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 | ||
43f3cec2 BB |
21 | /* |
22 | * pmemblk engine | |
23 | * | |
24 | * IO engine that uses libpmemblk to read and write data | |
25 | * | |
f1d85c6e BB |
26 | * To use: |
27 | * ioengine=pmemblk | |
43f3cec2 BB |
28 | * |
29 | * Other relevant settings: | |
30 | * iodepth=1 | |
31 | * direct=1 | |
32 | * thread=1 REQUIRED | |
33 | * unlink=1 | |
34 | * filename=/pmem0/fiotestfile,BSIZE,FSIZEMB | |
35 | * | |
36 | * thread must be set to 1 for pmemblk as multiple processes cannot | |
37 | * open the same block pool file. | |
38 | * | |
39 | * iodepth should be set to 1 as pmemblk is always synchronous. | |
40 | * Use numjobs to scale up. | |
41 | * | |
42 | * direct=1 is implied as pmemblk is always direct. | |
43 | * | |
44 | * Can set unlink to 1 to remove the block pool file after testing. | |
45 | * | |
46 | * When specifying the filename, if the block pool file does not already | |
47 | * exist, then the pmemblk engine can create the pool file if you specify | |
48 | * the block and file sizes. BSIZE is the block size in bytes. | |
49 | * FSIZEMB is the pool file size in MB. | |
50 | * | |
f1d85c6e BB |
51 | * See examples/pmemblk.fio for more. |
52 | * | |
43f3cec2 BB |
53 | * libpmemblk.so |
54 | * By default, the pmemblk engine will let the system find the libpmemblk.so | |
55 | * that it uses. You can use an alternative libpmemblk by setting the | |
56 | * FIO_PMEMBLK_LIB environment variable to the full path to the desired | |
57 | * libpmemblk.so. | |
58 | * | |
59 | */ | |
60 | ||
61 | #include <stdio.h> | |
62 | #include <stdlib.h> | |
63 | #include <unistd.h> | |
64 | #include <sys/uio.h> | |
65 | #include <errno.h> | |
66 | #include <assert.h> | |
67 | #include <dlfcn.h> | |
68 | #include <string.h> | |
69 | ||
70 | #include "../fio.h" | |
71 | ||
43f3cec2 BB |
72 | /* |
73 | * libpmemblk | |
74 | */ | |
75 | struct PMEMblkpool_s; | |
76 | typedef struct PMEMblkpool_s PMEMblkpool; | |
77 | ||
c086b314 JA |
78 | PMEMblkpool *(*pmemblk_create) (const char *, size_t, size_t, mode_t) = NULL; |
79 | PMEMblkpool *(*pmemblk_open) (const char *, size_t) = NULL; | |
80 | void (*pmemblk_close) (PMEMblkpool *) = NULL; | |
81 | size_t(*pmemblk_nblock) (PMEMblkpool *) = NULL; | |
82 | size_t(*pmemblk_bsize) (PMEMblkpool *) = NULL; | |
83 | int (*pmemblk_read) (PMEMblkpool *, void *, off_t) = NULL; | |
84 | int (*pmemblk_write) (PMEMblkpool *, const void *, off_t) = NULL; | |
85 | ||
86 | int load_libpmemblk(const char *path) | |
43f3cec2 | 87 | { |
c086b314 | 88 | void *dl; |
43f3cec2 BB |
89 | |
90 | if (NULL == path) | |
91 | path = "libpmemblk.so"; | |
92 | ||
93 | dl = dlopen(path, RTLD_NOW | RTLD_NODELETE); | |
94 | if (NULL == dl) | |
95 | goto errorout; | |
96 | ||
dfec8168 | 97 | if (NULL == (pmemblk_create = dlsym(dl, "pmemblk_create"))) |
43f3cec2 | 98 | goto errorout; |
dfec8168 | 99 | if (NULL == (pmemblk_open = dlsym(dl, "pmemblk_open"))) |
43f3cec2 | 100 | goto errorout; |
dfec8168 | 101 | if (NULL == (pmemblk_close = dlsym(dl, "pmemblk_close"))) |
43f3cec2 | 102 | goto errorout; |
dfec8168 | 103 | if (NULL == (pmemblk_nblock = dlsym(dl, "pmemblk_nblock"))) |
43f3cec2 | 104 | goto errorout; |
dfec8168 | 105 | if (NULL == (pmemblk_bsize = dlsym(dl, "pmemblk_bsize"))) |
43f3cec2 | 106 | goto errorout; |
dfec8168 | 107 | if (NULL == (pmemblk_read = dlsym(dl, "pmemblk_read"))) |
