| 1 | /* |
| 2 | * device DAX engine |
| 3 | * |
| 4 | * IO engine that reads/writes from files by doing memcpy to/from |
| 5 | * a memory mapped region of DAX enabled device. |
| 6 | * |
| 7 | * Copyright (C) 2016 Intel Corp |
| 8 | * |
| 9 | * This program is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU General Public License, |
| 11 | * version 2 as published by the Free Software Foundation.. |
| 12 | * |
| 13 | * This program is distributed in the hope that it will be useful, |
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | * GNU General Public License for more details. |
| 17 | * |
| 18 | */ |
| 19 | |
| 20 | /* |
| 21 | * device dax engine |
| 22 | * IO engine that access a DAX device directly for read and write data |
| 23 | * |
| 24 | * To use: |
| 25 | * ioengine=dev-dax |
| 26 | * |
| 27 | * Other relevant settings: |
| 28 | * iodepth=1 |
| 29 | * direct=0 REQUIRED |
| 30 | * filename=/dev/daxN.N |
| 31 | * bs=2m |
| 32 | * |
| 33 | * direct should be left to 0. Using dev-dax implies that memory access |
| 34 | * is direct. However, dev-dax does not support O_DIRECT flag by design |
| 35 | * since it is not necessary. |
| 36 | * |
| 37 | * bs should adhere to the device dax alignment at minimally. |
| 38 | * |
| 39 | * libpmem.so |
| 40 | * By default, the dev-dax engine will let the system find the libpmem.so |
| 41 | * that it uses. You can use an alternative libpmem by setting the |
| 42 | * FIO_PMEM_LIB environment variable to the full path to the desired |
| 43 | * libpmem.so. |
| 44 | */ |
| 45 | |
| 46 | #include <stdio.h> |
| 47 | #include <limits.h> |
| 48 | #include <stdlib.h> |
| 49 | #include <unistd.h> |
| 50 | #include <errno.h> |
| 51 | #include <sys/mman.h> |
| 52 | #include <sys/stat.h> |
| 53 | #include <sys/sysmacros.h> |
| 54 | #include <libgen.h> |
| 55 | #include <libpmem.h> |
| 56 | |
| 57 | #include "../fio.h" |
| 58 | #include "../verify.h" |
| 59 | |
| 60 | /* |
| 61 | * Limits us to 1GiB of mapped files in total to model after |
| 62 | * mmap engine behavior |
| 63 | */ |
| 64 | #define MMAP_TOTAL_SZ (1 * 1024 * 1024 * 1024UL) |
| 65 | |
| 66 | struct fio_devdax_data { |
| 67 | void *devdax_ptr; |
| 68 | size_t devdax_sz; |
| 69 | off_t devdax_off; |
| 70 | }; |
| 71 | |
| 72 | static int fio_devdax_file(struct thread_data *td, struct fio_file *f, |
| 73 | size_t length, off_t off) |
| 74 | { |
| 75 | struct fio_devdax_data *fdd = FILE_ENG_DATA(f); |
| 76 | int flags = 0; |
| 77 | |
| 78 | if (td_rw(td)) |
| 79 | flags = PROT_READ | PROT_WRITE; |
| 80 | else if (td_write(td)) { |
| 81 | flags = PROT_WRITE; |
| 82 | |
| 83 | if (td->o.verify != VERIFY_NONE) |
| 84 | flags |= PROT_READ; |
| 85 | } else |
| 86 | flags = PROT_READ; |
| 87 | |
| 88 | fdd->devdax_ptr = mmap(NULL, length, flags, MAP_SHARED, f->fd, off); |
| 89 | if (fdd->devdax_ptr == MAP_FAILED) { |
| 90 | fdd->devdax_ptr = NULL; |
| 91 | td_verror(td, errno, "mmap"); |
| 92 | } |
| 93 | |
| 94 | if (td->error && fdd->devdax_ptr) |
| 95 | munmap(fdd->devdax_ptr, length); |
| 96 | |
| 97 | return td->error; |
| 98 | } |
| 99 | |
| 100 | /* |
| 101 | * Just mmap an appropriate portion, we cannot mmap the full extent |
| 102 | */ |
| 103 | static int fio_devdax_prep_limited(struct thread_data *td, struct io_u *io_u) |
| 104 | { |
| 105 | struct fio_file *f = io_u->file; |
| 106 | struct fio_devdax_data *fdd = FILE_ENG_DATA(f); |
| 107 | |
| 108 | if (io_u->buflen > f->real_file_size) { |
| 109 | log_err("dev-dax: bs too big for dev-dax engine\n"); |
| 110 | return EIO; |
