1/*
  2 * Copyright (C) 2020 Red Hat GmbH
  3 *
  4 * This file is released under the GPL.
  5 *
  6 * Device-mapper target to emulate smaller logical block
  7 * size on backing devices exposing (natively) larger ones.
  8 *
  9 * E.g. 512 byte sector emulation on 4K native disks.
 10 */
 11
 12#include "dm.h"
 13#include <linux/module.h>
 14#include <linux/workqueue.h>
 15#include <linux/dm-bufio.h>
 16
 17#define DM_MSG_PREFIX "ebs"
 18
 19static void ebs_dtr(struct dm_target *ti);
 20
 21/* Emulated block size context. */
 22struct ebs_c {
 23	struct dm_dev *dev;		/* Underlying device to emulate block size on. */
 24	struct dm_bufio_client *bufio;	/* Use dm-bufio for read and read-modify-write processing. */
 25	struct workqueue_struct *wq;	/* Workqueue for ^ processing of bios. */
 26	struct work_struct ws;		/* Work item used for ^. */
 27	struct bio_list bios_in;	/* Worker bios input list. */
 28	spinlock_t lock;		/* Guard bios input list above. */
 29	sector_t start;			/* <start> table line argument, see ebs_ctr below. */
 30	unsigned int e_bs;		/* Emulated block size in sectors exposed to upper layer. */
 31	unsigned int u_bs;		/* Underlying block size in sectors retrieved from/set on lower layer device. */
 32	unsigned char block_shift;	/* bitshift sectors -> blocks used in dm-bufio API. */
 33	bool u_bs_set:1;		/* Flag to indicate underlying block size is set on table line. */
 34};
 35
 36static inline sector_t __sector_to_block(struct ebs_c *ec, sector_t sector)
 37{
 38	return sector >> ec->block_shift;
 39}
 40
 41static inline sector_t __block_mod(sector_t sector, unsigned int bs)
 42{
 43	return sector & (bs - 1);
 44}
 45
 46/* Return number of blocks for a bio, accounting for misalignment of start and end sectors. */
 47static inline unsigned int __nr_blocks(struct ebs_c *ec, struct bio *bio)
 48{
 49	sector_t end_sector = __block_mod(bio->bi_iter.bi_sector, ec->u_bs) + bio_sectors(bio);
 50
 51	return __sector_to_block(ec, end_sector) + (__block_mod(end_sector, ec->u_bs) ? 1 : 0);
 52}
 53
 54static inline bool __ebs_check_bs(unsigned int bs)
 55{
 56	return bs && is_power_of_2(bs);
 57}
 58
 59/*
 60 * READ/WRITE:
 61 *
 62 * copy blocks between bufio blocks and bio vector's (partial/overlapping) pages.
 63 */
 64static int __ebs_rw_bvec(struct ebs_c *ec, int rw, struct bio_vec *bv, struct bvec_iter *iter)
 65{
 66	int r = 0;
 67	unsigned char *ba, *pa;
 68	unsigned int cur_len;
 69	unsigned int bv_len = bv->bv_len;
 70	unsigned int buf_off = to_bytes(__block_mod(iter->bi_sector, ec->u_bs));
 71	sector_t block = __sector_to_block(ec, iter->bi_sector);
 72	struct dm_buffer *b;
 73
 74	if (unlikely(!bv->bv_page || !bv_len))
 75		return -EIO;
 76
 77	pa = page_address(bv->bv_page) + bv->bv_offset;
 78
 79	/* Handle overlapping page <-> blocks */
 80	while (bv_len) {
 81		cur_len = min(dm_bufio_get_block_size(ec->bufio) - buf_off, bv_len);
 82
 83		/* Avoid reading for writes in case bio vector's page overwrites block completely. */
 84		if (rw == READ || buf_off || bv_len < dm_bufio_get_block_size(ec->bufio))
 85			ba = dm_bufio_read(ec->bufio, block, &b);
 86		else
 87			ba = dm_bufio_new(ec->bufio, block, &b);
 88
 89		if (IS_ERR(ba)) {
 90			/*
 91			 * Carry on with next buffer, if any, to issue all possible
 92			 * data but return error.
