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 | ||
410a071c DLM |
25 | static bool is_valid_offset(const struct fio_file *f, uint64_t offset) |
26 | { | |
27 | return (uint64_t)(offset - f->file_offset) < f->io_size; | |
28 | } | |
29 | ||
30 | static inline unsigned int zbd_zone_nr(const struct fio_file *f, | |
31 | struct fio_zone_info *zone) | |
32 | { | |
33 | return zone - f->zbd_info->zone_info; | |
34 | } | |
35 | ||
36 | /** | |
37 | * zbd_zone_idx - convert an offset into a zone number | |
38 | * @f: file pointer. | |
39 | * @offset: offset in bytes. If this offset is in the first zone_size bytes | |
40 | * past the disk size then the index of the sentinel is returned. | |
41 | */ | |
42 | static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) | |
43 | { | |
44 | uint32_t zone_idx; | |
45 | ||
46 | if (f->zbd_info->zone_size_log2 > 0) | |
47 | zone_idx = offset >> f->zbd_info->zone_size_log2; | |
48 | else | |
49 | zone_idx = offset / f->zbd_info->zone_size; | |
50 | ||
51 | return min(zone_idx, f->zbd_info->nr_zones); | |
52 | } | |
53 | ||
54 | /** | |
55 | * zbd_zone_end - Return zone end location | |
56 | * @z: zone info pointer. | |
57 | */ | |
58 | static inline uint64_t zbd_zone_end(const struct fio_zone_info *z) | |
59 | { | |
60 | return (z+1)->start; | |
61 | } | |
62 | ||
63 | /** | |
64 | * zbd_zone_capacity_end - Return zone capacity limit end location | |
65 | * @z: zone info pointer. | |
66 | */ | |
67 | static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z) | |
68 | { | |
69 | return z->start + z->capacity; | |
70 | } | |
71 | ||
72 | /** | |
73 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
74 | * @f: file pointer. | |
75 | * @z: zone info pointer. | |
76 | * @required: minimum number of bytes that must remain in a zone. | |
77 | * | |
78 | * The caller must hold z->mutex. | |
79 | */ | |
80 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
81 | uint64_t required) | |
82 | { | |
83 | assert((required & 511) == 0); | |
84 | ||
85 | return z->has_wp && | |
86 | z->wp + required > zbd_zone_capacity_end(z); | |
87 | } | |
88 | ||
89 | static void zone_lock(struct thread_data *td, const struct fio_file *f, | |
90 | struct fio_zone_info *z) | |
91 | { | |
92 | struct zoned_block_device_info *zbd = f->zbd_info; | |
93 | uint32_t nz = z - zbd->zone_info; | |
94 | ||
95 | /* A thread should never lock zones outside its working area. */ | |
96 | assert(f->min_zone <= nz && nz < f->max_zone); | |
97 | ||
98 | assert(z->has_wp); | |
99 | ||
100 | /* | |
101 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
102 | * is changed or when io_u completes and zbd_put_io() executed. | |
103 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
104 | * other waiting for zone locks when building an io_u batch, first | |
105 | * only trylock the zone. If the zone is already locked by another job, | |
106 | * process the currently queued I/Os so that I/O progress is made and | |
107 | * zones unlocked. | |
108 | */ | |
109 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
110 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
111 | io_u_quiesce(td); | |
112 | pthread_mutex_lock(&z->mutex); | |
113 | } | |
114 | } | |
115 | ||
116 | static inline void zone_unlock(struct fio_zone_info *z) | |
117 | { | |
118 | int ret; | |
119 | ||
120 | assert(z->has_wp); | |
121 | ret = pthread_mutex_unlock(&z->mutex); | |
122 | assert(!ret); | |
123 | } | |
124 | ||
125 | static inline struct fio_zone_info *get_zone(const struct fio_file *f, | |
126 | unsigned int zone_nr) | |
127 | { | |
128 | return &f->zbd_info->zone_info[zone_nr]; | |
129 | } | |
130 | ||
b7694961 DLM |
131 | /** |
132 | * zbd_get_zoned_model - Get a device zoned model | |
133 | * @td: FIO thread data | |
134 | * @f: FIO file for which to get model information | |
135 | */ | |
38334c13 DLM |
136 | static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, |
137 | enum zbd_zoned_model *model) | |
b7694961 DLM |
138 | { |
139 | int ret; | |
140 | ||
50cc48d5 NC |
141 | if (f->filetype == FIO_TYPE_PIPE) { |
142 | log_err("zonemode=zbd does not support pipes\n"); | |
143 | return -EINVAL; | |
144 | } | |
145 | ||
9db0cde8 NC |
146 | /* If regular file, always emulate zones inside the file. */ |
147 | if (f->filetype == FIO_TYPE_FILE) { | |
148 | *model = ZBD_NONE; | |
149 | return 0; | |
150 | } | |
151 | ||
6c5b11d3 DLM |
152 | if (td->io_ops && td->io_ops->get_zoned_model) |
153 | ret = td->io_ops->get_zoned_model(td, f, model); | |
154 | else | |
155 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
156 | if (ret < 0) { |
157 | td_verror(td, errno, "get zoned model failed"); | |
158 | log_err("%s: get zoned model failed (%d).\n", | |
159 | f->file_name, errno); | |
160 | } | |
161 | ||
162 | return ret; | |
163 | } | |
164 | ||
165 | /** | |
166 | * zbd_report_zones - Get zone information | |
167 | * @td: FIO thread data. | |
168 | * @f: FIO file for which to get zone information | |
169 | * @offset: offset from which to report zones | |
170 | * @zones: Array of struct zbd_zone | |
171 | * @nr_zones: Size of @zones array | |
172 | * | |
173 | * Get zone information into @zones starting from the zone at offset @offset | |
174 | * for the device specified by @f. | |
175 | * | |
176 | * Returns the number of zones reported upon success and a negative error code | |
177 | * upon failure. If the zone report is empty, always assume an error (device | |
178 | * problem) and return -EIO. | |
179 | */ | |
38334c13 DLM |
180 | static int zbd_report_zones(struct thread_data *td, struct fio_file *f, |
181 | uint64_t offset, struct zbd_zone *zones, | |
182 | unsigned int nr_zones) | |
b7694961 DLM |
183 | { |
184 | int ret; | |
185 | ||
6c5b11d3 DLM |
186 | if (td->io_ops && td->io_ops->report_zones) |
187 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
188 | else | |
189 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
190 | if (ret < 0) { |
191 | td_verror(td, errno, "report zones failed"); | |
ee5e3436 SK |
192 | log_err("%s: report zones from sector %"PRIu64" failed (%d).\n", |
193 | f->file_name, offset >> 9, errno); | |
b7694961 DLM |
194 | } else if (ret == 0) { |
195 | td_verror(td, errno, "Empty zone report"); | |
ee5e3436 SK |
196 | log_err("%s: report zones from sector %"PRIu64" is empty.\n", |
197 | f->file_name, offset >> 9); | |
b7694961 DLM |
198 | ret = -EIO; |
199 | } | |
200 | ||
201 | return ret; | |
202 | } | |
203 | ||
204 | /** | |
205 | * zbd_reset_wp - reset the write pointer of a range of zones | |
206 | * @td: FIO thread data. | |
207 | * @f: FIO file for which to reset zones | |
208 | * @offset: Starting offset of the first zone to reset | |
209 | * @length: Length of the range of zones to reset | |
210 | * | |
211 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
212 | * Returns 0 upon success and a negative error code upon failure. | |
213 | */ | |
38334c13 DLM |
214 | static int zbd_reset_wp(struct thread_data *td, struct fio_file *f, |
215 | uint64_t offset, uint64_t length) | |
b7694961 DLM |
216 | { |
217 | int ret; | |
218 | ||
6c5b11d3 DLM |
219 | if (td->io_ops && td->io_ops->reset_wp) |
220 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
221 | else | |
222 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
223 | if (ret < 0) { |
224 | td_verror(td, errno, "resetting wp failed"); | |
ee5e3436 SK |
225 | log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n", |
226 | f->file_name, length >> 9, offset >> 9, errno); | |
b7694961 DLM |
227 | } |
228 | ||
229 | return ret; | |
230 | } | |
231 | ||
410a071c DLM |
232 | /** |
233 | * zbd_reset_zone - reset the write pointer of a single zone | |
234 | * @td: FIO thread data. | |
235 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
236 | * @z: Zone to reset. | |
237 | * | |
238 | * Returns 0 upon success and a negative error code upon failure. | |
239 | * | |
240 | * The caller must hold z->mutex. | |
241 | */ | |
242 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, | |
243 | struct fio_zone_info *z) | |
244 | { | |
245 | uint64_t offset = z->start; | |
246 | uint64_t length = (z+1)->start - offset; | |
247 | uint64_t data_in_zone = z->wp - z->start; | |
248 | int ret = 0; | |
249 | ||
250 | if (!data_in_zone) | |
251 | return 0; | |
252 | ||
253 | assert(is_valid_offset(f, offset + length - 1)); | |
254 | ||
255 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name, | |
256 | zbd_zone_nr(f, z)); | |
257 | switch (f->zbd_info->model) { | |
258 | case ZBD_HOST_AWARE: | |
259 | case ZBD_HOST_MANAGED: | |
260 | ret = zbd_reset_wp(td, f, offset, length); | |
261 | if (ret < 0) | |
262 | return ret; | |
263 | break; | |
264 | default: | |
265 | break; | |
266 | } | |
267 | ||
268 | pthread_mutex_lock(&f->zbd_info->mutex); | |
269 | f->zbd_info->sectors_with_data -= data_in_zone; | |
270 | f->zbd_info->wp_sectors_with_data -= data_in_zone; | |
271 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
272 | z->wp = z->start; | |
273 | z->verify_block = 0; | |
274 | ||
275 | td->ts.nr_zone_resets++; | |
276 | ||
277 | return ret; | |
278 | } | |
279 | ||
280 | /** | |
281 | * zbd_close_zone - Remove a zone from the open zones array. | |
282 | * @td: FIO thread data. | |
283 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
284 | * @zone_idx: Index of the zone to remove. | |
285 | * | |
286 | * The caller must hold f->zbd_info->mutex. | |
287 | */ | |
