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 | ||
81 | PMEMblkpool* (*pmemblk_create_ptr)(const char*, size_t, size_t, mode_t) = NULL; | |
82 | PMEMblkpool* (*pmemblk_open_ptr)(const char*, size_t) = NULL; | |
83 | void (*pmemblk_close_ptr)(PMEMblkpool*) = NULL; | |
84 | size_t (*pmemblk_nblock_ptr)(PMEMblkpool*) = NULL; | |
85 | size_t (*pmemblk_bsize_ptr)(PMEMblkpool*) = NULL; | |
86 | int (*pmemblk_read_ptr)(PMEMblkpool*, void*, off_t) = NULL; | |
87 | int (*pmemblk_write_ptr)(PMEMblkpool*, const void*, off_t) = NULL; | |
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 | ||
103 | if (NULL == (pmemblk_create_ptr = dlsym(dl, "pmemblk_create"))) | |
104 | goto errorout; | |
105 | if (NULL == (pmemblk_open_ptr = dlsym(dl, "pmemblk_open"))) | |
106 | goto errorout; | |
107 | if (NULL == (pmemblk_close_ptr = dlsym(dl, "pmemblk_close"))) | |
108 | goto errorout; | |
109 | if (NULL == (pmemblk_nblock_ptr = dlsym(dl, "pmemblk_nblock"))) | |
110 | goto errorout; | |
111 | if (NULL == (pmemblk_bsize_ptr = dlsym(dl, "pmemblk_bsize"))) | |
112 | goto errorout; | |
113 | if (NULL == (pmemblk_read_ptr = dlsym(dl, "pmemblk_read"))) | |
114 | goto errorout; | |
115 | if (NULL == (pmemblk_write_ptr = dlsym(dl, "pmemblk_write"))) | |
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 | ||
129 | #define pmemblk_create pmemblk_create_ptr | |
130 | #define pmemblk_open pmemblk_open_ptr | |
131 | #define pmemblk_close pmemblk_close_ptr | |
132 | #define pmemblk_nblock pmemblk_nblock_ptr | |
133 | #define pmemblk_bsize pmemblk_bsize_ptr | |
134 | #define pmemblk_read pmemblk_read_ptr | |
135 | #define pmemblk_write pmemblk_write_ptr | |
136 | ||
137 | ||
138 | typedef struct fio_pmemblk_file* fio_pmemblk_file_t; | |
139 | struct fio_pmemblk_file { | |
140 | fio_pmemblk_file_t pmb_next; | |
141 | char* pmb_filename; | |
142 | uint64_t pmb_refcnt; | |
143 | PMEMblkpool* pmb_pool; | |
144 | size_t pmb_bsize; | |
145 | size_t pmb_nblocks; | |
146 | }; | |
147 | #define FIOFILEPMBSET(_f, _v) do { \ | |
148 | (_f)->engine_data = (uint64_t)(uintptr_t)(_v); \ | |
149 | } while(0) | |
150 | #define FIOFILEPMBGET(_f) ((fio_pmemblk_file_t)((_f)->engine_data)) | |
151 | ||
152 | static fio_pmemblk_file_t Cache = NULL; | |
153 | ||
154 | static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER; | |
155 | #define CACHE_LOCK() \ | |
156 | (void)pthread_mutex_lock(&CacheLock) | |
157 | #define CACHE_UNLOCK() \ | |
158 | (void)pthread_mutex_unlock(&CacheLock) | |
159 | ||
160 | #define PMB_CREATE (0x0001) /* should create file */ | |
161 | ||
162 | ||
163 | fio_pmemblk_file_t | |
164 | fio_pmemblk_cache_lookup( | |
165 | const char* filename | |
166 | ) | |
167 | { | |
168 | fio_pmemblk_file_t i; | |
169 | ||
170 | for (i = Cache; i != NULL; i = i->pmb_next) | |
171 | if (0 == strcmp(filename, i->pmb_filename)) | |
172 | return i; | |
173 | ||
174 | return NULL; | |
175 | ||
176 | } /* fio_pmemblk_cache_lookup() */ | |
177 | ||
178 | ||
179 | static void | |
180 | fio_pmemblk_cache_insert( | |
181 | fio_pmemblk_file_t pmb | |
182 | ) | |
183 | { | |
184 | pmb->pmb_next = Cache; | |
185 | Cache = pmb; | |
186 | ||
187 | return; | |
188 | ||
189 | } /* fio_pmemblk_cache_insert() */ | |
190 | ||
191 | ||
192 | static void | |
193 | fio_pmemblk_cache_remove( | |
194 | fio_pmemblk_file_t pmb | |
195 | ) | |
196 | { | |
197 | fio_pmemblk_file_t i; | |
198 | ||
199 | if (pmb == Cache) { | |
200 | Cache = Cache->pmb_next; | |
201 | pmb->pmb_next = NULL; | |
202 | return; | |
203 | } | |
204 | ||
