1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/vmalloc.h>
3 #include <linux/bitmap.h>
6 #define CREATE_TRACE_POINTS
10 #define pr_fmt(fmt) "null_blk: " fmt
12 static inline sector_t mb_to_sects(unsigned long mb)
14 return ((sector_t)mb * SZ_1M) >> SECTOR_SHIFT;
17 static inline unsigned int null_zone_no(struct nullb_device *dev, sector_t sect)
19 return sect >> ilog2(dev->zone_size_sects);
22 static inline void null_lock_zone_res(struct nullb_device *dev)
24 if (dev->need_zone_res_mgmt)
25 spin_lock_irq(&dev->zone_res_lock);
28 static inline void null_unlock_zone_res(struct nullb_device *dev)
30 if (dev->need_zone_res_mgmt)
31 spin_unlock_irq(&dev->zone_res_lock);
34 static inline void null_init_zone_lock(struct nullb_device *dev,
35 struct nullb_zone *zone)
37 if (!dev->memory_backed)
38 spin_lock_init(&zone->spinlock);
40 mutex_init(&zone->mutex);
43 static inline void null_lock_zone(struct nullb_device *dev,
44 struct nullb_zone *zone)
46 if (!dev->memory_backed)
47 spin_lock_irq(&zone->spinlock);
49 mutex_lock(&zone->mutex);
52 static inline void null_unlock_zone(struct nullb_device *dev,
53 struct nullb_zone *zone)
55 if (!dev->memory_backed)
56 spin_unlock_irq(&zone->spinlock);
58 mutex_unlock(&zone->mutex);
61 int null_init_zoned_dev(struct nullb_device *dev, struct request_queue *q)
63 sector_t dev_capacity_sects, zone_capacity_sects;
64 struct nullb_zone *zone;
68 if (!is_power_of_2(dev->zone_size)) {
69 pr_err("zone_size must be power-of-two\n");
72 if (dev->zone_size > dev->size) {
73 pr_err("Zone size larger than device capacity\n");
77 if (!dev->zone_capacity)
78 dev->zone_capacity = dev->zone_size;
80 if (dev->zone_capacity > dev->zone_size) {
81 pr_err("zone capacity (%lu MB) larger than zone size (%lu MB)\n",
82 dev->zone_capacity, dev->zone_size);
86 zone_capacity_sects = mb_to_sects(dev->zone_capacity);
87 dev_capacity_sects = mb_to_sects(dev->size);
88 dev->zone_size_sects = mb_to_sects(dev->zone_size);
89 dev->nr_zones = round_up(dev_capacity_sects, dev->zone_size_sects)
90 >> ilog2(dev->zone_size_sects);
92 dev->zones = kvmalloc_array(dev->nr_zones, sizeof(struct nullb_zone),
93 GFP_KERNEL | __GFP_ZERO);
97 spin_lock_init(&dev->zone_res_lock);
99 if (dev->zone_nr_conv >= dev->nr_zones) {
100 dev->zone_nr_conv = dev->nr_zones - 1;
101 pr_info("changed the number of conventional zones to %u",
105 /* Max active zones has to be < nbr of seq zones in order to be enforceable */
106 if (dev->zone_max_active >= dev->nr_zones - dev->zone_nr_conv) {
107 dev->zone_max_active = 0;
108 pr_info("zone_max_active limit disabled, limit >= zone count\n");
111 /* Max open zones has to be <= max active zones */
112 if (dev->zone_max_active && dev->zone_max_open > dev->zone_max_active) {
113 dev->zone_max_open = dev->zone_max_active;
114 pr_info("changed the maximum number of open zones to %u\n",
116 } else if (dev->zone_max_open >= dev->nr_zones - dev->zone_nr_conv) {
117 dev->zone_max_open = 0;