43f3cec2 | 108 | goto errorout; |
dfec8168 | 109 | if (NULL == (pmemblk_write = dlsym(dl, "pmemblk_write"))) |
43f3cec2 | 110 | goto errorout; |
c086b314 | 111 | |
43f3cec2 BB |
112 | return 0; |
113 | ||
114 | errorout: | |
115 | log_err("fio: unable to load libpmemblk: %s\n", dlerror()); | |
116 | if (NULL != dl) | |
117 | dlclose(dl); | |
118 | ||
119 | return (-1); | |
43f3cec2 | 120 | |
c086b314 | 121 | } /* load_libpmemblk() */ |
43f3cec2 | 122 | |
c086b314 | 123 | typedef struct fio_pmemblk_file *fio_pmemblk_file_t; |
43f3cec2 | 124 | struct fio_pmemblk_file { |
c086b314 JA |
125 | fio_pmemblk_file_t pmb_next; |
126 | char *pmb_filename; | |
127 | uint64_t pmb_refcnt; | |
128 | PMEMblkpool *pmb_pool; | |
129 | size_t pmb_bsize; | |
130 | size_t pmb_nblocks; | |
43f3cec2 BB |
131 | }; |
132 | #define FIOFILEPMBSET(_f, _v) do { \ | |
133 | (_f)->engine_data = (uint64_t)(uintptr_t)(_v); \ | |
134 | } while(0) | |
135 | #define FIOFILEPMBGET(_f) ((fio_pmemblk_file_t)((_f)->engine_data)) | |
136 | ||
c086b314 | 137 | static fio_pmemblk_file_t Cache = NULL; |
43f3cec2 | 138 | |
c086b314 | 139 | static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER; |
43f3cec2 | 140 | |
c086b314 | 141 | #define PMB_CREATE (0x0001) /* should create file */ |
43f3cec2 | 142 | |
c086b314 | 143 | fio_pmemblk_file_t fio_pmemblk_cache_lookup(const char *filename) |
43f3cec2 | 144 | { |
c086b314 | 145 | fio_pmemblk_file_t i; |
43f3cec2 BB |
146 | |
147 | for (i = Cache; i != NULL; i = i->pmb_next) | |
148 | if (0 == strcmp(filename, i->pmb_filename)) | |
149 | return i; | |
150 | ||
151 | return NULL; | |
152 | ||
c086b314 | 153 | } /* fio_pmemblk_cache_lookup() */ |
43f3cec2 | 154 | |
c086b314 | 155 | static void fio_pmemblk_cache_insert(fio_pmemblk_file_t pmb) |
43f3cec2 BB |
156 | { |
157 | pmb->pmb_next = Cache; | |
158 | Cache = pmb; | |
159 | ||
160 | return; | |
161 | ||
c086b314 | 162 | } /* fio_pmemblk_cache_insert() */ |
43f3cec2 | 163 | |
c086b314 | 164 | static void fio_pmemblk_cache_remove(fio_pmemblk_file_t pmb) |
43f3cec2 | 165 | { |
c086b314 | 166 | fio_pmemblk_file_t i; |
43f3cec2 BB |
167 | |
168 | if (pmb == Cache) { | |
169 | Cache = Cache->pmb_next; | |
170 | pmb->pmb_next = NULL; | |
171 | return; | |
172 | } | |
173 | ||
174 | for (i = Cache; i != NULL; i = i->pmb_next) | |
175 | if (pmb == i->pmb_next) { | |
176 | i->pmb_next = i->pmb_next->pmb_next; | |
177 | pmb->pmb_next = NULL; | |
178 | return; | |
179 | } | |
180 | ||
181 | return; | |
182 | ||
c086b314 | 183 | } /* fio_pmemblk_cache_remove() */ |
43f3cec2 BB |
184 | |
185 | /* | |
186 | * to control block size and gross file size at the libpmemblk | |
187 | * level, we allow the block size and file size to be appended | |
188 | * to the file name: | |
189 | * | |
190 | * path[,bsize,fsizemb] | |
191 | * | |
192 | * note that we do not use the fio option "filesize" to dictate | |
193 | * the file size because we can only give libpmemblk the gross | |
194 | * file size, which is different from the net or usable file | |
195 | * size (which is probably what fio wants). | |
196 | * | |
197 | * the final path without the parameters is returned in ppath. | |
198 | * the block size and file size are returned in pbsize and fsize. | |
199 | * | |
200 | * note that the user should specify the file size in MiB, but | |
