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 | ||
dc8a3d62 DLM |
30 | static inline unsigned int zbd_zone_idx(const struct fio_file *f, |
31 | struct fio_zone_info *zone) | |
410a071c DLM |
32 | { |
33 | return zone - f->zbd_info->zone_info; | |
34 | } | |
35 | ||
36 | /** | |
dc8a3d62 | 37 | * zbd_offset_to_zone_idx - convert an offset into a zone number |
410a071c DLM |
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 | */ | |
dc8a3d62 DLM |
42 | static unsigned int zbd_offset_to_zone_idx(const struct fio_file *f, |
43 | uint64_t offset) | |
410a071c DLM |
44 | { |
45 | uint32_t zone_idx; | |
46 | ||
47 | if (f->zbd_info->zone_size_log2 > 0) | |
48 | zone_idx = offset >> f->zbd_info->zone_size_log2; | |
49 | else | |
50 | zone_idx = offset / f->zbd_info->zone_size; | |
51 | ||
52 | return min(zone_idx, f->zbd_info->nr_zones); | |
53 | } | |
54 | ||
55 | /** | |
56 | * zbd_zone_end - Return zone end location | |
57 | * @z: zone info pointer. | |
58 | */ | |
59 | static inline uint64_t zbd_zone_end(const struct fio_zone_info *z) | |
60 | { | |
61 | return (z+1)->start; | |
62 | } | |
63 | ||
64 | /** | |
65 | * zbd_zone_capacity_end - Return zone capacity limit end location | |
66 | * @z: zone info pointer. | |
67 | */ | |
68 | static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z) | |
69 | { | |
70 | return z->start + z->capacity; | |
71 | } | |
72 | ||
df67bf1e SK |
73 | /** |
74 | * zbd_zone_remainder - Return the number of bytes that are still available for | |
75 | * writing before the zone gets full | |
76 | * @z: zone info pointer. | |
77 | */ | |
78 | static inline uint64_t zbd_zone_remainder(struct fio_zone_info *z) | |
79 | { | |
80 | if (z->wp >= zbd_zone_capacity_end(z)) | |
81 | return 0; | |
82 | ||
83 | return zbd_zone_capacity_end(z) - z->wp; | |
84 | } | |
85 | ||
410a071c DLM |
86 | /** |
87 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
88 | * @f: file pointer. | |
89 | * @z: zone info pointer. | |
90 | * @required: minimum number of bytes that must remain in a zone. | |
91 | * | |
92 | * The caller must hold z->mutex. | |
93 | */ | |
94 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
95 | uint64_t required) | |
96 | { | |
97 | assert((required & 511) == 0); | |
98 | ||
df67bf1e | 99 | return z->has_wp && required > zbd_zone_remainder(z); |
410a071c DLM |
100 | } |
101 | ||
102 | static void zone_lock(struct thread_data *td, const struct fio_file *f, | |
103 | struct fio_zone_info *z) | |
104 | { | |
105 | struct zoned_block_device_info *zbd = f->zbd_info; | |
106 | uint32_t nz = z - zbd->zone_info; | |
107 | ||
108 | /* A thread should never lock zones outside its working area. */ | |
109 | assert(f->min_zone <= nz && nz < f->max_zone); | |
110 | ||
111 | assert(z->has_wp); | |
112 | ||
113 | /* | |
114 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
115 | * is changed or when io_u completes and zbd_put_io() executed. | |
116 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
117 | * other waiting for zone locks when building an io_u batch, first | |
118 | * only trylock the zone. If the zone is already locked by another job, | |
119 | * process the currently queued I/Os so that I/O progress is made and | |
120 | * zones unlocked. | |
121 | */ | |
122 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
123 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
124 | io_u_quiesce(td); | |
125 | pthread_mutex_lock(&z->mutex); | |
126 | } | |
127 | } | |
128 | ||
129 | static inline void zone_unlock(struct fio_zone_info *z) | |
130 | { | |
131 | int ret; | |
132 | ||
133 | assert(z->has_wp); | |
134 | ret = pthread_mutex_unlock(&z->mutex); | |
135 | assert(!ret); | |
136 | } | |
137 | ||
39e06ee7 DLM |
138 | static inline struct fio_zone_info *zbd_get_zone(const struct fio_file *f, |
139 | unsigned int zone_idx) | |
410a071c | 140 | { |
39e06ee7 | 141 | return &f->zbd_info->zone_info[zone_idx]; |
410a071c DLM |
142 | } |
143 | ||
53aa6171 DLM |
144 | static inline struct fio_zone_info * |
145 | zbd_offset_to_zone(const struct fio_file *f, uint64_t offset) | |
146 | { | |
147 | return zbd_get_zone(f, zbd_offset_to_zone_idx(f, offset)); | |
148 | } | |
149 | ||
2fb29f27 SK |
150 | static bool accounting_vdb(struct thread_data *td, const struct fio_file *f) |
151 | { | |
152 | return td->o.zrt.u.f && td_write(td); | |
153 | } | |
154 | ||
b7694961 DLM |
155 | /** |
156 | * zbd_get_zoned_model - Get a device zoned model | |
157 | * @td: FIO thread data | |
158 | * @f: FIO file for which to get model information | |
159 | */ | |
38334c13 DLM |
160 | static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, |
161 | enum zbd_zoned_model *model) | |
b7694961 DLM |
162 | { |
163 | int ret; | |
164 | ||
50cc48d5 NC |
165 | if (f->filetype == FIO_TYPE_PIPE) { |
166 | log_err("zonemode=zbd does not support pipes\n"); | |
167 | return -EINVAL; | |
168 | } | |
169 | ||
9db0cde8 NC |
170 | /* If regular file, always emulate zones inside the file. */ |
171 | if (f->filetype == FIO_TYPE_FILE) { | |
172 | *model = ZBD_NONE; | |
173 | return 0; | |
174 | } | |
175 | ||
6c5b11d3 DLM |
176 | if (td->io_ops && td->io_ops->get_zoned_model) |
177 | ret = td->io_ops->get_zoned_model(td, f, model); | |
178 | else | |
179 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
180 | if (ret < 0) { |
181 | td_verror(td, errno, "get zoned model failed"); | |
182 | log_err("%s: get zoned model failed (%d).\n", | |
183 | f->file_name, errno); | |
184 | } | |
185 | ||
186 | return ret; | |
187 | } | |
188 | ||
189 | /** | |
190 | * zbd_report_zones - Get zone information | |
191 | * @td: FIO thread data. | |
192 | * @f: FIO file for which to get zone information | |
193 | * @offset: offset from which to report zones | |
194 | * @zones: Array of struct zbd_zone | |
195 | * @nr_zones: Size of @zones array | |
196 | * | |
197 | * Get zone information into @zones starting from the zone at offset @offset | |
198 | * for the device specified by @f. | |
199 | * | |
200 | * Returns the number of zones reported upon success and a negative error code | |
201 | * upon failure. If the zone report is empty, always assume an error (device | |
202 | * problem) and return -EIO. | |
203 | */ | |
38334c13 DLM |
204 | static int zbd_report_zones(struct thread_data *td, struct fio_file *f, |
205 | uint64_t offset, struct zbd_zone *zones, | |
206 | unsigned int nr_zones) | |
b7694961 DLM |
207 | { |
208 | int ret; | |
209 | ||
6c5b11d3 DLM |
210 | if (td->io_ops && td->io_ops->report_zones) |
211 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
212 | else | |
213 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
214 | if (ret < 0) { |
215 | td_verror(td, errno, "report zones failed"); | |
362ce037 BVA |
216 | log_err("%s: report zones from sector %"PRIu64" failed (nr_zones=%d; errno=%d).\n", |
217 | f->file_name, offset >> 9, nr_zones, errno); | |
b7694961 DLM |
218 | } else if (ret == 0) { |
219 | td_verror(td, errno, "Empty zone report"); | |
ee5e3436 SK |
220 | log_err("%s: report zones from sector %"PRIu64" is empty.\n", |
221 | f->file_name, offset >> 9); | |
b7694961 DLM |
222 | ret = -EIO; |
223 | } | |
224 | ||
225 | return ret; | |
226 | } | |
227 | ||
228 | /** | |
229 | * zbd_reset_wp - reset the write pointer of a range of zones | |
230 | * @td: FIO thread data. | |
231 | * @f: FIO file for which to reset zones | |
232 | * @offset: Starting offset of the first zone to reset | |
233 | * @length: Length of the range of zones to reset | |
234 | * | |
235 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
236 | * Returns 0 upon success and a negative error code upon failure. | |
237 | */ | |
38334c13 DLM |
238 | static int zbd_reset_wp(struct thread_data *td, struct fio_file *f, |
239 | uint64_t offset, uint64_t length) | |
b7694961 DLM |
240 | { |
241 | int ret; | |
242 | ||
6c5b11d3 DLM |
243 | if (td->io_ops && td->io_ops->reset_wp) |
244 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
245 | else | |
246 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
247 | if (ret < 0) { |
248 | td_verror(td, errno, "resetting wp failed"); | |
ee5e3436 SK |
249 | log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n", |
250 | f->file_name, length >> 9, offset >> 9, errno); | |
b7694961 DLM |
251 | } |
252 | ||
253 | return ret; | |
254 | } | |
255 | ||
410a071c | 256 | /** |
67282020 | 257 | * __zbd_reset_zone - reset the write pointer of a single zone |
410a071c DLM |
258 | * @td: FIO thread data. |
259 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
260 | * @z: Zone to reset. | |
261 | * | |
262 | * Returns 0 upon success and a negative error code upon failure. | |
263 | * | |
264 | * The caller must hold z->mutex. | |
265 | */ | |
67282020 SK |
266 | static int __zbd_reset_zone(struct thread_data *td, struct fio_file *f, |
267 | struct fio_zone_info *z) | |
410a071c DLM |
268 | { |
269 | uint64_t offset = z->start; | |
270 | uint64_t length = (z+1)->start - offset; | |
271 | uint64_t data_in_zone = z->wp - z->start; | |
272 | int ret = 0; | |
273 | ||
274 | if (!data_in_zone) | |
275 | return 0; | |
276 | ||
277 | assert(is_valid_offset(f, offset + length - 1)); | |
278 | ||
139d8dc6 | 279 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", |
dc8a3d62 | 280 | f->file_name, zbd_zone_idx(f, z)); |
139d8dc6 | 281 | |
410a071c DLM |
282 | switch (f->zbd_info->model) { |
283 | case ZBD_HOST_AWARE: | |
284 | case ZBD_HOST_MANAGED: | |
285 | ret = zbd_reset_wp(td, f, offset, length); | |
286 | if (ret < 0) | |
287 | return ret; | |
288 | break; | |
289 | default: | |
290 | break; | |
291 | } | |
292 | ||
2fb29f27 SK |
293 | if (accounting_vdb(td, f)) { |
294 | pthread_mutex_lock(&f->zbd_info->mutex); | |
295 | f->zbd_info->wp_valid_data_bytes -= data_in_zone; | |
296 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
297 | } | |
139d8dc6 | 298 | |
410a071c | 299 | z->wp = z->start; |
410a071c DLM |
300 | |
301 | td->ts.nr_zone_resets++; | |
302 | ||
303 | return ret; | |
304 | } | |
305 | ||
306 | /** | |
a4807046 | 307 | * zbd_write_zone_put - Remove a zone from the write target zones array. |
410a071c | 308 | * @td: FIO thread data. |
a4807046 | 309 | * @f: FIO file that has the write zones array to remove. |
410a071c DLM |
310 | * @zone_idx: Index of the zone to remove. |
311 | * | |