288 | static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, | |
289 | unsigned int zone_idx) | |
290 | { | |
291 | uint32_t open_zone_idx = 0; | |
292 | ||
293 | for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) { | |
294 | if (f->zbd_info->open_zones[open_zone_idx] == zone_idx) | |
295 | break; | |
296 | } | |
297 | if (open_zone_idx == f->zbd_info->num_open_zones) | |
298 | return; | |
299 | ||
300 | dprint(FD_ZBD, "%s: closing zone %d\n", f->file_name, zone_idx); | |
301 | memmove(f->zbd_info->open_zones + open_zone_idx, | |
302 | f->zbd_info->open_zones + open_zone_idx + 1, | |
303 | (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) * | |
304 | sizeof(f->zbd_info->open_zones[0])); | |
305 | f->zbd_info->num_open_zones--; | |
306 | td->num_open_zones--; | |
307 | get_zone(f, zone_idx)->open = 0; | |
308 | } | |
309 | ||
310 | /** | |
311 | * zbd_reset_zones - Reset a range of zones. | |
312 | * @td: fio thread data. | |
313 | * @f: fio file for which to reset zones | |
314 | * @zb: first zone to reset. | |
315 | * @ze: first zone not to reset. | |
316 | * | |
317 | * Returns 0 upon success and 1 upon failure. | |
318 | */ | |
319 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
320 | struct fio_zone_info *const zb, | |
321 | struct fio_zone_info *const ze) | |
322 | { | |
323 | struct fio_zone_info *z; | |
324 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
325 | int res = 0; | |
326 | ||
327 | assert(min_bs); | |
328 | ||
329 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name, | |
330 | zbd_zone_nr(f, zb), zbd_zone_nr(f, ze)); | |
331 | for (z = zb; z < ze; z++) { | |
332 | uint32_t nz = zbd_zone_nr(f, z); | |
333 | ||
334 | if (!z->has_wp) | |
335 | continue; | |
336 | zone_lock(td, f, z); | |
337 | pthread_mutex_lock(&f->zbd_info->mutex); | |
338 | zbd_close_zone(td, f, nz); | |
339 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
340 | if (z->wp != z->start) { | |
341 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
342 | f->file_name, zbd_zone_nr(f, z)); | |
343 | if (zbd_reset_zone(td, f, z) < 0) | |
344 | res = 1; | |
345 | } | |
346 | zone_unlock(z); | |
347 | } | |
348 | ||
349 | return res; | |
350 | } | |
351 | ||
d2f442bc NC |
352 | /** |
353 | * zbd_get_max_open_zones - Get the maximum number of open zones | |
354 | * @td: FIO thread data | |
355 | * @f: FIO file for which to get max open zones | |
356 | * @max_open_zones: Upon success, result will be stored here. | |
357 | * | |
358 | * A @max_open_zones value set to zero means no limit. | |
359 | * | |
360 | * Returns 0 upon success and a negative error code upon failure. | |
361 | */ | |
38334c13 DLM |
362 | static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f, |
363 | unsigned int *max_open_zones) | |
d2f442bc NC |
364 | { |
365 | int ret; | |
366 | ||
367 | if (td->io_ops && td->io_ops->get_max_open_zones) | |
368 | ret = td->io_ops->get_max_open_zones(td, f, max_open_zones); | |
369 | else | |
370 | ret = blkzoned_get_max_open_zones(td, f, max_open_zones); | |
371 | if (ret < 0) { | |
372 | td_verror(td, errno, "get max open zones failed"); | |
373 | log_err("%s: get max open zones failed (%d).\n", | |
374 | f->file_name, errno); | |
375 | } | |
376 | ||
377 | return ret; | |
378 | } | |
379 | ||
bfbdd35b | 380 | /** |
410a071c DLM |
381 | * zbd_open_zone - Add a zone to the array of open zones. |
382 | * @td: fio thread data. | |
383 | * @f: fio file that has the open zones to add. | |
384 | * @zone_idx: Index of the zone to add. | |
bfbdd35b | 385 | * |
410a071c DLM |
386 | * Open a ZBD zone if it is not already open. Returns true if either the zone |
387 | * was already open or if the zone was successfully added to the array of open | |
388 | * zones without exceeding the maximum number of open zones. Returns false if | |
389 | * the zone was not already open and opening the zone would cause the zone limit | |
390 | * to be exceeded. | |
bfbdd35b | 391 | */ |
410a071c DLM |
392 | static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f, |
393 | uint32_t zone_idx) | |
1f57803b | 394 | { |
410a071c DLM |
395 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; |
396 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
397 | struct fio_zone_info *z = get_zone(f, zone_idx); | |
398 | bool res = true; | |
fae3b9a0 | 399 | |
410a071c DLM |
400 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
401 | return false; | |
43bcbd5b | 402 | |
1f57803b | 403 | /* |
410a071c DLM |
404 | * Skip full zones with data verification enabled because resetting a |
405 | * zone causes data loss and hence causes verification to fail. | |
1f57803b | 406 | */ |
410a071c DLM |
407 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) |
408 | return false; | |
4d4c71e6 | 409 | |
410a071c DLM |
410 | /* |
411 | * zbdi->max_open_zones == 0 means that there is no limit on the maximum | |
412 | * number of open zones. In this case, do no track open zones in | |
413 | * zbdi->open_zones array. | |
414 | */ | |
415 | if (!zbdi->max_open_zones) | |
416 | return true; | |
4d4c71e6 | 417 | |
410a071c | 418 | pthread_mutex_lock(&zbdi->mutex); |
b5a0f7ce DLM |
419 | |
420 | if (z->open) { | |
410a071c | 421 | /* |
b5a0f7ce DLM |
422 | * If the zone is going to be completely filled by writes |
423 | * already in-flight, handle it as a full zone instead of an | |
424 | * open zone. | |
410a071c DLM |
425 | */ |
426 | if (z->wp >= zbd_zone_capacity_end(z)) | |
427 | res = false; | |
428 | goto out; | |
429 | } | |
430 | res = false; | |
431 | /* Zero means no limit */ | |
432 | if (td->o.job_max_open_zones > 0 && | |
433 | td->num_open_zones >= td->o.job_max_open_zones) | |
434 | goto out; | |
435 | if (zbdi->num_open_zones >= zbdi->max_open_zones) | |
436 | goto out; | |
437 | dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx); | |
438 | zbdi->open_zones[zbdi->num_open_zones++] = zone_idx; | |
439 | td->num_open_zones++; | |
440 | z->open = 1; | |
441 | res = true; | |
bfbdd35b | 442 | |
410a071c DLM |
443 | out: |
444 | pthread_mutex_unlock(&zbdi->mutex); | |
445 | return res; | |
923f7c1e DF |
446 | } |
447 | ||
bfbdd35b BVA |
448 | /* Verify whether direct I/O is used for all host-managed zoned drives. */ |
449 | static bool zbd_using_direct_io(void) | |
450 | { | |
451 | struct thread_data *td; | |
452 | struct fio_file *f; | |
453 | int i, j; | |
454 | ||
455 | for_each_td(td, i) { | |
456 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) | |
457 | continue; | |
458 | for_each_file(td, f, j) { | |
459 | if (f->zbd_info && | |
b7694961 | 460 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
461 | return false; |
462 | } | |
463 | } | |
464 | ||
465 | return true; | |
466 | } | |
467 | ||
468 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
469 | static bool zbd_is_seq_job(struct fio_file *f) | |
470 | { | |
471 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
472 | ||
473 | assert(f->zbd_info); | |
474 | if (f->io_size == 0) | |
475 | return false; | |
476 | zone_idx_b = zbd_zone_idx(f, f->file_offset); | |
477 | zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1); | |
478 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) | |
be7a6bae | 479 | if (get_zone(f, zone_idx)->has_wp) |
bfbdd35b BVA |
480 | return true; |
481 | ||
482 | return false; | |
483 | } | |
484 | ||
485 | /* | |
486 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
487 | */ | |
488 | static bool zbd_verify_sizes(void) | |
489 | { | |
490 | const struct fio_zone_info *z; | |
491 | struct thread_data *td; | |
492 | struct fio_file *f; | |
493 | uint64_t new_offset, new_end; | |
494 | uint32_t zone_idx; | |
495 | int i, j; | |
496 | ||
497 | for_each_td(td, i) { | |
498 | for_each_file(td, f, j) { | |
499 | if (!f->zbd_info) | |
500 | continue; | |
501 | if (f->file_offset >= f->real_file_size) | |
502 | continue; | |
503 | if (!zbd_is_seq_job(f)) | |
504 | continue; | |
4d37720a | 505 | |
fc52e449 DLM |
506 | if (!td->o.zone_size) { |
507 | td->o.zone_size = f->zbd_info->zone_size; | |
508 | if (!td->o.zone_size) { | |
509 | log_err("%s: invalid 0 zone size\n", | |
510 | f->file_name); | |
511 | return false; | |
512 | } | |
971d6a22 | 513 | } else if (td->o.zone_size != f->zbd_info->zone_size) { |
ee5e3436 SK |
514 | log_err("%s: job parameter zonesize %llu does not match disk zone size %"PRIu64".\n", |
515 | f->file_name, td->o.zone_size, | |
516 | f->zbd_info->zone_size); | |
971d6a22 | 517 | return false; |
fc52e449 DLM |
518 | } |
519 | ||
4316f640 | 520 | if (td->o.zone_skip % td->o.zone_size) { |
4d37720a | 521 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", |
ee5e3436 SK |
522 | f->file_name, td->o.zone_skip, |
523 | td->o.zone_size); | |
4d37720a DLM |
524 | return false; |
525 | } | |
526 | ||
bfbdd35b | 527 | zone_idx = zbd_zone_idx(f, f->file_offset); |
923f7c1e | 528 | z = get_zone(f, zone_idx); |
f09a7773 PL |
529 | if ((f->file_offset != z->start) && |
530 | (td->o.td_ddir != TD_DDIR_READ)) { | |
236d23a8 | 531 | new_offset = zbd_zone_end(z); |
bfbdd35b BVA |
532 | if (new_offset >= f->file_offset + f->io_size) { |
533 | log_info("%s: io_size must be at least one zone\n", | |
534 | f->file_name); | |
535 | return false; | |
536 | } | |
ee5e3436 SK |
537 | log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n", |
538 | f->file_name, f->file_offset, | |
539 | new_offset); | |
bfbdd35b BVA |
540 | f->io_size -= (new_offset - f->file_offset); |
541 | f->file_offset = new_offset; | |
