dm-io --- diff/drivers/md/Makefile 2004-03-16 09:40:20.978986720 +0000 +++ source/drivers/md/Makefile 2004-03-16 09:40:39.682143408 +0000 @@ -3,7 +3,7 @@ # dm-mod-objs := dm.o dm-table.o dm-target.o dm-linear.o dm-stripe.o \ - dm-ioctl.o + dm-ioctl.o dm-io.o dm-multipath-objs := dm-path-selector.o dm-mpath.o raid6-objs := raid6main.o raid6algos.o raid6recov.o raid6tables.o \ raid6int1.o raid6int2.o raid6int4.o \ --- diff/mm/mempool.c 2004-03-16 09:37:58.060713600 +0000 +++ source/mm/mempool.c 2004-03-16 09:40:39.683143256 +0000 @@ -89,11 +89,6 @@ mempool_t * mempool_create(int min_nr, m } EXPORT_SYMBOL(mempool_create); -/* - * mempool_resize is disabled for now, because it has no callers. Feel free - * to turn it back on if needed. - */ -#if 0 /** * mempool_resize - resize an existing memory pool * @pool: pointer to the memory pool which was allocated via @@ -163,7 +158,6 @@ out: return 0; } EXPORT_SYMBOL(mempool_resize); -#endif /** * mempool_destroy - deallocate a memory pool --- diff/drivers/md/dm-io.c 1970-01-01 00:00:00.000000000 +0000 +++ source/drivers/md/dm-io.c 2004-03-16 09:40:39.684143104 +0000 @@ -0,0 +1,578 @@ +/* + * Copyright (C) 2003 Sistina Software + * + * This file is released under the GPL. + */ + +#include "dm-io.h" + +#include +#include +#include +#include +#include + +#define BIO_POOL_SIZE 256 + + +/*----------------------------------------------------------------- + * Bio set, move this to bio.c + *---------------------------------------------------------------*/ +#define BV_NAME_SIZE 16 +struct biovec_pool { + int nr_vecs; + char name[BV_NAME_SIZE]; + kmem_cache_t *slab; + mempool_t *pool; + atomic_t allocated; /* FIXME: debug */ +}; + +#define BIOVEC_NR_POOLS 6 +struct bio_set { + char name[BV_NAME_SIZE]; + kmem_cache_t *bio_slab; + mempool_t *bio_pool; + struct biovec_pool pools[BIOVEC_NR_POOLS]; +}; + +static void bio_set_exit(struct bio_set *bs) +{ + unsigned i; + struct biovec_pool *bp; + + if (bs->bio_pool) + mempool_destroy(bs->bio_pool); + + if (bs->bio_slab) + kmem_cache_destroy(bs->bio_slab); + + for (i = 0; i < BIOVEC_NR_POOLS; i++) { + bp = bs->pools + i; + if (bp->pool) + mempool_destroy(bp->pool); + + if (bp->slab) + kmem_cache_destroy(bp->slab); + } +} + +static void mk_name(char *str, size_t len, const char *prefix, unsigned count) +{ + snprintf(str, len, "%s-%u", prefix, count); +} + +static int bio_set_init(struct bio_set *bs, const char *slab_prefix, + unsigned pool_entries, unsigned scale) +{ + /* FIXME: this must match bvec_index(), why not go the + * whole hog and have a pool per power of 2 ? */ + static unsigned _vec_lengths[BIOVEC_NR_POOLS] = { + 1, 4, 16, 64, 128, BIO_MAX_PAGES + }; + + + unsigned i, size; + struct biovec_pool *bp; + + /* zero the bs so we can tear down properly on error */ + memset(bs, 0, sizeof(*bs)); + + /* + * Set up the bio pool. + */ + snprintf(bs->name, sizeof(bs->name), "%s-bio", slab_prefix); + + bs->bio_slab = kmem_cache_create(bs->name, sizeof(struct bio), 0, + SLAB_HWCACHE_ALIGN, NULL, NULL); + if (!bs->bio_slab) { + DMWARN("can't init bio slab"); + goto bad; + } + + bs->bio_pool = mempool_create(pool_entries, mempool_alloc_slab, + mempool_free_slab, bs->bio_slab); + if (!bs->bio_pool) { + DMWARN("can't init bio pool"); + goto bad; + } + + /* + * Set up the biovec pools. + */ + for (i = 0; i < BIOVEC_NR_POOLS; i++) { + bp = bs->pools + i; + bp->nr_vecs = _vec_lengths[i]; + atomic_set(&bp->allocated, 1); /* FIXME: debug */ + + + size = bp->nr_vecs * sizeof(struct bio_vec); + + mk_name(bp->name, sizeof(bp->name), slab_prefix, i); + bp->slab = kmem_cache_create(bp->name, size, 0, + SLAB_HWCACHE_ALIGN, NULL, NULL); + if (!bp->slab) { + DMWARN("can't init biovec slab cache"); + goto bad; + } + + if (i >= scale) + pool_entries >>= 1; + + bp->pool = mempool_create(pool_entries, mempool_alloc_slab, + mempool_free_slab, bp->slab); + if (!bp->pool) { + DMWARN("can't init biovec mempool"); + goto bad; + } + } + + return 0; + + bad: + bio_set_exit(bs); + return -ENOMEM; +} + +/* FIXME: blech */ +static inline unsigned bvec_index(unsigned nr) +{ + switch (nr) { + case 1: return 0; + case 2 ... 