Commit | Line | Data |
---|---|---|
bfbdd35b BVA |
1 | /* |
2 | * Copyright (C) 2018 Western Digital Corporation or its affiliates. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include <errno.h> | |
8 | #include <string.h> | |
9 | #include <stdlib.h> | |
bfbdd35b | 10 | #include <fcntl.h> |
bfbdd35b BVA |
11 | #include <sys/stat.h> |
12 | #include <unistd.h> | |
f5bff36e | 13 | |
cf42d79e | 14 | #include "os/os.h" |
bfbdd35b BVA |
15 | #include "file.h" |
16 | #include "fio.h" | |
17 | #include "lib/pow2.h" | |
18 | #include "log.h" | |
f5bff36e | 19 | #include "oslib/asprintf.h" |
bfbdd35b BVA |
20 | #include "smalloc.h" |
21 | #include "verify.h" | |
22 | #include "zbd.h" | |
23 | ||
b7694961 DLM |
24 | /** |
25 | * zbd_get_zoned_model - Get a device zoned model | |
26 | * @td: FIO thread data | |
27 | * @f: FIO file for which to get model information | |
28 | */ | |
29 | int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, | |
30 | enum zbd_zoned_model *model) | |
31 | { | |
32 | int ret; | |
33 | ||
6c5b11d3 DLM |
34 | if (td->io_ops && td->io_ops->get_zoned_model) |
35 | ret = td->io_ops->get_zoned_model(td, f, model); | |
36 | else | |
37 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
38 | if (ret < 0) { |
39 | td_verror(td, errno, "get zoned model failed"); | |
40 | log_err("%s: get zoned model failed (%d).\n", | |
41 | f->file_name, errno); | |
42 | } | |
43 | ||
44 | return ret; | |
45 | } | |
46 | ||
47 | /** | |
48 | * zbd_report_zones - Get zone information | |
49 | * @td: FIO thread data. | |
50 | * @f: FIO file for which to get zone information | |
51 | * @offset: offset from which to report zones | |
52 | * @zones: Array of struct zbd_zone | |
53 | * @nr_zones: Size of @zones array | |
54 | * | |
55 | * Get zone information into @zones starting from the zone at offset @offset | |
56 | * for the device specified by @f. | |
57 | * | |
58 | * Returns the number of zones reported upon success and a negative error code | |
59 | * upon failure. If the zone report is empty, always assume an error (device | |
60 | * problem) and return -EIO. | |
61 | */ | |
62 | int zbd_report_zones(struct thread_data *td, struct fio_file *f, | |
63 | uint64_t offset, struct zbd_zone *zones, | |
64 | unsigned int nr_zones) | |
65 | { | |
66 | int ret; | |
67 | ||
6c5b11d3 DLM |
68 | if (td->io_ops && td->io_ops->report_zones) |
69 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
70 | else | |
71 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
72 | if (ret < 0) { |
73 | td_verror(td, errno, "report zones failed"); | |
74 | log_err("%s: report zones from sector %llu failed (%d).\n", | |
75 | f->file_name, (unsigned long long)offset >> 9, errno); | |
76 | } else if (ret == 0) { | |
77 | td_verror(td, errno, "Empty zone report"); | |
78 | log_err("%s: report zones from sector %llu is empty.\n", | |
79 | f->file_name, (unsigned long long)offset >> 9); | |
80 | ret = -EIO; | |
81 | } | |
82 | ||
83 | return ret; | |
84 | } | |
85 | ||
86 | /** | |
87 | * zbd_reset_wp - reset the write pointer of a range of zones | |
88 | * @td: FIO thread data. | |
89 | * @f: FIO file for which to reset zones | |
90 | * @offset: Starting offset of the first zone to reset | |
91 | * @length: Length of the range of zones to reset | |
92 | * | |
93 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
94 | * Returns 0 upon success and a negative error code upon failure. | |
95 | */ | |
96 | int zbd_reset_wp(struct thread_data *td, struct fio_file *f, | |
97 | uint64_t offset, uint64_t length) | |
98 | { | |
99 | int ret; | |
100 | ||
6c5b11d3 DLM |
101 | if (td->io_ops && td->io_ops->reset_wp) |
102 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
103 | else | |
104 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
105 | if (ret < 0) { |
106 | td_verror(td, errno, "resetting wp failed"); | |
107 | log_err("%s: resetting wp for %llu sectors at sector %llu failed (%d).\n", | |
108 | f->file_name, (unsigned long long)length >> 9, | |
109 | (unsigned long long)offset >> 9, errno); | |
110 | } | |
111 | ||
112 | return ret; | |
113 | } | |
114 | ||
bfbdd35b BVA |
115 | /** |
116 | * zbd_zone_idx - convert an offset into a zone number | |
117 | * @f: file pointer. | |
118 | * @offset: offset in bytes. If this offset is in the first zone_size bytes | |
119 | * past the disk size then the index of the sentinel is returned. | |
120 | */ | |
121 | static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) | |
122 | { | |
123 | uint32_t zone_idx; | |
124 | ||
cd775e06 | 125 | if (f->zbd_info->zone_size_log2 > 0) |
bfbdd35b BVA |
126 | zone_idx = offset >> f->zbd_info->zone_size_log2; |
127 | else | |
ee3696bd | 128 | zone_idx = offset / f->zbd_info->zone_size; |
bfbdd35b BVA |
129 | |
130 | return min(zone_idx, f->zbd_info->nr_zones); | |
131 | } | |
132 | ||
b7694961 DLM |
133 | /** |
134 | * zbd_zone_swr - Test whether a zone requires sequential writes | |
135 | * @z: zone info pointer. | |
136 | */ | |
137 | static inline bool zbd_zone_swr(struct fio_zone_info *z) | |
138 | { | |
139 | return z->type == ZBD_ZONE_TYPE_SWR; | |
140 | } | |
141 | ||
bfbdd35b BVA |
142 | /** |
143 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
144 | * @f: file pointer. | |
145 | * @z: zone info pointer. | |
146 | * @required: minimum number of bytes that must remain in a zone. | |
147 | * | |
148 | * The caller must hold z->mutex. | |
149 | */ | |
150 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
151 | uint64_t required) | |
152 | { | |
153 | assert((required & 511) == 0); | |
154 | ||
b7694961 | 155 | return zbd_zone_swr(z) && |
ee3696bd | 156 | z->wp + required > z->start + f->zbd_info->zone_size; |
bfbdd35b BVA |
157 | } |
158 | ||
fae3b9a0 | 159 | static void zone_lock(struct thread_data *td, struct fio_file *f, struct fio_zone_info *z) |
1f57803b | 160 | { |
fae3b9a0 AD |
161 | struct zoned_block_device_info *zbd = f->zbd_info; |
162 | uint32_t nz = z - zbd->zone_info; | |
163 | ||
164 | /* A thread should never lock zones outside its working area. */ | |
165 | assert(f->min_zone <= nz && nz < f->max_zone); | |
166 | ||
1f57803b DLM |
167 | /* |
168 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
169 | * is changed or when io_u completes and zbd_put_io() executed. | |
170 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
171 | * other waiting for zone locks when building an io_u batch, first | |
172 | * only trylock the zone. If the zone is already locked by another job, | |
173 | * process the currently queued I/Os so that I/O progress is made and | |
174 | * zones unlocked. | |
175 | */ | |
176 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
177 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
178 | io_u_quiesce(td); | |
179 | pthread_mutex_lock(&z->mutex); | |
180 | } | |
181 | } | |
182 | ||
bfbdd35b BVA |
183 | static bool is_valid_offset(const struct fio_file *f, uint64_t offset) |
184 | { | |
185 | return (uint64_t)(offset - f->file_offset) < f->io_size; | |
186 | } | |
187 | ||
188 | /* Verify whether direct I/O is used for all host-managed zoned drives. */ | |
189 | static bool zbd_using_direct_io(void) | |
190 | { | |
191 | struct thread_data *td; | |
192 | struct fio_file *f; | |
193 | int i, j; | |
194 | ||
195 | for_each_td(td, i) { | |
196 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) | |
197 | continue; | |
198 | for_each_file(td, f, j) { | |
199 | if (f->zbd_info && | |
b7694961 | 200 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