205 | for (i = Cache; i != NULL; i = i->pmb_next) | |
206 | if (pmb == i->pmb_next) { | |
207 | i->pmb_next = i->pmb_next->pmb_next; | |
208 | pmb->pmb_next = NULL; | |
209 | return; | |
210 | } | |
211 | ||
212 | return; | |
213 | ||
214 | } /* fio_pmemblk_cache_remove() */ | |
215 | ||
216 | ||
217 | /* | |
218 | * to control block size and gross file size at the libpmemblk | |
219 | * level, we allow the block size and file size to be appended | |
220 | * to the file name: | |
221 | * | |
222 | * path[,bsize,fsizemb] | |
223 | * | |
224 | * note that we do not use the fio option "filesize" to dictate | |
225 | * the file size because we can only give libpmemblk the gross | |
226 | * file size, which is different from the net or usable file | |
227 | * size (which is probably what fio wants). | |
228 | * | |
229 | * the final path without the parameters is returned in ppath. | |
230 | * the block size and file size are returned in pbsize and fsize. | |
231 | * | |
232 | * note that the user should specify the file size in MiB, but | |
233 | * we return bytes from here. | |
234 | */ | |
235 | static void | |
236 | pmb_parse_path( | |
237 | const char* pathspec, | |
238 | char** ppath, | |
239 | uint64_t* pbsize, | |
240 | uint64_t* pfsize | |
241 | ) | |
242 | { | |
243 | char* path; | |
244 | char* s; | |
245 | uint64_t bsize; | |
246 | uint64_t fsizemb; | |
247 | ||
248 | path = strdup(pathspec); | |
249 | if (NULL == path) { | |
250 | *ppath = NULL; | |
251 | return; | |
252 | } | |
253 | ||
254 | /* extract sizes, if given */ | |
255 | s = strrchr(path, ','); | |
256 | if (s && (fsizemb = strtoull(s+1, NULL, 10))) { | |
257 | *s = 0; | |
258 | s = strrchr(path, ','); | |
259 | if (s && (bsize = strtoull(s+1, NULL, 10))) { | |
260 | *s = 0; | |
261 | *ppath = path; | |
262 | *pbsize = bsize; | |
263 | *pfsize = fsizemb << 20; | |
264 | return; | |
265 | } | |
266 | } | |
267 | ||
268 | /* size specs not found */ | |
269 | strcpy(path, pathspec); | |
270 | *ppath = path; | |
271 | *pbsize = 0; | |
272 | *pfsize = 0; | |
273 | return; | |
274 | ||
275 | } /* pmb_parse_path() */ | |
276 | ||
277 | ||
278 | static | |
279 | fio_pmemblk_file_t | |
280 | pmb_open( | |
281 | const char* pathspec, | |
282 | int flags | |
283 | ) | |
284 | { | |
285 | fio_pmemblk_file_t pmb; | |
286 | char* path = NULL; | |
287 | uint64_t bsize = 0; | |
288 | uint64_t fsize = 0; | |
289 | ||
290 | pmb_parse_path(pathspec, &path, &bsize, &fsize); | |
291 | if (NULL == path) | |
292 | return NULL; | |
293 | ||
294 | CACHE_LOCK(); | |
295 | ||
296 | pmb = fio_pmemblk_cache_lookup(path); | |
297 | ||
298 | if (NULL == pmb) { | |
299 | /* load libpmemblk if needed */ | |
300 | if (NULL == pmemblk_open) | |
301 | if (0 != load_libpmemblk(getenv("FIO_PMEMBLK_LIB"))) | |
302 | goto error; | |
303 | ||
304 | pmb = malloc(sizeof(*pmb)); | |
305 | if (NULL == pmb) | |
306 | goto error; | |
307 | ||
308 | /* try opening existing first, create it if needed */ | |
309 | pmb->pmb_pool = pmemblk_open(path, bsize); | |
310 | if ((NULL == pmb->pmb_pool) && | |
311 | (ENOENT == errno) && | |
312 | (flags & PMB_CREATE) && | |
313 | (0 < fsize) && | |
314 | (0 < bsize)) { | |
315 | pmb->pmb_pool = pmemblk_create(path, bsize, fsize, 0644); | |
316 | } | |
317 | if (NULL == pmb->pmb_pool) { | |
318 | log_err("fio: enable to open pmemblk pool file (errno %d)\n", | |
319 | errno); | |
320 | goto error; | |
321 | } | |
322 | ||
323 | pmb->pmb_filename = path; | |
324 | pmb->pmb_next = NULL; | |
325 | pmb->pmb_refcnt = 0; | |