118 pr_info("zone_max_open limit disabled, limit >= zone count\n");
120 dev->need_zone_res_mgmt = dev->zone_max_active || dev->zone_max_open;
121 dev->imp_close_zone_no = dev->zone_nr_conv;
123 for (i = 0; i < dev->zone_nr_conv; i++) {
124 zone = &dev->zones[i];
126 null_init_zone_lock(dev, zone);
127 zone->start = sector;
128 zone->len = dev->zone_size_sects;
129 zone->capacity = zone->len;
130 zone->wp = zone->start + zone->len;
131 zone->type = BLK_ZONE_TYPE_CONVENTIONAL;
132 zone->cond = BLK_ZONE_COND_NOT_WP;
134 sector += dev->zone_size_sects;
137 for (i = dev->zone_nr_conv; i < dev->nr_zones; i++) {
138 zone = &dev->zones[i];
140 null_init_zone_lock(dev, zone);
141 zone->start = zone->wp = sector;
142 if (zone->start + dev->zone_size_sects > dev_capacity_sects)
143 zone->len = dev_capacity_sects - zone->start;
145 zone->len = dev->zone_size_sects;
147 min_t(sector_t, zone->len, zone_capacity_sects);
148 zone->type = BLK_ZONE_TYPE_SEQWRITE_REQ;
149 zone->cond = BLK_ZONE_COND_EMPTY;
151 sector += dev->zone_size_sects;
157 int null_register_zoned_dev(struct nullb *nullb)
159 struct nullb_device *dev = nullb->dev;
160 struct request_queue *q = nullb->q;
162 blk_queue_set_zoned(nullb->disk, BLK_ZONED_HM);
163 blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
164 blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
166 if (queue_is_mq(q)) {
167 int ret = blk_revalidate_disk_zones(nullb->disk, NULL);
172 blk_queue_chunk_sectors(q, dev->zone_size_sects);
173 q->nr_zones = blkdev_nr_zones(nullb->disk);
176 blk_queue_max_zone_append_sectors(q, dev->zone_size_sects);
177 blk_queue_max_open_zones(q, dev->zone_max_open);
178 blk_queue_max_active_zones(q, dev->zone_max_active);
183 void null_free_zoned_dev(struct nullb_device *dev)
189 int null_report_zones(struct gendisk *disk, sector_t sector,
190 unsigned int nr_zones, report_zones_cb cb, void *data)
192 struct nullb *nullb = disk->private_data;
193 struct nullb_device *dev = nullb->dev;
194 unsigned int first_zone, i;
195 struct nullb_zone *zone;
196 struct blk_zone blkz;
199 first_zone = null_zone_no(dev, sector);
200 if (first_zone >= dev->nr_zones)
203 nr_zones = min(nr_zones, dev->nr_zones - first_zone);
204 trace_nullb_report_zones(nullb, nr_zones);
206 memset(&blkz, 0, sizeof(struct blk_zone));
207 zone = &dev->zones[first_zone];
208 for (i = 0; i < nr_zones; i++, zone++) {
210 * Stacked DM target drivers will remap the zone information by
211 * modifying the zone information passed to the report callback.
212 * So use a local copy to avoid corruption of the device zone
215 null_lock_zone(dev, zone);
216 blkz.start = zone->start;
217 blkz.len = zone->len;
219 blkz.type = zone->type;
220 blkz.cond = zone->cond;
221 blkz.capacity = zone->capacity;
222 null_unlock_zone(dev, zone);
224 error = cb(&blkz, i, data);
233 * This is called in the case of memory backing from null_process_cmd()
234 * with the target zone already locked.
