1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * lustre/ptlrpc/sec_bulk.c
37 *
38 * Author: Eric Mei <ericm@clusterfs.com>
39 */
40
41 #define DEBUG_SUBSYSTEM S_SEC
42
43 #include "../../include/linux/libcfs/libcfs.h"
44 #include <linux/crypto.h>
45
46 #include "../include/obd.h"
47 #include "../include/obd_cksum.h"
48 #include "../include/obd_class.h"
49 #include "../include/obd_support.h"
50 #include "../include/lustre_net.h"
51 #include "../include/lustre_import.h"
52 #include "../include/lustre_dlm.h"
53 #include "../include/lustre_sec.h"
54
55 #include "ptlrpc_internal.h"
56
57 /****************************************
58 * bulk encryption page pools *
59 ****************************************/
60
61 #define POINTERS_PER_PAGE (PAGE_CACHE_SIZE / sizeof(void *))
62 #define PAGES_PER_POOL (POINTERS_PER_PAGE)
63
64 #define IDLE_IDX_MAX (100)
65 #define IDLE_IDX_WEIGHT (3)
66
67 #define CACHE_QUIESCENT_PERIOD (20)
68
69 static struct ptlrpc_enc_page_pool {
70 /*
71 * constants
72 */
73 unsigned long epp_max_pages; /* maximum pages can hold, const */
74 unsigned int epp_max_pools; /* number of pools, const */
75
76 /*
77 * wait queue in case of not enough free pages.
78 */
79 wait_queue_head_t epp_waitq; /* waiting threads */
80 unsigned int epp_waitqlen; /* wait queue length */
81 unsigned long epp_pages_short; /* # of pages wanted of in-q users */
82 unsigned int epp_growing:1; /* during adding pages */
83
84 /*
85 * indicating how idle the pools are, from 0 to MAX_IDLE_IDX
86 * this is counted based on each time when getting pages from
87 * the pools, not based on time. which means in case that system
88 * is idled for a while but the idle_idx might still be low if no
89 * activities happened in the pools.
90 */
91 unsigned long epp_idle_idx;
92
93 /* last shrink time due to mem tight */
94 time64_t epp_last_shrink;
95 time64_t epp_last_access;
96
97 /*
98 * in-pool pages bookkeeping
99 */
100 spinlock_t epp_lock; /* protect following fields */
101 unsigned long epp_total_pages; /* total pages in pools */
102 unsigned long epp_free_pages; /* current pages available */
103
104 /*
105 * statistics
106 */
107 unsigned long epp_st_max_pages; /* # of pages ever reached */
108 unsigned int epp_st_grows; /* # of grows */
109 unsigned int epp_st_grow_fails; /* # of add pages failures */
110 unsigned int epp_st_shrinks; /* # of shrinks */
111 unsigned long epp_st_access; /* # of access */
112 unsigned long epp_st_missings; /* # of cache missing */
113 unsigned long epp_st_lowfree; /* lowest free pages reached */
114 unsigned int epp_st_max_wqlen; /* highest waitqueue length */
115 unsigned long epp_st_max_wait; /* in jiffies */
116 /*
117 * pointers to pools
118 */
119 struct page ***epp_pools;
120 } page_pools;
121
122 /*
123 * /proc/fs/lustre/sptlrpc/encrypt_page_pools
124 */
sptlrpc_proc_enc_pool_seq_show(struct seq_file * m,void * v)125 int sptlrpc_proc_enc_pool_seq_show(struct seq_file *m, void *v)
126 {
127 spin_lock(&page_pools.epp_lock);
128
129 seq_printf(m,
130 "physical pages: %lu\n"
131 "pages per pool: %lu\n"
132 "max pages: %lu\n"
133 "max pools: %u\n"
134 "total pages: %lu\n"
135 "total free: %lu\n"
136 "idle index: %lu/100\n"
137 "last shrink: %lds\n"