312 | * The caller must hold f->zbd_info->mutex. | |
313 | */ | |
a4807046 SK |
314 | static void zbd_write_zone_put(struct thread_data *td, const struct fio_file *f, |
315 | struct fio_zone_info *z) | |
410a071c | 316 | { |
a4807046 | 317 | uint32_t zi; |
410a071c | 318 | |
a4807046 | 319 | if (!z->write) |
a23411bb DLM |
320 | return; |
321 | ||
a4807046 SK |
322 | for (zi = 0; zi < f->zbd_info->num_write_zones; zi++) { |
323 | if (zbd_get_zone(f, f->zbd_info->write_zones[zi]) == z) | |
410a071c DLM |
324 | break; |
325 | } | |
a4807046 | 326 | if (zi == f->zbd_info->num_write_zones) |
410a071c DLM |
327 | return; |
328 | ||
a4807046 | 329 | dprint(FD_ZBD, "%s: removing zone %u from write zone array\n", |
dc8a3d62 | 330 | f->file_name, zbd_zone_idx(f, z)); |
139d8dc6 | 331 | |
a4807046 SK |
332 | memmove(f->zbd_info->write_zones + zi, |
333 | f->zbd_info->write_zones + zi + 1, | |
334 | (ZBD_MAX_WRITE_ZONES - (zi + 1)) * | |
335 | sizeof(f->zbd_info->write_zones[0])); | |
139d8dc6 | 336 | |
a4807046 SK |
337 | f->zbd_info->num_write_zones--; |
338 | td->num_write_zones--; | |
339 | z->write = 0; | |
410a071c DLM |
340 | } |
341 | ||
67282020 SK |
342 | /** |
343 | * zbd_reset_zone - reset the write pointer of a single zone and remove the zone | |
344 | * from the array of write zones. | |
345 | * @td: FIO thread data. | |
346 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
347 | * @z: Zone to reset. | |
348 | * | |
349 | * Returns 0 upon success and a negative error code upon failure. | |
350 | * | |
351 | * The caller must hold z->mutex. | |
352 | */ | |
353 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, | |
354 | struct fio_zone_info *z) | |
355 | { | |
356 | int ret; | |
357 | ||
358 | ret = __zbd_reset_zone(td, f, z); | |
359 | if (ret) | |
360 | return ret; | |
361 | ||
362 | pthread_mutex_lock(&f->zbd_info->mutex); | |
363 | zbd_write_zone_put(td, f, z); | |
364 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
365 | return 0; | |
366 | } | |
367 | ||
e1a1b59b SK |
368 | /** |
369 | * zbd_finish_zone - finish the specified zone | |
370 | * @td: FIO thread data. | |
371 | * @f: FIO file for which to finish a zone | |
372 | * @z: Zone to finish. | |
373 | * | |
374 | * Finish the zone at @offset with open or close status. | |
375 | */ | |
376 | static int zbd_finish_zone(struct thread_data *td, struct fio_file *f, | |
377 | struct fio_zone_info *z) | |
378 | { | |
379 | uint64_t offset = z->start; | |
380 | uint64_t length = f->zbd_info->zone_size; | |
381 | int ret = 0; | |
382 | ||
383 | switch (f->zbd_info->model) { | |
384 | case ZBD_HOST_AWARE: | |
385 | case ZBD_HOST_MANAGED: | |
386 | if (td->io_ops && td->io_ops->finish_zone) | |
387 | ret = td->io_ops->finish_zone(td, f, offset, length); | |
388 | else | |
389 | ret = blkzoned_finish_zone(td, f, offset, length); | |
390 | break; | |
391 | default: | |
392 | break; | |
393 | } | |
394 | ||
395 | if (ret < 0) { | |
396 | td_verror(td, errno, "finish zone failed"); | |
397 | log_err("%s: finish zone at sector %"PRIu64" failed (%d).\n", | |
398 | f->file_name, offset >> 9, errno); | |
399 | } else { | |
400 | z->wp = (z+1)->start; | |
401 | } | |
402 | ||
403 | return ret; | |
404 | } | |
405 | ||
410a071c DLM |
406 | /** |
407 | * zbd_reset_zones - Reset a range of zones. | |
408 | * @td: fio thread data. | |
409 | * @f: fio file for which to reset zones | |
410 | * @zb: first zone to reset. | |
411 | * @ze: first zone not to reset. | |
412 | * | |
413 | * Returns 0 upon success and 1 upon failure. | |
414 | */ | |
415 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
416 | struct fio_zone_info *const zb, | |
417 | struct fio_zone_info *const ze) | |
418 | { | |
419 | struct fio_zone_info *z; | |
420 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
421 | int res = 0; | |
422 | ||
423 | assert(min_bs); | |
424 | ||
139d8dc6 | 425 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", |
dc8a3d62 | 426 | f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze)); |
139d8dc6 | 427 | |
410a071c | 428 | for (z = zb; z < ze; z++) { |
410a071c DLM |
429 | if (!z->has_wp) |
430 | continue; | |
139d8dc6 | 431 | |
410a071c | 432 | zone_lock(td, f, z); |
139d8dc6 | 433 | |
410a071c DLM |
434 | if (z->wp != z->start) { |
435 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
dc8a3d62 | 436 | f->file_name, zbd_zone_idx(f, z)); |
410a071c DLM |
437 | if (zbd_reset_zone(td, f, z) < 0) |
438 | res = 1; | |
439 | } | |
139d8dc6 | 440 | |
410a071c DLM |
441 | zone_unlock(z); |
442 | } | |
443 | ||
444 | return res; | |
445 | } | |
446 | ||
d2f442bc NC |
447 | /** |
448 | * zbd_get_max_open_zones - Get the maximum number of open zones | |
449 | * @td: FIO thread data | |
450 | * @f: FIO file for which to get max open zones | |
451 | * @max_open_zones: Upon success, result will be stored here. | |
452 | * | |
453 | * A @max_open_zones value set to zero means no limit. | |
454 | * | |
455 | * Returns 0 upon success and a negative error code upon failure. | |
456 | */ | |
38334c13 DLM |
457 | static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f, |
458 | unsigned int *max_open_zones) | |
d2f442bc NC |
459 | { |
460 | int ret; | |
461 | ||
462 | if (td->io_ops && td->io_ops->get_max_open_zones) | |
463 | ret = td->io_ops->get_max_open_zones(td, f, max_open_zones); | |
464 | else | |
465 | ret = blkzoned_get_max_open_zones(td, f, max_open_zones); | |
466 | if (ret < 0) { | |
467 | td_verror(td, errno, "get max open zones failed"); | |
468 | log_err("%s: get max open zones failed (%d).\n", | |
469 | f->file_name, errno); | |
470 | } | |
471 | ||
472 | return ret; | |
473 | } | |
474 | ||
bfbdd35b | 475 | /** |
f539b98c | 476 | * __zbd_write_zone_get - Add a zone to the array of write zones. |
410a071c | 477 | * @td: fio thread data. |
a4807046 | 478 | * @f: fio file that has the write zones array to add. |
410a071c | 479 | * @zone_idx: Index of the zone to add. |
bfbdd35b | 480 | * |
f539b98c SK |
481 | * Do same operation as @zbd_write_zone_get, except it adds the zone at |
482 | * @zone_idx to write target zones array even when it does not have remainder | |
483 | * space to write one block. | |
bfbdd35b | 484 | */ |
f539b98c SK |
485 | static bool __zbd_write_zone_get(struct thread_data *td, |
486 | const struct fio_file *f, | |
487 | struct fio_zone_info *z) | |
1f57803b | 488 | { |
410a071c | 489 | struct zoned_block_device_info *zbdi = f->zbd_info; |
dc8a3d62 | 490 | uint32_t zone_idx = zbd_zone_idx(f, z); |
410a071c | 491 | bool res = true; |
fae3b9a0 | 492 | |
410a071c DLM |
493 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
494 | return false; | |
43bcbd5b | 495 | |
1f57803b | 496 | /* |
410a071c DLM |
497 | * Skip full zones with data verification enabled because resetting a |
498 | * zone causes data loss and hence causes verification to fail. | |
1f57803b | 499 | */ |
f539b98c | 500 | if (td->o.verify != VERIFY_NONE && zbd_zone_remainder(z) == 0) |
410a071c | 501 | return false; |
4d4c71e6 | 502 | |
410a071c | 503 | /* |
a4807046 SK |
504 | * zbdi->max_write_zones == 0 means that there is no limit on the |
505 | * maximum number of write target zones. In this case, do no track write | |
506 | * target zones in zbdi->write_zones array. | |
410a071c | 507 | */ |
a4807046 | 508 | if (!zbdi->max_write_zones) |
410a071c | 509 | return true; |
4d4c71e6 | 510 | |
410a071c | 511 | pthread_mutex_lock(&zbdi->mutex); |
b5a0f7ce | 512 | |
a4807046 | 513 | if (z->write) { |
410a071c | 514 | /* |
b5a0f7ce | 515 | * If the zone is going to be completely filled by writes |
a4807046 SK |
516 | * already in-flight, handle it as a full zone instead of a |
517 | * write target zone. | |
410a071c | 518 | */ |
df67bf1e | 519 | if (!zbd_zone_remainder(z)) |
410a071c DLM |
520 | res = false; |
521 | goto out; | |
522 | } | |
139d8dc6 | 523 | |
410a071c DLM |
524 | res = false; |
525 | /* Zero means no limit */ | |
526 | if (td->o.job_max_open_zones > 0 && | |
a4807046 | 527 | td->num_write_zones >= td->o.job_max_open_zones) |
410a071c | 528 | goto out; |
a4807046 | 529 | if (zbdi->num_write_zones >= zbdi->max_write_zones) |
410a071c | 530 | goto out; |
139d8dc6 | 531 | |
a4807046 | 532 | dprint(FD_ZBD, "%s: adding zone %u to write zone array\n", |
139d8dc6 DLM |
533 | f->file_name, zone_idx); |
534 | ||
a4807046 SK |
535 | zbdi->write_zones[zbdi->num_write_zones++] = zone_idx; |
536 | td->num_write_zones++; | |
537 | z->write = 1; | |
410a071c | 538 | res = true; |
bfbdd35b | 539 | |
410a071c DLM |
540 | out: |
541 | pthread_mutex_unlock(&zbdi->mutex); | |
542 | return res; | |
923f7c1e DF |
543 | } |
544 | ||
f539b98c SK |
545 | /** |
546 | * zbd_write_zone_get - Add a zone to the array of write zones. | |
547 | * @td: fio thread data. | |
548 | * @f: fio file that has the open zones to add. | |
549 | * @zone_idx: Index of the zone to add. | |
550 | * | |
551 | * Add a ZBD zone to write target zones array, if it is not yet added. Returns | |
552 | * true if either the zone was already added or if the zone was successfully | |
553 | * added to the array without exceeding the maximum number of write zones. | |
554 | * Returns false if the zone was not already added and addition of the zone | |
555 | * would cause the zone limit to be exceeded. | |
556 | */ | |
557 | static bool zbd_write_zone_get(struct thread_data *td, const struct fio_file *f, | |
558 | struct fio_zone_info *z) | |
559 | { | |
560 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
561 | ||
562 | /* | |
563 | * Skip full zones with data verification enabled because resetting a | |
564 | * zone causes data loss and hence causes verification to fail. | |
565 | */ | |
566 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) | |
567 | return false; | |
568 | ||
569 | return __zbd_write_zone_get(td, f, z); | |
570 | } | |
571 | ||
59c3200d | 572 | /* Verify whether direct I/O is used for all host-managed zoned block drives. */ |
bfbdd35b BVA |
573 | static bool zbd_using_direct_io(void) |
574 | { | |
bfbdd35b | 575 | struct fio_file *f; |
da8f124f | 576 | int j; |
bfbdd35b | 577 | |
da8f124f | 578 | for_each_td(td) { |
bfbdd35b BVA |
579 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) |
580 | continue; | |
581 | for_each_file(td, f, j) { | |
59c3200d | 582 | if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK && |
