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" | |
44ec32cb | 22 | #include "pshared.h" |
bfbdd35b BVA |
23 | #include "zbd.h" |
24 | ||
b7694961 DLM |
25 | /** |
26 | * zbd_get_zoned_model - Get a device zoned model | |
27 | * @td: FIO thread data | |
28 | * @f: FIO file for which to get model information | |
29 | */ | |
30 | int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, | |
31 | enum zbd_zoned_model *model) | |
32 | { | |
33 | int ret; | |
34 | ||
6c5b11d3 DLM |
35 | if (td->io_ops && td->io_ops->get_zoned_model) |
36 | ret = td->io_ops->get_zoned_model(td, f, model); | |
37 | else | |
38 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
39 | if (ret < 0) { |
40 | td_verror(td, errno, "get zoned model failed"); | |
41 | log_err("%s: get zoned model failed (%d).\n", | |
42 | f->file_name, errno); | |
43 | } | |
44 | ||
45 | return ret; | |
46 | } | |
47 | ||
48 | /** | |
49 | * zbd_report_zones - Get zone information | |
50 | * @td: FIO thread data. | |
51 | * @f: FIO file for which to get zone information | |
52 | * @offset: offset from which to report zones | |
53 | * @zones: Array of struct zbd_zone | |
54 | * @nr_zones: Size of @zones array | |
55 | * | |
56 | * Get zone information into @zones starting from the zone at offset @offset | |
57 | * for the device specified by @f. | |
58 | * | |
59 | * Returns the number of zones reported upon success and a negative error code | |
60 | * upon failure. If the zone report is empty, always assume an error (device | |
61 | * problem) and return -EIO. | |
62 | */ | |
63 | int zbd_report_zones(struct thread_data *td, struct fio_file *f, | |
64 | uint64_t offset, struct zbd_zone *zones, | |
65 | unsigned int nr_zones) | |
66 | { | |
67 | int ret; | |
68 | ||
6c5b11d3 DLM |
69 | if (td->io_ops && td->io_ops->report_zones) |
70 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
71 | else | |
72 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
73 | if (ret < 0) { |
74 | td_verror(td, errno, "report zones failed"); | |
75 | log_err("%s: report zones from sector %llu failed (%d).\n", | |
76 | f->file_name, (unsigned long long)offset >> 9, errno); | |
77 | } else if (ret == 0) { | |
78 | td_verror(td, errno, "Empty zone report"); | |
79 | log_err("%s: report zones from sector %llu is empty.\n", | |
80 | f->file_name, (unsigned long long)offset >> 9); | |
81 | ret = -EIO; | |
82 | } | |
83 | ||
84 | return ret; | |
85 | } | |
86 | ||
87 | /** | |
88 | * zbd_reset_wp - reset the write pointer of a range of zones | |
89 | * @td: FIO thread data. | |
90 | * @f: FIO file for which to reset zones | |
91 | * @offset: Starting offset of the first zone to reset | |
92 | * @length: Length of the range of zones to reset | |
93 | * | |
94 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
95 | * Returns 0 upon success and a negative error code upon failure. | |
96 | */ | |
97 | int zbd_reset_wp(struct thread_data *td, struct fio_file *f, | |
98 | uint64_t offset, uint64_t length) | |
99 | { | |
100 | int ret; | |
101 | ||
6c5b11d3 DLM |
102 | if (td->io_ops && td->io_ops->reset_wp) |
103 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
104 | else | |
105 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
106 | if (ret < 0) { |
107 | td_verror(td, errno, "resetting wp failed"); | |
108 | log_err("%s: resetting wp for %llu sectors at sector %llu failed (%d).\n", | |
109 | f->file_name, (unsigned long long)length >> 9, | |
110 | (unsigned long long)offset >> 9, errno); | |
111 | } | |
112 | ||
113 | return ret; | |
114 | } | |
115 | ||
bfbdd35b BVA |
116 | /** |
117 | * zbd_zone_idx - convert an offset into a zone number | |
118 | * @f: file pointer. | |
119 | * @offset: offset in bytes. If this offset is in the first zone_size bytes | |
120 | * past the disk size then the index of the sentinel is returned. | |
121 | */ | |
122 | static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) | |
123 | { | |
124 | uint32_t zone_idx; | |
125 | ||
cd775e06 | 126 | if (f->zbd_info->zone_size_log2 > 0) |
bfbdd35b BVA |
127 | zone_idx = offset >> f->zbd_info->zone_size_log2; |
128 | else | |
ee3696bd | 129 | zone_idx = offset / f->zbd_info->zone_size; |
bfbdd35b BVA |
130 | |
131 | return min(zone_idx, f->zbd_info->nr_zones); | |
132 | } | |
133 | ||
b7694961 DLM |
134 | /** |
135 | * zbd_zone_swr - Test whether a zone requires sequential writes | |
136 | * @z: zone info pointer. | |
137 | */ | |
138 | static inline bool zbd_zone_swr(struct fio_zone_info *z) | |
139 | { | |
140 | return z->type == ZBD_ZONE_TYPE_SWR; | |
141 | } | |
142 | ||
bfbdd35b BVA |
143 | /** |
144 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
145 | * @f: file pointer. | |
146 | * @z: zone info pointer. | |
147 | * @required: minimum number of bytes that must remain in a zone. | |
148 | * | |
149 | * The caller must hold z->mutex. | |
150 | */ | |
151 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
152 | uint64_t required) | |
153 | { | |
154 | assert((required & 511) == 0); | |
155 | ||
b7694961 | 156 | return zbd_zone_swr(z) && |
ee3696bd | 157 | z->wp + required > z->start + f->zbd_info->zone_size; |
bfbdd35b BVA |
158 | } |
159 | ||
fae3b9a0 | 160 | static void zone_lock(struct thread_data *td, struct fio_file *f, struct fio_zone_info *z) |
1f57803b | 161 | { |
fae3b9a0 AD |
162 | struct zoned_block_device_info *zbd = f->zbd_info; |
163 | uint32_t nz = z - zbd->zone_info; | |
164 | ||
165 | /* A thread should never lock zones outside its working area. */ | |
166 | assert(f->min_zone <= nz && nz < f->max_zone); | |
167 | ||
1f57803b DLM |
168 | /* |
169 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
170 | * is changed or when io_u completes and zbd_put_io() executed. | |
171 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
172 | * other waiting for zone locks when building an io_u batch, first | |
173 | * only trylock the zone. If the zone is already locked by another job, | |
174 | * process the currently queued I/Os so that I/O progress is made and | |
175 | * zones unlocked. | |
176 | */ | |
177 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
178 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
179 | io_u_quiesce(td); | |
180 | pthread_mutex_lock(&z->mutex); | |
181 | } | |
182 | } | |
183 | ||
bfbdd35b BVA |
184 | static bool is_valid_offset(const struct fio_file *f, uint64_t offset) |
185 | { | |
186 | return (uint64_t)(offset - f->file_offset) < f->io_size; | |
187 | } | |
188 | ||
189 | /* Verify whether direct I/O is used for all host-managed zoned drives. */ | |
190 | static bool zbd_using_direct_io(void) | |
191 | { | |
192 | struct thread_data *td; | |
193 | struct fio_file *f; | |
194 | int i, j; | |
195 | ||
196 | for_each_td(td, i) { | |
197 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) | |
198 | continue; | |
199 | for_each_file(td, f, j) { | |
200 | if (f->zbd_info && | |
b7694961 | 201 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
202 | return false; |
203 | } | |
204 | } | |
205 | ||
206 | return true; | |
207 | } | |
208 | ||
209 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
210 | static bool zbd_is_seq_job(struct fio_file *f) | |
211 | { | |
212 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
213 | ||
214 | assert(f->zbd_info); | |
215 | if (f->io_size == 0) | |