138 "last access: %lds\n"
139 "max pages reached: %lu\n"
140 "grows: %u\n"
141 "grows failure: %u\n"
142 "shrinks: %u\n"
143 "cache access: %lu\n"
144 "cache missing: %lu\n"
145 "low free mark: %lu\n"
146 "max waitqueue depth: %u\n"
147 "max wait time: %ld/%u\n",
148 totalram_pages,
149 PAGES_PER_POOL,
150 page_pools.epp_max_pages,
151 page_pools.epp_max_pools,
152 page_pools.epp_total_pages,
153 page_pools.epp_free_pages,
154 page_pools.epp_idle_idx,
155 (long)(ktime_get_seconds() - page_pools.epp_last_shrink),
156 (long)(ktime_get_seconds() - page_pools.epp_last_access),
157 page_pools.epp_st_max_pages,
158 page_pools.epp_st_grows,
159 page_pools.epp_st_grow_fails,
160 page_pools.epp_st_shrinks,
161 page_pools.epp_st_access,
162 page_pools.epp_st_missings,
163 page_pools.epp_st_lowfree,
164 page_pools.epp_st_max_wqlen,
165 page_pools.epp_st_max_wait,
166 HZ);
167
168 spin_unlock(&page_pools.epp_lock);
169
170 return 0;
171 }
172
enc_pools_release_free_pages(long npages)173 static void enc_pools_release_free_pages(long npages)
174 {
175 int p_idx, g_idx;
176 int p_idx_max1, p_idx_max2;
177
178 LASSERT(npages > 0);
179 LASSERT(npages <= page_pools.epp_free_pages);
180 LASSERT(page_pools.epp_free_pages <= page_pools.epp_total_pages);
181
182 /* max pool index before the release */
183 p_idx_max2 = (page_pools.epp_total_pages - 1) / PAGES_PER_POOL;
184
185 page_pools.epp_free_pages -= npages;
186 page_pools.epp_total_pages -= npages;
187
188 /* max pool index after the release */
189 p_idx_max1 = page_pools.epp_total_pages == 0 ? -1 :
190 ((page_pools.epp_total_pages - 1) / PAGES_PER_POOL);
191
192 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
193 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
194 LASSERT(page_pools.epp_pools[p_idx]);
195
196 while (npages--) {
197 LASSERT(page_pools.epp_pools[p_idx]);
198 LASSERT(page_pools.epp_pools[p_idx][g_idx] != NULL);
199
200 __free_page(page_pools.epp_pools[p_idx][g_idx]);
201 page_pools.epp_pools[p_idx][g_idx] = NULL;
202
203 if (++g_idx == PAGES_PER_POOL) {
204 p_idx++;
205 g_idx = 0;
206 }
207 }
208
209 /* free unused pools */
210 while (p_idx_max1 < p_idx_max2) {
211 LASSERT(page_pools.epp_pools[p_idx_max2]);
212 kfree(page_pools.epp_pools[p_idx_max2]);
213 page_pools.epp_pools[p_idx_max2] = NULL;
214 p_idx_max2--;
215 }
216 }
217
218 /*
219 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
220 */
enc_pools_shrink_count(struct shrinker * s,struct shrink_control * sc)221 static unsigned long enc_pools_shrink_count(struct shrinker *s,
222 struct shrink_control *sc)
223 {
224 /*
225 * if no pool access for a long time, we consider it's fully idle.
226 * a little race here is fine.
227 */
228 if (unlikely(ktime_get_seconds() - page_pools.epp_last_access >
229 CACHE_QUIESCENT_PERIOD)) {
230 spin_lock(&page_pools.epp_lock);
231 page_pools.epp_idle_idx = IDLE_IDX_MAX;
232 spin_unlock(&page_pools.epp_lock);
233 }
234
235 LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX);
236 return max((int)page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES, 0) *
237 (IDLE_IDX_MAX - page_pools.epp_idle_idx) / IDLE_IDX_MAX;
238 }
239
240 /*
241 * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool.