201 | * we return bytes from here. | |
202 | */ | |
203 | static void | |
c086b314 JA |
204 | pmb_parse_path(const char *pathspec, |
205 | char **ppath, uint64_t * pbsize, uint64_t * pfsize) | |
43f3cec2 | 206 | { |
c086b314 JA |
207 | char *path; |
208 | char *s; | |
209 | uint64_t bsize; | |
210 | uint64_t fsizemb; | |
43f3cec2 BB |
211 | |
212 | path = strdup(pathspec); | |
213 | if (NULL == path) { | |
214 | *ppath = NULL; | |
215 | return; | |
216 | } | |
217 | ||
218 | /* extract sizes, if given */ | |
219 | s = strrchr(path, ','); | |
c086b314 | 220 | if (s && (fsizemb = strtoull(s + 1, NULL, 10))) { |
43f3cec2 BB |
221 | *s = 0; |
222 | s = strrchr(path, ','); | |
c086b314 | 223 | if (s && (bsize = strtoull(s + 1, NULL, 10))) { |
43f3cec2 | 224 | *s = 0; |
c086b314 | 225 | *ppath = path; |
43f3cec2 BB |
226 | *pbsize = bsize; |
227 | *pfsize = fsizemb << 20; | |
228 | return; | |
229 | } | |
230 | } | |
231 | ||
232 | /* size specs not found */ | |
233 | strcpy(path, pathspec); | |
c086b314 | 234 | *ppath = path; |
43f3cec2 BB |
235 | *pbsize = 0; |
236 | *pfsize = 0; | |
237 | return; | |
238 | ||
c086b314 | 239 | } /* pmb_parse_path() */ |
43f3cec2 BB |
240 | |
241 | static | |
c086b314 | 242 | fio_pmemblk_file_t pmb_open(const char *pathspec, int flags) |
43f3cec2 | 243 | { |
c086b314 JA |
244 | fio_pmemblk_file_t pmb; |
245 | char *path = NULL; | |
246 | uint64_t bsize = 0; | |
247 | uint64_t fsize = 0; | |
43f3cec2 BB |
248 | |
249 | pmb_parse_path(pathspec, &path, &bsize, &fsize); | |
250 | if (NULL == path) | |
251 | return NULL; | |
252 | ||
de88999f | 253 | pthread_mutex_lock(&CacheLock); |
43f3cec2 BB |
254 | |
255 | pmb = fio_pmemblk_cache_lookup(path); | |
256 | ||
257 | if (NULL == pmb) { | |
258 | /* load libpmemblk if needed */ | |
259 | if (NULL == pmemblk_open) | |
260 | if (0 != load_libpmemblk(getenv("FIO_PMEMBLK_LIB"))) | |
261 | goto error; | |
262 | ||
263 | pmb = malloc(sizeof(*pmb)); | |
264 | if (NULL == pmb) | |
265 | goto error; | |
266 | ||
267 | /* try opening existing first, create it if needed */ | |
268 | pmb->pmb_pool = pmemblk_open(path, bsize); | |
269 | if ((NULL == pmb->pmb_pool) && | |
270 | (ENOENT == errno) && | |
c086b314 JA |
271 | (flags & PMB_CREATE) && (0 < fsize) && (0 < bsize)) { |
272 | pmb->pmb_pool = | |
273 | pmemblk_create(path, bsize, fsize, 0644); | |
43f3cec2 BB |
274 | } |
275 | if (NULL == pmb->pmb_pool) { | |
c086b314 JA |
276 | log_err |
277 | ("fio: enable to open pmemblk pool file (errno %d)\n", | |
278 | errno); | |
43f3cec2 BB |
279 | goto error; |
280 | } | |
281 | ||
282 | pmb->pmb_filename = path; | |
c086b314 JA |
283 | pmb->pmb_next = NULL; |
284 | pmb->pmb_refcnt = 0; | |
285 | pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool); | |
286 | pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool); | |
43f3cec2 BB |
287 | |
288 | fio_pmemblk_cache_insert(pmb); | |
289 | } | |
290 | ||
291 | pmb->pmb_refcnt += 1; | |
292 | ||
de88999f | 293 | pthread_mutex_unlock(&CacheLock); |
c086b314 | 294 | |
43f3cec2 BB |
295 | return pmb; |
296 | ||
297 | error: | |
298 | if (NULL != pmb) { | |
299 | if (NULL != pmb->pmb_pool) | |
300 | pmemblk_close(pmb->pmb_pool); | |
c086b314 | 301 | pmb->pmb_pool = NULL; |
43f3cec2 BB |
302 | pmb->pmb_filename = NULL; |
303 | free(pmb); | |
304 | } | |
305 | if (NULL != path) | |
306 | free(path); | |
de88999f JA |
307 | |
