From 08bafc0341f2f7920e9045bc32c40299cac8c21b Mon Sep 17 00:00:00 2001 From: Keith Mannthey Date: Tue, 25 Nov 2008 10:24:35 +0100 Subject: block: Supress Buffer I/O errors when SCSI REQ_QUIET flag set Allow the scsi request REQ_QUIET flag to be propagated to the buffer file system layer. The basic ideas is to pass the flag from the scsi request to the bio (block IO) and then to the buffer layer. The buffer layer can then suppress needless printks. This patch declutters the kernel log by removed the 40-50 (per lun) buffer io error messages seen during a boot in my multipath setup . It is a good chance any real errors will be missed in the "noise" it the logs without this patch. During boot I see blocks of messages like " __ratelimit: 211 callbacks suppressed Buffer I/O error on device sdm, logical block 5242879 Buffer I/O error on device sdm, logical block 5242879 Buffer I/O error on device sdm, logical block 5242847 Buffer I/O error on device sdm, logical block 1 Buffer I/O error on device sdm, logical block 5242878 Buffer I/O error on device sdm, logical block 5242879 Buffer I/O error on device sdm, logical block 5242879 Buffer I/O error on device sdm, logical block 5242879 Buffer I/O error on device sdm, logical block 5242879 Buffer I/O error on device sdm, logical block 5242872 " in my logs. My disk environment is multipath fiber channel using the SCSI_DH_RDAC code and multipathd. This topology includes an "active" and "ghost" path for each lun. IO's to the "ghost" path will never complete and the SCSI layer, via the scsi device handler rdac code, quick returns the IOs to theses paths and sets the REQ_QUIET scsi flag to suppress the scsi layer messages. I am wanting to extend the QUIET behavior to include the buffer file system layer to deal with these errors as well. I have been running this patch for a while now on several boxes without issue. A few runs of bonnie++ show no noticeable difference in performance in my setup. Thanks for John Stultz for the quiet_error finalization. Submitted-by: Keith Mannthey Signed-off-by: Jens Axboe --- fs/buffer.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-) (limited to 'fs') diff --git a/fs/buffer.c b/fs/buffer.c index 10179cfa115..776ae091d3b 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -99,10 +99,18 @@ __clear_page_buffers(struct page *page) page_cache_release(page); } + +static int quiet_error(struct buffer_head *bh) +{ + if (!test_bit(BH_Quiet, &bh->b_state) && printk_ratelimit()) + return 0; + return 1; +} + + static void buffer_io_error(struct buffer_head *bh) { char b[BDEVNAME_SIZE]; - printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n", bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); @@ -144,7 +152,7 @@ void end_buffer_write_sync(struct buffer_head *bh, int uptodate) if (uptodate) { set_buffer_uptodate(bh); } else { - if (!buffer_eopnotsupp(bh) && printk_ratelimit()) { + if (!buffer_eopnotsupp(bh) && !quiet_error(bh)) { buffer_io_error(bh); printk(KERN_WARNING "lost page write due to " "I/O error on %s\n", @@ -394,7 +402,7 @@ static void end_buffer_async_read(struct buffer_head *bh, int uptodate) set_buffer_uptodate(bh); } else { clear_buffer_uptodate(bh); - if (printk_ratelimit()) + if (!quiet_error(bh)) buffer_io_error(bh); SetPageError(page); } @@ -455,7 +463,7 @@ static void end_buffer_async_write(struct buffer_head *bh, int uptodate) if (uptodate) { set_buffer_uptodate(bh); } else { - if (printk_ratelimit()) { + if (!quiet_error(bh)) { buffer_io_error(bh); printk(KERN_WARNING "lost page write due to " "I/O error on %s\n", @@ -2913,6 +2921,9 @@ static void end_bio_bh_io_sync(struct bio *bio, int err) set_bit(BH_Eopnotsupp, &bh->b_state); } + if (unlikely (test_bit(BIO_QUIET,&bio->bi_flags))) + set_bit(BH_Quiet, &bh->b_state); + bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags)); bio_put(bio); } -- cgit v1.2.3-70-g09d2 From 7ff9345ffac56743b5001561bc2dc1e041b79149 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Thu, 11 Dec 2008 11:53:43 +0100 Subject: bio: only mempool back the largest bio_vec slab cache We only very rarely need the mempool backing, so it makes sense to get rid of all but one of the mempool in a bio_set. So keep the largest bio_vec count mempool so we can always honor the largest allocation, and "upgrade" callers that fail. Signed-off-by: Jens Axboe --- fs/bio.c | 125 ++++++++++++++++++++++++++++++---------------------- include/linux/bio.h | 3 +- 2 files changed, 74 insertions(+), 54 deletions(-) (limited to 'fs') diff --git a/fs/bio.c b/fs/bio.c index df99c882b80..eb6b4683a26 100644 --- a/fs/bio.c +++ b/fs/bio.c @@ -58,7 +58,8 @@ unsigned int bvec_nr_vecs(unsigned short idx) return bvec_slabs[idx].nr_vecs; } -struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) +struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, + struct bio_set *bs) { struct bio_vec *bvl; @@ -67,60 +68,90 @@ struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct * If not, this is a bio_kmalloc() allocation and just do a * kzalloc() for the exact number of vecs right away. */ - if (bs) { + if (!bs) + bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask); + + /* + * see comment near bvec_array define! + */ + switch (nr) { + case 1: + *idx = 0; + break; + case 2 ... 