2 * Copyright (C) 2020 Red Hat GmbH
4 * This file is released under the GPL.
6 * Device-mapper target to emulate smaller logical block
7 * size on backing devices exposing (natively) larger ones.
9 * E.g. 512 byte sector emulation on 4K native disks.
13 #include <linux/module.h>
14 #include <linux/workqueue.h>
15 #include <linux/dm-bufio.h>
17 #define DM_MSG_PREFIX "ebs"
19 static void ebs_dtr(struct dm_target *ti);
21 /* Emulated block size context. */
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. */
36 static inline sector_t __sector_to_block(struct ebs_c *ec, sector_t sector)
38 return sector >> ec->block_shift;
41 static inline sector_t __block_mod(sector_t sector, unsigned int bs)
43 return sector & (bs - 1);
46 /* Return number of blocks for a bio, accounting for misalignment of start and end sectors. */
47 static inline unsigned int __nr_blocks(struct ebs_c *ec, struct bio *bio)
49 sector_t end_sector = __block_mod(bio->bi_iter.bi_sector, ec->u_bs) + bio_sectors(bio);
51 return __sector_to_block(ec, end_sector) + (__block_mod(end_sector, ec->u_bs) ? 1 : 0);
54 static inline bool __ebs_check_bs(unsigned int bs)
56 return bs && is_power_of_2(bs);
62 * copy blocks between bufio blocks and bio vector's (partial/overlapping) pages.
64 static int __ebs_rw_bvec(struct ebs_c *ec, int rw, struct bio_vec *bv, struct bvec_iter *iter)
67 unsigned char *ba, *pa;
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);
74 if (unlikely(!bv->bv_page || !bv_len))
79 /* Handle overlapping page <-> blocks */
81 cur_len = min(dm_bufio_get_block_size(ec->bufio) - buf_off, bv_len);
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);
87 ba = dm_bufio_new(ec->bufio, block, &b);
91 * Carry on with next buffer, if any, to issue all possible
92 * data but return error.
96 /* Copy data to/from bio to buffer if read/new was successful above. */
99 memcpy(pa, ba, cur_len);
100 flush_dcache_page(bv->bv_page);
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);
119 /* READ/WRITE: iterate bio vector's copying between (partial) pages and bufio blocks. */
120 static int __ebs_rw_bio(struct ebs_c *ec, int rw, struct bio *bio)
124 struct bvec_iter iter;
126 bio_for_each_bvec(bv, bio, iter) {
127 rr = __ebs_rw_bvec(ec, rw, &bv, &iter);
136 * Discard bio's blocks, i.e. pass discards down.
138 * Avoid discarding partial blocks at beginning and end;
139 * return 0 in case no blocks can be discarded as a result.
141 static int __ebs_discard_bio(struct ebs_c *ec, struct bio *bio)
143 sector_t block, blocks, sector = bio->bi_iter.bi_sector;
145 block = __sector_to_block(ec, sector);
146 blocks = __nr_blocks(ec, bio);
149 * Partial first underlying block (__nr_blocks() may have
150 * resulted in one block).
152 if (__block_mod(sector, ec->u_bs)) {
157 /* Partial last underlying block if any. */
158 if (blocks && __block_mod(bio_end_sector(bio), ec->u_bs))
161 return blocks ? dm_bufio_issue_discard(ec->bufio, block, blocks) : 0;
164 /* Release blocks them from the bufio cache. */
165 static void __ebs_forget_bio(struct ebs_c *ec, struct bio *bio)
167 sector_t blocks, sector = bio->bi_iter.bi_sector;
169 blocks = __nr_blocks(ec, bio);
171 dm_bufio_forget_buffers(ec->bufio, __sector_to_block(ec, sector), blocks);
174 /* Worker function to process incoming bios. */
175 static void __ebs_process_bios(struct work_struct *ws)
179 sector_t block1, block2;
180 struct ebs_c *ec = container_of(ws, struct ebs_c, ws);
182 struct bio_list bios;
184 bio_list_init(&bios);
186 spin_lock_irq(&ec->lock);
188 bio_list_init(&ec->bios_in);
189 spin_unlock_irq(&ec->lock);
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);
205 bio_list_for_each(bio, &bios) {
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) {
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);
218 bio->bi_status = errno_to_blk_status(r);
222 * We write dirty buffers after processing I/O on them
223 * but before we endio thus addressing REQ_FUA/REQ_SYNC.
