| 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 | * pmemblk engine |
| 23 | * |
| 24 | * IO engine that uses libpmemblk to read and write data |
| 25 | * |
| 26 | * To use: |
| 27 | * ioengine=pmemblk |
| 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 | * |
| 51 | * See examples/pmemblk.fio for more. |
| 52 | * |
| 53 | */ |
| 54 | |
| 55 | #include <stdio.h> |
| 56 | #include <stdlib.h> |
| 57 | #include <unistd.h> |
| 58 | #include <sys/uio.h> |
| 59 | #include <errno.h> |
| 60 | #include <assert.h> |
| 61 | #include <string.h> |
| 62 | #include <libpmem.h> |
| 63 | #include <libpmemblk.h> |
| 64 | |
| 65 | #include "../fio.h" |
| 66 | |
| 67 | /* |
| 68 | * libpmemblk |
| 69 | */ |
| 70 | typedef struct fio_pmemblk_file *fio_pmemblk_file_t; |
| 71 | |
| 72 | struct fio_pmemblk_file { |
| 73 | fio_pmemblk_file_t pmb_next; |
| 74 | char *pmb_filename; |
| 75 | uint64_t pmb_refcnt; |
| 76 | PMEMblkpool *pmb_pool; |
| 77 | size_t pmb_bsize; |
| 78 | size_t pmb_nblocks; |
| 79 | }; |
| 80 | #define FIOFILEPMBSET(_f, _v) do { \ |
| 81 | (_f)->engine_data = (uint64_t)(uintptr_t)(_v); \ |
| 82 | } while(0) |
| 83 | #define FIOFILEPMBGET(_f) ((fio_pmemblk_file_t)((_f)->engine_data)) |
| 84 | |
| 85 | static fio_pmemblk_file_t Cache; |
| 86 | |
| 87 | static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER; |
| 88 | |
| 89 | #define PMB_CREATE (0x0001) /* should create file */ |
| 90 | |
| 91 | fio_pmemblk_file_t fio_pmemblk_cache_lookup(const char *filename) |
| 92 | { |
| 93 | fio_pmemblk_file_t i; |
| 94 | |
| 95 | for (i = Cache; i != NULL; i = i->pmb_next) |
| 96 | if (!strcmp(filename, i->pmb_filename)) |
| 97 | return i; |
| 98 | |
| 99 | return NULL; |
| 100 | } |
| 101 | |
| 102 | static void fio_pmemblk_cache_insert(fio_pmemblk_file_t pmb) |
| 103 | { |
| 104 | pmb->pmb_next = Cache; |
| 105 | Cache = pmb; |
| 106 | } |
| 107 | |
| 108 | static void fio_pmemblk_cache_remove(fio_pmemblk_file_t pmb) |
| 109 | { |
| 110 | fio_pmemblk_file_t i; |
| 111 | |
| 112 | if (pmb == Cache) { |
| 113 | Cache = Cache->pmb_next; |
| 114 | pmb->pmb_next = NULL; |
| 115 | return; |
| 116 | } |
| 117 | |
| 118 | for (i = Cache; i != NULL; i = i->pmb_next) |
| 119 | if (pmb == i->pmb_next) { |
| 120 | i->pmb_next = i->pmb_next->pmb_next; |
| 121 | pmb->pmb_next = NULL; |
| 122 | return; |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | /* |
| 127 | * to control block size and gross file size at the libpmemblk |
| 128 | * level, we allow the block size and file size to be appended |
| 129 | * to the file name: |
| 130 | * |
| 131 | * path[,bsize,fsizemb] |
| 132 | * |
| 133 | * note that we do not use the fio option "filesize" to dictate |
| 134 | * the file size because we can only give libpmemblk the gross |
| 135 | * file size, which is different from the net or usable file |
| 136 | * size (which is probably what fio wants). |
| 137 | * |
