2 * Copyright (C) 2003 Sistina Software
3 * Copyright (C) 2006 Red Hat GmbH
5 * This file is released under the GPL.
8 #include <linux/device-mapper.h>
10 #include <linux/bio.h>
11 #include <linux/mempool.h>
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <linux/dm-io.h>
22 /* FIXME: can we shrink this ? */
24 unsigned long error_bits;
25 unsigned long eopnotsupp_bits;
27 struct task_struct *sleeper;
28 struct dm_io_client *client;
29 io_notify_fn callback;
34 * io contexts are only dynamically allocated for asynchronous
35 * io. Since async io is likely to be the majority of io we'll
36 * have the same number of io contexts as bios! (FIXME: must reduce this).
39 static unsigned int pages_to_ios(unsigned int pages)
41 return 4 * pages; /* too many ? */
45 * Create a client with mempool and bioset.
47 struct dm_io_client *dm_io_client_create(unsigned num_pages)
49 unsigned ios = pages_to_ios(num_pages);
50 struct dm_io_client *client;
52 client = kmalloc(sizeof(*client), GFP_KERNEL);
54 return ERR_PTR(-ENOMEM);
56 client->pool = mempool_create_kmalloc_pool(ios, sizeof(struct io));
60 client->bios = bioset_create(16, 0);
68 mempool_destroy(client->pool);
70 return ERR_PTR(-ENOMEM);
72 EXPORT_SYMBOL(dm_io_client_create);
74 int dm_io_client_resize(unsigned num_pages, struct dm_io_client *client)
76 return mempool_resize(client->pool, pages_to_ios(num_pages),
79 EXPORT_SYMBOL(dm_io_client_resize);
81 void dm_io_client_destroy(struct dm_io_client *client)
83 mempool_destroy(client->pool);
84 bioset_free(client->bios);
87 EXPORT_SYMBOL(dm_io_client_destroy);
89 /*-----------------------------------------------------------------
90 * We need to keep track of which region a bio is doing io for.
91 * In order to save a memory allocation we store this the last
92 * bvec which we know is unused (blech).
93 * XXX This is ugly and can OOPS with some configs... find another way.
94 *---------------------------------------------------------------*/
95 static inline void bio_set_region(struct bio *bio, unsigned region)
97 bio->bi_io_vec[bio->bi_max_vecs].bv_len = region;
100 static inline unsigned bio_get_region(struct bio *bio)
102 return bio->bi_io_vec[bio->bi_max_vecs].bv_len;
105 /*-----------------------------------------------------------------
106 * We need an io object to keep track of the number of bios that
107 * have been dispatched for a particular io.
108 *---------------------------------------------------------------*/
109 static void dec_count(struct io *io, unsigned int region, int error)
112 set_bit(region, &io->error_bits);
113 if (error == -EOPNOTSUPP)
114 set_bit(region, &io->eopnotsupp_bits);
117 if (atomic_dec_and_test(&io->count)) {
119 wake_up_process(io->sleeper);
122 unsigned long r = io->error_bits;
123 io_notify_fn fn = io->callback;
124 void *context = io->context;
126 mempool_free(io, io->client->pool);
132 static void endio(struct bio *bio, int error)
137 if (error && bio_data_dir(bio) == READ)
141 * The bio destructor in bio_put() may use the io object.
143 io = bio->bi_private;
144 region = bio_get_region(bio);
149 dec_count(io, region, error);
152 /*-----------------------------------------------------------------
153 * These little objects provide an abstraction for getting a new
154 * destination page for io.
155 *---------------------------------------------------------------*/
157 void (*get_page)(struct dpages *dp,
158 struct page **p, unsigned long *len, unsigned *offset);
159 void (*next_page)(struct dpages *dp);
166 * Functions for getting the pages from a list.
168 static void list_get_page(struct dpages *dp,
169 struct page **p, unsigned long *len, unsigned *offset)
171 unsigned o = dp->context_u;
172 struct page_list *pl = (struct page_list *) dp->context_ptr;
175 *len = PAGE_SIZE - o;
179 static void list_next_page(struct dpages *dp)
181 struct page_list *pl = (struct page_list *) dp->context_ptr;
182 dp->context_ptr = pl->next;
186 static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset)
188 dp->get_page = list_get_page;
189 dp->next_page = list_next_page;
190 dp->context_u = offset;
191 dp->context_ptr = pl;
195 * Functions for getting the pages from a bvec.
197 static void bvec_get_page(struct dpages *dp,
198 struct page **p, unsigned long *len, unsigned *offset)
200 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
203 *offset = bvec->bv_offset;
206 static void bvec_next_page(struct dpages *dp)
208 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
209 dp->context_ptr = bvec + 1;
212 static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec)
214 dp->get_page = bvec_get_page;
215 dp->next_page = bvec_next_page;
216 dp->context_ptr = bvec;
220 * Functions for getting the pages from a VMA.
222 static void vm_get_page(struct dpages *dp,
223 struct page **p, unsigned long *len, unsigned *offset)
225 *p = vmalloc_to_page(dp->context_ptr);
226 *offset = dp->context_u;
227 *len = PAGE_SIZE - dp->context_u;
230 static void vm_next_page(struct dpages *dp)
232 dp->context_ptr += PAGE_SIZE - dp->context_u;
236 static void vm_dp_init(struct dpages *dp, void *data)
238 dp->get_page = vm_get_page;
239 dp->next_page = vm_next_page;
240 dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
241 dp->context_ptr = data;
244 static void dm_bio_destructor(struct bio *bio)
246 struct io *io = bio->bi_private;
248 bio_free(bio, io->client->bios);
252 * Functions for getting the pages from kernel memory.
