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) 2007, 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/ptlrpc/sec_bulk.c
38 * Author: Eric Mei <ericm@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_SEC
46 #include <libcfs/libcfs.h>
48 #include <liblustre.h>
49 #include <libcfs/list.h>
51 #include <linux/crypto.h>
55 #include <obd_cksum.h>
56 #include <obd_class.h>
57 #include <obd_support.h>
58 #include <lustre_net.h>
59 #include <lustre_import.h>
60 #include <lustre_dlm.h>
61 #include <lustre_sec.h>
63 #include "ptlrpc_internal.h"
65 /****************************************
66 * bulk encryption page pools *
67 ****************************************/
71 #define PTRS_PER_PAGE (CFS_PAGE_SIZE / sizeof(void *))
72 #define PAGES_PER_POOL (PTRS_PER_PAGE)
74 #define IDLE_IDX_MAX (100)
75 #define IDLE_IDX_WEIGHT (3)
77 #define CACHE_QUIESCENT_PERIOD (20)
79 static struct ptlrpc_enc_page_pool {
83 unsigned long epp_max_pages; /* maximum pages can hold, const */
84 unsigned int epp_max_pools; /* number of pools, const */
87 * wait queue in case of not enough free pages.
89 cfs_waitq_t epp_waitq; /* waiting threads */
90 unsigned int epp_waitqlen; /* wait queue length */
91 unsigned long epp_pages_short; /* # of pages wanted of in-q users */
92 unsigned int epp_growing:1; /* during adding pages */
95 * indicating how idle the pools are, from 0 to MAX_IDLE_IDX
96 * this is counted based on each time when getting pages from
97 * the pools, not based on time. which means in case that system
98 * is idled for a while but the idle_idx might still be low if no
99 * activities happened in the pools.
101 unsigned long epp_idle_idx;
103 /* last shrink time due to mem tight */
104 long epp_last_shrink;
105 long epp_last_access;
108 * in-pool pages bookkeeping
110 cfs_spinlock_t epp_lock; /* protect following fields */
111 unsigned long epp_total_pages; /* total pages in pools */
112 unsigned long epp_free_pages; /* current pages available */
117 unsigned long epp_st_max_pages; /* # of pages ever reached */
118 unsigned int epp_st_grows; /* # of grows */
119 unsigned int epp_st_grow_fails; /* # of add pages failures */
120 unsigned int epp_st_shrinks; /* # of shrinks */
121 unsigned long epp_st_access; /* # of access */
122 unsigned long epp_st_missings; /* # of cache missing */
123 unsigned long epp_st_lowfree; /* lowest free pages reached */
124 unsigned int epp_st_max_wqlen; /* highest waitqueue length */
125 cfs_time_t epp_st_max_wait; /* in jeffies */
129 cfs_page_t ***epp_pools;
135 const int pools_shrinker_seeks = CFS_DEFAULT_SEEKS;
136 static struct cfs_shrinker *pools_shrinker = NULL;
140 * /proc/fs/lustre/sptlrpc/encrypt_page_pools
142 int sptlrpc_proc_read_enc_pool(char *page, char **start, off_t off, int count,
143 int *eof, void *data)
147 cfs_spin_lock(&page_pools.epp_lock);
149 rc = snprintf(page, count,
150 "physical pages: %lu\n"
151 "pages per pool: %lu\n"
156 "idle index: %lu/100\n"
157 "last shrink: %lds\n"
158 "last access: %lds\n"
159 "max pages reached: %lu\n"
161 "grows failure: %u\n"
163 "cache access: %lu\n"
164 "cache missing: %lu\n"
165 "low free mark: %lu\n"
166 "max waitqueue depth: %u\n"
167 "max wait time: "CFS_TIME_T"/%u\n"
171 page_pools.epp_max_pages,
172 page_pools.epp_max_pools,
173 page_pools.epp_total_pages,
174 page_pools.epp_free_pages,
175 page_pools.epp_idle_idx,
176 cfs_time_current_sec() - page_pools.epp_last_shrink,