b7694961 | 583 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
584 | return false; |
585 | } | |
da8f124f | 586 | } end_for_each(); |
bfbdd35b BVA |
587 | |
588 | return true; | |
589 | } | |
590 | ||
591 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
b3e9bd03 | 592 | static bool zbd_is_seq_job(const struct fio_file *f) |
bfbdd35b BVA |
593 | { |
594 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
595 | ||
596 | assert(f->zbd_info); | |
139d8dc6 | 597 | |
bfbdd35b BVA |
598 | if (f->io_size == 0) |
599 | return false; | |
139d8dc6 | 600 | |
dc8a3d62 DLM |
601 | zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset); |
602 | zone_idx_e = | |
603 | zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1); | |
bfbdd35b | 604 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) |
39e06ee7 | 605 | if (zbd_get_zone(f, zone_idx)->has_wp) |
bfbdd35b BVA |
606 | return true; |
607 | ||
608 | return false; | |
609 | } | |
610 | ||
0bf93a1a DLM |
611 | /* |
612 | * Verify whether the file offset and size parameters are aligned with zone | |
613 | * boundaries. If the file offset is not aligned, align it down to the start of | |
614 | * the zone containing the start offset and align up the file io_size parameter. | |
615 | */ | |
616 | static bool zbd_zone_align_file_sizes(struct thread_data *td, | |
617 | struct fio_file *f) | |
618 | { | |
619 | const struct fio_zone_info *z; | |
620 | uint64_t new_offset, new_end; | |
0bf93a1a DLM |
621 | |
622 | if (!f->zbd_info) | |
623 | return true; | |
624 | if (f->file_offset >= f->real_file_size) | |
625 | return true; | |
626 | if (!zbd_is_seq_job(f)) | |
627 | return true; | |
628 | ||
629 | if (!td->o.zone_size) { | |
630 | td->o.zone_size = f->zbd_info->zone_size; | |
631 | if (!td->o.zone_size) { | |
632 | log_err("%s: invalid 0 zone size\n", | |
633 | f->file_name); | |
634 | return false; | |
635 | } | |
636 | } else if (td->o.zone_size != f->zbd_info->zone_size) { | |
637 | log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n", | |
638 | f->file_name, td->o.zone_size, | |
639 | f->zbd_info->zone_size); | |
640 | return false; | |
641 | } | |
642 | ||
643 | if (td->o.zone_skip % td->o.zone_size) { | |
644 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", | |
645 | f->file_name, td->o.zone_skip, | |
646 | td->o.zone_size); | |
647 | return false; | |
648 | } | |
649 | ||
53aa6171 | 650 | z = zbd_offset_to_zone(f, f->file_offset); |
0bf93a1a DLM |
651 | if ((f->file_offset != z->start) && |
652 | (td->o.td_ddir != TD_DDIR_READ)) { | |
653 | new_offset = zbd_zone_end(z); | |
654 | if (new_offset >= f->file_offset + f->io_size) { | |
655 | log_info("%s: io_size must be at least one zone\n", | |
656 | f->file_name); | |
657 | return false; | |
658 | } | |
659 | log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n", | |
660 | f->file_name, f->file_offset, | |
661 | new_offset); | |
662 | f->io_size -= (new_offset - f->file_offset); | |
663 | f->file_offset = new_offset; | |
664 | } | |
665 | ||
53aa6171 | 666 | z = zbd_offset_to_zone(f, f->file_offset + f->io_size); |
0bf93a1a DLM |
667 | new_end = z->start; |
668 | if ((td->o.td_ddir != TD_DDIR_READ) && | |
669 | (f->file_offset + f->io_size != new_end)) { | |
670 | if (new_end <= f->file_offset) { | |
671 | log_info("%s: io_size must be at least one zone\n", | |
672 | f->file_name); | |
673 | return false; | |
674 | } | |
675 | log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n", | |
676 | f->file_name, f->io_size, | |
677 | new_end - f->file_offset); | |
678 | f->io_size = new_end - f->file_offset; | |
679 | } | |
680 | ||
681 | return true; | |
682 | } | |
683 | ||
bfbdd35b BVA |
684 | /* |
685 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
686 | */ | |
687 | static bool zbd_verify_sizes(void) | |
688 | { | |
bfbdd35b | 689 | struct fio_file *f; |
da8f124f | 690 | int j; |
bfbdd35b | 691 | |
da8f124f | 692 | for_each_td(td) { |
bfbdd35b | 693 | for_each_file(td, f, j) { |
0bf93a1a | 694 | if (!zbd_zone_align_file_sizes(td, f)) |
4d37720a | 695 | return false; |
bfbdd35b | 696 | } |
da8f124f | 697 | } end_for_each(); |
bfbdd35b BVA |
698 | |
699 | return true; | |
700 | } | |
701 | ||
702 | static bool zbd_verify_bs(void) | |
703 | { | |
bfbdd35b | 704 | struct fio_file *f; |
da8f124f | 705 | int j; |
bfbdd35b | 706 | |
da8f124f | 707 | for_each_td(td) { |
e3be810b SK |
708 | if (td_trim(td) && |
709 | (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] || | |
710 | td->o.bssplit_nr[DDIR_TRIM])) { | |
711 | log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n"); | |
712 | return false; | |
713 | } | |
bfbdd35b | 714 | for_each_file(td, f, j) { |
1ddd225e AD |
715 | uint64_t zone_size; |
716 | ||
bfbdd35b BVA |
717 | if (!f->zbd_info) |
718 | continue; | |
139d8dc6 | 719 | |
bfbdd35b | 720 | zone_size = f->zbd_info->zone_size; |
e3be810b | 721 | if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) { |
ee5e3436 | 722 | log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n", |
e3be810b | 723 | f->file_name, td->o.bs[DDIR_TRIM], |
ee5e3436 | 724 | zone_size); |
e3be810b SK |
725 | return false; |
726 | } | |
bfbdd35b | 727 | } |
da8f124f | 728 | } end_for_each(); |
bfbdd35b BVA |
729 | return true; |
730 | } | |
731 | ||
bfbdd35b BVA |
732 | static int ilog2(uint64_t i) |
733 | { | |
734 | int log = -1; | |
735 | ||
736 | while (i) { | |
737 | i >>= 1; | |
738 | log++; | |
739 | } | |
740 | return log; | |
741 | } | |
742 | ||
743 | /* | |
744 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
745 | * allows to execute a ZBD workload against a non-ZBD device. | |
746 | */ | |
747 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
748 | { | |
749 | uint32_t nr_zones; | |
750 | struct fio_zone_info *p; | |
a4b7f12b | 751 | uint64_t zone_size = td->o.zone_size; |
b8dd9750 | 752 | uint64_t zone_capacity = td->o.zone_capacity; |
bfbdd35b | 753 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
754 | int i; |
755 | ||
a4b7f12b | 756 | if (zone_size == 0) { |
9db0cde8 | 757 | log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n", |
a4b7f12b DLM |
758 | f->file_name); |
759 | return 1; | |
760 | } | |
761 | ||
762 | if (zone_size < 512) { | |
763 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
764 | f->file_name); | |
765 | return 1; | |
766 | } | |
767 | ||
b8dd9750 HH |
768 | if (zone_capacity == 0) |
769 | zone_capacity = zone_size; | |
770 | ||
771 | if (zone_capacity > zone_size) { | |
772 | log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", | |
ee5e3436 | 773 | f->file_name, td->o.zone_capacity, td->o.zone_size); |
b8dd9750 HH |
774 | return 1; |
775 | } | |
776 | ||
9db0cde8 NC |
777 | if (f->real_file_size < zone_size) { |
778 | log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n", | |
779 | f->file_name, f->real_file_size, zone_size); | |
780 | return -EINVAL; | |
781 | } | |
782 | ||
ee3696bd | 783 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
784 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
785 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
786 | if (!zbd_info) | |
787 | return -ENOMEM; | |
788 | ||
44ec32cb | 789 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
790 | zbd_info->refcount = 1; |
791 | p = &zbd_info->zone_info[0]; | |
792 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
793 | mutex_init_pshared_with_type(&p->mutex, |
794 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b | 795 | p->start = i * zone_size; |
b14651a2 | 796 | p->wp = p->start; |
b7694961 DLM |
797 | p->type = ZBD_ZONE_TYPE_SWR; |
798 | p->cond = ZBD_ZONE_COND_EMPTY; | |
b8dd9750 | 799 | p->capacity = zone_capacity; |
be7a6bae | 800 | p->has_wp = 1; |
bfbdd35b BVA |
801 | } |
802 | /* a sentinel */ | |
803 | p->start = nr_zones * zone_size; | |
804 | ||
805 | f->zbd_info = zbd_info; | |
806 | f->zbd_info->zone_size = zone_size; | |
807 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 808 | ilog2(zone_size) : 0; |
bfbdd35b | 809 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
810 | return 0; |
811 | } | |
812 | ||
813 | /* | |
b7694961 DLM |
814 | * Maximum number of zones to report in one operation. |
815 | */ | |
816 | #define ZBD_REPORT_MAX_ZONES 8192U | |
817 | ||
818 | /* | |
819 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
820 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
821 | */ |
822 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
823 | { | |
b7694961 DLM |
824 | int nr_zones, nrz; |
825 | struct zbd_zone *zones, *z; | |
bfbdd35b | 826 | struct fio_zone_info *p; |
04f9090b BVA |
827 | uint64_t zone_size, offset, capacity; |
828 | bool same_zone_cap = true; | |
bfbdd35b | 829 | struct zoned_block_device_info *zbd_info = NULL; |
d060babc | 830 | int i, j, ret = -ENOMEM; |
bfbdd35b | 831 | |
b7694961 DLM |
832 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
833 | if (!zones) | |
bfbdd35b BVA |
834 | goto out; |
835 | ||
b7694961 DLM |
836 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
837 | if (nrz < 0) { | |
838 | ret = nrz; | |
839 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
840 | f->file_name, -ret); | |
841 | goto out; | |
bfbdd35b BVA |
842 | } |
843 | ||
b7694961 | 844 | zone_size = zones[0].len; |
04f9090b | 845 | capacity = zones[0].capacity; |
ee3696bd | 846 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
847 | |
848 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
849 | td->o.zone_size = zone_size; |
850 | } else if (td->o.zone_size != zone_size) { | |
ee5e3436 SK |
851 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n", |
852 | f->file_name, td->o.zone_size, zone_size); | |
bfbdd35b | 853 | ret = -EINVAL; |
b7694961 | 854 | goto out; |
bfbdd35b BVA |
855 | } |
856 | ||
9724b4f5 NC |
857 | dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", |
858 | f->file_name, nr_zones, zone_size / 1024); | |
bfbdd35b BVA |
859 | |
860 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
861 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
bfbdd35b | 862 | if (!zbd_info) |
b7694961 | 863 | goto out; |
44ec32cb | 864 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
865 | zbd_info->refcount = 1; |
866 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
867 | for (offset = 0, j = 0; j < nr_zones;) { |
868 | z = &zones[0]; | |
869 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
