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: | |
0cbb3f53 | 30 | * thread=1 REQUIRED |
43f3cec2 BB |
31 | * iodepth=1 |
32 | * direct=1 | |
43f3cec2 | 33 | * unlink=1 |
0cbb3f53 | 34 | * filename=/mnt/pmem0/fiotestfile,BSIZE,FSIZEMiB |
43f3cec2 BB |
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 | * | |
0cbb3f53 RE |
42 | * direct=1 is implied as pmemblk is always direct. A warning message |
43 | * is printed if this is not specified. | |
44 | * | |
45 | * unlink=1 removes the block pool file after testing, and is optional. | |
46 | * | |
47 | * The pmem device must have a DAX-capable filesystem and be mounted | |
48 | * with DAX enabled. filename must point to a file on that filesystem. | |
43f3cec2 | 49 | * |
0cbb3f53 RE |
50 | * Example: |
51 | * mkfs.xfs /dev/pmem0 | |
52 | * mkdir /mnt/pmem0 | |
53 | * mount -o dax /dev/pmem0 /mnt/pmem0 | |
43f3cec2 BB |
54 | * |
55 | * When specifying the filename, if the block pool file does not already | |
0cbb3f53 | 56 | * exist, then the pmemblk engine creates the pool file if you specify |
43f3cec2 | 57 | * the block and file sizes. BSIZE is the block size in bytes. |
0cbb3f53 | 58 | * FSIZEMB is the pool file size in MiB. |
43f3cec2 | 59 | * |
f1d85c6e BB |
60 | * See examples/pmemblk.fio for more. |
61 | * | |
43f3cec2 BB |
62 | */ |
63 | ||
64 | #include <stdio.h> | |
65 | #include <stdlib.h> | |
66 | #include <unistd.h> | |
67 | #include <sys/uio.h> | |
68 | #include <errno.h> | |
69 | #include <assert.h> | |
43f3cec2 | 70 | #include <string.h> |
cf8775b8 RE |
71 | #include <libpmem.h> |
72 | #include <libpmemblk.h> | |
43f3cec2 BB |
73 | |
74 | #include "../fio.h" | |
75 | ||
43f3cec2 BB |
76 | /* |
77 | * libpmemblk | |
78 | */ | |
c086b314 | 79 | typedef struct fio_pmemblk_file *fio_pmemblk_file_t; |
630431cc | 80 | |
43f3cec2 | 81 | struct fio_pmemblk_file { |
c086b314 JA |
82 | fio_pmemblk_file_t pmb_next; |
83 | char *pmb_filename; | |
84 | uint64_t pmb_refcnt; | |
85 | PMEMblkpool *pmb_pool; | |
86 | size_t pmb_bsize; | |
87 | size_t pmb_nblocks; | |
43f3cec2 BB |
88 | }; |
89 | #define FIOFILEPMBSET(_f, _v) do { \ | |
90 | (_f)->engine_data = (uint64_t)(uintptr_t)(_v); \ | |
91 | } while(0) | |
92 | #define FIOFILEPMBGET(_f) ((fio_pmemblk_file_t)((_f)->engine_data)) | |
93 | ||
630431cc | 94 | static fio_pmemblk_file_t Cache; |
43f3cec2 | 95 | |
c086b314 | 96 | static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER; |
43f3cec2 | 97 | |
c086b314 | 98 | #define PMB_CREATE (0x0001) /* should create file */ |
43f3cec2 | 99 | |
c086b314 | 100 | fio_pmemblk_file_t fio_pmemblk_cache_lookup(const char *filename) |
43f3cec2 | 101 | { |
c086b314 | 102 | fio_pmemblk_file_t i; |
43f3cec2 BB |
103 | |
104 | for (i = Cache; i != NULL; i = i->pmb_next) | |
630431cc | 105 | if (!strcmp(filename, i->pmb_filename)) |
43f3cec2 BB |
106 | return i; |
107 | ||
108 | return NULL; | |