542 | } | |
543 | zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size); | |
923f7c1e | 544 | z = get_zone(f, zone_idx); |
ee3696bd | 545 | new_end = z->start; |
f09a7773 PL |
546 | if ((td->o.td_ddir != TD_DDIR_READ) && |
547 | (f->file_offset + f->io_size != new_end)) { | |
bfbdd35b BVA |
548 | if (new_end <= f->file_offset) { |
549 | log_info("%s: io_size must be at least one zone\n", | |
550 | f->file_name); | |
551 | return false; | |
552 | } | |
ee5e3436 SK |
553 | log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n", |
554 | f->file_name, f->io_size, | |
555 | new_end - f->file_offset); | |
bfbdd35b BVA |
556 | f->io_size = new_end - f->file_offset; |
557 | } | |
558 | } | |
559 | } | |
560 | ||
561 | return true; | |
562 | } | |
563 | ||
564 | static bool zbd_verify_bs(void) | |
565 | { | |
566 | struct thread_data *td; | |
567 | struct fio_file *f; | |
bfbdd35b BVA |
568 | int i, j, k; |
569 | ||
570 | for_each_td(td, i) { | |
e3be810b SK |
571 | if (td_trim(td) && |
572 | (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] || | |
573 | td->o.bssplit_nr[DDIR_TRIM])) { | |
574 | log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n"); | |
575 | return false; | |
576 | } | |
bfbdd35b | 577 | for_each_file(td, f, j) { |
1ddd225e AD |
578 | uint64_t zone_size; |
579 | ||
bfbdd35b BVA |
580 | if (!f->zbd_info) |
581 | continue; | |
582 | zone_size = f->zbd_info->zone_size; | |
e3be810b | 583 | if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) { |
ee5e3436 | 584 | log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n", |
e3be810b | 585 | f->file_name, td->o.bs[DDIR_TRIM], |
ee5e3436 | 586 | zone_size); |
e3be810b SK |
587 | return false; |
588 | } | |
59f94d26 | 589 | for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) { |
bfbdd35b | 590 | if (td->o.verify != VERIFY_NONE && |
ee3696bd | 591 | zone_size % td->o.bs[k] != 0) { |
ee5e3436 | 592 | log_info("%s: block size %llu is not a divisor of the zone size %"PRIu64"\n", |
bfbdd35b | 593 | f->file_name, td->o.bs[k], |
ee5e3436 | 594 | zone_size); |
bfbdd35b BVA |
595 | return false; |
596 | } | |
597 | } | |
598 | } | |
599 | } | |
600 | return true; | |
601 | } | |
602 | ||
bfbdd35b BVA |
603 | static int ilog2(uint64_t i) |
604 | { | |
605 | int log = -1; | |
606 | ||
607 | while (i) { | |
608 | i >>= 1; | |
609 | log++; | |
610 | } | |
611 | return log; | |
612 | } | |
613 | ||
614 | /* | |
615 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
616 | * allows to execute a ZBD workload against a non-ZBD device. | |
617 | */ | |
618 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
619 | { | |
620 | uint32_t nr_zones; | |
621 | struct fio_zone_info *p; | |
a4b7f12b | 622 | uint64_t zone_size = td->o.zone_size; |
b8dd9750 | 623 | uint64_t zone_capacity = td->o.zone_capacity; |
bfbdd35b | 624 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
625 | int i; |
626 | ||
a4b7f12b | 627 | if (zone_size == 0) { |
9db0cde8 | 628 | log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n", |
a4b7f12b DLM |
629 | f->file_name); |
630 | return 1; | |
631 | } | |
632 | ||
633 | if (zone_size < 512) { | |
634 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
635 | f->file_name); | |
636 | return 1; | |
637 | } | |
638 | ||
b8dd9750 HH |
639 | if (zone_capacity == 0) |
640 | zone_capacity = zone_size; | |
641 | ||
642 | if (zone_capacity > zone_size) { | |
643 | log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", | |
ee5e3436 | 644 | f->file_name, td->o.zone_capacity, td->o.zone_size); |
b8dd9750 HH |
645 | return 1; |
646 | } | |
647 | ||
9db0cde8 NC |
648 | if (f->real_file_size < zone_size) { |
649 | log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n", | |
650 | f->file_name, f->real_file_size, zone_size); | |
651 | return -EINVAL; | |
652 | } | |
653 | ||
ee3696bd | 654 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
655 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
656 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
657 | if (!zbd_info) | |
658 | return -ENOMEM; | |
659 | ||
44ec32cb | 660 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
661 | zbd_info->refcount = 1; |
662 | p = &zbd_info->zone_info[0]; | |
663 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
664 | mutex_init_pshared_with_type(&p->mutex, |
665 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b | 666 | p->start = i * zone_size; |
b14651a2 | 667 | p->wp = p->start; |
b7694961 DLM |
668 | p->type = ZBD_ZONE_TYPE_SWR; |
669 | p->cond = ZBD_ZONE_COND_EMPTY; | |
b8dd9750 | 670 | p->capacity = zone_capacity; |
be7a6bae | 671 | p->has_wp = 1; |
bfbdd35b BVA |
672 | } |
673 | /* a sentinel */ | |
674 | p->start = nr_zones * zone_size; | |
675 | ||
676 | f->zbd_info = zbd_info; | |
677 | f->zbd_info->zone_size = zone_size; | |
678 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 679 | ilog2(zone_size) : 0; |
bfbdd35b | 680 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
681 | return 0; |
682 | } | |
683 | ||
684 | /* | |
b7694961 DLM |
685 | * Maximum number of zones to report in one operation. |
686 | */ | |
687 | #define ZBD_REPORT_MAX_ZONES 8192U | |
688 | ||
689 | /* | |
690 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
691 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
692 | */ |
693 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
694 | { | |
b7694961 DLM |
695 | int nr_zones, nrz; |
696 | struct zbd_zone *zones, *z; | |
bfbdd35b | 697 | struct fio_zone_info *p; |
b7694961 | 698 | uint64_t zone_size, offset; |
bfbdd35b | 699 | struct zoned_block_device_info *zbd_info = NULL; |
d060babc | 700 | int i, j, ret = -ENOMEM; |
bfbdd35b | 701 | |
b7694961 DLM |
702 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
703 | if (!zones) | |
bfbdd35b BVA |
704 | goto out; |
705 | ||
b7694961 DLM |
706 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
707 | if (nrz < 0) { | |
708 | ret = nrz; | |
709 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
710 | f->file_name, -ret); | |
711 | goto out; | |
bfbdd35b BVA |
712 | } |
713 | ||
b7694961 | 714 | zone_size = zones[0].len; |
ee3696bd | 715 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
716 | |
717 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
718 | td->o.zone_size = zone_size; |
719 | } else if (td->o.zone_size != zone_size) { | |
ee5e3436 SK |
720 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n", |
721 | f->file_name, td->o.zone_size, zone_size); | |
bfbdd35b | 722 | ret = -EINVAL; |
b7694961 | 723 | goto out; |
bfbdd35b BVA |
724 | } |
725 | ||
ee5e3436 SK |
726 | dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", f->file_name, |
727 | nr_zones, zone_size / 1024); | |
bfbdd35b BVA |
728 | |
729 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
730 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
bfbdd35b | 731 | if (!zbd_info) |
b7694961 | 732 | goto out; |
44ec32cb | 733 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
734 | zbd_info->refcount = 1; |
735 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
736 | for (offset = 0, j = 0; j < nr_zones;) { |
737 | z = &zones[0]; | |
738 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
739 | mutex_init_pshared_with_type(&p->mutex, |
740 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 741 | p->start = z->start; |
236d23a8 | 742 | p->capacity = z->capacity; |
bfbdd35b | 743 | switch (z->cond) { |
b7694961 DLM |
744 | case ZBD_ZONE_COND_NOT_WP: |
745 | case ZBD_ZONE_COND_FULL: | |
236d23a8 | 746 | p->wp = p->start + p->capacity; |
bfbdd35b BVA |
747 | break; |
748 | default: | |
749 | assert(z->start <= z->wp); | |
b7694961 DLM |
750 | assert(z->wp <= z->start + zone_size); |
751 | p->wp = z->wp; | |
bfbdd35b BVA |
752 | break; |
753 | } | |
be7a6bae DF |
754 | |
755 | switch (z->type) { | |
756 | case ZBD_ZONE_TYPE_SWR: | |
757 | p->has_wp = 1; | |
758 | break; | |
759 | default: | |
760 | p->has_wp = 0; | |
761 | } | |
bfbdd35b BVA |
762 | p->type = z->type; |
763 | p->cond = z->cond; | |
be7a6bae | 764 | |
bfbdd35b BVA |
765 | if (j > 0 && p->start != p[-1].start + zone_size) { |
766 | log_info("%s: invalid zone data\n", | |
767 | f->file_name); | |
768 | ret = -EINVAL; | |
b7694961 | 769 | goto out; |
bfbdd35b BVA |
770 | } |
771 | } | |
772 | z--; | |
b7694961 | 773 | offset = z->start + z->len; |
bfbdd35b BVA |
774 | if (j >= nr_zones) |
775 | break; | |
6c3f1cc1 DF |
776 | nrz = zbd_report_zones(td, f, offset, zones, |
777 | min((uint32_t)(nr_zones - j), | |
778 | ZBD_REPORT_MAX_ZONES)); | |
b7694961 DLM |
779 | if (nrz < 0) { |
780 | ret = nrz; | |
ee5e3436 SK |
781 | log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n", |
782 | offset, f->file_name, -ret); | |
b7694961 | 783 | goto out; |
bfbdd35b BVA |
784 | } |
785 | } | |
b7694961 | 786 | |
bfbdd35b | 787 | /* a sentinel */ |
b7694961 | 788 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
789 | |
790 | f->zbd_info = zbd_info; | |
791 | f->zbd_info->zone_size = zone_size; | |
792 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 793 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
794 | f->zbd_info->nr_zones = nr_zones; |
795 | zbd_info = NULL; | |
796 | ret = 0; | |
797 | ||
bfbdd35b | 798 | out: |
b7694961 DLM |
799 | sfree(zbd_info); |
800 | free(zones); | |
bfbdd35b BVA |
801 | return ret; |
802 | } | |
803 | ||
d2f442bc NC |
804 | static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f) |
805 | { | |
