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, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ptlrpc/sec_bulk.c
34 * Author: Eric Mei <ericm@clusterfs.com>
37 #define DEBUG_SUBSYSTEM S_SEC
39 #include <libcfs/libcfs.h>
42 #include <obd_cksum.h>
43 #include <obd_class.h>
44 #include <obd_support.h>
45 #include <lustre_net.h>
46 #include <lustre_import.h>
47 #include <lustre_dlm.h>
48 #include <lustre_sec.h>
50 #include "ptlrpc_internal.h"
52 static int mult = 20 - PAGE_SHIFT;
53 static int enc_pool_max_memory_mb;
54 module_param(enc_pool_max_memory_mb, int, 0644);
55 MODULE_PARM_DESC(enc_pool_max_memory_mb,
56 "Encoding pool max memory (MB), 1/8 of total physical memory by default");
58 /****************************************
59 * bulk encryption page pools *
60 ****************************************/
63 #define PTRS_PER_PAGE (PAGE_SIZE / sizeof(void *))
64 #define PAGES_PER_POOL (PTRS_PER_PAGE)
66 #define IDLE_IDX_MAX (100)
67 #define IDLE_IDX_WEIGHT (3)
69 #define CACHE_QUIESCENT_PERIOD (20)
71 static struct ptlrpc_enc_page_pool {
75 unsigned long epp_max_pages; /* maximum pages can hold, const */
76 unsigned int epp_max_pools; /* number of pools, const */
79 * wait queue in case of not enough free pages.
81 wait_queue_head_t epp_waitq; /* waiting threads */
82 unsigned int epp_waitqlen; /* wait queue length */
83 unsigned long epp_pages_short; /* # of pages wanted of in-q users */
84 unsigned int epp_growing:1; /* during adding pages */
87 * indicating how idle the pools are, from 0 to MAX_IDLE_IDX
88 * this is counted based on each time when getting pages from
89 * the pools, not based on time. which means in case that system
90 * is idled for a while but the idle_idx might still be low if no
91 * activities happened in the pools.
93 unsigned long epp_idle_idx;
95 /* last shrink time due to mem tight */
96 time64_t epp_last_shrink;
97 time64_t epp_last_access;
100 * in-pool pages bookkeeping
102 spinlock_t epp_lock; /* protect following fields */
103 unsigned long epp_total_pages; /* total pages in pools */
104 unsigned long epp_free_pages; /* current pages available */
109 unsigned long epp_st_max_pages; /* # of pages ever reached */
110 unsigned int epp_st_grows; /* # of grows */
111 unsigned int epp_st_grow_fails; /* # of add pages failures */
112 unsigned int epp_st_shrinks; /* # of shrinks */
113 unsigned long epp_st_access; /* # of access */
114 unsigned long epp_st_missings; /* # of cache missing */
115 unsigned long epp_st_lowfree; /* lowest free pages reached */
116 unsigned int epp_st_max_wqlen; /* highest waitqueue length */
117 cfs_time_t epp_st_max_wait; /* in jeffies */
118 unsigned long epp_st_outofmem; /* # of out of mem requests */
120 * pointers to pools, may be vmalloc'd
122 struct page ***epp_pools;
128 static const int pools_shrinker_seeks = DEFAULT_SEEKS;
129 static struct shrinker *pools_shrinker;
133 * /proc/fs/lustre/sptlrpc/encrypt_page_pools
135 int sptlrpc_proc_enc_pool_seq_show(struct seq_file *m, void *v)
137 spin_lock(&page_pools.epp_lock);
139 seq_printf(m, "physical pages: %lu\n"
140 "pages per pool: %lu\n"
145 "idle index: %lu/100\n"
146 "last shrink: %lds\n"
147 "last access: %lds\n"
148 "max pages reached: %lu\n"
150 "grows failure: %u\n"
152 "cache access: %lu\n"
153 "cache missing: %lu\n"
154 "low free mark: %lu\n"
155 "max waitqueue depth: %u\n"
156 "max wait time: "CFS_TIME_T"/%lu\n"
158 totalram_pages, PAGES_PER_POOL,
159 page_pools.epp_max_pages,
160 page_pools.epp_max_pools,
161 page_pools.epp_total_pages,
162 page_pools.epp_free_pages,