308 | pthread_mutex_unlock(&CacheLock); | |
43f3cec2 BB |
309 | return NULL; |
310 | ||
c086b314 | 311 | } /* pmb_open() */ |
43f3cec2 | 312 | |
c086b314 | 313 | static void pmb_close(fio_pmemblk_file_t pmb, const int keep) |
43f3cec2 | 314 | { |
de88999f | 315 | pthread_mutex_lock(&CacheLock); |
43f3cec2 BB |
316 | |
317 | pmb->pmb_refcnt--; | |
318 | ||
319 | if (!keep && (0 == pmb->pmb_refcnt)) { | |
320 | pmemblk_close(pmb->pmb_pool); | |
321 | pmb->pmb_pool = NULL; | |
322 | free(pmb->pmb_filename); | |
323 | pmb->pmb_filename = NULL; | |
324 | fio_pmemblk_cache_remove(pmb); | |
325 | free(pmb); | |
326 | } | |
327 | ||
de88999f | 328 | pthread_mutex_unlock(&CacheLock); |
43f3cec2 | 329 | |
c086b314 | 330 | } /* pmb_close() */ |
43f3cec2 | 331 | |
c086b314 | 332 | static int pmb_get_flags(struct thread_data *td, uint64_t * pflags) |
43f3cec2 | 333 | { |
c086b314 | 334 | static int thread_warned = 0; |
43f3cec2 BB |
335 | static int odirect_warned = 0; |
336 | ||
c086b314 | 337 | uint64_t flags = 0; |
43f3cec2 BB |
338 | |
339 | if (!td->o.use_thread) { | |
340 | if (!thread_warned) { | |
341 | thread_warned = 1; | |
342 | log_err("fio: must set thread=1 for pmemblk engine\n"); | |
343 | } | |
344 | return 1; | |
345 | } | |
346 | ||
347 | if (!td->o.odirect && !odirect_warned) { | |
348 | odirect_warned = 1; | |
349 | log_info("fio: direct == 0, but pmemblk is always direct\n"); | |
350 | } | |
351 | ||
352 | if (td->o.allow_create) | |
353 | flags |= PMB_CREATE; | |
354 | ||
355 | (*pflags) = flags; | |
356 | return 0; | |
357 | ||
c086b314 | 358 | } /* pmb_get_flags() */ |
43f3cec2 | 359 | |
c086b314 | 360 | static int fio_pmemblk_open_file(struct thread_data *td, struct fio_file *f) |
43f3cec2 | 361 | { |
c086b314 JA |
362 | uint64_t flags = 0; |
363 | fio_pmemblk_file_t pmb; | |
43f3cec2 BB |
364 | |
365 | if (0 != pmb_get_flags(td, &flags)) | |
366 | return 1; | |
367 | ||
368 | pmb = pmb_open(f->file_name, flags); | |
369 | if (NULL == pmb) | |
370 | return 1; | |
371 | ||
372 | FIOFILEPMBSET(f, pmb); | |
373 | ||
374 | return 0; | |
375 | ||
c086b314 | 376 | } /* fio_pmemblk_open_file() */ |
43f3cec2 BB |
377 | |
378 | static int | |
c086b314 | 379 | fio_pmemblk_close_file(struct thread_data fio_unused * td, struct fio_file *f) |
43f3cec2 | 380 | { |
c086b314 | 381 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); |
43f3cec2 BB |
382 | |
383 | if (pmb) | |
384 | pmb_close(pmb, 0); | |
385 | ||
386 | FIOFILEPMBSET(f, NULL); | |
387 | ||
388 | return 0; | |
389 | ||
c086b314 | 390 | } /* fio_pmemblk_close_file() */ |
43f3cec2 | 391 | |
c086b314 | 392 | static int fio_pmemblk_get_file_size(struct thread_data *td, struct fio_file *f) |
43f3cec2 | 393 | { |
c086b314 JA |
394 | uint64_t flags = 0; |
395 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
43f3cec2 BB |
396 | |
397 | if (fio_file_size_known(f)) | |
398 | return 0; | |
399 | ||
400 | if (NULL == pmb) { | |
401 | if (0 != pmb_get_flags(td, &flags)) | |
402 | return 1; | |
403 | pmb = pmb_open(f->file_name, flags); | |
404 | if (NULL == pmb) | |
405 | return 1; | |
406 | } | |
407 | ||
408 | f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks; | |
409 | ||
410 | fio_file_set_size_known(f); | |
411 | ||
412 | if (NULL == FIOFILEPMBGET(f)) | |
413 | pmb_close(pmb, 1); | |
414 | ||
415 | return 0; | |
416 | ||
