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 | ||
73 | /** | |
74 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
75 | * @f: file pointer. | |
76 | * @z: zone info pointer. | |
77 | * @required: minimum number of bytes that must remain in a zone. | |
78 | * | |
79 | * The caller must hold z->mutex. | |
80 | */ | |
81 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
82 | uint64_t required) | |
83 | { | |
84 | assert((required & 511) == 0); | |
85 | ||
86 | return z->has_wp && | |
87 | z->wp + required > zbd_zone_capacity_end(z); | |
88 | } | |
89 | ||
90 | static void zone_lock(struct thread_data *td, const struct fio_file *f, | |
91 | struct fio_zone_info *z) | |
92 | { | |
93 | struct zoned_block_device_info *zbd = f->zbd_info; | |
94 | uint32_t nz = z - zbd->zone_info; | |
95 | ||
96 | /* A thread should never lock zones outside its working area. */ | |
97 | assert(f->min_zone <= nz && nz < f->max_zone); | |
98 | ||
99 | assert(z->has_wp); | |
100 | ||
101 | /* | |
102 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
103 | * is changed or when io_u completes and zbd_put_io() executed. | |
104 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
105 | * other waiting for zone locks when building an io_u batch, first | |
106 | * only trylock the zone. If the zone is already locked by another job, | |
107 | * process the currently queued I/Os so that I/O progress is made and | |
108 | * zones unlocked. | |
109 | */ | |
110 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
111 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
112 | io_u_quiesce(td); | |
113 | pthread_mutex_lock(&z->mutex); | |
114 | } | |
115 | } | |
116 | ||
117 | static inline void zone_unlock(struct fio_zone_info *z) | |
118 | { | |
119 | int ret; | |
120 | ||
121 | assert(z->has_wp); | |
122 | ret = pthread_mutex_unlock(&z->mutex); | |
123 | assert(!ret); | |
124 | } | |
125 | ||
126 | static inline struct fio_zone_info *get_zone(const struct fio_file *f, | |
127 | unsigned int zone_nr) | |
128 | { | |
129 | return &f->zbd_info->zone_info[zone_nr]; | |
130 | } | |
131 | ||
b7694961 DLM |
132 | /** |
133 | * zbd_get_zoned_model - Get a device zoned model | |
134 | * @td: FIO thread data | |
135 | * @f: FIO file for which to get model information | |
136 | */ | |
38334c13 DLM |
137 | static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, |
138 | enum zbd_zoned_model *model) | |
b7694961 DLM |
139 | { |
140 | int ret; | |
141 | ||
50cc48d5 NC |
142 | if (f->filetype == FIO_TYPE_PIPE) { |
143 | log_err("zonemode=zbd does not support pipes\n"); | |
144 | return -EINVAL; | |
145 | } | |
146 | ||
9db0cde8 NC |
147 | /* If regular file, always emulate zones inside the file. */ |
148 | if (f->filetype == FIO_TYPE_FILE) { | |
149 | *model = ZBD_NONE; | |
150 | return 0; | |
151 | } | |
152 | ||
6c5b11d3 DLM |
153 | if (td->io_ops && td->io_ops->get_zoned_model) |
154 | ret = td->io_ops->get_zoned_model(td, f, model); | |
155 | else | |
156 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
157 | if (ret < 0) { |
158 | td_verror(td, errno, "get zoned model failed"); | |
159 | log_err("%s: get zoned model failed (%d).\n", | |
160 | f->file_name, errno); | |
161 | } | |
162 | ||
163 | return ret; | |
164 | } | |
165 | ||
166 | /** | |
167 | * zbd_report_zones - Get zone information | |
168 | * @td: FIO thread data. | |
169 | * @f: FIO file for which to get zone information | |
170 | * @offset: offset from which to report zones | |
171 | * @zones: Array of struct zbd_zone | |
172 | * @nr_zones: Size of @zones array | |
173 | * | |
174 | * Get zone information into @zones starting from the zone at offset @offset | |
175 | * for the device specified by @f. | |
176 | * | |
177 | * Returns the number of zones reported upon success and a negative error code | |
178 | * upon failure. If the zone report is empty, always assume an error (device | |
179 | * problem) and return -EIO. | |
180 | */ | |
38334c13 DLM |
181 | static int zbd_report_zones(struct thread_data *td, struct fio_file *f, |
182 | uint64_t offset, struct zbd_zone *zones, | |
183 | unsigned int nr_zones) | |
b7694961 DLM |
184 | { |
185 | int ret; | |
186 | ||
6c5b11d3 DLM |
187 | if (td->io_ops && td->io_ops->report_zones) |
188 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
189 | else | |
190 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
191 | if (ret < 0) { |
192 | td_verror(td, errno, "report zones failed"); | |
ee5e3436 SK |
193 | log_err("%s: report zones from sector %"PRIu64" failed (%d).\n", |
194 | f->file_name, offset >> 9, errno); | |
b7694961 DLM |
195 | } else if (ret == 0) { |
196 | td_verror(td, errno, "Empty zone report"); | |
ee5e3436 SK |
197 | log_err("%s: report zones from sector %"PRIu64" is empty.\n", |
198 | f->file_name, offset >> 9); | |
b7694961 DLM |
199 | ret = -EIO; |
200 | } | |
201 | ||
202 | return ret; | |
203 | } | |
204 | ||
205 | /** | |
206 | * zbd_reset_wp - reset the write pointer of a range of zones | |
207 | * @td: FIO thread data. | |
208 | * @f: FIO file for which to reset zones | |
209 | * @offset: Starting offset of the first zone to reset | |
210 | * @length: Length of the range of zones to reset | |
211 | * | |
212 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
213 | * Returns 0 upon success and a negative error code upon failure. | |
214 | */ | |
38334c13 DLM |
215 | static int zbd_reset_wp(struct thread_data *td, struct fio_file *f, |
216 | uint64_t offset, uint64_t length) | |
b7694961 DLM |
217 | { |
218 | int ret; | |
219 | ||
6c5b11d3 DLM |
220 | if (td->io_ops && td->io_ops->reset_wp) |
221 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
222 | else | |
223 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
224 | if (ret < 0) { |
225 | td_verror(td, errno, "resetting wp failed"); | |
ee5e3436 SK |
226 | log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n", |
227 | f->file_name, length >> 9, offset >> 9, errno); | |
b7694961 DLM |
228 | } |
229 | ||
230 | return ret; | |
231 | } | |
232 | ||
410a071c DLM |
233 | /** |
234 | * zbd_reset_zone - reset the write pointer of a single zone | |
235 | * @td: FIO thread data. | |
236 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
237 | * @z: Zone to reset. | |
238 | * | |
239 | * Returns 0 upon success and a negative error code upon failure. | |
240 | * | |
241 | * The caller must hold z->mutex. | |
242 | */ | |
243 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, | |
244 | struct fio_zone_info *z) | |
245 | { | |
246 | uint64_t offset = z->start; | |
247 | uint64_t length = (z+1)->start - offset; | |
248 | uint64_t data_in_zone = z->wp - z->start; | |
249 | int ret = 0; | |
250 | ||
251 | if (!data_in_zone) | |
252 | return 0; | |
253 | ||
254 | assert(is_valid_offset(f, offset + length - 1)); | |
255 | ||
139d8dc6 | 256 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", |
dc8a3d62 | 257 | f->file_name, zbd_zone_idx(f, z)); |
139d8dc6 | 258 | |
410a071c DLM |
259 | switch (f->zbd_info->model) { |
260 | case ZBD_HOST_AWARE: | |
261 | case ZBD_HOST_MANAGED: | |
262 | ret = zbd_reset_wp(td, f, offset, length); | |
263 | if (ret < 0) | |
264 | return ret; | |
265 | break; | |
266 | default: | |
267 | break; | |
268 | } | |
269 | ||
270 | pthread_mutex_lock(&f->zbd_info->mutex); | |
271 | f->zbd_info->sectors_with_data -= data_in_zone; | |
272 | f->zbd_info->wp_sectors_with_data -= data_in_zone; | |
273 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
139d8dc6 | 274 | |
410a071c DLM |
275 | z->wp = z->start; |
276 | z->verify_block = 0; | |
277 | ||
278 | td->ts.nr_zone_resets++; | |
279 | ||
280 | return ret; | |
281 | } | |
282 | ||
283 | /** | |
284 | * zbd_close_zone - Remove a zone from the open zones array. | |
285 | * @td: FIO thread data. | |
286 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
287 | * @zone_idx: Index of the zone to remove. | |
288 | * | |
289 | * The caller must hold f->zbd_info->mutex. | |
290 | */ | |
291 | static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, | |
a23411bb | 292 | struct fio_zone_info *z) |
410a071c | 293 | { |
a23411bb | 294 | uint32_t ozi; |
410a071c | 295 | |
a23411bb DLM |
296 | if (!z->open) |
297 | return; | |
298 | ||
299 | for (ozi = 0; ozi < f->zbd_info->num_open_zones; ozi++) { | |
300 | if (get_zone(f, f->zbd_info->open_zones[ozi]) == z) | |
410a071c DLM |
301 | break; |
302 | } | |
a23411bb | 303 | if (ozi == f->zbd_info->num_open_zones) |
410a071c DLM |
304 | return; |
305 | ||
a23411bb | 306 | dprint(FD_ZBD, "%s: closing zone %u\n", |
dc8a3d62 | 307 | f->file_name, zbd_zone_idx(f, z)); |
139d8dc6 | 308 | |
a23411bb DLM |
309 | memmove(f->zbd_info->open_zones + ozi, |
310 | f->zbd_info->open_zones + ozi + 1, | |
311 | (ZBD_MAX_OPEN_ZONES - (ozi + 1)) * | |
410a071c | 312 | sizeof(f->zbd_info->open_zones[0])); |
139d8dc6 | 313 | |
410a071c DLM |
314 | f->zbd_info->num_open_zones--; |
315 | td->num_open_zones--; | |
a23411bb | 316 | z->open = 0; |
410a071c DLM |
317 | } |
318 | ||
319 | /** | |
320 | * zbd_reset_zones - Reset a range of zones. | |
321 | * @td: fio thread data. | |
322 | * @f: fio file for which to reset zones | |
323 | * @zb: first zone to reset. | |
324 | * @ze: first zone not to reset. | |
325 | * | |
326 | * Returns 0 upon success and 1 upon failure. | |
327 | */ | |
328 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
329 | struct fio_zone_info *const zb, | |
330 | struct fio_zone_info *const ze) | |
331 | { | |
332 | struct fio_zone_info *z; | |
333 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
334 | int res = 0; | |
335 | ||