4: + *idx = 1; + break; + case 5 ... 16: + *idx = 2; + break; + case 17 ... 64: + *idx = 3; + break; + case 65 ... 128: + *idx = 4; + break; + case 129 ... BIO_MAX_PAGES: + *idx = 5; + break; + default: + return NULL; + } + + /* + * idx now points to the pool we want to allocate from. only the + * 1-vec entry pool is mempool backed. + */ + if (*idx == BIOVEC_MAX_IDX) { +fallback: + bvl = mempool_alloc(bs->bvec_pool, gfp_mask); + } else { + struct biovec_slab *bvs = bvec_slabs + *idx; + gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); + /* - * see comment near bvec_array define! + * Make this allocation restricted and don't dump info on + * allocation failures, since we'll fallback to the mempool + * in case of failure. */ - switch (nr) { - case 1: - *idx = 0; - break; - case 2 ... 4: - *idx = 1; - break; - case 5 ... 16: - *idx = 2; - break; - case 17 ... 64: - *idx = 3; - break; - case 65 ... 128: - *idx = 4; - break; - case 129 ... BIO_MAX_PAGES: - *idx = 5; - break; - default: - return NULL; - } + __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; /* - * idx now points to the pool we want to allocate from + * Try a slab allocation. If this fails and __GFP_WAIT + * is set, retry with the 1-entry mempool */ - bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); - if (bvl) - memset(bvl, 0, - bvec_nr_vecs(*idx) * sizeof(struct bio_vec)); - } else - bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask); + bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); + if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { + *idx = BIOVEC_MAX_IDX; + goto fallback; + } + } + + if (bvl) + memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec)); return bvl; } -void bio_free(struct bio *bio, struct bio_set *bio_set) +void bio_free(struct bio *bio, struct bio_set *bs) { if (bio->bi_io_vec) { const int pool_idx = BIO_POOL_IDX(bio); BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); - mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); + if (pool_idx == BIOVEC_MAX_IDX) + mempool_free(bio->bi_io_vec, bs->bvec_pool); + else { + struct biovec_slab *bvs = bvec_slabs + pool_idx; + + kmem_cache_free(bvs->slab, bio->bi_io_vec); + } } if (bio_integrity(bio)) - bio_integrity_free(bio, bio_set); + bio_integrity_free(bio, bs); - mempool_free(bio, bio_set->bio_pool); + mempool_free(bio, bs->bio_pool); } /* @@ -1346,30 +1377,18 @@ EXPORT_SYMBOL(bio_sector_offset); */ static int biovec_create_pools(struct bio_set *bs, int pool_entries) { - int i; + struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; - for (i = 0; i < BIOVEC_NR_POOLS; i++) { - struct biovec_slab *bp = bvec_slabs + i; - mempool_t **bvp = bs->bvec_pools + i; + bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab); + if (!bs->bvec_pool) + return -ENOMEM; - *bvp = mempool_create_slab_pool(pool_entries, bp->slab); - if (!*bvp) - return -ENOMEM; - } return 0; } static void biovec_free_pools(struct bio_set *bs) { - int i; - - for (i = 0; i < BIOVEC_NR_POOLS; i++) { - mempool_t *bvp = bs->bvec_pools[i]; - - if (bvp) - mempool_destroy(bvp); - } - + mempool_destroy(bs->bvec_pool); } void bioset_free(struct bio_set *bs) diff --git a/include/linux/bio.h b/include/linux/bio.h index 3ed714eb54d..d76e4bf22f2 100644 --- a/include/linux/bio.h +++ b/include/linux/bio.h @@ -397,13 +397,14 @@ static inline void bio_set_completion_cpu(struct bio *bio, unsigned int cpu) */ #define BIO_POOL_SIZE 2 #define BIOVEC_NR_POOLS 6 +#define BIOVEC_MAX_IDX (BIOVEC_NR_POOLS - 1) struct bio_set { mempool_t *bio_pool; #if defined(CONFIG_BLK_DEV_INTEGRITY) mempool_t *bio_integrity_pool; #endif - mempool_t *bvec_pools[BIOVEC_NR_POOLS]; + mempool_t *bvec_pool; }; struct biovec_slab { -- cgit v1.2.3-70-g09d2 From 1b4344986926da324b5cd10b683e5a1a5e1b7db3 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Wed, 22 Oct 2008 20:32:58 +0200 Subject: bio: move the slab pointer inside the bio_set In preparation for adding differently sized bios. Signed-off-by: Jens Axboe --- fs/bio.c | 7 +++++-- include/linux/bio.h | 1 + 2 files changed, 6 insertions(+), 2 deletions(-) (limited to 'fs') diff --git a/fs/bio.c b/fs/bio.c index eb6b4683a26..1ab8986b041 100644 --- a/fs/bio.c +++ b/fs/bio.c @@ -1404,12 +1404,15 @@ void bioset_free(struct bio_set *bs) struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) { - struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); + struct bio_set *bs; + bs = kzalloc(sizeof(*bs), GFP_KERNEL); if (!bs) return NULL; - bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); + bs->bio_slab = bio_slab; + + bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bs->bio_slab); if (!bs->bio_pool) goto bad; diff --git a/include/linux/bio.h b/include/linux/bio.h index d76e4bf22f2..9340098d75d 100644 --- a/include/linux/bio.h +++ b/include/linux/bio.h @@ -400,6 +400,7 @@ static inline void bio_set_completion_cpu(struct bio *bio, unsigned int cpu) #define BIOVEC_MAX_IDX (BIOVEC_NR_POOLS - 1) struct bio_set { + struct kmem_cache *bio_slab; mempool_t *bio_pool; #if defined(CONFIG_BLK_DEV_INTEGRITY) mempool_t *bio_integrity_pool; -- cgit v1.2.3-70-g09d2 From bb799ca0202a360fa74d5f17039b9100caebdde7 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Wed, 10 Dec 2008 15:35:05 +0100 Subject: bio: allow individual slabs in the bio_set Instead of having a global bio slab cache, add a reference to one in each bio_set that is created. This allows for personalized slabs in each bio_set, so that they can have bios of different sizes. This means we can personalize the bios we return. File systems may want to embed the bio inside another structure, to avoid allocation more items (and stuffing them in ->bi_private) after the get a bio. Or we may want to embed a number of bio_vecs directly at the end of a bio, to avoid doing two allocations to return a bio. This is now possible. Signed-off-by: Jens Axboe --- drivers/md/dm-crypt.c | 2 +- drivers/md/dm-io.c | 2 +- drivers/md/dm.c | 2 +- fs/bio-integrity.c | 2 +- fs/bio.c | 191 ++++++++++++++++++++++++++++++++++++++++++-------- include/linux/bio.h | 6 +- 6 files changed, 170 insertions(+), 35 deletions(-) (limited to 'fs') diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c index ce26c84af06..3326750ec02 100644 --- a/drivers/md/dm-crypt.c +++ b/drivers/md/dm-crypt.c @@ -1060,7 +1060,7 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) goto bad_page_pool; } - cc->bs = bioset_create(MIN_IOS, MIN_IOS); + cc->bs = bioset_create(MIN_IOS, 0); if (!cc->bs) { ti->error = "Cannot allocate crypt bioset"; goto bad_bs; diff --git a/drivers/md/dm-io.c b/drivers/md/dm-io.c index 2fd6d445063..a34338567a2 100644 --- a/drivers/md/dm-io.c +++ b/drivers/md/dm-io.c @@ -56,7 +56,7 @@ struct dm_io_client *dm_io_client_create(unsigned num_pages) if (!client->pool) goto bad; - client->bios = bioset_create(16, 16); + client->bios = bioset_create(16, 0); if (!client->bios) goto bad; diff --git a/drivers/md/dm.c b/drivers/md/dm.c index 343094c3fee..421c9f02d8c 100644 --- a/drivers/md/dm.c +++ b/drivers/md/dm.c @@ -1093,7 +1093,7 @@ static struct mapped_device *alloc_dev(int minor) if (!md->tio_pool) goto bad_tio_pool; - md->bs = bioset_create(16, 16); + md->bs = bioset_create(16, 0); if (!md->bs) goto bad_no_bioset; diff --git a/fs/bio-integrity.c b/fs/bio-integrity.c index 19caf7c962a..77ebc3c263d 100644 --- a/fs/bio-integrity.c +++ b/fs/bio-integrity.c @@ -111,7 +111,7 @@ void bio_integrity_free(struct bio *bio, struct bio_set *bs) && bip->bip_buf != NULL) kfree(bip->bip_buf); - mempool_free(bip->bip_vec, bs->bvec_pools[bip->bip_pool]); + bvec_free_bs(bs, bip->bip_vec, bip->bip_pool); mempool_free(bip, bs->bio_integrity_pool); bio->bi_integrity = NULL; diff --git a/fs/bio.c b/fs/bio.c index 1ab8986b041..0146f80789e 100644 --- a/fs/bio.c +++ b/fs/bio.c @@ -31,8 +31,6 @@ DEFINE_TRACE(block_split); -static struct kmem_cache *bio_slab __read_mostly; - static mempool_t *bio_split_pool __read_mostly; /* @@ -40,9 +38,8 @@ static mempool_t *bio_split_pool __read_mostly; * break badly! cannot be bigger than what you can fit into an * unsigned short */ - #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } -static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { +struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), }; #undef BV @@ -53,11 +50,119 @@ static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { */ struct bio_set *fs_bio_set; +/* + * Our slab pool management + */ +struct bio_slab { + struct kmem_cache *slab; + unsigned int slab_ref; + unsigned int slab_size; + char name[8]; +}; +static DEFINE_MUTEX(bio_slab_lock); +static struct bio_slab *bio_slabs; +static unsigned int bio_slab_nr, bio_slab_max; + +static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) +{ + unsigned int sz = sizeof(struct bio) + extra_size; + struct kmem_cache *slab = NULL; + struct bio_slab *bslab; + unsigned int i, entry = -1; + + mutex_lock(&bio_slab_lock); + + i = 0; + while (i < bio_slab_nr) { + struct bio_slab *bslab = &bio_slabs[i]; + + if (!bslab->slab && entry == -1) + entry = i; + else if (bslab->slab_size == sz) { + slab = bslab->slab; + bslab->slab_ref++; + break; + } + i++; + } + + if (slab) + goto out_unlock; + + if (bio_slab_nr == bio_slab_max && entry == -1) { + bio_slab_max <<= 1; + bio_slabs = krealloc(bio_slabs, + bio_slab_max * sizeof(struct bio_slab), + GFP_KERNEL); + if (!bio_slabs) + goto out_unlock; + } + if (entry == -1) + entry = bio_slab_nr++; + + bslab = &bio_slabs[entry]; + + snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); + slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL); + if (!slab) + goto out_unlock; + + printk("bio: create slab <%s> at %d\n", bslab->name, entry); + bslab->slab = slab; + bslab->slab_ref = 1; + bslab->slab_size = sz; +out_unlock: + mutex_unlock(&bio_slab_lock); + return slab; +} + +static void bio_put_slab(struct bio_set *bs) +{ + struct bio_slab *bslab = NULL; + unsigned int i; + + mutex_lock(&bio_slab_lock); + + for (i = 0; i < bio_slab_nr; i++) { + if (bs->bio_slab == bio_slabs[i].slab) { + bslab = &bio_slabs[i]; + break; + } + } + + if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) + goto out; + + WARN_ON(!bslab->slab_ref); + + if (--bslab->slab_ref) + goto out; + + kmem_cache_destroy(bslab->slab); + bslab->slab = NULL; + +out: + mutex_unlock(&bio_slab_lock); +} + unsigned int bvec_nr_vecs(unsigned short idx) { return bvec_slabs[idx].nr_vecs; } +void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx) +{ + BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); + + if (idx == BIOVEC_MAX_IDX) + mempool_free(bv, bs->bvec_pool); + else { + struct biovec_slab *bvs = bvec_slabs + idx; + + kmem_cache_free(bvs->slab, bv); + } +} + struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) { @@ -134,24 +239,22 @@ fallback: void bio_free(struct bio *bio, struct bio_set *bs) { - if (bio->bi_io_vec) { - const int pool_idx = BIO_POOL_IDX(bio); + void *p; - BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); - - if (pool_idx == BIOVEC_MAX_IDX) - mempool_free(bio->bi_io_vec, bs->bvec_pool); - else { - struct biovec_slab *bvs = bvec_slabs + pool_idx; - - kmem_cache_free(bvs->slab, bio->bi_io_vec); - } - } + if (bio->bi_io_vec) + bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio)); if (bio_integrity(bio)) bio_integrity_free(bio, bs); - mempool_free(bio, bs->bio_pool); + /* + * If we have front padding, adjust the bio pointer before freeing + */ + p = bio; + if (bs->front_pad) + p -= bs->front_pad; + + mempool_free(p, bs->bio_pool); } /* @@ -188,16 +291,20 @@ void bio_init(struct bio *bio) * for a &struct bio to become free. If a %NULL @bs is passed in, we will * fall back to just using @kmalloc to allocate the required memory. * - * allocate bio and iovecs from the memory pools specified by the - * bio_set structure, or @kmalloc if none given. + * Note that the caller must set ->bi_destructor on succesful return + * of a bio, to do the appropriate freeing of the bio once the reference + * count drops to zero. **/ struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) { - struct bio *bio; + struct bio *bio = NULL; + + if (bs) { + void *p = mempool_alloc(bs->bio_pool, gfp_mask); - if (bs) - bio = mempool_alloc(bs->bio_pool, gfp_mask); - else + if (p) + bio = p + bs->front_pad; + } else bio = kmalloc(sizeof(*bio), gfp_mask); if (likely(bio)) { @@ -1398,11 +1505,25 @@ void bioset_free(struct bio_set *bs) bioset_integrity_free(bs); biovec_free_pools(bs); + bio_put_slab(bs); kfree(bs); } -struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) +/** + * bioset_create - Create a bio_set + * @pool_size: Number of bio and bio_vecs to cache in the mempool + * @front_pad: Number of bytes to allocate in front of the returned bio + * + * Description: + * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller + * to ask for a number of bytes to be allocated in front of the bio. + * Front pad allocation is useful for embedding the bio inside + * another structure, to avoid allocating extra data to go with the bio. + * Note that the bio must be embedded at the END of that structure always, + * or things will break badly. + */ +struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) { struct bio_set *bs; @@ -1410,16 +1531,22 @@ struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) if (!bs) return NULL; - bs->bio_slab = bio_slab; + bs->front_pad = front_pad; - bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bs->bio_slab); + bs->bio_slab = bio_find_or_create_slab(front_pad); + if (!bs->bio_slab) { + kfree(bs); + return NULL; + } + + bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); if (!bs->bio_pool) goto