630431cc | 109 | } |
43f3cec2 | 110 | |
c086b314 | 111 | static void fio_pmemblk_cache_insert(fio_pmemblk_file_t pmb) |
43f3cec2 BB |
112 | { |
113 | pmb->pmb_next = Cache; | |
114 | Cache = pmb; | |
630431cc | 115 | } |
43f3cec2 | 116 | |
c086b314 | 117 | static void fio_pmemblk_cache_remove(fio_pmemblk_file_t pmb) |
43f3cec2 | 118 | { |
c086b314 | 119 | fio_pmemblk_file_t i; |
43f3cec2 BB |
120 | |
121 | if (pmb == Cache) { | |
122 | Cache = Cache->pmb_next; | |
123 | pmb->pmb_next = NULL; | |
124 | return; | |
125 | } | |
126 | ||
127 | for (i = Cache; i != NULL; i = i->pmb_next) | |
128 | if (pmb == i->pmb_next) { | |
129 | i->pmb_next = i->pmb_next->pmb_next; | |
130 | pmb->pmb_next = NULL; | |
131 | return; | |
132 | } | |
630431cc | 133 | } |
43f3cec2 BB |
134 | |
135 | /* | |
136 | * to control block size and gross file size at the libpmemblk | |
137 | * level, we allow the block size and file size to be appended | |
138 | * to the file name: | |
139 | * | |
0cbb3f53 | 140 | * path[,bsize,fsizemib] |
43f3cec2 BB |
141 | * |
142 | * note that we do not use the fio option "filesize" to dictate | |
143 | * the file size because we can only give libpmemblk the gross | |
144 | * file size, which is different from the net or usable file | |
145 | * size (which is probably what fio wants). | |
146 | * | |
147 | * the final path without the parameters is returned in ppath. | |
148 | * the block size and file size are returned in pbsize and fsize. | |
149 | * | |
0cbb3f53 | 150 | * note that the user specifies the file size in MiB, but |
43f3cec2 BB |
151 | * we return bytes from here. |
152 | */ | |
630431cc JA |
153 | static void pmb_parse_path(const char *pathspec, char **ppath, uint64_t *pbsize, |
154 | uint64_t *pfsize) | |
43f3cec2 | 155 | { |
c086b314 JA |
156 | char *path; |
157 | char *s; | |
158 | uint64_t bsize; | |
0cbb3f53 | 159 | uint64_t fsizemib; |
43f3cec2 BB |
160 | |
161 | path = strdup(pathspec); | |
630431cc | 162 | if (!path) { |
43f3cec2 BB |
163 | *ppath = NULL; |
164 | return; | |
165 | } | |
166 | ||
167 | /* extract sizes, if given */ | |
168 | s = strrchr(path, ','); | |
0cbb3f53 | 169 | if (s && (fsizemib = strtoull(s + 1, NULL, 10))) { |
43f3cec2 BB |
170 | *s = 0; |
171 | s = strrchr(path, ','); | |
c086b314 | 172 | if (s && (bsize = strtoull(s + 1, NULL, 10))) { |
43f3cec2 | 173 | *s = 0; |
c086b314 | 174 | *ppath = path; |
43f3cec2 | 175 | *pbsize = bsize; |
0cbb3f53 | 176 | *pfsize = fsizemib << 20; |
43f3cec2 BB |
177 | return; |
178 | } | |
179 | } | |
180 | ||
181 | /* size specs not found */ | |
182 | strcpy(path, pathspec); | |
c086b314 | 183 | *ppath = path; |
43f3cec2 BB |
184 | *pbsize = 0; |
185 | *pfsize = 0; | |
630431cc | 186 | } |
43f3cec2 | 187 | |
630431cc | 188 | static fio_pmemblk_file_t pmb_open(const char *pathspec, int flags) |
43f3cec2 | 189 | { |
c086b314 JA |
190 | fio_pmemblk_file_t pmb; |