163 page_pools.epp_idle_idx,
164 (long)(ktime_get_seconds() - page_pools.epp_last_shrink),
165 (long)(ktime_get_seconds() - page_pools.epp_last_access),
166 page_pools.epp_st_max_pages,
167 page_pools.epp_st_grows,
168 page_pools.epp_st_grow_fails,
169 page_pools.epp_st_shrinks,
170 page_pools.epp_st_access,
171 page_pools.epp_st_missings,
172 page_pools.epp_st_lowfree,
173 page_pools.epp_st_max_wqlen,
174 page_pools.epp_st_max_wait,
175 msecs_to_jiffies(MSEC_PER_SEC),
176 page_pools.epp_st_outofmem);
178 spin_unlock(&page_pools.epp_lock);
182 static void enc_pools_release_free_pages(long npages)
185 int p_idx_max1, p_idx_max2;
188 LASSERT(npages <= page_pools.epp_free_pages);
189 LASSERT(page_pools.epp_free_pages <= page_pools.epp_total_pages);
191 /* max pool index before the release */
192 p_idx_max2 = (page_pools.epp_total_pages - 1) / PAGES_PER_POOL;
194 page_pools.epp_free_pages -= npages;
195 page_pools.epp_total_pages -= npages;
197 /* max pool index after the release */
198 p_idx_max1 = page_pools.epp_total_pages == 0 ? -1 :
199 ((page_pools.epp_total_pages - 1) / PAGES_PER_POOL);
201 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
202 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
203 LASSERT(page_pools.epp_pools[p_idx]);
206 LASSERT(page_pools.epp_pools[p_idx]);
207 LASSERT(page_pools.epp_pools[p_idx][g_idx] != NULL);
209 __free_page(page_pools.epp_pools[p_idx][g_idx]);
210 page_pools.epp_pools[p_idx][g_idx] = NULL;
212 if (++g_idx == PAGES_PER_POOL) {
218 /* free unused pools */
219 while (p_idx_max1 < p_idx_max2) {
220 LASSERT(page_pools.epp_pools[p_idx_max2]);
221 OBD_FREE(page_pools.epp_pools[p_idx_max2], PAGE_SIZE);
222 page_pools.epp_pools[p_idx_max2] = NULL;
228 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
230 static unsigned long enc_pools_shrink_count(struct shrinker *s,
231 struct shrink_control *sc)
234 * if no pool access for a long time, we consider it's fully idle.
235 * a little race here is fine.
237 if (unlikely(ktime_get_real_seconds() - page_pools.epp_last_access >
238 CACHE_QUIESCENT_PERIOD)) {
239 spin_lock(&page_pools.epp_lock);
240 page_pools.epp_idle_idx = IDLE_IDX_MAX;
241 spin_unlock(&page_pools.epp_lock);
244 LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX);
245 return max((int)page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES, 0) *
246 (IDLE_IDX_MAX - page_pools.epp_idle_idx) / IDLE_IDX_MAX;
250 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
252 static unsigned long enc_pools_shrink_scan(struct shrinker *s,
253 struct shrink_control *sc)
255 spin_lock(&page_pools.epp_lock);
256 sc->nr_to_scan = min_t(unsigned long, sc->nr_to_scan,
257 page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES);
258 if (sc->nr_to_scan > 0) {
259 enc_pools_release_free_pages(sc->nr_to_scan);
260 CDEBUG(D_SEC, "released %ld pages, %ld left\n",
261 (long)sc->nr_to_scan, page_pools.epp_free_pages);
263 page_pools.epp_st_shrinks++;
264 page_pools.epp_last_shrink = ktime_get_real_seconds();
266 spin_unlock(&page_pools.epp_lock);
269 * if no pool access for a long time, we consider it's fully idle.
270 * a little race here is fine.
272 if (unlikely(ktime_get_real_seconds() - page_pools.epp_last_access >
273 CACHE_QUIESCENT_PERIOD)) {
274 spin_lock(&page_pools.epp_lock);
275 page_pools.epp_idle_idx = IDLE_IDX_MAX;
276 spin_unlock(&page_pools.epp_lock);
279 LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX);
280 return sc->nr_to_scan;
283 #ifndef HAVE_SHRINKER_COUNT
285 * could be called frequently for query (@nr_to_scan == 0).