216 | return false; | |
217 | zone_idx_b = zbd_zone_idx(f, f->file_offset); | |
218 | zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1); | |
219 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) | |
b7694961 | 220 | if (zbd_zone_swr(&f->zbd_info->zone_info[zone_idx])) |
bfbdd35b BVA |
221 | return true; |
222 | ||
223 | return false; | |
224 | } | |
225 | ||
226 | /* | |
227 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
228 | */ | |
229 | static bool zbd_verify_sizes(void) | |
230 | { | |
231 | const struct fio_zone_info *z; | |
232 | struct thread_data *td; | |
233 | struct fio_file *f; | |
234 | uint64_t new_offset, new_end; | |
235 | uint32_t zone_idx; | |
236 | int i, j; | |
237 | ||
238 | for_each_td(td, i) { | |
239 | for_each_file(td, f, j) { | |
240 | if (!f->zbd_info) | |
241 | continue; | |
242 | if (f->file_offset >= f->real_file_size) | |
243 | continue; | |
244 | if (!zbd_is_seq_job(f)) | |
245 | continue; | |
4d37720a | 246 | |
fc52e449 DLM |
247 | if (!td->o.zone_size) { |
248 | td->o.zone_size = f->zbd_info->zone_size; | |
249 | if (!td->o.zone_size) { | |
250 | log_err("%s: invalid 0 zone size\n", | |
251 | f->file_name); | |
252 | return false; | |
253 | } | |
971d6a22 DLM |
254 | } else if (td->o.zone_size != f->zbd_info->zone_size) { |
255 | log_err("%s: job parameter zonesize %llu does not match disk zone size %llu.\n", | |
256 | f->file_name, (unsigned long long) td->o.zone_size, | |
257 | (unsigned long long) f->zbd_info->zone_size); | |
258 | return false; | |
fc52e449 DLM |
259 | } |
260 | ||
4d37720a DLM |
261 | if (td->o.zone_skip && |
262 | (td->o.zone_skip < td->o.zone_size || | |
263 | td->o.zone_skip % td->o.zone_size)) { | |
264 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", | |
265 | f->file_name, (unsigned long long) td->o.zone_skip, | |
266 | (unsigned long long) td->o.zone_size); | |
267 | return false; | |
268 | } | |
269 | ||
bfbdd35b BVA |
270 | zone_idx = zbd_zone_idx(f, f->file_offset); |
271 | z = &f->zbd_info->zone_info[zone_idx]; | |
f09a7773 PL |
272 | if ((f->file_offset != z->start) && |
273 | (td->o.td_ddir != TD_DDIR_READ)) { | |
ee3696bd | 274 | new_offset = (z+1)->start; |
bfbdd35b BVA |
275 | if (new_offset >= f->file_offset + f->io_size) { |
276 | log_info("%s: io_size must be at least one zone\n", | |
277 | f->file_name); | |
278 | return false; | |
279 | } | |
a0c84dd4 JA |
280 | log_info("%s: rounded up offset from %llu to %llu\n", |
281 | f->file_name, (unsigned long long) f->file_offset, | |
282 | (unsigned long long) new_offset); | |
bfbdd35b BVA |
283 | f->io_size -= (new_offset - f->file_offset); |
284 | f->file_offset = new_offset; | |
285 | } | |
286 | zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size); | |
287 | z = &f->zbd_info->zone_info[zone_idx]; | |
ee3696bd | 288 | new_end = z->start; |
f09a7773 PL |
289 | if ((td->o.td_ddir != TD_DDIR_READ) && |
290 | (f->file_offset + f->io_size != new_end)) { | |
bfbdd35b BVA |
291 | if (new_end <= f->file_offset) { |
292 | log_info("%s: io_size must be at least one zone\n", | |
293 | f->file_name); | |
294 | return false; | |
295 | } | |
a0c84dd4 JA |
296 | log_info("%s: rounded down io_size from %llu to %llu\n", |
297 | f->file_name, (unsigned long long) f->io_size, | |
298 | (unsigned long long) new_end - f->file_offset); | |
bfbdd35b BVA |
299 | f->io_size = new_end - f->file_offset; |
300 | } | |
fae3b9a0 AD |
301 | |
302 | f->min_zone = zbd_zone_idx(f, f->file_offset); | |
303 | f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size); | |
bfbdd35b BVA |
304 | } |
305 | } | |
306 | ||
307 | return true; | |
308 | } | |
309 | ||
310 | static bool zbd_verify_bs(void) | |
311 | { | |
312 | struct thread_data *td; | |
313 | struct fio_file *f; | |
314 | uint32_t zone_size; | |
315 | int i, j, k; | |
316 | ||
317 | for_each_td(td, i) { | |
318 | for_each_file(td, f, j) { | |
319 | if (!f->zbd_info) | |
320 | continue; | |
321 | zone_size = f->zbd_info->zone_size; | |
322 | for (k = 0; k < ARRAY_SIZE(td->o.bs); k++) { | |
323 | if (td->o.verify != VERIFY_NONE && | |
ee3696bd | 324 | zone_size % td->o.bs[k] != 0) { |
bfbdd35b BVA |
325 | log_info("%s: block size %llu is not a divisor of the zone size %d\n", |
326 | f->file_name, td->o.bs[k], | |
ee3696bd | 327 | zone_size); |
bfbdd35b BVA |
328 | return false; |
329 | } | |
330 | } | |
331 | } | |
332 | } | |
333 | return true; | |
334 | } | |
335 | ||
bfbdd35b BVA |
336 | static int ilog2(uint64_t i) |
337 | { | |
338 | int log = -1; | |
339 | ||
340 | while (i) { | |
341 | i >>= 1; | |
342 | log++; | |
343 | } | |
344 | return log; | |
345 | } | |
346 | ||
347 | /* | |
348 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
349 | * allows to execute a ZBD workload against a non-ZBD device. | |
350 | */ | |
351 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
352 | { | |
353 | uint32_t nr_zones; | |
354 | struct fio_zone_info *p; | |
a4b7f12b | 355 | uint64_t zone_size = td->o.zone_size; |
bfbdd35b | 356 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
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 | ||
44ec32cb | 377 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
378 | zbd_info->refcount = 1; |
379 | p = &zbd_info->zone_info[0]; | |
380 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
381 | mutex_init_pshared_with_type(&p->mutex, |
382 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b BVA |
383 | p->start = i * zone_size; |
384 | p->wp = p->start + zone_size; | |
b7694961 DLM |
385 | p->type = ZBD_ZONE_TYPE_SWR; |
386 | p->cond = ZBD_ZONE_COND_EMPTY; | |
bfbdd35b BVA |
387 | } |
388 | /* a sentinel */ | |
389 | p->start = nr_zones * zone_size; | |
390 | ||
391 | f->zbd_info = zbd_info; | |
392 | f->zbd_info->zone_size = zone_size; | |
393 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 394 | ilog2(zone_size) : 0; |
bfbdd35b | 395 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
396 | return 0; |
397 | } | |
398 | ||
399 | /* | |
b7694961 DLM |
400 | * Maximum number of zones to report in one operation. |
401 | */ | |
402 | #define ZBD_REPORT_MAX_ZONES 8192U | |
403 | ||
404 | /* | |
405 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
406 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
407 | */ |
408 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
409 | { | |
b7694961 DLM |
410 | int nr_zones, nrz; |
411 | struct zbd_zone *zones, *z; | |
bfbdd35b | 412 | struct fio_zone_info *p; |
b7694961 | 413 | uint64_t zone_size, offset; |
bfbdd35b | 414 | struct zoned_block_device_info *zbd_info = NULL; |
b7694961 | 415 | int i, j, ret = 0; |
bfbdd35b | 416 | |
b7694961 DLM |
417 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
418 | if (!zones) | |
bfbdd35b BVA |
419 | goto out; |
420 | ||
b7694961 DLM |
421 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
422 | if (nrz < 0) { | |
423 | ret = nrz; | |
424 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
425 | f->file_name, -ret); | |
426 | goto out; | |
bfbdd35b BVA |
427 | } |
428 | ||
b7694961 | 429 | zone_size = zones[0].len; |
ee3696bd | 430 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
431 | |
432 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
433 | td->o.zone_size = zone_size; |
434 | } else if (td->o.zone_size != zone_size) { | |