336 | assert(min_bs); | |
337 | ||
139d8dc6 | 338 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", |
dc8a3d62 | 339 | f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze)); |
139d8dc6 | 340 | |
410a071c | 341 | for (z = zb; z < ze; z++) { |
410a071c DLM |
342 | if (!z->has_wp) |
343 | continue; | |
139d8dc6 | 344 | |
410a071c DLM |
345 | zone_lock(td, f, z); |
346 | pthread_mutex_lock(&f->zbd_info->mutex); | |
a23411bb | 347 | zbd_close_zone(td, f, z); |
410a071c | 348 | pthread_mutex_unlock(&f->zbd_info->mutex); |
139d8dc6 | 349 | |
410a071c DLM |
350 | if (z->wp != z->start) { |
351 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
dc8a3d62 | 352 | f->file_name, zbd_zone_idx(f, z)); |
410a071c DLM |
353 | if (zbd_reset_zone(td, f, z) < 0) |
354 | res = 1; | |
355 | } | |
139d8dc6 | 356 | |
410a071c DLM |
357 | zone_unlock(z); |
358 | } | |
359 | ||
360 | return res; | |
361 | } | |
362 | ||
d2f442bc NC |
363 | /** |
364 | * zbd_get_max_open_zones - Get the maximum number of open zones | |
365 | * @td: FIO thread data | |
366 | * @f: FIO file for which to get max open zones | |
367 | * @max_open_zones: Upon success, result will be stored here. | |
368 | * | |
369 | * A @max_open_zones value set to zero means no limit. | |
370 | * | |
371 | * Returns 0 upon success and a negative error code upon failure. | |
372 | */ | |
38334c13 DLM |
373 | static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f, |
374 | unsigned int *max_open_zones) | |
d2f442bc NC |
375 | { |
376 | int ret; | |
377 | ||
378 | if (td->io_ops && td->io_ops->get_max_open_zones) | |
379 | ret = td->io_ops->get_max_open_zones(td, f, max_open_zones); | |
380 | else | |
381 | ret = blkzoned_get_max_open_zones(td, f, max_open_zones); | |
382 | if (ret < 0) { | |
383 | td_verror(td, errno, "get max open zones failed"); | |
384 | log_err("%s: get max open zones failed (%d).\n", | |
385 | f->file_name, errno); | |
386 | } | |
387 | ||
388 | return ret; | |
389 | } | |
390 | ||
bfbdd35b | 391 | /** |
410a071c DLM |
392 | * zbd_open_zone - Add a zone to the array of open zones. |
393 | * @td: fio thread data. | |
394 | * @f: fio file that has the open zones to add. | |
395 | * @zone_idx: Index of the zone to add. | |
bfbdd35b | 396 | * |
410a071c DLM |
397 | * Open a ZBD zone if it is not already open. Returns true if either the zone |
398 | * was already open or if the zone was successfully added to the array of open | |
399 | * zones without exceeding the maximum number of open zones. Returns false if | |
400 | * the zone was not already open and opening the zone would cause the zone limit | |
401 | * to be exceeded. | |
bfbdd35b | 402 | */ |
410a071c | 403 | static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f, |
aad7c276 | 404 | struct fio_zone_info *z) |
1f57803b | 405 | { |
410a071c DLM |
406 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; |
407 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
dc8a3d62 | 408 | uint32_t zone_idx = zbd_zone_idx(f, z); |
410a071c | 409 | bool res = true; |
fae3b9a0 | 410 | |
410a071c DLM |
411 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
412 | return false; | |
43bcbd5b | 413 | |
1f57803b | 414 | /* |
410a071c DLM |
415 | * Skip full zones with data verification enabled because resetting a |
416 | * zone causes data loss and hence causes verification to fail. | |
1f57803b | 417 | */ |
410a071c DLM |
418 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) |
419 | return false; | |
4d4c71e6 | 420 | |
410a071c DLM |
421 | /* |
422 | * zbdi->max_open_zones == 0 means that there is no limit on the maximum | |
423 | * number of open zones. In this case, do no track open zones in | |
424 | * zbdi->open_zones array. | |
425 | */ | |
426 | if (!zbdi->max_open_zones) | |
427 | return true; | |
4d4c71e6 | 428 | |
410a071c | 429 | pthread_mutex_lock(&zbdi->mutex); |
b5a0f7ce DLM |
430 | |
431 | if (z->open) { | |
410a071c | 432 | /* |
b5a0f7ce DLM |
433 | * If the zone is going to be completely filled by writes |
434 | * already in-flight, handle it as a full zone instead of an | |
435 | * open zone. | |
410a071c DLM |
436 | */ |
437 | if (z->wp >= zbd_zone_capacity_end(z)) | |
438 | res = false; | |
439 | goto out; | |
440 | } | |
139d8dc6 | 441 | |
410a071c DLM |
442 | res = false; |
443 | /* Zero means no limit */ | |
444 | if (td->o.job_max_open_zones > 0 && | |
445 | td->num_open_zones >= td->o.job_max_open_zones) | |
446 | goto out; | |
447 | if (zbdi->num_open_zones >= zbdi->max_open_zones) | |
448 | goto out; | |
139d8dc6 | 449 | |
aad7c276 | 450 | dprint(FD_ZBD, "%s: opening zone %u\n", |
139d8dc6 DLM |
451 | f->file_name, zone_idx); |
452 | ||
410a071c DLM |
453 | zbdi->open_zones[zbdi->num_open_zones++] = zone_idx; |
454 | td->num_open_zones++; | |
455 | z->open = 1; | |
456 | res = true; | |
bfbdd35b | 457 | |
410a071c DLM |
458 | out: |
459 | pthread_mutex_unlock(&zbdi->mutex); | |
460 | return res; | |
923f7c1e DF |
461 | } |
462 | ||
bfbdd35b BVA |
463 | /* Verify whether direct I/O is used for all host-managed zoned drives. */ |
464 | static bool zbd_using_direct_io(void) | |
465 | { | |
466 | struct thread_data *td; | |
467 | struct fio_file *f; | |
468 | int i, j; | |
469 | ||
470 | for_each_td(td, i) { | |
471 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) | |
472 | continue; | |
473 | for_each_file(td, f, j) { | |
474 | if (f->zbd_info && | |
b7694961 | 475 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
476 | return false; |
477 | } | |
478 | } | |
479 | ||
480 | return true; | |
481 | } | |
482 | ||
483 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
484 | static bool zbd_is_seq_job(struct fio_file *f) | |
485 | { | |
486 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
487 | ||
488 | assert(f->zbd_info); | |
139d8dc6 | 489 | |
bfbdd35b BVA |
490 | if (f->io_size == 0) |
491 | return false; | |
139d8dc6 | 492 | |
dc8a3d62 DLM |
493 | zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset); |
494 | zone_idx_e = | |
495 | zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1); | |
bfbdd35b | 496 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) |
be7a6bae | 497 | if (get_zone(f, zone_idx)->has_wp) |
bfbdd35b BVA |
498 | return true; |
499 | ||
500 | return false; | |
501 | } | |
502 | ||
0bf93a1a DLM |
503 | /* |
504 | * Verify whether the file offset and size parameters are aligned with zone | |
505 | * boundaries. If the file offset is not aligned, align it down to the start of | |
506 | * the zone containing the start offset and align up the file io_size parameter. | |
507 | */ | |
508 | static bool zbd_zone_align_file_sizes(struct thread_data *td, | |
509 | struct fio_file *f) | |
510 | { | |
511 | const struct fio_zone_info *z; | |
512 | uint64_t new_offset, new_end; | |
513 | uint32_t zone_idx; | |
514 | ||
515 | if (!f->zbd_info) | |
516 | return true; | |
517 | if (f->file_offset >= f->real_file_size) | |
518 | return true; | |
519 | if (!zbd_is_seq_job(f)) | |
520 | return true; | |
521 | ||
522 | if (!td->o.zone_size) { | |
523 | td->o.zone_size = f->zbd_info->zone_size; | |
524 | if (!td->o.zone_size) { | |
525 | log_err("%s: invalid 0 zone size\n", | |
526 | f->file_name); | |
527 | return false; | |
528 | } | |
529 | } else if (td->o.zone_size != f->zbd_info->zone_size) { | |
530 | log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n", | |
531 | f->file_name, td->o.zone_size, | |
532 | f->zbd_info->zone_size); | |
533 | return false; | |
534 | } | |
535 | ||
536 | if (td->o.zone_skip % td->o.zone_size) { | |
537 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", | |
538 | f->file_name, td->o.zone_skip, | |
539 | td->o.zone_size); | |
540 | return false; | |
541 | } | |
542 | ||
dc8a3d62 | 543 | zone_idx = zbd_offset_to_zone_idx(f, f->file_offset); |
0bf93a1a DLM |
544 | z = get_zone(f, zone_idx); |
545 | if ((f->file_offset != z->start) && | |
546 | (td->o.td_ddir != TD_DDIR_READ)) { | |
547 | new_offset = zbd_zone_end(z); | |
548 | if (new_offset >= f->file_offset + f->io_size) { | |
549 | log_info("%s: io_size must be at least one zone\n", | |
550 | f->file_name); | |
551 | return false; | |
552 | } | |
553 | log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n", | |
554 | f->file_name, f->file_offset, | |
555 | new_offset); | |
556 | f->io_size -= (new_offset - f->file_offset); | |
557 | f->file_offset = new_offset; | |
558 | } | |
559 | ||
dc8a3d62 | 560 | zone_idx = zbd_offset_to_zone_idx(f, f->file_offset + f->io_size); |
0bf93a1a DLM |
561 | z = get_zone(f, zone_idx); |
562 | new_end = z->start; | |
563 | if ((td->o.td_ddir != TD_DDIR_READ) && | |
564 | (f->file_offset + f->io_size != new_end)) { | |
565 | if (new_end <= f->file_offset) { | |
566 | log_info("%s: io_size must be at least one zone\n", | |
567 | f->file_name); | |
568 | return false; | |
569 | } | |
570 | log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n", | |
571 | f->file_name, f->io_size, | |
572 | new_end - f->file_offset); | |
573 | f->io_size = new_end - f->file_offset; | |
574 | } | |
575 | ||
576 | return true; | |
577 | } | |
578 | ||
bfbdd35b BVA |
579 | /* |
580 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
581 | */ | |
582 | static bool zbd_verify_sizes(void) | |
583 | { | |
bfbdd35b BVA |
584 | struct thread_data *td; |
585 | struct fio_file *f; | |
bfbdd35b BVA |
586 | int i, j; |
587 | ||
588 | for_each_td(td, i) { | |
589 | for_each_file(td, f, j) { | |
0bf93a1a | 590 | if (!zbd_zone_align_file_sizes(td, f)) |
4d37720a | 591 | return false; |
bfbdd35b BVA |
592 | } |
593 | } | |
594 | ||
595 | return true; | |
596 | } | |
597 | ||