254 static void km_get_page(struct dpages *dp, struct page **p, unsigned long *len,
257 *p = virt_to_page(dp->context_ptr);
258 *offset = dp->context_u;
259 *len = PAGE_SIZE - dp->context_u;
262 static void km_next_page(struct dpages *dp)
264 dp->context_ptr += PAGE_SIZE - dp->context_u;
268 static void km_dp_init(struct dpages *dp, void *data)
270 dp->get_page = km_get_page;
271 dp->next_page = km_next_page;
272 dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
273 dp->context_ptr = data;
276 /*-----------------------------------------------------------------
277 * IO routines that accept a list of pages.
278 *---------------------------------------------------------------*/
279 static void do_region(int rw, unsigned region, struct dm_io_region *where,
280 struct dpages *dp, struct io *io)
287 sector_t remaining = where->count;
291 * Allocate a suitably sized-bio: we add an extra
292 * bvec for bio_get/set_region() and decrement bi_max_vecs
293 * to hide it from bio_add_page().
295 num_bvecs = dm_sector_div_up(remaining,
296 (PAGE_SIZE >> SECTOR_SHIFT));
297 num_bvecs = 1 + min_t(int, bio_get_nr_vecs(where->bdev),
299 if (unlikely(num_bvecs > BIO_MAX_PAGES))
300 num_bvecs = BIO_MAX_PAGES;
301 bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
302 bio->bi_sector = where->sector + (where->count - remaining);
303 bio->bi_bdev = where->bdev;
304 bio->bi_end_io = endio;
305 bio->bi_private = io;
306 bio->bi_destructor = dm_bio_destructor;
308 bio_set_region(bio, region);
311 * Try and add as many pages as possible.
314 dp->get_page(dp, &page, &len, &offset);
315 len = min(len, to_bytes(remaining));
316 if (!bio_add_page(bio, page, len, offset))
320 remaining -= to_sector(len);
324 atomic_inc(&io->count);
329 static void dispatch_io(int rw, unsigned int num_regions,
330 struct dm_io_region *where, struct dpages *dp,
331 struct io *io, int sync)
334 struct dpages old_pages = *dp;
337 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
340 * For multiple regions we need to be careful to rewind
341 * the dp object for each call to do_region.
343 for (i = 0; i < num_regions; i++) {
346 do_region(rw, i, where + i, dp, io);
350 * Drop the extra reference that we were holding to avoid
351 * the io being completed too early.
356 static int sync_io(struct dm_io_client *client, unsigned int num_regions,
357 struct dm_io_region *where, int rw, struct dpages *dp,
358 unsigned long *error_bits)
362 if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
369 io.eopnotsupp_bits = 0;
370 atomic_set(&io.count, 1); /* see dispatch_io() */
371 io.sleeper = current;
374 dispatch_io(rw, num_regions, where, dp, &io, 1);
377 set_current_state(TASK_UNINTERRUPTIBLE);
379 if (!atomic_read(&io.count))
384 set_current_state(TASK_RUNNING);
386 if (io.eopnotsupp_bits && (rw & (1 << BIO_RW_BARRIER))) {
387 rw &= ~(1 << BIO_RW_BARRIER);
392 *error_bits = io.error_bits;
394 return io.error_bits ? -EIO : 0;
397 static int async_io(struct dm_io_client *client, unsigned int num_regions,
398 struct dm_io_region *where, int rw, struct dpages *dp,
399 io_notify_fn fn, void *context)
403 if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
409 io = mempool_alloc(client->pool, GFP_NOIO);
411 io->eopnotsupp_bits = 0;
412 atomic_set(&io->count, 1); /* see dispatch_io() */
416 io->context = context;
418 dispatch_io(rw, num_regions, where, dp, io, 0);
422 static int dp_init(struct dm_io_request *io_req, struct dpages *dp)
424 /* Set up dpages based on memory type */
425 switch (io_req->mem.type) {
426 case DM_IO_PAGE_LIST:
427 list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
431 bvec_dp_init(dp, io_req->mem.ptr.bvec);
435 vm_dp_init(dp, io_req->mem.ptr.vma);
439 km_dp_init(dp, io_req->mem.ptr.addr);
450 * New collapsed (a)synchronous interface.
452 * If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
453 * the queue with blk_unplug() some time later or set the BIO_RW_SYNC bit in
454 * io_req->bi_rw. If you fail to do one of these, the IO will be submitted to
455 * the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
457 int dm_io(struct dm_io_request *io_req, unsigned num_regions,
458 struct dm_io_region *where, unsigned long *sync_error_bits)
463 r = dp_init(io_req, &dp);
467 if (!io_req->notify.fn)
468 return sync_io(io_req->client, num_regions, where,
469 io_req->bi_rw, &dp, sync_error_bits);
471 return async_io(io_req->client, num_regions, where, io_req->bi_rw,
472 &dp, io_req->notify.fn, io_req->notify.context);
474 EXPORT_SYMBOL(dm_io);