4580ff92 DLM |
435 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %llu.\n", |
436 | f->file_name, (unsigned long long) td->o.zone_size, | |
a0c84dd4 | 437 | (unsigned long long) zone_size); |
bfbdd35b | 438 | ret = -EINVAL; |
b7694961 | 439 | goto out; |
bfbdd35b BVA |
440 | } |
441 | ||
a0c84dd4 JA |
442 | dprint(FD_ZBD, "Device %s has %d zones of size %llu KB\n", f->file_name, |
443 | nr_zones, (unsigned long long) zone_size / 1024); | |
bfbdd35b BVA |
444 | |
445 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
446 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
447 | ret = -ENOMEM; | |
448 | if (!zbd_info) | |
b7694961 | 449 | goto out; |
44ec32cb | 450 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
451 | zbd_info->refcount = 1; |
452 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
453 | for (offset = 0, j = 0; j < nr_zones;) { |
454 | z = &zones[0]; | |
455 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
456 | mutex_init_pshared_with_type(&p->mutex, |
457 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 458 | p->start = z->start; |
bfbdd35b | 459 | switch (z->cond) { |
b7694961 DLM |
460 | case ZBD_ZONE_COND_NOT_WP: |
461 | case ZBD_ZONE_COND_FULL: | |
ee3696bd | 462 | p->wp = p->start + zone_size; |
bfbdd35b BVA |
463 | break; |
464 | default: | |
465 | assert(z->start <= z->wp); | |
b7694961 DLM |
466 | assert(z->wp <= z->start + zone_size); |
467 | p->wp = z->wp; | |
bfbdd35b BVA |
468 | break; |
469 | } | |
470 | p->type = z->type; | |
471 | p->cond = z->cond; | |
472 | if (j > 0 && p->start != p[-1].start + zone_size) { | |
473 | log_info("%s: invalid zone data\n", | |
474 | f->file_name); | |
475 | ret = -EINVAL; | |
b7694961 | 476 | goto out; |
bfbdd35b BVA |
477 | } |
478 | } | |
479 | z--; | |
b7694961 | 480 | offset = z->start + z->len; |
bfbdd35b BVA |
481 | if (j >= nr_zones) |
482 | break; | |
b7694961 DLM |
483 | nrz = zbd_report_zones(td, f, offset, |
484 | zones, ZBD_REPORT_MAX_ZONES); | |
485 | if (nrz < 0) { | |
486 | ret = nrz; | |
487 | log_info("fio: report zones (offset %llu) failed for %s (%d).\n", | |
488 | (unsigned long long)offset, | |
489 | f->file_name, -ret); | |
490 | goto out; | |
bfbdd35b BVA |
491 | } |
492 | } | |
b7694961 | 493 | |
bfbdd35b | 494 | /* a sentinel */ |
b7694961 | 495 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
496 | |
497 | f->zbd_info = zbd_info; | |
498 | f->zbd_info->zone_size = zone_size; | |
499 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 500 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
501 | f->zbd_info->nr_zones = nr_zones; |
502 | zbd_info = NULL; | |
503 | ret = 0; | |
504 | ||
bfbdd35b | 505 | out: |
b7694961 DLM |
506 | sfree(zbd_info); |
507 | free(zones); | |
bfbdd35b BVA |
508 | return ret; |
509 | } | |
510 | ||
511 | /* | |
512 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
513 | * | |
514 | * Returns 0 upon success and a negative error code upon failure. | |
515 | */ | |
379e5f09 | 516 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 517 | { |
b7694961 DLM |
518 | enum zbd_zoned_model zbd_model; |
519 | int ret; | |
bfbdd35b BVA |
520 | |
521 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
522 | ||
b7694961 DLM |
523 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
524 | if (ret) | |
525 | return ret; | |
526 | ||
bfbdd35b | 527 | switch (zbd_model) { |
b7694961 DLM |
528 | case ZBD_IGNORE: |
529 | return 0; | |
530 | case ZBD_HOST_AWARE: | |
531 | case ZBD_HOST_MANAGED: | |
bfbdd35b BVA |
532 | ret = parse_zone_info(td, f); |
533 | break; | |
b7694961 | 534 | case ZBD_NONE: |
bfbdd35b BVA |
535 | ret = init_zone_info(td, f); |
536 | break; | |
b7694961 DLM |
537 | default: |
538 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
539 | log_err("Unsupported zoned model\n"); | |
540 | return -EINVAL; | |
bfbdd35b | 541 | } |
b7694961 | 542 | |
219c662d | 543 | if (ret == 0) { |
bfbdd35b | 544 | f->zbd_info->model = zbd_model; |
219c662d AD |
545 | f->zbd_info->max_open_zones = td->o.max_open_zones; |
546 | } | |
bfbdd35b BVA |
547 | return ret; |
548 | } | |
549 | ||
550 | void zbd_free_zone_info(struct fio_file *f) | |
551 | { | |
552 | uint32_t refcount; | |
553 | ||
3c1dc34c | 554 | assert(f->zbd_info); |
bfbdd35b BVA |
555 | |
556 | pthread_mutex_lock(&f->zbd_info->mutex); | |
557 | refcount = --f->zbd_info->refcount; | |
558 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
559 | ||
560 | assert((int32_t)refcount >= 0); | |
561 | if (refcount == 0) | |
562 | sfree(f->zbd_info); | |
563 | f->zbd_info = NULL; | |
564 | } | |
565 | ||
566 | /* | |
567 | * Initialize f->zbd_info. | |
568 | * | |
569 | * Returns 0 upon success and a negative error code upon failure. | |
570 | * | |
571 | * Note: this function can only work correctly if it is called before the first | |
572 | * fio fork() call. | |
573 | */ | |
574 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
575 | { | |
576 | struct thread_data *td2; | |
577 | struct fio_file *f2; | |
578 | int i, j, ret; | |
579 | ||
580 | for_each_td(td2, i) { | |
581 | for_each_file(td2, f2, j) { | |
582 | if (td2 == td && f2 == file) | |
583 | continue; | |
584 | if (!f2->zbd_info || | |
585 | strcmp(f2->file_name, file->file_name) != 0) | |
586 | continue; | |
587 | file->zbd_info = f2->zbd_info; | |
588 | file->zbd_info->refcount++; | |
589 | return 0; | |
590 | } | |
591 | } | |
592 | ||
593 | ret = zbd_create_zone_info(td, file); | |
594 | if (ret < 0) | |
c5837eec | 595 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
bfbdd35b BVA |
596 | return ret; |
597 | } | |
598 | ||
3c1dc34c | 599 | int zbd_setup_files(struct thread_data *td) |
bfbdd35b BVA |
600 | { |
601 | struct fio_file *f; | |
602 | int i; | |
603 | ||
604 | for_each_file(td, f, i) { | |
a4b7f12b | 605 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 606 | return 1; |
bfbdd35b BVA |
607 | } |
608 | ||
609 | if (!zbd_using_direct_io()) { | |
610 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
611 | return 1; | |
612 | } | |
613 | ||
614 | if (!zbd_verify_sizes()) | |
615 | return 1; | |
616 | ||
617 | if (!zbd_verify_bs()) | |
618 | return 1; | |
619 | ||
219c662d AD |
620 | for_each_file(td, f, i) { |
621 | struct zoned_block_device_info *zbd = f->zbd_info; | |
622 | ||
623 | if (!zbd) | |
624 | continue; | |
625 | ||
626 | zbd->max_open_zones = zbd->max_open_zones ?: ZBD_MAX_OPEN_ZONES; | |
627 | ||
628 | if (td->o.max_open_zones > 0 && | |
629 | zbd->max_open_zones != td->o.max_open_zones) { | |
630 | log_err("Different 'max_open_zones' values\n"); | |
631 | return 1; | |
632 | } | |
633 | if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) { | |
634 | log_err("'max_open_zones' value is limited by %u\n", ZBD_MAX_OPEN_ZONES); | |
635 | return 1; | |
636 | } | |
637 | } | |
638 | ||
bfbdd35b BVA |
639 | return 0; |
640 | } | |
641 | ||
642 | /** | |
643 | * zbd_reset_range - reset zones for a range of sectors | |
644 | * @td: FIO thread data. | |
645 | * @f: Fio file for which to reset zones | |
646 | * @sector: Starting sector in units of 512 bytes | |
647 | * @nr_sectors: Number of sectors in units of 512 bytes | |
648 | * | |
649 | * Returns 0 upon success and a negative error code upon failure. | |
650 | */ | |