598 | static bool zbd_verify_bs(void) | |
599 | { | |
600 | struct thread_data *td; | |
601 | struct fio_file *f; | |
bfbdd35b BVA |
602 | int i, j, k; |
603 | ||
604 | for_each_td(td, i) { | |
e3be810b SK |
605 | if (td_trim(td) && |
606 | (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] || | |
607 | td->o.bssplit_nr[DDIR_TRIM])) { | |
608 | log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n"); | |
609 | return false; | |
610 | } | |
bfbdd35b | 611 | for_each_file(td, f, j) { |
1ddd225e AD |
612 | uint64_t zone_size; |
613 | ||
bfbdd35b BVA |
614 | if (!f->zbd_info) |
615 | continue; | |
139d8dc6 | 616 | |
bfbdd35b | 617 | zone_size = f->zbd_info->zone_size; |
e3be810b | 618 | if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) { |
ee5e3436 | 619 | log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n", |
e3be810b | 620 | f->file_name, td->o.bs[DDIR_TRIM], |
ee5e3436 | 621 | zone_size); |
e3be810b SK |
622 | return false; |
623 | } | |
59f94d26 | 624 | for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) { |
bfbdd35b | 625 | if (td->o.verify != VERIFY_NONE && |
ee3696bd | 626 | zone_size % td->o.bs[k] != 0) { |
ee5e3436 | 627 | log_info("%s: block size %llu is not a divisor of the zone size %"PRIu64"\n", |
bfbdd35b | 628 | f->file_name, td->o.bs[k], |
ee5e3436 | 629 | zone_size); |
bfbdd35b BVA |
630 | return false; |
631 | } | |
632 | } | |
633 | } | |
634 | } | |
635 | return true; | |
636 | } | |
637 | ||
bfbdd35b BVA |
638 | static int ilog2(uint64_t i) |
639 | { | |
640 | int log = -1; | |
641 | ||
642 | while (i) { | |
643 | i >>= 1; | |
644 | log++; | |
645 | } | |
646 | return log; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
651 | * allows to execute a ZBD workload against a non-ZBD device. | |
652 | */ | |
653 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
654 | { | |
655 | uint32_t nr_zones; | |
656 | struct fio_zone_info *p; | |
a4b7f12b | 657 | uint64_t zone_size = td->o.zone_size; |
b8dd9750 | 658 | uint64_t zone_capacity = td->o.zone_capacity; |
bfbdd35b | 659 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
660 | int i; |
661 | ||
a4b7f12b | 662 | if (zone_size == 0) { |
9db0cde8 | 663 | log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n", |
a4b7f12b DLM |
664 | f->file_name); |
665 | return 1; | |
666 | } | |
667 | ||
668 | if (zone_size < 512) { | |
669 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
670 | f->file_name); | |
671 | return 1; | |
672 | } | |
673 | ||
b8dd9750 HH |
674 | if (zone_capacity == 0) |
675 | zone_capacity = zone_size; | |
676 | ||
677 | if (zone_capacity > zone_size) { | |
678 | log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", | |
ee5e3436 | 679 | f->file_name, td->o.zone_capacity, td->o.zone_size); |
b8dd9750 HH |
680 | return 1; |
681 | } | |
682 | ||
9db0cde8 NC |
683 | if (f->real_file_size < zone_size) { |
684 | log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n", | |
685 | f->file_name, f->real_file_size, zone_size); | |
686 | return -EINVAL; | |
687 | } | |
688 | ||
ee3696bd | 689 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
690 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
691 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
692 | if (!zbd_info) | |
693 | return -ENOMEM; | |
694 | ||
44ec32cb | 695 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
696 | zbd_info->refcount = 1; |
697 | p = &zbd_info->zone_info[0]; | |
698 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
699 | mutex_init_pshared_with_type(&p->mutex, |
700 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b | 701 | p->start = i * zone_size; |
b14651a2 | 702 | p->wp = p->start; |
b7694961 DLM |
703 | p->type = ZBD_ZONE_TYPE_SWR; |
704 | p->cond = ZBD_ZONE_COND_EMPTY; | |
b8dd9750 | 705 | p->capacity = zone_capacity; |
be7a6bae | 706 | p->has_wp = 1; |
bfbdd35b BVA |
707 | } |
708 | /* a sentinel */ | |
709 | p->start = nr_zones * zone_size; | |
710 | ||
711 | f->zbd_info = zbd_info; | |
712 | f->zbd_info->zone_size = zone_size; | |
713 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 714 | ilog2(zone_size) : 0; |
bfbdd35b | 715 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
716 | return 0; |
717 | } | |
718 | ||
719 | /* | |
b7694961 DLM |
720 | * Maximum number of zones to report in one operation. |
721 | */ | |
722 | #define ZBD_REPORT_MAX_ZONES 8192U | |
723 | ||
724 | /* | |
725 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
726 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
727 | */ |
728 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
729 | { | |
b7694961 DLM |
730 | int nr_zones, nrz; |
731 | struct zbd_zone *zones, *z; | |
bfbdd35b | 732 | struct fio_zone_info *p; |
b7694961 | 733 | uint64_t zone_size, offset; |
bfbdd35b | 734 | struct zoned_block_device_info *zbd_info = NULL; |
d060babc | 735 | int i, j, ret = -ENOMEM; |
bfbdd35b | 736 | |
b7694961 DLM |
737 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
738 | if (!zones) | |
bfbdd35b BVA |
739 | goto out; |
740 | ||
b7694961 DLM |
741 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
742 | if (nrz < 0) { | |
743 | ret = nrz; | |
744 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
745 | f->file_name, -ret); | |
746 | goto out; | |
bfbdd35b BVA |
747 | } |
748 | ||
b7694961 | 749 | zone_size = zones[0].len; |
ee3696bd | 750 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
751 | |
752 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
753 | td->o.zone_size = zone_size; |
754 | } else if (td->o.zone_size != zone_size) { | |
ee5e3436 SK |
755 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n", |
756 | f->file_name, td->o.zone_size, zone_size); | |
bfbdd35b | 757 | ret = -EINVAL; |
b7694961 | 758 | goto out; |
bfbdd35b BVA |
759 | } |
760 | ||
139d8dc6 DLM |
761 | dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", |
762 | f->file_name, nr_zones, zone_size / 1024); | |
bfbdd35b BVA |
763 | |
764 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
765 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
bfbdd35b | 766 | if (!zbd_info) |
b7694961 | 767 | goto out; |
44ec32cb | 768 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
769 | zbd_info->refcount = 1; |
770 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
771 | for (offset = 0, j = 0; j < nr_zones;) { |
772 | z = &zones[0]; | |
773 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
774 | mutex_init_pshared_with_type(&p->mutex, |
775 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 776 | p->start = z->start; |
236d23a8 | 777 | p->capacity = z->capacity; |
139d8dc6 | 778 | |
bfbdd35b | 779 | switch (z->cond) { |
b7694961 DLM |
780 | case ZBD_ZONE_COND_NOT_WP: |
781 | case ZBD_ZONE_COND_FULL: | |
236d23a8 | 782 | p->wp = p->start + p->capacity; |
bfbdd35b BVA |
783 | break; |
784 | default: | |
785 | assert(z->start <= z->wp); | |
b7694961 DLM |
786 | assert(z->wp <= z->start + zone_size); |
787 | p->wp = z->wp; | |
bfbdd35b BVA |
788 | break; |
789 | } | |
be7a6bae DF |
790 | |
791 | switch (z->type) { | |
792 | case ZBD_ZONE_TYPE_SWR: | |
793 | p->has_wp = 1; | |
794 | break; | |
795 | default: | |
796 | p->has_wp = 0; | |
797 | } | |
bfbdd35b BVA |
798 | p->type = z->type; |
799 | p->cond = z->cond; | |
be7a6bae | 800 | |
bfbdd35b BVA |
801 | if (j > 0 && p->start != p[-1].start + zone_size) { |
802 | log_info("%s: invalid zone data\n", | |
803 | f->file_name); | |
804 | ret = -EINVAL; | |
b7694961 | 805 | goto out; |
bfbdd35b BVA |
806 | } |
807 | } | |
808 | z--; | |
b7694961 | 809 | offset = z->start + z->len; |
bfbdd35b BVA |
810 | if (j >= nr_zones) |
811 | break; | |
139d8dc6 | 812 | |
6c3f1cc1 DF |
813 | nrz = zbd_report_zones(td, f, offset, zones, |
814 | min((uint32_t)(nr_zones - j), | |
815 | ZBD_REPORT_MAX_ZONES)); | |
b7694961 DLM |
816 | if (nrz < 0) { |
817 | ret = nrz; | |
ee5e3436 SK |
818 | log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n", |
819 | offset, f->file_name, -ret); | |
b7694961 | 820 | goto out; |
bfbdd35b BVA |
821 | } |
822 | } | |
b7694961 | 823 | |
bfbdd35b | 824 | /* a sentinel */ |
b7694961 | 825 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
826 | |
827 | f->zbd_info = zbd_info; | |
828 | f->zbd_info->zone_size = zone_size; | |
829 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 830 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
831 | f->zbd_info->nr_zones = nr_zones; |
832 | zbd_info = NULL; | |
833 | ret = 0; | |
834 | ||
bfbdd35b | 835 | out: |
b7694961 DLM |
836 | sfree(zbd_info); |
837 | free(zones); | |
bfbdd35b BVA |
838 | return ret; |
839 | } | |
840 | ||
d2f442bc NC |
841 | static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f) |
842 | { | |
843 | struct zoned_block_device_info *zbd = f->zbd_info; | |
844 | unsigned int max_open_zones; | |
845 | int ret; | |
846 | ||
575686bb | 847 | if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) { |
d2f442bc NC |
848 | /* Only host-managed devices have a max open limit */ |
849 | zbd->max_open_zones = td->o.max_open_zones; | |
850 | goto out; | |
851 | } | |
852 | ||
853 | /* If host-managed, get the max open limit */ | |
854 | ret = zbd_get_max_open_zones(td, f, &max_open_zones); | |
855 | if (ret) | |
856 | return ret; | |
857 | ||
858 | if (!max_open_zones) { | |
859 | /* No device limit */ | |
860 | zbd->max_open_zones = td->o.max_open_zones; | |
861 | } else if (!td->o.max_open_zones) { | |
862 | /* No user limit. Set limit to device limit */ | |
863 | zbd->max_open_zones = max_open_zones; | |
864 | } else if (td->o.max_open_zones <= max_open_zones) { | |
