xref: /openbmc/linux/fs/btrfs/zoned.c (revision b9dd2add)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/slab.h>
4 #include <linux/blkdev.h>
5 #include "ctree.h"
6 #include "volumes.h"
7 #include "zoned.h"
8 #include "rcu-string.h"
9 
10 /* Maximum number of zones to report per blkdev_report_zones() call */
11 #define BTRFS_REPORT_NR_ZONES   4096
12 
13 /* Number of superblock log zones */
14 #define BTRFS_NR_SB_LOG_ZONES 2
15 
16 static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data)
17 {
18 	struct blk_zone *zones = data;
19 
20 	memcpy(&zones[idx], zone, sizeof(*zone));
21 
22 	return 0;
23 }
24 
25 static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones,
26 			    u64 *wp_ret)
27 {
28 	bool empty[BTRFS_NR_SB_LOG_ZONES];
29 	bool full[BTRFS_NR_SB_LOG_ZONES];
30 	sector_t sector;
31 
32 	ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL &&
33 	       zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL);
34 
35 	empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY);
36 	empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY);
37 	full[0] = (zones[0].cond == BLK_ZONE_COND_FULL);
38 	full[1] = (zones[1].cond == BLK_ZONE_COND_FULL);
39 
40 	/*
41 	 * Possible states of log buffer zones
42 	 *
43 	 *           Empty[0]  In use[0]  Full[0]
44 	 * Empty[1]         *          x        0
45 	 * In use[1]        0          x        0
46 	 * Full[1]          1          1        C
47 	 *
48 	 * Log position:
49 	 *   *: Special case, no superblock is written
50 	 *   0: Use write pointer of zones[0]
51 	 *   1: Use write pointer of zones[1]
52 	 *   C: Compare super blcoks from zones[0] and zones[1], use the latest
53 	 *      one determined by generation
54 	 *   x: Invalid state
55 	 */
56 
57 	if (empty[0] && empty[1]) {
58 		/* Special case to distinguish no superblock to read */
59 		*wp_ret = zones[0].start << SECTOR_SHIFT;
60 		return -ENOENT;
61 	} else if (full[0] && full[1]) {
62 		/* Compare two super blocks */
63 		struct address_space *mapping = bdev->bd_inode->i_mapping;
64 		struct page *page[BTRFS_NR_SB_LOG_ZONES];
65 		struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES];
66 		int i;
67 
68 		for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
69 			u64 bytenr;
70 
71 			bytenr = ((zones[i].start + zones[i].len)
72 				   << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE;
73 
74 			page[i] = read_cache_page_gfp(mapping,
75 					bytenr >> PAGE_SHIFT, GFP_NOFS);
76 			if (IS_ERR(page[i])) {
77 				if (i == 1)
78 					btrfs_release_disk_super(super[0]);
79 				return PTR_ERR(page[i]);
80 			}
81 			super[i] = page_address(page[i]);
82 		}
83 
84 		if (super[0]->generation > super[1]->generation)
85 			sector = zones[1].start;
86 		else
87 			sector = zones[0].start;
88 
89 		for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++)
90 			btrfs_release_disk_super(super[i]);
91 	} else if (!full[0] && (empty[1] || full[1])) {
92 		sector = zones[0].wp;
93 	} else if (full[0]) {
94 		sector = zones[1].wp;
95 	} else {
96 		return -EUCLEAN;
97 	}
98 	*wp_ret = sector << SECTOR_SHIFT;
99 	return 0;
100 }
101 
102 /*
103  * The following zones are reserved as the circular buffer on ZONED btrfs.