b7694961 | 651 | static int zbd_reset_range(struct thread_data *td, struct fio_file *f, |
ee3696bd | 652 | uint64_t offset, uint64_t length) |
bfbdd35b | 653 | { |
bfbdd35b BVA |
654 | uint32_t zone_idx_b, zone_idx_e; |
655 | struct fio_zone_info *zb, *ze, *z; | |
656 | int ret = 0; | |
657 | ||
ee3696bd | 658 | assert(is_valid_offset(f, offset + length - 1)); |
b7694961 | 659 | |
bfbdd35b | 660 | switch (f->zbd_info->model) { |
b7694961 DLM |
661 | case ZBD_HOST_AWARE: |
662 | case ZBD_HOST_MANAGED: | |
663 | ret = zbd_reset_wp(td, f, offset, length); | |
664 | if (ret < 0) | |
bfbdd35b | 665 | return ret; |
bfbdd35b | 666 | break; |
b7694961 | 667 | default: |
bfbdd35b BVA |
668 | break; |
669 | } | |
670 | ||
ee3696bd | 671 | zone_idx_b = zbd_zone_idx(f, offset); |
bfbdd35b | 672 | zb = &f->zbd_info->zone_info[zone_idx_b]; |
ee3696bd | 673 | zone_idx_e = zbd_zone_idx(f, offset + length); |
bfbdd35b BVA |
674 | ze = &f->zbd_info->zone_info[zone_idx_e]; |
675 | for (z = zb; z < ze; z++) { | |
676 | pthread_mutex_lock(&z->mutex); | |
a7c2b6fc BVA |
677 | pthread_mutex_lock(&f->zbd_info->mutex); |
678 | f->zbd_info->sectors_with_data -= z->wp - z->start; | |
679 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
bfbdd35b BVA |
680 | z->wp = z->start; |
681 | z->verify_block = 0; | |
682 | pthread_mutex_unlock(&z->mutex); | |
683 | } | |
684 | ||
fd5d733f BVA |
685 | td->ts.nr_zone_resets += ze - zb; |
686 | ||
bfbdd35b BVA |
687 | return ret; |
688 | } | |
689 | ||
a0c84dd4 JA |
690 | static unsigned int zbd_zone_nr(struct zoned_block_device_info *zbd_info, |
691 | struct fio_zone_info *zone) | |
692 | { | |
089ddd95 | 693 | return zone - zbd_info->zone_info; |
a0c84dd4 JA |
694 | } |
695 | ||
bfbdd35b BVA |
696 | /** |
697 | * zbd_reset_zone - reset the write pointer of a single zone | |
698 | * @td: FIO thread data. | |
699 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
700 | * @z: Zone to reset. | |
701 | * | |
702 | * Returns 0 upon success and a negative error code upon failure. | |
703 | */ | |
b7694961 | 704 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, |
bfbdd35b BVA |
705 | struct fio_zone_info *z) |
706 | { | |
a0c84dd4 JA |
707 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name, |
708 | zbd_zone_nr(f->zbd_info, z)); | |
d60be7d5 DLM |
709 | |
710 | return zbd_reset_range(td, f, z->start, (z+1)->start - z->start); | |
bfbdd35b BVA |
711 | } |
712 | ||
25dc6606 AD |
713 | /* The caller must hold f->zbd_info->mutex */ |
714 | static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, | |
715 | unsigned int open_zone_idx) | |
716 | { | |
717 | uint32_t zone_idx; | |
718 | ||
719 | assert(open_zone_idx < f->zbd_info->num_open_zones); | |
720 | zone_idx = f->zbd_info->open_zones[open_zone_idx]; | |
721 | memmove(f->zbd_info->open_zones + open_zone_idx, | |
722 | f->zbd_info->open_zones + open_zone_idx + 1, | |
723 | (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) * | |
724 | sizeof(f->zbd_info->open_zones[0])); | |
725 | f->zbd_info->num_open_zones--; | |
219c662d | 726 | td->num_open_zones--; |
25dc6606 AD |
727 | f->zbd_info->zone_info[zone_idx].open = 0; |
728 | } | |
729 | ||
bfbdd35b BVA |
730 | /* |
731 | * Reset a range of zones. Returns 0 upon success and 1 upon failure. | |
732 | * @td: fio thread data. | |
733 | * @f: fio file for which to reset zones | |
734 | * @zb: first zone to reset. | |
735 | * @ze: first zone not to reset. | |
736 | * @all_zones: whether to reset all zones or only those zones for which the | |
737 | * write pointer is not a multiple of td->o.min_bs[DDIR_WRITE]. | |
738 | */ | |
739 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
740 | struct fio_zone_info *const zb, | |
741 | struct fio_zone_info *const ze, bool all_zones) | |
742 | { | |
4803b841 | 743 | struct fio_zone_info *z; |
ee3696bd | 744 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; |
bfbdd35b BVA |
745 | bool reset_wp; |
746 | int res = 0; | |
747 | ||
adc6adcb DLM |
748 | assert(min_bs); |
749 | ||
a0c84dd4 JA |
750 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name, |
751 | zbd_zone_nr(f->zbd_info, zb), zbd_zone_nr(f->zbd_info, ze)); | |
bfbdd35b | 752 | for (z = zb; z < ze; z++) { |
25dc6606 AD |
753 | uint32_t nz = z - f->zbd_info->zone_info; |
754 | ||
b7694961 | 755 | if (!zbd_zone_swr(z)) |
1f57803b | 756 | continue; |
fae3b9a0 | 757 | zone_lock(td, f, z); |
25dc6606 AD |
758 | if (all_zones) { |
759 | unsigned int i; | |
760 | ||
761 | pthread_mutex_lock(&f->zbd_info->mutex); | |
762 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
763 | if (f->zbd_info->open_zones[i] == nz) | |
764 | zbd_close_zone(td, f, i); | |
765 | } | |
766 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
767 | ||
768 | reset_wp = z->wp != z->start; | |
769 | } else { | |
767d1372 | 770 | reset_wp = z->wp % min_bs != 0; |
25dc6606 | 771 | } |
1f57803b DLM |
772 | if (reset_wp) { |
773 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
774 | f->file_name, | |
775 | zbd_zone_nr(f->zbd_info, z)); | |
776 | if (zbd_reset_zone(td, f, z) < 0) | |
777 | res = 1; | |
bfbdd35b | 778 | } |
bfbdd35b | 779 | pthread_mutex_unlock(&z->mutex); |
4803b841 | 780 | } |
bfbdd35b BVA |
781 | |
782 | return res; | |
783 | } | |
784 | ||
a7c2b6fc BVA |
785 | /* |
786 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
787 | * zone reset. | |
788 | */ | |
1bb1bcad AD |
789 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
790 | const struct fio_file *f) | |
a7c2b6fc BVA |
791 | { |
792 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
793 | ||
a7c2b6fc BVA |
794 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
795 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
796 | } |
797 | ||
798 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
799 | const struct fio_file *f) | |
800 | { | |
801 | pthread_mutex_lock(&f->zbd_info->mutex); | |
802 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
803 | pthread_mutex_unlock(&f->zbd_info->mutex); |
804 | } | |
805 | ||
806 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
807 | const struct fio_file *f) | |
808 | { | |
809 | uint32_t write_cnt = 0; | |
810 | ||
811 | pthread_mutex_lock(&f->zbd_info->mutex); | |
812 | assert(f->zbd_info->write_cnt); | |
813 | if (f->zbd_info->write_cnt) | |
814 | write_cnt = --f->zbd_info->write_cnt; | |
815 | if (write_cnt == 0) | |
1bb1bcad | 816 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
817 | pthread_mutex_unlock(&f->zbd_info->mutex); |
818 | ||
819 | return write_cnt == 0; | |
820 | } | |
821 | ||
91d25131 BVA |
822 | enum swd_action { |
823 | CHECK_SWD, | |
824 | SET_SWD, | |
825 | }; | |
826 | ||
827 | /* Calculate the number of sectors with data (swd) and perform action 'a' */ | |
828 | static uint64_t zbd_process_swd(const struct fio_file *f, enum swd_action a) | |
615555bb | 829 | { |
615555bb BVA |
830 | struct fio_zone_info *zb, *ze, *z; |
831 | uint64_t swd = 0; | |
832 | ||
833 | zb = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)]; | |
834 | ze = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset + | |
835 | f->io_size)]; | |
836 | for (z = zb; z < ze; z++) { | |
837 | pthread_mutex_lock(&z->mutex); | |
838 | swd += z->wp - z->start; | |
839 | } | |
840 | pthread_mutex_lock(&f->zbd_info->mutex); | |
91d25131 BVA |
841 | switch (a) { |
