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 | * | |
27 | * To build as an external engine: | |
28 | * Copy pmemblk.c to the engines/ directory of the fio source code, then: | |
29 | * cd .../engines/ | |
30 | * gcc -Wall -shared -rdynamic -fPIC -o pmemblk.o -D_GNU_SOURCE pmemblk.c | |
31 | * | |
32 | * To use as an external engine: | |
33 | * ioengine=/full/path/to/pmemblk.o | |
34 | * | |
35 | * Other relevant settings: | |
36 | * iodepth=1 | |
37 | * direct=1 | |
38 | * thread=1 REQUIRED | |
39 | * unlink=1 | |
40 | * filename=/pmem0/fiotestfile,BSIZE,FSIZEMB | |
41 | * | |
42 | * thread must be set to 1 for pmemblk as multiple processes cannot | |
43 | * open the same block pool file. | |
44 | * | |
45 | * iodepth should be set to 1 as pmemblk is always synchronous. | |
46 | * Use numjobs to scale up. | |
47 | * | |
48 | * direct=1 is implied as pmemblk is always direct. | |
49 | * | |
50 | * Can set unlink to 1 to remove the block pool file after testing. | |
51 | * | |
52 | * When specifying the filename, if the block pool file does not already | |
53 | * exist, then the pmemblk engine can create the pool file if you specify | |
54 | * the block and file sizes. BSIZE is the block size in bytes. | |
55 | * FSIZEMB is the pool file size in MB. | |
56 | * | |
57 | * libpmemblk.so | |
58 | * By default, the pmemblk engine will let the system find the libpmemblk.so | |
59 | * that it uses. You can use an alternative libpmemblk by setting the | |
60 | * FIO_PMEMBLK_LIB environment variable to the full path to the desired | |
61 | * libpmemblk.so. | |
62 | * | |
63 | */ | |
64 | ||
65 | #include <stdio.h> | |
66 | #include <stdlib.h> | |
67 | #include <unistd.h> | |
68 | #include <sys/uio.h> | |
69 | #include <errno.h> | |
70 | #include <assert.h> | |
71 | #include <dlfcn.h> | |
72 | #include <string.h> | |
73 | ||
74 | #include "../fio.h" | |
75 | ||
76 | ||
77 | ||
78 | /* | |
79 | * libpmemblk | |
80 | */ | |
81 | struct PMEMblkpool_s; | |
82 | typedef struct PMEMblkpool_s PMEMblkpool; | |
83 | ||
84 | PMEMblkpool* (*pmemblk_create_ptr)(const char*, size_t, size_t, mode_t) = NULL; | |
85 | PMEMblkpool* (*pmemblk_open_ptr)(const char*, size_t) = NULL; | |
86 | void (*pmemblk_close_ptr)(PMEMblkpool*) = NULL; | |
87 | size_t (*pmemblk_nblock_ptr)(PMEMblkpool*) = NULL; | |
88 | size_t (*pmemblk_bsize_ptr)(PMEMblkpool*) = NULL; | |
89 | int (*pmemblk_read_ptr)(PMEMblkpool*, void*, off_t) = NULL; | |
90 | int (*pmemblk_write_ptr)(PMEMblkpool*, const void*, off_t) = NULL; | |
91 | ||
92 | int | |
93 | load_libpmemblk( | |
94 | const char* path | |
95 | ) | |
96 | { | |
97 | void* dl; | |
98 | ||
99 | if (NULL == path) | |
100 | path = "libpmemblk.so"; | |
101 | ||
102 | dl = dlopen(path, RTLD_NOW | RTLD_NODELETE); | |
103 | if (NULL == dl) | |
104 | goto errorout; | |
105 | ||
106 | if (NULL == (pmemblk_create_ptr = dlsym(dl, "pmemblk_create"))) | |
107 | goto errorout; | |
108 | if (NULL == (pmemblk_open_ptr = dlsym(dl, "pmemblk_open"))) | |
109 | goto errorout; | |
110 | if (NULL == (pmemblk_close_ptr = dlsym(dl, "pmemblk_close"))) | |
111 | goto errorout; | |
112 | if (NULL == (pmemblk_nblock_ptr = dlsym(dl, "pmemblk_nblock"))) | |
113 | goto errorout; | |
114 | if (NULL == (pmemblk_bsize_ptr = dlsym(dl, "pmemblk_bsize"))) | |
115 | goto errorout; | |
