| 1 | /* |
| 2 | * glusterfs engine |
| 3 | * |
| 4 | * common Glusterfs's gfapi interface |
| 5 | * |
| 6 | */ |
| 7 | |
| 8 | #include "gfapi.h" |
| 9 | #include "../optgroup.h" |
| 10 | |
| 11 | struct fio_option gfapi_options[] = { |
| 12 | { |
| 13 | .name = "volume", |
| 14 | .lname = "Glusterfs volume", |
| 15 | .type = FIO_OPT_STR_STORE, |
| 16 | .help = "Name of the Glusterfs volume", |
| 17 | .off1 = offsetof(struct gf_options, gf_vol), |
| 18 | .category = FIO_OPT_C_ENGINE, |
| 19 | .group = FIO_OPT_G_GFAPI, |
| 20 | }, |
| 21 | { |
| 22 | .name = "brick", |
| 23 | .lname = "Glusterfs brick name", |
| 24 | .type = FIO_OPT_STR_STORE, |
| 25 | .help = "Name of the Glusterfs brick to connect", |
| 26 | .off1 = offsetof(struct gf_options, gf_brick), |
| 27 | .category = FIO_OPT_C_ENGINE, |
| 28 | .group = FIO_OPT_G_GFAPI, |
| 29 | }, |
| 30 | { |
| 31 | .name = "single-instance", |
| 32 | .lname = "Single glusterfs instance", |
| 33 | .type = FIO_OPT_BOOL, |
| 34 | .help = "Only one glusterfs instance", |
| 35 | .off1 = offsetof(struct gf_options, gf_single_instance), |
| 36 | .category = FIO_OPT_C_ENGINE, |
| 37 | .group = FIO_OPT_G_GFAPI, |
| 38 | }, |
| 39 | { |
| 40 | .name = NULL, |
| 41 | }, |
| 42 | }; |
| 43 | |
| 44 | struct glfs_info { |
| 45 | struct flist_head list; |
| 46 | char *volume; |
| 47 | char *brick; |
| 48 | glfs_t *fs; |
| 49 | int refcount; |
| 50 | }; |
| 51 | |
| 52 | static pthread_mutex_t glfs_lock = PTHREAD_MUTEX_INITIALIZER; |
| 53 | static FLIST_HEAD(glfs_list_head); |
| 54 | |
| 55 | static glfs_t *fio_gf_new_fs(char *volume, char *brick) |
| 56 | { |
| 57 | int r = 0; |
| 58 | glfs_t *fs; |
| 59 | struct stat sb = { 0, }; |
| 60 | |
| 61 | fs = glfs_new(volume); |
| 62 | if (!fs) { |
| 63 | log_err("glfs_new failed.\n"); |
| 64 | goto out; |
| 65 | } |
| 66 | glfs_set_logging(fs, "/tmp/fio_gfapi.log", 7); |
| 67 | /* default to tcp */ |
| 68 | r = glfs_set_volfile_server(fs, "tcp", brick, 0); |
| 69 | if (r) { |
| 70 | log_err("glfs_set_volfile_server failed.\n"); |
| 71 | goto out; |
| 72 | } |
| 73 | r = glfs_init(fs); |
| 74 | if (r) { |
| 75 | log_err("glfs_init failed. Is glusterd running on brick?\n"); |
| 76 | goto out; |
| 77 | } |
| 78 | sleep(2); |
| 79 | r = glfs_lstat(fs, ".", &sb); |
| 80 | if (r) { |
| 81 | log_err("glfs_lstat failed.\n"); |
| 82 | goto out; |
| 83 | } |
| 84 | |
| 85 | out: |
| 86 | if (r) { |
| 87 | glfs_fini(fs); |
| 88 | fs = NULL; |
| 89 | } |
| 90 | return fs; |
| 91 | } |
| 92 | |
| 93 | static glfs_t *fio_gf_get_glfs(struct gf_options *opt, |
| 94 | char *volume, char *brick) |
| 95 | { |
| 96 | struct glfs_info *glfs = NULL; |
| 97 | struct glfs_info *tmp; |
| 98 | struct flist_head *entry; |
| 99 | |
| 100 | if (!opt->gf_single_instance) |
| 101 | return fio_gf_new_fs(volume, brick); |
| 102 | |
| 103 | pthread_mutex_lock (&glfs_lock); |
| 104 | |
| 105 | flist_for_each(entry, &glfs_list_head) { |
