| 1 | /* |
| 2 | * Copyright (C) 2020 Western Digital Corporation or its affiliates. |
| 3 | * |
| 4 | * This file is released under the GPL. |
| 5 | */ |
| 6 | #include <errno.h> |
| 7 | #include <string.h> |
| 8 | #include <stdlib.h> |
| 9 | #include <dirent.h> |
| 10 | #include <fcntl.h> |
| 11 | #include <sys/ioctl.h> |
| 12 | #include <sys/stat.h> |
| 13 | #include <unistd.h> |
| 14 | |
| 15 | #include "file.h" |
| 16 | #include "fio.h" |
| 17 | #include "lib/pow2.h" |
| 18 | #include "log.h" |
| 19 | #include "oslib/asprintf.h" |
| 20 | #include "smalloc.h" |
| 21 | #include "verify.h" |
| 22 | #include "zbd_types.h" |
| 23 | |
| 24 | #include <linux/blkzoned.h> |
| 25 | |
| 26 | /* |
| 27 | * If the uapi headers installed on the system lacks zone capacity support, |
| 28 | * use our local versions. If the installed headers are recent enough to |
| 29 | * support zone capacity, do not redefine any structs. |
| 30 | */ |
| 31 | #ifndef CONFIG_HAVE_REP_CAPACITY |
| 32 | #define BLK_ZONE_REP_CAPACITY (1 << 0) |
| 33 | |
| 34 | struct blk_zone_v2 { |
| 35 | __u64 start; /* Zone start sector */ |
| 36 | __u64 len; /* Zone length in number of sectors */ |
| 37 | __u64 wp; /* Zone write pointer position */ |
| 38 | __u8 type; /* Zone type */ |
| 39 | __u8 cond; /* Zone condition */ |
| 40 | __u8 non_seq; /* Non-sequential write resources active */ |
| 41 | __u8 reset; /* Reset write pointer recommended */ |
| 42 | __u8 resv[4]; |
| 43 | __u64 capacity; /* Zone capacity in number of sectors */ |
| 44 | __u8 reserved[24]; |
| 45 | }; |
| 46 | #define blk_zone blk_zone_v2 |
| 47 | |
| 48 | struct blk_zone_report_v2 { |
| 49 | __u64 sector; |
| 50 | __u32 nr_zones; |
| 51 | __u32 flags; |
| 52 | struct blk_zone zones[0]; |
| 53 | }; |
| 54 | #define blk_zone_report blk_zone_report_v2 |
| 55 | #endif /* CONFIG_HAVE_REP_CAPACITY */ |
| 56 | |
| 57 | /* |
| 58 | * Read up to 255 characters from the first line of a file. Strip the trailing |
| 59 | * newline. |
| 60 | */ |
| 61 | static char *read_file(const char *path) |
| 62 | { |
| 63 | char line[256], *p = line; |
| 64 | FILE *f; |
| 65 | |
| 66 | f = fopen(path, "rb"); |
| 67 | if (!f) |
| 68 | return NULL; |
| 69 | if (!fgets(line, sizeof(line), f)) |
| 70 | line[0] = '\0'; |
| 71 | strsep(&p, "\n"); |
| 72 | fclose(f); |
| 73 | |
| 74 | return strdup(line); |
| 75 | } |
| 76 | |
| 77 | /* |
| 78 | * Get the value of a sysfs attribute for a block device. |
| 79 | * |
| 80 | * Returns NULL on failure. |
| 81 | * Returns a pointer to a string on success. |
| 82 | * The caller is responsible for freeing the memory. |
| 83 | */ |
| 84 | static char *blkzoned_get_sysfs_attr(const char *file_name, const char *attr) |
| 85 | { |
| 86 | char *attr_path = NULL; |
| 87 | struct stat statbuf; |
| 88 | char *sys_devno_path = NULL; |
| 89 | char *part_attr_path = NULL; |
| 90 | char *part_str = NULL; |
| 91 | char sys_path[PATH_MAX]; |
| 92 | ssize_t sz; |
| 93 | char *delim = NULL; |
| 94 | char *attr_str = NULL; |
| 95 | |
| 96 | if (stat(file_name, &statbuf) < 0) |
| 97 | goto out; |
| 98 | |
| 99 | if (asprintf(&sys_devno_path, "/sys/dev/block/%d:%d", |
| 100 | major(statbuf.st_rdev), minor(statbuf.st_rdev)) < 0) |