842 | case CHECK_SWD: | |
843 | assert(f->zbd_info->sectors_with_data == swd); | |
844 | break; | |
845 | case SET_SWD: | |
846 | f->zbd_info->sectors_with_data = swd; | |
847 | break; | |
848 | } | |
615555bb BVA |
849 | pthread_mutex_unlock(&f->zbd_info->mutex); |
850 | for (z = zb; z < ze; z++) | |
851 | pthread_mutex_unlock(&z->mutex); | |
91d25131 BVA |
852 | |
853 | return swd; | |
854 | } | |
855 | ||
856 | /* | |
857 | * The swd check is useful for debugging but takes too much time to leave | |
858 | * it enabled all the time. Hence it is disabled by default. | |
859 | */ | |
860 | static const bool enable_check_swd = false; | |
861 | ||
862 | /* Check whether the value of zbd_info.sectors_with_data is correct. */ | |
863 | static void zbd_check_swd(const struct fio_file *f) | |
864 | { | |
865 | if (!enable_check_swd) | |
866 | return; | |
867 | ||
868 | zbd_process_swd(f, CHECK_SWD); | |
869 | } | |
870 | ||
871 | static void zbd_init_swd(struct fio_file *f) | |
872 | { | |
873 | uint64_t swd; | |
874 | ||
409a4f29 NA |
875 | if (!enable_check_swd) |
876 | return; | |
877 | ||
91d25131 BVA |
878 | swd = zbd_process_swd(f, SET_SWD); |
879 | dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name, | |
880 | swd); | |
615555bb BVA |
881 | } |
882 | ||
bfbdd35b BVA |
883 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
884 | { | |
91d25131 | 885 | struct fio_zone_info *zb, *ze; |
bfbdd35b | 886 | |
767d1372 | 887 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
888 | return; |
889 | ||
fae3b9a0 AD |
890 | zb = &f->zbd_info->zone_info[f->min_zone]; |
891 | ze = &f->zbd_info->zone_info[f->max_zone]; | |
91d25131 | 892 | zbd_init_swd(f); |
bfbdd35b BVA |
893 | /* |
894 | * If data verification is enabled reset the affected zones before | |
895 | * writing any data to avoid that a zone reset has to be issued while | |
896 | * writing data, which causes data loss. | |
897 | */ | |
898 | zbd_reset_zones(td, f, zb, ze, td->o.verify != VERIFY_NONE && | |
bfbdd35b | 899 | td->runstate != TD_VERIFYING); |
a7c2b6fc | 900 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
901 | } |
902 | ||
59b07544 BVA |
903 | /* The caller must hold f->zbd_info->mutex. */ |
904 | static bool is_zone_open(const struct thread_data *td, const struct fio_file *f, | |
905 | unsigned int zone_idx) | |
906 | { | |
907 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
908 | int i; | |
909 | ||
219c662d AD |
910 | assert(td->o.job_max_open_zones == 0 || td->num_open_zones <= td->o.job_max_open_zones); |
911 | assert(td->o.job_max_open_zones <= zbdi->max_open_zones); | |
912 | assert(zbdi->num_open_zones <= zbdi->max_open_zones); | |
59b07544 BVA |
913 | |
914 | for (i = 0; i < zbdi->num_open_zones; i++) | |
915 | if (zbdi->open_zones[i] == zone_idx) | |
916 | return true; | |
917 | ||
918 | return false; | |
919 | } | |
920 | ||
921 | /* | |
922 | * Open a ZBD zone if it was not yet open. Returns true if either the zone was | |
923 | * already open or if opening a new zone is allowed. Returns false if the zone | |
924 | * was not yet open and opening a new zone would cause the zone limit to be | |
925 | * exceeded. | |
926 | */ | |
927 | static bool zbd_open_zone(struct thread_data *td, const struct io_u *io_u, | |
928 | uint32_t zone_idx) | |
929 | { | |
930 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
931 | const struct fio_file *f = io_u->file; | |
932 | struct fio_zone_info *z = &f->zbd_info->zone_info[zone_idx]; | |
933 | bool res = true; | |
934 | ||
b7694961 | 935 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
59b07544 BVA |
936 | return false; |
937 | ||
938 | /* | |
939 | * Skip full zones with data verification enabled because resetting a | |
940 | * zone causes data loss and hence causes verification to fail. | |
941 | */ | |
942 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) | |
943 | return false; | |
944 | ||
59b07544 BVA |
945 | pthread_mutex_lock(&f->zbd_info->mutex); |
946 | if (is_zone_open(td, f, zone_idx)) | |
947 | goto out; | |
948 | res = false; | |
219c662d AD |
949 | /* Zero means no limit */ |
950 | if (td->o.job_max_open_zones > 0 && | |
951 | td->num_open_zones >= td->o.job_max_open_zones) | |
952 | goto out; | |
953 | if (f->zbd_info->num_open_zones >= f->zbd_info->max_open_zones) | |
59b07544 BVA |
954 | goto out; |
955 | dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx); | |
956 | f->zbd_info->open_zones[f->zbd_info->num_open_zones++] = zone_idx; | |
219c662d | 957 | td->num_open_zones++; |
59b07544 BVA |
958 | z->open = 1; |
959 | res = true; | |
960 | ||
961 | out: | |
962 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
963 | return res; | |
964 | } | |
965 | ||
6463db6c AD |
966 | /* Anything goes as long as it is not a constant. */ |
967 | static uint32_t pick_random_zone_idx(const struct fio_file *f, | |
968 | const struct io_u *io_u) | |
969 | { | |
970 | return io_u->offset * f->zbd_info->num_open_zones / f->real_file_size; | |
971 | } | |
972 | ||
59b07544 BVA |
973 | /* |
974 | * Modify the offset of an I/O unit that does not refer to an open zone such | |
975 | * that it refers to an open zone. Close an open zone and open a new zone if | |
976 | * necessary. This algorithm can only work correctly if all write pointers are | |
977 | * a multiple of the fio block size. The caller must neither hold z->mutex | |
978 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
979 | */ | |
379e5f09 BVA |
980 | static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, |
981 | struct io_u *io_u) | |
59b07544 BVA |
982 | { |
983 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 984 | struct fio_file *f = io_u->file; |
59b07544 BVA |
985 | struct fio_zone_info *z; |
986 | unsigned int open_zone_idx = -1; | |
987 | uint32_t zone_idx, new_zone_idx; | |
988 | int i; | |
989 | ||
990 | assert(is_valid_offset(f, io_u->offset)); | |
991 | ||
00ca8df5 | 992 | if (td->o.max_open_zones || td->o.job_max_open_zones) { |
59b07544 BVA |
993 | /* |
994 | * This statement accesses f->zbd_info->open_zones[] on purpose | |
995 | * without locking. | |
996 | */ | |
6463db6c | 997 | zone_idx = f->zbd_info->open_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 BVA |
998 | } else { |
999 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
1000 | } | |
fae3b9a0 AD |
1001 | if (zone_idx < f->min_zone) |
1002 | zone_idx = f->min_zone; | |
1003 | else if (zone_idx >= f->max_zone) | |
1004 | zone_idx = f->max_zone - 1; | |
59b07544 BVA |
1005 | dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", |
1006 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); | |
1007 | ||
1008 | /* | |
1009 | * Since z->mutex is the outer lock and f->zbd_info->mutex the inner | |
1010 | * lock it can happen that the state of the zone with index zone_idx | |
1011 | * has changed after 'z' has been assigned and before f->zbd_info->mutex | |
1012 | * has been obtained. Hence the loop. | |
1013 | */ | |
1014 | for (;;) { | |
6463db6c AD |
1015 | uint32_t tmp_idx; |
1016 | ||
59b07544 BVA |
1017 | z = &f->zbd_info->zone_info[zone_idx]; |
1018 | ||
fae3b9a0 | 1019 | zone_lock(td, f, z); |
59b07544 | 1020 | pthread_mutex_lock(&f->zbd_info->mutex); |
00ca8df5 | 1021 | if (td->o.max_open_zones == 0 && td->o.job_max_open_zones == 0) |
59b07544 BVA |