116 | if (NULL == (pmemblk_read_ptr = dlsym(dl, "pmemblk_read"))) | |
117 | goto errorout; | |
118 | if (NULL == (pmemblk_write_ptr = dlsym(dl, "pmemblk_write"))) | |
119 | goto errorout; | |
120 | ||
121 | return 0; | |
122 | ||
123 | errorout: | |
124 | log_err("fio: unable to load libpmemblk: %s\n", dlerror()); | |
125 | if (NULL != dl) | |
126 | dlclose(dl); | |
127 | ||
128 | return (-1); | |
129 | ||
130 | } /* load_libpmemblk() */ | |
131 | ||
132 | #define pmemblk_create pmemblk_create_ptr | |
133 | #define pmemblk_open pmemblk_open_ptr | |
134 | #define pmemblk_close pmemblk_close_ptr | |
135 | #define pmemblk_nblock pmemblk_nblock_ptr | |
136 | #define pmemblk_bsize pmemblk_bsize_ptr | |
137 | #define pmemblk_read pmemblk_read_ptr | |
138 | #define pmemblk_write pmemblk_write_ptr | |
139 | ||
140 | ||
141 | typedef struct fio_pmemblk_file* fio_pmemblk_file_t; | |
142 | struct fio_pmemblk_file { | |
143 | fio_pmemblk_file_t pmb_next; | |
144 | char* pmb_filename; | |
145 | uint64_t pmb_refcnt; | |
146 | PMEMblkpool* pmb_pool; | |
147 | size_t pmb_bsize; | |
148 | size_t pmb_nblocks; | |
149 | }; | |
150 | #define FIOFILEPMBSET(_f, _v) do { \ | |
151 | (_f)->engine_data = (uint64_t)(uintptr_t)(_v); \ | |
152 | } while(0) | |
153 | #define FIOFILEPMBGET(_f) ((fio_pmemblk_file_t)((_f)->engine_data)) | |
154 | ||
155 | static fio_pmemblk_file_t Cache = NULL; | |
156 | ||
157 | static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER; | |
158 | #define CACHE_LOCK() \ | |
159 | (void)pthread_mutex_lock(&CacheLock) | |
160 | #define CACHE_UNLOCK() \ | |
161 | (void)pthread_mutex_unlock(&CacheLock) | |
162 | ||
163 | #define PMB_CREATE (0x0001) /* should create file */ | |
164 | ||
165 | ||
166 | fio_pmemblk_file_t | |
167 | fio_pmemblk_cache_lookup( | |
168 | const char* filename | |
169 | ) | |
170 | { | |
171 | fio_pmemblk_file_t i; | |
172 | ||
173 | for (i = Cache; i != NULL; i = i->pmb_next) | |
174 | if (0 == strcmp(filename, i->pmb_filename)) | |
175 | return i; | |
176 | ||
177 | return NULL; | |
178 | ||
179 | } /* fio_pmemblk_cache_lookup() */ | |
180 | ||
181 | ||
182 | static void | |
183 | fio_pmemblk_cache_insert( | |
184 | fio_pmemblk_file_t pmb | |
185 | ) | |
186 | { | |
187 | pmb->pmb_next = Cache; | |
188 | Cache = pmb; | |
189 | ||
190 | return; | |
191 | ||
192 | } /* fio_pmemblk_cache_insert() */ | |
193 | ||
194 | ||
195 | static void | |
196 | fio_pmemblk_cache_remove( | |
197 | fio_pmemblk_file_t pmb | |
198 | ) | |
199 | { | |
200 | fio_pmemblk_file_t i; | |
201 | ||
202 | if (pmb == Cache) { | |
203 | Cache = Cache->pmb_next; | |
204 | pmb->pmb_next = NULL; | |
205 | return; | |
206 | } | |
207 | ||
208 | for (i = Cache; i != NULL; i = i->pmb_next) | |
209 | if (pmb == i->pmb_next) { | |
210 | i->pmb_next = i->pmb_next->pmb_next; | |
211 | pmb->pmb_next = NULL; | |
212 | return; | |
213 | } | |
214 | ||
215 | return; | |
216 | ||
217 | } /* fio_pmemblk_cache_remove() */ | |
218 | ||
219 | ||
220 | /* | |
221 | * to control block size and gross file size at the libpmemblk | |
222 | * level, we allow the block size and file size to be appended | |
223 | * to the file name: | |
224 | * | |
225 | * path[,bsize,fsizemb] | |
226 | * | |
227 | * note that we do not use the fio option "filesize" to dictate | |
