Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1763240AbXHaWpp (ORCPT ); Fri, 31 Aug 2007 18:45:45 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1762829AbXHaWow (ORCPT ); Fri, 31 Aug 2007 18:44:52 -0400 Received: from mx1.redhat.com ([66.187.233.31]:52022 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1760968AbXHaWos (ORCPT ); Fri, 31 Aug 2007 18:44:48 -0400 Date: Fri, 31 Aug 2007 18:44:27 -0400 (EDT) Message-Id: <20070831.184427.112285010.k-ueda@ct.jp.nec.com> To: linux-kernel@vger.kernel.org, linux-scsi@vger.kernel.org, linux-ide@vger.kernel.org, jens.axboe@oracle.com Cc: dm-devel@redhat.com, j-nomura@ce.jp.nec.com, k-ueda@ct.jp.nec.com Subject: [APPENDIX PATCH 1/5] blk_end_request: request-based dm core From: Kiyoshi Ueda X-Mailer: Mew version 3.3 on Emacs 21.3 / Mule 5.0 (SAKAKI) Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 32135 Lines: 1065 This patch is an examle of block device stacking at request level, showing the necessity of blk_end_request() and how the new rq->end_io() hook is used. Request-based dm itself is still under development and not ready for inclusion. This patch adds request-based dm feature to dm core. Request-based dm hooks clone's ->end_io() to check errors of clone returned from device drivers. (See clone_end_request()) # Currently, request-based dm can be turned on by ioctl at dm device # creation time, so the userspace patches are needed. # The ioctl from userspace is ignored if kernel doesn't support it, # so please update userspace tools first when you try this. # (If kernel was updated first, you would hit kernel panic.) Signed-off-by: Kiyoshi Ueda Signed-off-by: Jun'ichi Nomura --- block/ll_rw_blk.c | 9 drivers/md/dm-hw-handler.h | 1 drivers/md/dm-ioctl.c | 5 drivers/md/dm-table.c | 23 + drivers/md/dm.c | 514 +++++++++++++++++++++++++++++++++++++++++- drivers/md/dm.h | 13 + drivers/scsi/scsi_lib.c | 38 +++ include/linux/blkdev.h | 6 include/linux/device-mapper.h | 35 ++ include/linux/dm-ioctl.h | 9 10 files changed, 638 insertions(+), 15 deletions(-) diff -rupN 07-change-end-io/block/ll_rw_blk.c a1-rqdm-core/block/ll_rw_blk.c --- 07-change-end-io/block/ll_rw_blk.c 2007-08-24 12:31:41.000000000 -0400 +++ a1-rqdm-core/block/ll_rw_blk.c 2007-08-29 13:53:12.000000000 -0400 @@ -177,6 +177,13 @@ void blk_queue_softirq_done(struct reque EXPORT_SYMBOL(blk_queue_softirq_done); +void blk_queue_device_congested(struct request_queue *q, device_congested_fn *fn) +{ + q->device_congested_fn = fn; +} + +EXPORT_SYMBOL_GPL(blk_queue_device_congested); + /** * blk_queue_make_request - define an alternate make_request function for a device * @q: the request queue for the device to be affected @@ -3692,7 +3699,7 @@ int blk_end_io(struct request *rq, int u struct request_queue *q = rq->q; unsigned long flags = 0UL; - if (blk_fs_request(rq) || blk_pc_request(rq)) { + if ((blk_fs_request(rq) || blk_pc_request(rq)) && !blk_cloned_rq(rq)) { if (__end_that_request_first(rq, uptodate, nr_bytes)) return 1; } diff -rupN 07-change-end-io/drivers/md/dm.c a1-rqdm-core/drivers/md/dm.c --- 07-change-end-io/drivers/md/dm.c 2007-08-13 00:25:24.000000000 -0400 +++ a1-rqdm-core/drivers/md/dm.c 2007-08-30 11:19:30.000000000 -0400 @@ -51,6 +51,22 @@ struct dm_target_io { union map_info info; }; +/* + * For request based dm. + * One of these is allocated per request. + * + * Since assuming "original request : cloned request = 1 : 1" and + * a counter for number of clones like struct dm_io.io_count isn't needed, + * struct dm_io and struct target_io can merge. + */ +struct dm_rq_target_io { + struct mapped_device *md; + int error; + struct request *rq; + struct dm_target *ti; + union map_info info; +}; + union map_info *dm_get_mapinfo(struct bio *bio) { if (bio && bio->bi_private) @@ -58,6 +74,14 @@ union map_info *dm_get_mapinfo(struct bi return