| 101 | goto out; |
| 102 | |
| 103 | sz = readlink(sys_devno_path, sys_path, sizeof(sys_path) - 1); |
| 104 | if (sz < 0) |
| 105 | goto out; |
| 106 | sys_path[sz] = '\0'; |
| 107 | |
| 108 | /* |
| 109 | * If the device is a partition device, cut the device name in the |
| 110 | * canonical sysfs path to obtain the sysfs path of the holder device. |
| 111 | * e.g.: /sys/devices/.../sda/sda1 -> /sys/devices/.../sda |
| 112 | */ |
| 113 | if (asprintf(&part_attr_path, "/sys/dev/block/%s/partition", |
| 114 | sys_path) < 0) |
| 115 | goto out; |
| 116 | part_str = read_file(part_attr_path); |
| 117 | if (part_str && *part_str == '1') { |
| 118 | delim = strrchr(sys_path, '/'); |
| 119 | if (!delim) |
| 120 | goto out; |
| 121 | *delim = '\0'; |
| 122 | } |
| 123 | |
| 124 | if (asprintf(&attr_path, |
| 125 | "/sys/dev/block/%s/%s", sys_path, attr) < 0) |
| 126 | goto out; |
| 127 | |
| 128 | attr_str = read_file(attr_path); |
| 129 | out: |
| 130 | free(attr_path); |
| 131 | free(part_str); |
| 132 | free(part_attr_path); |
| 133 | free(sys_devno_path); |
| 134 | |
| 135 | return attr_str; |
| 136 | } |
| 137 | |
| 138 | int blkzoned_get_zoned_model(struct thread_data *td, struct fio_file *f, |
| 139 | enum zbd_zoned_model *model) |
| 140 | { |
| 141 | char *model_str = NULL; |
| 142 | |
| 143 | if (f->filetype != FIO_TYPE_BLOCK) |
| 144 | return -EINVAL; |
| 145 | |
| 146 | *model = ZBD_NONE; |
| 147 | |
| 148 | model_str = blkzoned_get_sysfs_attr(f->file_name, "queue/zoned"); |
| 149 | if (!model_str) |
| 150 | return 0; |
| 151 | |
| 152 | dprint(FD_ZBD, "%s: zbd model string: %s\n", f->file_name, model_str); |
| 153 | if (strcmp(model_str, "host-aware") == 0) |
| 154 | *model = ZBD_HOST_AWARE; |
| 155 | else if (strcmp(model_str, "host-managed") == 0) |
| 156 | *model = ZBD_HOST_MANAGED; |
| 157 | |
| 158 | free(model_str); |
| 159 | |
| 160 | return 0; |
| 161 | } |
| 162 | |
| 163 | int blkzoned_get_max_open_zones(struct thread_data *td, struct fio_file *f, |
| 164 | unsigned int *max_open_zones) |
| 165 | { |
| 166 | char *max_open_str; |
| 167 | |
| 168 | if (f->filetype != FIO_TYPE_BLOCK) |
| 169 | return -EIO; |
| 170 | |
| 171 | max_open_str = blkzoned_get_sysfs_attr(f->file_name, "queue/max_open_zones"); |
| 172 | if (!max_open_str) { |
| 173 | *max_open_zones = 0; |
| 174 | return 0; |
| 175 | } |
| 176 | |
| 177 | dprint(FD_ZBD, "%s: max open zones supported by device: %s\n", |
| 178 | f->file_name, max_open_str); |
| 179 | *max_open_zones = atoll(max_open_str); |
| 180 | |
| 181 | free(max_open_str); |
| 182 | |
| 183 | return 0; |
| 184 | } |
| 185 | |
| 186 | static uint64_t zone_capacity(struct blk_zone_report *hdr, |
| 187 | struct blk_zone *blkz) |
| 188 | { |
| 189 | if (hdr->flags & BLK_ZONE_REP_CAPACITY) |
| 190 | return blkz->capacity << 9; |
| 191 | return blkz->len << 9; |
| 192 | } |
| 193 | |
| 194 | int blkzoned_report_zones(struct thread_data *td, struct fio_file *f, |
| 195 | uint64_t offset, struct zbd_zone *zones, |
| 196 | unsigned int nr_zones) |
| 197 | { |
| 198 | struct blk_zone_report *hdr = NULL; |
| 199 | struct blk_zone *blkz; |
| 200 | struct zbd_zone *z; |
| 201 | unsigned int i; |