177 cfs_time_current_sec() - page_pools.epp_last_access,
178 page_pools.epp_st_max_pages,
179 page_pools.epp_st_grows,
180 page_pools.epp_st_grow_fails,
181 page_pools.epp_st_shrinks,
182 page_pools.epp_st_access,
183 page_pools.epp_st_missings,
184 page_pools.epp_st_lowfree,
185 page_pools.epp_st_max_wqlen,
186 page_pools.epp_st_max_wait, CFS_HZ
189 cfs_spin_unlock(&page_pools.epp_lock);
193 static void enc_pools_release_free_pages(long npages)
196 int p_idx_max1, p_idx_max2;
199 LASSERT(npages <= page_pools.epp_free_pages);
200 LASSERT(page_pools.epp_free_pages <= page_pools.epp_total_pages);
202 /* max pool index before the release */
203 p_idx_max2 = (page_pools.epp_total_pages - 1) / PAGES_PER_POOL;
205 page_pools.epp_free_pages -= npages;
206 page_pools.epp_total_pages -= npages;
208 /* max pool index after the release */
209 p_idx_max1 = page_pools.epp_total_pages == 0 ? -1 :
210 ((page_pools.epp_total_pages - 1) / PAGES_PER_POOL);
212 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
213 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
214 LASSERT(page_pools.epp_pools[p_idx]);
217 LASSERT(page_pools.epp_pools[p_idx]);
218 LASSERT(page_pools.epp_pools[p_idx][g_idx] != NULL);
220 cfs_free_page(page_pools.epp_pools[p_idx][g_idx]);
221 page_pools.epp_pools[p_idx][g_idx] = NULL;
223 if (++g_idx == PAGES_PER_POOL) {
229 /* free unused pools */
230 while (p_idx_max1 < p_idx_max2) {
231 LASSERT(page_pools.epp_pools[p_idx_max2]);
232 OBD_FREE(page_pools.epp_pools[p_idx_max2], CFS_PAGE_SIZE);
233 page_pools.epp_pools[p_idx_max2] = NULL;
239 * could be called frequently for query (@nr_to_scan == 0).
240 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
242 static int enc_pools_shrink(SHRINKER_FIRST_ARG int nr_to_scan,
243 unsigned int gfp_mask)
245 if (unlikely(nr_to_scan != 0)) {
246 cfs_spin_lock(&page_pools.epp_lock);
247 nr_to_scan = min(nr_to_scan, (int) page_pools.epp_free_pages -
248 PTLRPC_MAX_BRW_PAGES);
249 if (nr_to_scan > 0) {
250 enc_pools_release_free_pages(nr_to_scan);
251 CDEBUG(D_SEC, "released %d pages, %ld left\n",
252 nr_to_scan, page_pools.epp_free_pages);
254 page_pools.epp_st_shrinks++;
255 page_pools.epp_last_shrink = cfs_time_current_sec();
257 cfs_spin_unlock(&page_pools.epp_lock);
261 * if no pool access for a long time, we consider it's fully idle.
262 * a little race here is fine.
264 if (unlikely(cfs_time_current_sec() - page_pools.epp_last_access >
265 CACHE_QUIESCENT_PERIOD)) {
266 cfs_spin_lock(&page_pools.epp_lock);
267 page_pools.epp_idle_idx = IDLE_IDX_MAX;
268 cfs_spin_unlock(&page_pools.epp_lock);
271 LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX);
272 return max((int) page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES, 0) *
273 (IDLE_IDX_MAX - page_pools.epp_idle_idx) / IDLE_IDX_MAX;
277 int npages_to_npools(unsigned long npages)
279 return (int) ((npages + PAGES_PER_POOL - 1) / PAGES_PER_POOL);
283 * return how many pages cleaned up.
285 static unsigned long enc_pools_cleanup(cfs_page_t ***pools, int npools)
287 unsigned long cleaned = 0;
290 for (i = 0; i < npools; i++) {
292 for (j = 0; j < PAGES_PER_POOL; j++) {
294 cfs_free_page(pools[i][j]);
298 OBD_FREE(pools[i], CFS_PAGE_SIZE);
307 * merge @npools pointed by @pools which contains @npages new pages
308 * into current pools.
310 * we have options to avoid most memory copy with some tricks. but we choose
311 * the simplest way to avoid complexity. It's not frequently called.