228 | * the file size because we can only give libpmemblk the gross | |
229 | * file size, which is different from the net or usable file | |
230 | * size (which is probably what fio wants). | |
231 | * | |
232 | * the final path without the parameters is returned in ppath. | |
233 | * the block size and file size are returned in pbsize and fsize. | |
234 | * | |
235 | * note that the user should specify the file size in MiB, but | |
236 | * we return bytes from here. | |
237 | */ | |
238 | static void | |
239 | pmb_parse_path( | |
240 | const char* pathspec, | |
241 | char** ppath, | |
242 | uint64_t* pbsize, | |
243 | uint64_t* pfsize | |
244 | ) | |
245 | { | |
246 | char* path; | |
247 | char* s; | |
248 | uint64_t bsize; | |
249 | uint64_t fsizemb; | |
250 | ||
251 | path = strdup(pathspec); | |
252 | if (NULL == path) { | |
253 | *ppath = NULL; | |
254 | return; | |
255 | } | |
256 | ||
257 | /* extract sizes, if given */ | |
258 | s = strrchr(path, ','); | |
259 | if (s && (fsizemb = strtoull(s+1, NULL, 10))) { | |
260 | *s = 0; | |
261 | s = strrchr(path, ','); | |
262 | if (s && (bsize = strtoull(s+1, NULL, 10))) { | |
263 | *s = 0; | |
264 | *ppath = path; | |
265 | *pbsize = bsize; | |
266 | *pfsize = fsizemb << 20; | |
267 | return; | |
268 | } | |
269 | } | |
270 | ||
271 | /* size specs not found */ | |
272 | strcpy(path, pathspec); | |
273 | *ppath = path; | |
274 | *pbsize = 0; | |
275 | *pfsize = 0; | |
276 | return; | |
277 | ||
278 | } /* pmb_parse_path() */ | |
279 | ||
280 | ||
281 | static | |
282 | fio_pmemblk_file_t | |
283 | pmb_open( | |
284 | const char* pathspec, | |
285 | int flags | |
286 | ) | |
287 | { | |
288 | fio_pmemblk_file_t pmb; | |
289 | char* path = NULL; | |
290 | uint64_t bsize = 0; | |
291 | uint64_t fsize = 0; | |
292 | ||
293 | pmb_parse_path(pathspec, &path, &bsize, &fsize); | |
294 | if (NULL == path) | |
295 | return NULL; | |
296 | ||
297 | CACHE_LOCK(); | |
298 | ||
299 | pmb = fio_pmemblk_cache_lookup(path); | |
300 | ||
301 | if (NULL == pmb) { | |
302 | /* load libpmemblk if needed */ | |
303 | if (NULL == pmemblk_open) | |
304 | if (0 != load_libpmemblk(getenv("FIO_PMEMBLK_LIB"))) | |
305 | goto error; | |
306 | ||
307 | pmb = malloc(sizeof(*pmb)); | |
308 | if (NULL == pmb) | |
309 | goto error; | |
310 | ||
311 | /* try opening existing first, create it if needed */ | |
312 | pmb->pmb_pool = pmemblk_open(path, bsize); | |
313 | if ((NULL == pmb->pmb_pool) && | |
314 | (ENOENT == errno) && | |
315 | (flags & PMB_CREATE) && | |
316 | (0 < fsize) && | |
317 | (0 < bsize)) { | |
318 | pmb->pmb_pool = pmemblk_create(path, bsize, fsize, 0644); | |
319 | } | |
320 | if (NULL == pmb->pmb_pool) { | |
321 | log_err("fio: enable to open pmemblk pool file (errno %d)\n", | |
322 | errno); | |
323 | goto error; | |
324 | } | |
325 | ||
326 | pmb->pmb_filename = path; | |
327 | pmb->pmb_next = NULL; | |
328 | pmb->pmb_refcnt = 0; | |
329 | pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool); | |
330 | pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool); | |
331 | ||
332 | fio_pmemblk_cache_insert(pmb); | |
333 | } | |
334 | ||
335 | pmb->pmb_refcnt += 1; | |
336 | ||
337 | CACHE_UNLOCK(); | |
338 | ||
339 | return pmb; | |
340 | ||
341 | error: | |
342 | if (NULL != pmb) { | |
343 | if (NULL != pmb->pmb_pool) | |
344 | pmemblk_close(pmb->pmb_pool); | |