| 106 | tmp = flist_entry(entry, struct glfs_info, list); |
| 107 | if (!strcmp(volume, tmp->volume) && |
| 108 | !strcmp(brick, tmp->brick)) { |
| 109 | glfs = tmp; |
| 110 | break; |
| 111 | } |
| 112 | } |
| 113 | |
| 114 | if (glfs) { |
| 115 | glfs->refcount++; |
| 116 | } else { |
| 117 | glfs = malloc(sizeof(*glfs)); |
| 118 | if (!glfs) |
| 119 | goto out; |
| 120 | INIT_FLIST_HEAD(&glfs->list); |
| 121 | glfs->refcount = 0; |
| 122 | glfs->volume = strdup(volume); |
| 123 | glfs->brick = strdup(brick); |
| 124 | glfs->fs = fio_gf_new_fs(volume, brick); |
| 125 | if (!glfs->fs) { |
| 126 | free(glfs); |
| 127 | glfs = NULL; |
| 128 | goto out; |
| 129 | } |
| 130 | |
| 131 | flist_add_tail(&glfs->list, &glfs_list_head); |
| 132 | glfs->refcount = 1; |
| 133 | } |
| 134 | |
| 135 | out: |
| 136 | pthread_mutex_unlock (&glfs_lock); |
| 137 | |
| 138 | if (glfs) |
| 139 | return glfs->fs; |
| 140 | return NULL; |
| 141 | } |
| 142 | |
| 143 | static void fio_gf_put_glfs(struct gf_options *opt, glfs_t *fs) |
| 144 | { |
| 145 | struct glfs_info *glfs = NULL; |
| 146 | struct glfs_info *tmp; |
| 147 | struct flist_head *entry; |
| 148 | |
| 149 | if (!opt->gf_single_instance) { |
| 150 | glfs_fini(fs); |
| 151 | return; |
| 152 | } |
| 153 | |
| 154 | pthread_mutex_lock (&glfs_lock); |
| 155 | |
| 156 | flist_for_each(entry, &glfs_list_head) { |
| 157 | tmp = flist_entry(entry, struct glfs_info, list); |
| 158 | if (tmp->fs == fs) { |
| 159 | glfs = tmp; |
| 160 | break; |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | if (!glfs) { |
| 165 | log_err("glfs not found to fini.\n"); |
| 166 | } else { |
| 167 | glfs->refcount--; |
| 168 | |
| 169 | if (glfs->refcount == 0) { |
| 170 | glfs_fini(glfs->fs); |
| 171 | free(glfs->volume); |
| 172 | free(glfs->brick); |
| 173 | flist_del(&glfs->list); |
| 174 | } |
| 175 | } |
| 176 | |
| 177 | pthread_mutex_unlock (&glfs_lock); |
| 178 | } |
| 179 | |
| 180 | int fio_gf_setup(struct thread_data *td) |
| 181 | { |
| 182 | struct gf_data *g = NULL; |
| 183 | struct gf_options *opt = td->eo; |
| 184 | |
| 185 | dprint(FD_IO, "fio setup\n"); |
| 186 | |
| 187 | if (td->io_ops_data) |
| 188 | return 0; |
| 189 | |
| 190 | g = malloc(sizeof(struct gf_data)); |
| 191 | if (!g) { |
| 192 | log_err("malloc failed.\n"); |
| 193 | return -ENOMEM; |
| 194 | } |
| 195 | g->fd = NULL; |
| 196 | g->aio_events = NULL; |
| 197 | |
| 198 | g->fs = fio_gf_get_glfs(opt, opt->gf_vol, opt->gf_brick); |
| 199 | if (!g->fs) |
| 200 | goto cleanup; |
| 201 | |
| 202 | dprint(FD_FILE, "fio setup %p\n", g->fs); |
| 203 | td->io_ops_data = g; |
| 204 | return 0; |
| 205 | cleanup: |
| 206 | free(g); |
| 207 | td->io_ops_data = NULL; |
| 208 | return -EIO; |
| 209 | } |
| 210 | |
| 211 | void fio_gf_cleanup(struct thread_data *td) |
| 212 | { |
| 213 | struct gf_data *g = td->io_ops_data; |
| 214 | |
| 215 | if (g) { |
| 216 | if (g->aio_events) |