806 | struct zoned_block_device_info *zbd = f->zbd_info; | |
807 | unsigned int max_open_zones; | |
808 | int ret; | |
809 | ||
575686bb | 810 | if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) { |
d2f442bc NC |
811 | /* Only host-managed devices have a max open limit */ |
812 | zbd->max_open_zones = td->o.max_open_zones; | |
813 | goto out; | |
814 | } | |
815 | ||
816 | /* If host-managed, get the max open limit */ | |
817 | ret = zbd_get_max_open_zones(td, f, &max_open_zones); | |
818 | if (ret) | |
819 | return ret; | |
820 | ||
821 | if (!max_open_zones) { | |
822 | /* No device limit */ | |
823 | zbd->max_open_zones = td->o.max_open_zones; | |
824 | } else if (!td->o.max_open_zones) { | |
825 | /* No user limit. Set limit to device limit */ | |
826 | zbd->max_open_zones = max_open_zones; | |
827 | } else if (td->o.max_open_zones <= max_open_zones) { | |
828 | /* Both user limit and dev limit. User limit not too large */ | |
829 | zbd->max_open_zones = td->o.max_open_zones; | |
830 | } else { | |
831 | /* Both user limit and dev limit. User limit too large */ | |
832 | td_verror(td, EINVAL, | |
833 | "Specified --max_open_zones is too large"); | |
834 | log_err("Specified --max_open_zones (%d) is larger than max (%u)\n", | |
835 | td->o.max_open_zones, max_open_zones); | |
836 | return -EINVAL; | |
837 | } | |
838 | ||
839 | out: | |
840 | /* Ensure that the limit is not larger than FIO's internal limit */ | |
b346af90 NC |
841 | if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) { |
842 | td_verror(td, EINVAL, "'max_open_zones' value is too large"); | |
843 | log_err("'max_open_zones' value is larger than %u\n", ZBD_MAX_OPEN_ZONES); | |
844 | return -EINVAL; | |
845 | } | |
846 | ||
d2f442bc NC |
847 | dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n", |
848 | f->file_name, zbd->max_open_zones); | |
849 | ||
850 | return 0; | |
851 | } | |
852 | ||
bfbdd35b BVA |
853 | /* |
854 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
855 | * | |
856 | * Returns 0 upon success and a negative error code upon failure. | |
857 | */ | |
379e5f09 | 858 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 859 | { |
b7694961 DLM |
860 | enum zbd_zoned_model zbd_model; |
861 | int ret; | |
bfbdd35b BVA |
862 | |
863 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
864 | ||
b7694961 DLM |
865 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
866 | if (ret) | |
867 | return ret; | |
868 | ||
bfbdd35b | 869 | switch (zbd_model) { |
b7694961 DLM |
870 | case ZBD_HOST_AWARE: |
871 | case ZBD_HOST_MANAGED: | |
bfbdd35b | 872 | ret = parse_zone_info(td, f); |
d2f442bc NC |
873 | if (ret) |
874 | return ret; | |
bfbdd35b | 875 | break; |
b7694961 | 876 | case ZBD_NONE: |
bfbdd35b | 877 | ret = init_zone_info(td, f); |
d2f442bc NC |
878 | if (ret) |
879 | return ret; | |
bfbdd35b | 880 | break; |
b7694961 DLM |
881 | default: |
882 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
883 | log_err("Unsupported zoned model\n"); | |
884 | return -EINVAL; | |
bfbdd35b | 885 | } |
b7694961 | 886 | |
2c7dd23e | 887 | assert(f->zbd_info); |
d2f442bc NC |
888 | f->zbd_info->model = zbd_model; |
889 | ||
890 | ret = zbd_set_max_open_zones(td, f); | |
891 | if (ret) { | |
892 | zbd_free_zone_info(f); | |
893 | return ret; | |
219c662d | 894 | } |
d2f442bc NC |
895 | |
896 | return 0; | |
bfbdd35b BVA |
897 | } |
898 | ||
899 | void zbd_free_zone_info(struct fio_file *f) | |
900 | { | |
901 | uint32_t refcount; | |
902 | ||
3c1dc34c | 903 | assert(f->zbd_info); |
bfbdd35b BVA |
904 | |
905 | pthread_mutex_lock(&f->zbd_info->mutex); | |
906 | refcount = --f->zbd_info->refcount; | |
907 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
908 | ||
909 | assert((int32_t)refcount >= 0); | |
910 | if (refcount == 0) | |
911 | sfree(f->zbd_info); | |
912 | f->zbd_info = NULL; | |
913 | } | |
914 | ||
915 | /* | |
916 | * Initialize f->zbd_info. | |
917 | * | |
918 | * Returns 0 upon success and a negative error code upon failure. | |
919 | * | |
920 | * Note: this function can only work correctly if it is called before the first | |
921 | * fio fork() call. | |
922 | */ | |
923 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
924 | { | |
925 | struct thread_data *td2; | |
926 | struct fio_file *f2; | |
927 | int i, j, ret; | |
928 | ||
929 | for_each_td(td2, i) { | |
930 | for_each_file(td2, f2, j) { | |
931 | if (td2 == td && f2 == file) | |
932 | continue; | |
933 | if (!f2->zbd_info || | |
934 | strcmp(f2->file_name, file->file_name) != 0) | |
935 | continue; | |
936 | file->zbd_info = f2->zbd_info; | |
937 | file->zbd_info->refcount++; | |
938 | return 0; | |
939 | } | |
940 | } | |
941 | ||
942 | ret = zbd_create_zone_info(td, file); | |
943 | if (ret < 0) | |
c5837eec | 944 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
bfbdd35b BVA |
945 | return ret; |
946 | } | |
947 | ||
8f39afa7 | 948 | int zbd_init_files(struct thread_data *td) |
bfbdd35b BVA |
949 | { |
950 | struct fio_file *f; | |
951 | int i; | |
952 | ||
953 | for_each_file(td, f, i) { | |
a4b7f12b | 954 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 955 | return 1; |
bfbdd35b | 956 | } |
8f39afa7 AD |
957 | return 0; |
958 | } | |
959 | ||
960 | void zbd_recalc_options_with_zone_granularity(struct thread_data *td) | |
961 | { | |
962 | struct fio_file *f; | |
963 | int i; | |
964 | ||
965 | for_each_file(td, f, i) { | |
966 | struct zoned_block_device_info *zbd = f->zbd_info; | |
967 | // zonemode=strided doesn't get per-file zone size. | |
968 | uint64_t zone_size = zbd ? zbd->zone_size : td->o.zone_size; | |
969 | ||
970 | if (zone_size == 0) | |
971 | continue; | |
972 | ||
973 | if (td->o.size_nz > 0) { | |
974 | td->o.size = td->o.size_nz * zone_size; | |
975 | } | |
976 | if (td->o.io_size_nz > 0) { | |
977 | td->o.io_size = td->o.io_size_nz * zone_size; | |
978 | } | |
979 | if (td->o.start_offset_nz > 0) { | |
980 | td->o.start_offset = td->o.start_offset_nz * zone_size; | |
981 | } | |
982 | if (td->o.offset_increment_nz > 0) { | |
983 | td->o.offset_increment = td->o.offset_increment_nz * zone_size; | |
984 | } | |
985 | if (td->o.zone_skip_nz > 0) { | |
986 | td->o.zone_skip = td->o.zone_skip_nz * zone_size; | |
987 | } | |
988 | } | |
989 | } | |
990 | ||
991 | int zbd_setup_files(struct thread_data *td) | |
992 | { | |
993 | struct fio_file *f; | |
994 | int i; | |
bfbdd35b BVA |
995 | |
996 | if (!zbd_using_direct_io()) { | |
997 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
998 | return 1; | |
999 | } | |
1000 | ||
1001 | if (!zbd_verify_sizes()) | |
1002 | return 1; | |
1003 | ||
1004 | if (!zbd_verify_bs()) | |
1005 | return 1; | |
1006 | ||
219c662d AD |
1007 | for_each_file(td, f, i) { |
1008 | struct zoned_block_device_info *zbd = f->zbd_info; | |
954217b9 SK |
1009 | struct fio_zone_info *z; |
1010 | int zi; | |
219c662d | 1011 | |
5ddf46d0 | 1012 | assert(zbd); |
219c662d | 1013 | |
f952800a SK |
1014 | f->min_zone = zbd_zone_idx(f, f->file_offset); |
1015 | f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size); | |
1016 | ||
1017 | /* | |
1018 | * When all zones in the I/O range are conventional, io_size | |
1019 | * can be smaller than zone size, making min_zone the same | |
1020 | * as max_zone. This is why the assert below needs to be made | |
1021 | * conditional. | |
1022 | */ | |
1023 | if (zbd_is_seq_job(f)) | |
1024 | assert(f->min_zone < f->max_zone); | |
1025 | ||
219c662d AD |
1026 | if (td->o.max_open_zones > 0 && |
1027 | zbd->max_open_zones != td->o.max_open_zones) { | |
1028 | log_err("Different 'max_open_zones' values\n"); | |
1029 | return 1; | |
1030 | } | |
b346af90 NC |
1031 | |
1032 | /* | |
1033 | * The per job max open zones limit cannot be used without a | |
1034 | * global max open zones limit. (As the tracking of open zones | |
1035 | * is disabled when there is no global max open zones limit.) | |
1036 | */ | |
1037 | if (td->o.job_max_open_zones && !zbd->max_open_zones) { | |
1038 | log_err("'job_max_open_zones' cannot be used without a global open zones limit\n"); | |
219c662d AD |
1039 | return 1; |
1040 | } | |
954217b9 | 1041 | |
ea51055c NC |
1042 | /* |
1043 | * zbd->max_open_zones is the global limit shared for all jobs | |
1044 | * that target the same zoned block device. Force sync the per | |
1045 | * thread global limit with the actual global limit. (The real | |
1046 | * per thread/job limit is stored in td->o.job_max_open_zones). | |
1047 | */ | |
1048 | td->o.max_open_zones = zbd->max_open_zones; | |
1049 | ||
954217b9 SK |
1050 | for (zi = f->min_zone; zi < f->max_zone; zi++) { |
1051 | z = &zbd->zone_info[zi]; | |
1052 | if (z->cond != ZBD_ZONE_COND_IMP_OPEN && | |
1053 | z->cond != ZBD_ZONE_COND_EXP_OPEN) | |
1054 | continue; | |
1055 | if (zbd_open_zone(td, f, zi)) | |
1056 | continue; | |
1057 | /* | |
1058 | * If the number of open zones exceeds specified limits, | |
1059 | * reset all extra open zones. | |
1060 | */ | |
1061 | if (zbd_reset_zone(td, f, z) < 0) { | |
1062 | log_err("Failed to reest zone %d\n", zi); | |
1063 | return 1; | |
1064 | } | |
1065 | } | |
219c662d AD |
1066 | } |
1067 | ||
bfbdd35b BVA |
1068 | return 0; |
1069 | } | |
1070 | ||
a7c2b6fc BVA |
1071 | /* |
1072 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
1073 | * zone reset. | |
1074 | */ | |
1bb1bcad AD |
1075 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
1076 | const struct fio_file *f) | |
a7c2b6fc BVA |
1077 | { |