865 | /* Both user limit and dev limit. User limit not too large */ | |
866 | zbd->max_open_zones = td->o.max_open_zones; | |
867 | } else { | |
868 | /* Both user limit and dev limit. User limit too large */ | |
869 | td_verror(td, EINVAL, | |
870 | "Specified --max_open_zones is too large"); | |
871 | log_err("Specified --max_open_zones (%d) is larger than max (%u)\n", | |
872 | td->o.max_open_zones, max_open_zones); | |
873 | return -EINVAL; | |
874 | } | |
875 | ||
876 | out: | |
877 | /* Ensure that the limit is not larger than FIO's internal limit */ | |
b346af90 NC |
878 | if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) { |
879 | td_verror(td, EINVAL, "'max_open_zones' value is too large"); | |
139d8dc6 DLM |
880 | log_err("'max_open_zones' value is larger than %u\n", |
881 | ZBD_MAX_OPEN_ZONES); | |
b346af90 NC |
882 | return -EINVAL; |
883 | } | |
884 | ||
d2f442bc NC |
885 | dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n", |
886 | f->file_name, zbd->max_open_zones); | |
887 | ||
888 | return 0; | |
889 | } | |
890 | ||
bfbdd35b BVA |
891 | /* |
892 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
893 | * | |
894 | * Returns 0 upon success and a negative error code upon failure. | |
895 | */ | |
379e5f09 | 896 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 897 | { |
b7694961 DLM |
898 | enum zbd_zoned_model zbd_model; |
899 | int ret; | |
bfbdd35b BVA |
900 | |
901 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
902 | ||
b7694961 DLM |
903 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
904 | if (ret) | |
905 | return ret; | |
906 | ||
bfbdd35b | 907 | switch (zbd_model) { |
b7694961 DLM |
908 | case ZBD_HOST_AWARE: |
909 | case ZBD_HOST_MANAGED: | |
bfbdd35b | 910 | ret = parse_zone_info(td, f); |
d2f442bc NC |
911 | if (ret) |
912 | return ret; | |
bfbdd35b | 913 | break; |
b7694961 | 914 | case ZBD_NONE: |
bfbdd35b | 915 | ret = init_zone_info(td, f); |
d2f442bc NC |
916 | if (ret) |
917 | return ret; | |
bfbdd35b | 918 | break; |
b7694961 DLM |
919 | default: |
920 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
921 | log_err("Unsupported zoned model\n"); | |
922 | return -EINVAL; | |
bfbdd35b | 923 | } |
b7694961 | 924 | |
2c7dd23e | 925 | assert(f->zbd_info); |
d2f442bc NC |
926 | f->zbd_info->model = zbd_model; |
927 | ||
928 | ret = zbd_set_max_open_zones(td, f); | |
929 | if (ret) { | |
930 | zbd_free_zone_info(f); | |
931 | return ret; | |
219c662d | 932 | } |
d2f442bc NC |
933 | |
934 | return 0; | |
bfbdd35b BVA |
935 | } |
936 | ||
937 | void zbd_free_zone_info(struct fio_file *f) | |
938 | { | |
939 | uint32_t refcount; | |
940 | ||
3c1dc34c | 941 | assert(f->zbd_info); |
bfbdd35b BVA |
942 | |
943 | pthread_mutex_lock(&f->zbd_info->mutex); | |
944 | refcount = --f->zbd_info->refcount; | |
945 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
946 | ||
947 | assert((int32_t)refcount >= 0); | |
948 | if (refcount == 0) | |
949 | sfree(f->zbd_info); | |
950 | f->zbd_info = NULL; | |
951 | } | |
952 | ||
953 | /* | |
954 | * Initialize f->zbd_info. | |
955 | * | |
956 | * Returns 0 upon success and a negative error code upon failure. | |
957 | * | |
958 | * Note: this function can only work correctly if it is called before the first | |
959 | * fio fork() call. | |
960 | */ | |
961 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
962 | { | |
963 | struct thread_data *td2; | |
964 | struct fio_file *f2; | |
965 | int i, j, ret; | |
966 | ||
967 | for_each_td(td2, i) { | |
968 | for_each_file(td2, f2, j) { | |
969 | if (td2 == td && f2 == file) | |
970 | continue; | |
971 | if (!f2->zbd_info || | |
972 | strcmp(f2->file_name, file->file_name) != 0) | |
973 | continue; | |
974 | file->zbd_info = f2->zbd_info; | |
975 | file->zbd_info->refcount++; | |
976 | return 0; | |
977 | } | |
978 | } | |
979 | ||
980 | ret = zbd_create_zone_info(td, file); | |
981 | if (ret < 0) | |
c5837eec | 982 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
139d8dc6 | 983 | |
bfbdd35b BVA |
984 | return ret; |
985 | } | |
986 | ||
8f39afa7 | 987 | int zbd_init_files(struct thread_data *td) |
bfbdd35b BVA |
988 | { |
989 | struct fio_file *f; | |
990 | int i; | |
991 | ||
992 | for_each_file(td, f, i) { | |
a4b7f12b | 993 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 994 | return 1; |
bfbdd35b | 995 | } |
139d8dc6 | 996 | |
8f39afa7 AD |
997 | return 0; |
998 | } | |
999 | ||
1000 | void zbd_recalc_options_with_zone_granularity(struct thread_data *td) | |
1001 | { | |
1002 | struct fio_file *f; | |
1003 | int i; | |
1004 | ||
1005 | for_each_file(td, f, i) { | |
1006 | struct zoned_block_device_info *zbd = f->zbd_info; | |
139d8dc6 | 1007 | uint64_t zone_size; |
8f39afa7 | 1008 | |
139d8dc6 DLM |
1009 | /* zonemode=strided doesn't get per-file zone size. */ |
1010 | zone_size = zbd ? zbd->zone_size : td->o.zone_size; | |
8f39afa7 AD |
1011 | if (zone_size == 0) |
1012 | continue; | |
1013 | ||
139d8dc6 | 1014 | if (td->o.size_nz > 0) |
8f39afa7 | 1015 | td->o.size = td->o.size_nz * zone_size; |
139d8dc6 | 1016 | if (td->o.io_size_nz > 0) |
8f39afa7 | 1017 | td->o.io_size = td->o.io_size_nz * zone_size; |
139d8dc6 | 1018 | if (td->o.start_offset_nz > 0) |
8f39afa7 | 1019 | td->o.start_offset = td->o.start_offset_nz * zone_size; |
139d8dc6 DLM |
1020 | if (td->o.offset_increment_nz > 0) |
1021 | td->o.offset_increment = | |
1022 | td->o.offset_increment_nz * zone_size; | |
1023 | if (td->o.zone_skip_nz > 0) | |
8f39afa7 | 1024 | td->o.zone_skip = td->o.zone_skip_nz * zone_size; |
8f39afa7 AD |
1025 | } |
1026 | } | |
1027 | ||
1028 | int zbd_setup_files(struct thread_data *td) | |
1029 | { | |
1030 | struct fio_file *f; | |
1031 | int i; | |
bfbdd35b BVA |
1032 | |
1033 | if (!zbd_using_direct_io()) { | |
1034 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
1035 | return 1; | |
1036 | } | |
1037 | ||
1038 | if (!zbd_verify_sizes()) | |
1039 | return 1; | |
1040 | ||
1041 | if (!zbd_verify_bs()) | |
1042 | return 1; | |
1043 | ||
219c662d AD |
1044 | for_each_file(td, f, i) { |
1045 | struct zoned_block_device_info *zbd = f->zbd_info; | |
954217b9 SK |
1046 | struct fio_zone_info *z; |
1047 | int zi; | |
219c662d | 1048 | |
5ddf46d0 | 1049 | assert(zbd); |
219c662d | 1050 | |
dc8a3d62 DLM |
1051 | f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset); |
1052 | f->max_zone = | |
1053 | zbd_offset_to_zone_idx(f, f->file_offset + f->io_size); | |
f952800a SK |
1054 | |
1055 | /* | |
1056 | * When all zones in the I/O range are conventional, io_size | |
1057 | * can be smaller than zone size, making min_zone the same | |
1058 | * as max_zone. This is why the assert below needs to be made | |
1059 | * conditional. | |
1060 | */ | |
1061 | if (zbd_is_seq_job(f)) | |
1062 | assert(f->min_zone < f->max_zone); | |
1063 | ||
219c662d AD |
1064 | if (td->o.max_open_zones > 0 && |
1065 | zbd->max_open_zones != td->o.max_open_zones) { | |
1066 | log_err("Different 'max_open_zones' values\n"); | |
1067 | return 1; | |
1068 | } | |
b346af90 NC |
1069 | |
1070 | /* | |
1071 | * The per job max open zones limit cannot be used without a | |
1072 | * global max open zones limit. (As the tracking of open zones | |
1073 | * is disabled when there is no global max open zones limit.) | |
1074 | */ | |
1075 | if (td->o.job_max_open_zones && !zbd->max_open_zones) { | |
1076 | log_err("'job_max_open_zones' cannot be used without a global open zones limit\n"); | |
219c662d AD |
1077 | return 1; |
1078 | } | |
954217b9 | 1079 | |
ea51055c NC |
1080 | /* |
1081 | * zbd->max_open_zones is the global limit shared for all jobs | |
1082 | * that target the same zoned block device. Force sync the per | |
1083 | * thread global limit with the actual global limit. (The real | |
1084 | * per thread/job limit is stored in td->o.job_max_open_zones). | |
1085 | */ | |
1086 | td->o.max_open_zones = zbd->max_open_zones; | |
1087 | ||
954217b9 SK |
1088 | for (zi = f->min_zone; zi < f->max_zone; zi++) { |
1089 | z = &zbd->zone_info[zi]; | |
1090 | if (z->cond != ZBD_ZONE_COND_IMP_OPEN && | |
1091 | z->cond != ZBD_ZONE_COND_EXP_OPEN) | |
1092 | continue; | |
aad7c276 | 1093 | if (zbd_open_zone(td, f, z)) |
954217b9 SK |
1094 | continue; |
1095 | /* | |
1096 | * If the number of open zones exceeds specified limits, | |
1097 | * reset all extra open zones. | |
1098 | */ | |
1099 | if (zbd_reset_zone(td, f, z) < 0) { | |
1100 | log_err("Failed to reest zone %d\n", zi); | |
1101 | return 1; | |
1102 | } | |
1103 | } | |
219c662d AD |
1104 | } |
1105 | ||
bfbdd35b BVA |
1106 | return 0; |
1107 | } | |
1108 | ||
a7c2b6fc BVA |
1109 | /* |
1110 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
1111 | * zone reset. | |
1112 | */ | |
1bb1bcad AD |
1113 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
1114 | const struct fio_file *f) | |
a7c2b6fc BVA |
1115 | { |
1116 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
1117 | ||
a7c2b6fc BVA |
1118 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
1119 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
1120 | } |
1121 | ||
1122 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
1123 | const struct fio_file *f) | |
1124 | { | |
1125 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1126 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
1127 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1128 | } | |
1129 | ||
1130 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
1131 | const struct fio_file *f) | |
1132 | { | |
1133 | uint32_t write_cnt = 0; | |
1134 | ||
1135 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1136 | assert(f->zbd_info->write_cnt); | |