104  *  - The primary superblock: zones 0 and 1
105  *  - The first copy: zones 16 and 17
106  *  - The second copy: zones 1024 or zone at 256GB which is minimum, and
107  *                     the following one
108  */
109 static inline u32 sb_zone_number(int shift, int mirror)
110 {
111 	ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX);
112 
113 	switch (mirror) {
114 	case 0: return 0;
115 	case 1: return 16;
116 	case 2: return min_t(u64, btrfs_sb_offset(mirror) >> shift, 1024);
117 	}
118 
119 	return 0;
120 }
121 
122 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
123 			       struct blk_zone *zones, unsigned int *nr_zones)
124 {
125 	int ret;
126 
127 	if (!*nr_zones)
128 		return 0;
129 
130 	ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
131 				  copy_zone_info_cb, zones);
132 	if (ret < 0) {
133 		btrfs_err_in_rcu(device->fs_info,
134 				 "zoned: failed to read zone %llu on %s (devid %llu)",
135 				 pos, rcu_str_deref(device->name),
136 				 device->devid);
137 		return ret;
138 	}
139 	*nr_zones = ret;
140 	if (!ret)
141 		return -EIO;
142 
143 	return 0;
144 }
145 
146 int btrfs_get_dev_zone_info(struct btrfs_device *device)
147 {
148 	struct btrfs_zoned_device_info *zone_info = NULL;
149 	struct block_device *bdev = device->bdev;
150 	struct request_queue *queue = bdev_get_queue(bdev);
151 	sector_t nr_sectors;
152 	sector_t sector = 0;
153 	struct blk_zone *zones = NULL;
154 	unsigned int i, nreported = 0, nr_zones;
155 	unsigned int zone_sectors;
156 	int ret;
157 
158 	if (!bdev_is_zoned(bdev))
159 		return 0;
160 
161 	if (device->zone_info)
162 		return 0;
163 
164 	zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL);
165 	if (!zone_info)
166 		return -ENOMEM;
167 
168 	nr_sectors = bdev_nr_sectors(bdev);
169 	zone_sectors = bdev_zone_sectors(bdev);
170 	/* Check if it's power of 2 (see is_power_of_2) */
171 	ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0);
172 	zone_info->zone_size = zone_sectors << SECTOR_SHIFT;
173 	zone_info->zone_size_shift = ilog2(zone_info->zone_size);
174 	zone_info->max_zone_append_size =
175 		(u64)queue_max_zone_append_sectors(queue) << SECTOR_SHIFT;
176 	zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors);
177 	if (!IS_ALIGNED(nr_sectors, zone_sectors))
178 		zone_info->nr_zones++;
179 
180 	zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
181 	if (!zone_info->seq_zones) {
182 		ret = -ENOMEM;
183 		goto out;
184 	}
185 
186 	zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
187 	if (!zone_info->empty_zones) {
188 		ret = -ENOMEM;
189 		goto out;
190 	}
191 
192 	zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
193 	if (!zones) {
194 		ret = -ENOMEM;
195 		goto out;
196 	}
197 
198 	/* Get zones type */
199 	while (sector < nr_sectors) {
200 		nr_zones = BTRFS_REPORT_NR_ZONES;
201 		ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
202 					  &nr_zones);
203 		if (ret)
204 			goto out;
205 
206 		for (i = 0; i < nr_zones; i++) {
207 			if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
208 				__set_bit(nreported, zone_info->seq_zones);
209 			if (zones[i].cond == BLK_ZONE_COND_EMPTY)
210 				__set_bit(nreported, zone_info->empty_zones);
211 			nreported++;
212 		}
213 		sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
214 	}
215 
216 	if (nreported != zone_info->nr_zones) {
217 		btrfs_err_in_rcu(device->fs_info,
218 				 "inconsistent number of zones on %s (%u/%u)",
219 				 rcu_str_deref(device->name), nreported,
220 				 zone_info->nr_zones);
221 		ret = -EIO;
222 		goto out;
223 	}
224 
225 	/* Validate superblock log */
226 	nr_zones = BTRFS_NR_SB_LOG_ZONES;
227 	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
228 		u32 sb_zone;
229 		u64 sb_wp;
230 		int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
231 
232 		sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
233 		if (sb_zone + 1 >= zone_info->nr_zones)
234 			continue;
235 
236 		sector = sb_zone << (zone_info->zone_size_shift - SECTOR_SHIFT);
237 		ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT,
238 					  &zone_info->sb_zones[sb_pos],
239 					  &nr_zones);
240 		if (ret)
241 			goto out;
242 
243 		if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
244 			btrfs_err_in_rcu(device->fs_info,
245 	"zoned: failed to read super block log zone info at devid %llu zone %u",
246 					 device->devid, sb_zone);
247 			ret = -EUCLEAN;
248 			goto out;
249 		}
250 
251 		/*
252 		 * If zones[0] is conventional, always use the beggining of the
253 		 * zone to record superblock. No need to validate in that case.