201 | return false; |
202 | } | |
203 | } | |
204 | ||
205 | return true; | |
206 | } | |
207 | ||
208 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
209 | static bool zbd_is_seq_job(struct fio_file *f) | |
210 | { | |
211 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
212 | ||
213 | assert(f->zbd_info); | |
214 | if (f->io_size == 0) | |
215 | return false; | |
216 | zone_idx_b = zbd_zone_idx(f, f->file_offset); | |
217 | zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1); | |
218 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) | |
b7694961 | 219 | if (zbd_zone_swr(&f->zbd_info->zone_info[zone_idx])) |
bfbdd35b BVA |
220 | return true; |
221 | ||
222 | return false; | |
223 | } | |
224 | ||
225 | /* | |
226 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
227 | */ | |
228 | static bool zbd_verify_sizes(void) | |
229 | { | |
230 | const struct fio_zone_info *z; | |
231 | struct thread_data *td; | |
232 | struct fio_file *f; | |
233 | uint64_t new_offset, new_end; | |
234 | uint32_t zone_idx; | |
235 | int i, j; | |
236 | ||
237 | for_each_td(td, i) { | |
238 | for_each_file(td, f, j) { | |
239 | if (!f->zbd_info) | |
240 | continue; | |
241 | if (f->file_offset >= f->real_file_size) | |
242 | continue; | |
243 | if (!zbd_is_seq_job(f)) | |
244 | continue; | |
4d37720a | 245 | |
fc52e449 DLM |
246 | if (!td->o.zone_size) { |
247 | td->o.zone_size = f->zbd_info->zone_size; | |
248 | if (!td->o.zone_size) { | |
249 | log_err("%s: invalid 0 zone size\n", | |
250 | f->file_name); | |
251 | return false; | |
252 | } | |
971d6a22 DLM |
253 | } else if (td->o.zone_size != f->zbd_info->zone_size) { |
254 | log_err("%s: job parameter zonesize %llu does not match disk zone size %llu.\n", | |
255 | f->file_name, (unsigned long long) td->o.zone_size, | |
256 | (unsigned long long) f->zbd_info->zone_size); | |
257 | return false; | |
fc52e449 DLM |
258 | } |
259 | ||
4d37720a DLM |
260 | if (td->o.zone_skip && |
261 | (td->o.zone_skip < td->o.zone_size || | |
262 | td->o.zone_skip % td->o.zone_size)) { | |
263 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", | |
264 | f->file_name, (unsigned long long) td->o.zone_skip, | |
265 | (unsigned long long) td->o.zone_size); | |
266 | return false; | |
267 | } | |
268 | ||
bfbdd35b BVA |
269 | zone_idx = zbd_zone_idx(f, f->file_offset); |
270 | z = &f->zbd_info->zone_info[zone_idx]; | |
f09a7773 PL |
271 | if ((f->file_offset != z->start) && |
272 | (td->o.td_ddir != TD_DDIR_READ)) { | |
ee3696bd | 273 | new_offset = (z+1)->start; |
bfbdd35b BVA |
274 | if (new_offset >= f->file_offset + f->io_size) { |
275 | log_info("%s: io_size must be at least one zone\n", | |
276 | f->file_name); | |
277 | return false; | |
278 | } | |
a0c84dd4 JA |
279 | log_info("%s: rounded up offset from %llu to %llu\n", |
280 | f->file_name, (unsigned long long) f->file_offset, | |
281 | (unsigned long long) new_offset); | |
bfbdd35b BVA |
282 | f->io_size -= (new_offset - f->file_offset); |
283 | f->file_offset = new_offset; | |
284 | } | |
285 | zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size); | |
286 | z = &f->zbd_info->zone_info[zone_idx]; | |
ee3696bd | 287 | new_end = z->start; |
f09a7773 PL |
288 | if ((td->o.td_ddir != TD_DDIR_READ) && |
289 | (f->file_offset + f->io_size != new_end)) { | |
bfbdd35b BVA |
290 | if (new_end <= f->file_offset) { |
291 | log_info("%s: io_size must be at least one zone\n", | |
292 | f->file_name); | |
293 | return false; | |
294 | } | |
a0c84dd4 JA |
295 | log_info("%s: rounded down io_size from %llu to %llu\n", |
296 | f->file_name, (unsigned long long) f->io_size, | |
297 | (unsigned long long) new_end - f->file_offset); | |
bfbdd35b BVA |
298 | f->io_size = new_end - f->file_offset; |
299 | } | |
fae3b9a0 AD |
300 | |
301 | f->min_zone = zbd_zone_idx(f, f->file_offset); | |
302 | f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size); | |
bfbdd35b BVA |
303 | } |
304 | } | |
305 | ||
306 | return true; | |
307 | } | |
308 | ||
309 | static bool zbd_verify_bs(void) | |
310 | { | |
311 | struct thread_data *td; | |
312 | struct fio_file *f; | |
313 | uint32_t zone_size; | |
314 | int i, j, k; | |
315 | ||
316 | for_each_td(td, i) { | |
317 | for_each_file(td, f, j) { | |
318 | if (!f->zbd_info) | |
319 | continue; | |
320 | zone_size = f->zbd_info->zone_size; | |
321 | for (k = 0; k < ARRAY_SIZE(td->o.bs); k++) { | |
322 | if (td->o.verify != VERIFY_NONE && | |
ee3696bd | 323 | zone_size % td->o.bs[k] != 0) { |
bfbdd35b BVA |
324 | log_info("%s: block size %llu is not a divisor of the zone size %d\n", |
325 | f->file_name, td->o.bs[k], | |
ee3696bd | 326 | zone_size); |
bfbdd35b BVA |
327 | return false; |
328 | } | |
329 | } | |
330 | } | |
331 | } | |
332 | return true; | |
333 | } | |
334 | ||
bfbdd35b BVA |
335 | static int ilog2(uint64_t i) |
336 | { | |
337 | int log = -1; | |
338 | ||
339 | while (i) { | |
340 | i >>= 1; | |
341 | log++; | |
342 | } | |
343 | return log; | |
344 | } | |
345 | ||
346 | /* | |
347 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
348 | * allows to execute a ZBD workload against a non-ZBD device. | |
349 | */ | |
350 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
351 | { | |
352 | uint32_t nr_zones; | |
353 | struct fio_zone_info *p; | |
a4b7f12b | 354 | uint64_t zone_size = td->o.zone_size; |
bfbdd35b BVA |
355 | struct zoned_block_device_info *zbd_info = NULL; |
356 | pthread_mutexattr_t attr; | |
357 | int i; | |
358 | ||
a4b7f12b DLM |
359 | if (zone_size == 0) { |
360 | log_err("%s: Specifying the zone size is mandatory for regular block devices with --zonemode=zbd\n\n", | |
361 | f->file_name); | |
362 | return 1; | |
363 | } | |
364 | ||
365 | if (zone_size < 512) { | |
366 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
367 | f->file_name); | |
368 | return 1; | |
369 | } | |
370 | ||
ee3696bd | 371 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
372 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
373 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
374 | if (!zbd_info) | |
375 | return -ENOMEM; | |
376 | ||
377 | pthread_mutexattr_init(&attr); | |
bfbdd35b BVA |
378 | pthread_mutexattr_setpshared(&attr, true); |
379 | pthread_mutex_init(&zbd_info->mutex, &attr); | |
1bb1bcad | 380 | pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE); |
bfbdd35b BVA |
381 | zbd_info->refcount = 1; |
382 | p = &zbd_info->zone_info[0]; | |
383 | for (i = 0; i < nr_zones; i++, p++) { | |
384 | pthread_mutex_init(&p->mutex, &attr); | |
385 | p->start = i * zone_size; | |
386 | p->wp = p->start + zone_size; | |
b7694961 DLM |
387 | p->type = ZBD_ZONE_TYPE_SWR; |
388 | p->cond = ZBD_ZONE_COND_EMPTY; | |
bfbdd35b BVA |
389 | } |
390 | /* a sentinel */ | |
391 | p->start = nr_zones * zone_size; | |
392 | ||
393 | f->zbd_info = zbd_info; | |
394 | f->zbd_info->zone_size = zone_size; | |
395 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 396 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
397 | f->zbd_info->nr_zones = nr_zones; |
398 | pthread_mutexattr_destroy(&attr); | |
399 | return 0; | |
400 | } | |
401 | ||
402 | /* | |
b7694961 DLM |
403 | * Maximum number of zones to report in one operation. |
404 | */ | |
405 | #define ZBD_REPORT_MAX_ZONES 8192U | |
406 | ||
407 | /* | |
408 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
409 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
410 | */ |
411 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