| 217 | free(g->aio_events); |
| 218 | if (g->fd) |
| 219 | glfs_close(g->fd); |
| 220 | if (g->fs) |
| 221 | fio_gf_put_glfs(td->eo, g->fs); |
| 222 | free(g); |
| 223 | td->io_ops_data = NULL; |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | int fio_gf_get_file_size(struct thread_data *td, struct fio_file *f) |
| 228 | { |
| 229 | struct stat buf; |
| 230 | int ret; |
| 231 | struct gf_data *g = td->io_ops_data; |
| 232 | |
| 233 | dprint(FD_FILE, "get file size %s\n", f->file_name); |
| 234 | |
| 235 | if (!g || !g->fs) { |
| 236 | return 0; |
| 237 | } |
| 238 | if (fio_file_size_known(f)) |
| 239 | return 0; |
| 240 | |
| 241 | ret = glfs_lstat(g->fs, f->file_name, &buf); |
| 242 | if (ret < 0) { |
| 243 | log_err("glfs_lstat failed.\n"); |
| 244 | return ret; |
| 245 | } |
| 246 | |
| 247 | f->real_file_size = buf.st_size; |
| 248 | fio_file_set_size_known(f); |
| 249 | |
| 250 | return 0; |
| 251 | |
| 252 | } |
| 253 | |
| 254 | int fio_gf_open_file(struct thread_data *td, struct fio_file *f) |
| 255 | { |
| 256 | |
| 257 | int flags = 0; |
| 258 | int ret = 0; |
| 259 | struct gf_data *g = td->io_ops_data; |
| 260 | struct stat sb = { 0, }; |
| 261 | |
| 262 | if (td_write(td)) { |
| 263 | if (!read_only) |
| 264 | flags = O_RDWR; |
| 265 | } else if (td_read(td)) { |
| 266 | if (!read_only) |
| 267 | flags = O_RDWR; |
| 268 | else |
| 269 | flags = O_RDONLY; |
| 270 | } |
| 271 | |
| 272 | if (td->o.odirect) |
| 273 | flags |= OS_O_DIRECT; |
| 274 | if (td->o.sync_io) |
| 275 | flags |= O_SYNC; |
| 276 | |
| 277 | dprint(FD_FILE, "fio file %s open mode %s td rw %s\n", f->file_name, |
| 278 | flags & O_RDONLY ? "ro" : "rw", td_read(td) ? "read" : "write"); |
| 279 | g->fd = glfs_creat(g->fs, f->file_name, flags, 0644); |
| 280 | if (!g->fd) { |
| 281 | ret = errno; |
| 282 | log_err("glfs_creat failed.\n"); |
| 283 | return ret; |
| 284 | } |
| 285 | /* file for read doesn't exist or shorter than required, create/extend it */ |
| 286 | if (td_read(td)) { |
| 287 | if (glfs_lstat(g->fs, f->file_name, &sb) |
| 288 | || sb.st_size < f->real_file_size) { |
| 289 | dprint(FD_FILE, "fio extend file %s from %jd to %" PRIu64 "\n", |
| 290 | f->file_name, (intmax_t) sb.st_size, f->real_file_size); |
| 291 | ret = glfs_ftruncate(g->fd, f->real_file_size); |
| 292 | if (ret) { |
| 293 | log_err("failed fio extend file %s to %" PRIu64 "\n", |
| 294 | f->file_name, f->real_file_size); |
| 295 | } else { |
| 296 | unsigned long long left; |
| 297 | unsigned int bs; |
| 298 | char *b; |
| 299 | int r; |
| 300 | |
| 301 | /* fill the file, copied from extend_file */ |
| 302 | b = malloc(td->o.max_bs[DDIR_WRITE]); |
| 303 | |
| 304 | left = f->real_file_size; |
| 305 | while (left && !td->terminate) { |
| 306 | bs = td->o.max_bs[DDIR_WRITE]; |
| 307 | if (bs > left) |
| 308 | bs = left; |
| 309 | |
| 310 | fill_io_buffer(td, b, bs, bs); |
| 311 | |
| 312 | r = glfs_write(g->fd, b, bs, 0); |