870 | mutex_init_pshared_with_type(&p->mutex, |
871 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 872 | p->start = z->start; |
236d23a8 | 873 | p->capacity = z->capacity; |
04f9090b BVA |
874 | if (capacity != z->capacity) |
875 | same_zone_cap = false; | |
139d8dc6 | 876 | |
bfbdd35b | 877 | switch (z->cond) { |
b7694961 DLM |
878 | case ZBD_ZONE_COND_NOT_WP: |
879 | case ZBD_ZONE_COND_FULL: | |
236d23a8 | 880 | p->wp = p->start + p->capacity; |
bfbdd35b BVA |
881 | break; |
882 | default: | |
883 | assert(z->start <= z->wp); | |
b7694961 DLM |
884 | assert(z->wp <= z->start + zone_size); |
885 | p->wp = z->wp; | |
bfbdd35b BVA |
886 | break; |
887 | } | |
be7a6bae DF |
888 | |
889 | switch (z->type) { | |
890 | case ZBD_ZONE_TYPE_SWR: | |
891 | p->has_wp = 1; | |
892 | break; | |
893 | default: | |
894 | p->has_wp = 0; | |
895 | } | |
bfbdd35b BVA |
896 | p->type = z->type; |
897 | p->cond = z->cond; | |
be7a6bae | 898 | |
bfbdd35b | 899 | if (j > 0 && p->start != p[-1].start + zone_size) { |
adfa7b7c BVA |
900 | log_info("%s: invalid zone data [%d:%d]: %"PRIu64" + %"PRIu64" != %"PRIu64"\n", |
901 | f->file_name, j, i, | |
902 | p[-1].start, zone_size, p->start); | |
bfbdd35b | 903 | ret = -EINVAL; |
b7694961 | 904 | goto out; |
bfbdd35b BVA |
905 | } |
906 | } | |
907 | z--; | |
b7694961 | 908 | offset = z->start + z->len; |
bfbdd35b BVA |
909 | if (j >= nr_zones) |
910 | break; | |
139d8dc6 | 911 | |
6c3f1cc1 DF |
912 | nrz = zbd_report_zones(td, f, offset, zones, |
913 | min((uint32_t)(nr_zones - j), | |
914 | ZBD_REPORT_MAX_ZONES)); | |
b7694961 DLM |
915 | if (nrz < 0) { |
916 | ret = nrz; | |
ee5e3436 SK |
917 | log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n", |
918 | offset, f->file_name, -ret); | |
b7694961 | 919 | goto out; |
bfbdd35b BVA |
920 | } |
921 | } | |
b7694961 | 922 | |
bfbdd35b | 923 | /* a sentinel */ |
b7694961 | 924 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
925 | |
926 | f->zbd_info = zbd_info; | |
927 | f->zbd_info->zone_size = zone_size; | |
928 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 929 | ilog2(zone_size) : 0; |
bfbdd35b | 930 | f->zbd_info->nr_zones = nr_zones; |
04f9090b BVA |
931 | |
932 | if (same_zone_cap) | |
933 | dprint(FD_ZBD, "Zone capacity = %"PRIu64" KB\n", | |
934 | capacity / 1024); | |
935 | ||
bfbdd35b BVA |
936 | zbd_info = NULL; |
937 | ret = 0; | |
938 | ||
bfbdd35b | 939 | out: |
b7694961 DLM |
940 | sfree(zbd_info); |
941 | free(zones); | |
bfbdd35b BVA |
942 | return ret; |
943 | } | |
944 | ||
a4807046 | 945 | static int zbd_set_max_write_zones(struct thread_data *td, struct fio_file *f) |
d2f442bc NC |
946 | { |
947 | struct zoned_block_device_info *zbd = f->zbd_info; | |
948 | unsigned int max_open_zones; | |
949 | int ret; | |
950 | ||
575686bb | 951 | if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) { |
d2f442bc | 952 | /* Only host-managed devices have a max open limit */ |
a4807046 | 953 | zbd->max_write_zones = td->o.max_open_zones; |
d2f442bc NC |
954 | goto out; |
955 | } | |
956 | ||
957 | /* If host-managed, get the max open limit */ | |
958 | ret = zbd_get_max_open_zones(td, f, &max_open_zones); | |
959 | if (ret) | |
960 | return ret; | |
961 | ||
962 | if (!max_open_zones) { | |
963 | /* No device limit */ | |
a4807046 | 964 | zbd->max_write_zones = td->o.max_open_zones; |
d2f442bc NC |
965 | } else if (!td->o.max_open_zones) { |
966 | /* No user limit. Set limit to device limit */ | |
a4807046 | 967 | zbd->max_write_zones = max_open_zones; |
d2f442bc NC |
968 | } else if (td->o.max_open_zones <= max_open_zones) { |
969 | /* Both user limit and dev limit. User limit not too large */ | |
a4807046 | 970 | zbd->max_write_zones = td->o.max_open_zones; |
d2f442bc NC |
971 | } else { |
972 | /* Both user limit and dev limit. User limit too large */ | |
973 | td_verror(td, EINVAL, | |
974 | "Specified --max_open_zones is too large"); | |
975 | log_err("Specified --max_open_zones (%d) is larger than max (%u)\n", | |
976 | td->o.max_open_zones, max_open_zones); | |
977 | return -EINVAL; | |
978 | } | |
979 | ||
980 | out: | |
981 | /* Ensure that the limit is not larger than FIO's internal limit */ | |
a4807046 | 982 | if (zbd->max_write_zones > ZBD_MAX_WRITE_ZONES) { |
b346af90 | 983 | td_verror(td, EINVAL, "'max_open_zones' value is too large"); |
139d8dc6 | 984 | log_err("'max_open_zones' value is larger than %u\n", |
a4807046 | 985 | ZBD_MAX_WRITE_ZONES); |
b346af90 NC |
986 | return -EINVAL; |
987 | } | |
988 | ||
a4807046 SK |
989 | dprint(FD_ZBD, "%s: using max write zones limit: %"PRIu32"\n", |
990 | f->file_name, zbd->max_write_zones); | |
d2f442bc NC |
991 | |
992 | return 0; | |
993 | } | |
994 | ||
bfbdd35b BVA |
995 | /* |
996 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
997 | * | |
998 | * Returns 0 upon success and a negative error code upon failure. | |
999 | */ | |
379e5f09 | 1000 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 1001 | { |
b7694961 DLM |
1002 | enum zbd_zoned_model zbd_model; |
1003 | int ret; | |
bfbdd35b BVA |
1004 | |
1005 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1006 | ||
b7694961 DLM |
1007 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
1008 | if (ret) | |
1009 | return ret; | |
1010 | ||
bfbdd35b | 1011 | switch (zbd_model) { |
b7694961 DLM |
1012 | case ZBD_HOST_AWARE: |
1013 | case ZBD_HOST_MANAGED: | |
bfbdd35b | 1014 | ret = parse_zone_info(td, f); |
d2f442bc NC |
1015 | if (ret) |
1016 | return ret; | |
bfbdd35b | 1017 | break; |
b7694961 | 1018 | case ZBD_NONE: |
bfbdd35b | 1019 | ret = init_zone_info(td, f); |
d2f442bc NC |
1020 | if (ret) |
1021 | return ret; | |
bfbdd35b | 1022 | break; |
b7694961 DLM |
1023 | default: |
1024 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
1025 | log_err("Unsupported zoned model\n"); | |
1026 | return -EINVAL; | |
bfbdd35b | 1027 | } |
b7694961 | 1028 | |
2c7dd23e | 1029 | assert(f->zbd_info); |
d2f442bc NC |
1030 | f->zbd_info->model = zbd_model; |
1031 | ||
a4807046 | 1032 | ret = zbd_set_max_write_zones(td, f); |
d2f442bc NC |
1033 | if (ret) { |
1034 | zbd_free_zone_info(f); | |
1035 | return ret; | |
219c662d | 1036 | } |
d2f442bc NC |
1037 | |
1038 | return 0; | |
bfbdd35b BVA |
1039 | } |
1040 | ||
1041 | void zbd_free_zone_info(struct fio_file *f) | |
1042 | { | |
1043 | uint32_t refcount; | |
1044 | ||
3c1dc34c | 1045 | assert(f->zbd_info); |
bfbdd35b BVA |
1046 | |
1047 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1048 | refcount = --f->zbd_info->refcount; | |
1049 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1050 | ||
1051 | assert((int32_t)refcount >= 0); | |
1052 | if (refcount == 0) | |
1053 | sfree(f->zbd_info); | |
1054 | f->zbd_info = NULL; | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * Initialize f->zbd_info. | |
1059 | * | |
1060 | * Returns 0 upon success and a negative error code upon failure. | |
1061 | * | |
1062 | * Note: this function can only work correctly if it is called before the first | |
1063 | * fio fork() call. | |
1064 | */ | |
1065 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
1066 | { | |
bfbdd35b | 1067 | struct fio_file *f2; |
da8f124f | 1068 | int j, ret; |
bfbdd35b | 1069 | |
da8f124f | 1070 | for_each_td(td2) { |
bfbdd35b BVA |
1071 | for_each_file(td2, f2, j) { |
1072 | if (td2 == td && f2 == file) | |
1073 | continue; | |
1074 | if (!f2->zbd_info || | |
1075 | strcmp(f2->file_name, file->file_name) != 0) | |
1076 | continue; | |
1077 | file->zbd_info = f2->zbd_info; | |
1078 | file->zbd_info->refcount++; | |
1079 | return 0; | |
1080 | } | |
da8f124f | 1081 | } end_for_each(); |
bfbdd35b BVA |
1082 | |
1083 | ret = zbd_create_zone_info(td, file); | |
1084 | if (ret < 0) | |
c5837eec | 1085 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
139d8dc6 | 1086 | |
bfbdd35b BVA |
1087 | return ret; |
1088 | } | |
1089 | ||
8f39afa7 | 1090 | int zbd_init_files(struct thread_data *td) |
bfbdd35b BVA |
1091 | { |
1092 | struct fio_file *f; | |
1093 | int i; | |
1094 | ||
1095 | for_each_file(td, f, i) { | |
a4b7f12b | 1096 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 1097 | return 1; |
bfbdd35b | 1098 | } |
139d8dc6 | 1099 | |
8f39afa7 AD |
1100 | return 0; |
1101 | } | |
1102 | ||
1103 | void zbd_recalc_options_with_zone_granularity(struct thread_data *td) | |
1104 | { | |
1105 | struct fio_file *f; | |
1106 | int i; | |
1107 | ||
1108 | for_each_file(td, f, i) { | |
1109 | struct zoned_block_device_info *zbd = f->zbd_info; | |
139d8dc6 | 1110 | uint64_t zone_size; |
8f39afa7 | 1111 | |
139d8dc6 DLM |
1112 | /* zonemode=strided doesn't get per-file zone size. */ |
1113 | zone_size = zbd ? zbd->zone_size : td->o.zone_size; | |
8f39afa7 AD |
1114 | if (zone_size == 0) |
1115 | continue; | |
1116 | ||
139d8dc6 | 1117 | if (td->o.size_nz > 0) |
8f39afa7 | 1118 | td->o.size = td->o.size_nz * zone_size; |
139d8dc6 | 1119 | if (td->o.io_size_nz > 0) |
8f39afa7 | 1120 | td->o.io_size = td->o.io_size_nz * zone_size; |
139d8dc6 | 1121 | if (td->o.start_offset_nz > 0) |
8f39afa7 | 1122 | td->o.start_offset = td->o.start_offset_nz * zone_size; |
139d8dc6 DLM |
1123 | if (td->o.offset_increment_nz > 0) |
1124 | td->o.offset_increment = | |
1125 | td->o.offset_increment_nz * zone_size; | |
1126 | if (td->o.zone_skip_nz > 0) | |
8f39afa7 | 1127 | td->o.zone_skip = td->o.zone_skip_nz * zone_size; |
8f39afa7 AD |
1128 | } |
1129 | } | |
1130 | ||
9fb714da SK |
1131 | static uint64_t zbd_verify_and_set_vdb(struct thread_data *td, |
1132 | const struct fio_file *f) | |
1133 | { | |
1134 | struct fio_zone_info *zb, *ze, *z; | |
1135 | uint64_t wp_vdb = 0; | |
1136 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
1137 | ||
1138 | assert(td->runstate < TD_RUNNING); | |
1139 | assert(zbdi); | |
1140 | ||
1141 | if (!accounting_vdb(td, f)) | |
1142 | return 0; | |
1143 | ||
1144 | /* | |
1145 | * Ensure that the I/O range includes one or more sequential zones so | |
1146 | * that f->min_zone and f->max_zone have different values. | |
1147 | */ | |
1148 | if (!zbd_is_seq_job(f)) | |
1149 | return 0; | |
1150 | ||
1151 | if (zbdi->write_min_zone != zbdi->write_max_zone) { | |
1152 | if (zbdi->write_min_zone != f->min_zone || | |
1153 | zbdi->write_max_zone != f->max_zone) { | |
1154 | td_verror(td, EINVAL, | |