1078 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
1079 | ||
a7c2b6fc BVA |
1080 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
1081 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
1082 | } |
1083 | ||
1084 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
1085 | const struct fio_file *f) | |
1086 | { | |
1087 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1088 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
1089 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1090 | } | |
1091 | ||
1092 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
1093 | const struct fio_file *f) | |
1094 | { | |
1095 | uint32_t write_cnt = 0; | |
1096 | ||
1097 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1098 | assert(f->zbd_info->write_cnt); | |
1099 | if (f->zbd_info->write_cnt) | |
1100 | write_cnt = --f->zbd_info->write_cnt; | |
1101 | if (write_cnt == 0) | |
1bb1bcad | 1102 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
1103 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1104 | ||
1105 | return write_cnt == 0; | |
1106 | } | |
1107 | ||
91d25131 BVA |
1108 | enum swd_action { |
1109 | CHECK_SWD, | |
1110 | SET_SWD, | |
1111 | }; | |
1112 | ||
1113 | /* Calculate the number of sectors with data (swd) and perform action 'a' */ | |
11afb212 DF |
1114 | static uint64_t zbd_process_swd(struct thread_data *td, |
1115 | const struct fio_file *f, enum swd_action a) | |
615555bb | 1116 | { |
615555bb BVA |
1117 | struct fio_zone_info *zb, *ze, *z; |
1118 | uint64_t swd = 0; | |
403ea795 | 1119 | uint64_t wp_swd = 0; |
615555bb | 1120 | |
923f7c1e DF |
1121 | zb = get_zone(f, f->min_zone); |
1122 | ze = get_zone(f, f->max_zone); | |
615555bb | 1123 | for (z = zb; z < ze; z++) { |
403ea795 | 1124 | if (z->has_wp) { |
11afb212 | 1125 | zone_lock(td, f, z); |
403ea795 SK |
1126 | wp_swd += z->wp - z->start; |
1127 | } | |
615555bb BVA |
1128 | swd += z->wp - z->start; |
1129 | } | |
1130 | pthread_mutex_lock(&f->zbd_info->mutex); | |
91d25131 BVA |
1131 | switch (a) { |
1132 | case CHECK_SWD: | |
1133 | assert(f->zbd_info->sectors_with_data == swd); | |
403ea795 | 1134 | assert(f->zbd_info->wp_sectors_with_data == wp_swd); |
91d25131 BVA |
1135 | break; |
1136 | case SET_SWD: | |
1137 | f->zbd_info->sectors_with_data = swd; | |
403ea795 | 1138 | f->zbd_info->wp_sectors_with_data = wp_swd; |
91d25131 BVA |
1139 | break; |
1140 | } | |
615555bb BVA |
1141 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1142 | for (z = zb; z < ze; z++) | |
403ea795 SK |
1143 | if (z->has_wp) |
1144 | zone_unlock(z); | |
91d25131 BVA |
1145 | |
1146 | return swd; | |
1147 | } | |
1148 | ||
1149 | /* | |
1150 | * The swd check is useful for debugging but takes too much time to leave | |
1151 | * it enabled all the time. Hence it is disabled by default. | |
1152 | */ | |
1153 | static const bool enable_check_swd = false; | |
1154 | ||
403ea795 | 1155 | /* Check whether the values of zbd_info.*sectors_with_data are correct. */ |
11afb212 | 1156 | static void zbd_check_swd(struct thread_data *td, const struct fio_file *f) |
91d25131 BVA |
1157 | { |
1158 | if (!enable_check_swd) | |
1159 | return; | |
1160 | ||
11afb212 | 1161 | zbd_process_swd(td, f, CHECK_SWD); |
615555bb BVA |
1162 | } |
1163 | ||
bfbdd35b BVA |
1164 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
1165 | { | |
91d25131 | 1166 | struct fio_zone_info *zb, *ze; |
11afb212 | 1167 | uint64_t swd; |
bfbdd35b | 1168 | |
767d1372 | 1169 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
1170 | return; |
1171 | ||
923f7c1e DF |
1172 | zb = get_zone(f, f->min_zone); |
1173 | ze = get_zone(f, f->max_zone); | |
11afb212 DF |
1174 | swd = zbd_process_swd(td, f, SET_SWD); |
1175 | dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name, | |
1176 | swd); | |
bfbdd35b BVA |
1177 | /* |
1178 | * If data verification is enabled reset the affected zones before | |
1179 | * writing any data to avoid that a zone reset has to be issued while | |
1180 | * writing data, which causes data loss. | |
1181 | */ | |
57f88219 SK |
1182 | if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING) |
1183 | zbd_reset_zones(td, f, zb, ze); | |
a7c2b6fc | 1184 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
1185 | } |
1186 | ||
cb765e41 | 1187 | /* Return random zone index for one of the open zones. */ |
6463db6c AD |
1188 | static uint32_t pick_random_zone_idx(const struct fio_file *f, |
1189 | const struct io_u *io_u) | |
1190 | { | |
cb765e41 AK |
1191 | return (io_u->offset - f->file_offset) * f->zbd_info->num_open_zones / |
1192 | f->io_size; | |
6463db6c AD |
1193 | } |
1194 | ||
0f77c977 SK |
1195 | static bool any_io_in_flight(void) |
1196 | { | |
1197 | struct thread_data *td; | |
1198 | int i; | |
1199 | ||
1200 | for_each_td(td, i) { | |
1201 | if (td->io_u_in_flight) | |
1202 | return true; | |
1203 | } | |
1204 | ||
1205 | return false; | |
1206 | } | |
1207 | ||
59b07544 BVA |
1208 | /* |
1209 | * Modify the offset of an I/O unit that does not refer to an open zone such | |
1210 | * that it refers to an open zone. Close an open zone and open a new zone if | |
21c0c884 SK |
1211 | * necessary. The open zone is searched across sequential zones. |
1212 | * This algorithm can only work correctly if all write pointers are | |
59b07544 BVA |
1213 | * a multiple of the fio block size. The caller must neither hold z->mutex |
1214 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
1215 | */ | |
379e5f09 BVA |
1216 | static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, |
1217 | struct io_u *io_u) | |
59b07544 | 1218 | { |
07fc3f57 | 1219 | const uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
fae3b9a0 | 1220 | struct fio_file *f = io_u->file; |
af94a8c3 | 1221 | struct zoned_block_device_info *zbdi = f->zbd_info; |
59b07544 BVA |
1222 | struct fio_zone_info *z; |
1223 | unsigned int open_zone_idx = -1; | |
1224 | uint32_t zone_idx, new_zone_idx; | |
1225 | int i; | |
b2da58c4 | 1226 | bool wait_zone_close; |
0f77c977 SK |
1227 | bool in_flight; |
1228 | bool should_retry = true; | |
59b07544 BVA |
1229 | |
1230 | assert(is_valid_offset(f, io_u->offset)); | |
1231 | ||
ea51055c | 1232 | if (zbdi->max_open_zones || td->o.job_max_open_zones) { |
59b07544 | 1233 | /* |
af94a8c3 | 1234 | * This statement accesses zbdi->open_zones[] on purpose |
59b07544 BVA |
1235 | * without locking. |
1236 | */ | |
af94a8c3 | 1237 | zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 BVA |
1238 | } else { |
1239 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
1240 | } | |
fae3b9a0 AD |
1241 | if (zone_idx < f->min_zone) |
1242 | zone_idx = f->min_zone; | |
1243 | else if (zone_idx >= f->max_zone) | |
1244 | zone_idx = f->max_zone - 1; | |
59b07544 BVA |
1245 | dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", |
1246 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); | |
1247 | ||
1248 | /* | |
af94a8c3 | 1249 | * Since z->mutex is the outer lock and zbdi->mutex the inner |
59b07544 | 1250 | * lock it can happen that the state of the zone with index zone_idx |
af94a8c3 | 1251 | * has changed after 'z' has been assigned and before zbdi->mutex |
59b07544 BVA |
1252 | * has been obtained. Hence the loop. |
1253 | */ | |
1254 | for (;;) { | |
6463db6c AD |
1255 | uint32_t tmp_idx; |
1256 | ||
923f7c1e | 1257 | z = get_zone(f, zone_idx); |
14351148 DF |
1258 | if (z->has_wp) |
1259 | zone_lock(td, f, z); | |
af94a8c3 | 1260 | pthread_mutex_lock(&zbdi->mutex); |
14351148 DF |
1261 | if (z->has_wp) { |
1262 | if (z->cond != ZBD_ZONE_COND_OFFLINE && | |
ea51055c | 1263 | zbdi->max_open_zones == 0 && td->o.job_max_open_zones == 0) |
14351148 | 1264 | goto examine_zone; |
af94a8c3 | 1265 | if (zbdi->num_open_zones == 0) { |
14351148 DF |
1266 | dprint(FD_ZBD, "%s(%s): no zones are open\n", |
1267 | __func__, f->file_name); | |
1268 | goto open_other_zone; | |
1269 | } | |
59b07544 | 1270 | } |
6463db6c AD |
1271 | |
1272 | /* | |
1273 | * List of opened zones is per-device, shared across all threads. | |
1274 | * Start with quasi-random candidate zone. | |
1275 | * Ignore zones which don't belong to thread's offset/size area. | |
1276 | */ | |
1277 | open_zone_idx = pick_random_zone_idx(f, io_u); | |
1a8d510a | 1278 | assert(!open_zone_idx || |
af94a8c3 | 1279 | open_zone_idx < zbdi->num_open_zones); |
6463db6c | 1280 | tmp_idx = open_zone_idx; |
af94a8c3 | 1281 | for (i = 0; i < zbdi->num_open_zones; i++) { |
6463db6c AD |
1282 | uint32_t tmpz; |
1283 | ||
af94a8c3 | 1284 | if (tmp_idx >= zbdi->num_open_zones) |
6463db6c | 1285 | tmp_idx = 0; |
af94a8c3 | 1286 | tmpz = zbdi->open_zones[tmp_idx]; |
fae3b9a0 | 1287 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
6463db6c AD |
1288 | open_zone_idx = tmp_idx; |
1289 | goto found_candidate_zone; | |
1290 | } | |
1291 | ||
1292 | tmp_idx++; | |
1293 | } | |
1294 | ||
1295 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1296 | __func__, f->file_name); | |
af94a8c3 | 1297 | pthread_mutex_unlock(&zbdi->mutex); |
14351148 DF |
1298 | if (z->has_wp) |
1299 | zone_unlock(z); | |
6463db6c AD |
1300 | return NULL; |
1301 | ||
1302 | found_candidate_zone: | |
af94a8c3 | 1303 | new_zone_idx = zbdi->open_zones[open_zone_idx]; |
59b07544 BVA |
1304 | if (new_zone_idx == zone_idx) |
1305 | break; | |
1306 | zone_idx = new_zone_idx; | |
af94a8c3 | 1307 | pthread_mutex_unlock(&zbdi->mutex); |
14351148 DF |
1308 | if (z->has_wp) |
1309 | zone_unlock(z); | |
59b07544 BVA |
1310 | } |