412 | { | |
b7694961 DLM |
413 | int nr_zones, nrz; |
414 | struct zbd_zone *zones, *z; | |
bfbdd35b | 415 | struct fio_zone_info *p; |
b7694961 | 416 | uint64_t zone_size, offset; |
bfbdd35b BVA |
417 | struct zoned_block_device_info *zbd_info = NULL; |
418 | pthread_mutexattr_t attr; | |
b7694961 | 419 | int i, j, ret = 0; |
bfbdd35b BVA |
420 | |
421 | pthread_mutexattr_init(&attr); | |
bfbdd35b BVA |
422 | pthread_mutexattr_setpshared(&attr, true); |
423 | ||
b7694961 DLM |
424 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
425 | if (!zones) | |
bfbdd35b BVA |
426 | goto out; |
427 | ||
b7694961 DLM |
428 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
429 | if (nrz < 0) { | |
430 | ret = nrz; | |
431 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
432 | f->file_name, -ret); | |
433 | goto out; | |
bfbdd35b BVA |
434 | } |
435 | ||
b7694961 | 436 | zone_size = zones[0].len; |
ee3696bd | 437 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
438 | |
439 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
440 | td->o.zone_size = zone_size; |
441 | } else if (td->o.zone_size != zone_size) { | |
4580ff92 DLM |
442 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %llu.\n", |
443 | f->file_name, (unsigned long long) td->o.zone_size, | |
a0c84dd4 | 444 | (unsigned long long) zone_size); |
bfbdd35b | 445 | ret = -EINVAL; |
b7694961 | 446 | goto out; |
bfbdd35b BVA |
447 | } |
448 | ||
a0c84dd4 JA |
449 | dprint(FD_ZBD, "Device %s has %d zones of size %llu KB\n", f->file_name, |
450 | nr_zones, (unsigned long long) zone_size / 1024); | |
bfbdd35b BVA |
451 | |
452 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
453 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
454 | ret = -ENOMEM; | |
455 | if (!zbd_info) | |
b7694961 | 456 | goto out; |
bfbdd35b | 457 | pthread_mutex_init(&zbd_info->mutex, &attr); |
1bb1bcad | 458 | pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE); |
bfbdd35b BVA |
459 | zbd_info->refcount = 1; |
460 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
461 | for (offset = 0, j = 0; j < nr_zones;) { |
462 | z = &zones[0]; | |
463 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
bfbdd35b | 464 | pthread_mutex_init(&p->mutex, &attr); |
b7694961 | 465 | p->start = z->start; |
bfbdd35b | 466 | switch (z->cond) { |
b7694961 DLM |
467 | case ZBD_ZONE_COND_NOT_WP: |
468 | case ZBD_ZONE_COND_FULL: | |
ee3696bd | 469 | p->wp = p->start + zone_size; |
bfbdd35b BVA |
470 | break; |
471 | default: | |
472 | assert(z->start <= z->wp); | |
b7694961 DLM |
473 | assert(z->wp <= z->start + zone_size); |
474 | p->wp = z->wp; | |
bfbdd35b BVA |
475 | break; |
476 | } | |
477 | p->type = z->type; | |
478 | p->cond = z->cond; | |
479 | if (j > 0 && p->start != p[-1].start + zone_size) { | |
480 | log_info("%s: invalid zone data\n", | |
481 | f->file_name); | |
482 | ret = -EINVAL; | |
b7694961 | 483 | goto out; |
bfbdd35b BVA |
484 | } |
485 | } | |
486 | z--; | |
b7694961 | 487 | offset = z->start + z->len; |
bfbdd35b BVA |
488 | if (j >= nr_zones) |
489 | break; | |
b7694961 DLM |
490 | nrz = zbd_report_zones(td, f, offset, |
491 | zones, ZBD_REPORT_MAX_ZONES); | |
492 | if (nrz < 0) { | |
493 | ret = nrz; | |
494 | log_info("fio: report zones (offset %llu) failed for %s (%d).\n", | |
495 | (unsigned long long)offset, | |
496 | f->file_name, -ret); | |
497 | goto out; | |
bfbdd35b BVA |
498 | } |
499 | } | |
b7694961 | 500 | |
bfbdd35b | 501 | /* a sentinel */ |
b7694961 | 502 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
503 | |
504 | f->zbd_info = zbd_info; | |
505 | f->zbd_info->zone_size = zone_size; | |
506 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 507 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
508 | f->zbd_info->nr_zones = nr_zones; |
509 | zbd_info = NULL; | |
510 | ret = 0; | |
511 | ||
bfbdd35b | 512 | out: |
b7694961 DLM |
513 | sfree(zbd_info); |
514 | free(zones); | |
bfbdd35b BVA |
515 | pthread_mutexattr_destroy(&attr); |
516 | return ret; | |
517 | } | |
518 | ||
519 | /* | |
520 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
521 | * | |
522 | * Returns 0 upon success and a negative error code upon failure. | |
523 | */ | |
379e5f09 | 524 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 525 | { |
b7694961 DLM |
526 | enum zbd_zoned_model zbd_model; |
527 | int ret; | |
bfbdd35b BVA |
528 | |
529 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
530 | ||
b7694961 DLM |
531 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
532 | if (ret) | |
533 | return ret; | |
534 | ||
bfbdd35b | 535 | switch (zbd_model) { |
b7694961 DLM |
536 | case ZBD_IGNORE: |
537 | return 0; | |
538 | case ZBD_HOST_AWARE: | |
539 | case ZBD_HOST_MANAGED: | |
bfbdd35b BVA |
540 | ret = parse_zone_info(td, f); |
541 | break; | |
b7694961 | 542 | case ZBD_NONE: |
bfbdd35b BVA |
543 | ret = init_zone_info(td, f); |
544 | break; | |
b7694961 DLM |
545 | default: |
546 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
547 | log_err("Unsupported zoned model\n"); | |
548 | return -EINVAL; | |
bfbdd35b | 549 | } |
b7694961 | 550 | |
219c662d | 551 | if (ret == 0) { |
bfbdd35b | 552 | f->zbd_info->model = zbd_model; |
219c662d AD |
553 | f->zbd_info->max_open_zones = td->o.max_open_zones; |
554 | } | |
bfbdd35b BVA |
555 | return ret; |
556 | } | |
557 | ||
558 | void zbd_free_zone_info(struct fio_file *f) | |
559 | { | |
560 | uint32_t refcount; | |
561 | ||
3c1dc34c | 562 | assert(f->zbd_info); |
bfbdd35b BVA |
563 | |
564 | pthread_mutex_lock(&f->zbd_info->mutex); | |
565 | refcount = --f->zbd_info->refcount; | |
566 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
567 | ||
568 | assert((int32_t)refcount >= 0); | |
569 | if (refcount == 0) | |
570 | sfree(f->zbd_info); | |
571 | f->zbd_info = NULL; | |
572 | } | |
573 | ||
574 | /* | |
575 | * Initialize f->zbd_info. | |
576 | * | |
577 | * Returns 0 upon success and a negative error code upon failure. | |
578 | * | |
579 | * Note: this function can only work correctly if it is called before the first | |
580 | * fio fork() call. | |
581 | */ | |
582 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
583 | { | |
584 | struct thread_data *td2; | |
585 | struct fio_file *f2; | |
586 | int i, j, ret; | |
587 | ||
588 | for_each_td(td2, i) { | |
589 | for_each_file(td2, f2, j) { | |
590 | if (td2 == td && f2 == file) | |
591 | continue; | |
592 | if (!f2->zbd_info || | |
593 | strcmp(f2->file_name, file->file_name) != 0) | |
594 | continue; | |
595 | file->zbd_info = f2->zbd_info; | |
596 | file->zbd_info->refcount++; | |
597 | return 0; | |
598 | } | |
599 | } | |
600 | ||
601 | ret = zbd_create_zone_info(td, file); | |
602 | if (ret < 0) | |
c5837eec | 603 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
bfbdd35b BVA |
604 | return ret; |
605 | } | |
606 | ||
3c1dc34c | 607 | int zbd_setup_files(struct thread_data *td) |
bfbdd35b BVA |
608 | { |
609 | struct fio_file *f; | |
610 | int i; | |
611 | ||
612 | for_each_file(td, f, i) { | |
a4b7f12b | 613 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 614 | return 1; |
bfbdd35b BVA |
615 | } |
616 | ||
617 | if (!zbd_using_direct_io()) { | |
618 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
619 | return 1; | |
620 | } | |
621 | ||
622 | if (!zbd_verify_sizes()) | |
623 | return 1; | |
624 | ||
625 | if (!zbd_verify_bs()) | |
626 | return 1; | |
627 | ||
219c662d AD |
628 | for_each_file(td, f, i) { |
629 | struct zoned_block_device_info *zbd = f->zbd_info; | |