 93			 */
 94			r = PTR_ERR(ba);
 95		} else {
 96			/* Copy data to/from bio to buffer if read/new was successful above. */
 97			ba += buf_off;
 98			if (rw == READ) {
 99				memcpy(pa, ba, cur_len);
100				flush_dcache_page(bv->bv_page);
101			} else {
102				flush_dcache_page(bv->bv_page);
103				memcpy(ba, pa, cur_len);
104				dm_bufio_mark_partial_buffer_dirty(b, buf_off, buf_off + cur_len);
105			}
106
107			dm_bufio_release(b);
108		}
109
110		pa += cur_len;
111		bv_len -= cur_len;
112		buf_off = 0;
113		block++;
114	}
115
116	return r;
117}
118
119/* READ/WRITE: iterate bio vector's copying between (partial) pages and bufio blocks. */
120static int __ebs_rw_bio(struct ebs_c *ec, int rw, struct bio *bio)
121{
122	int r = 0, rr;
123	struct bio_vec bv;
124	struct bvec_iter iter;
125
126	bio_for_each_bvec(bv, bio, iter) {
127		rr = __ebs_rw_bvec(ec, rw, &bv, &iter);
128		if (rr)
129			r = rr;
130	}
131
132	return r;
133}
134
135/*
136 * Discard bio's blocks, i.e. pass discards down.
137 *
138 * Avoid discarding partial blocks at beginning and end;
139 * return 0 in case no blocks can be discarded as a result.
140 */
141static int __ebs_discard_bio(struct ebs_c *ec, struct bio *bio)
142{
143	sector_t block, blocks, sector = bio->bi_iter.bi_sector;
144
145	block = __sector_to_block(ec, sector);
146	blocks = __nr_blocks(ec, bio);
147
148	/*
149	 * Partial first underlying block (__nr_blocks() may have
150	 * resulted in one block).
151	 */
152	if (__block_mod(sector, ec->u_bs)) {
153		block++;
154		blocks--;
155	}
156
157	/* Partial last underlying block if any. */
158	if (blocks && __block_mod(bio_end_sector(bio), ec->u_bs))
159		blocks--;
160
161	return blocks ? dm_bufio_issue_discard(ec->bufio, block, blocks) : 0;
162}
163
164/* Release blocks them from the bufio cache. */
165static void __ebs_forget_bio(struct ebs_c *ec, struct bio *bio)
166{
167	sector_t blocks, sector = bio->bi_iter.bi_sector;
168
169	blocks = __nr_blocks(ec, bio);
170
171	dm_bufio_forget_buffers(ec->bufio, __sector_to_block(ec, sector), blocks);
172}
173
174/* Worker function to process incoming bios. */
175static void __ebs_process_bios(struct work_struct *ws)
176{
177	int r;
178	bool write = false;
179	sector_t block1, block2;
180	struct ebs_c *ec = container_of(ws, struct ebs_c, ws);
181	struct bio *bio;
182	struct bio_list bios;
183
184	bio_list_init(&bios);
185
186	spin_lock_irq(&ec->lock);
187	bios = ec->bios_in;
188	bio_list_init(&ec->bios_in);
189	spin_unlock_irq(&ec->lock);
190
191	/* Prefetch all read and any mis-aligned write buffers */
192	bio_list_for_each(bio, &bios) {
193		block1 = __sector_to_block(ec, bio->bi_iter.bi_sector);
194		if (bio_op(bio) == REQ_OP_READ)
195			dm_bufio_prefetch(ec->bufio, block1, __nr_blocks(ec, bio));
196		else if (bio_op(bio) == REQ_OP_WRITE && !(bio->bi_opf & REQ_PREFLUSH)) {
197			block2 = __sector_to_block(ec, bio_end_sector(bio));
198			if (__block_mod(bio->bi_iter.bi_sector, ec->u_bs))
199				dm_bufio_prefetch(ec->bufio, block1, 1);
200			if (__block_mod(bio_end_sector(bio), ec->u_bs) && block2 != block1)
201				dm_bufio_prefetch(ec->bufio, block2, 1);
202		}
203	}
204
205	bio_list_for_each(bio, &bios) {
206		r = -EIO;
207		if (bio_op(bio) == REQ_OP_READ)
208			r = __ebs_rw_bio(ec, READ, bio);
209		else if (bio_op(bio) == REQ_OP_WRITE) {
210			write = true;
211			r = __ebs_rw_bio(ec, WRITE, bio);
212		} else if (bio_op(bio) == REQ_OP_DISCARD) {
213			__ebs_forget_bio(ec, bio);
214			r = __ebs_discard_bio(ec, bio);
215		}
216
217		if (r < 0)
218			bio->bi_status = errno_to_blk_status(r);
219	}
220
221	/*
222	 * We write dirty buffers after processing I/O on them
223	 * but before we endio thus addressing REQ_FUA/REQ_SYNC.