1311 | ||
af94a8c3 | 1312 | /* Both z->mutex and zbdi->mutex are held. */ |
59b07544 BVA |
1313 | |
1314 | examine_zone: | |
236d23a8 | 1315 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) { |
af94a8c3 | 1316 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1317 | goto out; |
1318 | } | |
b2da58c4 SK |
1319 | |
1320 | open_other_zone: | |
1321 | /* Check if number of open zones reaches one of limits. */ | |
1322 | wait_zone_close = | |
af94a8c3 | 1323 | zbdi->num_open_zones == f->max_zone - f->min_zone || |
ea51055c NC |
1324 | (zbdi->max_open_zones && |
1325 | zbdi->num_open_zones == zbdi->max_open_zones) || | |
b2da58c4 SK |
1326 | (td->o.job_max_open_zones && |
1327 | td->num_open_zones == td->o.job_max_open_zones); | |
1328 | ||
af94a8c3 | 1329 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1330 | |
1331 | /* Only z->mutex is held. */ | |
1332 | ||
b2da58c4 SK |
1333 | /* |
1334 | * When number of open zones reaches to one of limits, wait for | |
1335 | * zone close before opening a new zone. | |
1336 | */ | |
1337 | if (wait_zone_close) { | |
1338 | dprint(FD_ZBD, "%s(%s): quiesce to allow open zones to close\n", | |
1339 | __func__, f->file_name); | |
1340 | io_u_quiesce(td); | |
1341 | } | |
1342 | ||
0f77c977 | 1343 | retry: |
59b07544 | 1344 | /* Zone 'z' is full, so try to open a new zone. */ |
af94a8c3 | 1345 | for (i = f->io_size / zbdi->zone_size; i > 0; i--) { |
59b07544 | 1346 | zone_idx++; |
21c0c884 SK |
1347 | if (z->has_wp) |
1348 | zone_unlock(z); | |
59b07544 | 1349 | z++; |
ee3696bd | 1350 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1351 | /* Wrap-around. */ |
fae3b9a0 | 1352 | zone_idx = f->min_zone; |
923f7c1e | 1353 | z = get_zone(f, zone_idx); |
59b07544 | 1354 | } |
ee3696bd | 1355 | assert(is_valid_offset(f, z->start)); |
21c0c884 SK |
1356 | if (!z->has_wp) |
1357 | continue; | |
fae3b9a0 | 1358 | zone_lock(td, f, z); |
59b07544 BVA |
1359 | if (z->open) |
1360 | continue; | |
954217b9 | 1361 | if (zbd_open_zone(td, f, zone_idx)) |
59b07544 BVA |
1362 | goto out; |
1363 | } | |
1364 | ||
1365 | /* Only z->mutex is held. */ | |
1366 | ||
1367 | /* Check whether the write fits in any of the already opened zones. */ | |
af94a8c3 NC |
1368 | pthread_mutex_lock(&zbdi->mutex); |
1369 | for (i = 0; i < zbdi->num_open_zones; i++) { | |
1370 | zone_idx = zbdi->open_zones[i]; | |
fae3b9a0 AD |
1371 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1372 | continue; | |
af94a8c3 | 1373 | pthread_mutex_unlock(&zbdi->mutex); |
4d4c71e6 | 1374 | zone_unlock(z); |
59b07544 | 1375 | |
923f7c1e | 1376 | z = get_zone(f, zone_idx); |
59b07544 | 1377 | |
fae3b9a0 | 1378 | zone_lock(td, f, z); |
236d23a8 | 1379 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) |
59b07544 | 1380 | goto out; |
af94a8c3 | 1381 | pthread_mutex_lock(&zbdi->mutex); |
59b07544 | 1382 | } |
0f77c977 SK |
1383 | |
1384 | /* | |
1385 | * When any I/O is in-flight or when all I/Os in-flight get completed, | |
1386 | * the I/Os might have closed zones then retry the steps to open a zone. | |
1387 | * Before retry, call io_u_quiesce() to complete in-flight writes. | |
1388 | */ | |
1389 | in_flight = any_io_in_flight(); | |
1390 | if (in_flight || should_retry) { | |
1391 | dprint(FD_ZBD, "%s(%s): wait zone close and retry open zones\n", | |
1392 | __func__, f->file_name); | |
1393 | pthread_mutex_unlock(&zbdi->mutex); | |
1394 | zone_unlock(z); | |
1395 | io_u_quiesce(td); | |
1396 | zone_lock(td, f, z); | |
1397 | should_retry = in_flight; | |
1398 | goto retry; | |
1399 | } | |
1400 | ||
af94a8c3 | 1401 | pthread_mutex_unlock(&zbdi->mutex); |
4d4c71e6 | 1402 | zone_unlock(z); |
59b07544 BVA |
1403 | dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__, |
1404 | f->file_name); | |
1405 | return NULL; | |
1406 | ||
1407 | out: | |
1408 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name, | |
1409 | zone_idx); | |
ee3696bd | 1410 | io_u->offset = z->start; |
21c0c884 | 1411 | assert(z->has_wp); |
8a866de7 | 1412 | assert(z->cond != ZBD_ZONE_COND_OFFLINE); |
59b07544 BVA |
1413 | return z; |
1414 | } | |
1415 | ||
bfbdd35b | 1416 | /* The caller must hold z->mutex. */ |
59b07544 BVA |
1417 | static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, |
1418 | struct io_u *io_u, | |
1419 | struct fio_zone_info *z) | |
bfbdd35b BVA |
1420 | { |
1421 | const struct fio_file *f = io_u->file; | |
07fc3f57 | 1422 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; |
bfbdd35b | 1423 | |
122c3539 | 1424 | if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) { |
4d4c71e6 | 1425 | zone_unlock(z); |
59b07544 BVA |
1426 | z = zbd_convert_to_open_zone(td, io_u); |
1427 | assert(z); | |
1428 | } | |
1429 | ||
0cac10b4 | 1430 | if (z->verify_block * min_bs >= z->capacity) { |
07fc3f57 | 1431 | log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n", f->file_name, z->verify_block, |
ee5e3436 | 1432 | min_bs, z->capacity); |
0cac10b4 DF |
1433 | /* |
1434 | * If the assertion below fails during a test run, adding | |
1435 | * "--experimental_verify=1" to the command line may help. | |
1436 | */ | |
1437 | assert(false); | |
1438 | } | |
1439 | io_u->offset = z->start + z->verify_block * min_bs; | |
1440 | if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
ee5e3436 SK |
1441 | log_err("%s: %llu + %llu >= %"PRIu64"\n", f->file_name, io_u->offset, |
1442 | io_u->buflen, zbd_zone_capacity_end(z)); | |
0cac10b4 DF |
1443 | assert(false); |
1444 | } | |
1445 | z->verify_block += io_u->buflen / min_bs; | |
1446 | ||
59b07544 | 1447 | return z; |
bfbdd35b BVA |
1448 | } |
1449 | ||
1450 | /* | |
5c86fdf6 SK |
1451 | * Find another zone which has @min_bytes of readable data. Search in zones |
1452 | * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf. | |
bfbdd35b | 1453 | * |
21c0c884 SK |
1454 | * Either returns NULL or returns a zone pointer. When the zone has write |
1455 | * pointer, hold the mutex for the zone. | |
bfbdd35b BVA |
1456 | */ |
1457 | static struct fio_zone_info * | |
07fc3f57 | 1458 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes, |
bfbdd35b BVA |
1459 | struct fio_zone_info *zb, struct fio_zone_info *zl) |
1460 | { | |
fae3b9a0 | 1461 | struct fio_file *f = io_u->file; |
bfbdd35b | 1462 | struct fio_zone_info *z1, *z2; |
923f7c1e | 1463 | const struct fio_zone_info *const zf = get_zone(f, f->min_zone); |
bfbdd35b BVA |
1464 | |
1465 | /* | |
1466 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1467 | * the nearest non-empty zone in case of random I/O. | |
1468 | */ | |
1469 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1470 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1471 | if (z1->has_wp) |
1472 | zone_lock(td, f, z1); | |
5c86fdf6 | 1473 | if (z1->start + min_bytes <= z1->wp) |
bfbdd35b | 1474 | return z1; |
21c0c884 SK |
1475 | if (z1->has_wp) |
1476 | zone_unlock(z1); | |
bfbdd35b BVA |
1477 | } else if (!td_random(td)) { |
1478 | break; | |
1479 | } | |
1480 | if (td_random(td) && z2 >= zf && | |
b7694961 | 1481 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1482 | if (z2->has_wp) |
1483 | zone_lock(td, f, z2); | |
5c86fdf6 | 1484 | if (z2->start + min_bytes <= z2->wp) |
bfbdd35b | 1485 | return z2; |
21c0c884 SK |
1486 | if (z2->has_wp) |
1487 | zone_unlock(z2); | |
bfbdd35b BVA |
1488 | } |
1489 | } | |
07fc3f57 | 1490 | dprint(FD_ZBD, "%s: no zone has %"PRIu64" bytes of readable data\n", |
5c86fdf6 | 1491 | f->file_name, min_bytes); |
bfbdd35b BVA |
1492 | return NULL; |
1493 | } | |
1494 | ||
b2da58c4 SK |
1495 | /** |
1496 | * zbd_end_zone_io - update zone status at command completion | |
1497 | * @io_u: I/O unit | |
1498 | * @z: zone info pointer | |
1499 | * | |
1500 | * If the write command made the zone full, close it. | |
1501 | * | |
1502 | * The caller must hold z->mutex. | |
1503 | */ | |
1504 | static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, | |
1505 | struct fio_zone_info *z) | |
1506 | { | |
1507 | const struct fio_file *f = io_u->file; | |
1508 | ||
1509 | if (io_u->ddir == DDIR_WRITE && | |
1510 | io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1511 | pthread_mutex_lock(&f->zbd_info->mutex); | |
122c3539 | 1512 | zbd_close_zone(td, f, zbd_zone_nr(f, z)); |
b2da58c4 SK |
1513 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1514 | } | |
1515 | } | |
1516 | ||
bfbdd35b | 1517 | /** |
d9ed3e63 | 1518 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1519 | * @io_u: I/O unit |
d9ed3e63 DLM |
1520 | * @success: Whether or not the I/O unit has been queued successfully |
1521 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1522 | * |
d9ed3e63 DLM |
1523 | * For write and trim operations, update the write pointer of the I/O unit |
1524 | * target zone. | |
bfbdd35b | 1525 | */ |
b2da58c4 SK |
1526 | static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, |
1527 | bool success) | |
bfbdd35b | 1528 | { |
d9ed3e63 DLM |
1529 | const struct fio_file *f = io_u->file; |
1530 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b BVA |
1531 | struct fio_zone_info *z; |
1532 | uint32_t zone_idx; | |
d9ed3e63 | 1533 | uint64_t zone_end; |
bfbdd35b | 1534 | |
5ddf46d0 | 1535 | assert(zbd_info); |
bfbdd35b | 1536 | |
d9ed3e63 | 1537 | zone_idx = zbd_zone_idx(f, io_u->offset); |
bfbdd35b | 1538 | assert(zone_idx < zbd_info->nr_zones); |
923f7c1e | 1539 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1540 | |
43bcbd5b | 1541 | assert(z->has_wp); |
d9ed3e63 | 1542 | |
