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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
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>
41 #define DEBUG_SUBSYSTEM S_SEC
43 #include "../../include/linux/libcfs/libcfs.h"
45 #include "../include/obd.h"
46 #include "../include/obd_cksum.h"
47 #include "../include/obd_class.h"
48 #include "../include/obd_support.h"
49 #include "../include/lustre_net.h"
50 #include "../include/lustre_import.h"
51 #include "../include/lustre_dlm.h"
52 #include "../include/lustre_sec.h"
54 #include "ptlrpc_internal.h"
56 /****************************************
57 * bulk encryption page pools *
58 ****************************************/
60 #define POINTERS_PER_PAGE (PAGE_SIZE / sizeof(void *))
61 #define PAGES_PER_POOL (POINTERS_PER_PAGE)
63 #define IDLE_IDX_MAX (100)
64 #define IDLE_IDX_WEIGHT (3)
66 #define CACHE_QUIESCENT_PERIOD (20)
68 static struct ptlrpc_enc_page_pool {
72 unsigned long epp_max_pages; /* maximum pages can hold, const */
73 unsigned int epp_max_pools; /* number of pools, const */
76 * wait queue in case of not enough free pages.
78 wait_queue_head_t epp_waitq; /* waiting threads */
79 unsigned int epp_waitqlen; /* wait queue length */
80 unsigned long epp_pages_short; /* # of pages wanted of in-q users */
81 unsigned int epp_growing:1; /* during adding pages */
84 * indicating how idle the pools are, from 0 to MAX_IDLE_IDX
85 * this is counted based on each time when getting pages from
86 * the pools, not based on time. which means in case that system
87 * is idled for a while but the idle_idx might still be low if no
88 * activities happened in the pools.
90 unsigned long epp_idle_idx;
92 /* last shrink time due to mem tight */
93 time64_t epp_last_shrink;
94 time64_t epp_last_access;
97 * in-pool pages bookkeeping
99 spinlock_t epp_lock; /* protect following fields */
100 unsigned long epp_total_pages; /* total pages in pools */
101 unsigned long epp_free_pages; /* current pages available */
106 unsigned long epp_st_max_pages; /* # of pages ever reached */
107 unsigned int epp_st_grows; /* # of grows */
108 unsigned int epp_st_grow_fails; /* # of add pages failures */
109 unsigned int epp_st_shrinks; /* # of shrinks */
110 unsigned long epp_st_access; /* # of access */
111 unsigned long epp_st_missings; /* # of cache missing */
112 unsigned long epp_st_lowfree; /* lowest free pages reached */
113 unsigned int epp_st_max_wqlen; /* highest waitqueue length */
114 unsigned long epp_st_max_wait; /* in jiffies */
118 struct page ***epp_pools;
122 * /sys/kernel/debug/lustre/sptlrpc/encrypt_page_pools
124 int sptlrpc_proc_enc_pool_seq_show(struct seq_file *m, void *v)
126 spin_lock(&page_pools.epp_lock);
129 "physical pages: %lu\n"
130 "pages per pool: %lu\n"
135 "idle index: %lu/100\n"
136 "last shrink: %lds\n"
137 "last access: %lds\n"
138 "max pages reached: %lu\n"
140 "grows failure: %u\n"
142 "cache access: %lu\n"
143 "cache missing: %lu\n"
144 "low free mark: %lu\n"
145 "max waitqueue depth: %u\n"
146 "max wait time: %ld/%u\n",
149 page_pools.epp_max_pages,
150 page_pools.epp_max_pools,
151 page_pools.epp_total_pages,
152 page_pools.epp_free_pages,
153 page_pools.epp_idle_idx,
154 (long)(ktime_get_seconds() - page_pools.epp_last_shrink),
155 (long)(ktime_get_seconds() - page_pools.epp_last_access),