242 */
enc_pools_shrink_scan(struct shrinker * s,struct shrink_control * sc)243 static unsigned long enc_pools_shrink_scan(struct shrinker *s,
244 struct shrink_control *sc)
245 {
246 spin_lock(&page_pools.epp_lock);
247 sc->nr_to_scan = min_t(unsigned long, sc->nr_to_scan,
248 page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES);
249 if (sc->nr_to_scan > 0) {
250 enc_pools_release_free_pages(sc->nr_to_scan);
251 CDEBUG(D_SEC, "released %ld pages, %ld left\n",
252 (long)sc->nr_to_scan, page_pools.epp_free_pages);
253
254 page_pools.epp_st_shrinks++;
255 page_pools.epp_last_shrink = ktime_get_seconds();
256 }
257 spin_unlock(&page_pools.epp_lock);
258
259 /*
260 * if no pool access for a long time, we consider it's fully idle.
261 * a little race here is fine.
262 */
263 if (unlikely(ktime_get_seconds() - page_pools.epp_last_access >
264 CACHE_QUIESCENT_PERIOD)) {
265 spin_lock(&page_pools.epp_lock);
266 page_pools.epp_idle_idx = IDLE_IDX_MAX;
267 spin_unlock(&page_pools.epp_lock);
268 }
269
270 LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX);
271 return sc->nr_to_scan;
272 }
273
274 static inline
npages_to_npools(unsigned long npages)275 int npages_to_npools(unsigned long npages)
276 {
277 return (int) ((npages + PAGES_PER_POOL - 1) / PAGES_PER_POOL);
278 }
279
280 /*
281 * return how many pages cleaned up.
282 */
enc_pools_cleanup(struct page *** pools,int npools)283 static unsigned long enc_pools_cleanup(struct page ***pools, int npools)
284 {
285 unsigned long cleaned = 0;
286 int i, j;
287
288 for (i = 0; i < npools; i++) {
289 if (pools[i]) {
290 for (j = 0; j < PAGES_PER_POOL; j++) {
291 if (pools[i][j]) {
292 __free_page(pools[i][j]);
293 cleaned++;
294 }
295 }
296 kfree(pools[i]);
297 pools[i] = NULL;
298 }
299 }
300
301 return cleaned;
302 }
303
enc_pools_wakeup(void)304 static inline void enc_pools_wakeup(void)
305 {
306 assert_spin_locked(&page_pools.epp_lock);
307 LASSERT(page_pools.epp_waitqlen >= 0);
308
309 if (unlikely(page_pools.epp_waitqlen)) {
310 LASSERT(waitqueue_active(&page_pools.epp_waitq));
311 wake_up_all(&page_pools.epp_waitq);
312 }
313 }
314
sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc * desc)315 void sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc *desc)
316 {
317 int p_idx, g_idx;
318 int i;
319
320 if (desc->bd_enc_iov == NULL)
321 return;
322
323 LASSERT(desc->bd_iov_count > 0);
324
325 spin_lock(&page_pools.epp_lock);
326
327 p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
328 g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
329
330 LASSERT(page_pools.epp_free_pages + desc->bd_iov_count <=
331 page_pools.epp_total_pages);
332 LASSERT(page_pools.epp_pools[p_idx]);
333
334 for (i = 0; i < desc->bd_iov_count; i++) {
335 LASSERT(desc->bd_enc_iov[i].kiov_page != NULL);
336 LASSERT(g_idx != 0 || page_pools.epp_pools[p_idx]);
337 LASSERT(page_pools.epp_pools[p_idx][g_idx] == NULL);
338
339 page_pools.epp_pools[p_idx][g_idx] =
340 desc->bd_enc_iov[i].kiov_page;
341
342 if (++g_idx == PAGES_PER_POOL) {
343 p_idx++;
344 g_idx = 0;
345 }
346 }
347
348 page_pools.epp_free_pages += desc->bd_iov_count;
349
350 enc_pools_wakeup();
351
352 spin_unlock(&page_pools.epp_lock);
353
354 kfree(desc->bd_enc_iov);
355 desc->bd_enc_iov = NULL;
356 }
357 EXPORT_SYMBOL(sptlrpc_enc_pool_put_pages);
358
enc_pools_alloc(void)359 static inline void enc_pools_alloc(void)
360 {
361 LASSERT(page_pools.epp_max_pools);
362 page_pools.epp_pools =
363 libcfs_kvzalloc(page_pools.epp_max_pools *
364 sizeof(*page_pools.epp_pools),
365 GFP_NOFS);
366 }
367
enc_pools_free(void)368 static inline void enc_pools_free(void)
369 {
370 LASSERT(page_pools.epp_max_pools);
371 LASSERT(page_pools.epp_pools);
372
373 kvfree(page_pools.epp_pools);
374 }
375
376 static struct shrinker pools_shrinker = {
377 .count_objects = enc_pools_shrink_count,
378 .scan_objects = enc_pools_shrink_scan,
379 .seeks = DEFAULT_SEEKS,
380 };
381
sptlrpc_enc_pool_init(void)382 int sptlrpc_enc_pool_init(void)
383 {
384 /*
385 * maximum capacity is 1/8 of total physical memory.