326 | pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool); | |
327 | pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool); | |
328 | ||
329 | fio_pmemblk_cache_insert(pmb); | |
330 | } | |
331 | ||
332 | pmb->pmb_refcnt += 1; | |
333 | ||
334 | CACHE_UNLOCK(); | |
335 | ||
336 | return pmb; | |
337 | ||
338 | error: | |
339 | if (NULL != pmb) { | |
340 | if (NULL != pmb->pmb_pool) | |
341 | pmemblk_close(pmb->pmb_pool); | |
342 | pmb->pmb_pool = NULL; | |
343 | pmb->pmb_filename = NULL; | |
344 | free(pmb); | |
345 | } | |
346 | if (NULL != path) | |
347 | free(path); | |
348 | CACHE_UNLOCK(); | |
349 | return NULL; | |
350 | ||
351 | } /* pmb_open() */ | |
352 | ||
353 | ||
354 | static void | |
355 | pmb_close( | |
356 | fio_pmemblk_file_t pmb, | |
357 | const int keep | |
358 | ) | |
359 | { | |
360 | CACHE_LOCK(); | |
361 | ||
362 | pmb->pmb_refcnt--; | |
363 | ||
364 | if (!keep && (0 == pmb->pmb_refcnt)) { | |
365 | pmemblk_close(pmb->pmb_pool); | |
366 | pmb->pmb_pool = NULL; | |
367 | free(pmb->pmb_filename); | |
368 | pmb->pmb_filename = NULL; | |
369 | fio_pmemblk_cache_remove(pmb); | |
370 | free(pmb); | |
371 | } | |
372 | ||
373 | CACHE_UNLOCK(); | |
374 | ||
375 | } /* pmb_close() */ | |
376 | ||
377 | ||
378 | static int | |
379 | pmb_get_flags( | |
380 | struct thread_data* td, | |
381 | uint64_t* pflags | |
382 | ) | |
383 | { | |
384 | static int thread_warned = 0; | |
385 | static int odirect_warned = 0; | |
386 | ||
387 | uint64_t flags = 0; | |
388 | ||
389 | if (!td->o.use_thread) { | |
390 | if (!thread_warned) { | |
391 | thread_warned = 1; | |
392 | log_err("fio: must set thread=1 for pmemblk engine\n"); | |
393 | } | |
394 | return 1; | |
395 | } | |
396 | ||
397 | if (!td->o.odirect && !odirect_warned) { | |
398 | odirect_warned = 1; | |
399 | log_info("fio: direct == 0, but pmemblk is always direct\n"); | |
400 | } | |
401 | ||
402 | if (td->o.allow_create) | |
403 | flags |= PMB_CREATE; | |
404 | ||
405 | (*pflags) = flags; | |
406 | return 0; | |
407 | ||
408 | } /* pmb_get_flags() */ | |
409 | ||
410 | ||
411 | static int | |
412 | fio_pmemblk_open_file( | |
413 | struct thread_data* td, | |
414 | struct fio_file* f) | |
415 | { | |
416 | uint64_t flags = 0; | |
417 | fio_pmemblk_file_t pmb; | |
418 | ||
419 | if (0 != pmb_get_flags(td, &flags)) | |
420 | return 1; | |
421 | ||
422 | pmb = pmb_open(f->file_name, flags); | |
423 | if (NULL == pmb) | |
424 | return 1; | |
425 | ||
426 | FIOFILEPMBSET(f, pmb); | |
427 | ||
428 | return 0; | |
429 | ||
430 | } /* fio_pmemblk_open_file() */ | |
431 | ||
432 | ||
433 | static int | |
434 | fio_pmemblk_close_file( | |
435 | struct thread_data fio_unused* td, | |
436 | struct fio_file* f | |
437 | ) | |
438 | { | |
439 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
440 | ||
441 | if (pmb) | |
442 | pmb_close(pmb, 0); | |
443 | ||
444 | FIOFILEPMBSET(f, NULL); | |
445 | ||
446 | return 0; | |
447 | ||
448 | } /* fio_pmemblk_close_file() */ | |
449 | ||
450 | ||
451 | static int | |
452 | fio_pmemblk_get_file_size( | |
453 | struct thread_data* td, | |
454 | struct fio_file* f | |
455 | ) | |
456 | { | |
457 | uint64_t flags = 0; | |
458 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
459 | ||
460 | if (fio_file_size_known(f)) | |
461 | return 0; | |
462 | ||
463 | if (NULL == pmb) { | |
464 | if (0 != pmb_get_flags(td, &flags)) | |
465 | return 1; | |
466 | pmb = pmb_open(f->file_name, flags); | |
467 | if (NULL == pmb) | |
468 | return 1; | |
469 | } | |
470 | ||