| 111 | } |
| 112 | |
| 113 | fdd->devdax_sz = min(MMAP_TOTAL_SZ, f->real_file_size); |
| 114 | if (fdd->devdax_sz > f->io_size) |
| 115 | fdd->devdax_sz = f->io_size; |
| 116 | |
| 117 | fdd->devdax_off = io_u->offset; |
| 118 | |
| 119 | return fio_devdax_file(td, f, fdd->devdax_sz, fdd->devdax_off); |
| 120 | } |
| 121 | |
| 122 | /* |
| 123 | * Attempt to mmap the entire file |
| 124 | */ |
| 125 | static int fio_devdax_prep_full(struct thread_data *td, struct io_u *io_u) |
| 126 | { |
| 127 | struct fio_file *f = io_u->file; |
| 128 | struct fio_devdax_data *fdd = FILE_ENG_DATA(f); |
| 129 | int ret; |
| 130 | |
| 131 | if (fio_file_partial_mmap(f)) |
| 132 | return EINVAL; |
| 133 | |
| 134 | if (io_u->offset != (size_t) io_u->offset || |
| 135 | f->io_size != (size_t) f->io_size) { |
| 136 | fio_file_set_partial_mmap(f); |
| 137 | return EINVAL; |
| 138 | } |
| 139 | |
| 140 | fdd->devdax_sz = f->io_size; |
| 141 | fdd->devdax_off = 0; |
| 142 | |
| 143 | ret = fio_devdax_file(td, f, fdd->devdax_sz, fdd->devdax_off); |
| 144 | if (ret) |
| 145 | fio_file_set_partial_mmap(f); |
| 146 | |
| 147 | return ret; |
| 148 | } |
| 149 | |
| 150 | static int fio_devdax_prep(struct thread_data *td, struct io_u *io_u) |
| 151 | { |
| 152 | struct fio_file *f = io_u->file; |
| 153 | struct fio_devdax_data *fdd = FILE_ENG_DATA(f); |
| 154 | int ret; |
| 155 | |
| 156 | /* |
| 157 | * It fits within existing mapping, use it |
| 158 | */ |
| 159 | if (io_u->offset >= fdd->devdax_off && |
| 160 | io_u->offset + io_u->buflen < fdd->devdax_off + fdd->devdax_sz) |
| 161 | goto done; |
| 162 | |
| 163 | /* |
| 164 | * unmap any existing mapping |
| 165 | */ |
| 166 | if (fdd->devdax_ptr) { |
| 167 | if (munmap(fdd->devdax_ptr, fdd->devdax_sz) < 0) |
| 168 | return errno; |
| 169 | fdd->devdax_ptr = NULL; |
| 170 | } |
| 171 | |
| 172 | if (fio_devdax_prep_full(td, io_u)) { |
| 173 | td_clear_error(td); |
| 174 | ret = fio_devdax_prep_limited(td, io_u); |
| 175 | if (ret) |
| 176 | return ret; |
| 177 | } |
| 178 | |
| 179 | done: |
| 180 | io_u->mmap_data = fdd->devdax_ptr + io_u->offset - fdd->devdax_off - |
| 181 | f->file_offset; |
| 182 | return 0; |
| 183 | } |
| 184 | |
| 185 | static int fio_devdax_queue(struct thread_data *td, struct io_u *io_u) |
| 186 | { |
| 187 | fio_ro_check(td, io_u); |
| 188 | io_u->error = 0; |
| 189 | |
| 190 | switch (io_u->ddir) { |
| 191 | case DDIR_READ: |
| 192 | memcpy(io_u->xfer_buf, io_u->mmap_data, io_u->xfer_buflen); |
| 193 | break; |
| 194 | case DDIR_WRITE: |
| 195 | pmem_memcpy_persist(io_u->mmap_data, io_u->xfer_buf, |
| 196 | io_u->xfer_buflen); |
| 197 | break; |
| 198 | case DDIR_SYNC: |
| 199 | case DDIR_DATASYNC: |
| 200 | case DDIR_SYNC_FILE_RANGE: |
| 201 | break; |
| 202 | default: |
| 203 | io_u->error = EINVAL; |
| 204 | break; |
| 205 | } |
| 206 | |
| 207 | return FIO_Q_COMPLETED; |
| 208 | } |
| 209 | |
| 210 | static int fio_devdax_init(struct thread_data *td) |
| 211 | { |
| 212 | struct thread_options *o = &td->o; |
| 213 | |
| 214 | if ((o->rw_min_bs & page_mask) && |
| 215 | (o->fsync_blocks || o->fdatasync_blocks)) { |
| 216 | log_err("dev-dax: mmap options dictate a minimum block size of %llu bytes\n", |
| 217 | (unsigned long long) page_size); |
| 218 | return 1; |
| 219 | } |
| 220 | |
| 221 | return 0; |
| 222 | } |
| 223 | |
| 224 | static int fio_devdax_open_file(struct thread_data *td, struct fio_file *f) |
| 225 | { |
| 226 | struct fio_devdax_data *fdd; |
| 227 | int ret; |
| 228 | |
| 229 | ret = generic_open_file(td, f); |