286 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
288 static int enc_pools_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
290 struct shrink_control scv = {
291 .nr_to_scan = shrink_param(sc, nr_to_scan),
292 .gfp_mask = shrink_param(sc, gfp_mask)
294 #if !defined(HAVE_SHRINKER_WANT_SHRINK_PTR) && !defined(HAVE_SHRINK_CONTROL)
295 struct shrinker* shrinker = NULL;
298 enc_pools_shrink_scan(shrinker, &scv);
300 return enc_pools_shrink_count(shrinker, &scv);
303 #endif /* HAVE_SHRINKER_COUNT */
306 int npages_to_npools(unsigned long npages)
308 return (int) ((npages + PAGES_PER_POOL - 1) / PAGES_PER_POOL);
312 * return how many pages cleaned up.
314 static unsigned long enc_pools_cleanup(struct page ***pools, int npools)
316 unsigned long cleaned = 0;
319 for (i = 0; i < npools; i++) {
321 for (j = 0; j < PAGES_PER_POOL; j++) {
323 __free_page(pools[i][j]);
327 OBD_FREE(pools[i], PAGE_SIZE);
336 * merge @npools pointed by @pools which contains @npages new pages
337 * into current pools.
339 * we have options to avoid most memory copy with some tricks. but we choose
340 * the simplest way to avoid complexity. It's not frequently called.
342 static void enc_pools_insert(struct page ***pools, int npools, int npages)
345 int op_idx, np_idx, og_idx, ng_idx;
346 int cur_npools, end_npools;
349 LASSERT(page_pools.epp_total_pages+npages <= page_pools.epp_max_pages);
350 LASSERT(npages_to_npools(npages) == npools);
351 LASSERT(page_pools.epp_growing);
353 spin_lock(&page_pools.epp_lock);
356 * (1) fill all the free slots of current pools.
358 /* free slots are those left by rent pages, and the extra ones with
359 * index >= total_pages, locate at the tail of last pool. */
360 freeslot = page_pools.epp_total_pages % PAGES_PER_POOL;
362 freeslot = PAGES_PER_POOL - freeslot;
363 freeslot += page_pools.epp_total_pages - page_pools.epp_free_pages;
365 op_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
366 og_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
368 ng_idx = (npages - 1) % PAGES_PER_POOL;
371 LASSERT(page_pools.epp_pools[op_idx][og_idx] == NULL);
372 LASSERT(pools[np_idx][ng_idx] != NULL);
374 page_pools.epp_pools[op_idx][og_idx] = pools[np_idx][ng_idx];
375 pools[np_idx][ng_idx] = NULL;
379 if (++og_idx == PAGES_PER_POOL) {
387 ng_idx = PAGES_PER_POOL - 1;
392 * (2) add pools if needed.
394 cur_npools = (page_pools.epp_total_pages + PAGES_PER_POOL - 1) /
396 end_npools = (page_pools.epp_total_pages + npages + PAGES_PER_POOL -1) /
398 LASSERT(end_npools <= page_pools.epp_max_pools);
401 while (cur_npools < end_npools) {
402 LASSERT(page_pools.epp_pools[cur_npools] == NULL);
403 LASSERT(np_idx < npools);
404 LASSERT(pools[np_idx] != NULL);
406 page_pools.epp_pools[cur_npools++] = pools[np_idx];
407 pools[np_idx++] = NULL;
410 page_pools.epp_total_pages += npages;
411 page_pools.epp_free_pages += npages;
412 page_pools.epp_st_lowfree = page_pools.epp_free_pages;
414 if (page_pools.epp_total_pages > page_pools.epp_st_max_pages)
415 page_pools.epp_st_max_pages = page_pools.epp_total_pages;
417 CDEBUG(D_SEC, "add %d pages to total %lu\n", npages,
418 page_pools.epp_total_pages);
420 spin_unlock(&page_pools.epp_lock);
423 static int enc_pools_add_pages(int npages)
425 static DEFINE_MUTEX(add_pages_mutex);
426 struct page ***pools;
427 int npools, alloced = 0;
428 int i, j, rc = -ENOMEM;
430 if (npages < PTLRPC_MAX_BRW_PAGES)
431 npages = PTLRPC_MAX_BRW_PAGES;
433 mutex_lock(&add_pages_mutex);
435 if (npages + page_pools.epp_total_pages > page_pools.epp_max_pages)
436 npages = page_pools.epp_max_pages - page_pools.epp_total_pages;
439 page_pools.epp_st_grows++;