| 138 | * the final path without the parameters is returned in ppath. |
| 139 | * the block size and file size are returned in pbsize and fsize. |
| 140 | * |
| 141 | * note that the user should specify the file size in MiB, but |
| 142 | * we return bytes from here. |
| 143 | */ |
| 144 | static void pmb_parse_path(const char *pathspec, char **ppath, uint64_t *pbsize, |
| 145 | uint64_t *pfsize) |
| 146 | { |
| 147 | char *path; |
| 148 | char *s; |
| 149 | uint64_t bsize; |
| 150 | uint64_t fsizemb; |
| 151 | |
| 152 | path = strdup(pathspec); |
| 153 | if (!path) { |
| 154 | *ppath = NULL; |
| 155 | return; |
| 156 | } |
| 157 | |
| 158 | /* extract sizes, if given */ |
| 159 | s = strrchr(path, ','); |
| 160 | if (s && (fsizemb = strtoull(s + 1, NULL, 10))) { |
| 161 | *s = 0; |
| 162 | s = strrchr(path, ','); |
| 163 | if (s && (bsize = strtoull(s + 1, NULL, 10))) { |
| 164 | *s = 0; |
| 165 | *ppath = path; |
| 166 | *pbsize = bsize; |
| 167 | *pfsize = fsizemb << 20; |
| 168 | return; |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | /* size specs not found */ |
| 173 | strcpy(path, pathspec); |
| 174 | *ppath = path; |
| 175 | *pbsize = 0; |
| 176 | *pfsize = 0; |
| 177 | } |
| 178 | |
| 179 | static fio_pmemblk_file_t pmb_open(const char *pathspec, int flags) |
| 180 | { |
| 181 | fio_pmemblk_file_t pmb; |
| 182 | char *path = NULL; |
| 183 | uint64_t bsize = 0; |
| 184 | uint64_t fsize = 0; |
| 185 | |
| 186 | pmb_parse_path(pathspec, &path, &bsize, &fsize); |
| 187 | if (!path) |
| 188 | return NULL; |
| 189 | |
| 190 | pthread_mutex_lock(&CacheLock); |
| 191 | |
| 192 | pmb = fio_pmemblk_cache_lookup(path); |
| 193 | if (!pmb) { |
| 194 | pmb = malloc(sizeof(*pmb)); |
| 195 | if (!pmb) |
| 196 | goto error; |
| 197 | |
| 198 | /* try opening existing first, create it if needed */ |
| 199 | pmb->pmb_pool = pmemblk_open(path, bsize); |
| 200 | if (!pmb->pmb_pool && (errno == ENOENT) && |
| 201 | (flags & PMB_CREATE) && (0 < fsize) && (0 < bsize)) { |
| 202 | pmb->pmb_pool = |
| 203 | pmemblk_create(path, bsize, fsize, 0644); |
| 204 | } |
| 205 | if (!pmb->pmb_pool) { |
| 206 | log_err |
| 207 | ("fio: enable to open pmemblk pool file (errno %d)\n", |
| 208 | errno); |
| 209 | goto error; |
| 210 | } |
| 211 | |
| 212 | pmb->pmb_filename = path; |
| 213 | pmb->pmb_next = NULL; |
| 214 | pmb->pmb_refcnt = 0; |
| 215 | pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool); |
| 216 | pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool); |
| 217 | |
| 218 | fio_pmemblk_cache_insert(pmb); |
| 219 | } |
| 220 | |
| 221 | pmb->pmb_refcnt += 1; |
| 222 | |
| 223 | pthread_mutex_unlock(&CacheLock); |
| 224 | |
| 225 | return pmb; |
| 226 | |
| 227 | error: |
| 228 | if (pmb) { |
| 229 | if (pmb->pmb_pool) |
| 230 | pmemblk_close(pmb->pmb_pool); |
| 231 | pmb->pmb_pool = NULL; |
| 232 | pmb->pmb_filename = NULL; |
| 233 | free(pmb); |
| 234 | } |
| 235 | if (path) |
| 236 | free(path); |
| 237 | |