386 * is the 1/8 a good number?
387 */
388 page_pools.epp_max_pages = totalram_pages / 8;
389 page_pools.epp_max_pools = npages_to_npools(page_pools.epp_max_pages);
390
391 init_waitqueue_head(&page_pools.epp_waitq);
392 page_pools.epp_waitqlen = 0;
393 page_pools.epp_pages_short = 0;
394
395 page_pools.epp_growing = 0;
396
397 page_pools.epp_idle_idx = 0;
398 page_pools.epp_last_shrink = ktime_get_seconds();
399 page_pools.epp_last_access = ktime_get_seconds();
400
401 spin_lock_init(&page_pools.epp_lock);
402 page_pools.epp_total_pages = 0;
403 page_pools.epp_free_pages = 0;
404
405 page_pools.epp_st_max_pages = 0;
406 page_pools.epp_st_grows = 0;
407 page_pools.epp_st_grow_fails = 0;
408 page_pools.epp_st_shrinks = 0;
409 page_pools.epp_st_access = 0;
410 page_pools.epp_st_missings = 0;
411 page_pools.epp_st_lowfree = 0;
412 page_pools.epp_st_max_wqlen = 0;
413 page_pools.epp_st_max_wait = 0;
414
415 enc_pools_alloc();
416 if (page_pools.epp_pools == NULL)
417 return -ENOMEM;
418
419 register_shrinker(&pools_shrinker);
420
421 return 0;
422 }
423
sptlrpc_enc_pool_fini(void)424 void sptlrpc_enc_pool_fini(void)
425 {
426 unsigned long cleaned, npools;
427
428 LASSERT(page_pools.epp_pools);
429 LASSERT(page_pools.epp_total_pages == page_pools.epp_free_pages);
430
431 unregister_shrinker(&pools_shrinker);
432
433 npools = npages_to_npools(page_pools.epp_total_pages);
434 cleaned = enc_pools_cleanup(page_pools.epp_pools, npools);
435 LASSERT(cleaned == page_pools.epp_total_pages);
436
437 enc_pools_free();
438
439 if (page_pools.epp_st_access > 0) {
440 CDEBUG(D_SEC,
441 "max pages %lu, grows %u, grow fails %u, shrinks %u, access %lu, missing %lu, max qlen %u, max wait %ld/%d\n",
442 page_pools.epp_st_max_pages, page_pools.epp_st_grows,
443 page_pools.epp_st_grow_fails,
444 page_pools.epp_st_shrinks, page_pools.epp_st_access,
445 page_pools.epp_st_missings, page_pools.epp_st_max_wqlen,
446 page_pools.epp_st_max_wait, HZ);
447 }
448 }
449
450 static int cfs_hash_alg_id[] = {
451 [BULK_HASH_ALG_NULL] = CFS_HASH_ALG_NULL,
452 [BULK_HASH_ALG_ADLER32] = CFS_HASH_ALG_ADLER32,
453 [BULK_HASH_ALG_CRC32] = CFS_HASH_ALG_CRC32,
454 [BULK_HASH_ALG_MD5] = CFS_HASH_ALG_MD5,
455 [BULK_HASH_ALG_SHA1] = CFS_HASH_ALG_SHA1,
456 [BULK_HASH_ALG_SHA256] = CFS_HASH_ALG_SHA256,
457 [BULK_HASH_ALG_SHA384] = CFS_HASH_ALG_SHA384,
458 [BULK_HASH_ALG_SHA512] = CFS_HASH_ALG_SHA512,
459 };
460
sptlrpc_get_hash_name(__u8 hash_alg)461 const char *sptlrpc_get_hash_name(__u8 hash_alg)
462 {
463 return cfs_crypto_hash_name(cfs_hash_alg_id[hash_alg]);
464 }
