Commit | Line | Data |
---|---|---|
43f3cec2 BB |
1 | /* |
2 | * pmemblk: IO engine that uses NVML libpmemblk to read and write data | |
3 | * | |
4 | * Copyright (C) 2016 Hewlett Packard Enterprise Development LP | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License, | |
8 | * version 2 as published by the Free Software Foundation.. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public | |
16 | * License along with this program; if not, write to the Free | |
17 | * Software Foundation, Inc., 59 Temple Place, Suite 330, | |
18 | * Boston, MA 02111-1307 USA | |
19 | */ | |
20 | ||
21 | ||
22 | /* | |
23 | * pmemblk engine | |
24 | * | |
25 | * IO engine that uses libpmemblk to read and write data | |
26 | * | |
f1d85c6e BB |
27 | * To use: |
28 | * ioengine=pmemblk | |
43f3cec2 BB |
29 | * |
30 | * Other relevant settings: | |
31 | * iodepth=1 | |
32 | * direct=1 | |
33 | * thread=1 REQUIRED | |
34 | * unlink=1 | |
35 | * filename=/pmem0/fiotestfile,BSIZE,FSIZEMB | |
36 | * | |
37 | * thread must be set to 1 for pmemblk as multiple processes cannot | |
38 | * open the same block pool file. | |
39 | * | |
40 | * iodepth should be set to 1 as pmemblk is always synchronous. | |
41 | * Use numjobs to scale up. | |
42 | * | |
43 | * direct=1 is implied as pmemblk is always direct. | |
44 | * | |
45 | * Can set unlink to 1 to remove the block pool file after testing. | |
46 | * | |
47 | * When specifying the filename, if the block pool file does not already | |
48 | * exist, then the pmemblk engine can create the pool file if you specify | |
49 | * the block and file sizes. BSIZE is the block size in bytes. | |
50 | * FSIZEMB is the pool file size in MB. | |
51 | * | |
f1d85c6e BB |
52 | * See examples/pmemblk.fio for more. |
53 | * | |
43f3cec2 BB |
54 | * libpmemblk.so |
55 | * By default, the pmemblk engine will let the system find the libpmemblk.so | |
56 | * that it uses. You can use an alternative libpmemblk by setting the | |
57 | * FIO_PMEMBLK_LIB environment variable to the full path to the desired | |
58 | * libpmemblk.so. | |
59 | * | |
60 | */ | |
61 | ||
62 | #include <stdio.h> | |
63 | #include <stdlib.h> | |
64 | #include <unistd.h> | |
65 | #include <sys/uio.h> | |
66 | #include <errno.h> | |
67 | #include <assert.h> | |
68 | #include <dlfcn.h> | |
69 | #include <string.h> | |
70 | ||
71 | #include "../fio.h" | |
72 | ||
73 | ||
74 | ||
75 | /* | |
76 | * libpmemblk | |
77 | */ | |
78 | struct PMEMblkpool_s; | |
79 | typedef struct PMEMblkpool_s PMEMblkpool; | |
80 | ||
dfec8168 BB |
81 | PMEMblkpool* (*pmemblk_create)(const char*, size_t, size_t, mode_t) = NULL; |
82 | PMEMblkpool* (*pmemblk_open)(const char*, size_t) = NULL; | |
83 | void (*pmemblk_close)(PMEMblkpool*) = NULL; | |
84 | size_t (*pmemblk_nblock)(PMEMblkpool*) = NULL; | |
85 | size_t (*pmemblk_bsize)(PMEMblkpool*) = NULL; | |
86 | int (*pmemblk_read)(PMEMblkpool*, void*, off_t) = NULL; | |
87 | int (*pmemblk_write)(PMEMblkpool*, const void*, off_t) = NULL; | |
43f3cec2 BB |
88 | |
89 | int | |
90 | load_libpmemblk( | |
91 | const char* path | |
92 | ) | |
93 | { | |
94 | void* dl; | |
95 | ||
96 | if (NULL == path) | |
97 | path = "libpmemblk.so"; | |
98 | ||
99 | dl = dlopen(path, RTLD_NOW | RTLD_NODELETE); | |
100 | if (NULL == dl) | |
101 | goto errorout; | |
102 | ||
dfec8168 | 103 | if (NULL == (pmemblk_create = dlsym(dl, "pmemblk_create"))) |