313 static void enc_pools_insert(cfs_page_t ***pools, int npools, int npages)
316 int op_idx, np_idx, og_idx, ng_idx;
317 int cur_npools, end_npools;
320 LASSERT(page_pools.epp_total_pages+npages <= page_pools.epp_max_pages);
321 LASSERT(npages_to_npools(npages) == npools);
322 LASSERT(page_pools.epp_growing);
324 cfs_spin_lock(&page_pools.epp_lock);
327 * (1) fill all the free slots of current pools.
329 /* free slots are those left by rent pages, and the extra ones with
330 * index >= total_pages, locate at the tail of last pool. */
331 freeslot = page_pools.epp_total_pages % PAGES_PER_POOL;
333 freeslot = PAGES_PER_POOL - freeslot;
334 freeslot += page_pools.epp_total_pages - page_pools.epp_free_pages;
336 op_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
337 og_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
339 ng_idx = (npages - 1) % PAGES_PER_POOL;
342 LASSERT(page_pools.epp_pools[op_idx][og_idx] == NULL);
343 LASSERT(pools[np_idx][ng_idx] != NULL);
345 page_pools.epp_pools[op_idx][og_idx] = pools[np_idx][ng_idx];
346 pools[np_idx][ng_idx] = NULL;
350 if (++og_idx == PAGES_PER_POOL) {
358 ng_idx = PAGES_PER_POOL - 1;
363 * (2) add pools if needed.
365 cur_npools = (page_pools.epp_total_pages + PAGES_PER_POOL - 1) /
367 end_npools = (page_pools.epp_total_pages + npages + PAGES_PER_POOL -1) /
369 LASSERT(end_npools <= page_pools.epp_max_pools);
372 while (cur_npools < end_npools) {
373 LASSERT(page_pools.epp_pools[cur_npools] == NULL);
374 LASSERT(np_idx < npools);
375 LASSERT(pools[np_idx] != NULL);
377 page_pools.epp_pools[cur_npools++] = pools[np_idx];
378 pools[np_idx++] = NULL;
381 page_pools.epp_total_pages += npages;
382 page_pools.epp_free_pages += npages;
383 page_pools.epp_st_lowfree = page_pools.epp_free_pages;
385 if (page_pools.epp_total_pages > page_pools.epp_st_max_pages)
386 page_pools.epp_st_max_pages = page_pools.epp_total_pages;
388 CDEBUG(D_SEC, "add %d pages to total %lu\n", npages,
389 page_pools.epp_total_pages);
391 cfs_spin_unlock(&page_pools.epp_lock);
394 static int enc_pools_add_pages(int npages)
396 static CFS_DECLARE_MUTEX(sem_add_pages);
398 int npools, alloced = 0;
399 int i, j, rc = -ENOMEM;
401 if (npages < PTLRPC_MAX_BRW_PAGES)
402 npages = PTLRPC_MAX_BRW_PAGES;
404 cfs_down(&sem_add_pages);
406 if (npages + page_pools.epp_total_pages > page_pools.epp_max_pages)
407 npages = page_pools.epp_max_pages - page_pools.epp_total_pages;
410 page_pools.epp_st_grows++;
412 npools = npages_to_npools(npages);
413 OBD_ALLOC(pools, npools * sizeof(*pools));
417 for (i = 0; i < npools; i++) {
418 OBD_ALLOC(pools[i], CFS_PAGE_SIZE);
419 if (pools[i] == NULL)
422 for (j = 0; j < PAGES_PER_POOL && alloced < npages; j++) {
423 pools[i][j] = cfs_alloc_page(CFS_ALLOC_IO |
425 if (pools[i][j] == NULL)
431 LASSERT(alloced == npages);
433 enc_pools_insert(pools, npools, npages);
434 CDEBUG(D_SEC, "added %d pages into pools\n", npages);
438 enc_pools_cleanup(pools, npools);
439 OBD_FREE(pools, npools * sizeof(*pools));
442 page_pools.epp_st_grow_fails++;
443 CERROR("Failed to allocate %d enc pages\n", npages);
446 cfs_up(&sem_add_pages);
450 static inline void enc_pools_wakeup(void)
452 LASSERT_SPIN_LOCKED(&page_pools.epp_lock);
453 LASSERT(page_pools.epp_waitqlen >= 0);
455 if (unlikely(page_pools.epp_waitqlen)) {
456 LASSERT(cfs_waitq_active(&page_pools.epp_waitq));
457 cfs_waitq_broadcast(&page_pools.epp_waitq);
461 static int enc_pools_should_grow(int page_needed, long now)
463 /* don't grow if someone else is growing the pools right now,
464 * or the pools has reached its full capacity
466 if (page_pools.epp_growing ||
467 page_pools.epp_total_pages == page_pools.epp_max_pages)
470 /* if total pages is not enough, we need to grow */
471 if (page_pools.epp_total_pages < page_needed)
475 * we wanted to return 0 here if there was a shrink just happened
476 * moment ago, but this may cause deadlock if both client and ost
477 * live on single node.