465 EXPORT_SYMBOL(sptlrpc_get_hash_name);
466
sptlrpc_get_hash_alg(const char * algname)467 __u8 sptlrpc_get_hash_alg(const char *algname)
468 {
469 return cfs_crypto_hash_alg(algname);
470 }
471 EXPORT_SYMBOL(sptlrpc_get_hash_alg);
472
bulk_sec_desc_unpack(struct lustre_msg * msg,int offset,int swabbed)473 int bulk_sec_desc_unpack(struct lustre_msg *msg, int offset, int swabbed)
474 {
475 struct ptlrpc_bulk_sec_desc *bsd;
476 int size = msg->lm_buflens[offset];
477
478 bsd = lustre_msg_buf(msg, offset, sizeof(*bsd));
479 if (bsd == NULL) {
480 CERROR("Invalid bulk sec desc: size %d\n", size);
481 return -EINVAL;
482 }
483
484 if (swabbed)
485 __swab32s(&bsd->bsd_nob);
486
487 if (unlikely(bsd->bsd_version != 0)) {
488 CERROR("Unexpected version %u\n", bsd->bsd_version);
489 return -EPROTO;
490 }
491
492 if (unlikely(bsd->bsd_type >= SPTLRPC_BULK_MAX)) {
493 CERROR("Invalid type %u\n", bsd->bsd_type);
494 return -EPROTO;
495 }
496
497 /* FIXME more sanity check here */
498
499 if (unlikely(bsd->bsd_svc != SPTLRPC_BULK_SVC_NULL &&
500 bsd->bsd_svc != SPTLRPC_BULK_SVC_INTG &&
501 bsd->bsd_svc != SPTLRPC_BULK_SVC_PRIV)) {
502 CERROR("Invalid svc %u\n", bsd->bsd_svc);
503 return -EPROTO;
504 }
505
506 return 0;
507 }
508 EXPORT_SYMBOL(bulk_sec_desc_unpack);
509
sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc * desc,__u8 alg,void * buf,int buflen)510 int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg,
511 void *buf, int buflen)
512 {
513 struct cfs_crypto_hash_desc *hdesc;
514 int hashsize;
515 char hashbuf[64];
516 unsigned int bufsize;
517 int i, err;
518
519 LASSERT(alg > BULK_HASH_ALG_NULL && alg < BULK_HASH_ALG_MAX);
520 LASSERT(buflen >= 4);
521
522 hdesc = cfs_crypto_hash_init(cfs_hash_alg_id[alg], NULL, 0);
523 if (IS_ERR(hdesc)) {
524 CERROR("Unable to initialize checksum hash %s\n",
525 cfs_crypto_hash_name(cfs_hash_alg_id[alg]));
526 return PTR_ERR(hdesc);
527 }
528
529 hashsize = cfs_crypto_hash_digestsize(cfs_hash_alg_id[alg]);
530
531 for (i = 0; i < desc->bd_iov_count; i++) {
532 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
533 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
534 desc->bd_iov[i].kiov_len);
535 }
536 if (hashsize > buflen) {
537 bufsize = sizeof(hashbuf);
538 err = cfs_crypto_hash_final(hdesc, (unsigned char *)hashbuf,
539 &bufsize);
540 memcpy(buf, hashbuf, buflen);
541 } else {
542 bufsize = buflen;
543 err = cfs_crypto_hash_final(hdesc, buf, &bufsize);
544 }
545
546 if (err)
547 cfs_crypto_hash_final(hdesc, NULL, NULL);
548 return err;
549 }
550 EXPORT_SYMBOL(sptlrpc_get_bulk_checksum);
551