254 		 */
255 		if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
256 		    BLK_ZONE_TYPE_CONVENTIONAL)
257 			continue;
258 
259 		ret = sb_write_pointer(device->bdev,
260 				       &zone_info->sb_zones[sb_pos], &sb_wp);
261 		if (ret != -ENOENT && ret) {
262 			btrfs_err_in_rcu(device->fs_info,
263 			"zoned: super block log zone corrupted devid %llu zone %u",
264 					 device->devid, sb_zone);
265 			ret = -EUCLEAN;
266 			goto out;
267 		}
268 	}
269 
270 
271 	kfree(zones);
272 
273 	device->zone_info = zone_info;
274 
275 	/* device->fs_info is not safe to use for printing messages */
276 	btrfs_info_in_rcu(NULL,
277 			"host-%s zoned block device %s, %u zones of %llu bytes",
278 			bdev_zoned_model(bdev) == BLK_ZONED_HM ? "managed" : "aware",
279 			rcu_str_deref(device->name), zone_info->nr_zones,
280 			zone_info->zone_size);
281 
282 	return 0;
283 
284 out:
285 	kfree(zones);
286 	bitmap_free(zone_info->empty_zones);
287 	bitmap_free(zone_info->seq_zones);
288 	kfree(zone_info);
289 
290 	return ret;
291 }
292 
293 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
294 {
295 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
296 
297 	if (!zone_info)
298 		return;
299 
300 	bitmap_free(zone_info->seq_zones);
301 	bitmap_free(zone_info->empty_zones);
302 	kfree(zone_info);
303 	device->zone_info = NULL;
304 }
305 
306 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
307 		       struct blk_zone *zone)
308 {
309 	unsigned int nr_zones = 1;
310 	int ret;
311 
312 	ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
313 	if (ret != 0 || !nr_zones)
314 		return ret ? ret : -EIO;
315 
316 	return 0;
317 }
318 
319 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
320 {
321 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
322 	struct btrfs_device *device;
323 	u64 zoned_devices = 0;
324 	u64 nr_devices = 0;
325 	u64 zone_size = 0;
326 	u64 max_zone_append_size = 0;
327 	const bool incompat_zoned = btrfs_is_zoned(fs_info);
328 	int ret = 0;
329 
330 	/* Count zoned devices */
331 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
332 		enum blk_zoned_model model;
333 
334 		if (!device->bdev)
335 			continue;
336 
337 		model = bdev_zoned_model(device->bdev);
338 		if (model == BLK_ZONED_HM ||
339 		    (model == BLK_ZONED_HA && incompat_zoned)) {
340 			struct btrfs_zoned_device_info *zone_info;
341 
342 			zone_info = device->zone_info;
343 			zoned_devices++;
344 			if (!zone_size) {
345 				zone_size = zone_info->zone_size;
346 			} else if (zone_info->zone_size != zone_size) {
347 				btrfs_err(fs_info,
348 		"zoned: unequal block device zone sizes: have %llu found %llu",
349 					  device->zone_info->zone_size,
350 					  zone_size);
351 				ret = -EINVAL;
352 				goto out;
353 			}
354 			if (!max_zone_append_size ||
355 			    (zone_info->max_zone_append_size &&
356 			     zone_info->max_zone_append_size < max_zone_append_size))
357 				max_zone_append_size =
358 					zone_info->max_zone_append_size;
359 		}
360 		nr_devices++;
361 	}
362 
363 	if (!zoned_devices && !incompat_zoned)
364 		goto out;
365 
366 	if (!zoned_devices && incompat_zoned) {
367 		/* No zoned block device found on ZONED filesystem */
368 		btrfs_err(fs_info,
369 			  "zoned: no zoned devices found on a zoned filesystem");
370 		ret = -EINVAL;
371 		goto out;
372 	}
373 
374 	if (zoned_devices && !incompat_zoned) {
375 		btrfs_err(fs_info,
376 			  "zoned: mode not enabled but zoned device found");
377 		ret = -EINVAL;
378 		goto out;
379 	}
380 
381 	if (zoned_devices != nr_devices) {
382 		btrfs_err(fs_info,
383 			  "zoned: cannot mix zoned and regular devices");
384 		ret = -EINVAL;
385 		goto out;
386 	}
387 
388 	/*
389 	 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
390 	 * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size,
391 	 * check the alignment here.