| 230 | if (ret) |
| 231 | return ret; |
| 232 | |
| 233 | fdd = calloc(1, sizeof(*fdd)); |
| 234 | if (!fdd) { |
| 235 | int fio_unused __ret; |
| 236 | __ret = generic_close_file(td, f); |
| 237 | return 1; |
| 238 | } |
| 239 | |
| 240 | FILE_SET_ENG_DATA(f, fdd); |
| 241 | |
| 242 | return 0; |
| 243 | } |
| 244 | |
| 245 | static int fio_devdax_close_file(struct thread_data *td, struct fio_file *f) |
| 246 | { |
| 247 | struct fio_devdax_data *fdd = FILE_ENG_DATA(f); |
| 248 | |
| 249 | FILE_SET_ENG_DATA(f, NULL); |
| 250 | free(fdd); |
| 251 | fio_file_clear_partial_mmap(f); |
| 252 | |
| 253 | return generic_close_file(td, f); |
| 254 | } |
| 255 | |
| 256 | static int |
| 257 | fio_devdax_get_file_size(struct thread_data *td, struct fio_file *f) |
| 258 | { |
| 259 | char spath[PATH_MAX]; |
| 260 | char npath[PATH_MAX]; |
| 261 | char *rpath; |
| 262 | FILE *sfile; |
| 263 | uint64_t size; |
| 264 | struct stat st; |
| 265 | int rc; |
| 266 | |
| 267 | if (fio_file_size_known(f)) |
| 268 | return 0; |
| 269 | |
| 270 | if (f->filetype != FIO_TYPE_CHAR) |
| 271 | return -EINVAL; |
| 272 | |
| 273 | rc = stat(f->file_name, &st); |
| 274 | if (rc < 0) { |
| 275 | log_err("%s: failed to stat file %s (%s)\n", |
| 276 | td->o.name, f->file_name, strerror(errno)); |
| 277 | return -errno; |
| 278 | } |
| 279 | |
| 280 | snprintf(spath, PATH_MAX, "/sys/dev/char/%d:%d/subsystem", |
| 281 | major(st.st_rdev), minor(st.st_rdev)); |
| 282 | |
| 283 | rpath = realpath(spath, npath); |
| 284 | if (!rpath) { |
| 285 | log_err("%s: realpath on %s failed (%s)\n", |
| 286 | td->o.name, spath, strerror(errno)); |
| 287 | return -errno; |
| 288 | } |
| 289 | |
| 290 | /* check if DAX device */ |
| 291 | if (strcmp("/sys/class/dax", rpath)) { |
| 292 | log_err("%s: %s not a DAX device!\n", |
| 293 | td->o.name, f->file_name); |
| 294 | } |
| 295 | |
| 296 | snprintf(spath, PATH_MAX, "/sys/dev/char/%d:%d/size", |
| 297 | major(st.st_rdev), minor(st.st_rdev)); |
| 298 | |
| 299 | sfile = fopen(spath, "r"); |
| 300 | if (!sfile) { |
| 301 | log_err("%s: fopen on %s failed (%s)\n", |
| 302 | td->o.name, spath, strerror(errno)); |
| 303 | return 1; |
| 304 | } |
| 305 | |
| 306 | rc = fscanf(sfile, "%lu", &size); |
| 307 | if (rc < 0) { |
| 308 | log_err("%s: fscanf on %s failed (%s)\n", |
| 309 | td->o.name, spath, strerror(errno)); |
| 310 | return 1; |
| 311 | } |
| 312 | |
| 313 | f->real_file_size = size; |
| 314 | |
| 315 | fclose(sfile); |
| 316 | |
| 317 | if (f->file_offset > f->real_file_size) { |
| 318 | log_err("%s: offset extends end (%llu > %llu)\n", td->o.name, |
| 319 | (unsigned long long) f->file_offset, |
| 320 | (unsigned long long) f->real_file_size); |
| 321 | return 1; |
| 322 | } |
| 323 | |
| 324 | fio_file_set_size_known(f); |
| 325 | return 0; |
| 326 | } |
| 327 | |
| 328 | static struct ioengine_ops ioengine = { |
| 329 | .name = "dev-dax", |
| 330 | .version = FIO_IOOPS_VERSION, |
| 331 | .init = fio_devdax_init, |
| 332 | .prep = fio_devdax_prep, |
| 333 | .queue = fio_devdax_queue, |
| 334 | .open_file = fio_devdax_open_file, |
| 335 | .close_file = fio_devdax_close_file, |
| 336 | .get_file_size = fio_devdax_get_file_size, |
| 337 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL, |
| 338 | }; |
| 339 | |
| 340 | static void fio_init fio_devdax_register(void) |
| 341 | { |
| 342 | register_ioengine(&ioengine); |
| 343 | } |
| 344 | |
| 345 | static void fio_exit fio_devdax_unregister(void) |
| 346 | { |
| 347 | unregister_ioengine(&ioengine); |
| 348 | } |