156 page_pools.epp_st_max_pages,
157 page_pools.epp_st_grows,
158 page_pools.epp_st_grow_fails,
159 page_pools.epp_st_shrinks,
160 page_pools.epp_st_access,
161 page_pools.epp_st_missings,
162 page_pools.epp_st_lowfree,
163 page_pools.epp_st_max_wqlen,
164 page_pools.epp_st_max_wait,
167 spin_unlock(&page_pools.epp_lock);
172 static void enc_pools_release_free_pages(long npages)
175 int p_idx_max1, p_idx_max2;
178 LASSERT(npages <= page_pools.epp_free_pages);
179 LASSERT(page_pools.epp_free_pages <= page_pools.epp_total_pages);
181 /* max pool index before the release */
182 p_idx_max2 = (page_pools.epp_total_pages - 1) / PAGES_PER_POOL;
184 page_pools.epp_free_pages -= npages;
185 page_pools.epp_total_pages -= npages;
187 /* max pool index after the release */
188 p_idx_max1 = page_pools.epp_total_pages == 0 ? -1 :
189 ((page_pools.epp_total_pages - 1) / PAGES_PER_POOL);
191 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
192 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
193 LASSERT(page_pools.epp_pools[p_idx]);
196 LASSERT(page_pools.epp_pools[p_idx]);
197 LASSERT(page_pools.epp_pools[p_idx][g_idx]);
199 __free_page(page_pools.epp_pools[p_idx][g_idx]);
200 page_pools.epp_pools[p_idx][g_idx] = NULL;
202 if (++g_idx == PAGES_PER_POOL) {
208 /* free unused pools */
209 while (p_idx_max1 < p_idx_max2) {
210 LASSERT(page_pools.epp_pools[p_idx_max2]);
211 kfree(page_pools.epp_pools[p_idx_max2]);
212 page_pools.epp_pools[p_idx_max2] = NULL;
218 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
220 static unsigned long enc_pools_shrink_count(struct shrinker *s,
221 struct shrink_control *sc)
224 * if no pool access for a long time, we consider it's fully idle.
225 * a little race here is fine.
227 if (unlikely(ktime_get_seconds() - page_pools.epp_last_access >
228 CACHE_QUIESCENT_PERIOD)) {
229 spin_lock(&page_pools.epp_lock);
230 page_pools.epp_idle_idx = IDLE_IDX_MAX;
231 spin_unlock(&page_pools.epp_lock);
234 LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX);
235 return max((int)page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES, 0) *
236 (IDLE_IDX_MAX - page_pools.epp_idle_idx) / IDLE_IDX_MAX;
240 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
242 static unsigned long enc_pools_shrink_scan(struct shrinker *s,
243 struct shrink_control *sc)
245 spin_lock(&page_pools.epp_lock);
246 sc->nr_to_scan = min_t(unsigned long, sc->nr_to_scan,
247 page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES);
248 if (sc->nr_to_scan > 0) {
249 enc_pools_release_free_pages(sc->nr_to_scan);
250 CDEBUG(D_SEC, "released %ld pages, %ld left\n",
251 (long)sc->nr_to_scan, page_pools.epp_free_pages);
253 page_pools.epp_st_shrinks++;
254 page_pools.epp_last_shrink = ktime_get_seconds();
256 spin_unlock(&page_pools.epp_lock);
259 * if no pool access for a long time, we consider it's fully idle.
260 * a little race here is fine.
262 if (unlikely(ktime_get_seconds() - page_pools.epp_last_access >
263 CACHE_QUIESCENT_PERIOD)) {
264 spin_lock(&page_pools.epp_lock);
265 page_pools.epp_idle_idx = IDLE_IDX_MAX;
266 spin_unlock(&page_pools.epp_lock);
269 LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX);
270 return sc->nr_to_scan;
274 int npages_to_npools(unsigned long npages)
276 return (int) ((npages + PAGES_PER_POOL - 1) / PAGES_PER_POOL);
280 * return how many pages cleaned up.