1137 | if (f->zbd_info->write_cnt) | |
1138 | write_cnt = --f->zbd_info->write_cnt; | |
1139 | if (write_cnt == 0) | |
1bb1bcad | 1140 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
1141 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1142 | ||
1143 | return write_cnt == 0; | |
1144 | } | |
1145 | ||
91d25131 BVA |
1146 | enum swd_action { |
1147 | CHECK_SWD, | |
1148 | SET_SWD, | |
1149 | }; | |
1150 | ||
1151 | /* Calculate the number of sectors with data (swd) and perform action 'a' */ | |
11afb212 DF |
1152 | static uint64_t zbd_process_swd(struct thread_data *td, |
1153 | const struct fio_file *f, enum swd_action a) | |
615555bb | 1154 | { |
615555bb BVA |
1155 | struct fio_zone_info *zb, *ze, *z; |
1156 | uint64_t swd = 0; | |
403ea795 | 1157 | uint64_t wp_swd = 0; |
615555bb | 1158 | |
923f7c1e DF |
1159 | zb = get_zone(f, f->min_zone); |
1160 | ze = get_zone(f, f->max_zone); | |
615555bb | 1161 | for (z = zb; z < ze; z++) { |
403ea795 | 1162 | if (z->has_wp) { |
11afb212 | 1163 | zone_lock(td, f, z); |
403ea795 SK |
1164 | wp_swd += z->wp - z->start; |
1165 | } | |
615555bb BVA |
1166 | swd += z->wp - z->start; |
1167 | } | |
139d8dc6 | 1168 | |
615555bb | 1169 | pthread_mutex_lock(&f->zbd_info->mutex); |
91d25131 BVA |
1170 | switch (a) { |
1171 | case CHECK_SWD: | |
1172 | assert(f->zbd_info->sectors_with_data == swd); | |
403ea795 | 1173 | assert(f->zbd_info->wp_sectors_with_data == wp_swd); |
91d25131 BVA |
1174 | break; |
1175 | case SET_SWD: | |
1176 | f->zbd_info->sectors_with_data = swd; | |
403ea795 | 1177 | f->zbd_info->wp_sectors_with_data = wp_swd; |
91d25131 BVA |
1178 | break; |
1179 | } | |
615555bb | 1180 | pthread_mutex_unlock(&f->zbd_info->mutex); |
139d8dc6 | 1181 | |
615555bb | 1182 | for (z = zb; z < ze; z++) |
403ea795 SK |
1183 | if (z->has_wp) |
1184 | zone_unlock(z); | |
91d25131 BVA |
1185 | |
1186 | return swd; | |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * The swd check is useful for debugging but takes too much time to leave | |
1191 | * it enabled all the time. Hence it is disabled by default. | |
1192 | */ | |
1193 | static const bool enable_check_swd = false; | |
1194 | ||
403ea795 | 1195 | /* Check whether the values of zbd_info.*sectors_with_data are correct. */ |
11afb212 | 1196 | static void zbd_check_swd(struct thread_data *td, const struct fio_file *f) |
91d25131 BVA |
1197 | { |
1198 | if (!enable_check_swd) | |
1199 | return; | |
1200 | ||
11afb212 | 1201 | zbd_process_swd(td, f, CHECK_SWD); |
615555bb BVA |
1202 | } |
1203 | ||
bfbdd35b BVA |
1204 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
1205 | { | |
91d25131 | 1206 | struct fio_zone_info *zb, *ze; |
11afb212 | 1207 | uint64_t swd; |
bfbdd35b | 1208 | |
767d1372 | 1209 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
1210 | return; |
1211 | ||
923f7c1e DF |
1212 | zb = get_zone(f, f->min_zone); |
1213 | ze = get_zone(f, f->max_zone); | |
11afb212 | 1214 | swd = zbd_process_swd(td, f, SET_SWD); |
139d8dc6 DLM |
1215 | |
1216 | dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", | |
1217 | __func__, f->file_name, swd); | |
1218 | ||
bfbdd35b BVA |
1219 | /* |
1220 | * If data verification is enabled reset the affected zones before | |
1221 | * writing any data to avoid that a zone reset has to be issued while | |
1222 | * writing data, which causes data loss. | |
1223 | */ | |
57f88219 SK |
1224 | if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING) |
1225 | zbd_reset_zones(td, f, zb, ze); | |
a7c2b6fc | 1226 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
1227 | } |
1228 | ||
cb765e41 | 1229 | /* Return random zone index for one of the open zones. */ |
6463db6c AD |
1230 | static uint32_t pick_random_zone_idx(const struct fio_file *f, |
1231 | const struct io_u *io_u) | |
1232 | { | |
139d8dc6 DLM |
1233 | return (io_u->offset - f->file_offset) * |
1234 | f->zbd_info->num_open_zones / f->io_size; | |
6463db6c AD |
1235 | } |
1236 | ||
0f77c977 SK |
1237 | static bool any_io_in_flight(void) |
1238 | { | |
1239 | struct thread_data *td; | |
1240 | int i; | |
1241 | ||
1242 | for_each_td(td, i) { | |
1243 | if (td->io_u_in_flight) | |
1244 | return true; | |
1245 | } | |
1246 | ||
1247 | return false; | |
1248 | } | |
1249 | ||
59b07544 BVA |
1250 | /* |
1251 | * Modify the offset of an I/O unit that does not refer to an open zone such | |
1252 | * that it refers to an open zone. Close an open zone and open a new zone if | |
21c0c884 SK |
1253 | * necessary. The open zone is searched across sequential zones. |
1254 | * This algorithm can only work correctly if all write pointers are | |
59b07544 BVA |
1255 | * a multiple of the fio block size. The caller must neither hold z->mutex |
1256 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
1257 | */ | |
379e5f09 BVA |
1258 | static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, |
1259 | struct io_u *io_u) | |
59b07544 | 1260 | { |
07fc3f57 | 1261 | const uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
fae3b9a0 | 1262 | struct fio_file *f = io_u->file; |
af94a8c3 | 1263 | struct zoned_block_device_info *zbdi = f->zbd_info; |
59b07544 BVA |
1264 | struct fio_zone_info *z; |
1265 | unsigned int open_zone_idx = -1; | |
1266 | uint32_t zone_idx, new_zone_idx; | |
1267 | int i; | |
b2da58c4 | 1268 | bool wait_zone_close; |
0f77c977 SK |
1269 | bool in_flight; |
1270 | bool should_retry = true; | |
59b07544 BVA |
1271 | |
1272 | assert(is_valid_offset(f, io_u->offset)); | |
1273 | ||
ea51055c | 1274 | if (zbdi->max_open_zones || td->o.job_max_open_zones) { |
59b07544 | 1275 | /* |
af94a8c3 | 1276 | * This statement accesses zbdi->open_zones[] on purpose |
59b07544 BVA |
1277 | * without locking. |
1278 | */ | |
af94a8c3 | 1279 | zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 | 1280 | } else { |
dc8a3d62 | 1281 | zone_idx = zbd_offset_to_zone_idx(f, io_u->offset); |
59b07544 | 1282 | } |
fae3b9a0 AD |
1283 | if (zone_idx < f->min_zone) |
1284 | zone_idx = f->min_zone; | |
1285 | else if (zone_idx >= f->max_zone) | |
1286 | zone_idx = f->max_zone - 1; | |
139d8dc6 DLM |
1287 | |
1288 | dprint(FD_ZBD, | |
1289 | "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", | |
59b07544 BVA |
1290 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); |
1291 | ||
1292 | /* | |
af94a8c3 | 1293 | * Since z->mutex is the outer lock and zbdi->mutex the inner |
59b07544 | 1294 | * lock it can happen that the state of the zone with index zone_idx |
af94a8c3 | 1295 | * has changed after 'z' has been assigned and before zbdi->mutex |
59b07544 BVA |
1296 | * has been obtained. Hence the loop. |
1297 | */ | |
1298 | for (;;) { | |
6463db6c AD |
1299 | uint32_t tmp_idx; |
1300 | ||
923f7c1e | 1301 | z = get_zone(f, zone_idx); |
14351148 DF |
1302 | if (z->has_wp) |
1303 | zone_lock(td, f, z); | |
139d8dc6 | 1304 | |
af94a8c3 | 1305 | pthread_mutex_lock(&zbdi->mutex); |
139d8dc6 | 1306 | |
14351148 DF |
1307 | if (z->has_wp) { |
1308 | if (z->cond != ZBD_ZONE_COND_OFFLINE && | |
139d8dc6 DLM |
1309 | zbdi->max_open_zones == 0 && |
1310 | td->o.job_max_open_zones == 0) | |
14351148 | 1311 | goto examine_zone; |
af94a8c3 | 1312 | if (zbdi->num_open_zones == 0) { |
14351148 DF |
1313 | dprint(FD_ZBD, "%s(%s): no zones are open\n", |
1314 | __func__, f->file_name); | |
1315 | goto open_other_zone; | |
1316 | } | |
59b07544 | 1317 | } |
6463db6c AD |
1318 | |
1319 | /* | |
139d8dc6 DLM |
1320 | * List of opened zones is per-device, shared across all |
1321 | * threads. Start with quasi-random candidate zone. Ignore | |
1322 | * zones which don't belong to thread's offset/size area. | |
6463db6c AD |
1323 | */ |
1324 | open_zone_idx = pick_random_zone_idx(f, io_u); | |
1a8d510a | 1325 | assert(!open_zone_idx || |
af94a8c3 | 1326 | open_zone_idx < zbdi->num_open_zones); |
6463db6c | 1327 | tmp_idx = open_zone_idx; |
139d8dc6 | 1328 | |
af94a8c3 | 1329 | for (i = 0; i < zbdi->num_open_zones; i++) { |
6463db6c AD |
1330 | uint32_t tmpz; |
1331 | ||
af94a8c3 | 1332 | if (tmp_idx >= zbdi->num_open_zones) |
6463db6c | 1333 | tmp_idx = 0; |
af94a8c3 | 1334 | tmpz = zbdi->open_zones[tmp_idx]; |
fae3b9a0 | 1335 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
6463db6c AD |
1336 | open_zone_idx = tmp_idx; |
1337 | goto found_candidate_zone; | |
1338 | } | |
1339 | ||
1340 | tmp_idx++; | |
1341 | } | |
1342 | ||
1343 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1344 | __func__, f->file_name); | |
139d8dc6 | 1345 | |
af94a8c3 | 1346 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1347 | |
14351148 DF |
1348 | if (z->has_wp) |
1349 | zone_unlock(z); | |
139d8dc6 | 1350 | |
6463db6c AD |
1351 | return NULL; |
1352 | ||
1353 | found_candidate_zone: | |
af94a8c3 | 1354 | new_zone_idx = zbdi->open_zones[open_zone_idx]; |
59b07544 BVA |
1355 | if (new_zone_idx == zone_idx) |
1356 | break; | |
1357 | zone_idx = new_zone_idx; | |
139d8dc6 | 1358 | |
af94a8c3 | 1359 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1360 | |
14351148 DF |
1361 | if (z->has_wp) |
1362 | zone_unlock(z); | |
59b07544 BVA |
1363 | } |
1364 | ||
af94a8c3 | 1365 | /* Both z->mutex and zbdi->mutex are held. */ |
59b07544 BVA |
1366 | |
1367 | examine_zone: | |
236d23a8 | 1368 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) { |
af94a8c3 | 1369 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1370 | goto out; |
1371 | } | |
b2da58c4 SK |
1372 | |
1373 | open_other_zone: | |
1374 | /* Check if number of open zones reaches one of limits. */ | |
1375 | wait_zone_close = | |
af94a8c3 | 1376 | zbdi->num_open_zones == f->max_zone - f->min_zone || |
ea51055c NC |
1377 | (zbdi->max_open_zones && |