NULL; } +union map_info *dm_get_rq_mapinfo(struct request *rq) +{ + if (rq && rq->end_io_data) + return &((struct dm_rq_target_io *)rq->end_io_data)->info; + return NULL; +} +EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); + #define MINOR_ALLOCED ((void *)-1) /* @@ -70,6 +94,13 @@ union map_info *dm_get_mapinfo(struct bi #define DMF_DELETING 4 #define DMF_NOFLUSH_SUSPENDING 5 +/* + * Bits for the md->features field. + */ +#define DM_FEAT_REQUEST_BASE (1 << 0) + +#define dm_feat_rq_base(md) ((md)->features & DM_FEAT_REQUEST_BASE) + struct mapped_device { struct rw_semaphore io_lock; struct semaphore suspend_lock; @@ -79,6 +110,7 @@ struct mapped_device { atomic_t open_count; unsigned long flags; + unsigned long features; struct request_queue *queue; struct gendisk *disk; @@ -121,11 +153,16 @@ struct mapped_device { /* forced geometry settings */ struct hd_geometry geometry; + + /* For saving the address of __make_request for request based dm */ + make_request_fn *saved_make_request_fn; }; #define MIN_IOS 256 static struct kmem_cache *_io_cache; static struct kmem_cache *_tio_cache; +static struct kmem_cache *_rq_cache; /* clone pool for request-based dm */ +static struct kmem_cache *_rq_tio_cache; /* target_io pool for request-based dm */ static int __init local_init(void) { @@ -143,9 +180,27 @@ static int __init local_init(void) return -ENOMEM; } + _rq_cache = kmem_cache_create("dm_rq", sizeof(struct request), + 0, 0, NULL); + if (!_rq_cache) { + kmem_cache_destroy(_tio_cache); + kmem_cache_destroy(_io_cache); + return -ENOMEM; + } + + _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); + if (!_rq_tio_cache) { + kmem_cache_destroy(_rq_cache); + kmem_cache_destroy(_tio_cache); + kmem_cache_destroy(_io_cache); + return -ENOMEM; + } + _major = major; r = register_blkdev(_major, _name); if (r < 0) { + kmem_cache_destroy(_rq_tio_cache); + kmem_cache_destroy(_rq_cache); kmem_cache_destroy(_tio_cache); kmem_cache_destroy(_io_cache); return r; @@ -159,6 +214,8 @@ static int __init local_init(void) static void local_exit(void) { + kmem_cache_destroy(_rq_tio_cache); + kmem_cache_destroy(_rq_cache); kmem_cache_destroy(_tio_cache); kmem_cache_destroy(_io_cache); unregister_blkdev(_major, _name); @@ -341,6 +398,27 @@ static void free_tio(struct mapped_devic mempool_free(tio, md->tio_pool); } +static inline struct request *alloc_rq(struct mapped_device *md) +{ + return mempool_alloc(md->io_pool, GFP_ATOMIC); +} + +static inline void free_rq(struct mapped_device *md, struct request *rq) +{ + mempool_free(rq, md->io_pool); +} + +static inline struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md) +{ + return mempool_alloc(md->tio_pool, GFP_ATOMIC); +} + +static inline void free_rq_tio(struct mapped_device *md, + struct dm_rq_target_io *tio) +{ + mempool_free(tio, md->tio_pool); +} + static void start_io_acct(struct dm_io *io) { struct mapped_device *md = io->md; @@ -434,6 +512,93 @@ int dm_set_geometry(struct mapped_device return 0; } +/* + * The queue is only valid as long as you have a reference + * count on 'md'. + */ +struct request_queue *dm_get_queue(struct mapped_device *md) +{ + if (blk_get_queue(md->queue)) + return NULL; + + return md->queue; +} +EXPORT_SYMBOL_GPL(dm_get_queue); + +void dm_put_queue(struct request_queue *q) +{ + blk_put_queue(q); +} +EXPORT_SYMBOL_GPL(dm_put_queue); + +void dm_stop_queue(struct request_queue *q) +{ + unsigned long flags; + + spin_lock_irqsave(q->queue_lock, flags); + blk_stop_queue(q); + spin_unlock_irqrestore(q->queue_lock, flags); +} +EXPORT_SYMBOL_GPL(dm_stop_queue); + +void dm_start_queue(struct request_queue *q) +{ + unsigned long flags; + + spin_lock_irqsave(q->queue_lock, flags); + if (blk_queue_stopped(q)) + blk_start_queue(q); + spin_unlock_irqrestore(q->queue_lock, flags); +} +EXPORT_SYMBOL_GPL(dm_start_queue); + +void __dm_requeue_request(struct request_queue *q, struct request *rq) +{ + if (elv_queue_empty(q)) + blk_plug_device(q); + blk_requeue_request(q, rq); +} +EXPORT_SYMBOL_GPL(__dm_requeue_request); + +void dm_requeue_request(struct request_queue *q, struct request *rq) +{ + unsigned long flags = 0UL; + + spin_lock_irqsave(q->queue_lock, flags); + __dm_requeue_request(q, rq); + spin_unlock_irqrestore(q->queue_lock, flags); +} +EXPORT_SYMBOL_GPL(dm_requeue_request); + +void dm_dispatch_request(struct request_queue *q, struct request *rq) +{ + int where = ELEVATOR_INSERT_BACK; + unsigned long flags; + + spin_lock_irqsave(q->queue_lock, flags); + + /* Initialize statistic stuff */ + disk_round_stats(rq->rq_disk); + rq->start_time = jiffies; + rq->rq_disk->in_flight++; + + __elv_add_request(q, rq, where, 0); + + spin_unlock_irqrestore(q->queue_lock, flags); +} +EXPORT_SYMBOL_GPL(dm_dispatch_request); + +static void kill_request(struct request *rq) +{ + int nr_bytes = rq->hard_nr_sectors << 9; + + if (!nr_bytes) + nr_bytes = rq->data_len; + + rq->cmd_flags |= REQ_QUIET; + blk_end_request(rq, 0, nr_bytes); +} + /*----------------------------------------------------------------- * CRUD START: * A more elegant soln is in the works that uses the queue @@ -533,6 +698,152 @@ static int clone_endio(struct bio *bio, return r; } +static void blk_update_cloned_rq(struct request *rq, struct request *clone) +{ + clone->nr_phys_segments = rq->nr_phys_segments; + clone->nr_hw_segments = rq->nr_hw_segments; + clone->current_nr_sectors = rq->current_nr_sectors; + clone->hard_cur_sectors = rq->hard_cur_sectors; + clone->hard_nr_sectors = rq->hard_nr_sectors; + clone->nr_sectors = rq->nr_sectors; + clone->hard_sector = rq->hard_sector; + clone->sector = rq->sector; + clone->data_len = rq->data_len; + clone->buffer = rq->buffer; + clone->data = rq->data; + clone->bio = rq->bio; + clone->biotail = rq->biotail; +} + +static void dec_rq_pending(struct dm_rq_target_io *tio) +{ + if (!atomic_dec_return(&tio->md->pending)) + /* nudge anyone waiting on suspend queue */ + wake_up(&tio->md->wait); +} + +static void finish_clone(struct request *clone, int needlock, + int (drv_callback)(struct request *)) +{ + if (!clone->q) + /* + * The clone was not dispatched into underlying devices and + * it means the caller is not underlying device driver, + * the caller should be dm. (e.g. dispatch_queued_ios() of + * dm-multipath) + * So no need to do here for this clone. + */ + return; + + /* + * The clone is *NOT* freed actually here because it is alloced from + * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. + * + * The uptodate and nr_bytes arguments of blk_end_io() don't matter + * because they aren't used for dm's clones. + */ + if (blk_end_io(clone, 1, 0, needlock, NULL, drv_callback)) + DMWARN("dm ignores the immediate return request of callback."); +} + +static void clean_clone(struct request *clone) +{ + clone->special = NULL; + clone->errors = 0; +} + +static int clone_end_request(struct request *clone, int uptodate, int nr_bytes, + int needlock, int (drv_callback)(struct request *)) +{ + int r = 0, error = 0, rw = rq_data_dir(clone); + struct dm_rq_target_io *tio = clone->end_io_data; + dm_request_endio_first_fn endio_first = tio->ti->type->rq_end_io_first; + dm_request_endio_fn endio = tio->ti->type->rq_end_io; + dm_request_queue_in_tgt_fn queue_in_tgt = tio->ti->type->queue_in_tgt; + struct request *orig = tio->rq; + struct request_queue *q = clone->q, *q_orig = orig->q; + + if (blk_fs_request(clone) && clone->rq_disk) + disk_stat_add(clone->rq_disk, sectors[rw], nr_bytes >> 9); + + if (end_io_error(uptodate)) + error = !uptodate ? -EIO : uptodate; + + if (endio_first) { + r = endio_first(tio->ti, clone, error, &tio->info); + switch (r) { + case 0: + /* succeeded */ + break; + case 1: + /* the target