480 if (now - page_pools.epp_last_shrink < 2)
485 * here we perhaps need consider other factors like wait queue
486 * length, idle index, etc. ?
489 /* grow the pools in any other cases */
494 * we allocate the requested pages atomically.
496 int sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc)
498 cfs_waitlink_t waitlink;
499 unsigned long this_idle = -1;
505 LASSERT(desc->bd_iov_count > 0);
506 LASSERT(desc->bd_iov_count <= page_pools.epp_max_pages);
508 /* resent bulk, enc iov might have been allocated previously */
509 if (desc->bd_enc_iov != NULL)
512 OBD_ALLOC(desc->bd_enc_iov,
513 desc->bd_iov_count * sizeof(*desc->bd_enc_iov));
514 if (desc->bd_enc_iov == NULL)
517 cfs_spin_lock(&page_pools.epp_lock);
519 page_pools.epp_st_access++;
521 if (unlikely(page_pools.epp_free_pages < desc->bd_iov_count)) {
523 tick = cfs_time_current();
525 now = cfs_time_current_sec();
527 page_pools.epp_st_missings++;
528 page_pools.epp_pages_short += desc->bd_iov_count;
530 if (enc_pools_should_grow(desc->bd_iov_count, now)) {
531 page_pools.epp_growing = 1;
533 cfs_spin_unlock(&page_pools.epp_lock);
534 enc_pools_add_pages(page_pools.epp_pages_short / 2);
535 cfs_spin_lock(&page_pools.epp_lock);
537 page_pools.epp_growing = 0;
541 if (++page_pools.epp_waitqlen >
542 page_pools.epp_st_max_wqlen)
543 page_pools.epp_st_max_wqlen =
544 page_pools.epp_waitqlen;
546 cfs_set_current_state(CFS_TASK_UNINT);
547 cfs_waitlink_init(&waitlink);
548 cfs_waitq_add(&page_pools.epp_waitq, &waitlink);
550 cfs_spin_unlock(&page_pools.epp_lock);
551 cfs_waitq_wait(&waitlink, CFS_TASK_UNINT);
552 cfs_waitq_del(&page_pools.epp_waitq, &waitlink);
553 LASSERT(page_pools.epp_waitqlen > 0);
554 cfs_spin_lock(&page_pools.epp_lock);
555 page_pools.epp_waitqlen--;
558 LASSERT(page_pools.epp_pages_short >= desc->bd_iov_count);
559 page_pools.epp_pages_short -= desc->bd_iov_count;
565 /* record max wait time */
566 if (unlikely(tick != 0)) {
567 tick = cfs_time_current() - tick;
568 if (tick > page_pools.epp_st_max_wait)
569 page_pools.epp_st_max_wait = tick;
572 /* proceed with rest of allocation */
573 page_pools.epp_free_pages -= desc->bd_iov_count;
575 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
576 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
578 for (i = 0; i < desc->bd_iov_count; i++) {
579 LASSERT(page_pools.epp_pools[p_idx][g_idx] != NULL);
580 desc->bd_enc_iov[i].kiov_page =
581 page_pools.epp_pools[p_idx][g_idx];
582 page_pools.epp_pools[p_idx][g_idx] = NULL;
584 if (++g_idx == PAGES_PER_POOL) {
590 if (page_pools.epp_free_pages < page_pools.epp_st_lowfree)
591 page_pools.epp_st_lowfree = page_pools.epp_free_pages;
594 * new idle index = (old * weight + new) / (weight + 1)