282 static unsigned long enc_pools_cleanup(struct page ***pools, int npools)
284 unsigned long cleaned = 0;
287 for (i = 0; i < npools; i++) {
289 for (j = 0; j < PAGES_PER_POOL; j++) {
291 __free_page(pools[i][j]);
303 static inline void enc_pools_wakeup(void)
305 assert_spin_locked(&page_pools.epp_lock);
307 if (unlikely(page_pools.epp_waitqlen)) {
308 LASSERT(waitqueue_active(&page_pools.epp_waitq));
309 wake_up_all(&page_pools.epp_waitq);
313 void sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc *desc)
318 if (!desc->bd_enc_iov)
321 LASSERT(desc->bd_iov_count > 0);
323 spin_lock(&page_pools.epp_lock);
325 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
326 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
328 LASSERT(page_pools.epp_free_pages + desc->bd_iov_count <=
329 page_pools.epp_total_pages);
330 LASSERT(page_pools.epp_pools[p_idx]);
332 for (i = 0; i < desc->bd_iov_count; i++) {
333 LASSERT(desc->bd_enc_iov[i].kiov_page);
334 LASSERT(g_idx != 0 || page_pools.epp_pools[p_idx]);
335 LASSERT(!page_pools.epp_pools[p_idx][g_idx]);
337 page_pools.epp_pools[p_idx][g_idx] =
338 desc->bd_enc_iov[i].kiov_page;
340 if (++g_idx == PAGES_PER_POOL) {
346 page_pools.epp_free_pages += desc->bd_iov_count;
350 spin_unlock(&page_pools.epp_lock);
352 kfree(desc->bd_enc_iov);
353 desc->bd_enc_iov = NULL;
355 EXPORT_SYMBOL(sptlrpc_enc_pool_put_pages);
357 static inline void enc_pools_alloc(void)
359 LASSERT(page_pools.epp_max_pools);
360 page_pools.epp_pools =
361 libcfs_kvzalloc(page_pools.epp_max_pools *
362 sizeof(*page_pools.epp_pools),
366 static inline void enc_pools_free(void)
368 LASSERT(page_pools.epp_max_pools);
369 LASSERT(page_pools.epp_pools);
371 kvfree(page_pools.epp_pools);
374 static struct shrinker pools_shrinker = {
375 .count_objects = enc_pools_shrink_count,
376 .scan_objects = enc_pools_shrink_scan,
377 .seeks = DEFAULT_SEEKS,
380 int sptlrpc_enc_pool_init(void)
383 * maximum capacity is 1/8 of total physical memory.
384 * is the 1/8 a good number?
386 page_pools.epp_max_pages = totalram_pages / 8;
387 page_pools.epp_max_pools = npages_to_npools(page_pools.epp_max_pages);
389 init_waitqueue_head(&page_pools.epp_waitq);
390 page_pools.epp_waitqlen = 0;
391 page_pools.epp_pages_short = 0;
393 page_pools.epp_growing = 0;
395 page_pools.epp_idle_idx = 0;
396 page_pools.epp_last_shrink = ktime_get_seconds();
397 page_pools.epp_last_access = ktime_get_seconds();
399 spin_lock_init(&page_pools.epp_lock);
400 page_pools.epp_total_pages = 0;
401 page_pools.epp_free_pages = 0;
403 page_pools.epp_st_max_pages = 0;
404 page_pools.epp_st_grows = 0;
405 page_pools.epp_st_grow_fails = 0;
406 page_pools.epp_st_shrinks = 0;
407 page_pools.epp_st_access = 0;
408 page_pools.epp_st_missings = 0;
409 page_pools.epp_st_lowfree = 0;
410 page_pools.epp_st_max_wqlen = 0;
411 page_pools.epp_st_max_wait = 0;
414 if (!page_pools.epp_pools)
417 register_shrinker(&pools_shrinker);
422 void sptlrpc_enc_pool_fini(void)
424 unsigned long cleaned, npools;
426 LASSERT(page_pools.epp_pools);
427 LASSERT(page_pools.epp_total_pages == page_pools.epp_free_pages);
429 unregister_shrinker(&pools_shrinker);