wants to handle the io without unmap */ + finish_clone(clone, needlock, drv_callback); + clean_clone(clone); + + if (!queue_in_tgt) { + DMERR("queue_in_tgt isn't implemented."); + BUG(); + } + queue_in_tgt(tio->ti, clone, &tio->info); + blk_run_queue(q_orig); + + return 0; + case 2: + /* The target wants to requeue the original request */ + if (needlock) + dm_requeue_request(q_orig, orig); + else + __dm_requeue_request(q_orig, orig); + + goto free_clone; + default: + if (r >= 0) { + DMWARN("unimplemented target endio return value: %d", r); + BUG(); + } + + /* + * the target detected error, but couldn't retry. + * direct the error to upper layer. + */ + uptodate = r; + break; + } + } + + /* Complete the original request's chunk */ + r = blk_end_request(orig, uptodate, nr_bytes); + + /* + * Recopy the original request fields that were updated + * in blk_end_request() to the clone. + */ + blk_update_cloned_rq(orig, clone); + + if (r) + /* The original request has leftover */ + return 1; + +free_clone: + /* + * Now the original request is completed and freed, or requeued. + * So + */ + + if (endio) + endio(tio->ti, clone, error, &tio->info); + + finish_clone(clone, needlock, drv_callback); + + blk_run_queue(q_orig); + + dec_rq_pending(tio); + free_rq(tio->md, clone); + free_rq_tio(tio->md, tio); + + return 0; +} + static sector_t max_io_len(struct mapped_device *md, sector_t sector, struct dm_target *ti) { @@ -844,6 +1155,166 @@ static int dm_request(struct request_que return 0; } +/* FIXME */ +static int dm_make_request(struct request_queue *q, struct bio *bio) +{ + int r = 0; + struct mapped_device *md = (struct mapped_device *)q->queuedata; + + r = md->saved_make_request_fn(q, bio); /* call __make_request() */ + + return r; +} + +static void setup_clone(struct request *clone, struct request *rq) +{ + INIT_LIST_HEAD(&clone->queuelist); + INIT_LIST_HEAD(&clone->donelist); + clone->q = NULL; + clone->cmd_flags = (rq_data_dir(rq) | REQ_NOMERGE | REQ_CLONED); + clone->cmd_type = rq->cmd_type; + clone->sector = rq->sector; + clone->hard_sector = rq->hard_sector; + clone->nr_sectors = rq->nr_sectors; + clone->hard_nr_sectors = rq->hard_nr_sectors; + clone->current_nr_sectors = rq->current_nr_sectors; + clone->hard_cur_sectors = rq->hard_cur_sectors; + clone->bio = rq->bio; + clone->biotail = rq->biotail; + INIT_HLIST_NODE(&clone->hash); +// RB_CLEAR_NODE(&clone->rb_node); + clone->completion_data = NULL; + clone->elevator_private = NULL; + clone->elevator_private2 = NULL; + clone->rq_disk = NULL; + clone->start_time = jiffies; + clone->nr_phys_segments = rq->nr_phys_segments; + clone->nr_hw_segments = rq->nr_hw_segments; + clone->ioprio = rq->ioprio; + clone->special = NULL; + clone->buffer = rq->buffer; + clone->tag = -1; + clone->errors = 0; + clone->ref_count = 1; + clone->cmd_len = rq->cmd_len; + memcpy(clone->cmd, rq->cmd, sizeof(rq->cmd)); + clone->data_len = rq->data_len; + clone->sense_len = rq->sense_len; + clone->data = rq->data; + clone->sense = rq->sense; + clone->timeout = 0; + clone->retries = 0; + clone->end_io = clone_end_request; + clone->end_io_data = NULL; +} + +int dm_underlying_device_congested(struct request_queue *q) +{ + if (q->device_congested_fn) + return q->device_congested_fn(q); + + return 0; +} +EXPORT_SYMBOL_GPL(dm_underlying_device_congested); + +static int clone_and_map_request(struct dm_target *ti, struct request *rq, + struct mapped_device *md) +{ + int r; + struct request *clone; + struct dm_rq_target_io *tio; + + tio = alloc_rq_tio(md); /* only one for each original request */ + if (!tio) + /* -ENOMEM */ + goto requeue; + tio->md = md; + tio->error = 0; + tio->rq = rq; + tio->ti = ti; + memset(&tio->info, 0, sizeof(tio->info)); + + clone = alloc_rq(md); + if (!clone) { + /* -ENOMEM */ + free_rq_tio(md, tio); + goto