596 if (this_idle == -1) {
597 this_idle = page_pools.epp_free_pages * IDLE_IDX_MAX /
598 page_pools.epp_total_pages;
600 page_pools.epp_idle_idx = (page_pools.epp_idle_idx * IDLE_IDX_WEIGHT +
602 (IDLE_IDX_WEIGHT + 1);
604 page_pools.epp_last_access = cfs_time_current_sec();
606 cfs_spin_unlock(&page_pools.epp_lock);
609 EXPORT_SYMBOL(sptlrpc_enc_pool_get_pages);
611 void sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc *desc)
616 if (desc->bd_enc_iov == NULL)
619 LASSERT(desc->bd_iov_count > 0);
621 cfs_spin_lock(&page_pools.epp_lock);
623 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
624 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
626 LASSERT(page_pools.epp_free_pages + desc->bd_iov_count <=
627 page_pools.epp_total_pages);
628 LASSERT(page_pools.epp_pools[p_idx]);
630 for (i = 0; i < desc->bd_iov_count; i++) {
631 LASSERT(desc->bd_enc_iov[i].kiov_page != NULL);
632 LASSERT(g_idx != 0 || page_pools.epp_pools[p_idx]);
633 LASSERT(page_pools.epp_pools[p_idx][g_idx] == NULL);
635 page_pools.epp_pools[p_idx][g_idx] =
636 desc->bd_enc_iov[i].kiov_page;
638 if (++g_idx == PAGES_PER_POOL) {
644 page_pools.epp_free_pages += desc->bd_iov_count;
648 cfs_spin_unlock(&page_pools.epp_lock);
650 OBD_FREE(desc->bd_enc_iov,
651 desc->bd_iov_count * sizeof(*desc->bd_enc_iov));
652 desc->bd_enc_iov = NULL;
654 EXPORT_SYMBOL(sptlrpc_enc_pool_put_pages);
657 * we don't do much stuff for add_user/del_user anymore, except adding some
658 * initial pages in add_user() if current pools are empty, rest would be
659 * handled by the pools's self-adaption.
661 int sptlrpc_enc_pool_add_user(void)
665 cfs_spin_lock(&page_pools.epp_lock);
666 if (page_pools.epp_growing == 0 && page_pools.epp_total_pages == 0) {
667 page_pools.epp_growing = 1;
670 cfs_spin_unlock(&page_pools.epp_lock);
673 enc_pools_add_pages(PTLRPC_MAX_BRW_PAGES +
674 PTLRPC_MAX_BRW_PAGES);
676 cfs_spin_lock(&page_pools.epp_lock);
677 page_pools.epp_growing = 0;
679 cfs_spin_unlock(&page_pools.epp_lock);
683 EXPORT_SYMBOL(sptlrpc_enc_pool_add_user);
685 int sptlrpc_enc_pool_del_user(void)
689 EXPORT_SYMBOL(sptlrpc_enc_pool_del_user);
691 static inline void enc_pools_alloc(void)
693 LASSERT(page_pools.epp_max_pools);
694 OBD_ALLOC_LARGE(page_pools.epp_pools,
695 page_pools.epp_max_pools *
696 sizeof(*page_pools.epp_pools));
699 static inline void enc_pools_free(void)
701 LASSERT(page_pools.epp_max_pools);
702 LASSERT(page_pools.epp_pools);
704 OBD_FREE_LARGE(page_pools.epp_pools,
705 page_pools.epp_max_pools *
706 sizeof(*page_pools.epp_pools));
709 int sptlrpc_enc_pool_init(void)
712 * maximum capacity is 1/8 of total physical memory.
713 * is the 1/8 a good number?