43f3cec2 | 104 | goto errorout; |
dfec8168 | 105 | if (NULL == (pmemblk_open = dlsym(dl, "pmemblk_open"))) |
43f3cec2 | 106 | goto errorout; |
dfec8168 | 107 | if (NULL == (pmemblk_close = dlsym(dl, "pmemblk_close"))) |
43f3cec2 | 108 | goto errorout; |
dfec8168 | 109 | if (NULL == (pmemblk_nblock = dlsym(dl, "pmemblk_nblock"))) |
43f3cec2 | 110 | goto errorout; |
dfec8168 | 111 | if (NULL == (pmemblk_bsize = dlsym(dl, "pmemblk_bsize"))) |
43f3cec2 | 112 | goto errorout; |
dfec8168 | 113 | if (NULL == (pmemblk_read = dlsym(dl, "pmemblk_read"))) |
43f3cec2 | 114 | goto errorout; |
dfec8168 | 115 | if (NULL == (pmemblk_write = dlsym(dl, "pmemblk_write"))) |
43f3cec2 BB |
116 | goto errorout; |
117 | ||
118 | return 0; | |
119 | ||
120 | errorout: | |
121 | log_err("fio: unable to load libpmemblk: %s\n", dlerror()); | |
122 | if (NULL != dl) | |
123 | dlclose(dl); | |
124 | ||
125 | return (-1); | |
126 | ||
127 | } /* load_libpmemblk() */ | |
128 | ||
43f3cec2 BB |
129 | |
130 | typedef struct fio_pmemblk_file* fio_pmemblk_file_t; | |
131 | struct fio_pmemblk_file { | |
132 | fio_pmemblk_file_t pmb_next; | |
133 | char* pmb_filename; | |
134 | uint64_t pmb_refcnt; | |
135 | PMEMblkpool* pmb_pool; | |
136 | size_t pmb_bsize; | |
137 | size_t pmb_nblocks; | |
138 | }; | |
139 | #define FIOFILEPMBSET(_f, _v) do { \ | |
140 | (_f)->engine_data = (uint64_t)(uintptr_t)(_v); \ | |
141 | } while(0) | |
142 | #define FIOFILEPMBGET(_f) ((fio_pmemblk_file_t)((_f)->engine_data)) | |
143 | ||
144 | static fio_pmemblk_file_t Cache = NULL; | |
145 | ||
146 | static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER; | |
147 | #define CACHE_LOCK() \ | |
148 | (void)pthread_mutex_lock(&CacheLock) | |
149 | #define CACHE_UNLOCK() \ | |
150 | (void)pthread_mutex_unlock(&CacheLock) | |
151 | ||
152 | #define PMB_CREATE (0x0001) /* should create file */ | |
153 | ||
154 | ||
155 | fio_pmemblk_file_t | |
156 | fio_pmemblk_cache_lookup( | |
157 | const char* filename | |
158 | ) | |
159 | { | |
160 | fio_pmemblk_file_t i; | |
161 | ||
162 | for (i = Cache; i != NULL; i = i->pmb_next) | |
163 | if (0 == strcmp(filename, i->pmb_filename)) | |
164 | return i; | |
165 | ||
166 | return NULL; | |
167 | ||
168 | } /* fio_pmemblk_cache_lookup() */ | |
169 | ||
170 | ||
171 | static void | |
172 | fio_pmemblk_cache_insert( | |
173 | fio_pmemblk_file_t pmb | |
174 | ) | |
175 | { | |
176 | pmb->pmb_next = Cache; | |
177 | Cache = pmb; | |
178 | ||
179 | return; | |
180 | ||
181 | } /* fio_pmemblk_cache_insert() */ | |
182 | ||
183 | ||
184 | static void | |
185 | fio_pmemblk_cache_remove( | |
186 | fio_pmemblk_file_t pmb | |
187 | ) | |
188 | { | |
189 | fio_pmemblk_file_t i; | |
190 | ||
191 | if (pmb == Cache) { | |
192 | Cache = Cache->pmb_next; | |
193 | pmb->pmb_next = NULL; | |
194 | return; | |
195 | } | |
196 | ||
197 | for (i = Cache; i != NULL; i = i->pmb_next) | |
198 | if (pmb == i->pmb_next) { | |
199 | i->pmb_next = i->pmb_next->pmb_next; | |
200 | pmb->pmb_next = NULL; | |
201 | return; | |
202 | } | |
203 | ||
204 | return; | |
205 | ||
206 | } /* fio_pmemblk_cache_remove() */ | |
207 | ||
208 | ||
209 | /* | |
210 | * to control block size and gross file size at the libpmemblk | |
211 | * level, we allow the block size and file size to be appended | |
212 | * to the file name: | |
213 | * | |