630 | ||
631 | if (!zbd) | |
632 | continue; | |
633 | ||
634 | zbd->max_open_zones = zbd->max_open_zones ?: ZBD_MAX_OPEN_ZONES; | |
635 | ||
636 | if (td->o.max_open_zones > 0 && | |
637 | zbd->max_open_zones != td->o.max_open_zones) { | |
638 | log_err("Different 'max_open_zones' values\n"); | |
639 | return 1; | |
640 | } | |
641 | if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) { | |
642 | log_err("'max_open_zones' value is limited by %u\n", ZBD_MAX_OPEN_ZONES); | |
643 | return 1; | |
644 | } | |
645 | } | |
646 | ||
bfbdd35b BVA |
647 | return 0; |
648 | } | |
649 | ||
650 | /** | |
651 | * zbd_reset_range - reset zones for a range of sectors | |
652 | * @td: FIO thread data. | |
653 | * @f: Fio file for which to reset zones | |
654 | * @sector: Starting sector in units of 512 bytes | |
655 | * @nr_sectors: Number of sectors in units of 512 bytes | |
656 | * | |
657 | * Returns 0 upon success and a negative error code upon failure. | |
658 | */ | |
b7694961 | 659 | static int zbd_reset_range(struct thread_data *td, struct fio_file *f, |
ee3696bd | 660 | uint64_t offset, uint64_t length) |
bfbdd35b | 661 | { |
bfbdd35b BVA |
662 | uint32_t zone_idx_b, zone_idx_e; |
663 | struct fio_zone_info *zb, *ze, *z; | |
664 | int ret = 0; | |
665 | ||
ee3696bd | 666 | assert(is_valid_offset(f, offset + length - 1)); |
b7694961 | 667 | |
bfbdd35b | 668 | switch (f->zbd_info->model) { |
b7694961 DLM |
669 | case ZBD_HOST_AWARE: |
670 | case ZBD_HOST_MANAGED: | |
671 | ret = zbd_reset_wp(td, f, offset, length); | |
672 | if (ret < 0) | |
bfbdd35b | 673 | return ret; |
bfbdd35b | 674 | break; |
b7694961 | 675 | default: |
bfbdd35b BVA |
676 | break; |
677 | } | |
678 | ||
ee3696bd | 679 | zone_idx_b = zbd_zone_idx(f, offset); |
bfbdd35b | 680 | zb = &f->zbd_info->zone_info[zone_idx_b]; |
ee3696bd | 681 | zone_idx_e = zbd_zone_idx(f, offset + length); |
bfbdd35b BVA |
682 | ze = &f->zbd_info->zone_info[zone_idx_e]; |
683 | for (z = zb; z < ze; z++) { | |
684 | pthread_mutex_lock(&z->mutex); | |
a7c2b6fc BVA |
685 | pthread_mutex_lock(&f->zbd_info->mutex); |
686 | f->zbd_info->sectors_with_data -= z->wp - z->start; | |
687 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
bfbdd35b BVA |
688 | z->wp = z->start; |
689 | z->verify_block = 0; | |
690 | pthread_mutex_unlock(&z->mutex); | |
691 | } | |
692 | ||
fd5d733f BVA |
693 | td->ts.nr_zone_resets += ze - zb; |
694 | ||
bfbdd35b BVA |
695 | return ret; |
696 | } | |
697 | ||
a0c84dd4 JA |
698 | static unsigned int zbd_zone_nr(struct zoned_block_device_info *zbd_info, |
699 | struct fio_zone_info *zone) | |
700 | { | |
089ddd95 | 701 | return zone - zbd_info->zone_info; |
a0c84dd4 JA |
702 | } |
703 | ||
bfbdd35b BVA |
704 | /** |
705 | * zbd_reset_zone - reset the write pointer of a single zone | |
706 | * @td: FIO thread data. | |
707 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
708 | * @z: Zone to reset. | |
709 | * | |
710 | * Returns 0 upon success and a negative error code upon failure. | |
711 | */ | |
b7694961 | 712 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, |
bfbdd35b BVA |
713 | struct fio_zone_info *z) |
714 | { | |
a0c84dd4 JA |
715 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name, |
716 | zbd_zone_nr(f->zbd_info, z)); | |
d60be7d5 DLM |
717 | |
718 | return zbd_reset_range(td, f, z->start, (z+1)->start - z->start); | |
bfbdd35b BVA |
719 | } |
720 | ||
25dc6606 AD |
721 | /* The caller must hold f->zbd_info->mutex */ |
722 | static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, | |
723 | unsigned int open_zone_idx) | |
724 | { | |
725 | uint32_t zone_idx; | |
726 | ||
727 | assert(open_zone_idx < f->zbd_info->num_open_zones); | |
728 | zone_idx = f->zbd_info->open_zones[open_zone_idx]; | |
729 | memmove(f->zbd_info->open_zones + open_zone_idx, | |
730 | f->zbd_info->open_zones + open_zone_idx + 1, | |
731 | (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) * | |
732 | sizeof(f->zbd_info->open_zones[0])); | |
733 | f->zbd_info->num_open_zones--; | |
219c662d | 734 | td->num_open_zones--; |
25dc6606 AD |
735 | f->zbd_info->zone_info[zone_idx].open = 0; |
736 | } | |
737 | ||
bfbdd35b BVA |
738 | /* |
739 | * Reset a range of zones. Returns 0 upon success and 1 upon failure. | |
740 | * @td: fio thread data. | |
741 | * @f: fio file for which to reset zones | |
742 | * @zb: first zone to reset. | |
743 | * @ze: first zone not to reset. | |
744 | * @all_zones: whether to reset all zones or only those zones for which the | |
745 | * write pointer is not a multiple of td->o.min_bs[DDIR_WRITE]. | |
746 | */ | |
747 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
748 | struct fio_zone_info *const zb, | |
749 | struct fio_zone_info *const ze, bool all_zones) | |
750 | { | |
4803b841 | 751 | struct fio_zone_info *z; |
ee3696bd | 752 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; |
bfbdd35b BVA |
753 | bool reset_wp; |
754 | int res = 0; | |
755 | ||
adc6adcb DLM |
756 | assert(min_bs); |
757 | ||
a0c84dd4 JA |
758 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name, |
759 | zbd_zone_nr(f->zbd_info, zb), zbd_zone_nr(f->zbd_info, ze)); | |
bfbdd35b | 760 | for (z = zb; z < ze; z++) { |
25dc6606 AD |
761 | uint32_t nz = z - f->zbd_info->zone_info; |
762 | ||
b7694961 | 763 | if (!zbd_zone_swr(z)) |
1f57803b | 764 | continue; |
fae3b9a0 | 765 | zone_lock(td, f, z); |
25dc6606 AD |
766 | if (all_zones) { |
767 | unsigned int i; | |
768 | ||
769 | pthread_mutex_lock(&f->zbd_info->mutex); | |
770 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
771 | if (f->zbd_info->open_zones[i] == nz) | |
772 | zbd_close_zone(td, f, i); | |
773 | } | |
774 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
775 | ||
776 | reset_wp = z->wp != z->start; | |
777 | } else { | |
767d1372 | 778 | reset_wp = z->wp % min_bs != 0; |
25dc6606 | 779 | } |
1f57803b DLM |
780 | if (reset_wp) { |
781 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
782 | f->file_name, | |
783 | zbd_zone_nr(f->zbd_info, z)); | |
784 | if (zbd_reset_zone(td, f, z) < 0) | |
785 | res = 1; | |
bfbdd35b | 786 | } |
bfbdd35b | 787 | pthread_mutex_unlock(&z->mutex); |
4803b841 | 788 | } |
bfbdd35b BVA |
789 | |
790 | return res; | |
791 | } | |
792 | ||
a7c2b6fc BVA |
793 | /* |
794 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
795 | * zone reset. | |
796 | */ | |
1bb1bcad AD |
797 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
798 | const struct fio_file *f) | |
a7c2b6fc BVA |
799 | { |
800 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
801 | ||
a7c2b6fc BVA |
802 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
803 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
804 | } |
805 | ||
806 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
807 | const struct fio_file *f) | |
808 | { | |
809 | pthread_mutex_lock(&f->zbd_info->mutex); | |
810 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
811 | pthread_mutex_unlock(&f->zbd_info->mutex); |
812 | } | |
813 | ||
814 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
815 | const struct fio_file *f) | |
816 | { | |
817 | uint32_t write_cnt = 0; | |
818 | ||
819 | pthread_mutex_lock(&f->zbd_info->mutex); | |
820 | assert(f->zbd_info->write_cnt); | |
821 | if (f->zbd_info->write_cnt) | |
822 | write_cnt = --f->zbd_info->write_cnt; | |
823 | if (write_cnt == 0) | |
1bb1bcad | 824 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
825 | pthread_mutex_unlock(&f->zbd_info->mutex); |
826 | ||