1378 | zbdi->num_open_zones == zbdi->max_open_zones) || | |
b2da58c4 SK |
1379 | (td->o.job_max_open_zones && |
1380 | td->num_open_zones == td->o.job_max_open_zones); | |
1381 | ||
af94a8c3 | 1382 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1383 | |
1384 | /* Only z->mutex is held. */ | |
1385 | ||
b2da58c4 SK |
1386 | /* |
1387 | * When number of open zones reaches to one of limits, wait for | |
1388 | * zone close before opening a new zone. | |
1389 | */ | |
1390 | if (wait_zone_close) { | |
139d8dc6 DLM |
1391 | dprint(FD_ZBD, |
1392 | "%s(%s): quiesce to allow open zones to close\n", | |
b2da58c4 SK |
1393 | __func__, f->file_name); |
1394 | io_u_quiesce(td); | |
1395 | } | |
1396 | ||
0f77c977 | 1397 | retry: |
59b07544 | 1398 | /* Zone 'z' is full, so try to open a new zone. */ |
af94a8c3 | 1399 | for (i = f->io_size / zbdi->zone_size; i > 0; i--) { |
59b07544 | 1400 | zone_idx++; |
21c0c884 SK |
1401 | if (z->has_wp) |
1402 | zone_unlock(z); | |
59b07544 | 1403 | z++; |
ee3696bd | 1404 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1405 | /* Wrap-around. */ |
fae3b9a0 | 1406 | zone_idx = f->min_zone; |
923f7c1e | 1407 | z = get_zone(f, zone_idx); |
59b07544 | 1408 | } |
ee3696bd | 1409 | assert(is_valid_offset(f, z->start)); |
21c0c884 SK |
1410 | if (!z->has_wp) |
1411 | continue; | |
fae3b9a0 | 1412 | zone_lock(td, f, z); |
59b07544 BVA |
1413 | if (z->open) |
1414 | continue; | |
aad7c276 | 1415 | if (zbd_open_zone(td, f, z)) |
59b07544 BVA |
1416 | goto out; |
1417 | } | |
1418 | ||
1419 | /* Only z->mutex is held. */ | |
1420 | ||
1421 | /* Check whether the write fits in any of the already opened zones. */ | |
af94a8c3 NC |
1422 | pthread_mutex_lock(&zbdi->mutex); |
1423 | for (i = 0; i < zbdi->num_open_zones; i++) { | |
1424 | zone_idx = zbdi->open_zones[i]; | |
fae3b9a0 AD |
1425 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1426 | continue; | |
af94a8c3 | 1427 | pthread_mutex_unlock(&zbdi->mutex); |
4d4c71e6 | 1428 | zone_unlock(z); |
59b07544 | 1429 | |
923f7c1e | 1430 | z = get_zone(f, zone_idx); |
59b07544 | 1431 | |
fae3b9a0 | 1432 | zone_lock(td, f, z); |
236d23a8 | 1433 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) |
59b07544 | 1434 | goto out; |
af94a8c3 | 1435 | pthread_mutex_lock(&zbdi->mutex); |
59b07544 | 1436 | } |
0f77c977 SK |
1437 | |
1438 | /* | |
1439 | * When any I/O is in-flight or when all I/Os in-flight get completed, | |
1440 | * the I/Os might have closed zones then retry the steps to open a zone. | |
1441 | * Before retry, call io_u_quiesce() to complete in-flight writes. | |
1442 | */ | |
1443 | in_flight = any_io_in_flight(); | |
1444 | if (in_flight || should_retry) { | |
139d8dc6 DLM |
1445 | dprint(FD_ZBD, |
1446 | "%s(%s): wait zone close and retry open zones\n", | |
0f77c977 SK |
1447 | __func__, f->file_name); |
1448 | pthread_mutex_unlock(&zbdi->mutex); | |
1449 | zone_unlock(z); | |
1450 | io_u_quiesce(td); | |
1451 | zone_lock(td, f, z); | |
1452 | should_retry = in_flight; | |
1453 | goto retry; | |
1454 | } | |
1455 | ||
af94a8c3 | 1456 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1457 | |
4d4c71e6 | 1458 | zone_unlock(z); |
139d8dc6 DLM |
1459 | |
1460 | dprint(FD_ZBD, "%s(%s): did not open another zone\n", | |
1461 | __func__, f->file_name); | |
1462 | ||
59b07544 BVA |
1463 | return NULL; |
1464 | ||
1465 | out: | |
139d8dc6 DLM |
1466 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", |
1467 | __func__, f->file_name, zone_idx); | |
1468 | ||
ee3696bd | 1469 | io_u->offset = z->start; |
21c0c884 | 1470 | assert(z->has_wp); |
8a866de7 | 1471 | assert(z->cond != ZBD_ZONE_COND_OFFLINE); |
139d8dc6 | 1472 | |
59b07544 BVA |
1473 | return z; |
1474 | } | |
1475 | ||
bfbdd35b | 1476 | /* The caller must hold z->mutex. */ |
59b07544 BVA |
1477 | static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, |
1478 | struct io_u *io_u, | |
1479 | struct fio_zone_info *z) | |
bfbdd35b BVA |
1480 | { |
1481 | const struct fio_file *f = io_u->file; | |
07fc3f57 | 1482 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; |
bfbdd35b | 1483 | |
aad7c276 | 1484 | if (!zbd_open_zone(td, f, z)) { |
4d4c71e6 | 1485 | zone_unlock(z); |
59b07544 BVA |
1486 | z = zbd_convert_to_open_zone(td, io_u); |
1487 | assert(z); | |
1488 | } | |
1489 | ||
0cac10b4 | 1490 | if (z->verify_block * min_bs >= z->capacity) { |
139d8dc6 DLM |
1491 | log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n", |
1492 | f->file_name, z->verify_block, min_bs, z->capacity); | |
0cac10b4 DF |
1493 | /* |
1494 | * If the assertion below fails during a test run, adding | |
1495 | * "--experimental_verify=1" to the command line may help. | |
1496 | */ | |
1497 | assert(false); | |
1498 | } | |
139d8dc6 | 1499 | |
0cac10b4 DF |
1500 | io_u->offset = z->start + z->verify_block * min_bs; |
1501 | if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
139d8dc6 DLM |
1502 | log_err("%s: %llu + %llu >= %"PRIu64"\n", |
1503 | f->file_name, io_u->offset, io_u->buflen, | |
1504 | zbd_zone_capacity_end(z)); | |
0cac10b4 DF |
1505 | assert(false); |
1506 | } | |
1507 | z->verify_block += io_u->buflen / min_bs; | |
1508 | ||
59b07544 | 1509 | return z; |
bfbdd35b BVA |
1510 | } |
1511 | ||
1512 | /* | |
5c86fdf6 SK |
1513 | * Find another zone which has @min_bytes of readable data. Search in zones |
1514 | * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf. | |
bfbdd35b | 1515 | * |
21c0c884 SK |
1516 | * Either returns NULL or returns a zone pointer. When the zone has write |
1517 | * pointer, hold the mutex for the zone. | |
bfbdd35b BVA |
1518 | */ |
1519 | static struct fio_zone_info * | |
07fc3f57 | 1520 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes, |
bfbdd35b BVA |
1521 | struct fio_zone_info *zb, struct fio_zone_info *zl) |
1522 | { | |
fae3b9a0 | 1523 | struct fio_file *f = io_u->file; |
bfbdd35b | 1524 | struct fio_zone_info *z1, *z2; |
923f7c1e | 1525 | const struct fio_zone_info *const zf = get_zone(f, f->min_zone); |
bfbdd35b BVA |
1526 | |
1527 | /* | |
1528 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1529 | * the nearest non-empty zone in case of random I/O. | |
1530 | */ | |
1531 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1532 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1533 | if (z1->has_wp) |
1534 | zone_lock(td, f, z1); | |
5c86fdf6 | 1535 | if (z1->start + min_bytes <= z1->wp) |
bfbdd35b | 1536 | return z1; |
21c0c884 SK |
1537 | if (z1->has_wp) |
1538 | zone_unlock(z1); | |
bfbdd35b BVA |
1539 | } else if (!td_random(td)) { |
1540 | break; | |
1541 | } | |
139d8dc6 | 1542 | |
bfbdd35b | 1543 | if (td_random(td) && z2 >= zf && |
b7694961 | 1544 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1545 | if (z2->has_wp) |
1546 | zone_lock(td, f, z2); | |
5c86fdf6 | 1547 | if (z2->start + min_bytes <= z2->wp) |
bfbdd35b | 1548 | return z2; |
21c0c884 SK |
1549 | if (z2->has_wp) |
1550 | zone_unlock(z2); | |
bfbdd35b BVA |
1551 | } |
1552 | } | |
139d8dc6 DLM |
1553 | |
1554 | dprint(FD_ZBD, | |
1555 | "%s: no zone has %"PRIu64" bytes of readable data\n", | |
5c86fdf6 | 1556 | f->file_name, min_bytes); |
139d8dc6 | 1557 | |
bfbdd35b BVA |
1558 | return NULL; |
1559 | } | |
1560 | ||
b2da58c4 SK |
1561 | /** |
1562 | * zbd_end_zone_io - update zone status at command completion | |
1563 | * @io_u: I/O unit | |
1564 | * @z: zone info pointer | |
1565 | * | |
1566 | * If the write command made the zone full, close it. | |
1567 | * | |
1568 | * The caller must hold z->mutex. | |
1569 | */ | |
1570 | static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, | |
1571 | struct fio_zone_info *z) | |
1572 | { | |
1573 | const struct fio_file *f = io_u->file; | |
1574 | ||
1575 | if (io_u->ddir == DDIR_WRITE && | |
1576 | io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1577 | pthread_mutex_lock(&f->zbd_info->mutex); | |
a23411bb | 1578 | zbd_close_zone(td, f, z); |
b2da58c4 SK |
1579 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1580 | } | |
1581 | } | |
1582 | ||
bfbdd35b | 1583 | /** |
d9ed3e63 | 1584 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1585 | * @io_u: I/O unit |
d9ed3e63 DLM |
1586 | * @success: Whether or not the I/O unit has been queued successfully |
1587 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1588 | * |
d9ed3e63 DLM |
1589 | * For write and trim operations, update the write pointer of the I/O unit |
1590 | * target zone. | |
bfbdd35b | 1591 | */ |
b2da58c4 SK |
1592 | static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, |
1593 | bool success) | |
bfbdd35b | 1594 | { |
d9ed3e63 DLM |
1595 | const struct fio_file *f = io_u->file; |
1596 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b BVA |
1597 | struct fio_zone_info *z; |
1598 | uint32_t zone_idx; | |
d9ed3e63 | 1599 | uint64_t zone_end; |
bfbdd35b | 1600 | |
5ddf46d0 | 1601 | assert(zbd_info); |
bfbdd35b | 1602 | |
dc8a3d62 | 1603 | zone_idx = zbd_offset_to_zone_idx(f, io_u->offset); |
bfbdd35b | 1604 | assert(zone_idx < zbd_info->nr_zones); |
923f7c1e | 1605 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1606 | |
43bcbd5b | 1607 | assert(z->has_wp); |
d9ed3e63 | 1608 | |
bfbdd35b BVA |
1609 | if (!success) |
1610 | goto unlock; | |
d9ed3e63 DLM |
1611 | |
1612 | dprint(FD_ZBD, | |
1613 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
1614 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1615 | ||
bfbdd35b BVA |
1616 | switch (io_u->ddir) { |
1617 | case DDIR_WRITE: | |
d9ed3e63 | 1618 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
236d23a8 | 1619 | zbd_zone_capacity_end(z)); |
139d8dc6 | 1620 | |