441 npools = npages_to_npools(npages);
442 OBD_ALLOC(pools, npools * sizeof(*pools));
446 for (i = 0; i < npools; i++) {
447 OBD_ALLOC(pools[i], PAGE_SIZE);
448 if (pools[i] == NULL)
451 for (j = 0; j < PAGES_PER_POOL && alloced < npages; j++) {
452 pools[i][j] = alloc_page(GFP_NOFS |
454 if (pools[i][j] == NULL)
460 LASSERT(alloced == npages);
462 enc_pools_insert(pools, npools, npages);
463 CDEBUG(D_SEC, "added %d pages into pools\n", npages);
467 enc_pools_cleanup(pools, npools);
468 OBD_FREE(pools, npools * sizeof(*pools));
471 page_pools.epp_st_grow_fails++;
472 CERROR("Failed to allocate %d enc pages\n", npages);
475 mutex_unlock(&add_pages_mutex);
479 static inline void enc_pools_wakeup(void)
481 assert_spin_locked(&page_pools.epp_lock);
483 if (unlikely(page_pools.epp_waitqlen)) {
484 LASSERT(waitqueue_active(&page_pools.epp_waitq));
485 wake_up_all(&page_pools.epp_waitq);
489 static int enc_pools_should_grow(int page_needed, time64_t now)
491 /* don't grow if someone else is growing the pools right now,
492 * or the pools has reached its full capacity
494 if (page_pools.epp_growing ||
495 page_pools.epp_total_pages == page_pools.epp_max_pages)
498 /* if total pages is not enough, we need to grow */
499 if (page_pools.epp_total_pages < page_needed)
503 * we wanted to return 0 here if there was a shrink just
504 * happened a moment ago, but this may cause deadlock if both
505 * client and ost live on single node.
509 * here we perhaps need consider other factors like wait queue
510 * length, idle index, etc. ?
513 /* grow the pools in any other cases */
518 * Export the number of free pages in the pool
520 int get_free_pages_in_pool(void)
522 return page_pools.epp_free_pages;
524 EXPORT_SYMBOL(get_free_pages_in_pool);
527 * Let outside world know if enc_pool full capacity is reached
529 int pool_is_at_full_capacity(void)
531 return (page_pools.epp_total_pages == page_pools.epp_max_pages);
533 EXPORT_SYMBOL(pool_is_at_full_capacity);
536 * we allocate the requested pages atomically.
538 int sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc)
540 wait_queue_t waitlink;
541 unsigned long this_idle = -1;
547 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
548 LASSERT(desc->bd_iov_count > 0);
549 LASSERT(desc->bd_iov_count <= page_pools.epp_max_pages);
551 /* resent bulk, enc iov might have been allocated previously */
552 if (GET_ENC_KIOV(desc) != NULL)
555 OBD_ALLOC_LARGE(GET_ENC_KIOV(desc),
556 desc->bd_iov_count * sizeof(*GET_ENC_KIOV(desc)));
557 if (GET_ENC_KIOV(desc) == NULL)
560 spin_lock(&page_pools.epp_lock);
562 page_pools.epp_st_access++;
564 if (unlikely(page_pools.epp_free_pages < desc->bd_iov_count)) {
566 tick = cfs_time_current();
568 now = ktime_get_real_seconds();
570 page_pools.epp_st_missings++;
571 page_pools.epp_pages_short += desc->bd_iov_count;
573 if (enc_pools_should_grow(desc->bd_iov_count, now)) {
574 page_pools.epp_growing = 1;
576 spin_unlock(&page_pools.epp_lock);
577 enc_pools_add_pages(page_pools.epp_pages_short / 2);
578 spin_lock(&page_pools.epp_lock);
580 page_pools.epp_growing = 0;
584 if (page_pools.epp_growing) {
585 if (++page_pools.epp_waitqlen >
586 page_pools.epp_st_max_wqlen)
587 page_pools.epp_st_max_wqlen =
588 page_pools.epp_waitqlen;
590 set_current_state(TASK_UNINTERRUPTIBLE);
591 init_waitqueue_entry(&waitlink, current);
592 add_wait_queue(&page_pools.epp_waitq,
595 spin_unlock(&page_pools.epp_lock);
597 remove_wait_queue(&page_pools.epp_waitq,
599 LASSERT(page_pools.epp_waitqlen > 0);
600 spin_lock(&page_pools.epp_lock);
601 page_pools.epp_waitqlen--;
603 /* ptlrpcd thread should not sleep in that case,
604 * or deadlock may occur!