bfbdd35b BVA |
1543 | if (!success) |
1544 | goto unlock; | |
d9ed3e63 DLM |
1545 | |
1546 | dprint(FD_ZBD, | |
1547 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
1548 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1549 | ||
bfbdd35b BVA |
1550 | switch (io_u->ddir) { |
1551 | case DDIR_WRITE: | |
d9ed3e63 | 1552 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
236d23a8 | 1553 | zbd_zone_capacity_end(z)); |
a7c2b6fc BVA |
1554 | pthread_mutex_lock(&zbd_info->mutex); |
1555 | /* | |
1556 | * z->wp > zone_end means that one or more I/O errors | |
1557 | * have occurred. | |
1558 | */ | |
403ea795 | 1559 | if (z->wp <= zone_end) { |
a7c2b6fc | 1560 | zbd_info->sectors_with_data += zone_end - z->wp; |
403ea795 SK |
1561 | zbd_info->wp_sectors_with_data += zone_end - z->wp; |
1562 | } | |
a7c2b6fc | 1563 | pthread_mutex_unlock(&zbd_info->mutex); |
bfbdd35b BVA |
1564 | z->wp = zone_end; |
1565 | break; | |
bfbdd35b BVA |
1566 | default: |
1567 | break; | |
1568 | } | |
d9ed3e63 | 1569 | |
b2da58c4 SK |
1570 | if (q == FIO_Q_COMPLETED && !io_u->error) |
1571 | zbd_end_zone_io(td, io_u, z); | |
1572 | ||
bfbdd35b | 1573 | unlock: |
d9ed3e63 DLM |
1574 | if (!success || q != FIO_Q_QUEUED) { |
1575 | /* BUSY or COMPLETED: unlock the zone */ | |
4d4c71e6 | 1576 | zone_unlock(z); |
d9ed3e63 DLM |
1577 | io_u->zbd_put_io = NULL; |
1578 | } | |
1579 | } | |
1580 | ||
1581 | /** | |
1582 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1583 | * @io_u: I/O unit | |
1584 | */ | |
b2da58c4 | 1585 | static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) |
d9ed3e63 DLM |
1586 | { |
1587 | const struct fio_file *f = io_u->file; | |
1588 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1589 | struct fio_zone_info *z; | |
1590 | uint32_t zone_idx; | |
1591 | ||
5ddf46d0 | 1592 | assert(zbd_info); |
615555bb | 1593 | |
d9ed3e63 DLM |
1594 | zone_idx = zbd_zone_idx(f, io_u->offset); |
1595 | assert(zone_idx < zbd_info->nr_zones); | |
923f7c1e | 1596 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1597 | |
43bcbd5b | 1598 | assert(z->has_wp); |
d9ed3e63 DLM |
1599 | |
1600 | dprint(FD_ZBD, | |
1601 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
1602 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1603 | ||
b2da58c4 SK |
1604 | zbd_end_zone_io(td, io_u, z); |
1605 | ||
4d4c71e6 | 1606 | zone_unlock(z); |
11afb212 | 1607 | zbd_check_swd(td, f); |
bfbdd35b BVA |
1608 | } |
1609 | ||
9d87c646 DLM |
1610 | /* |
1611 | * Windows and MacOS do not define this. | |
1612 | */ | |
1613 | #ifndef EREMOTEIO | |
1614 | #define EREMOTEIO 121 /* POSIX value */ | |
1615 | #endif | |
1616 | ||
bfbdd35b BVA |
1617 | bool zbd_unaligned_write(int error_code) |
1618 | { | |
1619 | switch (error_code) { | |
1620 | case EIO: | |
1621 | case EREMOTEIO: | |
1622 | return true; | |
1623 | } | |
1624 | return false; | |
1625 | } | |
1626 | ||
4d37720a DLM |
1627 | /** |
1628 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1629 | * @td: FIO thread data. | |
1630 | * @io_u: FIO I/O unit. | |
1631 | * | |
1632 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1633 | * no more IO can be performed in the current zone. | |
1634 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1635 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1636 | * - For write workloads, zoneskip is applied when the zone is full. | |
1637 | * This applies only to read and write operations. | |
1638 | */ | |
1639 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1640 | { | |
1641 | struct fio_file *f = io_u->file; | |
1642 | enum fio_ddir ddir = io_u->ddir; | |
1643 | struct fio_zone_info *z; | |
1644 | uint32_t zone_idx; | |
1645 | ||
1646 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1647 | assert(td->o.zone_size); | |
5ddf46d0 | 1648 | assert(f->zbd_info); |
4d37720a | 1649 | |
236d23a8 | 1650 | zone_idx = zbd_zone_idx(f, f->last_pos[ddir]); |
923f7c1e | 1651 | z = get_zone(f, zone_idx); |
236d23a8 SK |
1652 | |
1653 | /* | |
1654 | * When the zone capacity is smaller than the zone size and the I/O is | |
1655 | * sequential write, skip to zone end if the latest position is at the | |
1656 | * zone capacity limit. | |
1657 | */ | |
1658 | if (z->capacity < f->zbd_info->zone_size && !td_random(td) && | |
1659 | ddir == DDIR_WRITE && | |
1660 | f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { | |
1661 | dprint(FD_ZBD, | |
1662 | "%s: Jump from zone capacity limit to zone end:" | |
ee5e3436 SK |
1663 | " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n", |
1664 | f->file_name, f->last_pos[ddir], | |
1665 | zbd_zone_end(z), zone_idx, z->capacity); | |
236d23a8 SK |
1666 | td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; |
1667 | f->last_pos[ddir] = zbd_zone_end(z); | |
1668 | } | |
1669 | ||
4d37720a DLM |
1670 | /* |
1671 | * zone_skip is valid only for sequential workloads. | |
1672 | */ | |
1673 | if (td_random(td) || !td->o.zone_skip) | |
1674 | return; | |
1675 | ||
1676 | /* | |
1677 | * It is time to switch to a new zone if: | |
1678 | * - zone_bytes == zone_size bytes have already been accessed | |
1679 | * - The last position reached the end of the current zone. | |
1680 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1681 | * reached the zone write pointer. | |
1682 | */ | |
4d37720a | 1683 | if (td->zone_bytes >= td->o.zone_size || |
236d23a8 | 1684 | f->last_pos[ddir] >= zbd_zone_end(z) || |
4d37720a DLM |
1685 | (ddir == DDIR_READ && |
1686 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1687 | /* | |
1688 | * Skip zones. | |
1689 | */ | |
1690 | td->zone_bytes = 0; | |
1691 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1692 | ||
1693 | /* | |
1694 | * Wrap from the beginning, if we exceed the file size | |
1695 | */ | |
1696 | if (f->file_offset >= f->real_file_size) | |
1697 | f->file_offset = get_start_offset(td, f); | |
1698 | ||
1699 | f->last_pos[ddir] = f->file_offset; | |
1700 | td->io_skip_bytes += td->o.zone_skip; | |
1701 | } | |
1702 | } | |
1703 | ||
c65057f9 | 1704 | /** |
c7d5e152 | 1705 | * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. |
c65057f9 SK |
1706 | * |
1707 | * @td: FIO thread data. | |
1708 | * @io_u: FIO I/O unit. | |
1709 | * @ddir: I/O direction before adjustment. | |
1710 | * | |
1711 | * Return adjusted I/O direction. | |
1712 | */ | |
1713 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1714 | enum fio_ddir ddir) | |
1715 | { | |
1716 | /* | |
1717 | * In case read direction is chosen for the first random I/O, fio with | |
1718 | * zonemode=zbd stops because no data can be read from zoned block | |
1719 | * devices with all empty zones. Overwrite the first I/O direction as | |
1720 | * write to make sure data to read exists. | |
1721 | */ | |
5ddf46d0 | 1722 | assert(io_u->file->zbd_info); |
731461cc | 1723 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1724 | return ddir; |
1725 | ||
1726 | if (io_u->file->zbd_info->sectors_with_data || | |
1727 | td->o.read_beyond_wp) | |
1728 | return DDIR_READ; | |
1729 | ||
1730 | return DDIR_WRITE; | |
1731 | } | |
1732 | ||
bfbdd35b BVA |
1733 | /** |
1734 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1735 | * @td: FIO thread data. | |
1736 | * @io_u: FIO I/O unit. | |
1737 | * | |
1738 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1739 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1740 | * corresponds to io_u->offset at the end of this function. | |
1741 | */ | |
1742 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1743 | { | |
b7694961 | 1744 | struct fio_file *f = io_u->file; |
af94a8c3 | 1745 | struct zoned_block_device_info *zbdi = f->zbd_info; |
bfbdd35b | 1746 | uint32_t zone_idx_b; |
de65f7b7 | 1747 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b | 1748 | uint32_t orig_len = io_u->buflen; |
07fc3f57 | 1749 | uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
bfbdd35b BVA |
1750 | uint64_t new_len; |
1751 | int64_t range; | |
1752 | ||
af94a8c3 | 1753 | assert(zbdi); |
adc6adcb | 1754 | assert(min_bs); |
bfbdd35b BVA |
1755 | assert(is_valid_offset(f, io_u->offset)); |
1756 | assert(io_u->buflen); | |
1757 | zone_idx_b = zbd_zone_idx(f, io_u->offset); | |
923f7c1e | 1758 | zb = get_zone(f, zone_idx_b); |
de65f7b7 | 1759 | orig_zb = zb; |
bfbdd35b | 1760 | |
2efcf74b SK |
1761 | if (!zb->has_wp) { |
1762 | /* Accept non-write I/Os for conventional zones. */ | |
1763 | if (io_u->ddir != DDIR_WRITE) | |
1764 | return io_u_accept; | |
1765 | /* | |
1766 | * Make sure that writes to conventional zones | |
1767 | * don't cross over to any sequential zones. | |
1768 | */ | |
1769 | if (!(zb + 1)->has_wp || | |
1770 | io_u->offset + io_u->buflen <= (zb + 1)->start) | |
1771 | return io_u_accept; | |
1772 | ||
1773 | if (io_u->offset + min_bs > (zb + 1)->start) { | |
1774 | dprint(FD_IO, | |
07fc3f57 | 1775 | "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n", |
1e3d6e03 | 1776 | f->file_name, io_u->offset, |
ee5e3436 | 1777 | min_bs, (zb + 1)->start); |
2efcf74b SK |
1778 | io_u->offset = zb->start + (zb + 1)->start - io_u->offset; |
1779 | new_len = min(io_u->buflen, (zb + 1)->start - io_u->offset); | |
1780 | } else { | |
1781 | new_len = (zb + 1)->start - io_u->offset; | |
1782 | } | |
1783 | io_u->buflen = new_len / min_bs * min_bs; | |
bfbdd35b | 1784 | return io_u_accept; |
2efcf74b | 1785 | } |
bfbdd35b BVA |
1786 | |
1787 | /* | |
1788 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1789 | * is enabled. | |