1155 | "multi-jobs with different write ranges are " | |
1156 | "not supported with zone_reset_threshold"); | |
1157 | log_err("multi-jobs with different write ranges are " | |
1158 | "not supported with zone_reset_threshold\n"); | |
1159 | } | |
1160 | return 0; | |
1161 | } | |
1162 | ||
1163 | zbdi->write_min_zone = f->min_zone; | |
1164 | zbdi->write_max_zone = f->max_zone; | |
1165 | ||
1166 | zb = zbd_get_zone(f, f->min_zone); | |
1167 | ze = zbd_get_zone(f, f->max_zone); | |
1168 | for (z = zb; z < ze; z++) | |
1169 | if (z->has_wp) | |
1170 | wp_vdb += z->wp - z->start; | |
1171 | ||
1172 | zbdi->wp_valid_data_bytes = wp_vdb; | |
1173 | ||
1174 | return wp_vdb; | |
1175 | } | |
1176 | ||
8f39afa7 AD |
1177 | int zbd_setup_files(struct thread_data *td) |
1178 | { | |
1179 | struct fio_file *f; | |
1180 | int i; | |
bfbdd35b BVA |
1181 | |
1182 | if (!zbd_using_direct_io()) { | |
1183 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
1184 | return 1; | |
1185 | } | |
1186 | ||
1187 | if (!zbd_verify_sizes()) | |
1188 | return 1; | |
1189 | ||
1190 | if (!zbd_verify_bs()) | |
1191 | return 1; | |
1192 | ||
6e2da06a SK |
1193 | if (td->o.experimental_verify) { |
1194 | log_err("zonemode=zbd does not support experimental verify\n"); | |
1195 | return 1; | |
1196 | } | |
1197 | ||
219c662d AD |
1198 | for_each_file(td, f, i) { |
1199 | struct zoned_block_device_info *zbd = f->zbd_info; | |
954217b9 SK |
1200 | struct fio_zone_info *z; |
1201 | int zi; | |
9fb714da | 1202 | uint64_t vdb; |
219c662d | 1203 | |
5ddf46d0 | 1204 | assert(zbd); |
219c662d | 1205 | |
dc8a3d62 DLM |
1206 | f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset); |
1207 | f->max_zone = | |
1208 | zbd_offset_to_zone_idx(f, f->file_offset + f->io_size); | |
f952800a | 1209 | |
9fb714da SK |
1210 | vdb = zbd_verify_and_set_vdb(td, f); |
1211 | ||
1212 | dprint(FD_ZBD, "%s(%s): valid data bytes = %" PRIu64 "\n", | |
1213 | __func__, f->file_name, vdb); | |
1214 | ||
f952800a SK |
1215 | /* |
1216 | * When all zones in the I/O range are conventional, io_size | |
1217 | * can be smaller than zone size, making min_zone the same | |
1218 | * as max_zone. This is why the assert below needs to be made | |
1219 | * conditional. | |
1220 | */ | |
1221 | if (zbd_is_seq_job(f)) | |
1222 | assert(f->min_zone < f->max_zone); | |
1223 | ||
219c662d | 1224 | if (td->o.max_open_zones > 0 && |
a4807046 | 1225 | zbd->max_write_zones != td->o.max_open_zones) { |
219c662d AD |
1226 | log_err("Different 'max_open_zones' values\n"); |
1227 | return 1; | |
1228 | } | |
b346af90 NC |
1229 | |
1230 | /* | |
1231 | * The per job max open zones limit cannot be used without a | |
1232 | * global max open zones limit. (As the tracking of open zones | |
1233 | * is disabled when there is no global max open zones limit.) | |
1234 | */ | |
a4807046 | 1235 | if (td->o.job_max_open_zones && !zbd->max_write_zones) { |
b346af90 | 1236 | log_err("'job_max_open_zones' cannot be used without a global open zones limit\n"); |
219c662d AD |
1237 | return 1; |
1238 | } | |
954217b9 | 1239 | |
ea51055c | 1240 | /* |
a4807046 | 1241 | * zbd->max_write_zones is the global limit shared for all jobs |
ea51055c NC |
1242 | * that target the same zoned block device. Force sync the per |
1243 | * thread global limit with the actual global limit. (The real | |
1244 | * per thread/job limit is stored in td->o.job_max_open_zones). | |
1245 | */ | |
a4807046 | 1246 | td->o.max_open_zones = zbd->max_write_zones; |
ea51055c | 1247 | |
954217b9 SK |
1248 | for (zi = f->min_zone; zi < f->max_zone; zi++) { |
1249 | z = &zbd->zone_info[zi]; | |
1250 | if (z->cond != ZBD_ZONE_COND_IMP_OPEN && | |
1251 | z->cond != ZBD_ZONE_COND_EXP_OPEN) | |
1252 | continue; | |
f539b98c | 1253 | if (__zbd_write_zone_get(td, f, z)) |
954217b9 SK |
1254 | continue; |
1255 | /* | |
1256 | * If the number of open zones exceeds specified limits, | |
8ac76889 | 1257 | * error out. |
954217b9 | 1258 | */ |
8ac76889 SK |
1259 | log_err("Number of open zones exceeds max_open_zones limit\n"); |
1260 | return 1; | |
954217b9 | 1261 | } |
219c662d AD |
1262 | } |
1263 | ||
bfbdd35b BVA |
1264 | return 0; |
1265 | } | |
1266 | ||
a7c2b6fc BVA |
1267 | /* |
1268 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
1269 | * zone reset. | |
1270 | */ | |
1bb1bcad AD |
1271 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
1272 | const struct fio_file *f) | |
a7c2b6fc BVA |
1273 | { |
1274 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
1275 | ||
a7c2b6fc BVA |
1276 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
1277 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
1278 | } |
1279 | ||
1280 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
1281 | const struct fio_file *f) | |
1282 | { | |
1283 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1284 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
1285 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1286 | } | |
1287 | ||
1288 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
1289 | const struct fio_file *f) | |
1290 | { | |
1291 | uint32_t write_cnt = 0; | |
1292 | ||
1293 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1294 | assert(f->zbd_info->write_cnt); | |
1295 | if (f->zbd_info->write_cnt) | |
1296 | write_cnt = --f->zbd_info->write_cnt; | |
1297 | if (write_cnt == 0) | |
1bb1bcad | 1298 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
1299 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1300 | ||
1301 | return write_cnt == 0; | |
1302 | } | |
1303 | ||
bfbdd35b BVA |
1304 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
1305 | { | |
91d25131 | 1306 | struct fio_zone_info *zb, *ze; |
c5c8b92b | 1307 | bool verify_data_left = false; |
bfbdd35b | 1308 | |
767d1372 | 1309 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
1310 | return; |
1311 | ||
39e06ee7 DLM |
1312 | zb = zbd_get_zone(f, f->min_zone); |
1313 | ze = zbd_get_zone(f, f->max_zone); | |
139d8dc6 | 1314 | |
bfbdd35b BVA |
1315 | /* |
1316 | * If data verification is enabled reset the affected zones before | |
1317 | * writing any data to avoid that a zone reset has to be issued while | |
1318 | * writing data, which causes data loss. | |
1319 | */ | |
c5c8b92b SK |
1320 | if (td->o.verify != VERIFY_NONE) { |
1321 | verify_data_left = td->runstate == TD_VERIFYING || | |
1322 | td->io_hist_len || td->verify_batch; | |
1323 | if (td->io_hist_len && td->o.verify_backlog) | |
1324 | verify_data_left = | |
1325 | td->io_hist_len % td->o.verify_backlog; | |
1326 | if (!verify_data_left) | |
1327 | zbd_reset_zones(td, f, zb, ze); | |
1328 | } | |
1329 | ||
a7c2b6fc | 1330 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
1331 | } |
1332 | ||
a4807046 | 1333 | /* Return random zone index for one of the write target zones. */ |
6463db6c AD |
1334 | static uint32_t pick_random_zone_idx(const struct fio_file *f, |
1335 | const struct io_u *io_u) | |
1336 | { | |
139d8dc6 | 1337 | return (io_u->offset - f->file_offset) * |
a4807046 | 1338 | f->zbd_info->num_write_zones / f->io_size; |
6463db6c AD |
1339 | } |
1340 | ||
0f77c977 SK |
1341 | static bool any_io_in_flight(void) |
1342 | { | |
da8f124f | 1343 | for_each_td(td) { |
0f77c977 SK |
1344 | if (td->io_u_in_flight) |
1345 | return true; | |
da8f124f | 1346 | } end_for_each(); |
0f77c977 SK |
1347 | |
1348 | return false; | |
1349 | } | |
1350 | ||
59b07544 | 1351 | /* |
a4807046 SK |
1352 | * Modify the offset of an I/O unit that does not refer to a zone such that |
1353 | * in write target zones array. Add a zone to or remove a zone from the lsit if | |
1354 | * necessary. The write target zone is searched across sequential zones. | |
21c0c884 | 1355 | * This algorithm can only work correctly if all write pointers are |
59b07544 BVA |
1356 | * a multiple of the fio block size. The caller must neither hold z->mutex |
1357 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
1358 | */ | |
a4807046 SK |
1359 | static struct fio_zone_info *zbd_convert_to_write_zone(struct thread_data *td, |
1360 | struct io_u *io_u) | |
59b07544 | 1361 | { |
07fc3f57 | 1362 | const uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
fae3b9a0 | 1363 | struct fio_file *f = io_u->file; |
af94a8c3 | 1364 | struct zoned_block_device_info *zbdi = f->zbd_info; |
59b07544 | 1365 | struct fio_zone_info *z; |
a4807046 | 1366 | unsigned int write_zone_idx = -1; |
59b07544 BVA |
1367 | uint32_t zone_idx, new_zone_idx; |
1368 | int i; | |
a4807046 | 1369 | bool wait_zone_write; |
0f77c977 SK |
1370 | bool in_flight; |
1371 | bool should_retry = true; | |
59b07544 BVA |
1372 | |
1373 | assert(is_valid_offset(f, io_u->offset)); | |
1374 | ||
a4807046 | 1375 | if (zbdi->max_write_zones || td->o.job_max_open_zones) { |
59b07544 | 1376 | /* |
a4807046 | 1377 | * This statement accesses zbdi->write_zones[] on purpose |
59b07544 BVA |
1378 | * without locking. |
1379 | */ | |
a4807046 | 1380 | zone_idx = zbdi->write_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 | 1381 | } else { |
dc8a3d62 | 1382 | zone_idx = zbd_offset_to_zone_idx(f, io_u->offset); |
59b07544 | 1383 | } |
fae3b9a0 AD |
1384 | if (zone_idx < f->min_zone) |
1385 | zone_idx = f->min_zone; | |
1386 | else if (zone_idx >= f->max_zone) | |
1387 | zone_idx = f->max_zone - 1; | |
139d8dc6 DLM |
1388 | |
1389 | dprint(FD_ZBD, | |
1390 | "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", | |
59b07544 BVA |
1391 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); |
1392 | ||
1393 | /* | |
af94a8c3 | 1394 | * Since z->mutex is the outer lock and zbdi->mutex the inner |
59b07544 | 1395 | * lock it can happen that the state of the zone with index zone_idx |
af94a8c3 | 1396 | * has changed after 'z' has been assigned and before zbdi->mutex |
59b07544 BVA |
1397 | * has been obtained. Hence the loop. |
1398 | */ | |
1399 | for (;;) { | |
6463db6c AD |
1400 | uint32_t tmp_idx; |
1401 | ||
39e06ee7 | 1402 | z = zbd_get_zone(f, zone_idx); |
14351148 DF |
1403 | if (z->has_wp) |
1404 | zone_lock(td, f, z); | |
139d8dc6 | 1405 | |
af94a8c3 | 1406 | pthread_mutex_lock(&zbdi->mutex); |
139d8dc6 | 1407 | |
14351148 DF |
1408 | if (z->has_wp) { |
1409 | if (z->cond != ZBD_ZONE_COND_OFFLINE && | |
a4807046 | 1410 | zbdi->max_write_zones == 0 && |
139d8dc6 | 1411 | td->o.job_max_open_zones == 0) |
14351148 | 1412 | goto examine_zone; |