a7c2b6fc BVA |
1621 | /* |
1622 | * z->wp > zone_end means that one or more I/O errors | |
1623 | * have occurred. | |
1624 | */ | |
139d8dc6 | 1625 | pthread_mutex_lock(&zbd_info->mutex); |
403ea795 | 1626 | if (z->wp <= zone_end) { |
a7c2b6fc | 1627 | zbd_info->sectors_with_data += zone_end - z->wp; |
403ea795 SK |
1628 | zbd_info->wp_sectors_with_data += zone_end - z->wp; |
1629 | } | |
a7c2b6fc | 1630 | pthread_mutex_unlock(&zbd_info->mutex); |
bfbdd35b BVA |
1631 | z->wp = zone_end; |
1632 | break; | |
bfbdd35b BVA |
1633 | default: |
1634 | break; | |
1635 | } | |
d9ed3e63 | 1636 | |
b2da58c4 SK |
1637 | if (q == FIO_Q_COMPLETED && !io_u->error) |
1638 | zbd_end_zone_io(td, io_u, z); | |
1639 | ||
bfbdd35b | 1640 | unlock: |
d9ed3e63 DLM |
1641 | if (!success || q != FIO_Q_QUEUED) { |
1642 | /* BUSY or COMPLETED: unlock the zone */ | |
4d4c71e6 | 1643 | zone_unlock(z); |
d9ed3e63 DLM |
1644 | io_u->zbd_put_io = NULL; |
1645 | } | |
1646 | } | |
1647 | ||
1648 | /** | |
1649 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1650 | * @io_u: I/O unit | |
1651 | */ | |
b2da58c4 | 1652 | static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) |
d9ed3e63 DLM |
1653 | { |
1654 | const struct fio_file *f = io_u->file; | |
1655 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1656 | struct fio_zone_info *z; | |
1657 | uint32_t zone_idx; | |
1658 | ||
5ddf46d0 | 1659 | assert(zbd_info); |
615555bb | 1660 | |
dc8a3d62 | 1661 | zone_idx = zbd_offset_to_zone_idx(f, io_u->offset); |
d9ed3e63 | 1662 | assert(zone_idx < zbd_info->nr_zones); |
923f7c1e | 1663 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1664 | |
43bcbd5b | 1665 | assert(z->has_wp); |
d9ed3e63 DLM |
1666 | |
1667 | dprint(FD_ZBD, | |
1668 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
1669 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1670 | ||
b2da58c4 SK |
1671 | zbd_end_zone_io(td, io_u, z); |
1672 | ||
4d4c71e6 | 1673 | zone_unlock(z); |
11afb212 | 1674 | zbd_check_swd(td, f); |
bfbdd35b BVA |
1675 | } |
1676 | ||
9d87c646 DLM |
1677 | /* |
1678 | * Windows and MacOS do not define this. | |
1679 | */ | |
1680 | #ifndef EREMOTEIO | |
1681 | #define EREMOTEIO 121 /* POSIX value */ | |
1682 | #endif | |
1683 | ||
bfbdd35b BVA |
1684 | bool zbd_unaligned_write(int error_code) |
1685 | { | |
1686 | switch (error_code) { | |
1687 | case EIO: | |
1688 | case EREMOTEIO: | |
1689 | return true; | |
1690 | } | |
1691 | return false; | |
1692 | } | |
1693 | ||
4d37720a DLM |
1694 | /** |
1695 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1696 | * @td: FIO thread data. | |
1697 | * @io_u: FIO I/O unit. | |
1698 | * | |
1699 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1700 | * no more IO can be performed in the current zone. | |
1701 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1702 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1703 | * - For write workloads, zoneskip is applied when the zone is full. | |
1704 | * This applies only to read and write operations. | |
1705 | */ | |
1706 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1707 | { | |
1708 | struct fio_file *f = io_u->file; | |
1709 | enum fio_ddir ddir = io_u->ddir; | |
1710 | struct fio_zone_info *z; | |
1711 | uint32_t zone_idx; | |
1712 | ||
1713 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1714 | assert(td->o.zone_size); | |
5ddf46d0 | 1715 | assert(f->zbd_info); |
4d37720a | 1716 | |
dc8a3d62 | 1717 | zone_idx = zbd_offset_to_zone_idx(f, f->last_pos[ddir]); |
923f7c1e | 1718 | z = get_zone(f, zone_idx); |
236d23a8 SK |
1719 | |
1720 | /* | |
1721 | * When the zone capacity is smaller than the zone size and the I/O is | |
1722 | * sequential write, skip to zone end if the latest position is at the | |
1723 | * zone capacity limit. | |
1724 | */ | |
139d8dc6 DLM |
1725 | if (z->capacity < f->zbd_info->zone_size && |
1726 | !td_random(td) && ddir == DDIR_WRITE && | |
236d23a8 SK |
1727 | f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { |
1728 | dprint(FD_ZBD, | |
1729 | "%s: Jump from zone capacity limit to zone end:" | |
ee5e3436 SK |
1730 | " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n", |
1731 | f->file_name, f->last_pos[ddir], | |
1732 | zbd_zone_end(z), zone_idx, z->capacity); | |
236d23a8 SK |
1733 | td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; |
1734 | f->last_pos[ddir] = zbd_zone_end(z); | |
1735 | } | |
1736 | ||
4d37720a DLM |
1737 | /* |
1738 | * zone_skip is valid only for sequential workloads. | |
1739 | */ | |
1740 | if (td_random(td) || !td->o.zone_skip) | |
1741 | return; | |
1742 | ||
1743 | /* | |
1744 | * It is time to switch to a new zone if: | |
1745 | * - zone_bytes == zone_size bytes have already been accessed | |
1746 | * - The last position reached the end of the current zone. | |
1747 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1748 | * reached the zone write pointer. | |
1749 | */ | |
4d37720a | 1750 | if (td->zone_bytes >= td->o.zone_size || |
236d23a8 | 1751 | f->last_pos[ddir] >= zbd_zone_end(z) || |
4d37720a DLM |
1752 | (ddir == DDIR_READ && |
1753 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1754 | /* | |
1755 | * Skip zones. | |
1756 | */ | |
1757 | td->zone_bytes = 0; | |
1758 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1759 | ||
1760 | /* | |
1761 | * Wrap from the beginning, if we exceed the file size | |
1762 | */ | |
1763 | if (f->file_offset >= f->real_file_size) | |
1764 | f->file_offset = get_start_offset(td, f); | |
1765 | ||
1766 | f->last_pos[ddir] = f->file_offset; | |
1767 | td->io_skip_bytes += td->o.zone_skip; | |
1768 | } | |
1769 | } | |
1770 | ||
c65057f9 | 1771 | /** |
c7d5e152 | 1772 | * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. |
c65057f9 SK |
1773 | * |
1774 | * @td: FIO thread data. | |
1775 | * @io_u: FIO I/O unit. | |
1776 | * @ddir: I/O direction before adjustment. | |
1777 | * | |
1778 | * Return adjusted I/O direction. | |
1779 | */ | |
1780 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1781 | enum fio_ddir ddir) | |
1782 | { | |
1783 | /* | |
1784 | * In case read direction is chosen for the first random I/O, fio with | |
1785 | * zonemode=zbd stops because no data can be read from zoned block | |
1786 | * devices with all empty zones. Overwrite the first I/O direction as | |
1787 | * write to make sure data to read exists. | |
1788 | */ | |
5ddf46d0 | 1789 | assert(io_u->file->zbd_info); |
731461cc | 1790 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1791 | return ddir; |
1792 | ||
1793 | if (io_u->file->zbd_info->sectors_with_data || | |
1794 | td->o.read_beyond_wp) | |
1795 | return DDIR_READ; | |
1796 | ||
1797 | return DDIR_WRITE; | |
1798 | } | |
1799 | ||
bfbdd35b BVA |
1800 | /** |
1801 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1802 | * @td: FIO thread data. | |
1803 | * @io_u: FIO I/O unit. | |
1804 | * | |
1805 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1806 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1807 | * corresponds to io_u->offset at the end of this function. | |
1808 | */ | |
1809 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1810 | { | |
b7694961 | 1811 | struct fio_file *f = io_u->file; |
af94a8c3 | 1812 | struct zoned_block_device_info *zbdi = f->zbd_info; |
bfbdd35b | 1813 | uint32_t zone_idx_b; |
de65f7b7 | 1814 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b | 1815 | uint32_t orig_len = io_u->buflen; |
07fc3f57 | 1816 | uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
bfbdd35b BVA |
1817 | uint64_t new_len; |
1818 | int64_t range; | |
1819 | ||
af94a8c3 | 1820 | assert(zbdi); |
adc6adcb | 1821 | assert(min_bs); |
bfbdd35b BVA |
1822 | assert(is_valid_offset(f, io_u->offset)); |
1823 | assert(io_u->buflen); | |
139d8dc6 | 1824 | |
dc8a3d62 | 1825 | zone_idx_b = zbd_offset_to_zone_idx(f, io_u->offset); |
923f7c1e | 1826 | zb = get_zone(f, zone_idx_b); |
de65f7b7 | 1827 | orig_zb = zb; |
bfbdd35b | 1828 | |
2efcf74b SK |
1829 | if (!zb->has_wp) { |
1830 | /* Accept non-write I/Os for conventional zones. */ | |
1831 | if (io_u->ddir != DDIR_WRITE) | |
1832 | return io_u_accept; | |
139d8dc6 | 1833 | |
2efcf74b SK |
1834 | /* |
1835 | * Make sure that writes to conventional zones | |
1836 | * don't cross over to any sequential zones. | |
1837 | */ | |
1838 | if (!(zb + 1)->has_wp || | |
1839 | io_u->offset + io_u->buflen <= (zb + 1)->start) | |
1840 | return io_u_accept; | |
1841 | ||
1842 | if (io_u->offset + min_bs > (zb + 1)->start) { | |
1843 | dprint(FD_IO, | |
07fc3f57 | 1844 | "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n", |
1e3d6e03 | 1845 | f->file_name, io_u->offset, |
ee5e3436 | 1846 | min_bs, (zb + 1)->start); |
139d8dc6 DLM |
1847 | io_u->offset = |
1848 | zb->start + (zb + 1)->start - io_u->offset; | |
1849 | new_len = min(io_u->buflen, | |
1850 | (zb + 1)->start - io_u->offset); | |
2efcf74b SK |
1851 | } else { |
1852 | new_len = (zb + 1)->start - io_u->offset; | |
1853 | } | |
139d8dc6 | 1854 | |
2efcf74b | 1855 | io_u->buflen = new_len / min_bs * min_bs; |
139d8dc6 | 1856 | |
bfbdd35b | 1857 | return io_u_accept; |
2efcf74b | 1858 | } |
bfbdd35b BVA |
1859 | |
1860 | /* | |
1861 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1862 | * is enabled. | |
1863 | */ | |
b7694961 | 1864 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1865 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1866 | return io_u_accept; | |
1867 | ||
11afb212 | 1868 | zbd_check_swd(td, f); |
615555bb | 1869 | |
fae3b9a0 | 1870 | zone_lock(td, f, zb); |
6f0c6085 | 1871 | |
bfbdd35b BVA |
1872 | switch (io_u->ddir) { |
1873 | case DDIR_READ: | |