214 | * path[,bsize,fsizemb] | |
215 | * | |
216 | * note that we do not use the fio option "filesize" to dictate | |
217 | * the file size because we can only give libpmemblk the gross | |
218 | * file size, which is different from the net or usable file | |
219 | * size (which is probably what fio wants). | |
220 | * | |
221 | * the final path without the parameters is returned in ppath. | |
222 | * the block size and file size are returned in pbsize and fsize. | |
223 | * | |
224 | * note that the user should specify the file size in MiB, but | |
225 | * we return bytes from here. | |
226 | */ | |
227 | static void | |
228 | pmb_parse_path( | |
229 | const char* pathspec, | |
230 | char** ppath, | |
231 | uint64_t* pbsize, | |
232 | uint64_t* pfsize | |
233 | ) | |
234 | { | |
235 | char* path; | |
236 | char* s; | |
237 | uint64_t bsize; | |
238 | uint64_t fsizemb; | |
239 | ||
240 | path = strdup(pathspec); | |
241 | if (NULL == path) { | |
242 | *ppath = NULL; | |
243 | return; | |
244 | } | |
245 | ||
246 | /* extract sizes, if given */ | |
247 | s = strrchr(path, ','); | |
248 | if (s && (fsizemb = strtoull(s+1, NULL, 10))) { | |
249 | *s = 0; | |
250 | s = strrchr(path, ','); | |
251 | if (s && (bsize = strtoull(s+1, NULL, 10))) { | |
252 | *s = 0; | |
253 | *ppath = path; | |
254 | *pbsize = bsize; | |
255 | *pfsize = fsizemb << 20; | |
256 | return; | |
257 | } | |
258 | } | |
259 | ||
260 | /* size specs not found */ | |
261 | strcpy(path, pathspec); | |
262 | *ppath = path; | |
263 | *pbsize = 0; | |
264 | *pfsize = 0; | |
265 | return; | |
266 | ||
267 | } /* pmb_parse_path() */ | |
268 | ||
269 | ||
270 | static | |
271 | fio_pmemblk_file_t | |
272 | pmb_open( | |
273 | const char* pathspec, | |
274 | int flags | |
275 | ) | |
276 | { | |
277 | fio_pmemblk_file_t pmb; | |
278 | char* path = NULL; | |
279 | uint64_t bsize = 0; | |
280 | uint64_t fsize = 0; | |
281 | ||
282 | pmb_parse_path(pathspec, &path, &bsize, &fsize); | |
283 | if (NULL == path) | |
284 | return NULL; | |
285 | ||
286 | CACHE_LOCK(); | |
287 | ||
288 | pmb = fio_pmemblk_cache_lookup(path); | |
289 | ||
290 | if (NULL == pmb) { | |
291 | /* load libpmemblk if needed */ | |
292 | if (NULL == pmemblk_open) | |
293 | if (0 != load_libpmemblk(getenv("FIO_PMEMBLK_LIB"))) | |
294 | goto error; | |
295 | ||
296 | pmb = malloc(sizeof(*pmb)); | |
297 | if (NULL == pmb) | |
298 | goto error; | |
299 | ||
300 | /* try opening existing first, create it if needed */ | |
301 | pmb->pmb_pool = pmemblk_open(path, bsize); | |
302 | if ((NULL == pmb->pmb_pool) && | |
303 | (ENOENT == errno) && | |
304 | (flags & PMB_CREATE) && | |
305 | (0 < fsize) && | |
306 | (0 < bsize)) { | |
307 | pmb->pmb_pool = pmemblk_create(path, bsize, fsize, 0644); | |
308 | } | |
309 | if (NULL == pmb->pmb_pool) { | |
310 | log_err("fio: enable to open pmemblk pool file (errno %d)\n", | |
311 | errno); | |
312 | goto error; | |
313 | } | |
314 | ||
315 | pmb->pmb_filename = path; | |
316 | pmb->pmb_next = NULL; | |
317 | pmb->pmb_refcnt = 0; | |
318 | pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool); | |
319 | pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool); | |
320 | ||
321 | fio_pmemblk_cache_insert(pmb); | |
322 | } | |
323 | ||
324 | pmb->pmb_refcnt += 1; | |
325 | ||
326 | CACHE_UNLOCK(); | |
327 | ||
328 | return pmb; | |
329 | ||
