Commit | Line | Data |
---|---|---|
bfbdd35b BVA |
1 | /* |
2 | * Copyright (C) 2018 Western Digital Corporation or its affiliates. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include <errno.h> | |
8 | #include <string.h> | |
9 | #include <stdlib.h> | |
bfbdd35b | 10 | #include <fcntl.h> |
bfbdd35b BVA |
11 | #include <sys/stat.h> |
12 | #include <unistd.h> | |
f5bff36e | 13 | |
cf42d79e | 14 | #include "os/os.h" |
bfbdd35b BVA |
15 | #include "file.h" |
16 | #include "fio.h" | |
17 | #include "lib/pow2.h" | |
18 | #include "log.h" | |
f5bff36e | 19 | #include "oslib/asprintf.h" |
bfbdd35b BVA |
20 | #include "smalloc.h" |
21 | #include "verify.h" | |
44ec32cb | 22 | #include "pshared.h" |
bfbdd35b BVA |
23 | #include "zbd.h" |
24 | ||
410a071c DLM |
25 | static bool is_valid_offset(const struct fio_file *f, uint64_t offset) |
26 | { | |
27 | return (uint64_t)(offset - f->file_offset) < f->io_size; | |
28 | } | |
29 | ||
30 | static inline unsigned int zbd_zone_nr(const struct fio_file *f, | |
31 | struct fio_zone_info *zone) | |
32 | { | |
33 | return zone - f->zbd_info->zone_info; | |
34 | } | |
35 | ||
36 | /** | |
37 | * zbd_zone_idx - convert an offset into a zone number | |
38 | * @f: file pointer. | |
39 | * @offset: offset in bytes. If this offset is in the first zone_size bytes | |
40 | * past the disk size then the index of the sentinel is returned. | |
41 | */ | |
42 | static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) | |
43 | { | |
44 | uint32_t zone_idx; | |
45 | ||
46 | if (f->zbd_info->zone_size_log2 > 0) | |
47 | zone_idx = offset >> f->zbd_info->zone_size_log2; | |
48 | else | |
49 | zone_idx = offset / f->zbd_info->zone_size; | |
50 | ||
51 | return min(zone_idx, f->zbd_info->nr_zones); | |
52 | } | |
53 | ||
54 | /** | |
55 | * zbd_zone_end - Return zone end location | |
56 | * @z: zone info pointer. | |
57 | */ | |
58 | static inline uint64_t zbd_zone_end(const struct fio_zone_info *z) | |
59 | { | |
60 | return (z+1)->start; | |
61 | } | |
62 | ||
63 | /** | |
64 | * zbd_zone_capacity_end - Return zone capacity limit end location | |
65 | * @z: zone info pointer. | |
66 | */ | |
67 | static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z) | |
68 | { | |
69 | return z->start + z->capacity; | |
70 | } | |
71 | ||
72 | /** | |
73 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
74 | * @f: file pointer. | |
75 | * @z: zone info pointer. | |
76 | * @required: minimum number of bytes that must remain in a zone. | |
77 | * | |
78 | * The caller must hold z->mutex. | |
79 | */ | |
80 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
81 | uint64_t required) | |
82 | { | |
83 | assert((required & 511) == 0); | |
84 | ||
85 | return z->has_wp && | |
86 | z->wp + required > zbd_zone_capacity_end(z); | |
87 | } | |
88 | ||
89 | static void zone_lock(struct thread_data *td, const struct fio_file *f, | |
90 | struct fio_zone_info *z) | |
91 | { | |
92 | struct zoned_block_device_info *zbd = f->zbd_info; | |
93 | uint32_t nz = z - zbd->zone_info; | |
94 | ||
95 | /* A thread should never lock zones outside its working area. */ | |
96 | assert(f->min_zone <= nz && nz < f->max_zone); | |
97 | ||
98 | assert(z->has_wp); | |
99 | ||
100 | /* | |
101 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
102 | * is changed or when io_u completes and zbd_put_io() executed. | |
103 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
104 | * other waiting for zone locks when building an io_u batch, first | |
105 | * only trylock the zone. If the zone is already locked by another job, | |
106 | * process the currently queued I/Os so that I/O progress is made and | |
107 | * zones unlocked. | |
108 | */ | |
109 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
110 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
111 | io_u_quiesce(td); | |
112 | pthread_mutex_lock(&z->mutex); | |
113 | } | |
114 | } | |
115 | ||
116 | static inline void zone_unlock(struct fio_zone_info *z) | |
117 | { | |
118 | int ret; | |
119 | ||
120 | assert(z->has_wp); | |
121 | ret = pthread_mutex_unlock(&z->mutex); | |
122 | assert(!ret); | |
123 | } | |
124 | ||
125 | static inline struct fio_zone_info *get_zone(const struct fio_file *f, | |
126 | unsigned int zone_nr) | |
127 | { | |
128 | return &f->zbd_info->zone_info[zone_nr]; | |
129 | } | |
130 | ||
b7694961 DLM |
131 | /** |
132 | * zbd_get_zoned_model - Get a device zoned model | |
133 | * @td: FIO thread data | |
134 | * @f: FIO file for which to get model information | |
135 | */ | |
38334c13 DLM |
136 | static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, |
137 | enum zbd_zoned_model *model) | |
b7694961 DLM |
138 | { |
139 | int ret; | |
140 | ||
50cc48d5 NC |
141 | if (f->filetype == FIO_TYPE_PIPE) { |
142 | log_err("zonemode=zbd does not support pipes\n"); | |
143 | return -EINVAL; | |
144 | } | |
145 | ||
9db0cde8 NC |
146 | /* If regular file, always emulate zones inside the file. */ |
147 | if (f->filetype == FIO_TYPE_FILE) { | |
148 | *model = ZBD_NONE; | |
149 | return 0; | |
150 | } | |
151 | ||
6c5b11d3 DLM |
152 | if (td->io_ops && td->io_ops->get_zoned_model) |
153 | ret = td->io_ops->get_zoned_model(td, f, model); | |
154 | else | |
155 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
156 | if (ret < 0) { |
157 | td_verror(td, errno, "get zoned model failed"); | |
158 | log_err("%s: get zoned model failed (%d).\n", | |
159 | f->file_name, errno); | |
160 | } | |
161 | ||
162 | return ret; | |
163 | } | |
164 | ||
165 | /** | |
166 | * zbd_report_zones - Get zone information | |
167 | * @td: FIO thread data. | |
168 | * @f: FIO file for which to get zone information | |
169 | * @offset: offset from which to report zones | |
170 | * @zones: Array of struct zbd_zone | |
171 | * @nr_zones: Size of @zones array | |
172 | * | |
173 | * Get zone information into @zones starting from the zone at offset @offset | |
174 | * for the device specified by @f. | |
175 | * | |
176 | * Returns the number of zones reported upon success and a negative error code | |
177 | * upon failure. If the zone report is empty, always assume an error (device | |
178 | * problem) and return -EIO. | |
179 | */ | |
38334c13 DLM |
180 | static int zbd_report_zones(struct thread_data *td, struct fio_file *f, |
181 | uint64_t offset, struct zbd_zone *zones, | |
182 | unsigned int nr_zones) | |
b7694961 DLM |
183 | { |
184 | int ret; | |
185 | ||
6c5b11d3 DLM |
186 | if (td->io_ops && td->io_ops->report_zones) |
187 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
188 | else | |
189 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
190 | if (ret < 0) { |
191 | td_verror(td, errno, "report zones failed"); | |
ee5e3436 SK |
192 | log_err("%s: report zones from sector %"PRIu64" failed (%d).\n", |
193 | f->file_name, offset >> 9, errno); | |
b7694961 DLM |
194 | } else if (ret == 0) { |
195 | td_verror(td, errno, "Empty zone report"); | |
ee5e3436 SK |
196 | log_err("%s: report zones from sector %"PRIu64" is empty.\n", |
197 | f->file_name, offset >> 9); | |
b7694961 DLM |
198 | ret = -EIO; |
199 | } | |
200 | ||
201 | return ret; | |
202 | } | |
203 | ||
204 | /** | |
205 | * zbd_reset_wp - reset the write pointer of a range of zones | |
206 | * @td: FIO thread data. | |
207 | * @f: FIO file for which to reset zones | |
208 | * @offset: Starting offset of the first zone to reset | |
209 | * @length: Length of the range of zones to reset | |
210 | * | |
211 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
212 | * Returns 0 upon success and a negative error code upon failure. | |
213 | */ | |
38334c13 DLM |
214 | static int zbd_reset_wp(struct thread_data *td, struct fio_file *f, |
215 | uint64_t offset, uint64_t length) | |
b7694961 DLM |
216 | { |
217 | int ret; | |
218 | ||
6c5b11d3 DLM |
219 | if (td->io_ops && td->io_ops->reset_wp) |
220 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
221 | else | |
222 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
223 | if (ret < 0) { |
224 | td_verror(td, errno, "resetting wp failed"); | |
ee5e3436 SK |
225 | log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n", |
226 | f->file_name, length >> 9, offset >> 9, errno); | |
b7694961 DLM |
227 | } |
228 | ||
229 | return ret; | |
230 | } | |
231 | ||
410a071c DLM |
232 | /** |
233 | * zbd_reset_zone - reset the write pointer of a single zone | |
234 | * @td: FIO thread data. | |
235 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
236 | * @z: Zone to reset. | |
237 | * | |
238 | * Returns 0 upon success and a negative error code upon failure. | |
239 | * | |
240 | * The caller must hold z->mutex. | |
241 | */ | |
242 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, | |
243 | struct fio_zone_info *z) | |
244 | { | |
245 | uint64_t offset = z->start; | |
246 | uint64_t length = (z+1)->start - offset; | |
247 | uint64_t data_in_zone = z->wp - z->start; | |
248 | int ret = 0; | |
249 | ||
250 | if (!data_in_zone) | |
251 | return 0; | |
252 | ||
253 | assert(is_valid_offset(f, offset + length - 1)); | |
254 | ||
139d8dc6 DLM |
255 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", |
256 | f->file_name, zbd_zone_nr(f, z)); | |
257 | ||
410a071c DLM |
258 | switch (f->zbd_info->model) { |
259 | case ZBD_HOST_AWARE: | |
260 | case ZBD_HOST_MANAGED: | |
261 | ret = zbd_reset_wp(td, f, offset, length); | |
262 | if (ret < 0) | |
263 | return ret; | |
264 | break; | |
265 | default: | |
266 | break; | |
267 | } | |
268 | ||
269 | pthread_mutex_lock(&f->zbd_info->mutex); | |
270 | f->zbd_info->sectors_with_data -= data_in_zone; | |
271 | f->zbd_info->wp_sectors_with_data -= data_in_zone; | |
272 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
139d8dc6 | 273 | |
410a071c DLM |
274 | z->wp = z->start; |
275 | z->verify_block = 0; | |
276 | ||
277 | td->ts.nr_zone_resets++; | |
278 | ||