0ed7d55e AR |
1874 | if (td->runstate == TD_VERIFYING && td_write(td)) { |
1875 | zb = zbd_replay_write_order(td, io_u, zb); | |
bfbdd35b BVA |
1876 | goto accept; |
1877 | } | |
139d8dc6 | 1878 | |
bfbdd35b | 1879 | /* |
de65f7b7 DLM |
1880 | * Check that there is enough written data in the zone to do an |
1881 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1882 | * the I/O. | |
bfbdd35b | 1883 | */ |
b7694961 | 1884 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1885 | zb->wp - zb->start : 0; |
de65f7b7 | 1886 | if (range < min_bs || |
ee3696bd | 1887 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
4d4c71e6 | 1888 | zone_unlock(zb); |
923f7c1e | 1889 | zl = get_zone(f, f->max_zone); |
5c86fdf6 | 1890 | zb = zbd_find_zone(td, io_u, min_bs, zb, zl); |
bfbdd35b BVA |
1891 | if (!zb) { |
1892 | dprint(FD_ZBD, | |
1893 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1894 | f->file_name, io_u->offset, | |
1895 | io_u->buflen); | |
1896 | goto eof; | |
1897 | } | |
de65f7b7 DLM |
1898 | /* |
1899 | * zbd_find_zone() returned a zone with a range of at | |
1900 | * least min_bs. | |
1901 | */ | |
ee3696bd | 1902 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1903 | assert(range >= min_bs); |
1904 | ||
1905 | if (!td_random(td)) | |
ee3696bd | 1906 | io_u->offset = zb->start; |
bfbdd35b | 1907 | } |
139d8dc6 | 1908 | |
de65f7b7 DLM |
1909 | /* |
1910 | * Make sure the I/O is within the zone valid data range while | |
1911 | * maximizing the I/O size and preserving randomness. | |
1912 | */ | |
1913 | if (range <= io_u->buflen) | |
ee3696bd | 1914 | io_u->offset = zb->start; |
de65f7b7 | 1915 | else if (td_random(td)) |
ee3696bd DLM |
1916 | io_u->offset = zb->start + |
1917 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 | 1918 | (range - io_u->buflen)) / min_bs * min_bs; |
139d8dc6 | 1919 | |
43bcbd5b SK |
1920 | /* |
1921 | * When zbd_find_zone() returns a conventional zone, | |
1922 | * we can simply accept the new i/o offset here. | |
1923 | */ | |
1924 | if (!zb->has_wp) | |
1925 | return io_u_accept; | |
139d8dc6 | 1926 | |
de65f7b7 DLM |
1927 | /* |
1928 | * Make sure the I/O does not cross over the zone wp position. | |
1929 | */ | |
1930 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1931 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1932 | new_len = new_len / min_bs * min_bs; |
1933 | if (new_len < io_u->buflen) { | |
1934 | io_u->buflen = new_len; | |
1935 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1936 | orig_len, io_u->buflen); | |
bfbdd35b | 1937 | } |
139d8dc6 | 1938 | |
ee3696bd DLM |
1939 | assert(zb->start <= io_u->offset); |
1940 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
139d8dc6 | 1941 | |
bfbdd35b | 1942 | goto accept; |
139d8dc6 | 1943 | |
bfbdd35b | 1944 | case DDIR_WRITE: |
af94a8c3 | 1945 | if (io_u->buflen > zbdi->zone_size) { |
1c74aadc DF |
1946 | td_verror(td, EINVAL, "I/O buflen exceeds zone size"); |
1947 | dprint(FD_IO, | |
ee5e3436 SK |
1948 | "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n", |
1949 | f->file_name, io_u->buflen, zbdi->zone_size); | |
bfbdd35b | 1950 | goto eof; |
1c74aadc | 1951 | } |
139d8dc6 | 1952 | |
aad7c276 | 1953 | if (!zbd_open_zone(td, f, zb)) { |
4d4c71e6 | 1954 | zone_unlock(zb); |
59b07544 | 1955 | zb = zbd_convert_to_open_zone(td, io_u); |
1c74aadc DF |
1956 | if (!zb) { |
1957 | dprint(FD_IO, "%s: can't convert to open zone", | |
1958 | f->file_name); | |
59b07544 | 1959 | goto eof; |
1c74aadc | 1960 | } |
59b07544 | 1961 | } |
139d8dc6 | 1962 | |
a7c2b6fc BVA |
1963 | /* Check whether the zone reset threshold has been exceeded */ |
1964 | if (td->o.zrf.u.f) { | |
139d8dc6 DLM |
1965 | if (zbdi->wp_sectors_with_data >= f->io_size * td->o.zrt.u.f && |
1966 | zbd_dec_and_reset_write_cnt(td, f)) | |
a7c2b6fc | 1967 | zb->reset_zone = 1; |
a7c2b6fc | 1968 | } |
139d8dc6 | 1969 | |
bfbdd35b BVA |
1970 | /* Reset the zone pointer if necessary */ |
1971 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
1972 | assert(td->o.verify == VERIFY_NONE); | |
1973 | /* | |
1974 | * Since previous write requests may have been submitted | |
1975 | * asynchronously and since we will submit the zone | |
1976 | * reset synchronously, wait until previously submitted | |
1977 | * write requests have completed before issuing a | |
1978 | * zone reset. | |
1979 | */ | |
1980 | io_u_quiesce(td); | |
1981 | zb->reset_zone = 0; | |
1982 | if (zbd_reset_zone(td, f, zb) < 0) | |
1983 | goto eof; | |
236d23a8 SK |
1984 | |
1985 | if (zb->capacity < min_bs) { | |
1c74aadc | 1986 | td_verror(td, EINVAL, "ZCAP is less min_bs"); |
07fc3f57 | 1987 | log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n", |
ee5e3436 | 1988 | zb->capacity, min_bs); |
236d23a8 SK |
1989 | goto eof; |
1990 | } | |
bfbdd35b | 1991 | } |
139d8dc6 | 1992 | |
bfbdd35b BVA |
1993 | /* Make writes occur at the write pointer */ |
1994 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 1995 | io_u->offset = zb->wp; |
bfbdd35b | 1996 | if (!is_valid_offset(f, io_u->offset)) { |
1c74aadc DF |
1997 | td_verror(td, EINVAL, "invalid WP value"); |
1998 | dprint(FD_ZBD, "%s: dropped request with offset %llu\n", | |
1999 | f->file_name, io_u->offset); | |
bfbdd35b BVA |
2000 | goto eof; |
2001 | } | |
139d8dc6 | 2002 | |
bfbdd35b BVA |
2003 | /* |
2004 | * Make sure that the buflen is a multiple of the minimal | |
2005 | * block size. Give up if shrinking would make the request too | |
2006 | * small. | |
2007 | */ | |
2008 | new_len = min((unsigned long long)io_u->buflen, | |
236d23a8 | 2009 | zbd_zone_capacity_end(zb) - io_u->offset); |
bfbdd35b BVA |
2010 | new_len = new_len / min_bs * min_bs; |
2011 | if (new_len == io_u->buflen) | |
2012 | goto accept; | |
2013 | if (new_len >= min_bs) { | |
2014 | io_u->buflen = new_len; | |
2015 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
2016 | orig_len, io_u->buflen); | |
2017 | goto accept; | |
2018 | } | |
139d8dc6 | 2019 | |
1c74aadc | 2020 | td_verror(td, EIO, "zone remainder too small"); |
07fc3f57 | 2021 | log_err("zone remainder %lld smaller than min block size %"PRIu64"\n", |
1c74aadc | 2022 | (zbd_zone_capacity_end(zb) - io_u->offset), min_bs); |
139d8dc6 | 2023 | |
bfbdd35b | 2024 | goto eof; |
139d8dc6 | 2025 | |
bfbdd35b | 2026 | case DDIR_TRIM: |
e3be810b SK |
2027 | /* Check random trim targets a non-empty zone */ |
2028 | if (!td_random(td) || zb->wp > zb->start) | |
2029 | goto accept; | |
2030 | ||
2031 | /* Find out a non-empty zone to trim */ | |
2032 | zone_unlock(zb); | |
2033 | zl = get_zone(f, f->max_zone); | |
2034 | zb = zbd_find_zone(td, io_u, 1, zb, zl); | |
2035 | if (zb) { | |
2036 | io_u->offset = zb->start; | |
2037 | dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n", | |
2038 | f->file_name, io_u->offset); | |
2039 | goto accept; | |
2040 | } | |
139d8dc6 | 2041 | |
e3be810b | 2042 | goto eof; |
139d8dc6 | 2043 | |
bfbdd35b | 2044 | case DDIR_SYNC: |
e3be810b | 2045 | /* fall-through */ |
bfbdd35b BVA |
2046 | case DDIR_DATASYNC: |
2047 | case DDIR_SYNC_FILE_RANGE: | |
2048 | case DDIR_WAIT: | |
2049 | case DDIR_LAST: | |
2050 | case DDIR_INVAL: | |
2051 | goto accept; | |
2052 | } | |
2053 | ||
2054 | assert(false); | |
2055 | ||
2056 | accept: | |
43bcbd5b | 2057 | assert(zb->has_wp); |
b7694961 | 2058 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
2059 | assert(!io_u->zbd_queue_io); |
2060 | assert(!io_u->zbd_put_io); | |
139d8dc6 | 2061 | |
d9ed3e63 DLM |
2062 | io_u->zbd_queue_io = zbd_queue_io; |
2063 | io_u->zbd_put_io = zbd_put_io; | |
139d8dc6 | 2064 | |
2ef3c1b0 DF |
2065 | /* |
2066 | * Since we return with the zone lock still held, | |
2067 | * add an annotation to let Coverity know that it | |
2068 | * is intentional. | |
2069 | */ | |
2070 | /* coverity[missing_unlock] */ | |
139d8dc6 | 2071 | |
bfbdd35b BVA |
2072 | return io_u_accept; |
2073 | ||
2074 | eof: | |
43bcbd5b | 2075 | if (zb && zb->has_wp) |
4d4c71e6 | 2076 | zone_unlock(zb); |
139d8dc6 | 2077 | |
bfbdd35b BVA |
2078 | return io_u_eof; |
2079 | } | |
fd5d733f BVA |
2080 | |
2081 | /* Return a string with ZBD statistics */ | |
2082 | char *zbd_write_status(const struct thread_stat *ts) | |
2083 | { | |
2084 | char *res; | |
2085 | ||
ee5e3436 | 2086 | if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0) |
fd5d733f BVA |
2087 | return NULL; |
2088 | return res; | |
2089 | } | |
e3be810b SK |
2090 | |
2091 | /** | |
2092 | * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim | |
2093 | * | |
2094 | * @td: FIO thread data. | |
2095 | * @io_u: FIO I/O unit. | |
2096 | * | |
2097 | * It is assumed that z->mutex is already locked. | |
2098 | * Return io_u_completed when reset zone succeeds. Return 0 when the target zone | |
2099 | * does not have write pointer. On error, return negative errno. | |
2100 | */ | |
2101 | int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u) | |
2102 | { | |
2103 | struct fio_file *f = io_u->file; | |
2104 | struct fio_zone_info *z; | |
2105 | uint32_t zone_idx; | |
2106 | int ret; | |
2107 | ||
dc8a3d62 | 2108 | zone_idx = zbd_offset_to_zone_idx(f, io_u->offset); |
e3be810b SK |
2109 | z = get_zone(f, zone_idx); |
2110 | ||
2111 | if (!z->has_wp) | |
2112 | return 0; | |
2113 | ||
2114 | if (io_u->offset != z->start) { | |
139d8dc6 DLM |
2115 | log_err("Trim offset not at zone start (%lld)\n", |
2116 | io_u->offset); | |
e3be810b SK |
2117 | return -EINVAL; |
2118 | } | |
2119 | ||
2120 | ret = zbd_reset_zone((struct thread_data *)td, f, z); | |
2121 | if (ret < 0) | |
2122 | return ret; | |
2123 | ||
2124 | return io_u_completed; | |
2125 | } |