827 | return write_cnt == 0; | |
828 | } | |
829 | ||
91d25131 BVA |
830 | enum swd_action { |
831 | CHECK_SWD, | |
832 | SET_SWD, | |
833 | }; | |
834 | ||
835 | /* Calculate the number of sectors with data (swd) and perform action 'a' */ | |
836 | static uint64_t zbd_process_swd(const struct fio_file *f, enum swd_action a) | |
615555bb | 837 | { |
615555bb BVA |
838 | struct fio_zone_info *zb, *ze, *z; |
839 | uint64_t swd = 0; | |
840 | ||
841 | zb = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)]; | |
842 | ze = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset + | |
843 | f->io_size)]; | |
844 | for (z = zb; z < ze; z++) { | |
845 | pthread_mutex_lock(&z->mutex); | |
846 | swd += z->wp - z->start; | |
847 | } | |
848 | pthread_mutex_lock(&f->zbd_info->mutex); | |
91d25131 BVA |
849 | switch (a) { |
850 | case CHECK_SWD: | |
851 | assert(f->zbd_info->sectors_with_data == swd); | |
852 | break; | |
853 | case SET_SWD: | |
854 | f->zbd_info->sectors_with_data = swd; | |
855 | break; | |
856 | } | |
615555bb BVA |
857 | pthread_mutex_unlock(&f->zbd_info->mutex); |
858 | for (z = zb; z < ze; z++) | |
859 | pthread_mutex_unlock(&z->mutex); | |
91d25131 BVA |
860 | |
861 | return swd; | |
862 | } | |
863 | ||
864 | /* | |
865 | * The swd check is useful for debugging but takes too much time to leave | |
866 | * it enabled all the time. Hence it is disabled by default. | |
867 | */ | |
868 | static const bool enable_check_swd = false; | |
869 | ||
870 | /* Check whether the value of zbd_info.sectors_with_data is correct. */ | |
871 | static void zbd_check_swd(const struct fio_file *f) | |
872 | { | |
873 | if (!enable_check_swd) | |
874 | return; | |
875 | ||
876 | zbd_process_swd(f, CHECK_SWD); | |
877 | } | |
878 | ||
879 | static void zbd_init_swd(struct fio_file *f) | |
880 | { | |
881 | uint64_t swd; | |
882 | ||
409a4f29 NA |
883 | if (!enable_check_swd) |
884 | return; | |
885 | ||
91d25131 BVA |
886 | swd = zbd_process_swd(f, SET_SWD); |
887 | dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name, | |
888 | swd); | |
615555bb BVA |
889 | } |
890 | ||
bfbdd35b BVA |
891 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
892 | { | |
91d25131 | 893 | struct fio_zone_info *zb, *ze; |
bfbdd35b | 894 | |
767d1372 | 895 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
896 | return; |
897 | ||
fae3b9a0 AD |
898 | zb = &f->zbd_info->zone_info[f->min_zone]; |
899 | ze = &f->zbd_info->zone_info[f->max_zone]; | |
91d25131 | 900 | zbd_init_swd(f); |
bfbdd35b BVA |
901 | /* |
902 | * If data verification is enabled reset the affected zones before | |
903 | * writing any data to avoid that a zone reset has to be issued while | |
904 | * writing data, which causes data loss. | |
905 | */ | |
906 | zbd_reset_zones(td, f, zb, ze, td->o.verify != VERIFY_NONE && | |
bfbdd35b | 907 | td->runstate != TD_VERIFYING); |
a7c2b6fc | 908 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
909 | } |
910 | ||
59b07544 BVA |
911 | /* The caller must hold f->zbd_info->mutex. */ |
912 | static bool is_zone_open(const struct thread_data *td, const struct fio_file *f, | |
913 | unsigned int zone_idx) | |
914 | { | |
915 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
916 | int i; | |
917 | ||
219c662d AD |
918 | assert(td->o.job_max_open_zones == 0 || td->num_open_zones <= td->o.job_max_open_zones); |
919 | assert(td->o.job_max_open_zones <= zbdi->max_open_zones); | |
920 | assert(zbdi->num_open_zones <= zbdi->max_open_zones); | |
59b07544 BVA |
921 | |
922 | for (i = 0; i < zbdi->num_open_zones; i++) | |
923 | if (zbdi->open_zones[i] == zone_idx) | |
924 | return true; | |
925 | ||
926 | return false; | |
927 | } | |
928 | ||
929 | /* | |
930 | * Open a ZBD zone if it was not yet open. Returns true if either the zone was | |
931 | * already open or if opening a new zone is allowed. Returns false if the zone | |
932 | * was not yet open and opening a new zone would cause the zone limit to be | |
933 | * exceeded. | |
934 | */ | |
935 | static bool zbd_open_zone(struct thread_data *td, const struct io_u *io_u, | |
936 | uint32_t zone_idx) | |
937 | { | |
938 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
939 | const struct fio_file *f = io_u->file; | |
940 | struct fio_zone_info *z = &f->zbd_info->zone_info[zone_idx]; | |
941 | bool res = true; | |
942 | ||
b7694961 | 943 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
59b07544 BVA |
944 | return false; |
945 | ||
946 | /* | |
947 | * Skip full zones with data verification enabled because resetting a | |
948 | * zone causes data loss and hence causes verification to fail. | |
949 | */ | |
950 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) | |
951 | return false; | |
952 | ||
59b07544 BVA |
953 | pthread_mutex_lock(&f->zbd_info->mutex); |
954 | if (is_zone_open(td, f, zone_idx)) | |
955 | goto out; | |
956 | res = false; | |
219c662d AD |
957 | /* Zero means no limit */ |
958 | if (td->o.job_max_open_zones > 0 && | |
959 | td->num_open_zones >= td->o.job_max_open_zones) | |
960 | goto out; | |
961 | if (f->zbd_info->num_open_zones >= f->zbd_info->max_open_zones) | |
59b07544 BVA |
962 | goto out; |
963 | dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx); | |
964 | f->zbd_info->open_zones[f->zbd_info->num_open_zones++] = zone_idx; | |
219c662d | 965 | td->num_open_zones++; |
59b07544 BVA |
966 | z->open = 1; |
967 | res = true; | |
968 | ||
969 | out: | |
970 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
971 | return res; | |
972 | } | |
973 | ||
6463db6c AD |
974 | /* Anything goes as long as it is not a constant. */ |
975 | static uint32_t pick_random_zone_idx(const struct fio_file *f, | |
976 | const struct io_u *io_u) | |
977 | { | |
978 | return io_u->offset * f->zbd_info->num_open_zones / f->real_file_size; | |
979 | } | |
980 | ||
59b07544 BVA |
981 | /* |
982 | * Modify the offset of an I/O unit that does not refer to an open zone such | |
983 | * that it refers to an open zone. Close an open zone and open a new zone if | |
984 | * necessary. This algorithm can only work correctly if all write pointers are | |
985 | * a multiple of the fio block size. The caller must neither hold z->mutex | |
986 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
987 | */ | |
379e5f09 BVA |
988 | static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, |
989 | struct io_u *io_u) | |
59b07544 BVA |
990 | { |
991 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 992 | struct fio_file *f = io_u->file; |
59b07544 BVA |
993 | struct fio_zone_info *z; |
994 | unsigned int open_zone_idx = -1; | |
995 | uint32_t zone_idx, new_zone_idx; | |
996 | int i; | |
997 | ||
998 | assert(is_valid_offset(f, io_u->offset)); | |
999 | ||
219c662d | 1000 | if (td->o.job_max_open_zones) { |
59b07544 BVA |
1001 | /* |
1002 | * This statement accesses f->zbd_info->open_zones[] on purpose | |
1003 | * without locking. | |
1004 | */ | |
6463db6c | 1005 | zone_idx = f->zbd_info->open_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 BVA |
1006 | } else { |
1007 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
1008 | } | |
fae3b9a0 AD |
1009 | if (zone_idx < f->min_zone) |
1010 | zone_idx = f->min_zone; | |
1011 | else if (zone_idx >= f->max_zone) | |
1012 | zone_idx = f->max_zone - 1; | |
59b07544 BVA |
1013 | dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", |
1014 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); | |
1015 | ||
1016 | /* | |