279 | return ret; | |
280 | } | |
281 | ||
282 | /** | |
283 | * zbd_close_zone - Remove a zone from the open zones array. | |
284 | * @td: FIO thread data. | |
285 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
286 | * @zone_idx: Index of the zone to remove. | |
287 | * | |
288 | * The caller must hold f->zbd_info->mutex. | |
289 | */ | |
290 | static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, | |
291 | unsigned int zone_idx) | |
292 | { | |
293 | uint32_t open_zone_idx = 0; | |
294 | ||
295 | for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) { | |
296 | if (f->zbd_info->open_zones[open_zone_idx] == zone_idx) | |
297 | break; | |
298 | } | |
299 | if (open_zone_idx == f->zbd_info->num_open_zones) | |
300 | return; | |
301 | ||
139d8dc6 DLM |
302 | dprint(FD_ZBD, "%s: closing zone %d\n", |
303 | f->file_name, zone_idx); | |
304 | ||
410a071c DLM |
305 | memmove(f->zbd_info->open_zones + open_zone_idx, |
306 | f->zbd_info->open_zones + open_zone_idx + 1, | |
307 | (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) * | |
308 | sizeof(f->zbd_info->open_zones[0])); | |
139d8dc6 | 309 | |
410a071c DLM |
310 | f->zbd_info->num_open_zones--; |
311 | td->num_open_zones--; | |
312 | get_zone(f, zone_idx)->open = 0; | |
313 | } | |
314 | ||
315 | /** | |
316 | * zbd_reset_zones - Reset a range of zones. | |
317 | * @td: fio thread data. | |
318 | * @f: fio file for which to reset zones | |
319 | * @zb: first zone to reset. | |
320 | * @ze: first zone not to reset. | |
321 | * | |
322 | * Returns 0 upon success and 1 upon failure. | |
323 | */ | |
324 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
325 | struct fio_zone_info *const zb, | |
326 | struct fio_zone_info *const ze) | |
327 | { | |
328 | struct fio_zone_info *z; | |
329 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
330 | int res = 0; | |
331 | ||
332 | assert(min_bs); | |
333 | ||
139d8dc6 DLM |
334 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", |
335 | f->file_name, zbd_zone_nr(f, zb), zbd_zone_nr(f, ze)); | |
336 | ||
410a071c DLM |
337 | for (z = zb; z < ze; z++) { |
338 | uint32_t nz = zbd_zone_nr(f, z); | |
339 | ||
340 | if (!z->has_wp) | |
341 | continue; | |
139d8dc6 | 342 | |
410a071c DLM |
343 | zone_lock(td, f, z); |
344 | pthread_mutex_lock(&f->zbd_info->mutex); | |
345 | zbd_close_zone(td, f, nz); | |
346 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
139d8dc6 | 347 | |
410a071c DLM |
348 | if (z->wp != z->start) { |
349 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
350 | f->file_name, zbd_zone_nr(f, z)); | |
351 | if (zbd_reset_zone(td, f, z) < 0) | |
352 | res = 1; | |
353 | } | |
139d8dc6 | 354 | |
410a071c DLM |
355 | zone_unlock(z); |
356 | } | |
357 | ||
358 | return res; | |
359 | } | |
360 | ||
d2f442bc NC |
361 | /** |
362 | * zbd_get_max_open_zones - Get the maximum number of open zones | |
363 | * @td: FIO thread data | |
364 | * @f: FIO file for which to get max open zones | |
365 | * @max_open_zones: Upon success, result will be stored here. | |
366 | * | |
367 | * A @max_open_zones value set to zero means no limit. | |
368 | * | |
369 | * Returns 0 upon success and a negative error code upon failure. | |
370 | */ | |
38334c13 DLM |
371 | static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f, |
372 | unsigned int *max_open_zones) | |
d2f442bc NC |
373 | { |
374 | int ret; | |
375 | ||
376 | if (td->io_ops && td->io_ops->get_max_open_zones) | |
377 | ret = td->io_ops->get_max_open_zones(td, f, max_open_zones); | |
378 | else | |
379 | ret = blkzoned_get_max_open_zones(td, f, max_open_zones); | |
380 | if (ret < 0) { | |
381 | td_verror(td, errno, "get max open zones failed"); | |
382 | log_err("%s: get max open zones failed (%d).\n", | |
383 | f->file_name, errno); | |
384 | } | |
385 | ||
386 | return ret; | |
387 | } | |
388 | ||
bfbdd35b | 389 | /** |
410a071c DLM |
390 | * zbd_open_zone - Add a zone to the array of open zones. |
391 | * @td: fio thread data. | |
392 | * @f: fio file that has the open zones to add. | |
393 | * @zone_idx: Index of the zone to add. | |
bfbdd35b | 394 | * |
410a071c DLM |
395 | * Open a ZBD zone if it is not already open. Returns true if either the zone |
396 | * was already open or if the zone was successfully added to the array of open | |
397 | * zones without exceeding the maximum number of open zones. Returns false if | |
398 | * the zone was not already open and opening the zone would cause the zone limit | |
399 | * to be exceeded. | |
bfbdd35b | 400 | */ |
410a071c DLM |
401 | static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f, |
402 | uint32_t zone_idx) | |
1f57803b | 403 | { |
410a071c DLM |
404 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; |
405 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
406 | struct fio_zone_info *z = get_zone(f, zone_idx); | |
407 | bool res = true; | |
fae3b9a0 | 408 | |
410a071c DLM |
409 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
410 | return false; | |
43bcbd5b | 411 | |
1f57803b | 412 | /* |
410a071c DLM |
413 | * Skip full zones with data verification enabled because resetting a |
414 | * zone causes data loss and hence causes verification to fail. | |
1f57803b | 415 | */ |
410a071c DLM |
416 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) |
417 | return false; | |
4d4c71e6 | 418 | |
410a071c DLM |
419 | /* |
420 | * zbdi->max_open_zones == 0 means that there is no limit on the maximum | |
421 | * number of open zones. In this case, do no track open zones in | |
422 | * zbdi->open_zones array. | |
423 | */ | |
424 | if (!zbdi->max_open_zones) | |
425 | return true; | |
4d4c71e6 | 426 | |
410a071c | 427 | pthread_mutex_lock(&zbdi->mutex); |
b5a0f7ce DLM |
428 | |
429 | if (z->open) { | |
410a071c | 430 | /* |
b5a0f7ce DLM |
431 | * If the zone is going to be completely filled by writes |
432 | * already in-flight, handle it as a full zone instead of an | |
433 | * open zone. | |
410a071c DLM |
434 | */ |
435 | if (z->wp >= zbd_zone_capacity_end(z)) | |
436 | res = false; | |
437 | goto out; | |
438 | } | |
139d8dc6 | 439 | |
410a071c DLM |
440 | res = false; |
441 | /* Zero means no limit */ | |
442 | if (td->o.job_max_open_zones > 0 && | |
443 | td->num_open_zones >= td->o.job_max_open_zones) | |
444 | goto out; | |
445 | if (zbdi->num_open_zones >= zbdi->max_open_zones) | |
446 | goto out; | |
139d8dc6 DLM |
447 | |
448 | dprint(FD_ZBD, "%s: opening zone %d\n", | |
449 | f->file_name, zone_idx); | |
450 | ||
410a071c DLM |
451 | zbdi->open_zones[zbdi->num_open_zones++] = zone_idx; |
452 | td->num_open_zones++; | |
453 | z->open = 1; | |
454 | res = true; | |
bfbdd35b | 455 | |
410a071c DLM |
456 | out: |
457 | pthread_mutex_unlock(&zbdi->mutex); | |
458 | return res; | |
923f7c1e DF |
459 | } |
460 | ||
bfbdd35b BVA |
461 | /* Verify whether direct I/O is used for all host-managed zoned drives. */ |
462 | static bool zbd_using_direct_io(void) | |
463 | { | |
464 | struct thread_data *td; | |
465 | struct fio_file *f; | |
466 | int i, j; | |
467 | ||
468 | for_each_td(td, i) { | |
469 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) | |
470 | continue; | |
471 | for_each_file(td, f, j) { | |
472 | if (f->zbd_info && | |
b7694961 | 473 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
474 | return false; |
475 | } | |
476 | } | |
477 | ||
478 | return true; | |
479 | } | |
480 | ||
481 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
482 | static bool zbd_is_seq_job(struct fio_file *f) | |
483 | { | |
484 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
485 | ||
486 | assert(f->zbd_info); | |
139d8dc6 | 487 | |
bfbdd35b BVA |
488 | if (f->io_size == 0) |
489 | return false; | |
139d8dc6 | 490 | |
bfbdd35b BVA |
491 | zone_idx_b = zbd_zone_idx(f, f->file_offset); |
492 | zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1); | |
493 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) | |
be7a6bae | 494 | if (get_zone(f, zone_idx)->has_wp) |
bfbdd35b BVA |
495 | return true; |
496 | ||
497 | return false; | |
498 | } | |
499 | ||
0bf93a1a DLM |
500 | /* |
501 | * Verify whether the file offset and size parameters are aligned with zone | |
502 | * boundaries. If the file offset is not aligned, align it down to the start of | |
503 | * the zone containing the start offset and align up the file io_size parameter. | |
504 | */ | |
505 | static bool zbd_zone_align_file_sizes(struct thread_data *td, | |
506 | struct fio_file *f) | |
507 | { | |
508 | const struct fio_zone_info *z; | |
509 | uint64_t new_offset, new_end; | |
510 | uint32_t zone_idx; | |
511 | ||
512 | if (!f->zbd_info) | |
513 | return true; | |
514 | if (f->file_offset >= f->real_file_size) | |
515 | return true; | |
516 | if (!zbd_is_seq_job(f)) | |
517 | return true; | |
518 | ||
519 | if (!td->o.zone_size) { | |
520 | td->o.zone_size = f->zbd_info->zone_size; | |
521 | if (!td->o.zone_size) { | |
522 | log_err("%s: invalid 0 zone size\n", | |
523 | f->file_name); | |
524 | return false; | |
525 | } | |
526 | } else if (td->o.zone_size != f->zbd_info->zone_size) { | |
527 | log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n", | |
528 | f->file_name, td->o.zone_size, | |
529 | f->zbd_info->zone_size); | |
530 | return false; | |
531 | } | |
532 | ||
533 | if (td->o.zone_skip % td->o.zone_size) { | |
534 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", | |
535 | f->file_name, td->o.zone_skip, | |
536 | td->o.zone_size); | |
537 | return false; | |
538 | } | |
539 | ||
540 | zone_idx = zbd_zone_idx(f, f->file_offset); | |
541 | z = get_zone(f, zone_idx); | |
542 | if ((f->file_offset != z->start) && | |
543 | (td->o.td_ddir != TD_DDIR_READ)) { | |
544 | new_offset = zbd_zone_end(z); | |
545 | if (new_offset >= f->file_offset + f->io_size) { | |
546 | log_info("%s: io_size must be at least one zone\n", | |