392 	 */
393 	if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
394 		btrfs_err(fs_info,
395 			  "zoned: zone size %llu not aligned to stripe %u",
396 			  zone_size, BTRFS_STRIPE_LEN);
397 		ret = -EINVAL;
398 		goto out;
399 	}
400 
401 	if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
402 		btrfs_err(fs_info, "zoned: mixed block groups not supported");
403 		ret = -EINVAL;
404 		goto out;
405 	}
406 
407 	fs_info->zone_size = zone_size;
408 	fs_info->max_zone_append_size = max_zone_append_size;
409 
410 	btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
411 out:
412 	return ret;
413 }
414 
415 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
416 {
417 	if (!btrfs_is_zoned(info))
418 		return 0;
419 
420 	/*
421 	 * Space cache writing is not COWed. Disable that to avoid write errors
422 	 * in sequential zones.
423 	 */
424 	if (btrfs_test_opt(info, SPACE_CACHE)) {
425 		btrfs_err(info, "zoned: space cache v1 is not supported");
426 		return -EINVAL;
427 	}
428 
429 	if (btrfs_test_opt(info, NODATACOW)) {
430 		btrfs_err(info, "zoned: NODATACOW not supported");
431 		return -EINVAL;
432 	}
433 
434 	return 0;
435 }
436 
437 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
438 			   int rw, u64 *bytenr_ret)
439 {
440 	u64 wp;
441 	int ret;
442 
443 	if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
444 		*bytenr_ret = zones[0].start << SECTOR_SHIFT;
445 		return 0;
446 	}
447 
448 	ret = sb_write_pointer(bdev, zones, &wp);
449 	if (ret != -ENOENT && ret < 0)
450 		return ret;
451 
452 	if (rw == WRITE) {
453 		struct blk_zone *reset = NULL;
454 
455 		if (wp == zones[0].start << SECTOR_SHIFT)
456 			reset = &zones[0];
457 		else if (wp == zones[1].start << SECTOR_SHIFT)
458 			reset = &zones[1];
459 
460 		if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
461 			ASSERT(reset->cond == BLK_ZONE_COND_FULL);
462 
463 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
464 					       reset->start, reset->len,
465 					       GFP_NOFS);
466 			if (ret)
467 				return ret;
468 
469 			reset->cond = BLK_ZONE_COND_EMPTY;
470 			reset->wp = reset->start;
471 		}
472 	} else if (ret != -ENOENT) {
473 		/* For READ, we want the precious one */
474 		if (wp == zones[0].start << SECTOR_SHIFT)
475 			wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT;
476 		wp -= BTRFS_SUPER_INFO_SIZE;
477 	}
478 
479 	*bytenr_ret = wp;
480 	return 0;
481 
482 }
483 
484 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
485 			       u64 *bytenr_ret)
486 {
487 	struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
488 	unsigned int zone_sectors;
489 	u32 sb_zone;
490 	int ret;
491 	u64 zone_size;
492 	u8 zone_sectors_shift;
493 	sector_t nr_sectors;
494 	u32 nr_zones;
495 
496 	if (!bdev_is_zoned(bdev)) {
497 		*bytenr_ret = btrfs_sb_offset(mirror);
498 		return 0;
499 	}
500 
501 	ASSERT(rw == READ || rw == WRITE);
502 
503 	zone_sectors = bdev_zone_sectors(bdev);
504 	if (!is_power_of_2(zone_sectors))
505 		return -EINVAL;
506 	zone_size = zone_sectors << SECTOR_SHIFT;