1017 | * Since z->mutex is the outer lock and f->zbd_info->mutex the inner | |
1018 | * lock it can happen that the state of the zone with index zone_idx | |
1019 | * has changed after 'z' has been assigned and before f->zbd_info->mutex | |
1020 | * has been obtained. Hence the loop. | |
1021 | */ | |
1022 | for (;;) { | |
6463db6c AD |
1023 | uint32_t tmp_idx; |
1024 | ||
59b07544 BVA |
1025 | z = &f->zbd_info->zone_info[zone_idx]; |
1026 | ||
fae3b9a0 | 1027 | zone_lock(td, f, z); |
59b07544 | 1028 | pthread_mutex_lock(&f->zbd_info->mutex); |
219c662d | 1029 | if (td->o.job_max_open_zones == 0) |
59b07544 BVA |
1030 | goto examine_zone; |
1031 | if (f->zbd_info->num_open_zones == 0) { | |
1032 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1033 | pthread_mutex_unlock(&z->mutex); | |
1034 | dprint(FD_ZBD, "%s(%s): no zones are open\n", | |
1035 | __func__, f->file_name); | |
1036 | return NULL; | |
1037 | } | |
6463db6c AD |
1038 | |
1039 | /* | |
1040 | * List of opened zones is per-device, shared across all threads. | |
1041 | * Start with quasi-random candidate zone. | |
1042 | * Ignore zones which don't belong to thread's offset/size area. | |
1043 | */ | |
1044 | open_zone_idx = pick_random_zone_idx(f, io_u); | |
59b07544 | 1045 | assert(open_zone_idx < f->zbd_info->num_open_zones); |
6463db6c AD |
1046 | tmp_idx = open_zone_idx; |
1047 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1048 | uint32_t tmpz; | |
1049 | ||
1050 | if (tmp_idx >= f->zbd_info->num_open_zones) | |
1051 | tmp_idx = 0; | |
1052 | tmpz = f->zbd_info->open_zones[tmp_idx]; | |
fae3b9a0 | 1053 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
6463db6c AD |
1054 | open_zone_idx = tmp_idx; |
1055 | goto found_candidate_zone; | |
1056 | } | |
1057 | ||
1058 | tmp_idx++; | |
1059 | } | |
1060 | ||
1061 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1062 | __func__, f->file_name); | |
adc6adcb DLM |
1063 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1064 | pthread_mutex_unlock(&z->mutex); | |
6463db6c AD |
1065 | return NULL; |
1066 | ||
1067 | found_candidate_zone: | |
59b07544 BVA |
1068 | new_zone_idx = f->zbd_info->open_zones[open_zone_idx]; |
1069 | if (new_zone_idx == zone_idx) | |
1070 | break; | |
1071 | zone_idx = new_zone_idx; | |
1072 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1073 | pthread_mutex_unlock(&z->mutex); | |
1074 | } | |
1075 | ||
1076 | /* Both z->mutex and f->zbd_info->mutex are held. */ | |
1077 | ||
1078 | examine_zone: | |
ee3696bd | 1079 | if (z->wp + min_bs <= (z+1)->start) { |
59b07544 BVA |
1080 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1081 | goto out; | |
1082 | } | |
1083 | dprint(FD_ZBD, "%s(%s): closing zone %d\n", __func__, f->file_name, | |
1084 | zone_idx); | |
219c662d | 1085 | if (td->o.job_max_open_zones) |
59b07544 BVA |
1086 | zbd_close_zone(td, f, open_zone_idx); |
1087 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1088 | ||
1089 | /* Only z->mutex is held. */ | |
1090 | ||
1091 | /* Zone 'z' is full, so try to open a new zone. */ | |
1092 | for (i = f->io_size / f->zbd_info->zone_size; i > 0; i--) { | |
1093 | zone_idx++; | |
1094 | pthread_mutex_unlock(&z->mutex); | |
1095 | z++; | |
ee3696bd | 1096 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1097 | /* Wrap-around. */ |
fae3b9a0 | 1098 | zone_idx = f->min_zone; |
59b07544 BVA |
1099 | z = &f->zbd_info->zone_info[zone_idx]; |
1100 | } | |
ee3696bd | 1101 | assert(is_valid_offset(f, z->start)); |
fae3b9a0 | 1102 | zone_lock(td, f, z); |
59b07544 BVA |
1103 | if (z->open) |
1104 | continue; | |
1105 | if (zbd_open_zone(td, io_u, zone_idx)) | |
1106 | goto out; | |
1107 | } | |
1108 | ||
1109 | /* Only z->mutex is held. */ | |
1110 | ||
1111 | /* Check whether the write fits in any of the already opened zones. */ | |
1112 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1113 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1114 | zone_idx = f->zbd_info->open_zones[i]; | |
fae3b9a0 AD |
1115 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1116 | continue; | |
59b07544 BVA |
1117 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1118 | pthread_mutex_unlock(&z->mutex); | |
1119 | ||
1120 | z = &f->zbd_info->zone_info[zone_idx]; | |
1121 | ||
fae3b9a0 | 1122 | zone_lock(td, f, z); |
ee3696bd | 1123 | if (z->wp + min_bs <= (z+1)->start) |
59b07544 BVA |
1124 | goto out; |
1125 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1126 | } | |
1127 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1128 | pthread_mutex_unlock(&z->mutex); | |
1129 | dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__, | |
1130 | f->file_name); | |
1131 | return NULL; | |
1132 | ||
1133 | out: | |
1134 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name, | |
1135 | zone_idx); | |
ee3696bd | 1136 | io_u->offset = z->start; |
59b07544 BVA |
1137 | return z; |
1138 | } | |
1139 | ||
bfbdd35b | 1140 | /* The caller must hold z->mutex. */ |
59b07544 BVA |
1141 | static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, |
1142 | struct io_u *io_u, | |
1143 | struct fio_zone_info *z) | |
bfbdd35b BVA |
1144 | { |
1145 | const struct fio_file *f = io_u->file; | |
1146 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
1147 | ||
59b07544 BVA |
1148 | if (!zbd_open_zone(td, io_u, z - f->zbd_info->zone_info)) { |
1149 | pthread_mutex_unlock(&z->mutex); | |
1150 | z = zbd_convert_to_open_zone(td, io_u); | |
1151 | assert(z); | |
1152 | } | |
1153 | ||
bfbdd35b | 1154 | if (z->verify_block * min_bs >= f->zbd_info->zone_size) |
a0c84dd4 JA |
1155 | log_err("%s: %d * %d >= %llu\n", f->file_name, z->verify_block, |
1156 | min_bs, (unsigned long long) f->zbd_info->zone_size); | |
ee3696bd | 1157 | io_u->offset = z->start + z->verify_block++ * min_bs; |
59b07544 | 1158 | return z; |
bfbdd35b BVA |
1159 | } |
1160 | ||
1161 | /* | |
1162 | * Find another zone for which @io_u fits below the write pointer. Start | |
1163 | * searching in zones @zb + 1 .. @zl and continue searching in zones | |
1164 | * @zf .. @zb - 1. | |
1165 | * | |
1166 | * Either returns NULL or returns a zone pointer and holds the mutex for that | |
1167 | * zone. | |
1168 | */ | |
1169 | static struct fio_zone_info * | |
1170 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, | |
1171 | struct fio_zone_info *zb, struct fio_zone_info *zl) | |
1172 | { | |
1173 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 1174 | struct fio_file *f = io_u->file; |
bfbdd35b BVA |
1175 | struct fio_zone_info *z1, *z2; |
1176 | const struct fio_zone_info *const zf = | |
1177 | &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)]; | |
1178 | ||
1179 | /* | |
1180 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1181 | * the nearest non-empty zone in case of random I/O. | |
1182 | */ | |
1183 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1184 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
fae3b9a0 | 1185 | zone_lock(td, f, z1); |
ee3696bd | 1186 | if (z1->start + min_bs <= z1->wp) |
bfbdd35b BVA |
1187 | return z1; |
1188 | pthread_mutex_unlock(&z1->mutex); | |
1189 | } else if (!td_random(td)) { | |
1190 | break; | |
1191 | } | |
1192 | if (td_random(td) && z2 >= zf && | |
b7694961 | 1193 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
fae3b9a0 | 1194 | zone_lock(td, f, z2); |
ee3696bd | 1195 | if (z2->start + min_bs <= z2->wp) |
bfbdd35b BVA |
1196 | return z2; |
1197 | pthread_mutex_unlock(&z2->mutex); | |
1198 | } | |
1199 | } | |
1200 | dprint(FD_ZBD, "%s: adjusting random read offset failed\n", | |