547 | f->file_name); | |
548 | return false; | |
549 | } | |
550 | log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n", | |
551 | f->file_name, f->file_offset, | |
552 | new_offset); | |
553 | f->io_size -= (new_offset - f->file_offset); | |
554 | f->file_offset = new_offset; | |
555 | } | |
556 | ||
557 | zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size); | |
558 | z = get_zone(f, zone_idx); | |
559 | new_end = z->start; | |
560 | if ((td->o.td_ddir != TD_DDIR_READ) && | |
561 | (f->file_offset + f->io_size != new_end)) { | |
562 | if (new_end <= f->file_offset) { | |
563 | log_info("%s: io_size must be at least one zone\n", | |
564 | f->file_name); | |
565 | return false; | |
566 | } | |
567 | log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n", | |
568 | f->file_name, f->io_size, | |
569 | new_end - f->file_offset); | |
570 | f->io_size = new_end - f->file_offset; | |
571 | } | |
572 | ||
573 | return true; | |
574 | } | |
575 | ||
bfbdd35b BVA |
576 | /* |
577 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
578 | */ | |
579 | static bool zbd_verify_sizes(void) | |
580 | { | |
bfbdd35b BVA |
581 | struct thread_data *td; |
582 | struct fio_file *f; | |
bfbdd35b BVA |
583 | int i, j; |
584 | ||
585 | for_each_td(td, i) { | |
586 | for_each_file(td, f, j) { | |
0bf93a1a | 587 | if (!zbd_zone_align_file_sizes(td, f)) |
4d37720a | 588 | return false; |
bfbdd35b BVA |
589 | } |
590 | } | |
591 | ||
592 | return true; | |
593 | } | |
594 | ||
595 | static bool zbd_verify_bs(void) | |
596 | { | |
597 | struct thread_data *td; | |
598 | struct fio_file *f; | |
bfbdd35b BVA |
599 | int i, j, k; |
600 | ||
601 | for_each_td(td, i) { | |
e3be810b SK |
602 | if (td_trim(td) && |
603 | (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] || | |
604 | td->o.bssplit_nr[DDIR_TRIM])) { | |
605 | log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n"); | |
606 | return false; | |
607 | } | |
bfbdd35b | 608 | for_each_file(td, f, j) { |
1ddd225e AD |
609 | uint64_t zone_size; |
610 | ||
bfbdd35b BVA |
611 | if (!f->zbd_info) |
612 | continue; | |
139d8dc6 | 613 | |
bfbdd35b | 614 | zone_size = f->zbd_info->zone_size; |
e3be810b | 615 | if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) { |
ee5e3436 | 616 | log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n", |
e3be810b | 617 | f->file_name, td->o.bs[DDIR_TRIM], |
ee5e3436 | 618 | zone_size); |
e3be810b SK |
619 | return false; |
620 | } | |
59f94d26 | 621 | for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) { |
bfbdd35b | 622 | if (td->o.verify != VERIFY_NONE && |
ee3696bd | 623 | zone_size % td->o.bs[k] != 0) { |
ee5e3436 | 624 | log_info("%s: block size %llu is not a divisor of the zone size %"PRIu64"\n", |
bfbdd35b | 625 | f->file_name, td->o.bs[k], |
ee5e3436 | 626 | zone_size); |
bfbdd35b BVA |
627 | return false; |
628 | } | |
629 | } | |
630 | } | |
631 | } | |
632 | return true; | |
633 | } | |
634 | ||
bfbdd35b BVA |
635 | static int ilog2(uint64_t i) |
636 | { | |
637 | int log = -1; | |
638 | ||
639 | while (i) { | |
640 | i >>= 1; | |
641 | log++; | |
642 | } | |
643 | return log; | |
644 | } | |
645 | ||
646 | /* | |
647 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
648 | * allows to execute a ZBD workload against a non-ZBD device. | |
649 | */ | |
650 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
651 | { | |
652 | uint32_t nr_zones; | |
653 | struct fio_zone_info *p; | |
a4b7f12b | 654 | uint64_t zone_size = td->o.zone_size; |
b8dd9750 | 655 | uint64_t zone_capacity = td->o.zone_capacity; |
bfbdd35b | 656 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
657 | int i; |
658 | ||
a4b7f12b | 659 | if (zone_size == 0) { |
9db0cde8 | 660 | log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n", |
a4b7f12b DLM |
661 | f->file_name); |
662 | return 1; | |
663 | } | |
664 | ||
665 | if (zone_size < 512) { | |
666 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
667 | f->file_name); | |
668 | return 1; | |
669 | } | |
670 | ||
b8dd9750 HH |
671 | if (zone_capacity == 0) |
672 | zone_capacity = zone_size; | |
673 | ||
674 | if (zone_capacity > zone_size) { | |
675 | log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", | |
ee5e3436 | 676 | f->file_name, td->o.zone_capacity, td->o.zone_size); |
b8dd9750 HH |
677 | return 1; |
678 | } | |
679 | ||
9db0cde8 NC |
680 | if (f->real_file_size < zone_size) { |
681 | log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n", | |
682 | f->file_name, f->real_file_size, zone_size); | |
683 | return -EINVAL; | |
684 | } | |
685 | ||
ee3696bd | 686 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
687 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
688 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
689 | if (!zbd_info) | |
690 | return -ENOMEM; | |
691 | ||
44ec32cb | 692 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
693 | zbd_info->refcount = 1; |
694 | p = &zbd_info->zone_info[0]; | |
695 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
696 | mutex_init_pshared_with_type(&p->mutex, |
697 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b | 698 | p->start = i * zone_size; |
b14651a2 | 699 | p->wp = p->start; |
b7694961 DLM |
700 | p->type = ZBD_ZONE_TYPE_SWR; |
701 | p->cond = ZBD_ZONE_COND_EMPTY; | |
b8dd9750 | 702 | p->capacity = zone_capacity; |
be7a6bae | 703 | p->has_wp = 1; |
bfbdd35b BVA |
704 | } |
705 | /* a sentinel */ | |
706 | p->start = nr_zones * zone_size; | |
707 | ||
708 | f->zbd_info = zbd_info; | |
709 | f->zbd_info->zone_size = zone_size; | |
710 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 711 | ilog2(zone_size) : 0; |
bfbdd35b | 712 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
713 | return 0; |
714 | } | |
715 | ||
716 | /* | |
b7694961 DLM |
717 | * Maximum number of zones to report in one operation. |
718 | */ | |
719 | #define ZBD_REPORT_MAX_ZONES 8192U | |
720 | ||
721 | /* | |
722 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
723 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
724 | */ |
725 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
726 | { | |
b7694961 DLM |
727 | int nr_zones, nrz; |
728 | struct zbd_zone *zones, *z; | |
bfbdd35b | 729 | struct fio_zone_info *p; |
b7694961 | 730 | uint64_t zone_size, offset; |
bfbdd35b | 731 | struct zoned_block_device_info *zbd_info = NULL; |
d060babc | 732 | int i, j, ret = -ENOMEM; |
bfbdd35b | 733 | |
b7694961 DLM |
734 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
735 | if (!zones) | |
bfbdd35b BVA |
736 | goto out; |
737 | ||
b7694961 DLM |
738 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
739 | if (nrz < 0) { | |
740 | ret = nrz; | |
741 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
742 | f->file_name, -ret); | |
743 | goto out; | |
bfbdd35b BVA |
744 | } |
745 | ||
b7694961 | 746 | zone_size = zones[0].len; |
ee3696bd | 747 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
748 | |
749 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
750 | td->o.zone_size = zone_size; |
751 | } else if (td->o.zone_size != zone_size) { | |
ee5e3436 SK |
752 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n", |
753 | f->file_name, td->o.zone_size, zone_size); | |
bfbdd35b | 754 | ret = -EINVAL; |
b7694961 | 755 | goto out; |
bfbdd35b BVA |
756 | } |
757 | ||
139d8dc6 DLM |
758 | dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", |
759 | f->file_name, nr_zones, zone_size / 1024); | |
bfbdd35b BVA |
760 | |
761 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
762 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
bfbdd35b | 763 | if (!zbd_info) |
b7694961 | 764 | goto out; |
44ec32cb | 765 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
766 | zbd_info->refcount = 1; |
767 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
768 | for (offset = 0, j = 0; j < nr_zones;) { |
769 | z = &zones[0]; | |
770 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
771 | mutex_init_pshared_with_type(&p->mutex, |
772 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 773 | p->start = z->start; |
236d23a8 | 774 | p->capacity = z->capacity; |
139d8dc6 | 775 | |
bfbdd35b | 776 | switch (z->cond) { |
b7694961 DLM |
777 | case ZBD_ZONE_COND_NOT_WP: |
778 | case ZBD_ZONE_COND_FULL: | |
236d23a8 | 779 | p->wp = p->start + p->capacity; |
bfbdd35b BVA |
780 | break; |
781 | default: | |
782 | assert(z->start <= z->wp); | |
b7694961 DLM |
783 | assert(z->wp <= z->start + zone_size); |
784 | p->wp = z->wp; | |
bfbdd35b BVA |
785 | break; |
786 | } | |
be7a6bae DF |
787 | |
788 | switch (z->type) { | |
789 | case ZBD_ZONE_TYPE_SWR: | |
790 | p->has_wp = 1; | |
791 | break; | |
792 | default: | |
793 | p->has_wp = 0; | |
794 | } | |
bfbdd35b BVA |
795 | p->type = z->type; |
796 | p->cond = z->cond; | |
be7a6bae | 797 | |
bfbdd35b BVA |
798 | if (j > 0 && p->start != p[-1].start + zone_size) { |
799 | log_info("%s: invalid zone data\n", | |
800 | f->file_name); | |
801 | ret = -EINVAL; | |
b7694961 | 802 | goto out; |
bfbdd35b BVA |
803 | } |
804 | } | |
805 | z--; | |
b7694961 | 806 | offset = z->start + z->len; |
bfbdd35b BVA |
807 | if (j >= nr_zones) |
808 | break; | |
139d8dc6 | 809 | |
6c3f1cc1 DF |
810 | nrz = zbd_report_zones(td, f, offset, zones, |
811 | min((uint32_t)(nr_zones - j), | |
812 | ZBD_REPORT_MAX_ZONES)); | |
b7694961 DLM |
813 | if (nrz < 0) { |
814 | ret = nrz; | |
ee5e3436 SK |
815 | log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n", |
816 | offset, f->file_name, -ret); | |
b7694961 | 817 | goto out; |
bfbdd35b BVA |
818 | } |
819 | } | |
b7694961 | 820 | |
bfbdd35b | 821 | /* a sentinel */ |
b7694961 | 822 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
823 | |
824 | f->zbd_info = zbd_info; | |
825 | f->zbd_info->zone_size = zone_size; | |
826 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 827 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