| 202 | int fd = -1, ret; |
| 203 | |
| 204 | fd = open(f->file_name, O_RDONLY | O_LARGEFILE); |
| 205 | if (fd < 0) |
| 206 | return -errno; |
| 207 | |
| 208 | hdr = calloc(1, sizeof(struct blk_zone_report) + |
| 209 | nr_zones * sizeof(struct blk_zone)); |
| 210 | if (!hdr) { |
| 211 | ret = -ENOMEM; |
| 212 | goto out; |
| 213 | } |
| 214 | |
| 215 | hdr->nr_zones = nr_zones; |
| 216 | hdr->sector = offset >> 9; |
| 217 | ret = ioctl(fd, BLKREPORTZONE, hdr); |
| 218 | if (ret) { |
| 219 | ret = -errno; |
| 220 | goto out; |
| 221 | } |
| 222 | |
| 223 | nr_zones = hdr->nr_zones; |
| 224 | blkz = (void *) hdr + sizeof(*hdr); |
| 225 | z = &zones[0]; |
| 226 | for (i = 0; i < nr_zones; i++, z++, blkz++) { |
| 227 | z->start = blkz->start << 9; |
| 228 | z->wp = blkz->wp << 9; |
| 229 | z->len = blkz->len << 9; |
| 230 | z->capacity = zone_capacity(hdr, blkz); |
| 231 | |
| 232 | switch (blkz->type) { |
| 233 | case BLK_ZONE_TYPE_CONVENTIONAL: |
| 234 | z->type = ZBD_ZONE_TYPE_CNV; |
| 235 | break; |
| 236 | case BLK_ZONE_TYPE_SEQWRITE_REQ: |
| 237 | z->type = ZBD_ZONE_TYPE_SWR; |
| 238 | break; |
| 239 | case BLK_ZONE_TYPE_SEQWRITE_PREF: |
| 240 | z->type = ZBD_ZONE_TYPE_SWP; |
| 241 | break; |
| 242 | default: |
| 243 | td_verror(td, errno, "invalid zone type"); |
| 244 | log_err("%s: invalid type for zone at sector %llu.\n", |
| 245 | f->file_name, (unsigned long long)offset >> 9); |
| 246 | ret = -EIO; |
| 247 | goto out; |
| 248 | } |
| 249 | |
| 250 | switch (blkz->cond) { |
| 251 | case BLK_ZONE_COND_NOT_WP: |
| 252 | z->cond = ZBD_ZONE_COND_NOT_WP; |
| 253 | break; |
| 254 | case BLK_ZONE_COND_EMPTY: |
| 255 | z->cond = ZBD_ZONE_COND_EMPTY; |
| 256 | break; |
| 257 | case BLK_ZONE_COND_IMP_OPEN: |
| 258 | z->cond = ZBD_ZONE_COND_IMP_OPEN; |
| 259 | break; |
| 260 | case BLK_ZONE_COND_EXP_OPEN: |
| 261 | z->cond = ZBD_ZONE_COND_EXP_OPEN; |
| 262 | break; |
| 263 | case BLK_ZONE_COND_CLOSED: |
| 264 | z->cond = ZBD_ZONE_COND_CLOSED; |
| 265 | break; |
| 266 | case BLK_ZONE_COND_FULL: |
| 267 | z->cond = ZBD_ZONE_COND_FULL; |
| 268 | break; |
| 269 | case BLK_ZONE_COND_READONLY: |
| 270 | case BLK_ZONE_COND_OFFLINE: |
| 271 | default: |
| 272 | /* Treat all these conditions as offline (don't use!) */ |
| 273 | z->cond = ZBD_ZONE_COND_OFFLINE; |
| 274 | z->wp = z->start; |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | ret = nr_zones; |
| 279 | out: |
| 280 | free(hdr); |
| 281 | close(fd); |
| 282 | |
| 283 | return ret; |
| 284 | } |
| 285 | |
| 286 | int blkzoned_reset_wp(struct thread_data *td, struct fio_file *f, |
| 287 | uint64_t offset, uint64_t length) |
| 288 | { |
| 289 | struct blk_zone_range zr = { |
| 290 | .sector = offset >> 9, |
| 291 | .nr_sectors = length >> 9, |
| 292 | }; |
| 293 | int fd, ret = 0; |
| 294 | |
| 295 | /* If the file is not yet opened, open it for this function. */ |
| 296 | fd = f->fd; |
| 297 | if (fd < 0) { |
| 298 | fd = open(f->file_name, O_RDWR | O_LARGEFILE); |
| 299 | if (fd < 0) |
| 300 | return -errno; |
| 301 | } |
| 302 | |
| 303 | if (ioctl(fd, BLKRESETZONE, &zr) < 0) |
| 304 | ret = -errno; |
| 305 | |
| 306 | if (f->fd < 0) |
| 307 | close(fd); |
| 308 | |
| 309 | return ret; |
| 310 | } |