330 | error: | |
331 | if (NULL != pmb) { | |
332 | if (NULL != pmb->pmb_pool) | |
333 | pmemblk_close(pmb->pmb_pool); | |
334 | pmb->pmb_pool = NULL; | |
335 | pmb->pmb_filename = NULL; | |
336 | free(pmb); | |
337 | } | |
338 | if (NULL != path) | |
339 | free(path); | |
340 | CACHE_UNLOCK(); | |
341 | return NULL; | |
342 | ||
343 | } /* pmb_open() */ | |
344 | ||
345 | ||
346 | static void | |
347 | pmb_close( | |
348 | fio_pmemblk_file_t pmb, | |
349 | const int keep | |
350 | ) | |
351 | { | |
352 | CACHE_LOCK(); | |
353 | ||
354 | pmb->pmb_refcnt--; | |
355 | ||
356 | if (!keep && (0 == pmb->pmb_refcnt)) { | |
357 | pmemblk_close(pmb->pmb_pool); | |
358 | pmb->pmb_pool = NULL; | |
359 | free(pmb->pmb_filename); | |
360 | pmb->pmb_filename = NULL; | |
361 | fio_pmemblk_cache_remove(pmb); | |
362 | free(pmb); | |
363 | } | |
364 | ||
365 | CACHE_UNLOCK(); | |
366 | ||
367 | } /* pmb_close() */ | |
368 | ||
369 | ||
370 | static int | |
371 | pmb_get_flags( | |
372 | struct thread_data* td, | |
373 | uint64_t* pflags | |
374 | ) | |
375 | { | |
376 | static int thread_warned = 0; | |
377 | static int odirect_warned = 0; | |
378 | ||
379 | uint64_t flags = 0; | |
380 | ||
381 | if (!td->o.use_thread) { | |
382 | if (!thread_warned) { | |
383 | thread_warned = 1; | |
384 | log_err("fio: must set thread=1 for pmemblk engine\n"); | |
385 | } | |
386 | return 1; | |
387 | } | |
388 | ||
389 | if (!td->o.odirect && !odirect_warned) { | |
390 | odirect_warned = 1; | |
391 | log_info("fio: direct == 0, but pmemblk is always direct\n"); | |
392 | } | |
393 | ||
394 | if (td->o.allow_create) | |
395 | flags |= PMB_CREATE; | |
396 | ||
397 | (*pflags) = flags; | |
398 | return 0; | |
399 | ||
400 | } /* pmb_get_flags() */ | |
401 | ||
402 | ||
403 | static int | |
404 | fio_pmemblk_open_file( | |
405 | struct thread_data* td, | |
406 | struct fio_file* f) | |
407 | { | |
408 | uint64_t flags = 0; | |
409 | fio_pmemblk_file_t pmb; | |
410 | ||
411 | if (0 != pmb_get_flags(td, &flags)) | |
412 | return 1; | |
413 | ||
414 | pmb = pmb_open(f->file_name, flags); | |
415 | if (NULL == pmb) | |
416 | return 1; | |
417 | ||
418 | FIOFILEPMBSET(f, pmb); | |
419 | ||
420 | return 0; | |
421 | ||
422 | } /* fio_pmemblk_open_file() */ | |
423 | ||
424 | ||
425 | static int | |
426 | fio_pmemblk_close_file( | |
427 | struct thread_data fio_unused* td, | |
428 | struct fio_file* f | |
429 | ) | |
430 | { | |
431 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
432 | ||
433 | if (pmb) | |
434 | pmb_close(pmb, 0); | |
435 | ||
436 | FIOFILEPMBSET(f, NULL); | |
437 | ||
438 | return 0; | |
439 | ||
440 | } /* fio_pmemblk_close_file() */ | |
441 | ||
442 | ||
443 | static int | |
444 | fio_pmemblk_get_file_size( | |
445 | struct thread_data* td, | |
446 | struct fio_file* f | |
447 | ) | |
448 | { | |
449 | uint64_t flags = 0; | |
450 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
451 | ||
452 | if (fio_file_size_known(f)) | |
453 | return 0; | |
454 | ||
455 | if (NULL == pmb) { | |
456 | if (0 != pmb_get_flags(td, &flags)) | |
457 | return 1; | |
458 | pmb = pmb_open(f->file_name, flags); | |
459 | if (NULL == pmb) | |
460 | return 1; | |
461 | } | |
462 | ||
463 | f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks; | |
464 | ||
465 | fio_file_set_size_known(f); | |
466 | ||
467 | if (NULL == FIOFILEPMBGET(f)) | |