1201 | f->file_name); | |
1202 | return NULL; | |
1203 | } | |
1204 | ||
bfbdd35b | 1205 | /** |
d9ed3e63 | 1206 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1207 | * @io_u: I/O unit |
d9ed3e63 DLM |
1208 | * @success: Whether or not the I/O unit has been queued successfully |
1209 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1210 | * |
d9ed3e63 DLM |
1211 | * For write and trim operations, update the write pointer of the I/O unit |
1212 | * target zone. | |
bfbdd35b | 1213 | */ |
d9ed3e63 | 1214 | static void zbd_queue_io(struct io_u *io_u, int q, bool success) |
bfbdd35b | 1215 | { |
d9ed3e63 DLM |
1216 | const struct fio_file *f = io_u->file; |
1217 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b BVA |
1218 | struct fio_zone_info *z; |
1219 | uint32_t zone_idx; | |
d9ed3e63 | 1220 | uint64_t zone_end; |
bfbdd35b | 1221 | |
bfbdd35b BVA |
1222 | if (!zbd_info) |
1223 | return; | |
1224 | ||
d9ed3e63 | 1225 | zone_idx = zbd_zone_idx(f, io_u->offset); |
bfbdd35b | 1226 | assert(zone_idx < zbd_info->nr_zones); |
d9ed3e63 DLM |
1227 | z = &zbd_info->zone_info[zone_idx]; |
1228 | ||
b7694961 | 1229 | if (!zbd_zone_swr(z)) |
bfbdd35b | 1230 | return; |
d9ed3e63 | 1231 | |
bfbdd35b BVA |
1232 | if (!success) |
1233 | goto unlock; | |
d9ed3e63 DLM |
1234 | |
1235 | dprint(FD_ZBD, | |
1236 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
1237 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1238 | ||
bfbdd35b BVA |
1239 | switch (io_u->ddir) { |
1240 | case DDIR_WRITE: | |
d9ed3e63 DLM |
1241 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
1242 | (z + 1)->start); | |
a7c2b6fc BVA |
1243 | pthread_mutex_lock(&zbd_info->mutex); |
1244 | /* | |
1245 | * z->wp > zone_end means that one or more I/O errors | |
1246 | * have occurred. | |
1247 | */ | |
1248 | if (z->wp <= zone_end) | |
1249 | zbd_info->sectors_with_data += zone_end - z->wp; | |
1250 | pthread_mutex_unlock(&zbd_info->mutex); | |
bfbdd35b BVA |
1251 | z->wp = zone_end; |
1252 | break; | |
1253 | case DDIR_TRIM: | |
1254 | assert(z->wp == z->start); | |
1255 | break; | |
1256 | default: | |
1257 | break; | |
1258 | } | |
d9ed3e63 | 1259 | |
bfbdd35b | 1260 | unlock: |
d9ed3e63 DLM |
1261 | if (!success || q != FIO_Q_QUEUED) { |
1262 | /* BUSY or COMPLETED: unlock the zone */ | |
1263 | pthread_mutex_unlock(&z->mutex); | |
1264 | io_u->zbd_put_io = NULL; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | /** | |
1269 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1270 | * @io_u: I/O unit | |
1271 | */ | |
1272 | static void zbd_put_io(const struct io_u *io_u) | |
1273 | { | |
1274 | const struct fio_file *f = io_u->file; | |
1275 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1276 | struct fio_zone_info *z; | |
1277 | uint32_t zone_idx; | |
f59c2f8e | 1278 | int ret; |
d9ed3e63 DLM |
1279 | |
1280 | if (!zbd_info) | |
1281 | return; | |
615555bb | 1282 | |
d9ed3e63 DLM |
1283 | zone_idx = zbd_zone_idx(f, io_u->offset); |
1284 | assert(zone_idx < zbd_info->nr_zones); | |
1285 | z = &zbd_info->zone_info[zone_idx]; | |
1286 | ||
b7694961 | 1287 | if (!zbd_zone_swr(z)) |
d9ed3e63 DLM |
1288 | return; |
1289 | ||
1290 | dprint(FD_ZBD, | |
1291 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
1292 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1293 | ||
f59c2f8e AD |
1294 | ret = pthread_mutex_unlock(&z->mutex); |
1295 | assert(ret == 0); | |
d9ed3e63 | 1296 | zbd_check_swd(f); |
bfbdd35b BVA |
1297 | } |
1298 | ||
9d87c646 DLM |
1299 | /* |
1300 | * Windows and MacOS do not define this. | |
1301 | */ | |
1302 | #ifndef EREMOTEIO | |
1303 | #define EREMOTEIO 121 /* POSIX value */ | |
1304 | #endif | |
1305 | ||
bfbdd35b BVA |
1306 | bool zbd_unaligned_write(int error_code) |
1307 | { | |
1308 | switch (error_code) { | |
1309 | case EIO: | |
1310 | case EREMOTEIO: | |
1311 | return true; | |
1312 | } | |
1313 | return false; | |
1314 | } | |
1315 | ||
4d37720a DLM |
1316 | /** |
1317 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1318 | * @td: FIO thread data. | |
1319 | * @io_u: FIO I/O unit. | |
1320 | * | |
1321 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1322 | * no more IO can be performed in the current zone. | |
1323 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1324 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1325 | * - For write workloads, zoneskip is applied when the zone is full. | |
1326 | * This applies only to read and write operations. | |
1327 | */ | |
1328 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1329 | { | |
1330 | struct fio_file *f = io_u->file; | |
1331 | enum fio_ddir ddir = io_u->ddir; | |
1332 | struct fio_zone_info *z; | |
1333 | uint32_t zone_idx; | |
1334 | ||
1335 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1336 | assert(td->o.zone_size); | |
1337 | ||
1338 | /* | |
1339 | * zone_skip is valid only for sequential workloads. | |
1340 | */ | |
1341 | if (td_random(td) || !td->o.zone_skip) | |
1342 | return; | |
1343 | ||
1344 | /* | |
1345 | * It is time to switch to a new zone if: | |
1346 | * - zone_bytes == zone_size bytes have already been accessed | |
1347 | * - The last position reached the end of the current zone. | |
1348 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1349 | * reached the zone write pointer. | |
1350 | */ | |
1351 | zone_idx = zbd_zone_idx(f, f->last_pos[ddir]); | |
1352 | z = &f->zbd_info->zone_info[zone_idx]; | |
1353 | ||
1354 | if (td->zone_bytes >= td->o.zone_size || | |
1355 | f->last_pos[ddir] >= (z+1)->start || | |
1356 | (ddir == DDIR_READ && | |
1357 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1358 | /* | |
1359 | * Skip zones. | |
1360 | */ | |
1361 | td->zone_bytes = 0; | |
1362 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1363 | ||
1364 | /* | |
1365 | * Wrap from the beginning, if we exceed the file size | |
1366 | */ | |
1367 | if (f->file_offset >= f->real_file_size) | |
1368 | f->file_offset = get_start_offset(td, f); | |
1369 | ||
1370 | f->last_pos[ddir] = f->file_offset; | |
1371 | td->io_skip_bytes += td->o.zone_skip; | |
1372 | } | |
1373 | } | |
1374 | ||
c65057f9 SK |
1375 | /** |
1376 | * zbd_adjust_ddir - Adjust an I/O direction for zonedmode=zbd. | |
1377 | * | |
1378 | * @td: FIO thread data. | |
1379 | * @io_u: FIO I/O unit. | |
1380 | * @ddir: I/O direction before adjustment. | |
1381 | * | |
1382 | * Return adjusted I/O direction. | |
1383 | */ | |
1384 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1385 | enum fio_ddir ddir) | |
1386 | { | |
1387 | /* | |
1388 | * In case read direction is chosen for the first random I/O, fio with | |
1389 | * zonemode=zbd stops because no data can be read from zoned block | |
1390 | * devices with all empty zones. Overwrite the first I/O direction as | |
1391 | * write to make sure data to read exists. | |
1392 | */ | |
731461cc | 1393 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1394 | return ddir; |
1395 | ||
1396 | if (io_u->file->zbd_info->sectors_with_data || | |
1397 | td->o.read_beyond_wp) | |
1398 | return DDIR_READ; | |
1399 | ||
1400 | return DDIR_WRITE; | |
1401 | } | |
1402 | ||
bfbdd35b BVA |
1403 | /** |
1404 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1405 | * @td: FIO thread data. | |
1406 | * @io_u: FIO I/O unit. | |
1407 | * | |
1408 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1409 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1410 | * corresponds to io_u->offset at the end of this function. | |