1022 | goto examine_zone; |
1023 | if (f->zbd_info->num_open_zones == 0) { | |
1024 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1025 | pthread_mutex_unlock(&z->mutex); | |
1026 | dprint(FD_ZBD, "%s(%s): no zones are open\n", | |
1027 | __func__, f->file_name); | |
1028 | return NULL; | |
1029 | } | |
6463db6c AD |
1030 | |
1031 | /* | |
1032 | * List of opened zones is per-device, shared across all threads. | |
1033 | * Start with quasi-random candidate zone. | |
1034 | * Ignore zones which don't belong to thread's offset/size area. | |
1035 | */ | |
1036 | open_zone_idx = pick_random_zone_idx(f, io_u); | |
59b07544 | 1037 | assert(open_zone_idx < f->zbd_info->num_open_zones); |
6463db6c AD |
1038 | tmp_idx = open_zone_idx; |
1039 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1040 | uint32_t tmpz; | |
1041 | ||
1042 | if (tmp_idx >= f->zbd_info->num_open_zones) | |
1043 | tmp_idx = 0; | |
1044 | tmpz = f->zbd_info->open_zones[tmp_idx]; | |
fae3b9a0 | 1045 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
6463db6c AD |
1046 | open_zone_idx = tmp_idx; |
1047 | goto found_candidate_zone; | |
1048 | } | |
1049 | ||
1050 | tmp_idx++; | |
1051 | } | |
1052 | ||
1053 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1054 | __func__, f->file_name); | |
adc6adcb DLM |
1055 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1056 | pthread_mutex_unlock(&z->mutex); | |
6463db6c AD |
1057 | return NULL; |
1058 | ||
1059 | found_candidate_zone: | |
59b07544 BVA |
1060 | new_zone_idx = f->zbd_info->open_zones[open_zone_idx]; |
1061 | if (new_zone_idx == zone_idx) | |
1062 | break; | |
1063 | zone_idx = new_zone_idx; | |
1064 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1065 | pthread_mutex_unlock(&z->mutex); | |
1066 | } | |
1067 | ||
1068 | /* Both z->mutex and f->zbd_info->mutex are held. */ | |
1069 | ||
1070 | examine_zone: | |
ee3696bd | 1071 | if (z->wp + min_bs <= (z+1)->start) { |
59b07544 BVA |
1072 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1073 | goto out; | |
1074 | } | |
1075 | dprint(FD_ZBD, "%s(%s): closing zone %d\n", __func__, f->file_name, | |
1076 | zone_idx); | |
00ca8df5 | 1077 | if (td->o.max_open_zones || td->o.job_max_open_zones) |
59b07544 BVA |
1078 | zbd_close_zone(td, f, open_zone_idx); |
1079 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1080 | ||
1081 | /* Only z->mutex is held. */ | |
1082 | ||
1083 | /* Zone 'z' is full, so try to open a new zone. */ | |
1084 | for (i = f->io_size / f->zbd_info->zone_size; i > 0; i--) { | |
1085 | zone_idx++; | |
1086 | pthread_mutex_unlock(&z->mutex); | |
1087 | z++; | |
ee3696bd | 1088 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1089 | /* Wrap-around. */ |
fae3b9a0 | 1090 | zone_idx = f->min_zone; |
59b07544 BVA |
1091 | z = &f->zbd_info->zone_info[zone_idx]; |
1092 | } | |
ee3696bd | 1093 | assert(is_valid_offset(f, z->start)); |
fae3b9a0 | 1094 | zone_lock(td, f, z); |
59b07544 BVA |
1095 | if (z->open) |
1096 | continue; | |
1097 | if (zbd_open_zone(td, io_u, zone_idx)) | |
1098 | goto out; | |
1099 | } | |
1100 | ||
1101 | /* Only z->mutex is held. */ | |
1102 | ||
1103 | /* Check whether the write fits in any of the already opened zones. */ | |
1104 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1105 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1106 | zone_idx = f->zbd_info->open_zones[i]; | |
fae3b9a0 AD |
1107 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1108 | continue; | |
59b07544 BVA |
1109 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1110 | pthread_mutex_unlock(&z->mutex); | |
1111 | ||
1112 | z = &f->zbd_info->zone_info[zone_idx]; | |
1113 | ||
fae3b9a0 | 1114 | zone_lock(td, f, z); |
ee3696bd | 1115 | if (z->wp + min_bs <= (z+1)->start) |
59b07544 BVA |
1116 | goto out; |
1117 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1118 | } | |
1119 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1120 | pthread_mutex_unlock(&z->mutex); | |
1121 | dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__, | |
1122 | f->file_name); | |
1123 | return NULL; | |
1124 | ||
1125 | out: | |
1126 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name, | |
1127 | zone_idx); | |
ee3696bd | 1128 | io_u->offset = z->start; |
59b07544 BVA |
1129 | return z; |
1130 | } | |
1131 | ||
bfbdd35b | 1132 | /* The caller must hold z->mutex. */ |
59b07544 BVA |
1133 | static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, |
1134 | struct io_u *io_u, | |
1135 | struct fio_zone_info *z) | |
bfbdd35b BVA |
1136 | { |
1137 | const struct fio_file *f = io_u->file; | |
1138 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
1139 | ||
59b07544 BVA |
1140 | if (!zbd_open_zone(td, io_u, z - f->zbd_info->zone_info)) { |
1141 | pthread_mutex_unlock(&z->mutex); | |
1142 | z = zbd_convert_to_open_zone(td, io_u); | |
1143 | assert(z); | |
1144 | } | |
1145 | ||
bfbdd35b | 1146 | if (z->verify_block * min_bs >= f->zbd_info->zone_size) |
a0c84dd4 JA |
1147 | log_err("%s: %d * %d >= %llu\n", f->file_name, z->verify_block, |
1148 | min_bs, (unsigned long long) f->zbd_info->zone_size); | |
ee3696bd | 1149 | io_u->offset = z->start + z->verify_block++ * min_bs; |
59b07544 | 1150 | return z; |
bfbdd35b BVA |
1151 | } |
1152 | ||
1153 | /* | |
1154 | * Find another zone for which @io_u fits below the write pointer. Start | |
1155 | * searching in zones @zb + 1 .. @zl and continue searching in zones | |
1156 | * @zf .. @zb - 1. | |
1157 | * | |
1158 | * Either returns NULL or returns a zone pointer and holds the mutex for that | |
1159 | * zone. | |
1160 | */ | |
1161 | static struct fio_zone_info * | |
1162 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, | |
1163 | struct fio_zone_info *zb, struct fio_zone_info *zl) | |
1164 | { | |
1165 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 1166 | struct fio_file *f = io_u->file; |
bfbdd35b BVA |
1167 | struct fio_zone_info *z1, *z2; |
1168 | const struct fio_zone_info *const zf = | |
1169 | &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)]; | |
1170 | ||
1171 | /* | |
1172 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1173 | * the nearest non-empty zone in case of random I/O. | |
1174 | */ | |
1175 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1176 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
fae3b9a0 | 1177 | zone_lock(td, f, z1); |
ee3696bd | 1178 | if (z1->start + min_bs <= z1->wp) |
bfbdd35b BVA |
1179 | return z1; |
1180 | pthread_mutex_unlock(&z1->mutex); | |
1181 | } else if (!td_random(td)) { | |
1182 | break; | |
1183 | } | |
1184 | if (td_random(td) && z2 >= zf && | |
b7694961 | 1185 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
fae3b9a0 | 1186 | zone_lock(td, f, z2); |
ee3696bd | 1187 | if (z2->start + min_bs <= z2->wp) |
bfbdd35b BVA |
1188 | return z2; |
1189 | pthread_mutex_unlock(&z2->mutex); | |
1190 | } | |
1191 | } | |
1192 | dprint(FD_ZBD, "%s: adjusting random read offset failed\n", | |
1193 | f->file_name); | |
1194 | return NULL; | |
1195 | } | |
1196 | ||
bfbdd35b | 1197 | /** |
d9ed3e63 | 1198 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1199 | * @io_u: I/O unit |
d9ed3e63 DLM |
1200 | * @success: Whether or not the I/O unit has been queued successfully |
1201 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1202 | * |
d9ed3e63 DLM |