828 | f->zbd_info->nr_zones = nr_zones; |
829 | zbd_info = NULL; | |
830 | ret = 0; | |
831 | ||
bfbdd35b | 832 | out: |
b7694961 DLM |
833 | sfree(zbd_info); |
834 | free(zones); | |
bfbdd35b BVA |
835 | return ret; |
836 | } | |
837 | ||
d2f442bc NC |
838 | static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f) |
839 | { | |
840 | struct zoned_block_device_info *zbd = f->zbd_info; | |
841 | unsigned int max_open_zones; | |
842 | int ret; | |
843 | ||
575686bb | 844 | if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) { |
d2f442bc NC |
845 | /* Only host-managed devices have a max open limit */ |
846 | zbd->max_open_zones = td->o.max_open_zones; | |
847 | goto out; | |
848 | } | |
849 | ||
850 | /* If host-managed, get the max open limit */ | |
851 | ret = zbd_get_max_open_zones(td, f, &max_open_zones); | |
852 | if (ret) | |
853 | return ret; | |
854 | ||
855 | if (!max_open_zones) { | |
856 | /* No device limit */ | |
857 | zbd->max_open_zones = td->o.max_open_zones; | |
858 | } else if (!td->o.max_open_zones) { | |
859 | /* No user limit. Set limit to device limit */ | |
860 | zbd->max_open_zones = max_open_zones; | |
861 | } else if (td->o.max_open_zones <= max_open_zones) { | |
862 | /* Both user limit and dev limit. User limit not too large */ | |
863 | zbd->max_open_zones = td->o.max_open_zones; | |
864 | } else { | |
865 | /* Both user limit and dev limit. User limit too large */ | |
866 | td_verror(td, EINVAL, | |
867 | "Specified --max_open_zones is too large"); | |
868 | log_err("Specified --max_open_zones (%d) is larger than max (%u)\n", | |
869 | td->o.max_open_zones, max_open_zones); | |
870 | return -EINVAL; | |
871 | } | |
872 | ||
873 | out: | |
874 | /* Ensure that the limit is not larger than FIO's internal limit */ | |
b346af90 NC |
875 | if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) { |
876 | td_verror(td, EINVAL, "'max_open_zones' value is too large"); | |
139d8dc6 DLM |
877 | log_err("'max_open_zones' value is larger than %u\n", |
878 | ZBD_MAX_OPEN_ZONES); | |
b346af90 NC |
879 | return -EINVAL; |
880 | } | |
881 | ||
d2f442bc NC |
882 | dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n", |
883 | f->file_name, zbd->max_open_zones); | |
884 | ||
885 | return 0; | |
886 | } | |
887 | ||
bfbdd35b BVA |
888 | /* |
889 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
890 | * | |
891 | * Returns 0 upon success and a negative error code upon failure. | |
892 | */ | |
379e5f09 | 893 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 894 | { |
b7694961 DLM |
895 | enum zbd_zoned_model zbd_model; |
896 | int ret; | |
bfbdd35b BVA |
897 | |
898 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
899 | ||
b7694961 DLM |
900 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
901 | if (ret) | |
902 | return ret; | |
903 | ||
bfbdd35b | 904 | switch (zbd_model) { |
b7694961 DLM |
905 | case ZBD_HOST_AWARE: |
906 | case ZBD_HOST_MANAGED: | |
bfbdd35b | 907 | ret = parse_zone_info(td, f); |
d2f442bc NC |
908 | if (ret) |
909 | return ret; | |
bfbdd35b | 910 | break; |
b7694961 | 911 | case ZBD_NONE: |
bfbdd35b | 912 | ret = init_zone_info(td, f); |
d2f442bc NC |
913 | if (ret) |
914 | return ret; | |
bfbdd35b | 915 | break; |
b7694961 DLM |
916 | default: |
917 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
918 | log_err("Unsupported zoned model\n"); | |
919 | return -EINVAL; | |
bfbdd35b | 920 | } |
b7694961 | 921 | |
2c7dd23e | 922 | assert(f->zbd_info); |
d2f442bc NC |
923 | f->zbd_info->model = zbd_model; |
924 | ||
925 | ret = zbd_set_max_open_zones(td, f); | |
926 | if (ret) { | |
927 | zbd_free_zone_info(f); | |
928 | return ret; | |
219c662d | 929 | } |
d2f442bc NC |
930 | |
931 | return 0; | |
bfbdd35b BVA |
932 | } |
933 | ||
934 | void zbd_free_zone_info(struct fio_file *f) | |
935 | { | |
936 | uint32_t refcount; | |
937 | ||
3c1dc34c | 938 | assert(f->zbd_info); |
bfbdd35b BVA |
939 | |
940 | pthread_mutex_lock(&f->zbd_info->mutex); | |
941 | refcount = --f->zbd_info->refcount; | |
942 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
943 | ||
944 | assert((int32_t)refcount >= 0); | |
945 | if (refcount == 0) | |
946 | sfree(f->zbd_info); | |
947 | f->zbd_info = NULL; | |
948 | } | |
949 | ||
950 | /* | |
951 | * Initialize f->zbd_info. | |
952 | * | |
953 | * Returns 0 upon success and a negative error code upon failure. | |
954 | * | |
955 | * Note: this function can only work correctly if it is called before the first | |
956 | * fio fork() call. | |
957 | */ | |
958 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
959 | { | |
960 | struct thread_data *td2; | |
961 | struct fio_file *f2; | |
962 | int i, j, ret; | |
963 | ||
964 | for_each_td(td2, i) { | |
965 | for_each_file(td2, f2, j) { | |
966 | if (td2 == td && f2 == file) | |
967 | continue; | |
968 | if (!f2->zbd_info || | |
969 | strcmp(f2->file_name, file->file_name) != 0) | |
970 | continue; | |
971 | file->zbd_info = f2->zbd_info; | |
972 | file->zbd_info->refcount++; | |
973 | return 0; | |
974 | } | |
975 | } | |
976 | ||
977 | ret = zbd_create_zone_info(td, file); | |
978 | if (ret < 0) | |
c5837eec | 979 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
139d8dc6 | 980 | |
bfbdd35b BVA |
981 | return ret; |
982 | } | |
983 | ||
8f39afa7 | 984 | int zbd_init_files(struct thread_data *td) |
bfbdd35b BVA |
985 | { |
986 | struct fio_file *f; | |
987 | int i; | |
988 | ||
989 | for_each_file(td, f, i) { | |
a4b7f12b | 990 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 991 | return 1; |
bfbdd35b | 992 | } |
139d8dc6 | 993 | |
8f39afa7 AD |
994 | return 0; |
995 | } | |
996 | ||
997 | void zbd_recalc_options_with_zone_granularity(struct thread_data *td) | |
998 | { | |
999 | struct fio_file *f; | |
1000 | int i; | |
1001 | ||
1002 | for_each_file(td, f, i) { | |
1003 | struct zoned_block_device_info *zbd = f->zbd_info; | |
139d8dc6 | 1004 | uint64_t zone_size; |
8f39afa7 | 1005 | |
139d8dc6 DLM |
1006 | /* zonemode=strided doesn't get per-file zone size. */ |
1007 | zone_size = zbd ? zbd->zone_size : td->o.zone_size; | |
8f39afa7 AD |
1008 | if (zone_size == 0) |
1009 | continue; | |
1010 | ||
139d8dc6 | 1011 | if (td->o.size_nz > 0) |
8f39afa7 | 1012 | td->o.size = td->o.size_nz * zone_size; |
139d8dc6 | 1013 | if (td->o.io_size_nz > 0) |
8f39afa7 | 1014 | td->o.io_size = td->o.io_size_nz * zone_size; |
139d8dc6 | 1015 | if (td->o.start_offset_nz > 0) |
8f39afa7 | 1016 | td->o.start_offset = td->o.start_offset_nz * zone_size; |
139d8dc6 DLM |
1017 | if (td->o.offset_increment_nz > 0) |
1018 | td->o.offset_increment = | |
1019 | td->o.offset_increment_nz * zone_size; | |
1020 | if (td->o.zone_skip_nz > 0) | |
8f39afa7 | 1021 | td->o.zone_skip = td->o.zone_skip_nz * zone_size; |
8f39afa7 AD |
1022 | } |
1023 | } | |
1024 | ||
1025 | int zbd_setup_files(struct thread_data *td) | |
1026 | { | |
1027 | struct fio_file *f; | |
1028 | int i; | |
bfbdd35b BVA |
1029 | |
1030 | if (!zbd_using_direct_io()) { | |
1031 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
1032 | return 1; | |
1033 | } | |
1034 | ||
1035 | if (!zbd_verify_sizes()) | |
1036 | return 1; | |
1037 | ||
1038 | if (!zbd_verify_bs()) | |
1039 | return 1; | |
1040 | ||
219c662d AD |
1041 | for_each_file(td, f, i) { |
1042 | struct zoned_block_device_info *zbd = f->zbd_info; | |
954217b9 SK |
1043 | struct fio_zone_info *z; |
1044 | int zi; | |
219c662d | 1045 | |
5ddf46d0 | 1046 | assert(zbd); |
219c662d | 1047 | |
f952800a SK |
1048 | f->min_zone = zbd_zone_idx(f, f->file_offset); |
1049 | f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size); | |
1050 | ||
1051 | /* | |
1052 | * When all zones in the I/O range are conventional, io_size | |
1053 | * can be smaller than zone size, making min_zone the same | |
1054 | * as max_zone. This is why the assert below needs to be made | |
1055 | * conditional. | |
1056 | */ | |
1057 | if (zbd_is_seq_job(f)) | |
1058 | assert(f->min_zone < f->max_zone); | |
1059 | ||
219c662d AD |
1060 | if (td->o.max_open_zones > 0 && |
1061 | zbd->max_open_zones != td->o.max_open_zones) { | |
1062 | log_err("Different 'max_open_zones' values\n"); | |
1063 | return 1; | |
1064 | } | |
b346af90 NC |
1065 | |
1066 | /* | |
1067 | * The per job max open zones limit cannot be used without a | |
1068 | * global max open zones limit. (As the tracking of open zones | |
1069 | * is disabled when there is no global max open zones limit.) | |
1070 | */ | |
1071 | if (td->o.job_max_open_zones && !zbd->max_open_zones) { | |
1072 | log_err("'job_max_open_zones' cannot be used without a global open zones limit\n"); | |
219c662d AD |
1073 | return 1; |
1074 | } | |
954217b9 | 1075 | |
ea51055c NC |
1076 | /* |
1077 | * zbd->max_open_zones is the global limit shared for all jobs | |
1078 | * that target the same zoned block device. Force sync the per | |
1079 | * thread global limit with the actual global limit. (The real | |
1080 | * per thread/job limit is stored in td->o.job_max_open_zones). | |
1081 | */ | |
1082 | td->o.max_open_zones = zbd->max_open_zones; | |
1083 | ||
954217b9 SK |
1084 | for (zi = f->min_zone; zi < f->max_zone; zi++) { |
1085 | z = &zbd->zone_info[zi]; | |
1086 | if (z->cond != ZBD_ZONE_COND_IMP_OPEN && | |
1087 | z->cond != ZBD_ZONE_COND_EXP_OPEN) | |
1088 | continue; | |
1089 | if (zbd_open_zone(td, f, zi)) | |
1090 | continue; | |
1091 | /* | |
1092 | * If the number of open zones exceeds specified limits, | |
1093 | * reset all extra open zones. | |
1094 | */ | |
1095 | if (zbd_reset_zone(td, f, z) < 0) { | |
1096 | log_err("Failed to reest zone %d\n", zi); | |
1097 | return 1; | |
1098 | } | |
1099 | } | |
219c662d AD |
1100 | } |
1101 | ||
bfbdd35b BVA |
1102 | return 0; |
1103 | } | |
1104 | ||
a7c2b6fc BVA |
1105 | /* |
1106 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