1411 | */ | |
1412 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1413 | { | |
b7694961 | 1414 | struct fio_file *f = io_u->file; |
bfbdd35b | 1415 | uint32_t zone_idx_b; |
de65f7b7 | 1416 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b BVA |
1417 | uint32_t orig_len = io_u->buflen; |
1418 | uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
1419 | uint64_t new_len; | |
1420 | int64_t range; | |
1421 | ||
1422 | if (!f->zbd_info) | |
1423 | return io_u_accept; | |
1424 | ||
adc6adcb | 1425 | assert(min_bs); |
bfbdd35b BVA |
1426 | assert(is_valid_offset(f, io_u->offset)); |
1427 | assert(io_u->buflen); | |
1428 | zone_idx_b = zbd_zone_idx(f, io_u->offset); | |
1429 | zb = &f->zbd_info->zone_info[zone_idx_b]; | |
de65f7b7 | 1430 | orig_zb = zb; |
bfbdd35b BVA |
1431 | |
1432 | /* Accept the I/O offset for conventional zones. */ | |
b7694961 | 1433 | if (!zbd_zone_swr(zb)) |
bfbdd35b BVA |
1434 | return io_u_accept; |
1435 | ||
1436 | /* | |
1437 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1438 | * is enabled. | |
1439 | */ | |
b7694961 | 1440 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1441 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1442 | return io_u_accept; | |
1443 | ||
615555bb BVA |
1444 | zbd_check_swd(f); |
1445 | ||
fae3b9a0 | 1446 | zone_lock(td, f, zb); |
6f0c6085 | 1447 | |
bfbdd35b BVA |
1448 | switch (io_u->ddir) { |
1449 | case DDIR_READ: | |
1450 | if (td->runstate == TD_VERIFYING) { | |
ecc9e8f7 DLM |
1451 | if (td_write(td)) |
1452 | zb = zbd_replay_write_order(td, io_u, zb); | |
bfbdd35b BVA |
1453 | goto accept; |
1454 | } | |
1455 | /* | |
de65f7b7 DLM |
1456 | * Check that there is enough written data in the zone to do an |
1457 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1458 | * the I/O. | |
bfbdd35b | 1459 | */ |
b7694961 | 1460 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1461 | zb->wp - zb->start : 0; |
de65f7b7 | 1462 | if (range < min_bs || |
ee3696bd | 1463 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
bfbdd35b BVA |
1464 | pthread_mutex_unlock(&zb->mutex); |
1465 | zl = &f->zbd_info->zone_info[zbd_zone_idx(f, | |
1466 | f->file_offset + f->io_size)]; | |
1467 | zb = zbd_find_zone(td, io_u, zb, zl); | |
1468 | if (!zb) { | |
1469 | dprint(FD_ZBD, | |
1470 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1471 | f->file_name, io_u->offset, | |
1472 | io_u->buflen); | |
1473 | goto eof; | |
1474 | } | |
de65f7b7 DLM |
1475 | /* |
1476 | * zbd_find_zone() returned a zone with a range of at | |
1477 | * least min_bs. | |
1478 | */ | |
ee3696bd | 1479 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1480 | assert(range >= min_bs); |
1481 | ||
1482 | if (!td_random(td)) | |
ee3696bd | 1483 | io_u->offset = zb->start; |
bfbdd35b | 1484 | } |
de65f7b7 DLM |
1485 | /* |
1486 | * Make sure the I/O is within the zone valid data range while | |
1487 | * maximizing the I/O size and preserving randomness. | |
1488 | */ | |
1489 | if (range <= io_u->buflen) | |
ee3696bd | 1490 | io_u->offset = zb->start; |
de65f7b7 | 1491 | else if (td_random(td)) |
ee3696bd DLM |
1492 | io_u->offset = zb->start + |
1493 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 DLM |
1494 | (range - io_u->buflen)) / min_bs * min_bs; |
1495 | /* | |
1496 | * Make sure the I/O does not cross over the zone wp position. | |
1497 | */ | |
1498 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1499 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1500 | new_len = new_len / min_bs * min_bs; |
1501 | if (new_len < io_u->buflen) { | |
1502 | io_u->buflen = new_len; | |
1503 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1504 | orig_len, io_u->buflen); | |
bfbdd35b | 1505 | } |
ee3696bd DLM |
1506 | assert(zb->start <= io_u->offset); |
1507 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
bfbdd35b BVA |
1508 | goto accept; |
1509 | case DDIR_WRITE: | |
ee3696bd | 1510 | if (io_u->buflen > f->zbd_info->zone_size) |
bfbdd35b | 1511 | goto eof; |
59b07544 BVA |
1512 | if (!zbd_open_zone(td, io_u, zone_idx_b)) { |
1513 | pthread_mutex_unlock(&zb->mutex); | |
1514 | zb = zbd_convert_to_open_zone(td, io_u); | |
1515 | if (!zb) | |
1516 | goto eof; | |
1517 | zone_idx_b = zb - f->zbd_info->zone_info; | |
1518 | } | |
a7c2b6fc BVA |
1519 | /* Check whether the zone reset threshold has been exceeded */ |
1520 | if (td->o.zrf.u.f) { | |
ee3696bd | 1521 | if (f->zbd_info->sectors_with_data >= |
a7c2b6fc BVA |
1522 | f->io_size * td->o.zrt.u.f && |
1523 | zbd_dec_and_reset_write_cnt(td, f)) { | |
1524 | zb->reset_zone = 1; | |
1525 | } | |
1526 | } | |
bfbdd35b BVA |
1527 | /* Reset the zone pointer if necessary */ |
1528 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
1529 | assert(td->o.verify == VERIFY_NONE); | |
1530 | /* | |
1531 | * Since previous write requests may have been submitted | |
1532 | * asynchronously and since we will submit the zone | |
1533 | * reset synchronously, wait until previously submitted | |
1534 | * write requests have completed before issuing a | |
1535 | * zone reset. | |
1536 | */ | |
1537 | io_u_quiesce(td); | |
1538 | zb->reset_zone = 0; | |
1539 | if (zbd_reset_zone(td, f, zb) < 0) | |
1540 | goto eof; | |
1541 | } | |
1542 | /* Make writes occur at the write pointer */ | |
1543 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 1544 | io_u->offset = zb->wp; |
bfbdd35b BVA |
1545 | if (!is_valid_offset(f, io_u->offset)) { |
1546 | dprint(FD_ZBD, "Dropped request with offset %llu\n", | |
1547 | io_u->offset); | |
1548 | goto eof; | |
1549 | } | |
1550 | /* | |
1551 | * Make sure that the buflen is a multiple of the minimal | |
1552 | * block size. Give up if shrinking would make the request too | |
1553 | * small. | |
1554 | */ | |
1555 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1556 | (zb + 1)->start - io_u->offset); |
bfbdd35b BVA |
1557 | new_len = new_len / min_bs * min_bs; |
1558 | if (new_len == io_u->buflen) | |
1559 | goto accept; | |
1560 | if (new_len >= min_bs) { | |
1561 | io_u->buflen = new_len; | |
1562 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1563 | orig_len, io_u->buflen); | |
1564 | goto accept; | |
1565 | } | |
1566 | log_err("Zone remainder %lld smaller than minimum block size %d\n", | |
ee3696bd | 1567 | ((zb + 1)->start - io_u->offset), |
bfbdd35b BVA |
1568 | min_bs); |
1569 | goto eof; | |
1570 | case DDIR_TRIM: | |
1571 | /* fall-through */ | |
1572 | case DDIR_SYNC: | |
1573 | case DDIR_DATASYNC: | |
1574 | case DDIR_SYNC_FILE_RANGE: | |
1575 | case DDIR_WAIT: | |
1576 | case DDIR_LAST: | |
1577 | case DDIR_INVAL: | |
1578 | goto accept; | |
1579 | } | |
1580 | ||
1581 | assert(false); | |
1582 | ||
1583 | accept: | |
1584 | assert(zb); | |
b7694961 | 1585 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
1586 | assert(!io_u->zbd_queue_io); |
1587 | assert(!io_u->zbd_put_io); | |
1588 | io_u->zbd_queue_io = zbd_queue_io; | |
1589 | io_u->zbd_put_io = zbd_put_io; | |
bfbdd35b BVA |
1590 | return io_u_accept; |
1591 | ||
1592 | eof: | |
1593 | if (zb) | |
1594 | pthread_mutex_unlock(&zb->mutex); | |
1595 | return io_u_eof; | |
1596 | } | |
fd5d733f BVA |
1597 | |
1598 | /* Return a string with ZBD statistics */ | |
1599 | char *zbd_write_status(const struct thread_stat *ts) | |
1600 | { | |
1601 | char *res; | |
1602 | ||
a0c84dd4 | 1603 | if (asprintf(&res, "; %llu zone resets", (unsigned long long) ts->nr_zone_resets) < 0) |
fd5d733f BVA |
1604 | return NULL; |
1605 | return res; | |
1606 | } |