bad; - if (bioset_integrity_create(bs, bio_pool_size)) + if (bioset_integrity_create(bs, pool_size)) goto bad; - if (!biovec_create_pools(bs, bvec_pool_size)) + if (!biovec_create_pools(bs, pool_size)) return bs; bad: @@ -1443,12 +1570,16 @@ static void __init biovec_init_slabs(void) static int __init init_bio(void) { - bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC); + bio_slab_max = 2; + bio_slab_nr = 0; + bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); + if (!bio_slabs) + panic("bio: can't allocate bios\n"); bio_integrity_init_slab(); biovec_init_slabs(); - fs_bio_set = bioset_create(BIO_POOL_SIZE, 2); + fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); if (!fs_bio_set) panic("bio: can't allocate bios\n"); diff --git a/include/linux/bio.h b/include/linux/bio.h index 9340098d75d..4b80d3537f9 100644 --- a/include/linux/bio.h +++ b/include/linux/bio.h @@ -334,7 +334,7 @@ struct bio_pair { extern struct bio_pair *bio_split(struct bio *bi, int first_sectors); extern void bio_pair_release(struct bio_pair *dbio); -extern struct bio_set *bioset_create(int, int); +extern struct bio_set *bioset_create(unsigned int, unsigned int); extern void bioset_free(struct bio_set *); extern struct bio *bio_alloc(gfp_t, int); @@ -379,6 +379,7 @@ extern struct bio *bio_copy_user_iov(struct request_queue *, extern int bio_uncopy_user(struct bio *); void zero_fill_bio(struct bio *bio); extern struct bio_vec *bvec_alloc_bs(gfp_t, int, unsigned long *, struct bio_set *); +extern void bvec_free_bs(struct bio_set *, struct bio_vec *, unsigned int); extern unsigned int bvec_nr_vecs(unsigned short idx); /* @@ -401,6 +402,8 @@ static inline void bio_set_completion_cpu(struct bio *bio, unsigned int cpu) struct bio_set { struct kmem_cache *bio_slab; + unsigned int front_pad; + mempool_t *bio_pool; #if defined(CONFIG_BLK_DEV_INTEGRITY) mempool_t *bio_integrity_pool; @@ -415,6 +418,7 @@ struct biovec_slab { }; extern struct bio_set *fs_bio_set; +extern struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly; /* * a small number of entries is fine, not going to be performance critical. -- cgit v1.2.3-70-g09d2 From 392ddc32982a5c661dd90dd49a3cb37f1c68b782 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Tue, 23 Dec 2008 12:42:54 +0100 Subject: bio: add support for inlining a number of bio_vecs inside the bio When we go and allocate a bio for IO, we actually do two allocations. One for the bio itself, and one for the bi_io_vec that holds the actual pages we are interested in. This feature inlines a definable amount of io vecs inside the bio itself, so we eliminate the bio_vec array allocation for IO's up to a certain size. It defaults to 4 vecs, which is typically 16k of IO. Signed-off-by: Jens Axboe --- fs/bio.c | 27 ++++++++++++++++++++++----- include/linux/bio.h | 12 ++++++++++++ 2 files changed, 34 insertions(+), 5 deletions(-) (limited to 'fs') diff --git a/fs/bio.c b/fs/bio.c index 0146f80789e..75e6be18ecd 100644 --- a/fs/bio.c +++ b/fs/bio.c @@ -31,6 +31,12 @@ DEFINE_TRACE(block_split); +/* + * Test patch to inline a certain number of bi_io_vec's inside the bio + * itself, to shrink a bio data allocation from two mempool calls to one + */ +#define BIO_INLINE_VECS 4 + static mempool_t *bio_split_pool __read_mostly; /* @@ -241,7 +247,7 @@ void bio_free(struct bio *bio, struct bio_set *bs) { void *p; - if (bio->bi_io_vec) + if (bio_has_allocated_vec(bio)) bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio)); if (bio_integrity(bio)) @@ -267,7 +273,8 @@ static void bio_fs_destructor(struct bio *bio) static void bio_kmalloc_destructor(struct bio *bio) { - kfree(bio->bi_io_vec); + if (bio_has_allocated_vec(bio)) + kfree(bio->bi_io_vec); kfree(bio); } @@ -314,7 +321,16 @@ struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) if (likely(nr_iovecs)) { unsigned long uninitialized_var(idx); - bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); + if (nr_iovecs <= BIO_INLINE_VECS) { + idx = 0; + bvl = bio->bi_inline_vecs; + nr_iovecs = BIO_INLINE_VECS; + memset(bvl, 0, BIO_INLINE_VECS * sizeof(*bvl)); + } else { + bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, + bs); + nr_iovecs = bvec_nr_vecs(idx); + } if (unlikely(!bvl)) { if (bs) mempool_free(bio, bs->bio_pool); @@ -324,7 +340,7 @@ struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) goto out; } bio->bi_flags |= idx << BIO_POOL_OFFSET; - bio->bi_max_vecs = bvec_nr_vecs(idx); + bio->bi_max_vecs = nr_iovecs; } bio->bi_io_vec = bvl; } @@ -1525,6 +1541,7 @@ void bioset_free(struct bio_set *bs) */ struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) { + unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); struct bio_set *bs; bs = kzalloc(sizeof(*bs), GFP_KERNEL); @@ -1533,7 +1550,7 @@ struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) bs->front_pad = front_pad; - bs->bio_slab = bio_find_or_create_slab(front_pad); + bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); if (!bs->bio_slab) { kfree(bs); return NULL; diff --git a/include/linux/bio.h b/include/linux/bio.h index 4b80d3537f9..18462c5b8ff 100644 --- a/include/linux/bio.h +++ b/include/linux/bio.h @@ -102,6 +102,13 @@ struct bio { #endif bio_destructor_t *bi_destructor; /* destructor */ + + /* + * We can inline a number of vecs at the end of the bio, to avoid + * double allocations for a small number of bio_vecs. This member + * MUST obviously be kept at the very end of the bio. + */ + struct bio_vec bi_inline_vecs[0]; }; /* @@ -213,6 +220,11 @@ static inline void *bio_data(struct bio *bio) return NULL; } +static inline int bio_has_allocated_vec(struct bio *bio) +{ + return bio->bi_io_vec && bio->bi_io_vec != bio->bi_inline_vecs; +} + /* * will die */ -- cgit v1.2.3-70-g09d2 From abf137dd7712132ee56d5b3143c2ff61a72a5faa Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Tue, 9 Dec 2008 08:11:22 +0100 Subject: aio: make the lookup_ioctx() lockless The mm->ioctx_list is currently protected by a reader-writer lock, so we always grab that lock on the read side for doing ioctx lookups. As the workload is extremely reader biased, turn this into an rcu hlist so we can make lookup_ioctx() lockless. Get rid of the rwlock and use a spinlock for providing update side exclusion. There's usually only 1 entry on this list, so it doesn't make sense to look into fancier data structures. Reviewed-by: Jeff Moyer Signed-off-by: Jens Axboe --- arch/s390/mm/pgtable.c | 4 +- fs/aio.c | 100 ++++++++++++++++++++++++++--------------------- include/linux/aio.h | 5 ++- include/linux/mm_types.h | 5 ++- kernel/fork.c | 4 +- 5 files changed, 67 insertions(+), 51 deletions(-) (limited to 'fs') diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c index ef3635b52fc..0767827540b 100644 --- a/arch/s390/mm/pgtable.c +++ b/arch/s390/mm/pgtable.c @@ -263,7 +263,7 @@ int s390_enable_sie(void) /* lets check if we are allowed to replace the mm */ task_lock(tsk); if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 || - tsk->mm != tsk->active_mm || tsk->mm->ioctx_list) { + tsk->mm != tsk->active_mm || !hlist_empty(&tsk->mm->ioctx_list)) { task_unlock(tsk); return -EINVAL; } @@ -279,7 +279,7 @@ int s390_enable_sie(void) /* Now lets check again if something happened */ task_lock(tsk); if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 || - tsk->mm != tsk->active_mm || tsk->mm->ioctx_list) { + tsk->mm != tsk->active_mm || !hlist_empty(&tsk->mm->ioctx_list)) { mmput(mm); task_unlock(tsk); return -EINVAL; diff --git a/fs/aio.c b/fs/aio.c index f658441d566..d6f89d3c15e 100644 --- a/fs/aio.c +++ b/fs/aio.c @@ -191,6 +191,20 @@ static int aio_setup_ring(struct kioctx *ctx) kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \ } while(0) +static void ctx_rcu_free(struct rcu_head *head) +{ + struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); + unsigned nr_events = ctx->max_reqs; + + kmem_cache_free(kioctx_cachep, ctx); + + if (nr_events) { + spin_lock(&aio_nr_lock); + BUG_ON(aio_nr - nr_events > aio_nr); + aio_nr -= nr_events; + spin_unlock(&aio_nr_lock); + } +} /* __put_ioctx * Called when the last user of an aio context has gone away, @@ -198,8 +212,6 @@ static int aio_setup_ring(struct kioctx *ctx) */ static void __put_ioctx(struct kioctx *ctx) { - unsigned nr_events = ctx->max_reqs; - BUG_ON(ctx->reqs_active); cancel_delayed_work(&ctx->wq); @@ -208,14 +220,7 @@ static void __put_ioctx(struct kioctx *ctx) mmdrop(ctx->mm); ctx->mm = NULL; pr_debug("__put_ioctx: freeing %p\n", ctx); - kmem_cache_free(kioctx_cachep, ctx); - - if (nr_events) { - spin_lock(&aio_nr_lock); - BUG_ON(aio_nr - nr_events > aio_nr); - aio_nr -= nr_events; - spin_unlock(&aio_nr_lock); - } + call_rcu(&ctx->rcu_head, ctx_rcu_free); } #define get_ioctx(kioctx) do { \ @@ -235,6 +240,7 @@ static struct kioctx *ioctx_alloc(unsigned nr_events) { struct mm_struct *mm; struct kioctx *ctx; + int did_sync = 0; /* Prevent overflows */ if ((nr_events > (0x10000000U / sizeof(struct io_event))) || @@ -267,21 +273,30 @@ static struct kioctx *ioctx_alloc(unsigned nr_events) goto out_freectx; /* limit the number of system wide aios */ - spin_lock(&aio_nr_lock); - if (aio_nr + ctx->max_reqs > aio_max_nr || - aio_nr + ctx->max_reqs < aio_nr) - ctx->max_reqs = 0; - else - aio_nr += ctx->max_reqs; - spin_unlock(&aio_nr_lock); + do { + spin_lock_bh(&aio_nr_lock); + if (aio_nr + nr_events > aio_max_nr || + aio_nr + nr_events < aio_nr) + ctx->max_reqs = 0; + else + aio_nr += ctx->max_reqs; + spin_unlock_bh(&aio_nr_lock); + if (ctx->max_reqs || did_sync) + break; + + /* wait for rcu callbacks to have completed before giving up */ + synchronize_rcu(); + did_sync = 1; + ctx->max_reqs = nr_events; + } while (1); + if (ctx->max_reqs == 0) goto out_cleanup; /* now link into global list. */ - write_lock(&mm->ioctx_list_lock); - ctx->next = mm->ioctx_list; - mm->ioctx_list = ctx; - write_unlock(&mm->ioctx_list_lock); + spin_lock(&mm->ioctx_lock); + hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); + spin_unlock(&mm->ioctx_lock); dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", ctx, ctx->user_id, current->mm, ctx->ring_info.nr); @@ -375,11 +390,12 @@ ssize_t wait_on_sync_kiocb(struct kiocb *iocb) */ void exit_aio(struct mm_struct *mm) { - struct kioctx *ctx = mm->ioctx_list; - mm->ioctx_list = NULL; - while (ctx) { - struct kioctx *next = ctx->next; - ctx->next = NULL; + struct kioctx *ctx; + + while (!hlist_empty(&mm->ioctx_list)) { + ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list); + hlist_del_rcu(&ctx->list); + aio_cancel_all(ctx); wait_for_all_aios(ctx); @@ -394,7 +410,6 @@ void exit_aio(struct mm_struct *mm) atomic_read(&ctx->users), ctx->dead, ctx->reqs_active); put_ioctx(ctx); - ctx = next; } } @@ -555,19 +570,21 @@ int aio_put_req(struct kiocb *req) static struct kioctx *lookup_ioctx(unsigned long ctx_id) { - struct kioctx *ioctx; - struct mm_struct *mm; + struct mm_struct *mm = current->mm; + struct kioctx *ctx = NULL; + struct hlist_node *n; - mm = current->mm; - read_lock(&mm->ioctx_list_lock); - for (ioctx = mm->ioctx_list; ioctx; ioctx = ioctx->next) - if (likely(ioctx->user_id == ctx_id && !ioctx->dead)) { - get_ioctx(ioctx); + rcu_read_lock(); + + hlist_for_each_entry_rcu(ctx, n, &mm->ioctx_list, list) { + if (ctx->user_id == ctx_id && !ctx->dead) { + get_ioctx(ctx); break; } - read_unlock(&mm->ioctx_list_lock); + } - return ioctx; + rcu_read_unlock(); + return ctx; } /* @@ -1215,19 +1232,14 @@ out: static void io_destroy(struct kioctx *ioctx) { struct mm_struct *mm = current->mm; - struct kioctx **tmp; int was_dead; /* delete the entry from the list is someone else hasn't already */ - write_lock(&mm->ioctx_list_lock); + spin_lock(&mm->ioctx_lock); was_dead = ioctx->dead; ioctx->dead = 1; - for (tmp = &mm->ioctx_list; *tmp && *tmp != ioctx; - tmp = &(*tmp)->next) - ; - if (*tmp) - *tmp = ioctx->next; - write_unlock(&mm->ioctx_list_lock); + hlist_del_rcu(&ioctx->list); + spin_unlock(&mm->ioctx_lock); dprintk("aio_release(%p)\n", ioctx); if (likely(!was_dead)) diff --git a/include/linux/aio.h b/include/linux/aio.h index f6b8cf99b59..b16a957030f 100644 --- a/include/linux/aio.h +++ b/include/linux/aio.h @@ -5,6 +5,7 @@ #include #include #include +#include #include @@ -183,7 +184,7 @@ struct kioctx { /* This needs improving */ unsigned long user_id; - struct kioctx *next; + struct hlist_node list; wait_queue_head_t wait; @@ -199,6 +200,8 @@ struct kioctx { struct aio_ring_info ring_info; struct delayed_work wq; + + struct rcu_head rcu_head; }; /* prototypes */ diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index fe825471d5a..9cfc9b627fd 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -232,8 +232,9 @@ struct mm_struct { struct core_state *core_state; /* coredumping support */ /* aio bits */ - rwlock_t ioctx_list_lock; /* aio lock */ - struct kioctx *ioctx_list; + spinlock_t ioctx_lock; + struct hlist_head ioctx_list; + #ifdef CONFIG_MM_OWNER /* * "owner" points to a task that is regarded as the canonical diff --git a/kernel/fork.c b/kernel/fork.c index 6144b36cd89..43cbf30669e 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -415,8 +415,8 @@ static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p) set_mm_counter(mm, file_rss, 0); set_mm_counter(mm, anon_rss, 0); spin_lock_init(&mm->page_table_lock); - rwlock_init(&mm->ioctx_list_lock); - mm->ioctx_list = NULL; + spin_lock_init(&mm->ioctx_lock); + INIT_HLIST_HEAD(&mm->ioctx_list); mm->free_area_cache = TASK_UNMAPPED_BASE; mm->cached_hole_size = ~0UL; mm_init_owner(mm, p); -- cgit v1.2.3-70-g09d2 From b3a6ffe16b5cc48abe7db8d04882dc45280eb693 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Fri, 12 Dec 2008 09:51:16 +0100 Subject: Get rid of CONFIG_LSF We have two seperate config entries for large devices/files. One is CONFIG_LBD that guards just the devices, the other is CONFIG_LSF that handles large files. This doesn't make a lot of sense, you typically want both or none. So get rid of CONFIG_LSF and change CONFIG_LBD wording to indicate that it covers both. Acked-by: Jean Delvare Signed-off-by: Jens Axboe --- block/Kconfig | 23 +++++------------------ fs/ext4/super.c | 8 ++++---- include/linux/types.h | 11 +++-------- 3 files changed, 