a4807046 SK |
1413 | if (zbdi->num_write_zones == 0) { |
1414 | dprint(FD_ZBD, "%s(%s): no zone is write target\n", | |
14351148 | 1415 | __func__, f->file_name); |
a4807046 | 1416 | goto choose_other_zone; |
14351148 | 1417 | } |
59b07544 | 1418 | } |
6463db6c AD |
1419 | |
1420 | /* | |
a4807046 | 1421 | * Array of write target zones is per-device, shared across all |
139d8dc6 DLM |
1422 | * threads. Start with quasi-random candidate zone. Ignore |
1423 | * zones which don't belong to thread's offset/size area. | |
6463db6c | 1424 | */ |
a4807046 SK |
1425 | write_zone_idx = pick_random_zone_idx(f, io_u); |
1426 | assert(!write_zone_idx || | |
1427 | write_zone_idx < zbdi->num_write_zones); | |
1428 | tmp_idx = write_zone_idx; | |
139d8dc6 | 1429 | |
a4807046 | 1430 | for (i = 0; i < zbdi->num_write_zones; i++) { |
6463db6c AD |
1431 | uint32_t tmpz; |
1432 | ||
a4807046 | 1433 | if (tmp_idx >= zbdi->num_write_zones) |
6463db6c | 1434 | tmp_idx = 0; |
a4807046 | 1435 | tmpz = zbdi->write_zones[tmp_idx]; |
fae3b9a0 | 1436 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
a4807046 | 1437 | write_zone_idx = tmp_idx; |
6463db6c AD |
1438 | goto found_candidate_zone; |
1439 | } | |
1440 | ||
1441 | tmp_idx++; | |
1442 | } | |
1443 | ||
1444 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1445 | __func__, f->file_name); | |
139d8dc6 | 1446 | |
af94a8c3 | 1447 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1448 | |
14351148 DF |
1449 | if (z->has_wp) |
1450 | zone_unlock(z); | |
139d8dc6 | 1451 | |
6463db6c AD |
1452 | return NULL; |
1453 | ||
1454 | found_candidate_zone: | |
a4807046 | 1455 | new_zone_idx = zbdi->write_zones[write_zone_idx]; |
59b07544 BVA |
1456 | if (new_zone_idx == zone_idx) |
1457 | break; | |
1458 | zone_idx = new_zone_idx; | |
139d8dc6 | 1459 | |
af94a8c3 | 1460 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1461 | |
14351148 DF |
1462 | if (z->has_wp) |
1463 | zone_unlock(z); | |
59b07544 BVA |
1464 | } |
1465 | ||
af94a8c3 | 1466 | /* Both z->mutex and zbdi->mutex are held. */ |
59b07544 BVA |
1467 | |
1468 | examine_zone: | |
df67bf1e | 1469 | if (zbd_zone_remainder(z) >= min_bs) { |
af94a8c3 | 1470 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1471 | goto out; |
1472 | } | |
b2da58c4 | 1473 | |
a4807046 SK |
1474 | choose_other_zone: |
1475 | /* Check if number of write target zones reaches one of limits. */ | |
1476 | wait_zone_write = | |
1477 | zbdi->num_write_zones == f->max_zone - f->min_zone || | |
1478 | (zbdi->max_write_zones && | |
1479 | zbdi->num_write_zones == zbdi->max_write_zones) || | |
b2da58c4 | 1480 | (td->o.job_max_open_zones && |
a4807046 | 1481 | td->num_write_zones == td->o.job_max_open_zones); |
b2da58c4 | 1482 | |
af94a8c3 | 1483 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1484 | |
1485 | /* Only z->mutex is held. */ | |
1486 | ||
b2da58c4 | 1487 | /* |
a4807046 SK |
1488 | * When number of write target zones reaches to one of limits, wait for |
1489 | * zone write completion to one of them before trying a new zone. | |
b2da58c4 | 1490 | */ |
a4807046 | 1491 | if (wait_zone_write) { |
139d8dc6 | 1492 | dprint(FD_ZBD, |
a4807046 | 1493 | "%s(%s): quiesce to remove a zone from write target zones array\n", |
b2da58c4 SK |
1494 | __func__, f->file_name); |
1495 | io_u_quiesce(td); | |
1496 | } | |
1497 | ||
0f77c977 | 1498 | retry: |
a4807046 | 1499 | /* Zone 'z' is full, so try to choose a new zone. */ |
af94a8c3 | 1500 | for (i = f->io_size / zbdi->zone_size; i > 0; i--) { |
59b07544 | 1501 | zone_idx++; |
21c0c884 SK |
1502 | if (z->has_wp) |
1503 | zone_unlock(z); | |
59b07544 | 1504 | z++; |
ee3696bd | 1505 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1506 | /* Wrap-around. */ |
fae3b9a0 | 1507 | zone_idx = f->min_zone; |
39e06ee7 | 1508 | z = zbd_get_zone(f, zone_idx); |
59b07544 | 1509 | } |
ee3696bd | 1510 | assert(is_valid_offset(f, z->start)); |
21c0c884 SK |
1511 | if (!z->has_wp) |
1512 | continue; | |
fae3b9a0 | 1513 | zone_lock(td, f, z); |
a4807046 | 1514 | if (z->write) |
59b07544 | 1515 | continue; |
a4807046 | 1516 | if (zbd_write_zone_get(td, f, z)) |
59b07544 BVA |
1517 | goto out; |
1518 | } | |
1519 | ||
1520 | /* Only z->mutex is held. */ | |
1521 | ||
a4807046 | 1522 | /* Check whether the write fits in any of the write target zones. */ |
af94a8c3 | 1523 | pthread_mutex_lock(&zbdi->mutex); |
a4807046 SK |
1524 | for (i = 0; i < zbdi->num_write_zones; i++) { |
1525 | zone_idx = zbdi->write_zones[i]; | |
fae3b9a0 AD |
1526 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1527 | continue; | |
af94a8c3 | 1528 | pthread_mutex_unlock(&zbdi->mutex); |
4d4c71e6 | 1529 | zone_unlock(z); |
59b07544 | 1530 | |
39e06ee7 | 1531 | z = zbd_get_zone(f, zone_idx); |
59b07544 | 1532 | |
fae3b9a0 | 1533 | zone_lock(td, f, z); |
df67bf1e | 1534 | if (zbd_zone_remainder(z) >= min_bs) |
59b07544 | 1535 | goto out; |
af94a8c3 | 1536 | pthread_mutex_lock(&zbdi->mutex); |
59b07544 | 1537 | } |
0f77c977 SK |
1538 | |
1539 | /* | |
1540 | * When any I/O is in-flight or when all I/Os in-flight get completed, | |
a4807046 SK |
1541 | * the I/Os might have removed zones from the write target array then |
1542 | * retry the steps to choose a zone. Before retry, call io_u_quiesce() | |
1543 | * to complete in-flight writes. | |
0f77c977 SK |
1544 | */ |
1545 | in_flight = any_io_in_flight(); | |
1546 | if (in_flight || should_retry) { | |
139d8dc6 | 1547 | dprint(FD_ZBD, |
a4807046 | 1548 | "%s(%s): wait zone write and retry write target zone selection\n", |
0f77c977 | 1549 | __func__, f->file_name); |
62ac6649 | 1550 | should_retry = in_flight; |
0f77c977 SK |
1551 | pthread_mutex_unlock(&zbdi->mutex); |
1552 | zone_unlock(z); | |
1553 | io_u_quiesce(td); | |
1554 | zone_lock(td, f, z); | |
0f77c977 SK |
1555 | goto retry; |
1556 | } | |
1557 | ||
af94a8c3 | 1558 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1559 | |
4d4c71e6 | 1560 | zone_unlock(z); |
139d8dc6 | 1561 | |
a4807046 | 1562 | dprint(FD_ZBD, "%s(%s): did not choose another write zone\n", |
139d8dc6 DLM |
1563 | __func__, f->file_name); |
1564 | ||
59b07544 BVA |
1565 | return NULL; |
1566 | ||
1567 | out: | |
139d8dc6 DLM |
1568 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", |
1569 | __func__, f->file_name, zone_idx); | |
1570 | ||
ee3696bd | 1571 | io_u->offset = z->start; |
21c0c884 | 1572 | assert(z->has_wp); |
8a866de7 | 1573 | assert(z->cond != ZBD_ZONE_COND_OFFLINE); |
139d8dc6 | 1574 | |
59b07544 BVA |
1575 | return z; |
1576 | } | |
1577 | ||
bfbdd35b | 1578 | /* |
5c86fdf6 SK |
1579 | * Find another zone which has @min_bytes of readable data. Search in zones |
1580 | * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf. | |
bfbdd35b | 1581 | * |
21c0c884 SK |
1582 | * Either returns NULL or returns a zone pointer. When the zone has write |
1583 | * pointer, hold the mutex for the zone. | |
bfbdd35b BVA |
1584 | */ |
1585 | static struct fio_zone_info * | |
07fc3f57 | 1586 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes, |
bfbdd35b BVA |
1587 | struct fio_zone_info *zb, struct fio_zone_info *zl) |
1588 | { | |
fae3b9a0 | 1589 | struct fio_file *f = io_u->file; |
bfbdd35b | 1590 | struct fio_zone_info *z1, *z2; |
39e06ee7 | 1591 | const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone); |
bfbdd35b BVA |
1592 | |
1593 | /* | |
1594 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1595 | * the nearest non-empty zone in case of random I/O. | |
1596 | */ | |
1597 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1598 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1599 | if (z1->has_wp) |
1600 | zone_lock(td, f, z1); | |
5c86fdf6 | 1601 | if (z1->start + min_bytes <= z1->wp) |
bfbdd35b | 1602 | return z1; |
21c0c884 SK |
1603 | if (z1->has_wp) |
1604 | zone_unlock(z1); | |
bfbdd35b BVA |
1605 | } else if (!td_random(td)) { |
1606 | break; | |
1607 | } | |
139d8dc6 | 1608 | |
bfbdd35b | 1609 | if (td_random(td) && z2 >= zf && |
b7694961 | 1610 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1611 | if (z2->has_wp) |
1612 | zone_lock(td, f, z2); | |
5c86fdf6 | 1613 | if (z2->start + min_bytes <= z2->wp) |
bfbdd35b | 1614 | return z2; |
21c0c884 SK |
1615 | if (z2->has_wp) |
1616 | zone_unlock(z2); | |
bfbdd35b BVA |
1617 | } |
1618 | } | |
139d8dc6 DLM |
1619 | |
1620 | dprint(FD_ZBD, | |
1621 | "%s: no zone has %"PRIu64" bytes of readable data\n", | |
5c86fdf6 | 1622 | f->file_name, min_bytes); |
139d8dc6 | 1623 | |
bfbdd35b BVA |
1624 | return NULL; |
1625 | } | |
1626 | ||
b2da58c4 SK |
1627 | /** |
1628 | * zbd_end_zone_io - update zone status at command completion | |
1629 | * @io_u: I/O unit | |
1630 | * @z: zone info pointer | |
1631 | * | |
a4807046 SK |
1632 | * If the write command made the zone full, remove it from the write target |
1633 | * zones array. | |
b2da58c4 SK |
1634 | * |
1635 | * The caller must hold z->mutex. | |
1636 | */ | |
1637 | static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, | |
1638 | struct fio_zone_info *z) | |
1639 | { | |
1640 | const struct fio_file *f = io_u->file; | |
1641 | ||
1642 | if (io_u->ddir == DDIR_WRITE && | |
1643 | io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1644 | pthread_mutex_lock(&f->zbd_info->mutex); | |
a4807046 | 1645 | zbd_write_zone_put(td, f, z); |
b2da58c4 SK |
1646 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1647 | } | |
1648 | } | |
1649 | ||
bfbdd35b | 1650 | /** |
d9ed3e63 | 1651 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1652 | * @io_u: I/O unit |
d9ed3e63 DLM |
1653 | * @success: Whether or not the I/O unit has been queued successfully |
1654 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1655 | * |
d9ed3e63 DLM |
1656 | * For write and trim operations, update the write pointer of the I/O unit |
1657 | * target zone. | |
bfbdd35b | 1658 | */ |
b2da58c4 SK |
1659 | static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, |
1660 | bool success) | |
bfbdd35b | 1661 | { |
d9ed3e63 DLM |
1662 | const struct fio_file *f = io_u->file; |
1663 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b | 1664 | struct fio_zone_info *z; |
d9ed3e63 | 1665 | uint64_t zone_end; |
bfbdd35b | 1666 | |
5ddf46d0 | 1667 | assert(zbd_info); |
bfbdd35b | 1668 | |
53aa6171 | 1669 | z = zbd_offset_to_zone(f, io_u->offset); |
43bcbd5b | 1670 | assert(z->has_wp); |