| 313 | dprint(FD_IO, |
| 314 | "fio write %d of %" PRIu64 " file %s\n", |
| 315 | r, f->real_file_size, |
| 316 | f->file_name); |
| 317 | |
| 318 | if (r > 0) { |
| 319 | left -= r; |
| 320 | continue; |
| 321 | } else { |
| 322 | if (r < 0) { |
| 323 | int __e = errno; |
| 324 | |
| 325 | if (__e == ENOSPC) { |
| 326 | if (td->o. |
| 327 | fill_device) |
| 328 | break; |
| 329 | log_info |
| 330 | ("fio: ENOSPC on laying out " |
| 331 | "file, stopping\n"); |
| 332 | break; |
| 333 | } |
| 334 | td_verror(td, errno, |
| 335 | "write"); |
| 336 | } else |
| 337 | td_verror(td, EIO, |
| 338 | "write"); |
| 339 | |
| 340 | break; |
| 341 | } |
| 342 | } |
| 343 | |
| 344 | if (b) |
| 345 | free(b); |
| 346 | glfs_lseek(g->fd, 0, SEEK_SET); |
| 347 | |
| 348 | if (td->terminate && td->o.unlink) { |
| 349 | dprint(FD_FILE, "terminate unlink %s\n", |
| 350 | f->file_name); |
| 351 | glfs_unlink(g->fs, f->file_name); |
| 352 | } else if (td->o.create_fsync) { |
| 353 | if (glfs_fsync(g->fd) < 0) { |
| 354 | dprint(FD_FILE, |
| 355 | "failed to sync, close %s\n", |
| 356 | f->file_name); |
| 357 | td_verror(td, errno, "fsync"); |
| 358 | glfs_close(g->fd); |
| 359 | g->fd = NULL; |
| 360 | return 1; |
| 361 | } |
| 362 | } |
| 363 | } |
| 364 | } |
| 365 | } |
| 366 | #if defined(GFAPI_USE_FADVISE) |
| 367 | { |
| 368 | int r = 0; |
| 369 | if (td_random(td)) { |
| 370 | r = glfs_fadvise(g->fd, 0, f->real_file_size, |
| 371 | POSIX_FADV_RANDOM); |
| 372 | } else { |
| 373 | r = glfs_fadvise(g->fd, 0, f->real_file_size, |
| 374 | POSIX_FADV_SEQUENTIAL); |
| 375 | } |
| 376 | if (r) { |
| 377 | dprint(FD_FILE, "fio %p fadvise %s status %d\n", g->fs, |
| 378 | f->file_name, r); |
| 379 | } |
| 380 | } |
| 381 | #endif |
| 382 | dprint(FD_FILE, "fio %p created %s\n", g->fs, f->file_name); |
| 383 | f->fd = -1; |
| 384 | f->shadow_fd = -1; |
| 385 | td->o.open_files ++; |
| 386 | return ret; |
| 387 | } |
| 388 | |
| 389 | int fio_gf_close_file(struct thread_data *td, struct fio_file *f) |
| 390 | { |
| 391 | int ret = 0; |
| 392 | struct gf_data *g = td->io_ops_data; |
| 393 | |
| 394 | dprint(FD_FILE, "fd close %s\n", f->file_name); |
| 395 | |
| 396 | if (g) { |
| 397 | if (g->fd && glfs_close(g->fd) < 0) |
| 398 | ret = errno; |
| 399 | g->fd = NULL; |
| 400 | } |
| 401 | |
| 402 | return ret; |
| 403 | } |
| 404 | |
| 405 | int fio_gf_unlink_file(struct thread_data *td, struct fio_file *f) |
| 406 | { |
| 407 | int ret = 0; |
| 408 | struct gf_data *g = td->io_ops_data; |
| 409 | |
| 410 | dprint(FD_FILE, "fd unlink %s\n", f->file_name); |
| 411 | |
| 412 | if (g) { |
| 413 | if (g->fd && glfs_close(g->fd) < 0) |
| 414 | ret = errno; |
| 415 | |
| 416 | glfs_unlink(g->fs, f->file_name); |
| 417 | |
| 418 | if (g->fs) |
| 419 | glfs_fini(g->fs); |
| 420 | |
| 421 | g->fd = NULL; |
| 422 | free(g); |
| 423 | } |
| 424 | td->io_ops_data = NULL; |
| 425 | |
| 426 | return ret; |
| 427 | } |