605 * Instead, return -ENOMEM so that upper layers
606 * will put request back in queue. */
607 page_pools.epp_st_outofmem++;
608 spin_unlock(&page_pools.epp_lock);
609 OBD_FREE_LARGE(GET_ENC_KIOV(desc),
611 sizeof(*GET_ENC_KIOV(desc)));
612 GET_ENC_KIOV(desc) = NULL;
617 LASSERT(page_pools.epp_pages_short >= desc->bd_iov_count);
618 page_pools.epp_pages_short -= desc->bd_iov_count;
624 /* record max wait time */
625 if (unlikely(tick != 0)) {
626 tick = cfs_time_current() - tick;
627 if (tick > page_pools.epp_st_max_wait)
628 page_pools.epp_st_max_wait = tick;
631 /* proceed with rest of allocation */
632 page_pools.epp_free_pages -= desc->bd_iov_count;
634 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
635 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
637 for (i = 0; i < desc->bd_iov_count; i++) {
638 LASSERT(page_pools.epp_pools[p_idx][g_idx] != NULL);
639 BD_GET_ENC_KIOV(desc, i).kiov_page =
640 page_pools.epp_pools[p_idx][g_idx];
641 page_pools.epp_pools[p_idx][g_idx] = NULL;
643 if (++g_idx == PAGES_PER_POOL) {
649 if (page_pools.epp_free_pages < page_pools.epp_st_lowfree)
650 page_pools.epp_st_lowfree = page_pools.epp_free_pages;
653 * new idle index = (old * weight + new) / (weight + 1)
655 if (this_idle == -1) {
656 this_idle = page_pools.epp_free_pages * IDLE_IDX_MAX /
657 page_pools.epp_total_pages;
659 page_pools.epp_idle_idx = (page_pools.epp_idle_idx * IDLE_IDX_WEIGHT +
661 (IDLE_IDX_WEIGHT + 1);
663 page_pools.epp_last_access = ktime_get_real_seconds();
665 spin_unlock(&page_pools.epp_lock);
668 EXPORT_SYMBOL(sptlrpc_enc_pool_get_pages);
670 void sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc *desc)
675 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
677 if (GET_ENC_KIOV(desc) == NULL)
680 LASSERT(desc->bd_iov_count > 0);
682 spin_lock(&page_pools.epp_lock);
684 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
685 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
687 LASSERT(page_pools.epp_free_pages + desc->bd_iov_count <=
688 page_pools.epp_total_pages);
689 LASSERT(page_pools.epp_pools[p_idx]);
691 for (i = 0; i < desc->bd_iov_count; i++) {
692 LASSERT(BD_GET_ENC_KIOV(desc, i).kiov_page != NULL);
693 LASSERT(g_idx != 0 || page_pools.epp_pools[p_idx]);
694 LASSERT(page_pools.epp_pools[p_idx][g_idx] == NULL);
696 page_pools.epp_pools[p_idx][g_idx] =
697 BD_GET_ENC_KIOV(desc, i).kiov_page;
699 if (++g_idx == PAGES_PER_POOL) {
705 page_pools.epp_free_pages += desc->bd_iov_count;
709 spin_unlock(&page_pools.epp_lock);
711 OBD_FREE_LARGE(GET_ENC_KIOV(desc),
712 desc->bd_iov_count * sizeof(*GET_ENC_KIOV(desc)));
713 GET_ENC_KIOV(desc) = NULL;
717 * we don't do much stuff for add_user/del_user anymore, except adding some
718 * initial pages in add_user() if current pools are empty, rest would be
719 * handled by the pools's self-adaption.