4: return 1; + case 5 ... 16: return 2; + case 17 ... 64: return 3; + case 65 ... 128:return 4; + case 129 ... BIO_MAX_PAGES: return 5; + } + + BUG(); + return 0; +} + +static inline void bs_bio_init(struct bio *bio) +{ + bio->bi_next = NULL; + bio->bi_flags = 1 << BIO_UPTODATE; + bio->bi_rw = 0; + bio->bi_vcnt = 0; + bio->bi_idx = 0; + bio->bi_phys_segments = 0; + bio->bi_hw_segments = 0; + bio->bi_size = 0; + bio->bi_max_vecs = 0; + bio->bi_end_io = NULL; + atomic_set(&bio->bi_cnt, 1); + bio->bi_private = NULL; +} + +static unsigned _bio_count = 0; +struct bio *bio_set_alloc(struct bio_set *bs, int gfp_mask, int nr_iovecs) +{ + struct biovec_pool *bp; + struct bio_vec *bv = NULL; + unsigned long idx; + struct bio *bio; + + bio = mempool_alloc(bs->bio_pool, gfp_mask); + if (unlikely(!bio)) + return NULL; + + bio_init(bio); + + if (likely(nr_iovecs)) { + idx = bvec_index(nr_iovecs); + bp = bs->pools + idx; + bv = mempool_alloc(bp->pool, gfp_mask); + if (!bv) { + mempool_free(bio, bs->bio_pool); + return NULL; + } + + memset(bv, 0, bp->nr_vecs * sizeof(*bv)); + bio->bi_flags |= idx << BIO_POOL_OFFSET; + bio->bi_max_vecs = bp->nr_vecs; + atomic_inc(&bp->allocated); + } + + bio->bi_io_vec = bv; + return bio; +} + +static void bio_set_free(struct bio_set *bs, struct bio *bio) +{ + struct biovec_pool *bp = bs->pools + BIO_POOL_IDX(bio); + + if (atomic_dec_and_test(&bp->allocated)) + BUG(); + + mempool_free(bio->bi_io_vec, bp->pool); + mempool_free(bio, bs->bio_pool); +} + +/*----------------------------------------------------------------- + * dm-io proper + *---------------------------------------------------------------*/ +static struct bio_set _bios; + +/* FIXME: can we shrink this ? */ +struct io { + unsigned long error; + atomic_t count; + struct task_struct *sleeper; + io_notify_fn callback; + void *context; +}; + +/* + * io contexts are only dynamically allocated for asynchronous + * io. Since async io is likely to be the majority of io we'll + * have the same number of io contexts as buffer heads ! (FIXME: + * must reduce this). + */ +static unsigned _num_ios; +static mempool_t *_io_pool; + +static void *alloc_io(int gfp_mask, void *pool_data) +{ + return kmalloc(sizeof(struct io), gfp_mask); +} + +static void free_io(void *element, void *pool_data) +{ + kfree(element); +} + +static unsigned int pages_to_ios(unsigned int pages) +{ + return 4 * pages; /* too many ? */ +} + +static int resize_pool(unsigned int new_ios) +{ + int r = 0; + + if (_io_pool) { + if (new_ios == 0) { + /* free off the pool */ + mempool_destroy(_io_pool); + _io_pool = NULL; + bio_set_exit(&_bios); + + } else { + /* resize the pool */ + r = mempool_resize(_io_pool, new_ios, GFP_KERNEL); + } + + } else { + /* create new pool */ + _io_pool = mempool_create(new_ios, alloc_io, free_io, NULL); + if (!_io_pool) + r = -ENOMEM; + + r = bio_set_init(&_bios, "dm-io", 512, 1); + if (r) { + mempool_destroy(_io_pool); + _io_pool = NULL; + } + } + + if (!r) + _num_ios = new_ios; + + return r; +} + +int dm_io_get(unsigned int num_pages) +{ + return resize_pool(_num_ios + pages_to_ios(num_pages)); +} + +void dm_io_put(unsigned int num_pages) +{ + resize_pool(_num_ios - pages_to_ios(num_pages)); +} + +/*----------------------------------------------------------------- + * We need to keep track of which region a bio is doing io for. + * In order to save a memory allocation we store this the last + * bvec which we know is unused (blech). + *---------------------------------------------------------------*/ +static inline void bio_set_region(struct bio *bio, unsigned region) +{ + bio->bi_io_vec[bio->bi_max_vecs - 1].bv_len = region; +} + +static inline unsigned bio_get_region(struct bio *bio) +{ + return bio->bi_io_vec[bio->bi_max_vecs - 1].bv_len; +} + +/*----------------------------------------------------------------- + * We need an io object to keep track of the number of bios that + * have been dispatched for a particular io. + *---------------------------------------------------------------*/ +static void dec_count(struct io *io, unsigned int region, int error) +{ + if (error) + set_bit(region, &io->error); + + if (atomic_dec_and_test(&io->count)) { + if (io->sleeper) + wake_up_process(io->sleeper); + + else { + int r = io->error; + io_notify_fn fn = io->callback; + void *context = io->context; + + mempool_free(io, _io_pool); + fn(r, context); + } + } +} + +static int endio(struct bio *bio, unsigned int done, int error) +{ + struct io *io = (struct io *) bio->bi_private; + + /* keep going until we've finished */ + if (bio->bi_size) + return 1; + + /* FIXME: kcopyd needs pages zeroing on read failure ??? + * sounds like kcopyd is broken */ + dec_count(io, bio_get_region(bio), error); + bio_put(bio); + + return 0; +} + +static void bio_dtr(struct bio *bio) +{ + _bio_count--; + bio_set_free(&_bios, bio); +} + +/*----------------------------------------------------------------- + * These little objects provide an abstraction for getting a new + * destination page for io. + *---------------------------------------------------------------*/ +struct dpages { + void (*get_page)(struct dpages *dp, + struct page **p, unsigned long *len, unsigned *offset); + void (*next_page)(struct dpages *dp); + + unsigned context_u; + void *context_ptr; +}; + +/* + * Functions for getting the pages from a list. + */ +void list_get_page(struct dpages *dp, + struct page **p, unsigned long *len, unsigned *offset) +{ + unsigned o = dp->context_u; + + *p = (struct page *) dp->context_ptr; + *len = PAGE_SIZE - o; + *offset = o; +} + +void list_next_page(struct dpages *dp) +{ + struct page *page = (struct page *) dp->context_ptr; + dp->context_ptr = list_entry(page->list.next, struct page, list); + dp->context_u = 0; +} + +void list_dp_init(struct dpages *dp, struct page *page, unsigned offset) +{ + dp->get_page = list_get_page; + dp->next_page = list_next_page; + dp->context_u = offset; + dp->context_ptr = page; +} + +/* + * Functions for getting the pages from a bvec. + */ +void bvec_get_page(struct dpages *dp, + struct page **p, unsigned long *len, unsigned *offset) +{ + struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr; + *p = bvec->bv_page; + *len = bvec->bv_len; + *offset = bvec->bv_offset; +} + +void bvec_next_page(struct dpages *dp) +{ + struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr; + dp->context_ptr = bvec + 1; +} + +void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec) +{ + dp->get_page = bvec_get_page; + dp->next_page = bvec_next_page; + dp->context_ptr = bvec; +} + +/*----------------------------------------------------------------- + * IO routines that accept a list of pages. + *---------------------------------------------------------------*/ +static void do_region(int rw, unsigned int region, struct io_region *where, + struct dpages *dp, struct io *io) +{ + struct bio *bio; + struct page *page; + unsigned long len; + unsigned offset; + unsigned num_bvecs; + sector_t remaining = where->count; + + while (remaining) { + /* + * Allocate a suitably sized bio, we add an extra + * bvec for bio_get/set_region(). + */ + num_bvecs = (remaining / (PAGE_SIZE >> 9)) + 2; + _bio_count++; + bio = bio_set_alloc(&_bios, GFP_NOIO, num_bvecs); + bio->bi_sector = where->sector + (where->count - remaining); + bio->bi_bdev = where->bdev; + bio->bi_end_io = endio; + bio->bi_private = io; + bio->bi_destructor = bio_dtr; + bio_set_region(bio, region); + + /* + * Try and add as many pages as possible. + */ + while (remaining) { + dp->get_page(dp, &page, &len, &offset); + len = min(len, to_bytes(remaining)); + if (!bio_add_page(bio, page, len, offset)) + break; + + offset = 0; + remaining -= to_sector(len); + dp->next_page(dp); + } + + atomic_inc(&io->count); + submit_bio(rw, bio); + } +} + +static void dispatch_io(int rw, unsigned int num_regions, + struct io_region *where, struct dpages *dp, + struct io *io) +{ + int i; + struct dpages old_pages = *dp; + + /* + * For multiple regions we need to be careful to rewind + * the dp object for each call to do_region. + */ + for (i = 0; i < num_regions; i++) { + *dp = old_pages; + if (where[i].count) + do_region(rw, i, where + i, dp, io); + } + + /* + * Drop the extra refence that we were holding to avoid + * the io being completed too early. + */ + dec_count(io, 0, 0); +} + +int sync_io(unsigned int num_regions, struct io_region *where, + int rw, struct dpages *dp, unsigned long *error_bits) +{ + struct io io; + + BUG_ON(num_regions > 1 && rw != WRITE); + + io.error = 0; + atomic_set(&io.count, 1); /* see dispatch_io() */ + io.sleeper = current; + + dispatch_io(rw, num_regions, where, dp, &io); + blk_run_queues(); + + while (1) { + set_current_state(TASK_UNINTERRUPTIBLE); + + if (!atomic_read(&io.count) || signal_pending(current)) + break; + + io_schedule(); + } + set_current_state(TASK_RUNNING); + + if (atomic_read(&io.count)) + return -EINTR; + + *error_bits = io.error; + return io.error ? -EIO : 0; +} + +int async_io(unsigned int num_regions, struct io_region *where, int rw, + struct dpages *dp, io_notify_fn fn, void *context) +{ + struct io *io = mempool_alloc(_io_pool, GFP_NOIO); + + io->error = 0; + atomic_set(&io->count, 1); /* see dispatch_io() */ + io->sleeper = NULL; + io->callback = fn; + io->context = context; + + dispatch_io(rw, num_regions, where, dp, io); + return 0; +} + +int dm_io_sync(unsigned int num_regions, struct io_region *where, int rw, + struct page *pages, unsigned int offset, + unsigned long *error_bits) +{ + struct dpages dp; + list_dp_init(&dp, pages, offset); + return sync_io(num_regions, where, rw, &dp, error_bits); +} + +int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, unsigned long *error_bits) +{ + struct dpages dp; + bvec_dp_init(&dp, bvec); + return sync_io(num_regions, where, rw, &dp, error_bits); +} + +int dm_io_async(unsigned int num_regions, struct io_region *where, int rw, + struct page *pages, unsigned int offset, + io_notify_fn fn, void *context) +{ + struct dpages dp; + list_dp_init(&dp, pages, offset); + return async_io(num_regions, where, rw, &dp, fn, context); +} + +int dm_io_async_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, io_notify_fn fn, void *context) +{ + struct dpages dp; + bvec_dp_init(&dp, bvec); + return async_io(num_regions, where, rw, &dp, fn, context); +} + + +EXPORT_SYMBOL(dm_io_get); +EXPORT_SYMBOL(dm_io_put); +EXPORT_SYMBOL(dm_io_sync); +EXPORT_SYMBOL(dm_io_async); +EXPORT_SYMBOL(dm_io_async_bvec); --- diff/drivers/md/dm-io.h 1970-01-01 00:00:00.000000000 +0000 +++ source/drivers/md/dm-io.h 2004-03-16 09:40:39.684143104 +0000 @@ -0,0 +1,68 @@ +/* + * Copyright (C) 2003 Sistina Software + * + * This file is released under the GPL. + */ + +#ifndef _DM_IO_H +#define _DM_IO_H + +#include "dm.h" + +/* FIXME make this configurable */ +#define DM_MAX_IO_REGIONS 8 + +struct io_region { + struct block_device *bdev; + sector_t sector; + sector_t count; +}; + + +/* + * 'error' is a bitset, with each bit indicating whether an error + * occurred doing io to the corresponding region. + */ +typedef void (*io_notify_fn)(unsigned long error, void *context); + + +/* + * Before anyone uses the IO interface they should call + * dm_io_get(), specifying roughly how many pages they are + * expecting to perform io on concurrently. + * + * This function may block. + */ +int dm_io_get(unsigned int num_pages); +void dm_io_put(unsigned int num_pages); + + +/* + * Synchronous IO. + * + * Please ensure that the rw flag in the next two functions is + * either READ or WRITE, ie. we don't take READA. Any + * regions with a zero count field will be ignored. + */ +int dm_io_sync(unsigned int num_regions, struct io_region *where, int rw, + struct page *pages, unsigned int offset, + unsigned long *error_bits); + +int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, unsigned long *error_bits); + + +/* + * Aynchronous IO. + * + * The 'where' array may be safely allocated on the stack since + * the function takes a copy. + */ +int dm_io_async(unsigned int num_regions, struct io_region *where, int rw, + struct page *pages, unsigned int offset, + io_notify_fn fn, void *context); + +int dm_io_async_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, io_notify_fn fn, void *context); + +#endif