345 | pmb->pmb_pool = NULL; | |
346 | pmb->pmb_filename = NULL; | |
347 | free(pmb); | |
348 | } | |
349 | if (NULL != path) | |
350 | free(path); | |
351 | CACHE_UNLOCK(); | |
352 | return NULL; | |
353 | ||
354 | } /* pmb_open() */ | |
355 | ||
356 | ||
357 | static void | |
358 | pmb_close( | |
359 | fio_pmemblk_file_t pmb, | |
360 | const int keep | |
361 | ) | |
362 | { | |
363 | CACHE_LOCK(); | |
364 | ||
365 | pmb->pmb_refcnt--; | |
366 | ||
367 | if (!keep && (0 == pmb->pmb_refcnt)) { | |
368 | pmemblk_close(pmb->pmb_pool); | |
369 | pmb->pmb_pool = NULL; | |
370 | free(pmb->pmb_filename); | |
371 | pmb->pmb_filename = NULL; | |
372 | fio_pmemblk_cache_remove(pmb); | |
373 | free(pmb); | |
374 | } | |
375 | ||
376 | CACHE_UNLOCK(); | |
377 | ||
378 | } /* pmb_close() */ | |
379 | ||
380 | ||
381 | static int | |
382 | pmb_get_flags( | |
383 | struct thread_data* td, | |
384 | uint64_t* pflags | |
385 | ) | |
386 | { | |
387 | static int thread_warned = 0; | |
388 | static int odirect_warned = 0; | |
389 | ||
390 | uint64_t flags = 0; | |
391 | ||
392 | if (!td->o.use_thread) { | |
393 | if (!thread_warned) { | |
394 | thread_warned = 1; | |
395 | log_err("fio: must set thread=1 for pmemblk engine\n"); | |
396 | } | |
397 | return 1; | |
398 | } | |
399 | ||
400 | if (!td->o.odirect && !odirect_warned) { | |
401 | odirect_warned = 1; | |
402 | log_info("fio: direct == 0, but pmemblk is always direct\n"); | |
403 | } | |
404 | ||
405 | if (td->o.allow_create) | |
406 | flags |= PMB_CREATE; | |
407 | ||
408 | (*pflags) = flags; | |
409 | return 0; | |
410 | ||
411 | } /* pmb_get_flags() */ | |
412 | ||
413 | ||
414 | static int | |
415 | fio_pmemblk_open_file( | |
416 | struct thread_data* td, | |
417 | struct fio_file* f) | |
418 | { | |
419 | uint64_t flags = 0; | |
420 | fio_pmemblk_file_t pmb; | |
421 | ||
422 | if (0 != pmb_get_flags(td, &flags)) | |
423 | return 1; | |
424 | ||
425 | pmb = pmb_open(f->file_name, flags); | |
426 | if (NULL == pmb) | |
427 | return 1; | |
428 | ||
429 | FIOFILEPMBSET(f, pmb); | |
430 | ||
431 | return 0; | |
432 | ||
433 | } /* fio_pmemblk_open_file() */ | |
434 | ||
435 | ||
436 | static int | |
437 | fio_pmemblk_close_file( | |
438 | struct thread_data fio_unused* td, | |
439 | struct fio_file* f | |
440 | ) | |
441 | { | |
442 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
443 | ||
444 | if (pmb) | |
445 | pmb_close(pmb, 0); | |
446 | ||
447 | FIOFILEPMBSET(f, NULL); | |
448 | ||
449 | return 0; | |
450 | ||
451 | } /* fio_pmemblk_close_file() */ | |
452 | ||
453 | ||
454 | static int | |
455 | fio_pmemblk_get_file_size( | |
456 | struct thread_data* td, | |
457 | struct fio_file* f | |
458 | ) | |
459 | { | |
460 | uint64_t flags = 0; | |
461 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
462 | ||
463 | if (fio_file_size_known(f)) | |
464 | return 0; | |
465 | ||
466 | if (NULL == pmb) { | |
467 | if (0 != pmb_get_flags(td, &flags)) | |
468 | return 1; | |
469 | pmb = pmb_open(f->file_name, flags); | |
470 | if (NULL == pmb) | |
471 | return 1; | |
472 | } | |
473 | ||
474 | f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks; | |
475 | ||
476 | fio_file_set_size_known(f); | |
477 | ||
478 | if (NULL == FIOFILEPMBGET(f)) | |
479 | pmb_close(pmb, 1); | |
480 | ||
481 | return 0; | |