| 238 | pthread_mutex_unlock(&CacheLock); |
| 239 | return NULL; |
| 240 | } |
| 241 | |
| 242 | static void pmb_close(fio_pmemblk_file_t pmb, const bool keep) |
| 243 | { |
| 244 | pthread_mutex_lock(&CacheLock); |
| 245 | |
| 246 | pmb->pmb_refcnt--; |
| 247 | |
| 248 | if (!keep && !pmb->pmb_refcnt) { |
| 249 | pmemblk_close(pmb->pmb_pool); |
| 250 | pmb->pmb_pool = NULL; |
| 251 | free(pmb->pmb_filename); |
| 252 | pmb->pmb_filename = NULL; |
| 253 | fio_pmemblk_cache_remove(pmb); |
| 254 | free(pmb); |
| 255 | } |
| 256 | |
| 257 | pthread_mutex_unlock(&CacheLock); |
| 258 | } |
| 259 | |
| 260 | static int pmb_get_flags(struct thread_data *td, uint64_t *pflags) |
| 261 | { |
| 262 | static int thread_warned = 0; |
| 263 | static int odirect_warned = 0; |
| 264 | |
| 265 | uint64_t flags = 0; |
| 266 | |
| 267 | if (!td->o.use_thread) { |
| 268 | if (!thread_warned) { |
| 269 | thread_warned = 1; |
| 270 | log_err("fio: must set thread=1 for pmemblk engine\n"); |
| 271 | } |
| 272 | return 1; |
| 273 | } |
| 274 | |
| 275 | if (!td->o.odirect && !odirect_warned) { |
| 276 | odirect_warned = 1; |
| 277 | log_info("fio: direct == 0, but pmemblk is always direct\n"); |
| 278 | } |
| 279 | |
| 280 | if (td->o.allow_create) |
| 281 | flags |= PMB_CREATE; |
| 282 | |
| 283 | (*pflags) = flags; |
| 284 | return 0; |
| 285 | } |
| 286 | |
| 287 | static int fio_pmemblk_open_file(struct thread_data *td, struct fio_file *f) |
| 288 | { |
| 289 | uint64_t flags = 0; |
| 290 | fio_pmemblk_file_t pmb; |
| 291 | |
| 292 | if (pmb_get_flags(td, &flags)) |
| 293 | return 1; |
| 294 | |
| 295 | pmb = pmb_open(f->file_name, flags); |
| 296 | if (!pmb) |
| 297 | return 1; |
| 298 | |
| 299 | FIOFILEPMBSET(f, pmb); |
| 300 | return 0; |
| 301 | } |
| 302 | |
| 303 | static int fio_pmemblk_close_file(struct thread_data fio_unused *td, |
| 304 | struct fio_file *f) |
| 305 | { |
| 306 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); |
| 307 | |
| 308 | if (pmb) |
| 309 | pmb_close(pmb, false); |
| 310 | |
| 311 | FIOFILEPMBSET(f, NULL); |
| 312 | return 0; |
| 313 | } |
| 314 | |
| 315 | static int fio_pmemblk_get_file_size(struct thread_data *td, struct fio_file *f) |
| 316 | { |
| 317 | uint64_t flags = 0; |
| 318 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); |
| 319 | |
| 320 | if (fio_file_size_known(f)) |
| 321 | return 0; |
| 322 | |
| 323 | if (!pmb) { |
| 324 | if (pmb_get_flags(td, &flags)) |
| 325 | return 1; |
| 326 | pmb = pmb_open(f->file_name, flags); |
| 327 | if (!pmb) |
| 328 | return 1; |
| 329 | } |
| 330 | |
| 331 | f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks; |
| 332 | |
| 333 | fio_file_set_size_known(f); |
| 334 | |
| 335 | if (!FIOFILEPMBGET(f)) |
| 336 | pmb_close(pmb, true); |
| 337 | |
| 338 | return 0; |
| 339 | } |
| 340 | |
| 341 | static int fio_pmemblk_queue(struct thread_data *td, struct io_u *io_u) |
| 342 | { |