224	 */
225	r = write ? dm_bufio_write_dirty_buffers(ec->bufio) : 0;
226
227	while ((bio = bio_list_pop(&bios))) {
228		/* Any other request is endioed. */
229		if (unlikely(r && bio_op(bio) == REQ_OP_WRITE))
230			bio_io_error(bio);
231		else
232			bio_endio(bio);
233	}
234}
235
236/*
237 * Construct an emulated block size mapping: <dev_path> <offset> <ebs> [<ubs>]
238 *
239 * <dev_path>: path of the underlying device
240 * <offset>: offset in 512 bytes sectors into <dev_path>
241 * <ebs>: emulated block size in units of 512 bytes exposed to the upper layer
242 * [<ubs>]: underlying block size in units of 512 bytes imposed on the lower layer;
243 * 	    optional, if not supplied, retrieve logical block size from underlying device
244 */
245static int ebs_ctr(struct dm_target *ti, unsigned int argc, char **argv)
246{
247	int r;
248	unsigned short tmp1;
249	unsigned long long tmp;
250	char dummy;
251	struct ebs_c *ec;
252
253	if (argc < 3 || argc > 4) {
254		ti->error = "Invalid argument count";
255		return -EINVAL;
256	}
257
258	ec = ti->private = kzalloc(sizeof(*ec), GFP_KERNEL);
259	if (!ec) {
260		ti->error = "Cannot allocate ebs context";
261		return -ENOMEM;
262	}
263
264	r = -EINVAL;
265	if (sscanf(argv[1], "%llu%c", &tmp, &dummy) != 1 ||
266	    tmp != (sector_t)tmp ||
267	    (sector_t)tmp >= ti->len) {
268		ti->error = "Invalid device offset sector";
269		goto bad;
270	}
271	ec->start = tmp;
272
273	if (sscanf(argv[2], "%hu%c", &tmp1, &dummy) != 1 ||
274	    !__ebs_check_bs(tmp1) ||
275	    to_bytes(tmp1) > PAGE_SIZE) {
276		ti->error = "Invalid emulated block size";
277		goto bad;
278	}
279	ec->e_bs = tmp1;
280
281	if (argc > 3) {
282		if (sscanf(argv[3], "%hu%c", &tmp1, &dummy) != 1 || !__ebs_check_bs(tmp1)) {
283			ti->error = "Invalid underlying block size";
284			goto bad;
285		}
286		ec->u_bs = tmp1;
287		ec->u_bs_set = true;
288	} else
289		ec->u_bs_set = false;
290
291	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ec->dev);
292	if (r) {
293		ti->error = "Device lookup failed";
294		ec->dev = NULL;
295		goto bad;
296	}
297
298	r = -EINVAL;
299	if (!ec->u_bs_set) {
300		ec->u_bs = to_sector(bdev_logical_block_size(ec->dev->bdev));
301		if (!__ebs_check_bs(ec->u_bs)) {
302			ti->error = "Invalid retrieved underlying block size";
303			goto bad;
304		}
305	}
306
307	if (!ec->u_bs_set && ec->e_bs == ec->u_bs)
308		DMINFO("Emulation superfluous: emulated equal to underlying block size");
309
310	if (__block_mod(ec->start, ec->u_bs)) {
311		ti->error = "Device offset must be multiple of underlying block size";
312		goto bad;
313	}
314
315	ec->bufio = dm_bufio_client_create(ec->dev->bdev, to_bytes(ec->u_bs), 1, 0, NULL, NULL);
316	if (IS_ERR(ec->bufio)) {
317		ti->error = "Cannot create dm bufio client";
318		r = PTR_ERR(ec->bufio);
319		ec->bufio = NULL;
320		goto bad;
321	}
322
323	ec->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
324	if (!ec->wq) {
325		ti->error = "Cannot create dm-" DM_MSG_PREFIX " workqueue";
326		r = -ENOMEM;
327		goto bad;
328	}
329
330	ec->block_shift = __ffs(ec->u_bs);
331	INIT_WORK(&ec->ws, &__ebs_process_bios);
332	bio_list_init(&ec->bios_in);
333	spin_lock_init(&ec->lock);
334
335	ti->num_flush_bios = 1;
336	ti->num_discard_bios = 1;
337	ti->num_secure_erase_bios = 0;
338	ti->num_write_same_bios = 0;
339	ti->num_write_zeroes_bios = 0;
340	return 0;
341bad:
342	ebs_dtr(ti);
343	return r;
344}
345
346static void ebs_dtr(struct dm_target *ti)
347{
348	struct ebs_c *ec = ti->private;
349
350	if (ec->wq)
351		destroy_workqueue(ec->wq);
352	if (ec->bufio)
353		dm_bufio_client_destroy(ec->bufio);
354	if (ec->dev)
355		dm_put_device(ti, ec->dev);
356	kfree(ec);
357}
358
359static int ebs_map(struct dm_target *ti, struct bio *bio)
360{
361	struct ebs_c *ec = ti->private;
362
363	bio_set_dev(bio, ec->dev->bdev);
364	bio->bi_iter.bi_sector = ec->start + dm_target_offset(ti, bio->bi_iter.bi_sector);
365
366	if (unlikely(bio_op(bio) == REQ_OP_FLUSH))
367		return DM_MAPIO_REMAPPED;
368	/*
369	 * Only queue for bufio processing in case of partial or overlapping buffers
370	 * -or-
371	 * emulation with ebs == ubs aiming for tests of dm-bufio overhead.