715 page_pools.epp_max_pages = cfs_num_physpages / 8;
716 page_pools.epp_max_pools = npages_to_npools(page_pools.epp_max_pages);
718 cfs_waitq_init(&page_pools.epp_waitq);
719 page_pools.epp_waitqlen = 0;
720 page_pools.epp_pages_short = 0;
722 page_pools.epp_growing = 0;
724 page_pools.epp_idle_idx = 0;
725 page_pools.epp_last_shrink = cfs_time_current_sec();
726 page_pools.epp_last_access = cfs_time_current_sec();
728 cfs_spin_lock_init(&page_pools.epp_lock);
729 page_pools.epp_total_pages = 0;
730 page_pools.epp_free_pages = 0;
732 page_pools.epp_st_max_pages = 0;
733 page_pools.epp_st_grows = 0;
734 page_pools.epp_st_grow_fails = 0;
735 page_pools.epp_st_shrinks = 0;
736 page_pools.epp_st_access = 0;
737 page_pools.epp_st_missings = 0;
738 page_pools.epp_st_lowfree = 0;
739 page_pools.epp_st_max_wqlen = 0;
740 page_pools.epp_st_max_wait = 0;
743 if (page_pools.epp_pools == NULL)
746 pools_shrinker = cfs_set_shrinker(pools_shrinker_seeks,
748 if (pools_shrinker == NULL) {
756 void sptlrpc_enc_pool_fini(void)
758 unsigned long cleaned, npools;
760 LASSERT(pools_shrinker);
761 LASSERT(page_pools.epp_pools);
762 LASSERT(page_pools.epp_total_pages == page_pools.epp_free_pages);
764 cfs_remove_shrinker(pools_shrinker);
766 npools = npages_to_npools(page_pools.epp_total_pages);
767 cleaned = enc_pools_cleanup(page_pools.epp_pools, npools);
768 LASSERT(cleaned == page_pools.epp_total_pages);
772 if (page_pools.epp_st_access > 0) {
773 CWARN("max pages %lu, grows %u, grow fails %u, shrinks %u, "
774 "access %lu, missing %lu, max qlen %u, max wait "
776 page_pools.epp_st_max_pages, page_pools.epp_st_grows,
777 page_pools.epp_st_grow_fails,
778 page_pools.epp_st_shrinks, page_pools.epp_st_access,
779 page_pools.epp_st_missings, page_pools.epp_st_max_wqlen,
780 page_pools.epp_st_max_wait, CFS_HZ);
784 #else /* !__KERNEL__ */
786 int sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc)
791 void sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc *desc)
795 int sptlrpc_enc_pool_init(void)
800 void sptlrpc_enc_pool_fini(void)
805 /****************************************
806 * Helpers to assist policy modules to *
807 * implement checksum funcationality *
808 ****************************************/
810 static struct sptlrpc_hash_type hash_types[] = {
811 [BULK_HASH_ALG_NULL] = { "null", "null", 0 },
812 [BULK_HASH_ALG_ADLER32] = { "adler32", "adler32", 4 },
813 [BULK_HASH_ALG_CRC32] = { "crc32", "crc32", 4 },
814 [BULK_HASH_ALG_MD5] = { "md5", "md5", 16 },
815 [BULK_HASH_ALG_SHA1] = { "sha1", "sha1", 20 },
816 [BULK_HASH_ALG_SHA256] = { "sha256", "sha256", 32 },
817 [BULK_HASH_ALG_SHA384] = { "sha384", "sha384", 48 },
818 [BULK_HASH_ALG_SHA512] = { "sha512", "sha512", 64 },
821 const struct sptlrpc_hash_type *sptlrpc_get_hash_type(__u8 hash_alg)
823 struct sptlrpc_hash_type *ht;
825 if (hash_alg < BULK_HASH_ALG_MAX) {
826 ht = &hash_types[hash_alg];
827 if (ht->sht_tfm_name)
832 EXPORT_SYMBOL(sptlrpc_get_hash_type);
834 const char * sptlrpc_get_hash_name(__u8 hash_alg)
836 const struct sptlrpc_hash_type *ht;
838 ht = sptlrpc_get_hash_type(hash_alg);
844 EXPORT_SYMBOL(sptlrpc_get_hash_name);
846 __u8 sptlrpc_get_hash_alg(const char *algname)
850 for (i = 0; i < BULK_HASH_ALG_MAX; i++)
851 if (!strcmp(hash_types[i].sht_name, algname))
855 EXPORT_SYMBOL(sptlrpc_get_hash_alg);
857 int bulk_sec_desc_unpack(struct lustre_msg *msg, int offset, int swabbed)