requeue; + } + setup_clone(clone, rq); + clone->end_io_data = tio; + + atomic_inc(&md->pending); + r = ti->type->map_rq(ti, clone, &tio->info); + switch (r) { + case 0: + /* the target has taken the request to submit by itself */ + break; + case 1: + /* the clone has been remapped so dispatch it */ + dm_dispatch_request(clone->q, clone); + break; + case 2: + /* the target has requested to requeue the original request */ + dec_rq_pending(tio); + free_rq_tio(md, tio); + free_rq(md, clone); + goto requeue; + default: + if (r >= 0) { + DMWARN("unimplemented target map return value: %d", r); + BUG(); + } + + dec_rq_pending(tio); + free_rq_tio(md, tio); + free_rq(md, clone); + kill_request(rq); + break; + } + + return 0; + +requeue: + return 1; +} + +/* + * q->request_fn for request based dm. + * called with q->queue_lock held + */ +static void dm_request_fn(struct request_queue *q) +{ + int r; + struct mapped_device *md = (struct mapped_device *)q->queuedata; + struct dm_table *map = dm_get_table(md); + struct dm_target *ti; + dm_congested_fn congested; + struct request *rq; + + while (!blk_queue_plugged(q)) { + rq = elv_next_request(q); + if (!rq) + break; + + ti = dm_table_find_target(map, rq->sector); + congested = ti->type->congested; + if (congested && congested(ti)) + break; + + blkdev_dequeue_request(rq); + spin_unlock(q->queue_lock); + r = clone_and_map_request(ti, rq, md); + spin_lock_irq(q->queue_lock); + + if (r) + __dm_requeue_request(q, rq); + } + + dm_table_put(map); + + return; +} + static int dm_flush_all(struct request_queue *q, struct gendisk *disk, sector_t *error_sector) { @@ -865,6 +1336,9 @@ static void dm_unplug_all(struct request struct dm_table *map = dm_get_table(md); if (map) { + if (dm_feat_rq_base(md)) + generic_unplug_device(q); + dm_table_unplug_all(map); dm_table_put(map); } @@ -966,7 +1440,7 @@ static struct block_device_operations dm /* * Allocate and initialise a blank device with a given minor. */ -static struct mapped_device *alloc_dev(int minor) +static struct mapped_device *alloc_dev(int minor, int request_base) { int r; struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); @@ -997,23 +1471,35 @@ static struct mapped_device *alloc_dev(i atomic_set(&md->open_count, 0); atomic_set(&md->event_nr, 0); - md->queue = blk_alloc_queue(GFP_KERNEL); + if (request_base) { + md->features |= DM_FEAT_REQUEST_BASE; + md->queue = blk_init_queue(dm_request_fn, NULL); + } else { + md->queue = blk_alloc_queue(GFP_KERNEL); + } if (!md->queue) goto bad1_free_minor; md->queue->queuedata = md; - md->queue->backing_dev_info.congested_fn = dm_any_congested; - md->queue->backing_dev_info.congested_data = md; - blk_queue_make_request(md->queue, dm_request); + if (request_base) { + md->saved_make_request_fn = md->queue->make_request_fn; + blk_queue_make_request(md->queue, dm_make_request); + } else { + md->queue->backing_dev_info.congested_fn = dm_any_congested; + md->queue->backing_dev_info.congested_data = md; + blk_queue_make_request(md->queue, dm_request); + } blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); md->queue->unplug_fn = dm_unplug_all; md->queue->issue_flush_fn = dm_flush_all; - md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); + md->io_pool = mempool_create_slab_pool(MIN_IOS, + request_base ? _rq_cache : _io_cache); if (!md->io_pool) goto bad2; - md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); + md->tio_pool = mempool_create_slab_pool(MIN_IOS, + request_base ? _rq_tio_cache : _tio_cache); if (!md->tio_pool) goto bad3; @@ -1154,11 +1640,12 @@ static void __unbind(struct mapped_devic /* * Constructor for a new device. */ -int dm_create(int minor, struct mapped_device **result) +int dm_create(int minor, struct mapped_device **result, unsigned create_flags) { struct mapped_device *md; + int request_base = create_flags & DM_CREATE_REQUEST_BASE_FLAG ? 