191 | char *path = NULL; | |
192 | uint64_t bsize = 0; | |
193 | uint64_t fsize = 0; | |
43f3cec2 BB |
194 | |
195 | pmb_parse_path(pathspec, &path, &bsize, &fsize); | |
630431cc | 196 | if (!path) |
43f3cec2 BB |
197 | return NULL; |
198 | ||
de88999f | 199 | pthread_mutex_lock(&CacheLock); |
43f3cec2 BB |
200 | |
201 | pmb = fio_pmemblk_cache_lookup(path); | |
630431cc | 202 | if (!pmb) { |
43f3cec2 | 203 | pmb = malloc(sizeof(*pmb)); |
630431cc | 204 | if (!pmb) |
43f3cec2 BB |
205 | goto error; |
206 | ||
207 | /* try opening existing first, create it if needed */ | |
208 | pmb->pmb_pool = pmemblk_open(path, bsize); | |
630431cc | 209 | if (!pmb->pmb_pool && (errno == ENOENT) && |
c086b314 JA |
210 | (flags & PMB_CREATE) && (0 < fsize) && (0 < bsize)) { |
211 | pmb->pmb_pool = | |
212 | pmemblk_create(path, bsize, fsize, 0644); | |
43f3cec2 | 213 | } |
630431cc | 214 | if (!pmb->pmb_pool) { |
a0998600 RE |
215 | log_err("pmemblk: unable to open pmemblk pool file %s (%s)\n", |
216 | path, strerror(errno)); | |
43f3cec2 BB |
217 | goto error; |
218 | } | |
219 | ||
220 | pmb->pmb_filename = path; | |
c086b314 JA |
221 | pmb->pmb_next = NULL; |
222 | pmb->pmb_refcnt = 0; | |
223 | pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool); | |
224 | pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool); | |
43f3cec2 BB |
225 | |
226 | fio_pmemblk_cache_insert(pmb); | |
227 | } | |
228 | ||
229 | pmb->pmb_refcnt += 1; | |
230 | ||
de88999f | 231 | pthread_mutex_unlock(&CacheLock); |
c086b314 | 232 | |
43f3cec2 BB |
233 | return pmb; |
234 | ||
235 | error: | |
630431cc JA |
236 | if (pmb) { |
237 | if (pmb->pmb_pool) | |
43f3cec2 | 238 | pmemblk_close(pmb->pmb_pool); |
c086b314 | 239 | pmb->pmb_pool = NULL; |
43f3cec2 BB |
240 | pmb->pmb_filename = NULL; |
241 | free(pmb); | |
242 | } | |
630431cc | 243 | if (path) |
43f3cec2 | 244 | free(path); |
de88999f JA |
245 | |
246 | pthread_mutex_unlock(&CacheLock); | |
43f3cec2 | 247 | return NULL; |
630431cc | 248 | } |
43f3cec2 | 249 | |
630431cc | 250 | static void pmb_close(fio_pmemblk_file_t pmb, const bool keep) |
43f3cec2 | 251 | { |
de88999f | 252 | pthread_mutex_lock(&CacheLock); |
43f3cec2 BB |
253 | |
254 | pmb->pmb_refcnt--; | |
255 | ||
630431cc | 256 | if (!keep && !pmb->pmb_refcnt) { |
43f3cec2 BB |
257 | pmemblk_close(pmb->pmb_pool); |
258 | pmb->pmb_pool = NULL; | |
259 | free(pmb->pmb_filename); | |
260 | pmb->pmb_filename = NULL; | |
261 | fio_pmemblk_cache_remove(pmb); | |
262 | free(pmb); | |
263 | } | |
264 | ||
de88999f | 265 | pthread_mutex_unlock(&CacheLock); |
630431cc | 266 | } |
43f3cec2 | 267 | |
630431cc | 268 | static int pmb_get_flags(struct thread_data *td, uint64_t *pflags) |
43f3cec2 | 269 | { |
c086b314 | 270 | static int thread_warned = 0; |
43f3cec2 BB |
271 | static int odirect_warned = 0; |
272 | ||
c086b314 | 273 | uint64_t flags = 0; |