859 struct ptlrpc_bulk_sec_desc *bsd;
860 int size = msg->lm_buflens[offset];
862 bsd = lustre_msg_buf(msg, offset, sizeof(*bsd));
864 CERROR("Invalid bulk sec desc: size %d\n", size);
869 __swab32s(&bsd->bsd_nob);
872 if (unlikely(bsd->bsd_version != 0)) {
873 CERROR("Unexpected version %u\n", bsd->bsd_version);
877 if (unlikely(bsd->bsd_type >= SPTLRPC_BULK_MAX)) {
878 CERROR("Invalid type %u\n", bsd->bsd_type);
882 /* FIXME more sanity check here */
884 if (unlikely(bsd->bsd_svc != SPTLRPC_BULK_SVC_NULL &&
885 bsd->bsd_svc != SPTLRPC_BULK_SVC_INTG &&
886 bsd->bsd_svc != SPTLRPC_BULK_SVC_PRIV)) {
887 CERROR("Invalid svc %u\n", bsd->bsd_svc);
893 EXPORT_SYMBOL(bulk_sec_desc_unpack);
898 static int do_bulk_checksum_adler32(struct ptlrpc_bulk_desc *desc, void *buf)
906 for (i = 0; i < desc->bd_iov_count; i++) {
907 page = desc->bd_iov[i].kiov_page;
908 off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
909 ptr = cfs_kmap(page) + off;
910 len = desc->bd_iov[i].kiov_len;
912 adler32 = adler32(adler32, ptr, len);
917 adler32 = cpu_to_le32(adler32);
918 memcpy(buf, &adler32, sizeof(adler32));
923 static int do_bulk_checksum_crc32(struct ptlrpc_bulk_desc *desc, void *buf)
931 for (i = 0; i < desc->bd_iov_count; i++) {
932 page = desc->bd_iov[i].kiov_page;
933 off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
934 ptr = cfs_kmap(page) + off;
935 len = desc->bd_iov[i].kiov_len;
937 crc32 = crc32_le(crc32, ptr, len);
942 crc32 = cpu_to_le32(crc32);
943 memcpy(buf, &crc32, sizeof(crc32));
947 int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg,
948 void *buf, int buflen)
950 struct hash_desc hdesc;
953 struct scatterlist sl;
956 LASSERT(alg > BULK_HASH_ALG_NULL && alg < BULK_HASH_ALG_MAX);
957 LASSERT(buflen >= 4);
960 case BULK_HASH_ALG_ADLER32:
962 return do_bulk_checksum_adler32(desc, buf);
964 CERROR("Adler32 not supported\n");
967 case BULK_HASH_ALG_CRC32:
968 return do_bulk_checksum_crc32(desc, buf);
971 hdesc.tfm = ll_crypto_alloc_hash(hash_types[alg].sht_tfm_name, 0, 0);
972 if (hdesc.tfm == NULL) {
973 CERROR("Unable to allocate TFM %s\n", hash_types[alg].sht_name);
978 ll_crypto_hash_init(&hdesc);
980 hashsize = ll_crypto_hash_digestsize(hdesc.tfm);
982 for (i = 0; i < desc->bd_iov_count; i++) {
983 sg_set_page(&sl, desc->bd_iov[i].kiov_page,
984 desc->bd_iov[i].kiov_len,
985 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK);
986 ll_crypto_hash_update(&hdesc, &sl, sl.length);
989 if (hashsize > buflen) {
990 ll_crypto_hash_final(&hdesc, hashbuf);
991 memcpy(buf, hashbuf, buflen);
993 ll_crypto_hash_final(&hdesc, buf);
996 ll_crypto_free_hash(hdesc.tfm);
999 EXPORT_SYMBOL(sptlrpc_get_bulk_checksum);
1001 #else /* !__KERNEL__ */
1003 int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg,
1004 void *buf, int buflen)
1009 LASSERT(alg == BULK_HASH_ALG_ADLER32 || alg == BULK_HASH_ALG_CRC32);
1011 if (alg == BULK_HASH_ALG_ADLER32)
1016 for (i = 0; i < desc->bd_iov_count; i++) {
1017 unsigned char *ptr = desc->bd_iov[i].iov_base;
1018 int len = desc->bd_iov[i].iov_len;
1021 case BULK_HASH_ALG_ADLER32:
1023 csum32 = adler32(csum32, ptr, len);
1025 CERROR("Adler32 not supported\n");
1029 case BULK_HASH_ALG_CRC32:
1030 csum32 = crc32_le(csum32, ptr, len);
1035 csum32 = cpu_to_le32(csum32);
1036 memcpy(buf, &csum32, sizeof(csum32));
1040 #endif /* __KERNEL__ */