1107 | * zone reset. | |
1108 | */ | |
1bb1bcad AD |
1109 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
1110 | const struct fio_file *f) | |
a7c2b6fc BVA |
1111 | { |
1112 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
1113 | ||
a7c2b6fc BVA |
1114 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
1115 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
1116 | } |
1117 | ||
1118 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
1119 | const struct fio_file *f) | |
1120 | { | |
1121 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1122 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
1123 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1124 | } | |
1125 | ||
1126 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
1127 | const struct fio_file *f) | |
1128 | { | |
1129 | uint32_t write_cnt = 0; | |
1130 | ||
1131 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1132 | assert(f->zbd_info->write_cnt); | |
1133 | if (f->zbd_info->write_cnt) | |
1134 | write_cnt = --f->zbd_info->write_cnt; | |
1135 | if (write_cnt == 0) | |
1bb1bcad | 1136 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
1137 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1138 | ||
1139 | return write_cnt == 0; | |
1140 | } | |
1141 | ||
91d25131 BVA |
1142 | enum swd_action { |
1143 | CHECK_SWD, | |
1144 | SET_SWD, | |
1145 | }; | |
1146 | ||
1147 | /* Calculate the number of sectors with data (swd) and perform action 'a' */ | |
11afb212 DF |
1148 | static uint64_t zbd_process_swd(struct thread_data *td, |
1149 | const struct fio_file *f, enum swd_action a) | |
615555bb | 1150 | { |
615555bb BVA |
1151 | struct fio_zone_info *zb, *ze, *z; |
1152 | uint64_t swd = 0; | |
403ea795 | 1153 | uint64_t wp_swd = 0; |
615555bb | 1154 | |
923f7c1e DF |
1155 | zb = get_zone(f, f->min_zone); |
1156 | ze = get_zone(f, f->max_zone); | |
615555bb | 1157 | for (z = zb; z < ze; z++) { |
403ea795 | 1158 | if (z->has_wp) { |
11afb212 | 1159 | zone_lock(td, f, z); |
403ea795 SK |
1160 | wp_swd += z->wp - z->start; |
1161 | } | |
615555bb BVA |
1162 | swd += z->wp - z->start; |
1163 | } | |
139d8dc6 | 1164 | |
615555bb | 1165 | pthread_mutex_lock(&f->zbd_info->mutex); |
91d25131 BVA |
1166 | switch (a) { |
1167 | case CHECK_SWD: | |
1168 | assert(f->zbd_info->sectors_with_data == swd); | |
403ea795 | 1169 | assert(f->zbd_info->wp_sectors_with_data == wp_swd); |
91d25131 BVA |
1170 | break; |
1171 | case SET_SWD: | |
1172 | f->zbd_info->sectors_with_data = swd; | |
403ea795 | 1173 | f->zbd_info->wp_sectors_with_data = wp_swd; |
91d25131 BVA |
1174 | break; |
1175 | } | |
615555bb | 1176 | pthread_mutex_unlock(&f->zbd_info->mutex); |
139d8dc6 | 1177 | |
615555bb | 1178 | for (z = zb; z < ze; z++) |
403ea795 SK |
1179 | if (z->has_wp) |
1180 | zone_unlock(z); | |
91d25131 BVA |
1181 | |
1182 | return swd; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * The swd check is useful for debugging but takes too much time to leave | |
1187 | * it enabled all the time. Hence it is disabled by default. | |
1188 | */ | |
1189 | static const bool enable_check_swd = false; | |
1190 | ||
403ea795 | 1191 | /* Check whether the values of zbd_info.*sectors_with_data are correct. */ |
11afb212 | 1192 | static void zbd_check_swd(struct thread_data *td, const struct fio_file *f) |
91d25131 BVA |
1193 | { |
1194 | if (!enable_check_swd) | |
1195 | return; | |
1196 | ||
11afb212 | 1197 | zbd_process_swd(td, f, CHECK_SWD); |
615555bb BVA |
1198 | } |
1199 | ||
bfbdd35b BVA |
1200 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
1201 | { | |
91d25131 | 1202 | struct fio_zone_info *zb, *ze; |
11afb212 | 1203 | uint64_t swd; |
bfbdd35b | 1204 | |
767d1372 | 1205 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
1206 | return; |
1207 | ||
923f7c1e DF |
1208 | zb = get_zone(f, f->min_zone); |
1209 | ze = get_zone(f, f->max_zone); | |
11afb212 | 1210 | swd = zbd_process_swd(td, f, SET_SWD); |
139d8dc6 DLM |
1211 | |
1212 | dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", | |
1213 | __func__, f->file_name, swd); | |
1214 | ||
bfbdd35b BVA |
1215 | /* |
1216 | * If data verification is enabled reset the affected zones before | |
1217 | * writing any data to avoid that a zone reset has to be issued while | |
1218 | * writing data, which causes data loss. | |
1219 | */ | |
57f88219 SK |
1220 | if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING) |
1221 | zbd_reset_zones(td, f, zb, ze); | |
a7c2b6fc | 1222 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
1223 | } |
1224 | ||
cb765e41 | 1225 | /* Return random zone index for one of the open zones. */ |
6463db6c AD |
1226 | static uint32_t pick_random_zone_idx(const struct fio_file *f, |
1227 | const struct io_u *io_u) | |
1228 | { | |
139d8dc6 DLM |
1229 | return (io_u->offset - f->file_offset) * |
1230 | f->zbd_info->num_open_zones / f->io_size; | |
6463db6c AD |
1231 | } |
1232 | ||
0f77c977 SK |
1233 | static bool any_io_in_flight(void) |
1234 | { | |
1235 | struct thread_data *td; | |
1236 | int i; | |
1237 | ||
1238 | for_each_td(td, i) { | |
1239 | if (td->io_u_in_flight) | |
1240 | return true; | |
1241 | } | |
1242 | ||
1243 | return false; | |
1244 | } | |
1245 | ||
59b07544 BVA |
1246 | /* |
1247 | * Modify the offset of an I/O unit that does not refer to an open zone such | |
1248 | * that it refers to an open zone. Close an open zone and open a new zone if | |
21c0c884 SK |
1249 | * necessary. The open zone is searched across sequential zones. |
1250 | * This algorithm can only work correctly if all write pointers are | |
59b07544 BVA |
1251 | * a multiple of the fio block size. The caller must neither hold z->mutex |
1252 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
1253 | */ | |
379e5f09 BVA |
1254 | static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, |
1255 | struct io_u *io_u) | |
59b07544 | 1256 | { |
07fc3f57 | 1257 | const uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
fae3b9a0 | 1258 | struct fio_file *f = io_u->file; |
af94a8c3 | 1259 | struct zoned_block_device_info *zbdi = f->zbd_info; |
59b07544 BVA |
1260 | struct fio_zone_info *z; |
1261 | unsigned int open_zone_idx = -1; | |
1262 | uint32_t zone_idx, new_zone_idx; | |
1263 | int i; | |
b2da58c4 | 1264 | bool wait_zone_close; |
0f77c977 SK |
1265 | bool in_flight; |
1266 | bool should_retry = true; | |
59b07544 BVA |
1267 | |
1268 | assert(is_valid_offset(f, io_u->offset)); | |
1269 | ||
ea51055c | 1270 | if (zbdi->max_open_zones || td->o.job_max_open_zones) { |
59b07544 | 1271 | /* |
af94a8c3 | 1272 | * This statement accesses zbdi->open_zones[] on purpose |
59b07544 BVA |
1273 | * without locking. |
1274 | */ | |
af94a8c3 | 1275 | zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 BVA |
1276 | } else { |
1277 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
1278 | } | |
fae3b9a0 AD |
1279 | if (zone_idx < f->min_zone) |
1280 | zone_idx = f->min_zone; | |
1281 | else if (zone_idx >= f->max_zone) | |
1282 | zone_idx = f->max_zone - 1; | |
139d8dc6 DLM |
1283 | |
1284 | dprint(FD_ZBD, | |
1285 | "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", | |
59b07544 BVA |
1286 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); |
1287 | ||
1288 | /* | |
af94a8c3 | 1289 | * Since z->mutex is the outer lock and zbdi->mutex the inner |
59b07544 | 1290 | * lock it can happen that the state of the zone with index zone_idx |
af94a8c3 | 1291 | * has changed after 'z' has been assigned and before zbdi->mutex |
59b07544 BVA |
1292 | * has been obtained. Hence the loop. |
1293 | */ | |
1294 | for (;;) { | |
6463db6c AD |
1295 | uint32_t tmp_idx; |
1296 | ||
923f7c1e | 1297 | z = get_zone(f, zone_idx); |
14351148 DF |
1298 | if (z->has_wp) |
1299 | zone_lock(td, f, z); | |
139d8dc6 | 1300 | |
af94a8c3 | 1301 | pthread_mutex_lock(&zbdi->mutex); |
139d8dc6 | 1302 | |
14351148 DF |
1303 | if (z->has_wp) { |
1304 | if (z->cond != ZBD_ZONE_COND_OFFLINE && | |
139d8dc6 DLM |
1305 | zbdi->max_open_zones == 0 && |
1306 | td->o.job_max_open_zones == 0) | |
14351148 | 1307 | goto examine_zone; |
af94a8c3 | 1308 | if (zbdi->num_open_zones == 0) { |
14351148 DF |
1309 | dprint(FD_ZBD, "%s(%s): no zones are open\n", |
1310 | __func__, f->file_name); | |
1311 | goto open_other_zone; | |
1312 | } | |
59b07544 | 1313 | } |
6463db6c AD |
1314 | |
1315 | /* | |
139d8dc6 DLM |
1316 | * List of opened zones is per-device, shared across all |
1317 | * threads. Start with quasi-random candidate zone. Ignore | |
1318 | * zones which don't belong to thread's offset/size area. | |
6463db6c AD |
1319 | */ |
1320 | open_zone_idx = pick_random_zone_idx(f, io_u); | |
1a8d510a | 1321 | assert(!open_zone_idx || |
af94a8c3 | 1322 | open_zone_idx < zbdi->num_open_zones); |
6463db6c | 1323 | tmp_idx = open_zone_idx; |
139d8dc6 | 1324 | |
af94a8c3 | 1325 | for (i = 0; i < zbdi->num_open_zones; i++) { |
6463db6c AD |
1326 | uint32_t tmpz; |
1327 | ||
af94a8c3 | 1328 | if (tmp_idx >= zbdi->num_open_zones) |
6463db6c | 1329 | tmp_idx = 0; |
af94a8c3 | 1330 | tmpz = zbdi->open_zones[tmp_idx]; |
fae3b9a0 | 1331 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
6463db6c AD |
1332 | open_zone_idx = tmp_idx; |
1333 | goto found_candidate_zone; | |
1334 | } | |
1335 | ||
1336 | tmp_idx++; | |
1337 | } | |
1338 | ||
1339 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1340 | __func__, f->file_name); | |
139d8dc6 | 1341 | |
af94a8c3 | 1342 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1343 | |
14351148 DF |
1344 | if (z->has_wp) |
1345 | zone_unlock(z); | |
139d8dc6 | 1346 | |
6463db6c AD |
1347 | return NULL; |
1348 | ||
1349 | found_candidate_zone: | |
af94a8c3 | 1350 | new_zone_idx = zbdi->open_zones[open_zone_idx]; |
59b07544 BVA |
1351 | if (new_zone_idx == zone_idx) |
1352 | break; | |
1353 | zone_idx = new_zone_idx; | |
139d8dc6 | 1354 | |