| 343 | struct fio_file *f = io_u->file; |
| 344 | fio_pmemblk_file_t pmb = FIOFILEPMBGET(f); |
| 345 | |
| 346 | unsigned long long off; |
| 347 | unsigned long len; |
| 348 | void *buf; |
| 349 | |
| 350 | fio_ro_check(td, io_u); |
| 351 | |
| 352 | switch (io_u->ddir) { |
| 353 | case DDIR_READ: |
| 354 | case DDIR_WRITE: |
| 355 | off = io_u->offset; |
| 356 | len = io_u->xfer_buflen; |
| 357 | |
| 358 | io_u->error = EINVAL; |
| 359 | if (off % pmb->pmb_bsize) |
| 360 | break; |
| 361 | if (len % pmb->pmb_bsize) |
| 362 | break; |
| 363 | if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks) |
| 364 | break; |
| 365 | |
| 366 | io_u->error = 0; |
| 367 | buf = io_u->xfer_buf; |
| 368 | off /= pmb->pmb_bsize; |
| 369 | len /= pmb->pmb_bsize; |
| 370 | while (0 < len) { |
| 371 | if (io_u->ddir == DDIR_READ && |
| 372 | 0 != pmemblk_read(pmb->pmb_pool, buf, off)) { |
| 373 | io_u->error = errno; |
| 374 | break; |
| 375 | } else if (0 != pmemblk_write(pmb->pmb_pool, buf, off)) { |
| 376 | io_u->error = errno; |
| 377 | break; |
| 378 | } |
| 379 | buf += pmb->pmb_bsize; |
| 380 | off++; |
| 381 | len--; |
| 382 | } |
| 383 | off *= pmb->pmb_bsize; |
| 384 | len *= pmb->pmb_bsize; |
| 385 | io_u->resid = io_u->xfer_buflen - (off - io_u->offset); |
| 386 | break; |
| 387 | case DDIR_SYNC: |
| 388 | case DDIR_DATASYNC: |
| 389 | case DDIR_SYNC_FILE_RANGE: |
| 390 | /* we're always sync'd */ |
| 391 | io_u->error = 0; |
| 392 | break; |
| 393 | default: |
| 394 | io_u->error = EINVAL; |
| 395 | break; |
| 396 | } |
| 397 | |
| 398 | return FIO_Q_COMPLETED; |
| 399 | } |
| 400 | |
| 401 | static int fio_pmemblk_unlink_file(struct thread_data *td, struct fio_file *f) |
| 402 | { |
| 403 | char *path = NULL; |
| 404 | uint64_t bsize = 0; |
| 405 | uint64_t fsize = 0; |
| 406 | |
| 407 | /* |
| 408 | * we need our own unlink in case the user has specified |
| 409 | * the block and file sizes in the path name. we parse |
| 410 | * the file_name to determine the file name we actually used. |
| 411 | */ |
| 412 | |
| 413 | pmb_parse_path(f->file_name, &path, &bsize, &fsize); |
| 414 | if (!path) |
| 415 | return ENOENT; |
| 416 | |
| 417 | unlink(path); |
| 418 | free(path); |
| 419 | return 0; |
| 420 | } |
| 421 | |
| 422 | static struct ioengine_ops ioengine = { |
| 423 | .name = "pmemblk", |
| 424 | .version = FIO_IOOPS_VERSION, |
| 425 | .queue = fio_pmemblk_queue, |
| 426 | .open_file = fio_pmemblk_open_file, |
| 427 | .close_file = fio_pmemblk_close_file, |
| 428 | .get_file_size = fio_pmemblk_get_file_size, |
| 429 | .unlink_file = fio_pmemblk_unlink_file, |
| 430 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL, |
| 431 | }; |
| 432 | |
| 433 | static void fio_init fio_pmemblk_register(void) |
| 434 | { |
| 435 | register_ioengine(&ioengine); |
| 436 | } |
| 437 | |
| 438 | static void fio_exit fio_pmemblk_unregister(void) |
| 439 | { |
| 440 | unregister_ioengine(&ioengine); |
| 441 | } |