43f3cec2 BB |
274 | |
275 | if (!td->o.use_thread) { | |
276 | if (!thread_warned) { | |
277 | thread_warned = 1; | |
a0998600 | 278 | log_err("pmemblk: must set thread=1 for pmemblk engine\n"); |
43f3cec2 BB |
279 | } |
280 | return 1; | |
281 | } | |
282 | ||
283 | if (!td->o.odirect && !odirect_warned) { | |
284 | odirect_warned = 1; | |
a0998600 | 285 | log_info("pmemblk: direct == 0, but pmemblk is always direct\n"); |
43f3cec2 BB |
286 | } |
287 | ||
288 | if (td->o.allow_create) | |
289 | flags |= PMB_CREATE; | |
290 | ||
291 | (*pflags) = flags; | |
292 | return 0; | |
630431cc | 293 | } |
43f3cec2 | 294 | |
c086b314 | 295 | static int fio_pmemblk_open_file(struct thread_data *td, struct fio_file *f) |
43f3cec2 | 296 | { |
c086b314 JA |
297 | uint64_t flags = 0; |
298 | fio_pmemblk_file_t pmb; | |
43f3cec2 | 299 | |
630431cc | 300 | if (pmb_get_flags(td, &flags)) |
43f3cec2 BB |
301 | return 1; |
302 | ||
303 | pmb = pmb_open(f->file_name, flags); | |
630431cc | 304 | if (!pmb) |
43f3cec2 BB |
305 | return 1; |
306 | ||
307 | FIOFILEPMBSET(f, pmb); | |
43f3cec2 | 308 | return 0; |
630431cc | 309 | } |
43f3cec2 | 310 | |
630431cc JA |
311 | static int fio_pmemblk_close_file(struct thread_data fio_unused *td, |
312 | struct fio_file *f) | |
43f3cec2 | 313 | { |
c086b314 | 314 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); |
43f3cec2 BB |
315 | |
316 | if (pmb) | |
630431cc | 317 | pmb_close(pmb, false); |
43f3cec2 BB |
318 | |
319 | FIOFILEPMBSET(f, NULL); | |
43f3cec2 | 320 | return 0; |
630431cc | 321 | } |
43f3cec2 | 322 | |
c086b314 | 323 | static int fio_pmemblk_get_file_size(struct thread_data *td, struct fio_file *f) |
43f3cec2 | 324 | { |
c086b314 JA |
325 | uint64_t flags = 0; |
326 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
43f3cec2 BB |
327 | |
328 | if (fio_file_size_known(f)) | |
329 | return 0; | |
330 | ||
630431cc JA |
331 | if (!pmb) { |
332 | if (pmb_get_flags(td, &flags)) | |
43f3cec2 BB |
333 | return 1; |
334 | pmb = pmb_open(f->file_name, flags); | |
630431cc | 335 | if (!pmb) |
43f3cec2 BB |
336 | return 1; |
337 | } | |
338 | ||
339 | f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks; | |
340 | ||
341 | fio_file_set_size_known(f); | |
342 | ||
630431cc JA |
343 | if (!FIOFILEPMBGET(f)) |
344 | pmb_close(pmb, true); | |
43f3cec2 BB |
345 | |
346 | return 0; | |
630431cc | 347 | } |
43f3cec2 | 348 | |
c086b314 | 349 | static int fio_pmemblk_queue(struct thread_data *td, struct io_u *io_u) |
43f3cec2 | 350 | { |
c086b314 JA |
351 | struct fio_file *f = io_u->file; |
352 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); | |
43f3cec2 | 353 | |
c086b314 JA |
354 | unsigned long long off; |
355 | unsigned long len; | |
356 | void *buf; | |
43f3cec2 BB |
357 | |
358 | fio_ro_check(td, io_u); | |
359 | ||
360 | switch (io_u->ddir) { | |
361 | case DDIR_READ: | |
43f3cec2 BB |