c086b314 | 417 | } /* fio_pmemblk_get_file_size() */ |
43f3cec2 | 418 | |
c086b314 | 419 | static int fio_pmemblk_queue(struct thread_data *td, struct io_u *io_u) |
43f3cec2 | 420 | { |
c086b314 JA |
421 | struct fio_file *f = io_u->file; |
422 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
43f3cec2 | 423 | |
c086b314 JA |
424 | unsigned long long off; |
425 | unsigned long len; | |
426 | void *buf; | |
427 | int (*blkop) (PMEMblkpool *, void *, off_t) = (void *)pmemblk_write; | |
43f3cec2 BB |
428 | |
429 | fio_ro_check(td, io_u); | |
430 | ||
431 | switch (io_u->ddir) { | |
432 | case DDIR_READ: | |
433 | blkop = pmemblk_read; | |
434 | /* fall through */ | |
435 | case DDIR_WRITE: | |
436 | off = io_u->offset; | |
437 | len = io_u->xfer_buflen; | |
438 | ||
439 | io_u->error = EINVAL; | |
440 | if (0 != (off % pmb->pmb_bsize)) | |
441 | break; | |
442 | if (0 != (len % pmb->pmb_bsize)) | |
443 | break; | |
444 | if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks) | |
445 | break; | |
446 | ||
447 | io_u->error = 0; | |
448 | buf = io_u->xfer_buf; | |
449 | off /= pmb->pmb_bsize; | |
450 | len /= pmb->pmb_bsize; | |
451 | while (0 < len) { | |
452 | if (0 != blkop(pmb->pmb_pool, buf, off)) { | |
453 | io_u->error = errno; | |
454 | break; | |
455 | } | |
456 | buf += pmb->pmb_bsize; | |
457 | off++; | |
458 | len--; | |
459 | } | |
460 | off *= pmb->pmb_bsize; | |
461 | len *= pmb->pmb_bsize; | |
462 | io_u->resid = io_u->xfer_buflen - (off - io_u->offset); | |
463 | break; | |
464 | case DDIR_SYNC: | |
465 | case DDIR_DATASYNC: | |
466 | case DDIR_SYNC_FILE_RANGE: | |
467 | /* we're always sync'd */ | |
468 | io_u->error = 0; | |
469 | break; | |
470 | default: | |
471 | io_u->error = EINVAL; | |
472 | break; | |
473 | } | |
474 | ||
475 | return FIO_Q_COMPLETED; | |
476 | ||
c086b314 | 477 | } /* fio_pmemblk_queue() */ |
43f3cec2 | 478 | |
c086b314 | 479 | static int fio_pmemblk_unlink_file(struct thread_data *td, struct fio_file *f) |
43f3cec2 | 480 | { |
c086b314 JA |
481 | char *path = NULL; |
482 | uint64_t bsize = 0; | |
483 | uint64_t fsize = 0; | |
43f3cec2 BB |
484 | |
485 | /* | |
486 | * we need our own unlink in case the user has specified | |
487 | * the block and file sizes in the path name. we parse | |
488 | * the file_name to determine the file name we actually used. | |
489 | */ | |
490 | ||
491 | pmb_parse_path(f->file_name, &path, &bsize, &fsize); | |
492 | if (NULL == path) | |
493 | return 1; | |
494 | ||
495 | unlink(path); | |
496 | free(path); | |
497 | ||
498 | return 0; | |
499 | ||
c086b314 | 500 | } /* fio_pmemblk_unlink_file() */ |
43f3cec2 BB |
501 | |
502 | struct ioengine_ops ioengine = { | |
c086b314 JA |
503 | .name = "pmemblk", |
504 | .version = FIO_IOOPS_VERSION, | |
505 | .queue = fio_pmemblk_queue, | |
506 | .open_file = fio_pmemblk_open_file, | |
507 | .close_file = fio_pmemblk_close_file, | |
508 | .get_file_size = fio_pmemblk_get_file_size, | |
509 | .unlink_file = fio_pmemblk_unlink_file, | |
510 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL, | |
43f3cec2 BB |
511 | }; |
512 | ||
43f3cec2 BB |
513 | static void |
514 | fio_init fio_pmemblk_register(void) | |
515 | { | |
516 | register_ioengine(&ioengine); | |
517 | } | |
518 | ||
43f3cec2 BB |
519 | static void |
520 | fio_exit fio_pmemblk_unregister(void) | |
521 | { | |
522 | unregister_ioengine(&ioengine); | |
523 | } |