236 size_t null_zone_valid_read_len(struct nullb *nullb,
237 sector_t sector, unsigned int len)
239 struct nullb_device *dev = nullb->dev;
240 struct nullb_zone *zone = &dev->zones[null_zone_no(dev, sector)];
241 unsigned int nr_sectors = len >> SECTOR_SHIFT;
243 /* Read must be below the write pointer position */
244 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL ||
245 sector + nr_sectors <= zone->wp)
248 if (sector > zone->wp)
251 return (zone->wp - sector) << SECTOR_SHIFT;
254 static blk_status_t __null_close_zone(struct nullb_device *dev,
255 struct nullb_zone *zone)
257 switch (zone->cond) {
258 case BLK_ZONE_COND_CLOSED:
259 /* close operation on closed is not an error */
261 case BLK_ZONE_COND_IMP_OPEN:
262 dev->nr_zones_imp_open--;
264 case BLK_ZONE_COND_EXP_OPEN:
265 dev->nr_zones_exp_open--;
267 case BLK_ZONE_COND_EMPTY:
268 case BLK_ZONE_COND_FULL:
270 return BLK_STS_IOERR;
273 if (zone->wp == zone->start) {
274 zone->cond = BLK_ZONE_COND_EMPTY;
276 zone->cond = BLK_ZONE_COND_CLOSED;
277 dev->nr_zones_closed++;
283 static void null_close_imp_open_zone(struct nullb_device *dev)
285 struct nullb_zone *zone;
288 zno = dev->imp_close_zone_no;
289 if (zno >= dev->nr_zones)
290 zno = dev->zone_nr_conv;
292 for (i = dev->zone_nr_conv; i < dev->nr_zones; i++) {
293 zone = &dev->zones[zno];
295 if (zno >= dev->nr_zones)
296 zno = dev->zone_nr_conv;
298 if (zone->cond == BLK_ZONE_COND_IMP_OPEN) {
299 __null_close_zone(dev, zone);
300 dev->imp_close_zone_no = zno;
306 static blk_status_t null_check_active(struct nullb_device *dev)
308 if (!dev->zone_max_active)
311 if (dev->nr_zones_exp_open + dev->nr_zones_imp_open +
312 dev->nr_zones_closed < dev->zone_max_active)
315 return BLK_STS_ZONE_ACTIVE_RESOURCE;
318 static blk_status_t null_check_open(struct nullb_device *dev)
320 if (!dev->zone_max_open)
323 if (dev->nr_zones_exp_open + dev->nr_zones_imp_open < dev->zone_max_open)
326 if (dev->nr_zones_imp_open) {
327 if (null_check_active(dev) == BLK_STS_OK) {
328 null_close_imp_open_zone(dev);
333 return BLK_STS_ZONE_OPEN_RESOURCE;
337 * This function matches the manage open zone resources function in the ZBC standard,
338 * with the addition of max active zones support (added in the ZNS standard).
340 * The function determines if a zone can transition to implicit open or explicit open,
341 * while maintaining the max open zone (and max active zone) limit(s). It may close an
342 * implicit open zone in order to make additional zone resources available.
344 * ZBC states that an implicit open zone shall be closed only if there is not
345 * room within the open limit. However, with the addition of an active limit,
346 * it is not certain that closing an implicit open zone will allow a new zone
347 * to be opened, since we might already be at the active limit capacity.
349 static blk_status_t null_check_zone_resources(struct nullb_device *dev,
350 struct nullb_zone *zone)
354 switch (zone->cond) {
355 case BLK_ZONE_COND_EMPTY:
356 ret = null_check_active(dev);
357 if (ret != BLK_STS_OK)
360 case BLK_ZONE_COND_CLOSED:
361 return null_check_open(dev);
363 /* Should never be called for other states */
365 return BLK_STS_IOERR;
369 static blk_status_t null_zone_write(struct nullb_cmd *cmd, sector_t sector,
370 unsigned int nr_sectors, bool append)
372 struct nullb_device *dev = cmd->nq->dev;
373 unsigned int zno = null_zone_no(dev, sector);
374 struct nullb_zone *zone = &dev->zones[zno];
377 trace_nullb_zone_op(cmd, zno, zone->cond);
379 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL) {
381 return BLK_STS_IOERR;
382 return null_process_cmd(cmd, REQ_OP_WRITE, sector, nr_sectors);
385 null_lock_zone(dev, zone);
387 if (zone->cond == BLK_ZONE_COND_FULL) {
388 /* Cannot write to a full zone */
394 * Regular writes must be at the write pointer position.