1203 | * For write and trim operations, update the write pointer of the I/O unit |
1204 | * target zone. | |
bfbdd35b | 1205 | */ |
d9ed3e63 | 1206 | static void zbd_queue_io(struct io_u *io_u, int q, bool success) |
bfbdd35b | 1207 | { |
d9ed3e63 DLM |
1208 | const struct fio_file *f = io_u->file; |
1209 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b BVA |
1210 | struct fio_zone_info *z; |
1211 | uint32_t zone_idx; | |
d9ed3e63 | 1212 | uint64_t zone_end; |
bfbdd35b | 1213 | |
bfbdd35b BVA |
1214 | if (!zbd_info) |
1215 | return; | |
1216 | ||
d9ed3e63 | 1217 | zone_idx = zbd_zone_idx(f, io_u->offset); |
bfbdd35b | 1218 | assert(zone_idx < zbd_info->nr_zones); |
d9ed3e63 DLM |
1219 | z = &zbd_info->zone_info[zone_idx]; |
1220 | ||
b7694961 | 1221 | if (!zbd_zone_swr(z)) |
bfbdd35b | 1222 | return; |
d9ed3e63 | 1223 | |
bfbdd35b BVA |
1224 | if (!success) |
1225 | goto unlock; | |
d9ed3e63 DLM |
1226 | |
1227 | dprint(FD_ZBD, | |
1228 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
1229 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1230 | ||
bfbdd35b BVA |
1231 | switch (io_u->ddir) { |
1232 | case DDIR_WRITE: | |
d9ed3e63 DLM |
1233 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
1234 | (z + 1)->start); | |
a7c2b6fc BVA |
1235 | pthread_mutex_lock(&zbd_info->mutex); |
1236 | /* | |
1237 | * z->wp > zone_end means that one or more I/O errors | |
1238 | * have occurred. | |
1239 | */ | |
1240 | if (z->wp <= zone_end) | |
1241 | zbd_info->sectors_with_data += zone_end - z->wp; | |
1242 | pthread_mutex_unlock(&zbd_info->mutex); | |
bfbdd35b BVA |
1243 | z->wp = zone_end; |
1244 | break; | |
1245 | case DDIR_TRIM: | |
1246 | assert(z->wp == z->start); | |
1247 | break; | |
1248 | default: | |
1249 | break; | |
1250 | } | |
d9ed3e63 | 1251 | |
bfbdd35b | 1252 | unlock: |
d9ed3e63 DLM |
1253 | if (!success || q != FIO_Q_QUEUED) { |
1254 | /* BUSY or COMPLETED: unlock the zone */ | |
1255 | pthread_mutex_unlock(&z->mutex); | |
1256 | io_u->zbd_put_io = NULL; | |
1257 | } | |
1258 | } | |
1259 | ||
1260 | /** | |
1261 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1262 | * @io_u: I/O unit | |
1263 | */ | |
1264 | static void zbd_put_io(const struct io_u *io_u) | |
1265 | { | |
1266 | const struct fio_file *f = io_u->file; | |
1267 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1268 | struct fio_zone_info *z; | |
1269 | uint32_t zone_idx; | |
f59c2f8e | 1270 | int ret; |
d9ed3e63 DLM |
1271 | |
1272 | if (!zbd_info) | |
1273 | return; | |
615555bb | 1274 | |
d9ed3e63 DLM |
1275 | zone_idx = zbd_zone_idx(f, io_u->offset); |
1276 | assert(zone_idx < zbd_info->nr_zones); | |
1277 | z = &zbd_info->zone_info[zone_idx]; | |
1278 | ||
b7694961 | 1279 | if (!zbd_zone_swr(z)) |
d9ed3e63 DLM |
1280 | return; |
1281 | ||
1282 | dprint(FD_ZBD, | |
1283 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
1284 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1285 | ||
f59c2f8e AD |
1286 | ret = pthread_mutex_unlock(&z->mutex); |
1287 | assert(ret == 0); | |
d9ed3e63 | 1288 | zbd_check_swd(f); |
bfbdd35b BVA |
1289 | } |
1290 | ||
9d87c646 DLM |
1291 | /* |
1292 | * Windows and MacOS do not define this. | |
1293 | */ | |
1294 | #ifndef EREMOTEIO | |
1295 | #define EREMOTEIO 121 /* POSIX value */ | |
1296 | #endif | |
1297 | ||
bfbdd35b BVA |
1298 | bool zbd_unaligned_write(int error_code) |
1299 | { | |
1300 | switch (error_code) { | |
1301 | case EIO: | |
1302 | case EREMOTEIO: | |
1303 | return true; | |
1304 | } | |
1305 | return false; | |
1306 | } | |
1307 | ||
4d37720a DLM |
1308 | /** |
1309 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1310 | * @td: FIO thread data. | |
1311 | * @io_u: FIO I/O unit. | |
1312 | * | |
1313 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1314 | * no more IO can be performed in the current zone. | |
1315 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1316 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1317 | * - For write workloads, zoneskip is applied when the zone is full. | |
1318 | * This applies only to read and write operations. | |
1319 | */ | |
1320 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1321 | { | |
1322 | struct fio_file *f = io_u->file; | |
1323 | enum fio_ddir ddir = io_u->ddir; | |
1324 | struct fio_zone_info *z; | |
1325 | uint32_t zone_idx; | |
1326 | ||
1327 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1328 | assert(td->o.zone_size); | |
1329 | ||
1330 | /* | |
1331 | * zone_skip is valid only for sequential workloads. | |
1332 | */ | |
1333 | if (td_random(td) || !td->o.zone_skip) | |
1334 | return; | |
1335 | ||
1336 | /* | |
1337 | * It is time to switch to a new zone if: | |
1338 | * - zone_bytes == zone_size bytes have already been accessed | |
1339 | * - The last position reached the end of the current zone. | |
1340 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1341 | * reached the zone write pointer. | |
1342 | */ | |
1343 | zone_idx = zbd_zone_idx(f, f->last_pos[ddir]); | |
1344 | z = &f->zbd_info->zone_info[zone_idx]; | |
1345 | ||
1346 | if (td->zone_bytes >= td->o.zone_size || | |
1347 | f->last_pos[ddir] >= (z+1)->start || | |
1348 | (ddir == DDIR_READ && | |
1349 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1350 | /* | |
1351 | * Skip zones. | |
1352 | */ | |
1353 | td->zone_bytes = 0; | |
1354 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1355 | ||
1356 | /* | |
1357 | * Wrap from the beginning, if we exceed the file size | |
1358 | */ | |
1359 | if (f->file_offset >= f->real_file_size) | |
1360 | f->file_offset = get_start_offset(td, f); | |
1361 | ||
1362 | f->last_pos[ddir] = f->file_offset; | |
1363 | td->io_skip_bytes += td->o.zone_skip; | |
1364 | } | |
1365 | } | |
1366 | ||
c65057f9 SK |
1367 | /** |
1368 | * zbd_adjust_ddir - Adjust an I/O direction for zonedmode=zbd. | |
1369 | * | |
1370 | * @td: FIO thread data. | |
1371 | * @io_u: FIO I/O unit. | |
1372 | * @ddir: I/O direction before adjustment. | |
1373 | * | |
1374 | * Return adjusted I/O direction. | |
1375 | */ | |
1376 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1377 | enum fio_ddir ddir) | |
1378 | { | |
1379 | /* | |
1380 | * In case read direction is chosen for the first random I/O, fio with | |
1381 | * zonemode=zbd stops because no data can be read from zoned block | |
1382 | * devices with all empty zones. Overwrite the first I/O direction as | |
1383 | * write to make sure data to read exists. | |
1384 | */ | |
731461cc | 1385 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1386 | return ddir; |
1387 | ||
1388 | if (io_u->file->zbd_info->sectors_with_data || | |
1389 | td->o.read_beyond_wp) | |
1390 | return DDIR_READ; | |
1391 | ||
1392 | return DDIR_WRITE; | |
1393 | } | |
1394 | ||
bfbdd35b BVA |
1395 | /** |
1396 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1397 | * @td: FIO thread data. | |
1398 | * @io_u: FIO I/O unit. | |
1399 | * | |
1400 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1401 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1402 | * corresponds to io_u->offset at the end of this function. | |
1403 | */ | |
1404 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1405 | { | |
b7694961 | 1406 | struct fio_file *f = io_u->file; |
bfbdd35b | 1407 | uint32_t zone_idx_b; |