d9ed3e63 | 1671 | |
bfbdd35b BVA |
1672 | if (!success) |
1673 | goto unlock; | |
d9ed3e63 DLM |
1674 | |
1675 | dprint(FD_ZBD, | |
1676 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
53aa6171 | 1677 | f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z)); |
d9ed3e63 | 1678 | |
bfbdd35b BVA |
1679 | switch (io_u->ddir) { |
1680 | case DDIR_WRITE: | |
d9ed3e63 | 1681 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
236d23a8 | 1682 | zbd_zone_capacity_end(z)); |
139d8dc6 | 1683 | |
a7c2b6fc BVA |
1684 | /* |
1685 | * z->wp > zone_end means that one or more I/O errors | |
1686 | * have occurred. | |
1687 | */ | |
2fb29f27 SK |
1688 | if (accounting_vdb(td, f) && z->wp <= zone_end) { |
1689 | pthread_mutex_lock(&zbd_info->mutex); | |
d56a6df3 | 1690 | zbd_info->wp_valid_data_bytes += zone_end - z->wp; |
2fb29f27 SK |
1691 | pthread_mutex_unlock(&zbd_info->mutex); |
1692 | } | |
bfbdd35b BVA |
1693 | z->wp = zone_end; |
1694 | break; | |
bfbdd35b BVA |
1695 | default: |
1696 | break; | |
1697 | } | |
d9ed3e63 | 1698 | |
b2da58c4 SK |
1699 | if (q == FIO_Q_COMPLETED && !io_u->error) |
1700 | zbd_end_zone_io(td, io_u, z); | |
1701 | ||
bfbdd35b | 1702 | unlock: |
d9ed3e63 DLM |
1703 | if (!success || q != FIO_Q_QUEUED) { |
1704 | /* BUSY or COMPLETED: unlock the zone */ | |
4d4c71e6 | 1705 | zone_unlock(z); |
d9ed3e63 DLM |
1706 | io_u->zbd_put_io = NULL; |
1707 | } | |
1708 | } | |
1709 | ||
1710 | /** | |
1711 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1712 | * @io_u: I/O unit | |
1713 | */ | |
b2da58c4 | 1714 | static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) |
d9ed3e63 DLM |
1715 | { |
1716 | const struct fio_file *f = io_u->file; | |
1717 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1718 | struct fio_zone_info *z; | |
d9ed3e63 | 1719 | |
5ddf46d0 | 1720 | assert(zbd_info); |
615555bb | 1721 | |
53aa6171 | 1722 | z = zbd_offset_to_zone(f, io_u->offset); |
43bcbd5b | 1723 | assert(z->has_wp); |
d9ed3e63 DLM |
1724 | |
1725 | dprint(FD_ZBD, | |
1726 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
53aa6171 | 1727 | f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z)); |
d9ed3e63 | 1728 | |
b2da58c4 SK |
1729 | zbd_end_zone_io(td, io_u, z); |
1730 | ||
4d4c71e6 | 1731 | zone_unlock(z); |
bfbdd35b BVA |
1732 | } |
1733 | ||
9d87c646 DLM |
1734 | /* |
1735 | * Windows and MacOS do not define this. | |
1736 | */ | |
1737 | #ifndef EREMOTEIO | |
1738 | #define EREMOTEIO 121 /* POSIX value */ | |
1739 | #endif | |
1740 | ||
bfbdd35b BVA |
1741 | bool zbd_unaligned_write(int error_code) |
1742 | { | |
1743 | switch (error_code) { | |
1744 | case EIO: | |
1745 | case EREMOTEIO: | |
1746 | return true; | |
1747 | } | |
1748 | return false; | |
1749 | } | |
1750 | ||
4d37720a DLM |
1751 | /** |
1752 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1753 | * @td: FIO thread data. | |
1754 | * @io_u: FIO I/O unit. | |
1755 | * | |
1756 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1757 | * no more IO can be performed in the current zone. | |
1758 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1759 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1760 | * - For write workloads, zoneskip is applied when the zone is full. | |
1761 | * This applies only to read and write operations. | |
1762 | */ | |
1763 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1764 | { | |
1765 | struct fio_file *f = io_u->file; | |
1766 | enum fio_ddir ddir = io_u->ddir; | |
1767 | struct fio_zone_info *z; | |
4d37720a DLM |
1768 | |
1769 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1770 | assert(td->o.zone_size); | |
5ddf46d0 | 1771 | assert(f->zbd_info); |
4d37720a | 1772 | |
53aa6171 | 1773 | z = zbd_offset_to_zone(f, f->last_pos[ddir]); |
236d23a8 SK |
1774 | |
1775 | /* | |
1776 | * When the zone capacity is smaller than the zone size and the I/O is | |
1777 | * sequential write, skip to zone end if the latest position is at the | |
1778 | * zone capacity limit. | |
1779 | */ | |
139d8dc6 DLM |
1780 | if (z->capacity < f->zbd_info->zone_size && |
1781 | !td_random(td) && ddir == DDIR_WRITE && | |
236d23a8 SK |
1782 | f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { |
1783 | dprint(FD_ZBD, | |
1784 | "%s: Jump from zone capacity limit to zone end:" | |
ee5e3436 SK |
1785 | " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n", |
1786 | f->file_name, f->last_pos[ddir], | |
53aa6171 | 1787 | zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity); |
236d23a8 SK |
1788 | td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; |
1789 | f->last_pos[ddir] = zbd_zone_end(z); | |
1790 | } | |
1791 | ||
4d37720a DLM |
1792 | /* |
1793 | * zone_skip is valid only for sequential workloads. | |
1794 | */ | |
1795 | if (td_random(td) || !td->o.zone_skip) | |
1796 | return; | |
1797 | ||
1798 | /* | |
1799 | * It is time to switch to a new zone if: | |
1800 | * - zone_bytes == zone_size bytes have already been accessed | |
1801 | * - The last position reached the end of the current zone. | |
1802 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1803 | * reached the zone write pointer. | |
1804 | */ | |
4d37720a | 1805 | if (td->zone_bytes >= td->o.zone_size || |
236d23a8 | 1806 | f->last_pos[ddir] >= zbd_zone_end(z) || |
4d37720a DLM |
1807 | (ddir == DDIR_READ && |
1808 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1809 | /* | |
1810 | * Skip zones. | |
1811 | */ | |
1812 | td->zone_bytes = 0; | |
1813 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1814 | ||
1815 | /* | |
1816 | * Wrap from the beginning, if we exceed the file size | |
1817 | */ | |
1818 | if (f->file_offset >= f->real_file_size) | |
1819 | f->file_offset = get_start_offset(td, f); | |
1820 | ||
1821 | f->last_pos[ddir] = f->file_offset; | |
1822 | td->io_skip_bytes += td->o.zone_skip; | |
1823 | } | |
1824 | } | |
1825 | ||
c65057f9 | 1826 | /** |
c7d5e152 | 1827 | * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. |
c65057f9 SK |
1828 | * |
1829 | * @td: FIO thread data. | |
1830 | * @io_u: FIO I/O unit. | |
1831 | * @ddir: I/O direction before adjustment. | |
1832 | * | |
1833 | * Return adjusted I/O direction. | |
1834 | */ | |
1835 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1836 | enum fio_ddir ddir) | |
1837 | { | |
1838 | /* | |
1839 | * In case read direction is chosen for the first random I/O, fio with | |
1840 | * zonemode=zbd stops because no data can be read from zoned block | |
1841 | * devices with all empty zones. Overwrite the first I/O direction as | |
1842 | * write to make sure data to read exists. | |
1843 | */ | |
5ddf46d0 | 1844 | assert(io_u->file->zbd_info); |
731461cc | 1845 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1846 | return ddir; |
1847 | ||
440f63b1 | 1848 | if (io_u->file->last_start[DDIR_WRITE] != -1ULL || td->o.read_beyond_wp) |
c65057f9 SK |
1849 | return DDIR_READ; |
1850 | ||
1851 | return DDIR_WRITE; | |
1852 | } | |
1853 | ||
bfbdd35b BVA |
1854 | /** |
1855 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1856 | * @td: FIO thread data. | |
1857 | * @io_u: FIO I/O unit. | |
1858 | * | |
1859 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1860 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1861 | * corresponds to io_u->offset at the end of this function. | |
1862 | */ | |
1863 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1864 | { | |
b7694961 | 1865 | struct fio_file *f = io_u->file; |
af94a8c3 | 1866 | struct zoned_block_device_info *zbdi = f->zbd_info; |
de65f7b7 | 1867 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b | 1868 | uint32_t orig_len = io_u->buflen; |
07fc3f57 | 1869 | uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
bfbdd35b BVA |
1870 | uint64_t new_len; |
1871 | int64_t range; | |
1872 | ||
af94a8c3 | 1873 | assert(zbdi); |
adc6adcb | 1874 | assert(min_bs); |
bfbdd35b BVA |
1875 | assert(is_valid_offset(f, io_u->offset)); |
1876 | assert(io_u->buflen); | |
139d8dc6 | 1877 | |
53aa6171 | 1878 | zb = zbd_offset_to_zone(f, io_u->offset); |
de65f7b7 | 1879 | orig_zb = zb; |
bfbdd35b | 1880 | |
2efcf74b SK |
1881 | if (!zb->has_wp) { |
1882 | /* Accept non-write I/Os for conventional zones. */ | |
1883 | if (io_u->ddir != DDIR_WRITE) | |
1884 | return io_u_accept; | |
139d8dc6 | 1885 | |
2efcf74b SK |
1886 | /* |
1887 | * Make sure that writes to conventional zones | |
1888 | * don't cross over to any sequential zones. | |
1889 | */ | |
1890 | if (!(zb + 1)->has_wp || | |
1891 | io_u->offset + io_u->buflen <= (zb + 1)->start) | |
1892 | return io_u_accept; | |
1893 | ||
1894 | if (io_u->offset + min_bs > (zb + 1)->start) { | |
1895 | dprint(FD_IO, | |
07fc3f57 | 1896 | "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n", |
1e3d6e03 | 1897 | f->file_name, io_u->offset, |
ee5e3436 | 1898 | min_bs, (zb + 1)->start); |
139d8dc6 DLM |
1899 | io_u->offset = |
1900 | zb->start + (zb + 1)->start - io_u->offset; | |
1901 | new_len = min(io_u->buflen, | |
1902 | (zb + 1)->start - io_u->offset); | |
2efcf74b SK |
1903 | } else { |
1904 | new_len = (zb + 1)->start - io_u->offset; | |
1905 | } | |
139d8dc6 | 1906 | |
2efcf74b | 1907 | io_u->buflen = new_len / min_bs * min_bs; |
139d8dc6 | 1908 | |
bfbdd35b | 1909 | return io_u_accept; |
2efcf74b | 1910 | } |
bfbdd35b BVA |
1911 | |
1912 | /* | |
1913 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1914 | * is enabled. | |
1915 | */ | |
b7694961 | 1916 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1917 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1918 | return io_u_accept; | |
1919 | ||
fae3b9a0 | 1920 | zone_lock(td, f, zb); |
6f0c6085 | 1921 | |
bfbdd35b BVA |
1922 | switch (io_u->ddir) { |
1923 | case DDIR_READ: | |
6e2da06a | 1924 | if (td->runstate == TD_VERIFYING && td_write(td)) |
bfbdd35b | 1925 | goto accept; |
139d8dc6 | 1926 | |
bfbdd35b | 1927 | /* |
de65f7b7 DLM |
1928 | * Check that there is enough written data in the zone to do an |
1929 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1930 | * the I/O. | |
bfbdd35b | 1931 | */ |
b7694961 | 1932 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1933 | zb->wp - zb->start : 0; |
de65f7b7 | 1934 | if (range < min_bs || |
ee3696bd | 1935 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
4d4c71e6 | 1936 | zone_unlock(zb); |
39e06ee7 | 1937 | zl = zbd_get_zone(f, f->max_zone); |