395 * Zone append writes are automatically issued at the write
396 * pointer and the position returned using the request or BIO
402 cmd->bio->bi_iter.bi_sector = sector;
404 cmd->rq->__sector = sector;
405 } else if (sector != zone->wp) {
410 if (zone->wp + nr_sectors > zone->start + zone->capacity) {
415 if (zone->cond == BLK_ZONE_COND_CLOSED ||
416 zone->cond == BLK_ZONE_COND_EMPTY) {
417 null_lock_zone_res(dev);
419 ret = null_check_zone_resources(dev, zone);
420 if (ret != BLK_STS_OK) {
421 null_unlock_zone_res(dev);
424 if (zone->cond == BLK_ZONE_COND_CLOSED) {
425 dev->nr_zones_closed--;
426 dev->nr_zones_imp_open++;
427 } else if (zone->cond == BLK_ZONE_COND_EMPTY) {
428 dev->nr_zones_imp_open++;
431 if (zone->cond != BLK_ZONE_COND_EXP_OPEN)
432 zone->cond = BLK_ZONE_COND_IMP_OPEN;
434 null_unlock_zone_res(dev);
437 ret = null_process_cmd(cmd, REQ_OP_WRITE, sector, nr_sectors);
438 if (ret != BLK_STS_OK)
441 zone->wp += nr_sectors;
442 if (zone->wp == zone->start + zone->capacity) {
443 null_lock_zone_res(dev);
444 if (zone->cond == BLK_ZONE_COND_EXP_OPEN)
445 dev->nr_zones_exp_open--;
446 else if (zone->cond == BLK_ZONE_COND_IMP_OPEN)
447 dev->nr_zones_imp_open--;
448 zone->cond = BLK_ZONE_COND_FULL;
449 null_unlock_zone_res(dev);
455 null_unlock_zone(dev, zone);
460 static blk_status_t null_open_zone(struct nullb_device *dev,
461 struct nullb_zone *zone)
463 blk_status_t ret = BLK_STS_OK;
465 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
466 return BLK_STS_IOERR;
468 null_lock_zone_res(dev);
470 switch (zone->cond) {
471 case BLK_ZONE_COND_EXP_OPEN:
472 /* open operation on exp open is not an error */
474 case BLK_ZONE_COND_EMPTY:
475 ret = null_check_zone_resources(dev, zone);
476 if (ret != BLK_STS_OK)
479 case BLK_ZONE_COND_IMP_OPEN:
480 dev->nr_zones_imp_open--;
482 case BLK_ZONE_COND_CLOSED:
483 ret = null_check_zone_resources(dev, zone);
484 if (ret != BLK_STS_OK)
486 dev->nr_zones_closed--;
488 case BLK_ZONE_COND_FULL:
494 zone->cond = BLK_ZONE_COND_EXP_OPEN;
495 dev->nr_zones_exp_open++;
498 null_unlock_zone_res(dev);
503 static blk_status_t null_close_zone(struct nullb_device *dev,
504 struct nullb_zone *zone)
508 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
509 return BLK_STS_IOERR;
511 null_lock_zone_res(dev);
512 ret = __null_close_zone(dev, zone);
513 null_unlock_zone_res(dev);
518 static blk_status_t null_finish_zone(struct nullb_device *dev,
519 struct nullb_zone *zone)
521 blk_status_t ret = BLK_STS_OK;
523 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
524 return BLK_STS_IOERR;
526 null_lock_zone_res(dev);
528 switch (zone->cond) {
529 case BLK_ZONE_COND_FULL:
530 /* finish operation on full is not an error */
532 case BLK_ZONE_COND_EMPTY:
533 ret = null_check_zone_resources(dev, zone);
534 if (ret != BLK_STS_OK)
537 case BLK_ZONE_COND_IMP_OPEN:
538 dev->nr_zones_imp_open--;
540 case BLK_ZONE_COND_EXP_OPEN:
541 dev->nr_zones_exp_open--;
543 case BLK_ZONE_COND_CLOSED:
544 ret = null_check_zone_resources(dev, zone);
545 if (ret != BLK_STS_OK)
547 dev->nr_zones_closed--;
554 zone->cond = BLK_ZONE_COND_FULL;
555 zone->wp = zone->start + zone->len;
558 null_unlock_zone_res(dev);
563 static blk_status_t null_reset_zone(struct nullb_device *dev,
564 struct nullb_zone *zone)
566 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
567 return BLK_STS_IOERR;
569 null_lock_zone_res(dev);
571 switch (zone->cond) {
572 case BLK_ZONE_COND_EMPTY:
573 /* reset operation on empty is not an error */
574 null_unlock_zone_res(dev);
576 case BLK_ZONE_COND_IMP_OPEN:
577 dev->nr_zones_imp_open--;
579 case BLK_ZONE_COND_EXP_OPEN:
580 dev->nr_zones_exp_open--;
582 case BLK_ZONE_COND_CLOSED:
583 dev->nr_zones_closed--;
585 case BLK_ZONE_COND_FULL:
588 null_unlock_zone_res(dev);
589 return BLK_STS_IOERR;
592 zone->cond = BLK_ZONE_COND_EMPTY;
593 zone->wp = zone->start;
595 null_unlock_zone_res(dev);
597 if (dev->memory_backed)
598 return null_handle_discard(dev, zone->start, zone->len);
603 static blk_status_t null_zone_mgmt(struct nullb_cmd *cmd, enum req_opf op,
606 struct nullb_device *dev = cmd->nq->dev;
607 unsigned int zone_no;
608 struct nullb_zone *zone;
612 if (op == REQ_OP_ZONE_RESET_ALL) {
613 for (i = dev->zone_nr_conv; i < dev->nr_zones; i++) {
614 zone = &dev->zones[i];
615 null_lock_zone(dev, zone);
616 if (zone->cond != BLK_ZONE_COND_EMPTY) {
617 null_reset_zone(dev, zone);
618 trace_nullb_zone_op(cmd, i, zone->cond);
620 null_unlock_zone(dev, zone);
625 zone_no = null_zone_no(dev, sector);
626 zone = &dev->zones[zone_no];
628 null_lock_zone(dev, zone);
631 case REQ_OP_ZONE_RESET:
632 ret = null_reset_zone(dev, zone);
634 case REQ_OP_ZONE_OPEN:
635 ret = null_open_zone(dev, zone);
637 case REQ_OP_ZONE_CLOSE:
638 ret = null_close_zone(dev, zone);
640 case REQ_OP_ZONE_FINISH:
641 ret = null_finish_zone(dev, zone);
644 ret = BLK_STS_NOTSUPP;
648 if (ret == BLK_STS_OK)
649 trace_nullb_zone_op(cmd, zone_no, zone->cond);
651 null_unlock_zone(dev, zone);
656 blk_status_t null_process_zoned_cmd(struct nullb_cmd *cmd, enum req_opf op,
657 sector_t sector, sector_t nr_sectors)
659 struct nullb_device *dev;
660 struct nullb_zone *zone;
665 return null_zone_write(cmd, sector, nr_sectors, false);
666 case REQ_OP_ZONE_APPEND:
667 return null_zone_write(cmd, sector, nr_sectors, true);
668 case REQ_OP_ZONE_RESET:
669 case REQ_OP_ZONE_RESET_ALL:
670 case REQ_OP_ZONE_OPEN:
671 case REQ_OP_ZONE_CLOSE:
672 case REQ_OP_ZONE_FINISH:
673 return null_zone_mgmt(cmd, op, sector);
676 zone = &dev->zones[null_zone_no(dev, sector)];
678 null_lock_zone(dev, zone);
679 sts = null_process_cmd(cmd, op, sector, nr_sectors);
680 null_unlock_zone(dev, zone);