471 | f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks; | |
472 | ||
473 | fio_file_set_size_known(f); | |
474 | ||
475 | if (NULL == FIOFILEPMBGET(f)) | |
476 | pmb_close(pmb, 1); | |
477 | ||
478 | return 0; | |
479 | ||
480 | } /* fio_pmemblk_get_file_size() */ | |
481 | ||
482 | ||
483 | static int | |
484 | fio_pmemblk_queue( | |
485 | struct thread_data* td, | |
486 | struct io_u* io_u) | |
487 | { | |
488 | struct fio_file* f = io_u->file; | |
489 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
490 | ||
491 | unsigned long long off; | |
492 | unsigned long len; | |
493 | void* buf; | |
494 | int (*blkop)(PMEMblkpool*, void*, off_t) = (void*)pmemblk_write; | |
495 | ||
496 | fio_ro_check(td, io_u); | |
497 | ||
498 | switch (io_u->ddir) { | |
499 | case DDIR_READ: | |
500 | blkop = pmemblk_read; | |
501 | /* fall through */ | |
502 | case DDIR_WRITE: | |
503 | off = io_u->offset; | |
504 | len = io_u->xfer_buflen; | |
505 | ||
506 | io_u->error = EINVAL; | |
507 | if (0 != (off % pmb->pmb_bsize)) | |
508 | break; | |
509 | if (0 != (len % pmb->pmb_bsize)) | |
510 | break; | |
511 | if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks) | |
512 | break; | |
513 | ||
514 | io_u->error = 0; | |
515 | buf = io_u->xfer_buf; | |
516 | off /= pmb->pmb_bsize; | |
517 | len /= pmb->pmb_bsize; | |
518 | while (0 < len) { | |
519 | if (0 != blkop(pmb->pmb_pool, buf, off)) { | |
520 | io_u->error = errno; | |
521 | break; | |
522 | } | |
523 | buf += pmb->pmb_bsize; | |
524 | off++; | |
525 | len--; | |
526 | } | |
527 | off *= pmb->pmb_bsize; | |
528 | len *= pmb->pmb_bsize; | |
529 | io_u->resid = io_u->xfer_buflen - (off - io_u->offset); | |
530 | break; | |
531 | case DDIR_SYNC: | |
532 | case DDIR_DATASYNC: | |
533 | case DDIR_SYNC_FILE_RANGE: | |
534 | /* we're always sync'd */ | |
535 | io_u->error = 0; | |
536 | break; | |
537 | default: | |
538 | io_u->error = EINVAL; | |
539 | break; | |
540 | } | |
541 | ||
542 | return FIO_Q_COMPLETED; | |
543 | ||
544 | } /* fio_pmemblk_queue() */ | |
545 | ||
546 | ||
547 | static int | |
548 | fio_pmemblk_unlink_file( | |
549 | struct thread_data* td, | |
550 | struct fio_file* f | |
551 | ) | |
552 | { | |
553 | char* path = NULL; | |
554 | uint64_t bsize = 0; | |
555 | uint64_t fsize = 0; | |
556 | ||
557 | /* | |
558 | * we need our own unlink in case the user has specified | |
559 | * the block and file sizes in the path name. we parse | |
560 | * the file_name to determine the file name we actually used. | |
561 | */ | |
562 | ||
563 | pmb_parse_path(f->file_name, &path, &bsize, &fsize); | |
564 | if (NULL == path) | |
565 | return 1; | |
566 | ||
567 | unlink(path); | |
568 | free(path); | |
569 | ||
570 | return 0; | |
571 | ||
572 | } /* fio_pmemblk_unlink_file() */ | |
573 | ||
574 | ||
575 | struct ioengine_ops ioengine = { | |
576 | .name = "pmemblk", | |
577 | .version = FIO_IOOPS_VERSION, | |
578 | .queue = fio_pmemblk_queue, | |
579 | .open_file = fio_pmemblk_open_file, | |
580 | .close_file = fio_pmemblk_close_file, | |
581 | .get_file_size = fio_pmemblk_get_file_size, | |
582 | .unlink_file = fio_pmemblk_unlink_file, | |
583 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL, | |
584 | }; | |
585 | ||
586 | ||
587 | static void | |
588 | fio_init fio_pmemblk_register(void) | |
589 | { | |
590 | register_ioengine(&ioengine); | |
591 | } | |
592 | ||
593 | ||
594 | static void | |
595 | fio_exit fio_pmemblk_unregister(void) | |
596 | { | |
597 | unregister_ioengine(&ioengine); | |
598 | } | |
599 |