482 | ||
483 | } /* fio_pmemblk_get_file_size() */ | |
484 | ||
485 | ||
486 | static int | |
487 | fio_pmemblk_queue( | |
488 | struct thread_data* td, | |
489 | struct io_u* io_u) | |
490 | { | |
491 | struct fio_file* f = io_u->file; | |
492 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
493 | ||
494 | unsigned long long off; | |
495 | unsigned long len; | |
496 | void* buf; | |
497 | int (*blkop)(PMEMblkpool*, void*, off_t) = (void*)pmemblk_write; | |
498 | ||
499 | fio_ro_check(td, io_u); | |
500 | ||
501 | switch (io_u->ddir) { | |
502 | case DDIR_READ: | |
503 | blkop = pmemblk_read; | |
504 | /* fall through */ | |
505 | case DDIR_WRITE: | |
506 | off = io_u->offset; | |
507 | len = io_u->xfer_buflen; | |
508 | ||
509 | io_u->error = EINVAL; | |
510 | if (0 != (off % pmb->pmb_bsize)) | |
511 | break; | |
512 | if (0 != (len % pmb->pmb_bsize)) | |
513 | break; | |
514 | if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks) | |
515 | break; | |
516 | ||
517 | io_u->error = 0; | |
518 | buf = io_u->xfer_buf; | |
519 | off /= pmb->pmb_bsize; | |
520 | len /= pmb->pmb_bsize; | |
521 | while (0 < len) { | |
522 | if (0 != blkop(pmb->pmb_pool, buf, off)) { | |
523 | io_u->error = errno; | |
524 | break; | |
525 | } | |
526 | buf += pmb->pmb_bsize; | |
527 | off++; | |
528 | len--; | |
529 | } | |
530 | off *= pmb->pmb_bsize; | |
531 | len *= pmb->pmb_bsize; | |
532 | io_u->resid = io_u->xfer_buflen - (off - io_u->offset); | |
533 | break; | |
534 | case DDIR_SYNC: | |
535 | case DDIR_DATASYNC: | |
536 | case DDIR_SYNC_FILE_RANGE: | |
537 | /* we're always sync'd */ | |
538 | io_u->error = 0; | |
539 | break; | |
540 | default: | |
541 | io_u->error = EINVAL; | |
542 | break; | |
543 | } | |
544 | ||
545 | return FIO_Q_COMPLETED; | |
546 | ||
547 | } /* fio_pmemblk_queue() */ | |
548 | ||
549 | ||
550 | static int | |
551 | fio_pmemblk_unlink_file( | |
552 | struct thread_data* td, | |
553 | struct fio_file* f | |
554 | ) | |
555 | { | |
556 | char* path = NULL; | |
557 | uint64_t bsize = 0; | |
558 | uint64_t fsize = 0; | |
559 | ||
560 | /* | |
561 | * we need our own unlink in case the user has specified | |
562 | * the block and file sizes in the path name. we parse | |
563 | * the file_name to determine the file name we actually used. | |
564 | */ | |
565 | ||
566 | pmb_parse_path(f->file_name, &path, &bsize, &fsize); | |
567 | if (NULL == path) | |
568 | return 1; | |
569 | ||
570 | unlink(path); | |
571 | free(path); | |
572 | ||
573 | return 0; | |
574 | ||
575 | } /* fio_pmemblk_unlink_file() */ | |
576 | ||
577 | ||
578 | struct ioengine_ops ioengine = { | |
579 | .name = "pmemblk", | |
580 | .version = FIO_IOOPS_VERSION, | |
581 | .queue = fio_pmemblk_queue, | |
582 | .open_file = fio_pmemblk_open_file, | |
583 | .close_file = fio_pmemblk_close_file, | |
584 | .get_file_size = fio_pmemblk_get_file_size, | |
585 | .unlink_file = fio_pmemblk_unlink_file, | |
586 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL, | |
587 | }; | |
588 | ||
589 | ||
590 | static void | |
591 | fio_init fio_pmemblk_register(void) | |
592 | { | |
593 | register_ioengine(&ioengine); | |
594 | } | |
595 | ||
596 | ||
597 | static void | |
598 | fio_exit fio_pmemblk_unregister(void) | |
599 | { | |
600 | unregister_ioengine(&ioengine); | |
601 | } | |
602 |