af94a8c3 | 1355 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1356 | |
14351148 DF |
1357 | if (z->has_wp) |
1358 | zone_unlock(z); | |
59b07544 BVA |
1359 | } |
1360 | ||
af94a8c3 | 1361 | /* Both z->mutex and zbdi->mutex are held. */ |
59b07544 BVA |
1362 | |
1363 | examine_zone: | |
236d23a8 | 1364 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) { |
af94a8c3 | 1365 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1366 | goto out; |
1367 | } | |
b2da58c4 SK |
1368 | |
1369 | open_other_zone: | |
1370 | /* Check if number of open zones reaches one of limits. */ | |
1371 | wait_zone_close = | |
af94a8c3 | 1372 | zbdi->num_open_zones == f->max_zone - f->min_zone || |
ea51055c NC |
1373 | (zbdi->max_open_zones && |
1374 | zbdi->num_open_zones == zbdi->max_open_zones) || | |
b2da58c4 SK |
1375 | (td->o.job_max_open_zones && |
1376 | td->num_open_zones == td->o.job_max_open_zones); | |
1377 | ||
af94a8c3 | 1378 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1379 | |
1380 | /* Only z->mutex is held. */ | |
1381 | ||
b2da58c4 SK |
1382 | /* |
1383 | * When number of open zones reaches to one of limits, wait for | |
1384 | * zone close before opening a new zone. | |
1385 | */ | |
1386 | if (wait_zone_close) { | |
139d8dc6 DLM |
1387 | dprint(FD_ZBD, |
1388 | "%s(%s): quiesce to allow open zones to close\n", | |
b2da58c4 SK |
1389 | __func__, f->file_name); |
1390 | io_u_quiesce(td); | |
1391 | } | |
1392 | ||
0f77c977 | 1393 | retry: |
59b07544 | 1394 | /* Zone 'z' is full, so try to open a new zone. */ |
af94a8c3 | 1395 | for (i = f->io_size / zbdi->zone_size; i > 0; i--) { |
59b07544 | 1396 | zone_idx++; |
21c0c884 SK |
1397 | if (z->has_wp) |
1398 | zone_unlock(z); | |
59b07544 | 1399 | z++; |
ee3696bd | 1400 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1401 | /* Wrap-around. */ |
fae3b9a0 | 1402 | zone_idx = f->min_zone; |
923f7c1e | 1403 | z = get_zone(f, zone_idx); |
59b07544 | 1404 | } |
ee3696bd | 1405 | assert(is_valid_offset(f, z->start)); |
21c0c884 SK |
1406 | if (!z->has_wp) |
1407 | continue; | |
fae3b9a0 | 1408 | zone_lock(td, f, z); |
59b07544 BVA |
1409 | if (z->open) |
1410 | continue; | |
954217b9 | 1411 | if (zbd_open_zone(td, f, zone_idx)) |
59b07544 BVA |
1412 | goto out; |
1413 | } | |
1414 | ||
1415 | /* Only z->mutex is held. */ | |
1416 | ||
1417 | /* Check whether the write fits in any of the already opened zones. */ | |
af94a8c3 NC |
1418 | pthread_mutex_lock(&zbdi->mutex); |
1419 | for (i = 0; i < zbdi->num_open_zones; i++) { | |
1420 | zone_idx = zbdi->open_zones[i]; | |
fae3b9a0 AD |
1421 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1422 | continue; | |
af94a8c3 | 1423 | pthread_mutex_unlock(&zbdi->mutex); |
4d4c71e6 | 1424 | zone_unlock(z); |
59b07544 | 1425 | |
923f7c1e | 1426 | z = get_zone(f, zone_idx); |
59b07544 | 1427 | |
fae3b9a0 | 1428 | zone_lock(td, f, z); |
236d23a8 | 1429 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) |
59b07544 | 1430 | goto out; |
af94a8c3 | 1431 | pthread_mutex_lock(&zbdi->mutex); |
59b07544 | 1432 | } |
0f77c977 SK |
1433 | |
1434 | /* | |
1435 | * When any I/O is in-flight or when all I/Os in-flight get completed, | |
1436 | * the I/Os might have closed zones then retry the steps to open a zone. | |
1437 | * Before retry, call io_u_quiesce() to complete in-flight writes. | |
1438 | */ | |
1439 | in_flight = any_io_in_flight(); | |
1440 | if (in_flight || should_retry) { | |
139d8dc6 DLM |
1441 | dprint(FD_ZBD, |
1442 | "%s(%s): wait zone close and retry open zones\n", | |
0f77c977 SK |
1443 | __func__, f->file_name); |
1444 | pthread_mutex_unlock(&zbdi->mutex); | |
1445 | zone_unlock(z); | |
1446 | io_u_quiesce(td); | |
1447 | zone_lock(td, f, z); | |
1448 | should_retry = in_flight; | |
1449 | goto retry; | |
1450 | } | |
1451 | ||
af94a8c3 | 1452 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1453 | |
4d4c71e6 | 1454 | zone_unlock(z); |
139d8dc6 DLM |
1455 | |
1456 | dprint(FD_ZBD, "%s(%s): did not open another zone\n", | |
1457 | __func__, f->file_name); | |
1458 | ||
59b07544 BVA |
1459 | return NULL; |
1460 | ||
1461 | out: | |
139d8dc6 DLM |
1462 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", |
1463 | __func__, f->file_name, zone_idx); | |
1464 | ||
ee3696bd | 1465 | io_u->offset = z->start; |
21c0c884 | 1466 | assert(z->has_wp); |
8a866de7 | 1467 | assert(z->cond != ZBD_ZONE_COND_OFFLINE); |
139d8dc6 | 1468 | |
59b07544 BVA |
1469 | return z; |
1470 | } | |
1471 | ||
bfbdd35b | 1472 | /* The caller must hold z->mutex. */ |
59b07544 BVA |
1473 | static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, |
1474 | struct io_u *io_u, | |
1475 | struct fio_zone_info *z) | |
bfbdd35b BVA |
1476 | { |
1477 | const struct fio_file *f = io_u->file; | |
07fc3f57 | 1478 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; |
bfbdd35b | 1479 | |
122c3539 | 1480 | if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) { |
4d4c71e6 | 1481 | zone_unlock(z); |
59b07544 BVA |
1482 | z = zbd_convert_to_open_zone(td, io_u); |
1483 | assert(z); | |
1484 | } | |
1485 | ||
0cac10b4 | 1486 | if (z->verify_block * min_bs >= z->capacity) { |
139d8dc6 DLM |
1487 | log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n", |
1488 | f->file_name, z->verify_block, min_bs, z->capacity); | |
0cac10b4 DF |
1489 | /* |
1490 | * If the assertion below fails during a test run, adding | |
1491 | * "--experimental_verify=1" to the command line may help. | |
1492 | */ | |
1493 | assert(false); | |
1494 | } | |
139d8dc6 | 1495 | |
0cac10b4 DF |
1496 | io_u->offset = z->start + z->verify_block * min_bs; |
1497 | if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
139d8dc6 DLM |
1498 | log_err("%s: %llu + %llu >= %"PRIu64"\n", |
1499 | f->file_name, io_u->offset, io_u->buflen, | |
1500 | zbd_zone_capacity_end(z)); | |
0cac10b4 DF |
1501 | assert(false); |
1502 | } | |
1503 | z->verify_block += io_u->buflen / min_bs; | |
1504 | ||
59b07544 | 1505 | return z; |
bfbdd35b BVA |
1506 | } |
1507 | ||
1508 | /* | |
5c86fdf6 SK |
1509 | * Find another zone which has @min_bytes of readable data. Search in zones |
1510 | * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf. | |
bfbdd35b | 1511 | * |
21c0c884 SK |
1512 | * Either returns NULL or returns a zone pointer. When the zone has write |
1513 | * pointer, hold the mutex for the zone. | |
bfbdd35b BVA |
1514 | */ |
1515 | static struct fio_zone_info * | |
07fc3f57 | 1516 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes, |
bfbdd35b BVA |
1517 | struct fio_zone_info *zb, struct fio_zone_info *zl) |
1518 | { | |
fae3b9a0 | 1519 | struct fio_file *f = io_u->file; |
bfbdd35b | 1520 | struct fio_zone_info *z1, *z2; |
923f7c1e | 1521 | const struct fio_zone_info *const zf = get_zone(f, f->min_zone); |
bfbdd35b BVA |
1522 | |
1523 | /* | |
1524 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1525 | * the nearest non-empty zone in case of random I/O. | |
1526 | */ | |
1527 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1528 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1529 | if (z1->has_wp) |
1530 | zone_lock(td, f, z1); | |
5c86fdf6 | 1531 | if (z1->start + min_bytes <= z1->wp) |
bfbdd35b | 1532 | return z1; |
21c0c884 SK |
1533 | if (z1->has_wp) |
1534 | zone_unlock(z1); | |
bfbdd35b BVA |
1535 | } else if (!td_random(td)) { |
1536 | break; | |
1537 | } | |
139d8dc6 | 1538 | |
bfbdd35b | 1539 | if (td_random(td) && z2 >= zf && |
b7694961 | 1540 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1541 | if (z2->has_wp) |
1542 | zone_lock(td, f, z2); | |
5c86fdf6 | 1543 | if (z2->start + min_bytes <= z2->wp) |
bfbdd35b | 1544 | return z2; |
21c0c884 SK |
1545 | if (z2->has_wp) |
1546 | zone_unlock(z2); | |
bfbdd35b BVA |
1547 | } |
1548 | } | |
139d8dc6 DLM |
1549 | |
1550 | dprint(FD_ZBD, | |
1551 | "%s: no zone has %"PRIu64" bytes of readable data\n", | |
5c86fdf6 | 1552 | f->file_name, min_bytes); |
139d8dc6 | 1553 | |
bfbdd35b BVA |
1554 | return NULL; |
1555 | } | |
1556 | ||
b2da58c4 SK |
1557 | /** |
1558 | * zbd_end_zone_io - update zone status at command completion | |
1559 | * @io_u: I/O unit | |
1560 | * @z: zone info pointer | |
1561 | * | |
1562 | * If the write command made the zone full, close it. | |
1563 | * | |
1564 | * The caller must hold z->mutex. | |
1565 | */ | |
1566 | static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, | |
1567 | struct fio_zone_info *z) | |
1568 | { | |
1569 | const struct fio_file *f = io_u->file; | |
1570 | ||
1571 | if (io_u->ddir == DDIR_WRITE && | |
1572 | io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1573 | pthread_mutex_lock(&f->zbd_info->mutex); | |
122c3539 | 1574 | zbd_close_zone(td, f, zbd_zone_nr(f, z)); |
b2da58c4 SK |
1575 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1576 | } | |
1577 | } | |
1578 | ||
bfbdd35b | 1579 | /** |
d9ed3e63 | 1580 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1581 | * @io_u: I/O unit |
d9ed3e63 DLM |
1582 | * @success: Whether or not the I/O unit has been queued successfully |
1583 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1584 | * |
d9ed3e63 DLM |
1585 | * For write and trim operations, update the write pointer of the I/O unit |
1586 | * target zone. | |
bfbdd35b | 1587 | */ |
b2da58c4 SK |
1588 | static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, |
1589 | bool success) | |
bfbdd35b | 1590 | { |
d9ed3e63 DLM |
1591 | const struct fio_file *f = io_u->file; |
1592 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b BVA |
1593 | struct fio_zone_info *z; |
1594 | uint32_t zone_idx; | |
d9ed3e63 | 1595 | uint64_t zone_end; |
bfbdd35b | 1596 | |
5ddf46d0 | 1597 | assert(zbd_info); |
bfbdd35b | 1598 | |
d9ed3e63 | 1599 | zone_idx = zbd_zone_idx(f, io_u->offset); |
bfbdd35b | 1600 | assert(zone_idx < zbd_info->nr_zones); |
923f7c1e | 1601 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1602 | |