12 insertions(+), 30 deletions(-) (limited to 'fs') diff --git a/block/Kconfig b/block/Kconfig index 290b219fad9..ac0956f7778 100644 --- a/block/Kconfig +++ b/block/Kconfig @@ -24,21 +24,17 @@ menuconfig BLOCK if BLOCK config LBD - bool "Support for Large Block Devices" + bool "Support for large block devices and files" depends on !64BIT help - Enable block devices of size 2TB and larger. + Enable block devices or files of size 2TB and larger. This option is required to support the full capacity of large (2TB+) block devices, including RAID, disk, Network Block Device, Logical Volume Manager (LVM) and loopback. - - For example, RAID devices are frequently bigger than the capacity - of the largest individual hard drive. - - This option is not required if you have individual disk drives - which total 2TB+ and you are not aggregating the capacity into - a large block device (e.g. using RAID or LVM). + + This option also enables support for single files larger than + 2TB. If unsure, say N. @@ -58,15 +54,6 @@ config BLK_DEV_IO_TRACE If unsure, say N. -config LSF - bool "Support for Large Single Files" - depends on !64BIT - help - Say Y here if you want to be able to handle very large files (2TB - and larger), otherwise say N. - - If unsure, say Y. - config BLK_DEV_BSG bool "Block layer SG support v4 (EXPERIMENTAL)" depends on EXPERIMENTAL diff --git a/fs/ext4/super.c b/fs/ext4/super.c index e4a241c65db..04158ad74db 100644 --- a/fs/ext4/super.c +++ b/fs/ext4/super.c @@ -1721,7 +1721,7 @@ static loff_t ext4_max_size(int blkbits, int has_huge_files) /* small i_blocks in vfs inode? */ if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) { /* - * CONFIG_LSF is not enabled implies the inode + * CONFIG_LBD is not enabled implies the inode * i_block represent total blocks in 512 bytes * 32 == size of vfs inode i_blocks * 8 */ @@ -1764,7 +1764,7 @@ static loff_t ext4_max_bitmap_size(int bits, int has_huge_files) if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) { /* - * !has_huge_files or CONFIG_LSF is not enabled + * !has_huge_files or CONFIG_LBD is not enabled * implies the inode i_block represent total blocks in * 512 bytes 32 == size of vfs inode i_blocks * 8 */ @@ -2021,13 +2021,13 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent) if (has_huge_files) { /* * Large file size enabled file system can only be - * mount if kernel is build with CONFIG_LSF + * mount if kernel is build with CONFIG_LBD */ if (sizeof(root->i_blocks) < sizeof(u64) && !(sb->s_flags & MS_RDONLY)) { printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge " "files cannot be mounted read-write " - "without CONFIG_LSF.\n", sb->s_id); + "without CONFIG_LBD.\n", sb->s_id); goto failed_mount; } } diff --git a/include/linux/types.h b/include/linux/types.h index 1d98330b1f2..121f349cb7e 100644 --- a/include/linux/types.h +++ b/include/linux/types.h @@ -135,19 +135,14 @@ typedef __s64 int64_t; * * Linux always considers sectors to be 512 bytes long independently * of the devices real block size. + * + * blkcnt_t is the type of the inode's block count. */ #ifdef CONFIG_LBD typedef u64 sector_t; -#else -typedef unsigned long sector_t; -#endif - -/* - * The type of the inode's block count. - */ -#ifdef CONFIG_LSF typedef u64 blkcnt_t; #else +typedef unsigned long sector_t; typedef unsigned long blkcnt_t; #endif -- cgit v1.2.3-70-g09d2 From d3f761104b097738932afcc310fbbbbfb007ef92 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Tue, 23 Dec 2008 12:46:21 +0100 Subject: bio: get rid of bio_vec clearing We don't need to clear the memory used for adding bio_vec entries, since nobody should be looking at members unitialized. Any valid use should be below bio->bi_vcnt, and that members up until that count must be valid since they were added through bio_add_page(). Signed-off-by: Jens Axboe --- fs/bio.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) (limited to 'fs') diff --git a/fs/bio.c b/fs/bio.c index 75e6be18ecd..711cee10360 100644 --- a/fs/bio.c +++ b/fs/bio.c @@ -180,7 +180,7 @@ struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, * kzalloc() for the exact number of vecs right away. */ if (!bs) - bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask); + bvl = kmalloc(nr * sizeof(struct bio_vec), gfp_mask); /* * see comment near bvec_array define! @@ -237,9 +237,6 @@ fallback: } } - if (bvl) - memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec)); - return bvl; } @@ -325,7 +322,6 @@ struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) idx = 0; bvl = bio->bi_inline_vecs; nr_iovecs = BIO_INLINE_VECS; - memset(bvl, 0, BIO_INLINE_VECS * sizeof(*bvl)); } else { bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); -- cgit v1.2.3-70-g09d2