721 int sptlrpc_enc_pool_add_user(void)
725 spin_lock(&page_pools.epp_lock);
726 if (page_pools.epp_growing == 0 && page_pools.epp_total_pages == 0) {
727 page_pools.epp_growing = 1;
730 spin_unlock(&page_pools.epp_lock);
733 enc_pools_add_pages(PTLRPC_MAX_BRW_PAGES +
734 PTLRPC_MAX_BRW_PAGES);
736 spin_lock(&page_pools.epp_lock);
737 page_pools.epp_growing = 0;
739 spin_unlock(&page_pools.epp_lock);
743 EXPORT_SYMBOL(sptlrpc_enc_pool_add_user);
745 int sptlrpc_enc_pool_del_user(void)
749 EXPORT_SYMBOL(sptlrpc_enc_pool_del_user);
751 static inline void enc_pools_alloc(void)
753 LASSERT(page_pools.epp_max_pools);
754 OBD_ALLOC_LARGE(page_pools.epp_pools,
755 page_pools.epp_max_pools *
756 sizeof(*page_pools.epp_pools));
759 static inline void enc_pools_free(void)
761 LASSERT(page_pools.epp_max_pools);
762 LASSERT(page_pools.epp_pools);
764 OBD_FREE_LARGE(page_pools.epp_pools,
765 page_pools.epp_max_pools *
766 sizeof(*page_pools.epp_pools));
769 int sptlrpc_enc_pool_init(void)
771 DEF_SHRINKER_VAR(shvar, enc_pools_shrink,
772 enc_pools_shrink_count, enc_pools_shrink_scan);
774 page_pools.epp_max_pages = totalram_pages / 8;
775 if (enc_pool_max_memory_mb > 0 &&
776 enc_pool_max_memory_mb <= (totalram_pages >> mult))
777 page_pools.epp_max_pages = enc_pool_max_memory_mb << mult;
779 page_pools.epp_max_pools = npages_to_npools(page_pools.epp_max_pages);
781 init_waitqueue_head(&page_pools.epp_waitq);
782 page_pools.epp_waitqlen = 0;
783 page_pools.epp_pages_short = 0;
785 page_pools.epp_growing = 0;
787 page_pools.epp_idle_idx = 0;
788 page_pools.epp_last_shrink = ktime_get_real_seconds();
789 page_pools.epp_last_access = ktime_get_real_seconds();
791 spin_lock_init(&page_pools.epp_lock);
792 page_pools.epp_total_pages = 0;
793 page_pools.epp_free_pages = 0;
795 page_pools.epp_st_max_pages = 0;
796 page_pools.epp_st_grows = 0;
797 page_pools.epp_st_grow_fails = 0;
798 page_pools.epp_st_shrinks = 0;
799 page_pools.epp_st_access = 0;
800 page_pools.epp_st_missings = 0;
801 page_pools.epp_st_lowfree = 0;
802 page_pools.epp_st_max_wqlen = 0;
803 page_pools.epp_st_max_wait = 0;
804 page_pools.epp_st_outofmem = 0;
807 if (page_pools.epp_pools == NULL)
810 pools_shrinker = set_shrinker(pools_shrinker_seeks, &shvar);
811 if (pools_shrinker == NULL) {
819 void sptlrpc_enc_pool_fini(void)
821 unsigned long cleaned, npools;
823 LASSERT(pools_shrinker);
824 LASSERT(page_pools.epp_pools);
825 LASSERT(page_pools.epp_total_pages == page_pools.epp_free_pages);
827 remove_shrinker(pools_shrinker);
829 npools = npages_to_npools(page_pools.epp_total_pages);
830 cleaned = enc_pools_cleanup(page_pools.epp_pools, npools);
831 LASSERT(cleaned == page_pools.epp_total_pages);
835 if (page_pools.epp_st_access > 0) {
837 "max pages %lu, grows %u, grow fails %u, shrinks %u, "
838 "access %lu, missing %lu, max qlen %u, max wait "
839 CFS_TIME_T"/%lu, out of mem %lu\n",
840 page_pools.epp_st_max_pages, page_pools.epp_st_grows,
841 page_pools.epp_st_grow_fails,