372	 */
373	if (likely(__block_mod(bio->bi_iter.bi_sector, ec->u_bs) ||
374		   __block_mod(bio_end_sector(bio), ec->u_bs) ||
375		   ec->e_bs == ec->u_bs)) {
376		spin_lock_irq(&ec->lock);
377		bio_list_add(&ec->bios_in, bio);
378		spin_unlock_irq(&ec->lock);
379
380		queue_work(ec->wq, &ec->ws);
381
382		return DM_MAPIO_SUBMITTED;
383	}
384
385	/* Forget any buffer content relative to this direct backing device I/O. */
386	__ebs_forget_bio(ec, bio);
387
388	return DM_MAPIO_REMAPPED;
389}
390
391static void ebs_status(struct dm_target *ti, status_type_t type,
392		       unsigned status_flags, char *result, unsigned maxlen)
393{
394	struct ebs_c *ec = ti->private;
395
396	switch (type) {
397	case STATUSTYPE_INFO:
398		*result = '\0';
399		break;
400	case STATUSTYPE_TABLE:
401		snprintf(result, maxlen, ec->u_bs_set ? "%s %llu %u %u" : "%s %llu %u",
402			 ec->dev->name, (unsigned long long) ec->start, ec->e_bs, ec->u_bs);
403		break;
404	}
405}
406
407static int ebs_prepare_ioctl(struct dm_target *ti, struct block_device **bdev)
408{
409	struct ebs_c *ec = ti->private;
410	struct dm_dev *dev = ec->dev;
411
412	/*
413	 * Only pass ioctls through if the device sizes match exactly.
414	 */
415	*bdev = dev->bdev;
416	return !!(ec->start || ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT);
417}
418
419static void ebs_io_hints(struct dm_target *ti, struct queue_limits *limits)
420{
421	struct ebs_c *ec = ti->private;
422
423	limits->logical_block_size = to_bytes(ec->e_bs);
424	limits->physical_block_size = to_bytes(ec->u_bs);
425	limits->alignment_offset = limits->physical_block_size;
426	blk_limits_io_min(limits, limits->logical_block_size);
427}
428
429static int ebs_iterate_devices(struct dm_target *ti,
430				  iterate_devices_callout_fn fn, void *data)
431{
432	struct ebs_c *ec = ti->private;
433
434	return fn(ti, ec->dev, ec->start, ti->len, data);
435}
436
437static struct target_type ebs_target = {
438	.name		 = "ebs",
439	.version	 = {1, 0, 1},
440	.features	 = DM_TARGET_PASSES_INTEGRITY,
441	.module		 = THIS_MODULE,
442	.ctr		 = ebs_ctr,
443	.dtr		 = ebs_dtr,
444	.map		 = ebs_map,
445	.status		 = ebs_status,
446	.io_hints	 = ebs_io_hints,
447	.prepare_ioctl	 = ebs_prepare_ioctl,
448	.iterate_devices = ebs_iterate_devices,
449};
450
451static int __init dm_ebs_init(void)
452{
453	int r = dm_register_target(&ebs_target);
454
455	if (r < 0)
456		DMERR("register failed %d", r);
457
458	return r;
459}
460
461static void dm_ebs_exit(void)
462{
463	dm_unregister_target(&ebs_target);
464}
465
466module_init(dm_ebs_init);
467module_exit(dm_ebs_exit);
468
469MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
470MODULE_DESCRIPTION(DM_NAME " emulated block size target");
471MODULE_LICENSE("GPL");