5c86fdf6 | 1938 | zb = zbd_find_zone(td, io_u, min_bs, zb, zl); |
bfbdd35b BVA |
1939 | if (!zb) { |
1940 | dprint(FD_ZBD, | |
1941 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1942 | f->file_name, io_u->offset, | |
1943 | io_u->buflen); | |
1944 | goto eof; | |
1945 | } | |
de65f7b7 DLM |
1946 | /* |
1947 | * zbd_find_zone() returned a zone with a range of at | |
1948 | * least min_bs. | |
1949 | */ | |
ee3696bd | 1950 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1951 | assert(range >= min_bs); |
1952 | ||
1953 | if (!td_random(td)) | |
ee3696bd | 1954 | io_u->offset = zb->start; |
bfbdd35b | 1955 | } |
139d8dc6 | 1956 | |
de65f7b7 DLM |
1957 | /* |
1958 | * Make sure the I/O is within the zone valid data range while | |
1959 | * maximizing the I/O size and preserving randomness. | |
1960 | */ | |
1961 | if (range <= io_u->buflen) | |
ee3696bd | 1962 | io_u->offset = zb->start; |
de65f7b7 | 1963 | else if (td_random(td)) |
ee3696bd DLM |
1964 | io_u->offset = zb->start + |
1965 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 | 1966 | (range - io_u->buflen)) / min_bs * min_bs; |
139d8dc6 | 1967 | |
43bcbd5b SK |
1968 | /* |
1969 | * When zbd_find_zone() returns a conventional zone, | |
1970 | * we can simply accept the new i/o offset here. | |
1971 | */ | |
1972 | if (!zb->has_wp) | |
1973 | return io_u_accept; | |
139d8dc6 | 1974 | |
de65f7b7 DLM |
1975 | /* |
1976 | * Make sure the I/O does not cross over the zone wp position. | |
1977 | */ | |
1978 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1979 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1980 | new_len = new_len / min_bs * min_bs; |
1981 | if (new_len < io_u->buflen) { | |
1982 | io_u->buflen = new_len; | |
1983 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1984 | orig_len, io_u->buflen); | |
bfbdd35b | 1985 | } |
139d8dc6 | 1986 | |
ee3696bd DLM |
1987 | assert(zb->start <= io_u->offset); |
1988 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
139d8dc6 | 1989 | |
bfbdd35b | 1990 | goto accept; |
139d8dc6 | 1991 | |
bfbdd35b | 1992 | case DDIR_WRITE: |
af94a8c3 | 1993 | if (io_u->buflen > zbdi->zone_size) { |
1c74aadc DF |
1994 | td_verror(td, EINVAL, "I/O buflen exceeds zone size"); |
1995 | dprint(FD_IO, | |
ee5e3436 SK |
1996 | "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n", |
1997 | f->file_name, io_u->buflen, zbdi->zone_size); | |
bfbdd35b | 1998 | goto eof; |
1c74aadc | 1999 | } |
139d8dc6 | 2000 | |
e1a1b59b SK |
2001 | retry: |
2002 | if (zbd_zone_remainder(zb) > 0 && | |
2003 | zbd_zone_remainder(zb) < min_bs) { | |
2004 | pthread_mutex_lock(&f->zbd_info->mutex); | |
a4807046 | 2005 | zbd_write_zone_put(td, f, zb); |
e1a1b59b SK |
2006 | pthread_mutex_unlock(&f->zbd_info->mutex); |
2007 | dprint(FD_ZBD, | |
2008 | "%s: finish zone %d\n", | |
2009 | f->file_name, zbd_zone_idx(f, zb)); | |
2010 | io_u_quiesce(td); | |
2011 | zbd_finish_zone(td, f, zb); | |
2012 | if (zbd_zone_idx(f, zb) + 1 >= f->max_zone) { | |
2013 | if (!td_random(td)) | |
2014 | goto eof; | |
2015 | } | |
2016 | zone_unlock(zb); | |
2017 | ||
2018 | /* Find the next write pointer zone */ | |
2019 | do { | |
2020 | zb++; | |
2021 | if (zbd_zone_idx(f, zb) >= f->max_zone) | |
2022 | zb = zbd_get_zone(f, f->min_zone); | |
2023 | } while (!zb->has_wp); | |
2024 | ||
2025 | zone_lock(td, f, zb); | |
2026 | } | |
2027 | ||
a4807046 | 2028 | if (!zbd_write_zone_get(td, f, zb)) { |
4d4c71e6 | 2029 | zone_unlock(zb); |
a4807046 | 2030 | zb = zbd_convert_to_write_zone(td, io_u); |
1c74aadc | 2031 | if (!zb) { |
a4807046 | 2032 | dprint(FD_IO, "%s: can't convert to write target zone", |
1c74aadc | 2033 | f->file_name); |
59b07544 | 2034 | goto eof; |
1c74aadc | 2035 | } |
59b07544 | 2036 | } |
139d8dc6 | 2037 | |
e1a1b59b SK |
2038 | if (zbd_zone_remainder(zb) > 0 && |
2039 | zbd_zone_remainder(zb) < min_bs) | |
2040 | goto retry; | |
2041 | ||
a7c2b6fc BVA |
2042 | /* Check whether the zone reset threshold has been exceeded */ |
2043 | if (td->o.zrf.u.f) { | |
d56a6df3 SK |
2044 | if (zbdi->wp_valid_data_bytes >= |
2045 | f->io_size * td->o.zrt.u.f && | |
139d8dc6 | 2046 | zbd_dec_and_reset_write_cnt(td, f)) |
a7c2b6fc | 2047 | zb->reset_zone = 1; |
a7c2b6fc | 2048 | } |
139d8dc6 | 2049 | |
bfbdd35b BVA |
2050 | /* Reset the zone pointer if necessary */ |
2051 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
cef8006c SK |
2052 | if (td->o.verify != VERIFY_NONE) { |
2053 | /* | |
2054 | * Unset io-u->file to tell get_next_verify() | |
2055 | * that this IO is not requeue. | |
2056 | */ | |
2057 | io_u->file = NULL; | |
2058 | if (!get_next_verify(td, io_u)) { | |
2059 | zone_unlock(zb); | |
2060 | return io_u_accept; | |
2061 | } | |
2062 | io_u->file = f; | |
2063 | } | |
2064 | ||
bfbdd35b BVA |
2065 | /* |
2066 | * Since previous write requests may have been submitted | |
2067 | * asynchronously and since we will submit the zone | |
2068 | * reset synchronously, wait until previously submitted | |
2069 | * write requests have completed before issuing a | |
2070 | * zone reset. | |
2071 | */ | |
2072 | io_u_quiesce(td); | |
2073 | zb->reset_zone = 0; | |
67282020 | 2074 | if (__zbd_reset_zone(td, f, zb) < 0) |
bfbdd35b | 2075 | goto eof; |
236d23a8 SK |
2076 | |
2077 | if (zb->capacity < min_bs) { | |
1c74aadc | 2078 | td_verror(td, EINVAL, "ZCAP is less min_bs"); |
07fc3f57 | 2079 | log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n", |
ee5e3436 | 2080 | zb->capacity, min_bs); |
236d23a8 SK |
2081 | goto eof; |
2082 | } | |
bfbdd35b | 2083 | } |
139d8dc6 | 2084 | |
bfbdd35b BVA |
2085 | /* Make writes occur at the write pointer */ |
2086 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 2087 | io_u->offset = zb->wp; |
bfbdd35b | 2088 | if (!is_valid_offset(f, io_u->offset)) { |
1c74aadc DF |
2089 | td_verror(td, EINVAL, "invalid WP value"); |
2090 | dprint(FD_ZBD, "%s: dropped request with offset %llu\n", | |
2091 | f->file_name, io_u->offset); | |
bfbdd35b BVA |
2092 | goto eof; |
2093 | } | |
139d8dc6 | 2094 | |
bfbdd35b BVA |
2095 | /* |
2096 | * Make sure that the buflen is a multiple of the minimal | |
2097 | * block size. Give up if shrinking would make the request too | |
2098 | * small. | |
2099 | */ | |
2100 | new_len = min((unsigned long long)io_u->buflen, | |
236d23a8 | 2101 | zbd_zone_capacity_end(zb) - io_u->offset); |
bfbdd35b BVA |
2102 | new_len = new_len / min_bs * min_bs; |
2103 | if (new_len == io_u->buflen) | |
2104 | goto accept; | |
2105 | if (new_len >= min_bs) { | |
2106 | io_u->buflen = new_len; | |
2107 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
2108 | orig_len, io_u->buflen); | |
2109 | goto accept; | |
2110 | } | |
139d8dc6 | 2111 | |
1c74aadc | 2112 | td_verror(td, EIO, "zone remainder too small"); |
07fc3f57 | 2113 | log_err("zone remainder %lld smaller than min block size %"PRIu64"\n", |
1c74aadc | 2114 | (zbd_zone_capacity_end(zb) - io_u->offset), min_bs); |
139d8dc6 | 2115 | |
bfbdd35b | 2116 | goto eof; |
139d8dc6 | 2117 | |
bfbdd35b | 2118 | case DDIR_TRIM: |
e3be810b SK |
2119 | /* Check random trim targets a non-empty zone */ |
2120 | if (!td_random(td) || zb->wp > zb->start) | |
2121 | goto accept; | |
2122 | ||
2123 | /* Find out a non-empty zone to trim */ | |
2124 | zone_unlock(zb); | |
39e06ee7 | 2125 | zl = zbd_get_zone(f, f->max_zone); |
e3be810b SK |
2126 | zb = zbd_find_zone(td, io_u, 1, zb, zl); |
2127 | if (zb) { | |
2128 | io_u->offset = zb->start; | |
2129 | dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n", | |
2130 | f->file_name, io_u->offset); | |
2131 | goto accept; | |
2132 | } | |
139d8dc6 | 2133 | |
e3be810b | 2134 | goto eof; |
139d8dc6 | 2135 | |
bfbdd35b | 2136 | case DDIR_SYNC: |
e3be810b | 2137 | /* fall-through */ |
bfbdd35b BVA |
2138 | case DDIR_DATASYNC: |
2139 | case DDIR_SYNC_FILE_RANGE: | |
2140 | case DDIR_WAIT: | |
2141 | case DDIR_LAST: | |
2142 | case DDIR_INVAL: | |
2143 | goto accept; | |
2144 | } | |
2145 | ||
2146 | assert(false); | |
2147 | ||
2148 | accept: | |
43bcbd5b | 2149 | assert(zb->has_wp); |
b7694961 | 2150 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
2151 | assert(!io_u->zbd_queue_io); |
2152 | assert(!io_u->zbd_put_io); | |
139d8dc6 | 2153 | |
d9ed3e63 DLM |
2154 | io_u->zbd_queue_io = zbd_queue_io; |
2155 | io_u->zbd_put_io = zbd_put_io; | |
139d8dc6 | 2156 | |
2ef3c1b0 DF |
2157 | /* |
2158 | * Since we return with the zone lock still held, | |
2159 | * add an annotation to let Coverity know that it | |
2160 | * is intentional. | |
2161 | */ | |
2162 | /* coverity[missing_unlock] */ | |
139d8dc6 | 2163 | |
bfbdd35b BVA |
2164 | return io_u_accept; |
2165 | ||
2166 | eof: | |
43bcbd5b | 2167 | if (zb && zb->has_wp) |
4d4c71e6 | 2168 | zone_unlock(zb); |
139d8dc6 | 2169 | |
bfbdd35b BVA |
2170 | return io_u_eof; |
2171 | } | |
fd5d733f BVA |
2172 | |
2173 | /* Return a string with ZBD statistics */ | |
2174 | char *zbd_write_status(const struct thread_stat *ts) | |
2175 | { | |
2176 | char *res; | |
2177 | ||
ee5e3436 | 2178 | if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0) |
fd5d733f BVA |
2179 | return NULL; |
2180 | return res; | |
2181 | } | |
e3be810b SK |
2182 | |
2183 | /** | |
2184 | * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim | |
2185 | * | |
2186 | * @td: FIO thread data. | |
2187 | * @io_u: FIO I/O unit. | |
2188 | * | |
2189 | * It is assumed that z->mutex is already locked. | |
2190 | * Return io_u_completed when reset zone succeeds. Return 0 when the target zone | |
2191 | * does not have write pointer. On error, return negative errno. | |
2192 | */ | |
67282020 | 2193 | int zbd_do_io_u_trim(struct thread_data *td, struct io_u *io_u) |
e3be810b SK |
2194 | { |
2195 | struct fio_file *f = io_u->file; | |
2196 | struct fio_zone_info *z; | |
e3be810b SK |
2197 | int ret; |
2198 | ||
53aa6171 | 2199 | z = zbd_offset_to_zone(f, io_u->offset); |
e3be810b SK |
2200 | if (!z->has_wp) |
2201 | return 0; | |
2202 | ||
2203 | if (io_u->offset != z->start) { | |
139d8dc6 DLM |
2204 | log_err("Trim offset not at zone start (%lld)\n", |
2205 | io_u->offset); | |
e3be810b SK |
2206 | return -EINVAL; |
2207 | } | |
2208 | ||
2209 | ret = zbd_reset_zone((struct thread_data *)td, f, z); | |
2210 | if (ret < 0) | |
2211 | return ret; | |
2212 | ||
2213 | return io_u_completed; | |
2214 | } |