1790 | */ | |
b7694961 | 1791 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1792 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1793 | return io_u_accept; | |
1794 | ||
11afb212 | 1795 | zbd_check_swd(td, f); |
615555bb | 1796 | |
fae3b9a0 | 1797 | zone_lock(td, f, zb); |
6f0c6085 | 1798 | |
bfbdd35b BVA |
1799 | switch (io_u->ddir) { |
1800 | case DDIR_READ: | |
0ed7d55e AR |
1801 | if (td->runstate == TD_VERIFYING && td_write(td)) { |
1802 | zb = zbd_replay_write_order(td, io_u, zb); | |
bfbdd35b BVA |
1803 | goto accept; |
1804 | } | |
1805 | /* | |
de65f7b7 DLM |
1806 | * Check that there is enough written data in the zone to do an |
1807 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1808 | * the I/O. | |
bfbdd35b | 1809 | */ |
b7694961 | 1810 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1811 | zb->wp - zb->start : 0; |
de65f7b7 | 1812 | if (range < min_bs || |
ee3696bd | 1813 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
4d4c71e6 | 1814 | zone_unlock(zb); |
923f7c1e | 1815 | zl = get_zone(f, f->max_zone); |
5c86fdf6 | 1816 | zb = zbd_find_zone(td, io_u, min_bs, zb, zl); |
bfbdd35b BVA |
1817 | if (!zb) { |
1818 | dprint(FD_ZBD, | |
1819 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1820 | f->file_name, io_u->offset, | |
1821 | io_u->buflen); | |
1822 | goto eof; | |
1823 | } | |
de65f7b7 DLM |
1824 | /* |
1825 | * zbd_find_zone() returned a zone with a range of at | |
1826 | * least min_bs. | |
1827 | */ | |
ee3696bd | 1828 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1829 | assert(range >= min_bs); |
1830 | ||
1831 | if (!td_random(td)) | |
ee3696bd | 1832 | io_u->offset = zb->start; |
bfbdd35b | 1833 | } |
de65f7b7 DLM |
1834 | /* |
1835 | * Make sure the I/O is within the zone valid data range while | |
1836 | * maximizing the I/O size and preserving randomness. | |
1837 | */ | |
1838 | if (range <= io_u->buflen) | |
ee3696bd | 1839 | io_u->offset = zb->start; |
de65f7b7 | 1840 | else if (td_random(td)) |
ee3696bd DLM |
1841 | io_u->offset = zb->start + |
1842 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 | 1843 | (range - io_u->buflen)) / min_bs * min_bs; |
43bcbd5b SK |
1844 | /* |
1845 | * When zbd_find_zone() returns a conventional zone, | |
1846 | * we can simply accept the new i/o offset here. | |
1847 | */ | |
1848 | if (!zb->has_wp) | |
1849 | return io_u_accept; | |
de65f7b7 DLM |
1850 | /* |
1851 | * Make sure the I/O does not cross over the zone wp position. | |
1852 | */ | |
1853 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1854 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1855 | new_len = new_len / min_bs * min_bs; |
1856 | if (new_len < io_u->buflen) { | |
1857 | io_u->buflen = new_len; | |
1858 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1859 | orig_len, io_u->buflen); | |
bfbdd35b | 1860 | } |
ee3696bd DLM |
1861 | assert(zb->start <= io_u->offset); |
1862 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
bfbdd35b BVA |
1863 | goto accept; |
1864 | case DDIR_WRITE: | |
af94a8c3 | 1865 | if (io_u->buflen > zbdi->zone_size) { |
1c74aadc DF |
1866 | td_verror(td, EINVAL, "I/O buflen exceeds zone size"); |
1867 | dprint(FD_IO, | |
ee5e3436 SK |
1868 | "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n", |
1869 | f->file_name, io_u->buflen, zbdi->zone_size); | |
bfbdd35b | 1870 | goto eof; |
1c74aadc | 1871 | } |
954217b9 | 1872 | if (!zbd_open_zone(td, f, zone_idx_b)) { |
4d4c71e6 | 1873 | zone_unlock(zb); |
59b07544 | 1874 | zb = zbd_convert_to_open_zone(td, io_u); |
1c74aadc DF |
1875 | if (!zb) { |
1876 | dprint(FD_IO, "%s: can't convert to open zone", | |
1877 | f->file_name); | |
59b07544 | 1878 | goto eof; |
1c74aadc | 1879 | } |
59b07544 | 1880 | } |
a7c2b6fc BVA |
1881 | /* Check whether the zone reset threshold has been exceeded */ |
1882 | if (td->o.zrf.u.f) { | |
af94a8c3 | 1883 | if (zbdi->wp_sectors_with_data >= |
a7c2b6fc BVA |
1884 | f->io_size * td->o.zrt.u.f && |
1885 | zbd_dec_and_reset_write_cnt(td, f)) { | |
1886 | zb->reset_zone = 1; | |
1887 | } | |
1888 | } | |
bfbdd35b BVA |
1889 | /* Reset the zone pointer if necessary */ |
1890 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
1891 | assert(td->o.verify == VERIFY_NONE); | |
1892 | /* | |
1893 | * Since previous write requests may have been submitted | |
1894 | * asynchronously and since we will submit the zone | |
1895 | * reset synchronously, wait until previously submitted | |
1896 | * write requests have completed before issuing a | |
1897 | * zone reset. | |
1898 | */ | |
1899 | io_u_quiesce(td); | |
1900 | zb->reset_zone = 0; | |
1901 | if (zbd_reset_zone(td, f, zb) < 0) | |
1902 | goto eof; | |
236d23a8 SK |
1903 | |
1904 | if (zb->capacity < min_bs) { | |
1c74aadc | 1905 | td_verror(td, EINVAL, "ZCAP is less min_bs"); |
07fc3f57 | 1906 | log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n", |
ee5e3436 | 1907 | zb->capacity, min_bs); |
236d23a8 SK |
1908 | goto eof; |
1909 | } | |
bfbdd35b BVA |
1910 | } |
1911 | /* Make writes occur at the write pointer */ | |
1912 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 1913 | io_u->offset = zb->wp; |
bfbdd35b | 1914 | if (!is_valid_offset(f, io_u->offset)) { |
1c74aadc DF |
1915 | td_verror(td, EINVAL, "invalid WP value"); |
1916 | dprint(FD_ZBD, "%s: dropped request with offset %llu\n", | |
1917 | f->file_name, io_u->offset); | |
bfbdd35b BVA |
1918 | goto eof; |
1919 | } | |
1920 | /* | |
1921 | * Make sure that the buflen is a multiple of the minimal | |
1922 | * block size. Give up if shrinking would make the request too | |
1923 | * small. | |
1924 | */ | |
1925 | new_len = min((unsigned long long)io_u->buflen, | |
236d23a8 | 1926 | zbd_zone_capacity_end(zb) - io_u->offset); |
bfbdd35b BVA |
1927 | new_len = new_len / min_bs * min_bs; |
1928 | if (new_len == io_u->buflen) | |
1929 | goto accept; | |
1930 | if (new_len >= min_bs) { | |
1931 | io_u->buflen = new_len; | |
1932 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1933 | orig_len, io_u->buflen); | |
1934 | goto accept; | |
1935 | } | |
1c74aadc | 1936 | td_verror(td, EIO, "zone remainder too small"); |
07fc3f57 | 1937 | log_err("zone remainder %lld smaller than min block size %"PRIu64"\n", |
1c74aadc | 1938 | (zbd_zone_capacity_end(zb) - io_u->offset), min_bs); |
bfbdd35b BVA |
1939 | goto eof; |
1940 | case DDIR_TRIM: | |
e3be810b SK |
1941 | /* Check random trim targets a non-empty zone */ |
1942 | if (!td_random(td) || zb->wp > zb->start) | |
1943 | goto accept; | |
1944 | ||
1945 | /* Find out a non-empty zone to trim */ | |
1946 | zone_unlock(zb); | |
1947 | zl = get_zone(f, f->max_zone); | |
1948 | zb = zbd_find_zone(td, io_u, 1, zb, zl); | |
1949 | if (zb) { | |
1950 | io_u->offset = zb->start; | |
1951 | dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n", | |
1952 | f->file_name, io_u->offset); | |
1953 | goto accept; | |
1954 | } | |
1955 | goto eof; | |
bfbdd35b | 1956 | case DDIR_SYNC: |
e3be810b | 1957 | /* fall-through */ |
bfbdd35b BVA |
1958 | case DDIR_DATASYNC: |
1959 | case DDIR_SYNC_FILE_RANGE: | |
1960 | case DDIR_WAIT: | |
1961 | case DDIR_LAST: | |
1962 | case DDIR_INVAL: | |
1963 | goto accept; | |
1964 | } | |
1965 | ||
1966 | assert(false); | |
1967 | ||
1968 | accept: | |
43bcbd5b | 1969 | assert(zb->has_wp); |
b7694961 | 1970 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
1971 | assert(!io_u->zbd_queue_io); |
1972 | assert(!io_u->zbd_put_io); | |
1973 | io_u->zbd_queue_io = zbd_queue_io; | |
1974 | io_u->zbd_put_io = zbd_put_io; | |
2ef3c1b0 DF |
1975 | /* |
1976 | * Since we return with the zone lock still held, | |
1977 | * add an annotation to let Coverity know that it | |
1978 | * is intentional. | |
1979 | */ | |
1980 | /* coverity[missing_unlock] */ | |
bfbdd35b BVA |
1981 | return io_u_accept; |
1982 | ||
1983 | eof: | |
43bcbd5b | 1984 | if (zb && zb->has_wp) |
4d4c71e6 | 1985 | zone_unlock(zb); |
bfbdd35b BVA |
1986 | return io_u_eof; |
1987 | } | |
fd5d733f BVA |
1988 | |
1989 | /* Return a string with ZBD statistics */ | |
1990 | char *zbd_write_status(const struct thread_stat *ts) | |
1991 | { | |
1992 | char *res; | |
1993 | ||
ee5e3436 | 1994 | if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0) |
fd5d733f BVA |
1995 | return NULL; |
1996 | return res; | |
1997 | } | |
e3be810b SK |
1998 | |
1999 | /** | |
2000 | * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim | |
2001 | * | |
2002 | * @td: FIO thread data. | |
2003 | * @io_u: FIO I/O unit. | |
2004 | * | |
2005 | * It is assumed that z->mutex is already locked. | |
2006 | * Return io_u_completed when reset zone succeeds. Return 0 when the target zone | |
2007 | * does not have write pointer. On error, return negative errno. | |
2008 | */ | |
2009 | int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u) | |
2010 | { | |
2011 | struct fio_file *f = io_u->file; | |
2012 | struct fio_zone_info *z; | |
2013 | uint32_t zone_idx; | |
2014 | int ret; | |
2015 | ||
2016 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
2017 | z = get_zone(f, zone_idx); | |
2018 | ||
2019 | if (!z->has_wp) | |
2020 | return 0; | |
2021 | ||
2022 | if (io_u->offset != z->start) { | |
2023 | log_err("Trim offset not at zone start (%lld)\n", io_u->offset); | |
2024 | return -EINVAL; | |
2025 | } | |
2026 | ||
2027 | ret = zbd_reset_zone((struct thread_data *)td, f, z); | |
2028 | if (ret < 0) | |
2029 | return ret; | |
2030 | ||
2031 | return io_u_completed; | |
2032 | } |