507 	zone_sectors_shift = ilog2(zone_sectors);
508 	nr_sectors = bdev_nr_sectors(bdev);
509 	nr_zones = nr_sectors >> zone_sectors_shift;
510 
511 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
512 	if (sb_zone + 1 >= nr_zones)
513 		return -ENOENT;
514 
515 	ret = blkdev_report_zones(bdev, sb_zone << zone_sectors_shift,
516 				  BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
517 				  zones);
518 	if (ret < 0)
519 		return ret;
520 	if (ret != BTRFS_NR_SB_LOG_ZONES)
521 		return -EIO;
522 
523 	return sb_log_location(bdev, zones, rw, bytenr_ret);
524 }
525 
526 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
527 			  u64 *bytenr_ret)
528 {
529 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
530 	u32 zone_num;
531 
532 	if (!zinfo) {
533 		*bytenr_ret = btrfs_sb_offset(mirror);
534 		return 0;
535 	}
536 
537 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
538 	if (zone_num + 1 >= zinfo->nr_zones)
539 		return -ENOENT;
540 
541 	return sb_log_location(device->bdev,
542 			       &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
543 			       rw, bytenr_ret);
544 }
545 
546 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
547 				  int mirror)
548 {
549 	u32 zone_num;
550 
551 	if (!zinfo)
552 		return false;
553 
554 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
555 	if (zone_num + 1 >= zinfo->nr_zones)
556 		return false;
557 
558 	if (!test_bit(zone_num, zinfo->seq_zones))
559 		return false;
560 
561 	return true;
562 }
563 
564 void btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
565 {
566 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
567 	struct blk_zone *zone;
568 
569 	if (!is_sb_log_zone(zinfo, mirror))
570 		return;
571 
572 	zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
573 	if (zone->cond != BLK_ZONE_COND_FULL) {
574 		if (zone->cond == BLK_ZONE_COND_EMPTY)
575 			zone->cond = BLK_ZONE_COND_IMP_OPEN;
576 
577 		zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
578 
579 		if (zone->wp == zone->start + zone->len)
580 			zone->cond = BLK_ZONE_COND_FULL;
581 
582 		return;
583 	}
584 
585 	zone++;
586 	ASSERT(zone->cond != BLK_ZONE_COND_FULL);
587 	if (zone->cond == BLK_ZONE_COND_EMPTY)
588 		zone->cond = BLK_ZONE_COND_IMP_OPEN;
589 
590 	zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
591 
592 	if (zone->wp == zone->start + zone->len)
593 		zone->cond = BLK_ZONE_COND_FULL;
594 }
595 
596 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
597 {
598 	sector_t zone_sectors;
599 	sector_t nr_sectors;
600 	u8 zone_sectors_shift;
601 	u32 sb_zone;
602 	u32 nr_zones;
603 
604 	zone_sectors = bdev_zone_sectors(bdev);
605 	zone_sectors_shift = ilog2(zone_sectors);
606 	nr_sectors = bdev_nr_sectors(bdev);
607 	nr_zones = nr_sectors >> zone_sectors_shift;
608 
609 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
610 	if (sb_zone + 1 >= nr_zones)
611 		return -ENOENT;
612 
613 	return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
614 				sb_zone << zone_sectors_shift,
615 				zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
616 }
617