431 npools = npages_to_npools(page_pools.epp_total_pages);
432 cleaned = enc_pools_cleanup(page_pools.epp_pools, npools);
433 LASSERT(cleaned == page_pools.epp_total_pages);
437 if (page_pools.epp_st_access > 0) {
439 "max pages %lu, grows %u, grow fails %u, shrinks %u, access %lu, missing %lu, max qlen %u, max wait %ld/%d\n",
440 page_pools.epp_st_max_pages, page_pools.epp_st_grows,
441 page_pools.epp_st_grow_fails,
442 page_pools.epp_st_shrinks, page_pools.epp_st_access,
443 page_pools.epp_st_missings, page_pools.epp_st_max_wqlen,
444 page_pools.epp_st_max_wait, HZ);
448 static int cfs_hash_alg_id[] = {
449 [BULK_HASH_ALG_NULL] = CFS_HASH_ALG_NULL,
450 [BULK_HASH_ALG_ADLER32] = CFS_HASH_ALG_ADLER32,
451 [BULK_HASH_ALG_CRC32] = CFS_HASH_ALG_CRC32,
452 [BULK_HASH_ALG_MD5] = CFS_HASH_ALG_MD5,
453 [BULK_HASH_ALG_SHA1] = CFS_HASH_ALG_SHA1,
454 [BULK_HASH_ALG_SHA256] = CFS_HASH_ALG_SHA256,
455 [BULK_HASH_ALG_SHA384] = CFS_HASH_ALG_SHA384,
456 [BULK_HASH_ALG_SHA512] = CFS_HASH_ALG_SHA512,
459 const char *sptlrpc_get_hash_name(__u8 hash_alg)
461 return cfs_crypto_hash_name(cfs_hash_alg_id[hash_alg]);
463 EXPORT_SYMBOL(sptlrpc_get_hash_name);
465 __u8 sptlrpc_get_hash_alg(const char *algname)
467 return cfs_crypto_hash_alg(algname);
469 EXPORT_SYMBOL(sptlrpc_get_hash_alg);
471 int bulk_sec_desc_unpack(struct lustre_msg *msg, int offset, int swabbed)
473 struct ptlrpc_bulk_sec_desc *bsd;
474 int size = msg->lm_buflens[offset];
476 bsd = lustre_msg_buf(msg, offset, sizeof(*bsd));
478 CERROR("Invalid bulk sec desc: size %d\n", size);
483 __swab32s(&bsd->bsd_nob);
485 if (unlikely(bsd->bsd_version != 0)) {
486 CERROR("Unexpected version %u\n", bsd->bsd_version);
490 if (unlikely(bsd->bsd_type >= SPTLRPC_BULK_MAX)) {
491 CERROR("Invalid type %u\n", bsd->bsd_type);
495 /* FIXME more sanity check here */
497 if (unlikely(bsd->bsd_svc != SPTLRPC_BULK_SVC_NULL &&
498 bsd->bsd_svc != SPTLRPC_BULK_SVC_INTG &&
499 bsd->bsd_svc != SPTLRPC_BULK_SVC_PRIV)) {
500 CERROR("Invalid svc %u\n", bsd->bsd_svc);
506 EXPORT_SYMBOL(bulk_sec_desc_unpack);
508 int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg,
509 void *buf, int buflen)
511 struct cfs_crypto_hash_desc *hdesc;
513 unsigned int bufsize;
516 LASSERT(alg > BULK_HASH_ALG_NULL && alg < BULK_HASH_ALG_MAX);
517 LASSERT(buflen >= 4);
519 hdesc = cfs_crypto_hash_init(cfs_hash_alg_id[alg], NULL, 0);
521 CERROR("Unable to initialize checksum hash %s\n",
522 cfs_crypto_hash_name(cfs_hash_alg_id[alg]));
523 return PTR_ERR(hdesc);
526 hashsize = cfs_crypto_hash_digestsize(cfs_hash_alg_id[alg]);
528 for (i = 0; i < desc->bd_iov_count; i++) {
529 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
530 desc->bd_iov[i].kiov_offset & ~PAGE_MASK,
531 desc->bd_iov[i].kiov_len);
534 if (hashsize > buflen) {
535 unsigned char hashbuf[CFS_CRYPTO_HASH_DIGESTSIZE_MAX];
537 bufsize = sizeof(hashbuf);
538 LASSERTF(bufsize >= hashsize, "bufsize = %u < hashsize %u\n",
540 err = cfs_crypto_hash_final(hdesc, hashbuf, &bufsize);
541 memcpy(buf, hashbuf, buflen);
544 err = cfs_crypto_hash_final(hdesc, buf, &bufsize);
549 EXPORT_SYMBOL(sptlrpc_get_bulk_checksum);