de65f7b7 | 1408 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b BVA |
1409 | uint32_t orig_len = io_u->buflen; |
1410 | uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
1411 | uint64_t new_len; | |
1412 | int64_t range; | |
1413 | ||
1414 | if (!f->zbd_info) | |
1415 | return io_u_accept; | |
1416 | ||
adc6adcb | 1417 | assert(min_bs); |
bfbdd35b BVA |
1418 | assert(is_valid_offset(f, io_u->offset)); |
1419 | assert(io_u->buflen); | |
1420 | zone_idx_b = zbd_zone_idx(f, io_u->offset); | |
1421 | zb = &f->zbd_info->zone_info[zone_idx_b]; | |
de65f7b7 | 1422 | orig_zb = zb; |
bfbdd35b BVA |
1423 | |
1424 | /* Accept the I/O offset for conventional zones. */ | |
b7694961 | 1425 | if (!zbd_zone_swr(zb)) |
bfbdd35b BVA |
1426 | return io_u_accept; |
1427 | ||
1428 | /* | |
1429 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1430 | * is enabled. | |
1431 | */ | |
b7694961 | 1432 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1433 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1434 | return io_u_accept; | |
1435 | ||
615555bb BVA |
1436 | zbd_check_swd(f); |
1437 | ||
fae3b9a0 | 1438 | zone_lock(td, f, zb); |
6f0c6085 | 1439 | |
bfbdd35b BVA |
1440 | switch (io_u->ddir) { |
1441 | case DDIR_READ: | |
1442 | if (td->runstate == TD_VERIFYING) { | |
ecc9e8f7 DLM |
1443 | if (td_write(td)) |
1444 | zb = zbd_replay_write_order(td, io_u, zb); | |
bfbdd35b BVA |
1445 | goto accept; |
1446 | } | |
1447 | /* | |
de65f7b7 DLM |
1448 | * Check that there is enough written data in the zone to do an |
1449 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1450 | * the I/O. | |
bfbdd35b | 1451 | */ |
b7694961 | 1452 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1453 | zb->wp - zb->start : 0; |
de65f7b7 | 1454 | if (range < min_bs || |
ee3696bd | 1455 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
bfbdd35b BVA |
1456 | pthread_mutex_unlock(&zb->mutex); |
1457 | zl = &f->zbd_info->zone_info[zbd_zone_idx(f, | |
1458 | f->file_offset + f->io_size)]; | |
1459 | zb = zbd_find_zone(td, io_u, zb, zl); | |
1460 | if (!zb) { | |
1461 | dprint(FD_ZBD, | |
1462 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1463 | f->file_name, io_u->offset, | |
1464 | io_u->buflen); | |
1465 | goto eof; | |
1466 | } | |
de65f7b7 DLM |
1467 | /* |
1468 | * zbd_find_zone() returned a zone with a range of at | |
1469 | * least min_bs. | |
1470 | */ | |
ee3696bd | 1471 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1472 | assert(range >= min_bs); |
1473 | ||
1474 | if (!td_random(td)) | |
ee3696bd | 1475 | io_u->offset = zb->start; |
bfbdd35b | 1476 | } |
de65f7b7 DLM |
1477 | /* |
1478 | * Make sure the I/O is within the zone valid data range while | |
1479 | * maximizing the I/O size and preserving randomness. | |
1480 | */ | |
1481 | if (range <= io_u->buflen) | |
ee3696bd | 1482 | io_u->offset = zb->start; |
de65f7b7 | 1483 | else if (td_random(td)) |
ee3696bd DLM |
1484 | io_u->offset = zb->start + |
1485 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 DLM |
1486 | (range - io_u->buflen)) / min_bs * min_bs; |
1487 | /* | |
1488 | * Make sure the I/O does not cross over the zone wp position. | |
1489 | */ | |
1490 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1491 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1492 | new_len = new_len / min_bs * min_bs; |
1493 | if (new_len < io_u->buflen) { | |
1494 | io_u->buflen = new_len; | |
1495 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1496 | orig_len, io_u->buflen); | |
bfbdd35b | 1497 | } |
ee3696bd DLM |
1498 | assert(zb->start <= io_u->offset); |
1499 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
bfbdd35b BVA |
1500 | goto accept; |
1501 | case DDIR_WRITE: | |
ee3696bd | 1502 | if (io_u->buflen > f->zbd_info->zone_size) |
bfbdd35b | 1503 | goto eof; |
59b07544 BVA |
1504 | if (!zbd_open_zone(td, io_u, zone_idx_b)) { |
1505 | pthread_mutex_unlock(&zb->mutex); | |
1506 | zb = zbd_convert_to_open_zone(td, io_u); | |
1507 | if (!zb) | |
1508 | goto eof; | |
1509 | zone_idx_b = zb - f->zbd_info->zone_info; | |
1510 | } | |
a7c2b6fc BVA |
1511 | /* Check whether the zone reset threshold has been exceeded */ |
1512 | if (td->o.zrf.u.f) { | |
ee3696bd | 1513 | if (f->zbd_info->sectors_with_data >= |
a7c2b6fc BVA |
1514 | f->io_size * td->o.zrt.u.f && |
1515 | zbd_dec_and_reset_write_cnt(td, f)) { | |
1516 | zb->reset_zone = 1; | |
1517 | } | |
1518 | } | |
bfbdd35b BVA |
1519 | /* Reset the zone pointer if necessary */ |
1520 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
1521 | assert(td->o.verify == VERIFY_NONE); | |
1522 | /* | |
1523 | * Since previous write requests may have been submitted | |
1524 | * asynchronously and since we will submit the zone | |
1525 | * reset synchronously, wait until previously submitted | |
1526 | * write requests have completed before issuing a | |
1527 | * zone reset. | |
1528 | */ | |
1529 | io_u_quiesce(td); | |
1530 | zb->reset_zone = 0; | |
1531 | if (zbd_reset_zone(td, f, zb) < 0) | |
1532 | goto eof; | |
1533 | } | |
1534 | /* Make writes occur at the write pointer */ | |
1535 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 1536 | io_u->offset = zb->wp; |
bfbdd35b BVA |
1537 | if (!is_valid_offset(f, io_u->offset)) { |
1538 | dprint(FD_ZBD, "Dropped request with offset %llu\n", | |
1539 | io_u->offset); | |
1540 | goto eof; | |
1541 | } | |
1542 | /* | |
1543 | * Make sure that the buflen is a multiple of the minimal | |
1544 | * block size. Give up if shrinking would make the request too | |
1545 | * small. | |
1546 | */ | |
1547 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1548 | (zb + 1)->start - io_u->offset); |
bfbdd35b BVA |
1549 | new_len = new_len / min_bs * min_bs; |
1550 | if (new_len == io_u->buflen) | |
1551 | goto accept; | |
1552 | if (new_len >= min_bs) { | |
1553 | io_u->buflen = new_len; | |
1554 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1555 | orig_len, io_u->buflen); | |
1556 | goto accept; | |
1557 | } | |
1558 | log_err("Zone remainder %lld smaller than minimum block size %d\n", | |
ee3696bd | 1559 | ((zb + 1)->start - io_u->offset), |
bfbdd35b BVA |
1560 | min_bs); |
1561 | goto eof; | |
1562 | case DDIR_TRIM: | |
1563 | /* fall-through */ | |
1564 | case DDIR_SYNC: | |
1565 | case DDIR_DATASYNC: | |
1566 | case DDIR_SYNC_FILE_RANGE: | |
1567 | case DDIR_WAIT: | |
1568 | case DDIR_LAST: | |
1569 | case DDIR_INVAL: | |
1570 | goto accept; | |
1571 | } | |
1572 | ||
1573 | assert(false); | |
1574 | ||
1575 | accept: | |
1576 | assert(zb); | |
b7694961 | 1577 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
1578 | assert(!io_u->zbd_queue_io); |
1579 | assert(!io_u->zbd_put_io); | |
1580 | io_u->zbd_queue_io = zbd_queue_io; | |
1581 | io_u->zbd_put_io = zbd_put_io; | |
bfbdd35b BVA |
1582 | return io_u_accept; |
1583 | ||
1584 | eof: | |
1585 | if (zb) | |
1586 | pthread_mutex_unlock(&zb->mutex); | |
1587 | return io_u_eof; | |
1588 | } | |
fd5d733f BVA |
1589 | |
1590 | /* Return a string with ZBD statistics */ | |
1591 | char *zbd_write_status(const struct thread_stat *ts) | |
1592 | { | |
1593 | char *res; | |
1594 | ||
a0c84dd4 | 1595 | if (asprintf(&res, "; %llu zone resets", (unsigned long long) ts->nr_zone_resets) < 0) |
fd5d733f BVA |
1596 | return NULL; |
1597 | return res; | |
1598 | } |