362 | case DDIR_WRITE: |
363 | off = io_u->offset; | |
364 | len = io_u->xfer_buflen; | |
365 | ||
366 | io_u->error = EINVAL; | |
630431cc | 367 | if (off % pmb->pmb_bsize) |
43f3cec2 | 368 | break; |
630431cc | 369 | if (len % pmb->pmb_bsize) |
43f3cec2 BB |
370 | break; |
371 | if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks) | |
372 | break; | |
373 | ||
374 | io_u->error = 0; | |
375 | buf = io_u->xfer_buf; | |
376 | off /= pmb->pmb_bsize; | |
377 | len /= pmb->pmb_bsize; | |
378 | while (0 < len) { | |
cf8775b8 RE |
379 | if (io_u->ddir == DDIR_READ && |
380 | 0 != pmemblk_read(pmb->pmb_pool, buf, off)) { | |
381 | io_u->error = errno; | |
382 | break; | |
383 | } else if (0 != pmemblk_write(pmb->pmb_pool, buf, off)) { | |
43f3cec2 BB |
384 | io_u->error = errno; |
385 | break; | |
386 | } | |
387 | buf += pmb->pmb_bsize; | |
388 | off++; | |
389 | len--; | |
390 | } | |
391 | off *= pmb->pmb_bsize; | |
392 | len *= pmb->pmb_bsize; | |
393 | io_u->resid = io_u->xfer_buflen - (off - io_u->offset); | |
394 | break; | |
395 | case DDIR_SYNC: | |
396 | case DDIR_DATASYNC: | |
397 | case DDIR_SYNC_FILE_RANGE: | |
398 | /* we're always sync'd */ | |
399 | io_u->error = 0; | |
400 | break; | |
401 | default: | |
402 | io_u->error = EINVAL; | |
403 | break; | |
404 | } | |
405 | ||
406 | return FIO_Q_COMPLETED; | |
630431cc | 407 | } |
43f3cec2 | 408 | |
c086b314 | 409 | static int fio_pmemblk_unlink_file(struct thread_data *td, struct fio_file *f) |
43f3cec2 | 410 | { |
c086b314 JA |
411 | char *path = NULL; |
412 | uint64_t bsize = 0; | |
413 | uint64_t fsize = 0; | |
43f3cec2 BB |
414 | |
415 | /* | |
416 | * we need our own unlink in case the user has specified | |
417 | * the block and file sizes in the path name. we parse | |
418 | * the file_name to determine the file name we actually used. | |
419 | */ | |
420 | ||
421 | pmb_parse_path(f->file_name, &path, &bsize, &fsize); | |
630431cc | 422 | if (!path) |
2442c935 | 423 | return ENOENT; |
43f3cec2 BB |
424 | |
425 | unlink(path); | |
426 | free(path); | |
43f3cec2 | 427 | return 0; |
630431cc | 428 | } |
43f3cec2 | 429 | |
13690c10 | 430 | static struct ioengine_ops ioengine = { |
c086b314 JA |
431 | .name = "pmemblk", |
432 | .version = FIO_IOOPS_VERSION, | |
433 | .queue = fio_pmemblk_queue, | |
434 | .open_file = fio_pmemblk_open_file, | |
435 | .close_file = fio_pmemblk_close_file, | |
436 | .get_file_size = fio_pmemblk_get_file_size, | |
437 | .unlink_file = fio_pmemblk_unlink_file, | |
438 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL, | |
43f3cec2 BB |
439 | }; |
440 | ||
630431cc | 441 | static void fio_init fio_pmemblk_register(void) |
43f3cec2 BB |
442 | { |
443 | register_ioengine(&ioengine); | |
444 | } | |
445 | ||
630431cc | 446 | static void fio_exit fio_pmemblk_unregister(void) |
43f3cec2 BB |
447 | { |
448 | unregister_ioengine(&ioengine); | |
449 | } |