43bcbd5b | 1603 | assert(z->has_wp); |
d9ed3e63 | 1604 | |
bfbdd35b BVA |
1605 | if (!success) |
1606 | goto unlock; | |
d9ed3e63 DLM |
1607 | |
1608 | dprint(FD_ZBD, | |
1609 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
1610 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1611 | ||
bfbdd35b BVA |
1612 | switch (io_u->ddir) { |
1613 | case DDIR_WRITE: | |
d9ed3e63 | 1614 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
236d23a8 | 1615 | zbd_zone_capacity_end(z)); |
139d8dc6 | 1616 | |
a7c2b6fc BVA |
1617 | /* |
1618 | * z->wp > zone_end means that one or more I/O errors | |
1619 | * have occurred. | |
1620 | */ | |
139d8dc6 | 1621 | pthread_mutex_lock(&zbd_info->mutex); |
403ea795 | 1622 | if (z->wp <= zone_end) { |
a7c2b6fc | 1623 | zbd_info->sectors_with_data += zone_end - z->wp; |
403ea795 SK |
1624 | zbd_info->wp_sectors_with_data += zone_end - z->wp; |
1625 | } | |
a7c2b6fc | 1626 | pthread_mutex_unlock(&zbd_info->mutex); |
bfbdd35b BVA |
1627 | z->wp = zone_end; |
1628 | break; | |
bfbdd35b BVA |
1629 | default: |
1630 | break; | |
1631 | } | |
d9ed3e63 | 1632 | |
b2da58c4 SK |
1633 | if (q == FIO_Q_COMPLETED && !io_u->error) |
1634 | zbd_end_zone_io(td, io_u, z); | |
1635 | ||
bfbdd35b | 1636 | unlock: |
d9ed3e63 DLM |
1637 | if (!success || q != FIO_Q_QUEUED) { |
1638 | /* BUSY or COMPLETED: unlock the zone */ | |
4d4c71e6 | 1639 | zone_unlock(z); |
d9ed3e63 DLM |
1640 | io_u->zbd_put_io = NULL; |
1641 | } | |
1642 | } | |
1643 | ||
1644 | /** | |
1645 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1646 | * @io_u: I/O unit | |
1647 | */ | |
b2da58c4 | 1648 | static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) |
d9ed3e63 DLM |
1649 | { |
1650 | const struct fio_file *f = io_u->file; | |
1651 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1652 | struct fio_zone_info *z; | |
1653 | uint32_t zone_idx; | |
1654 | ||
5ddf46d0 | 1655 | assert(zbd_info); |
615555bb | 1656 | |
d9ed3e63 DLM |
1657 | zone_idx = zbd_zone_idx(f, io_u->offset); |
1658 | assert(zone_idx < zbd_info->nr_zones); | |
923f7c1e | 1659 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1660 | |
43bcbd5b | 1661 | assert(z->has_wp); |
d9ed3e63 DLM |
1662 | |
1663 | dprint(FD_ZBD, | |
1664 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
1665 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1666 | ||
b2da58c4 SK |
1667 | zbd_end_zone_io(td, io_u, z); |
1668 | ||
4d4c71e6 | 1669 | zone_unlock(z); |
11afb212 | 1670 | zbd_check_swd(td, f); |
bfbdd35b BVA |
1671 | } |
1672 | ||
9d87c646 DLM |
1673 | /* |
1674 | * Windows and MacOS do not define this. | |
1675 | */ | |
1676 | #ifndef EREMOTEIO | |
1677 | #define EREMOTEIO 121 /* POSIX value */ | |
1678 | #endif | |
1679 | ||
bfbdd35b BVA |
1680 | bool zbd_unaligned_write(int error_code) |
1681 | { | |
1682 | switch (error_code) { | |
1683 | case EIO: | |
1684 | case EREMOTEIO: | |
1685 | return true; | |
1686 | } | |
1687 | return false; | |
1688 | } | |
1689 | ||
4d37720a DLM |
1690 | /** |
1691 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1692 | * @td: FIO thread data. | |
1693 | * @io_u: FIO I/O unit. | |
1694 | * | |
1695 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1696 | * no more IO can be performed in the current zone. | |
1697 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1698 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1699 | * - For write workloads, zoneskip is applied when the zone is full. | |
1700 | * This applies only to read and write operations. | |
1701 | */ | |
1702 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1703 | { | |
1704 | struct fio_file *f = io_u->file; | |
1705 | enum fio_ddir ddir = io_u->ddir; | |
1706 | struct fio_zone_info *z; | |
1707 | uint32_t zone_idx; | |
1708 | ||
1709 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1710 | assert(td->o.zone_size); | |
5ddf46d0 | 1711 | assert(f->zbd_info); |
4d37720a | 1712 | |
236d23a8 | 1713 | zone_idx = zbd_zone_idx(f, f->last_pos[ddir]); |
923f7c1e | 1714 | z = get_zone(f, zone_idx); |
236d23a8 SK |
1715 | |
1716 | /* | |
1717 | * When the zone capacity is smaller than the zone size and the I/O is | |
1718 | * sequential write, skip to zone end if the latest position is at the | |
1719 | * zone capacity limit. | |
1720 | */ | |
139d8dc6 DLM |
1721 | if (z->capacity < f->zbd_info->zone_size && |
1722 | !td_random(td) && ddir == DDIR_WRITE && | |
236d23a8 SK |
1723 | f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { |
1724 | dprint(FD_ZBD, | |
1725 | "%s: Jump from zone capacity limit to zone end:" | |
ee5e3436 SK |
1726 | " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n", |
1727 | f->file_name, f->last_pos[ddir], | |
1728 | zbd_zone_end(z), zone_idx, z->capacity); | |
236d23a8 SK |
1729 | td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; |
1730 | f->last_pos[ddir] = zbd_zone_end(z); | |
1731 | } | |
1732 | ||
4d37720a DLM |
1733 | /* |
1734 | * zone_skip is valid only for sequential workloads. | |
1735 | */ | |
1736 | if (td_random(td) || !td->o.zone_skip) | |
1737 | return; | |
1738 | ||
1739 | /* | |
1740 | * It is time to switch to a new zone if: | |
1741 | * - zone_bytes == zone_size bytes have already been accessed | |
1742 | * - The last position reached the end of the current zone. | |
1743 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1744 | * reached the zone write pointer. | |
1745 | */ | |
4d37720a | 1746 | if (td->zone_bytes >= td->o.zone_size || |
236d23a8 | 1747 | f->last_pos[ddir] >= zbd_zone_end(z) || |
4d37720a DLM |
1748 | (ddir == DDIR_READ && |
1749 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1750 | /* | |
1751 | * Skip zones. | |
1752 | */ | |
1753 | td->zone_bytes = 0; | |
1754 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1755 | ||
1756 | /* | |
1757 | * Wrap from the beginning, if we exceed the file size | |
1758 | */ | |
1759 | if (f->file_offset >= f->real_file_size) | |
1760 | f->file_offset = get_start_offset(td, f); | |
1761 | ||
1762 | f->last_pos[ddir] = f->file_offset; | |
1763 | td->io_skip_bytes += td->o.zone_skip; | |
1764 | } | |
1765 | } | |
1766 | ||
c65057f9 | 1767 | /** |
c7d5e152 | 1768 | * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. |
c65057f9 SK |
1769 | * |
1770 | * @td: FIO thread data. | |
1771 | * @io_u: FIO I/O unit. | |
1772 | * @ddir: I/O direction before adjustment. | |
1773 | * | |
1774 | * Return adjusted I/O direction. | |
1775 | */ | |
1776 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1777 | enum fio_ddir ddir) | |
1778 | { | |
1779 | /* | |
1780 | * In case read direction is chosen for the first random I/O, fio with | |
1781 | * zonemode=zbd stops because no data can be read from zoned block | |
1782 | * devices with all empty zones. Overwrite the first I/O direction as | |
1783 | * write to make sure data to read exists. | |
1784 | */ | |
5ddf46d0 | 1785 | assert(io_u->file->zbd_info); |
731461cc | 1786 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1787 | return ddir; |
1788 | ||
1789 | if (io_u->file->zbd_info->sectors_with_data || | |
1790 | td->o.read_beyond_wp) | |
1791 | return DDIR_READ; | |
1792 | ||
1793 | return DDIR_WRITE; | |
1794 | } | |
1795 | ||
bfbdd35b BVA |
1796 | /** |
1797 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1798 | * @td: FIO thread data. | |
1799 | * @io_u: FIO I/O unit. | |
1800 | * | |
1801 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1802 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1803 | * corresponds to io_u->offset at the end of this function. | |
1804 | */ | |
1805 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1806 | { | |
b7694961 | 1807 | struct fio_file *f = io_u->file; |
af94a8c3 | 1808 | struct zoned_block_device_info *zbdi = f->zbd_info; |
bfbdd35b | 1809 | uint32_t zone_idx_b; |
de65f7b7 | 1810 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b | 1811 | uint32_t orig_len = io_u->buflen; |
07fc3f57 | 1812 | uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
bfbdd35b BVA |
1813 | uint64_t new_len; |
1814 | int64_t range; | |
1815 | ||
af94a8c3 | 1816 | assert(zbdi); |
adc6adcb | 1817 | assert(min_bs); |
bfbdd35b BVA |
1818 | assert(is_valid_offset(f, io_u->offset)); |
1819 | assert(io_u->buflen); | |
139d8dc6 | 1820 | |
bfbdd35b | 1821 | zone_idx_b = zbd_zone_idx(f, io_u->offset); |
923f7c1e | 1822 | zb = get_zone(f, zone_idx_b); |
de65f7b7 | 1823 | orig_zb = zb; |
bfbdd35b | 1824 | |
2efcf74b SK |
1825 | if (!zb->has_wp) { |
1826 | /* Accept non-write I/Os for conventional zones. */ | |
1827 | if (io_u->ddir != DDIR_WRITE) | |
1828 | return io_u_accept; | |
139d8dc6 | 1829 | |
2efcf74b SK |
1830 | /* |
1831 | * Make sure that writes to conventional zones | |
1832 | * don't cross over to any sequential zones. | |
1833 | */ | |
1834 | if (!(zb + 1)->has_wp || | |
1835 | io_u->offset + io_u->buflen <= (zb + 1)->start) | |
1836 | return io_u_accept; | |
1837 | ||
1838 | if (io_u->offset + min_bs > (zb + 1)->start) { | |
1839 | dprint(FD_IO, | |
07fc3f57 | 1840 | "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n", |
1e3d6e03 | 1841 | f->file_name, io_u->offset, |
ee5e3436 | 1842 | min_bs, (zb + 1)->start); |
139d8dc6 DLM |
1843 | io_u->offset = |
1844 | zb->start + (zb + 1)->start - io_u->offset; | |
1845 | new_len = min(io_u->buflen, | |
1846 | (zb + 1)->start - io_u->offset); | |
2efcf74b SK |
1847 | } else { |
1848 | new_len = (zb + 1)->start - io_u->offset; | |
1849 | } | |
139d8dc6 | 1850 | |
2efcf74b | 1851 | io_u->buflen = new_len / min_bs * min_bs; |
139d8dc6 | 1852 | |
bfbdd35b | 1853 | return io_u_accept; |
2efcf74b | 1854 | } |
bfbdd35b BVA |
1855 | |
1856 | /* | |
1857 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1858 | * is enabled. | |
1859 | */ | |
b7694961 | 1860 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1861 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1862 | return io_u_accept; | |