842 page_pools.epp_st_shrinks, page_pools.epp_st_access,
843 page_pools.epp_st_missings, page_pools.epp_st_max_wqlen,
844 page_pools.epp_st_max_wait,
845 msecs_to_jiffies(MSEC_PER_SEC),
846 page_pools.epp_st_outofmem);
851 static int cfs_hash_alg_id[] = {
852 [BULK_HASH_ALG_NULL] = CFS_HASH_ALG_NULL,
853 [BULK_HASH_ALG_ADLER32] = CFS_HASH_ALG_ADLER32,
854 [BULK_HASH_ALG_CRC32] = CFS_HASH_ALG_CRC32,
855 [BULK_HASH_ALG_MD5] = CFS_HASH_ALG_MD5,
856 [BULK_HASH_ALG_SHA1] = CFS_HASH_ALG_SHA1,
857 [BULK_HASH_ALG_SHA256] = CFS_HASH_ALG_SHA256,
858 [BULK_HASH_ALG_SHA384] = CFS_HASH_ALG_SHA384,
859 [BULK_HASH_ALG_SHA512] = CFS_HASH_ALG_SHA512,
861 const char * sptlrpc_get_hash_name(__u8 hash_alg)
863 return cfs_crypto_hash_name(cfs_hash_alg_id[hash_alg]);
866 __u8 sptlrpc_get_hash_alg(const char *algname)
868 return cfs_crypto_hash_alg(algname);
871 int bulk_sec_desc_unpack(struct lustre_msg *msg, int offset, int swabbed)
873 struct ptlrpc_bulk_sec_desc *bsd;
874 int size = msg->lm_buflens[offset];
876 bsd = lustre_msg_buf(msg, offset, sizeof(*bsd));
878 CERROR("Invalid bulk sec desc: size %d\n", size);
883 __swab32s(&bsd->bsd_nob);
886 if (unlikely(bsd->bsd_version != 0)) {
887 CERROR("Unexpected version %u\n", bsd->bsd_version);
891 if (unlikely(bsd->bsd_type >= SPTLRPC_BULK_MAX)) {
892 CERROR("Invalid type %u\n", bsd->bsd_type);
896 /* FIXME more sanity check here */
898 if (unlikely(bsd->bsd_svc != SPTLRPC_BULK_SVC_NULL &&
899 bsd->bsd_svc != SPTLRPC_BULK_SVC_INTG &&
900 bsd->bsd_svc != SPTLRPC_BULK_SVC_PRIV)) {
901 CERROR("Invalid svc %u\n", bsd->bsd_svc);
907 EXPORT_SYMBOL(bulk_sec_desc_unpack);
910 * Compute the checksum of an RPC buffer payload. If the return \a buflen
911 * is not large enough, truncate the result to fit so that it is possible
912 * to use a hash function with a large hash space, but only use a part of
913 * the resulting hash.
915 int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg,
916 void *buf, int buflen)
918 struct cfs_crypto_hash_desc *hdesc;
920 unsigned int bufsize;
923 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
924 LASSERT(alg > BULK_HASH_ALG_NULL && alg < BULK_HASH_ALG_MAX);
925 LASSERT(buflen >= 4);
927 hdesc = cfs_crypto_hash_init(cfs_hash_alg_id[alg], NULL, 0);
929 CERROR("Unable to initialize checksum hash %s\n",
930 cfs_crypto_hash_name(cfs_hash_alg_id[alg]));
931 return PTR_ERR(hdesc);
934 hashsize = cfs_crypto_hash_digestsize(cfs_hash_alg_id[alg]);
936 for (i = 0; i < desc->bd_iov_count; i++) {
937 cfs_crypto_hash_update_page(hdesc,
938 BD_GET_KIOV(desc, i).kiov_page,
939 BD_GET_KIOV(desc, i).kiov_offset &
941 BD_GET_KIOV(desc, i).kiov_len);
944 if (hashsize > buflen) {
945 unsigned char hashbuf[CFS_CRYPTO_HASH_DIGESTSIZE_MAX];
947 bufsize = sizeof(hashbuf);
948 LASSERTF(bufsize >= hashsize, "bufsize = %u < hashsize %u\n",
950 err = cfs_crypto_hash_final(hdesc, hashbuf, &bufsize);
951 memcpy(buf, hashbuf, buflen);
954 err = cfs_crypto_hash_final(hdesc, buf, &bufsize);