225 r = write ? dm_bufio_write_dirty_buffers(ec->bufio) : 0;
227 while ((bio = bio_list_pop(&bios))) {
228 /* Any other request is endioed. */
229 if (unlikely(r && bio_op(bio) == REQ_OP_WRITE))
237 * Construct an emulated block size mapping: <dev_path> <offset> <ebs> [<ubs>]
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
245 static int ebs_ctr(struct dm_target *ti, unsigned int argc, char **argv)
249 unsigned long long tmp;
253 if (argc < 3 || argc > 4) {
254 ti->error = "Invalid argument count";
258 ec = ti->private = kzalloc(sizeof(*ec), GFP_KERNEL);
260 ti->error = "Cannot allocate ebs context";
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";
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";
282 if (sscanf(argv[3], "%hu%c", &tmp1, &dummy) != 1 || !__ebs_check_bs(tmp1)) {
283 ti->error = "Invalid underlying block size";
289 ec->u_bs_set = false;
291 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ec->dev);
293 ti->error = "Device lookup failed";
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";
307 if (!ec->u_bs_set && ec->e_bs == ec->u_bs)
308 DMINFO("Emulation superfluous: emulated equal to underlying block size");
310 if (__block_mod(ec->start, ec->u_bs)) {
311 ti->error = "Device offset must be multiple of underlying block size";
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);
323 ec->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
325 ti->error = "Cannot create dm-" DM_MSG_PREFIX " workqueue";
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);
335 ti->num_flush_bios = 1;
336 ti->num_discard_bios = 1;
337 ti->num_secure_erase_bios = 0;
338 ti->num_write_zeroes_bios = 0;
345 static void ebs_dtr(struct dm_target *ti)
347 struct ebs_c *ec = ti->private;
350 destroy_workqueue(ec->wq);
352 dm_bufio_client_destroy(ec->bufio);
354 dm_put_device(ti, ec->dev);
358 static int ebs_map(struct dm_target *ti, struct bio *bio)
360 struct ebs_c *ec = ti->private;
362 bio_set_dev(bio, ec->dev->bdev);
363 bio->bi_iter.bi_sector = ec->start + dm_target_offset(ti, bio->bi_iter.bi_sector);
365 if (unlikely(bio_op(bio) == REQ_OP_FLUSH))
366 return DM_MAPIO_REMAPPED;
368 * Only queue for bufio processing in case of partial or overlapping buffers
370 * emulation with ebs == ubs aiming for tests of dm-bufio overhead.
372 if (likely(__block_mod(bio->bi_iter.bi_sector, ec->u_bs) ||
373 __block_mod(bio_end_sector(bio), ec->u_bs) ||
374 ec->e_bs == ec->u_bs)) {
375 spin_lock_irq(&ec->lock);
376 bio_list_add(&ec->bios_in, bio);
377 spin_unlock_irq(&ec->lock);
379 queue_work(ec->wq, &ec->ws);
381 return DM_MAPIO_SUBMITTED;
384 /* Forget any buffer content relative to this direct backing device I/O. */
385 __ebs_forget_bio(ec, bio);
387 return DM_MAPIO_REMAPPED;
390 static void ebs_status(struct dm_target *ti, status_type_t type,
391 unsigned status_flags, char *result, unsigned maxlen)
393 struct ebs_c *ec = ti->private;
396 case STATUSTYPE_INFO:
399 case STATUSTYPE_TABLE:
400 snprintf(result, maxlen, ec->u_bs_set ? "%s %llu %u %u" : "%s %llu %u",
401 ec->dev->name, (unsigned long long) ec->start, ec->e_bs, ec->u_bs);
409 static int ebs_prepare_ioctl(struct dm_target *ti, struct block_device **bdev)
411 struct ebs_c *ec = ti->private;
412 struct dm_dev *dev = ec->dev;
415 * Only pass ioctls through if the device sizes match exactly.
418 return !!(ec->start || ti->len != bdev_nr_sectors(dev->bdev));
421 static void ebs_io_hints(struct dm_target *ti, struct queue_limits *limits)
423 struct ebs_c *ec = ti->private;
425 limits->logical_block_size = to_bytes(ec->e_bs);
426 limits->physical_block_size = to_bytes(ec->u_bs);
427 limits->alignment_offset = limits->physical_block_size;
428 blk_limits_io_min(limits, limits->logical_block_size);
431 static int ebs_iterate_devices(struct dm_target *ti,
432 iterate_devices_callout_fn fn, void *data)
434 struct ebs_c *ec = ti->private;
436 return fn(ti, ec->dev, ec->start, ti->len, data);
439 static struct target_type ebs_target = {
441 .version = {1, 0, 1},
442 .features = DM_TARGET_PASSES_INTEGRITY,
443 .module = THIS_MODULE,
447 .status = ebs_status,
448 .io_hints = ebs_io_hints,
449 .prepare_ioctl = ebs_prepare_ioctl,
450 .iterate_devices = ebs_iterate_devices,
453 static int __init dm_ebs_init(void)
455 int r = dm_register_target(&ebs_target);
458 DMERR("register failed %d", r);
463 static void dm_ebs_exit(void)
465 dm_unregister_target(&ebs_target);
468 module_init(dm_ebs_init);
469 module_exit(dm_ebs_exit);
471 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
472 MODULE_DESCRIPTION(DM_NAME " emulated block size target");
473 MODULE_LICENSE("GPL");