1863 | ||
11afb212 | 1864 | zbd_check_swd(td, f); |
615555bb | 1865 | |
fae3b9a0 | 1866 | zone_lock(td, f, zb); |
6f0c6085 | 1867 | |
bfbdd35b BVA |
1868 | switch (io_u->ddir) { |
1869 | case DDIR_READ: | |
0ed7d55e AR |
1870 | if (td->runstate == TD_VERIFYING && td_write(td)) { |
1871 | zb = zbd_replay_write_order(td, io_u, zb); | |
bfbdd35b BVA |
1872 | goto accept; |
1873 | } | |
139d8dc6 | 1874 | |
bfbdd35b | 1875 | /* |
de65f7b7 DLM |
1876 | * Check that there is enough written data in the zone to do an |
1877 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1878 | * the I/O. | |
bfbdd35b | 1879 | */ |
b7694961 | 1880 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1881 | zb->wp - zb->start : 0; |
de65f7b7 | 1882 | if (range < min_bs || |
ee3696bd | 1883 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
4d4c71e6 | 1884 | zone_unlock(zb); |
923f7c1e | 1885 | zl = get_zone(f, f->max_zone); |
5c86fdf6 | 1886 | zb = zbd_find_zone(td, io_u, min_bs, zb, zl); |
bfbdd35b BVA |
1887 | if (!zb) { |
1888 | dprint(FD_ZBD, | |
1889 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1890 | f->file_name, io_u->offset, | |
1891 | io_u->buflen); | |
1892 | goto eof; | |
1893 | } | |
de65f7b7 DLM |
1894 | /* |
1895 | * zbd_find_zone() returned a zone with a range of at | |
1896 | * least min_bs. | |
1897 | */ | |
ee3696bd | 1898 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1899 | assert(range >= min_bs); |
1900 | ||
1901 | if (!td_random(td)) | |
ee3696bd | 1902 | io_u->offset = zb->start; |
bfbdd35b | 1903 | } |
139d8dc6 | 1904 | |
de65f7b7 DLM |
1905 | /* |
1906 | * Make sure the I/O is within the zone valid data range while | |
1907 | * maximizing the I/O size and preserving randomness. | |
1908 | */ | |
1909 | if (range <= io_u->buflen) | |
ee3696bd | 1910 | io_u->offset = zb->start; |
de65f7b7 | 1911 | else if (td_random(td)) |
ee3696bd DLM |
1912 | io_u->offset = zb->start + |
1913 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 | 1914 | (range - io_u->buflen)) / min_bs * min_bs; |
139d8dc6 | 1915 | |
43bcbd5b SK |
1916 | /* |
1917 | * When zbd_find_zone() returns a conventional zone, | |
1918 | * we can simply accept the new i/o offset here. | |
1919 | */ | |
1920 | if (!zb->has_wp) | |
1921 | return io_u_accept; | |
139d8dc6 | 1922 | |
de65f7b7 DLM |
1923 | /* |
1924 | * Make sure the I/O does not cross over the zone wp position. | |
1925 | */ | |
1926 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1927 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1928 | new_len = new_len / min_bs * min_bs; |
1929 | if (new_len < io_u->buflen) { | |
1930 | io_u->buflen = new_len; | |
1931 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1932 | orig_len, io_u->buflen); | |
bfbdd35b | 1933 | } |
139d8dc6 | 1934 | |
ee3696bd DLM |
1935 | assert(zb->start <= io_u->offset); |
1936 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
139d8dc6 | 1937 | |
bfbdd35b | 1938 | goto accept; |
139d8dc6 | 1939 | |
bfbdd35b | 1940 | case DDIR_WRITE: |
af94a8c3 | 1941 | if (io_u->buflen > zbdi->zone_size) { |
1c74aadc DF |
1942 | td_verror(td, EINVAL, "I/O buflen exceeds zone size"); |
1943 | dprint(FD_IO, | |
ee5e3436 SK |
1944 | "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n", |
1945 | f->file_name, io_u->buflen, zbdi->zone_size); | |
bfbdd35b | 1946 | goto eof; |
1c74aadc | 1947 | } |
139d8dc6 | 1948 | |
954217b9 | 1949 | if (!zbd_open_zone(td, f, zone_idx_b)) { |
4d4c71e6 | 1950 | zone_unlock(zb); |
59b07544 | 1951 | zb = zbd_convert_to_open_zone(td, io_u); |
1c74aadc DF |
1952 | if (!zb) { |
1953 | dprint(FD_IO, "%s: can't convert to open zone", | |
1954 | f->file_name); | |
59b07544 | 1955 | goto eof; |
1c74aadc | 1956 | } |
59b07544 | 1957 | } |
139d8dc6 | 1958 | |
a7c2b6fc BVA |
1959 | /* Check whether the zone reset threshold has been exceeded */ |
1960 | if (td->o.zrf.u.f) { | |
139d8dc6 DLM |
1961 | if (zbdi->wp_sectors_with_data >= f->io_size * td->o.zrt.u.f && |
1962 | zbd_dec_and_reset_write_cnt(td, f)) | |
a7c2b6fc | 1963 | zb->reset_zone = 1; |
a7c2b6fc | 1964 | } |
139d8dc6 | 1965 | |
bfbdd35b BVA |
1966 | /* Reset the zone pointer if necessary */ |
1967 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
1968 | assert(td->o.verify == VERIFY_NONE); | |
1969 | /* | |
1970 | * Since previous write requests may have been submitted | |
1971 | * asynchronously and since we will submit the zone | |
1972 | * reset synchronously, wait until previously submitted | |
1973 | * write requests have completed before issuing a | |
1974 | * zone reset. | |
1975 | */ | |
1976 | io_u_quiesce(td); | |
1977 | zb->reset_zone = 0; | |
1978 | if (zbd_reset_zone(td, f, zb) < 0) | |
1979 | goto eof; | |
236d23a8 SK |
1980 | |
1981 | if (zb->capacity < min_bs) { | |
1c74aadc | 1982 | td_verror(td, EINVAL, "ZCAP is less min_bs"); |
07fc3f57 | 1983 | log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n", |
ee5e3436 | 1984 | zb->capacity, min_bs); |
236d23a8 SK |
1985 | goto eof; |
1986 | } | |
bfbdd35b | 1987 | } |
139d8dc6 | 1988 | |
bfbdd35b BVA |
1989 | /* Make writes occur at the write pointer */ |
1990 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 1991 | io_u->offset = zb->wp; |
bfbdd35b | 1992 | if (!is_valid_offset(f, io_u->offset)) { |
1c74aadc DF |
1993 | td_verror(td, EINVAL, "invalid WP value"); |
1994 | dprint(FD_ZBD, "%s: dropped request with offset %llu\n", | |
1995 | f->file_name, io_u->offset); | |
bfbdd35b BVA |
1996 | goto eof; |
1997 | } | |
139d8dc6 | 1998 | |
bfbdd35b BVA |
1999 | /* |
2000 | * Make sure that the buflen is a multiple of the minimal | |
2001 | * block size. Give up if shrinking would make the request too | |
2002 | * small. | |
2003 | */ | |
2004 | new_len = min((unsigned long long)io_u->buflen, | |
236d23a8 | 2005 | zbd_zone_capacity_end(zb) - io_u->offset); |
bfbdd35b BVA |
2006 | new_len = new_len / min_bs * min_bs; |
2007 | if (new_len == io_u->buflen) | |
2008 | goto accept; | |
2009 | if (new_len >= min_bs) { | |
2010 | io_u->buflen = new_len; | |
2011 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
2012 | orig_len, io_u->buflen); | |
2013 | goto accept; | |
2014 | } | |
139d8dc6 | 2015 | |
1c74aadc | 2016 | td_verror(td, EIO, "zone remainder too small"); |
07fc3f57 | 2017 | log_err("zone remainder %lld smaller than min block size %"PRIu64"\n", |
1c74aadc | 2018 | (zbd_zone_capacity_end(zb) - io_u->offset), min_bs); |
139d8dc6 | 2019 | |
bfbdd35b | 2020 | goto eof; |
139d8dc6 | 2021 | |
bfbdd35b | 2022 | case DDIR_TRIM: |
e3be810b SK |
2023 | /* Check random trim targets a non-empty zone */ |
2024 | if (!td_random(td) || zb->wp > zb->start) | |
2025 | goto accept; | |
2026 | ||
2027 | /* Find out a non-empty zone to trim */ | |
2028 | zone_unlock(zb); | |
2029 | zl = get_zone(f, f->max_zone); | |
2030 | zb = zbd_find_zone(td, io_u, 1, zb, zl); | |
2031 | if (zb) { | |
2032 | io_u->offset = zb->start; | |
2033 | dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n", | |
2034 | f->file_name, io_u->offset); | |
2035 | goto accept; | |
2036 | } | |
139d8dc6 | 2037 | |
e3be810b | 2038 | goto eof; |
139d8dc6 | 2039 | |
bfbdd35b | 2040 | case DDIR_SYNC: |
e3be810b | 2041 | /* fall-through */ |
bfbdd35b BVA |
2042 | case DDIR_DATASYNC: |
2043 | case DDIR_SYNC_FILE_RANGE: | |
2044 | case DDIR_WAIT: | |
2045 | case DDIR_LAST: | |
2046 | case DDIR_INVAL: | |
2047 | goto accept; | |
2048 | } | |
2049 | ||
2050 | assert(false); | |
2051 | ||
2052 | accept: | |
43bcbd5b | 2053 | assert(zb->has_wp); |
b7694961 | 2054 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
2055 | assert(!io_u->zbd_queue_io); |
2056 | assert(!io_u->zbd_put_io); | |
139d8dc6 | 2057 | |
d9ed3e63 DLM |
2058 | io_u->zbd_queue_io = zbd_queue_io; |
2059 | io_u->zbd_put_io = zbd_put_io; | |
139d8dc6 | 2060 | |
2ef3c1b0 DF |
2061 | /* |
2062 | * Since we return with the zone lock still held, | |
2063 | * add an annotation to let Coverity know that it | |
2064 | * is intentional. | |
2065 | */ | |
2066 | /* coverity[missing_unlock] */ | |
139d8dc6 | 2067 | |
bfbdd35b BVA |
2068 | return io_u_accept; |
2069 | ||
2070 | eof: | |
43bcbd5b | 2071 | if (zb && zb->has_wp) |
4d4c71e6 | 2072 | zone_unlock(zb); |
139d8dc6 | 2073 | |
bfbdd35b BVA |
2074 | return io_u_eof; |
2075 | } | |
fd5d733f BVA |
2076 | |
2077 | /* Return a string with ZBD statistics */ | |
2078 | char *zbd_write_status(const struct thread_stat *ts) | |
2079 | { | |
2080 | char *res; | |
2081 | ||
ee5e3436 | 2082 | if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0) |
fd5d733f BVA |
2083 | return NULL; |
2084 | return res; | |
2085 | } | |
e3be810b SK |
2086 | |
2087 | /** | |
2088 | * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim | |
2089 | * | |
2090 | * @td: FIO thread data. | |
2091 | * @io_u: FIO I/O unit. | |
2092 | * | |
2093 | * It is assumed that z->mutex is already locked. | |
2094 | * Return io_u_completed when reset zone succeeds. Return 0 when the target zone | |
2095 | * does not have write pointer. On error, return negative errno. | |
2096 | */ | |
2097 | int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u) | |
2098 | { | |
2099 | struct fio_file *f = io_u->file; | |
2100 | struct fio_zone_info *z; | |
2101 | uint32_t zone_idx; | |
2102 | int ret; | |
2103 | ||
2104 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
2105 | z = get_zone(f, zone_idx); | |
2106 | ||
2107 | if (!z->has_wp) | |
2108 | return 0; | |
2109 | ||
2110 | if (io_u->offset != z->start) { | |
139d8dc6 DLM |
2111 | log_err("Trim offset not at zone start (%lld)\n", |
2112 | io_u->offset); | |
e3be810b SK |
2113 | return -EINVAL; |
2114 | } | |
2115 | ||
2116 | ret = zbd_reset_zone((struct thread_data *)td, f, z); | |
2117 | if (ret < 0) | |
2118 | return ret; | |
2119 | ||
2120 | return io_u_completed; | |
2121 | } |