capa(struct inode
*inode
,
443 struct obd_capa
*ocapa
)
445 struct ll_inode_info
*lli
= ll_i2info(inode
);
446 struct obd_capa
*tmp
;
447 struct list_head
*next
= NULL
;
449 /* capa is sorted in lli_oss_capas so lookup can always find the
451 list_for_each_entry(tmp
, &lli
->lli_oss_capas
, u
.cli
.lli_list
) {
452 if (cfs_time_after(ocapa
->c_expiry
, tmp
->c_expiry
)) {
453 next
= &tmp
->u
.cli
.lli_list
;
457 LASSERT(&ocapa
->u
.cli
.lli_list
!= next
);
458 list_move_tail(&ocapa
->u
.cli
.lli_list
, next
?: &lli
->lli_oss_capas
);
461 static struct obd_capa
*do_add_oss_capa(struct inode
*inode
,
462 struct obd_capa
*ocapa
)
464 struct obd_capa
*old
;
465 struct lustre_capa
*capa
= &ocapa
->c_capa
;
467 LASSERTF(S_ISREG(inode
->i_mode
),
468 "inode has oss capa, but not regular file, mode: %d\n",
471 /* FIXME: can't replace it so easily with fine-grained opc */
472 old
= do_lookup_oss_capa(inode
, capa_opc(capa
) & CAPA_OPC_OSS_ONLY
);
474 ocapa
->u
.cli
.inode
= inode
;
475 INIT_LIST_HEAD(&ocapa
->u
.cli
.lli_list
);
476 capa_count
[CAPA_SITE_CLIENT
]++;
478 DEBUG_CAPA(D_SEC
, capa
, "add OSS");
480 spin_lock(&old
->c_lock
);
482 spin_unlock(&old
->c_lock
);
484 DEBUG_CAPA(D_SEC
, capa
, "update OSS");
490 inode_add_oss_capa(inode
, ocapa
);
494 struct obd_capa
*ll_add_capa(struct inode
*inode
, struct obd_capa
*ocapa
)
496 spin_lock(&capa_lock
);
497 ocapa
= capa_for_mds(&ocapa
->c_capa
) ? do_add_mds_capa(inode
, ocapa
) :
498 do_add_oss_capa(inode
, ocapa
);
500 /* truncate capa won't renew */
501 if (ocapa
->c_capa
.lc_opc
!= CAPA_OPC_OSS_TRUNC
) {
502 set_capa_expiry(ocapa
);
503 list_del_init(&ocapa
->c_list
);
504 sort_add_capa(ocapa
, ll_capa_list
);
506 update_capa_timer(ocapa
, capa_renewal_time(ocapa
));
509 spin_unlock(&capa_lock
);
511 atomic_set(&ll_capa_debug
, 1);
515 static inline void delay_capa_renew(struct obd_capa
*oc
, unsigned long delay
)
517 /* NB: set a fake expiry for this capa to prevent it renew too soon */
518 oc
->c_expiry
= cfs_time_add(oc
->c_expiry
, cfs_time_seconds(delay
));
521 static int ll_update_capa(struct obd_capa
*ocapa
, struct lustre_capa
*capa
)
523 struct inode
*inode
= ocapa
->u
.cli
.inode
;
531 spin_lock(&capa_lock
);
533 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
,
534 "renewal canceled because object removed");
535 ll_capa_renewal_noent
++;
537 ll_capa_renewal_failed
++;
539 /* failed capa won't be renewed any longer, but if -EIO,
540 * client might be doing recovery, retry in 2 min. */
541 if (rc
== -EIO
&& !capa_is_expired(ocapa
)) {
542 delay_capa_renew(ocapa
, 120);
543 DEBUG_CAPA(D_ERROR
, &ocapa
->c_capa
,
544 "renewal failed: -EIO, retry in 2 mins");
545 ll_capa_renewal_retries
++;
548 DEBUG_CAPA(D_ERROR
, &ocapa
->c_capa
,
549 "renewal failed(rc: %d) for", rc
);
553 list_del_init(&ocapa
->c_list
);
554 sort_add_capa(ocapa
, &ll_idle_capas
);
555 spin_unlock(&capa_lock
);
562 spin_lock(&ocapa
->c_lock
);
563 LASSERT(!memcmp(&ocapa
->c_capa
, capa
,
564 offsetof(struct lustre_capa
, lc_opc
)));
565 ocapa
->c_capa
= *capa
;
566 set_capa_expiry(ocapa
);
567 spin_unlock(&ocapa
->c_lock
);
569 spin_lock(&capa_lock
);
570 if (capa_for_oss(capa
))
571 inode_add_oss_capa(inode
, ocapa
);
572 DEBUG_CAPA(D_SEC
, capa
, "renew");
574 list_del_init(&ocapa
->c_list
);
575 sort_add_capa(ocapa
, ll_capa_list
);
576 update_capa_timer(ocapa
, capa_renewal_time(ocapa
));
577 spin_unlock(&capa_lock
);
584 void ll_capa_open(struct inode
*inode
)
586 struct ll_inode_info
*lli
= ll_i2info(inode
);
588 if ((ll_i2sbi(inode
)->ll_flags
& (LL_SBI_MDS_CAPA
| LL_SBI_OSS_CAPA
))
592 if (!S_ISREG(inode
->i_mode
))
595 atomic_inc(&lli
->lli_open_count
);
598 void ll_capa_close(struct inode
*inode
)
600 struct ll_inode_info
*lli
= ll_i2info(inode
);
602 if ((ll_i2sbi(inode
)->ll_flags
& (LL_SBI_MDS_CAPA
| LL_SBI_OSS_CAPA
))
606 if (!S_ISREG(inode
->i_mode
))
609 atomic_dec(&lli
->lli_open_count
);
612 /* delete CAPA_OPC_OSS_TRUNC only */
613 void ll_truncate_free_capa(struct obd_capa
*ocapa
)
618 LASSERT(ocapa
->c_capa
.lc_opc
& CAPA_OPC_OSS_TRUNC
);
619 DEBUG_CAPA(D_SEC
, &ocapa
->c_capa
, "free truncate");
621 /* release ref when find */
623 if (likely(ocapa
->c_capa
.lc_opc
== CAPA_OPC_OSS_TRUNC
)) {
624 spin_lock(&capa_lock
);
625 ll_delete_capa(ocapa
);
626 spin_unlock(&capa_lock
);
630 void ll_clear_inode_capas(struct inode
*inode
)
632 struct ll_inode_info
*lli
= ll_i2info(inode
);
633 struct obd_capa
*ocapa
, *tmp
;
635 spin_lock(&capa_lock
);
636 ocapa
= lli
->lli_mds_capa
;
638 ll_delete_capa(ocapa
);
640 list_for_each_entry_safe(ocapa
, tmp
, &lli
->lli_oss_capas
,
642 ll_delete_capa(ocapa
);
643 spin_unlock(&capa_lock
);
646 void ll_print_capa_stat(struct ll_sb_info
*sbi
)
648 if (sbi
->ll_flags
& (LL_SBI_MDS_CAPA
| LL_SBI_OSS_CAPA
))
649 LCONSOLE_INFO("Fid capabilities renewed: %llu\n"
650 "Fid capabilities renewal ENOENT: %llu\n"
651 "Fid capabilities failed to renew: %llu\n"
652 "Fid capabilities renewal retries: %llu\n",
653 ll_capa_renewed
, ll_capa_renewal_noent
,
654 ll_capa_renewal_failed
, ll_capa_renewal_retries
);