468 | pmb_close(pmb, 1); | |
469 | ||
470 | return 0; | |
471 | ||
472 | } /* fio_pmemblk_get_file_size() */ | |
473 | ||
474 | ||
475 | static int | |
476 | fio_pmemblk_queue( | |
477 | struct thread_data* td, | |
478 | struct io_u* io_u) | |
479 | { | |
480 | struct fio_file* f = io_u->file; | |
481 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
482 | ||
483 | unsigned long long off; | |
484 | unsigned long len; | |
485 | void* buf; | |
486 | int (*blkop)(PMEMblkpool*, void*, off_t) = (void*)pmemblk_write; | |
487 | ||
488 | fio_ro_check(td, io_u); | |
489 | ||
490 | switch (io_u->ddir) { | |
491 | case DDIR_READ: | |
492 | blkop = pmemblk_read; | |
493 | /* fall through */ | |
494 | case DDIR_WRITE: | |
495 | off = io_u->offset; | |
496 | len = io_u->xfer_buflen; | |
497 | ||
498 | io_u->error = EINVAL; | |
499 | if (0 != (off % pmb->pmb_bsize)) | |
500 | break; | |
501 | if (0 != (len % pmb->pmb_bsize)) | |
502 | break; | |
503 | if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks) | |
504 | break; | |
505 | ||
506 | io_u->error = 0; | |
507 | buf = io_u->xfer_buf; | |
508 | off /= pmb->pmb_bsize; | |
509 | len /= pmb->pmb_bsize; | |
510 | while (0 < len) { | |
511 | if (0 != blkop(pmb->pmb_pool, buf, off)) { | |
512 | io_u->error = errno; | |
513 | break; | |
514 | } | |
515 | buf += pmb->pmb_bsize; | |
516 | off++; | |
517 | len--; | |
518 | } | |
519 | off *= pmb->pmb_bsize; | |
520 | len *= pmb->pmb_bsize; | |
521 | io_u->resid = io_u->xfer_buflen - (off - io_u->offset); | |
522 | break; | |
523 | case DDIR_SYNC: | |
524 | case DDIR_DATASYNC: | |
525 | case DDIR_SYNC_FILE_RANGE: | |
526 | /* we're always sync'd */ | |
527 | io_u->error = 0; | |
528 | break; | |
529 | default: | |
530 | io_u->error = EINVAL; | |
531 | break; | |
532 | } | |
533 | ||
534 | return FIO_Q_COMPLETED; | |
535 | ||
536 | } /* fio_pmemblk_queue() */ | |
537 | ||
538 | ||
539 | static int | |
540 | fio_pmemblk_unlink_file( | |
541 | struct thread_data* td, | |
542 | struct fio_file* f | |
543 | ) | |
544 | { | |
545 | char* path = NULL; | |
546 | uint64_t bsize = 0; | |
547 | uint64_t fsize = 0; | |
548 | ||
549 | /* | |
550 | * we need our own unlink in case the user has specified | |
551 | * the block and file sizes in the path name. we parse | |
552 | * the file_name to determine the file name we actually used. | |
553 | */ | |
554 | ||
555 | pmb_parse_path(f->file_name, &path, &bsize, &fsize); | |
556 | if (NULL == path) | |
557 | return 1; | |
558 | ||
559 | unlink(path); | |
560 | free(path); | |
561 | ||
562 | return 0; | |
563 | ||
564 | } /* fio_pmemblk_unlink_file() */ | |
565 | ||
566 | ||
567 | struct ioengine_ops ioengine = { | |
568 | .name = "pmemblk", | |
569 | .version = FIO_IOOPS_VERSION, | |
570 | .queue = fio_pmemblk_queue, | |
571 | .open_file = fio_pmemblk_open_file, | |
572 | .close_file = fio_pmemblk_close_file, | |
573 | .get_file_size = fio_pmemblk_get_file_size, | |
574 | .unlink_file = fio_pmemblk_unlink_file, | |
575 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL, | |
576 | }; | |
577 | ||
578 | ||
579 | static void | |
580 | fio_init fio_pmemblk_register(void) | |
581 | { | |
582 | register_ioengine(&ioengine); | |
583 | } | |
584 | ||
585 | ||
586 | static void | |
587 | fio_exit fio_pmemblk_unregister(void) | |
588 | { | |
589 | unregister_ioengine(&ioengine); | |
590 | } | |
591 |