1 : 0; - md = alloc_dev(minor); + md = alloc_dev(minor, request_base); if (!md) return -ENXIO; @@ -1383,9 +1870,13 @@ int dm_suspend(struct mapped_device *md, up_write(&md->io_lock); /* unplug */ - if (map) + if (map) { dm_table_unplug_all(map); + if (dm_feat_rq_base(md)) + dm_stop_queue(md->queue); + } + /* * Then we wait for the already mapped ios to * complete. @@ -1475,6 +1966,9 @@ int dm_resume(struct mapped_device *md) if (!map || !dm_table_get_size(map)) goto out; + if (dm_feat_rq_base(md)) + dm_start_queue(md->queue); + r = dm_table_resume_targets(map); if (r) goto out; diff -rupN 07-change-end-io/drivers/md/dm.h a1-rqdm-core/drivers/md/dm.h --- 07-change-end-io/drivers/md/dm.h 2007-08-13 00:25:24.000000000 -0400 +++ a1-rqdm-core/drivers/md/dm.h 2007-08-28 15:21:48.000000000 -0400 @@ -84,6 +84,11 @@ #define DM_SUSPEND_NOFLUSH_FLAG (1 << 1) /* + * Create feature flags + */ +#define DM_CREATE_REQUEST_BASE_FLAG (1 << 0) + +/* * List of devices that a metadevice uses and should open/close. */ struct dm_dev { @@ -112,6 +117,7 @@ int dm_table_resume_targets(struct dm_ta int dm_table_any_congested(struct dm_table *t, int bdi_bits); void dm_table_unplug_all(struct dm_table *t); int dm_table_flush_all(struct dm_table *t); +int dm_table_underlying_device_congested(struct dm_table *t); /*----------------------------------------------------------------- * A registry of target types. @@ -124,6 +130,12 @@ int dm_target_iterate(void (*iter_func)( void *param), void *param); /*----------------------------------------------------------------- + * Helper for block layer operations + *---------------------------------------------------------------*/ +void dm_dispatch_request(struct request_queue *q, struct request *rq); +int dm_underlying_device_congested(struct request_queue *q); + +/*----------------------------------------------------------------- * Useful inlines. *---------------------------------------------------------------*/ static inline int array_too_big(unsigned long fixed, unsigned long obj, @@ -180,6 +192,7 @@ void dm_stripe_exit(void); void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size); union map_info *dm_get_mapinfo(struct bio *bio); +union map_info *dm_get_rq_mapinfo(struct request *rq); int dm_open_count(struct mapped_device *md); int dm_lock_for_deletion(struct mapped_device *md); diff -rupN 07-change-end-io/drivers/md/dm-hw-handler.h a1-rqdm-core/drivers/md/dm-hw-handler.h --- 07-change-end-io/drivers/md/dm-hw-handler.h 2007-08-13 00:25:24.000000000 -0400 +++ a1-rqdm-core/drivers/md/dm-hw-handler.h 2007-08-28 15:21:48.000000000 -0400 @@ -35,6 +35,7 @@ struct hw_handler_type { void (*pg_init) (struct hw_handler *hwh, unsigned bypassed, struct dm_path *path); unsigned (*error) (struct hw_handler *hwh, struct bio *bio); + unsigned (*error_rq) (struct hw_handler *hwh, struct request *rq); int (*status) (struct hw_handler *hwh, status_type_t type, char *result, unsigned int maxlen); }; diff -rupN 07-change-end-io/drivers/md/dm-ioctl.c a1-rqdm-core/drivers/md/dm-ioctl.c --- 07-change-end-io/drivers/md/dm-ioctl.c 2007-08-13 00:25:24.000000000 -0400 +++ a1-rqdm-core/drivers/md/dm-ioctl.c 2007-08-28 15:21:48.000000000 -0400 @@ -565,6 +565,7 @@ static int dev_create(struct dm_ioctl *p { int r, m = DM_ANY_MINOR; struct mapped_device *md; + unsigned create_flags = 0; r = check_name(param->name); if (r) @@ -572,8 +573,10 @@ static int dev_create(struct dm_ioctl *p if (param->flags & DM_PERSISTENT_DEV_FLAG) m = MINOR(huge_decode_dev(param->dev)); + if (param->flags & DM_REQUEST_BASE_FLAG) + create_flags |= DM_CREATE_REQUEST_BASE_FLAG; - r = dm_create(m, &md); + r = dm_create(m, &md, create_flags); if (r) return r; diff -rupN 07-change-end-io/drivers/md/dm-table.c a1-rqdm-core/drivers/md/dm-table.c --- 07-change-end-io/drivers/md/dm-table.c 2007-08-22 18:54:04.000000000 -0400 +++ a1-rqdm-core/drivers/md/dm-table.c 2007-08-28 15:21:48.000000000 -0400 @@ -1025,6 +1025,29 @@ int dm_table_flush_all(struct dm_table * return ret; } +int dm_table_underlying_device_congested(struct dm_table *t) +{ + int r = 0; + struct list_head *d, *devices = dm_table_get_devices(t); + struct dm_dev *dd; + struct request_queue *q; + + for (d = devices->next; d != devices; d = d->next) { + dd = list_entry(d, struct dm_dev, list); + q = bdev_get_queue(dd->bdev); + + /* + * We should use "&=" and target driver should chose + * not congested device by using a table function like + * dm_table_uncongested_device(). + */ + r &= dm_underlying_device_congested(q); +// r |= dm_underlying_device_congested(q); + } + + return r; +} + struct mapped_device *dm_table_get_md(struct dm_table *t) { dm_get(t->md); diff -rupN 07-change-end-io/drivers/scsi/scsi_lib.c a1-rqdm-core/drivers/scsi/scsi_lib.c --- 07-change-end-io/drivers/scsi/scsi_lib.c 2007-08-24 12:26:06.000000000 -0400 +++ a1-rqdm-core/drivers/scsi/scsi_lib.c 2007-08-28 15:21:48.000000000 -0400 @@ -1401,6 +1401,43 @@ static void scsi_softirq_done(struct req } } +static int scsi_device_congested(struct request_queue *q) +{ + int r = 0; + struct scsi_device *sdev = q->queuedata; +// struct Scsi_Host *shost; +// unsigned long flags = 0UL; + + if (!sdev || !get_device(&sdev->sdev_gendev)) + /* + * Something may happened. We handle it in scsi_request_fn. + * Please dispatch the requests in the queue. + */ + return 0; + + if ((sdev->device_busy >= sdev->queue_depth) || sdev->device_blocked || + (sdev->single_lun && scsi_target(sdev)->starget_sdev_user && + scsi_target(sdev)->starget_sdev_user != sdev)) { + r = 1; + goto out; + } + +/* + shost = sdev->host; + spin_lock_irqsave(shost->host_lock, flags); + if (scsi_host_in_recovery(shost) || + (shost->can_queue > 0 && shost->host_busy >= shost->can_queue) || + shost->host_blocked || shost->host_self_blocked) + r = 1; + spin_unlock_irqrestore(shost->host_lock, flags); +*/ + +out: + put_device(&sdev->sdev_gendev); + + return r; +} + /* * Function: scsi_request_fn() * @@ -1590,6 +1627,7 @@ struct request_queue *scsi_alloc_queue(s blk_queue_prep_rq(q, scsi_prep_fn); blk_queue_softirq_done(q, scsi_softirq_done); + blk_queue_device_congested(q, scsi_device_congested); return q; } diff -rupN 07-change-end-io/include/linux/blkdev.h a1-rqdm-core/include/linux/blkdev.h --- 07-change-end-io/include/linux/blkdev.h 2007-08-24 12:32:44.000000000 -0400 +++ a1-rqdm-core/include/linux/blkdev.h 2007-08-29 13:53:12.000000000 -0400 @@ -203,6 +203,7 @@ enum rq_flag_bits { __REQ_RW_SYNC, /* request is sync (O_DIRECT) */ __REQ_ALLOCED, /* request came from our alloc pool */ __REQ_RW_META, /* metadata io request */ + __REQ_CLONED, /* request is a clone of another request */ __REQ_NR_BITS, /* stops here */ }; @@ -224,6 +225,7 @@ enum rq_flag_bits { #define REQ_RW_SYNC (1 << __REQ_RW_SYNC) #define REQ_ALLOCED (1 << __REQ_ALLOCED) #define REQ_RW_META (1 << __REQ_RW_META) +#define REQ_CLONED (1 << __REQ_CLONED) #define BLK_MAX_CDB 16 @@ -348,6 +350,7 @@ typedef int (merge_bvec_fn) (struct requ typedef int (issue_flush_fn) (struct request_queue *, struct gendisk *, sector_t *); typedef void (prepare_flush_fn) (struct request_queue *, struct request *); typedef void (softirq_done_fn)(struct request *); +typedef int (device_congested_fn) (struct request_queue *q); enum blk_queue_state { Queue_down, @@ -386,6 +389,7 @@ struct request_queue issue_flush_fn *issue_flush_fn; prepare_flush_fn *prepare_flush_fn; softirq_done_fn *softirq_done_fn; + device_congested_fn *device_congested_fn; /* * Dispatch queue sorting @@ -555,6 +559,7 @@ enum { #define blk_sorted_rq(rq) ((rq)->cmd_flags & REQ_SORTED) #define blk_barrier_rq(rq) ((rq)->cmd_flags & REQ_HARDBARRIER) #define blk_fua_rq(rq) ((rq)->cmd_flags & REQ_FUA) +#define blk_cloned_rq(rq) ((rq)->cmd_flags & REQ_CLONED) #define blk_bidi_rq(rq) ((rq)->next_rq != NULL) #define list_entry_rq(ptr) list_entry((ptr), struct request, queuelist) @@ -786,6 +791,7 @@ extern void blk_queue_prep_rq(struct req extern void blk_queue_merge_bvec(struct request_queue *, merge_bvec_fn *); extern void blk_queue_dma_alignment(struct request_queue *, int); extern void blk_queue_softirq_done(struct request_queue *, softirq_done_fn *); +extern void blk_queue_device_congested(struct request_queue *, device_congested_fn *); extern struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev); extern int blk_queue_ordered(struct request_queue *, unsigned, prepare_flush_fn *); extern void blk_queue_issue_flush_fn(struct request_queue *, issue_flush_fn *); diff -rupN 07-change-end-io/include/linux/device-mapper.h a1-rqdm-core/include/linux/device-mapper.h --- 07-change-end-io/include/linux/device-mapper.h 2007-08-13 00:25:24.000000000 -0400 +++ a1-rqdm-core/include/linux/device-mapper.h 2007-08-28 15:21:48.000000000 -0400 @@ -10,6 +10,8 @@ #ifdef __KERNEL__ +struct request; +struct request_queue; struct dm_target; struct dm_table; struct dm_dev; @@ -45,6 +47,9 @@ typedef void (*dm_dtr_fn) (struct dm_tar typedef int (*dm_map_fn) (struct dm_target *ti, struct bio *bio, union map_info *map_context); +typedef int (*dm_map_request_fn) (struct dm_target *ti, struct request *clone, + union map_info *map_context); + /* * Returns: * < 0 : error (currently ignored) @@ -57,6 +62,18 @@ typedef int (*dm_endio_fn) (struct dm_ta struct bio *bio, int error, union map_info *map_context); +typedef int (*dm_request_endio_first_fn) (struct dm_target *ti, + struct request *clone, int error, + union map_info *map_context); + +typedef int (*dm_request_endio_fn) (struct dm_target *ti, + struct request *clone, int error, + union map_info *map_context); + +typedef void (*dm_request_queue_in_tgt_fn) (struct dm_target *ti, + struct request *clone, + union map_info *map_context); + typedef void (*dm_flush_fn) (struct dm_target *ti); typedef void (*dm_presuspend_fn) (struct dm_target *ti); typedef void (*dm_postsuspend_fn) (struct dm_target *ti); @@ -71,6 +88,7 @@ typedef int (*dm_message_fn) (struct dm_ typedef int (*dm_ioctl_fn) (struct dm_target *ti, struct inode *inode, struct file *filp, unsigned int cmd, unsigned long arg); +typedef int (*dm_congested_fn) (struct dm_target *ti); void dm_error(const char *message); @@ -98,7 +116,11 @@ struct target_type { dm_ctr_fn ctr; dm_dtr_fn dtr; dm_map_fn map; + dm_map_request_fn map_rq; dm_endio_fn end_io; + dm_request_endio_first_fn rq_end_io_first; + dm_request_endio_fn rq_end_io; + dm_request_queue_in_tgt_fn queue_in_tgt; dm_flush_fn flush; dm_presuspend_fn presuspend; dm_postsuspend_fn postsuspend; @@ -107,6 +129,7 @@ struct target_type { dm_status_fn status; dm_message_fn message; dm_ioctl_fn ioctl; + dm_congested_fn congested; }; struct io_restrictions { @@ -142,6 +165,8 @@ struct dm_target { /* Used to provide an error string from the ctr */ char *error; + + struct request_queue *queue; }; int dm_register_target(struct target_type *t); @@ -157,7 +182,7 @@ int dm_unregister_target(struct target_t * DM_ANY_MINOR chooses the next available minor number. */ #define DM_ANY_MINOR (-1) -int dm_create(int minor, struct mapped_device **md); +int dm_create(int minor, struct mapped_device **md, unsigned create_flags); /* * Reference counting for md. @@ -167,6 +192,14 @@ void dm_get(struct mapped_device *md); void dm_put(struct mapped_device *md); /* + * Queue operations + */ +struct request_queue *dm_get_queue(struct mapped_device *md); +void dm_put_queue(struct request_queue *q); +void dm_stop_queue(struct request_queue *q); +void dm_start_queue(struct request_queue *q); + +/* * An arbitrary pointer may be stored alongside a mapped device. */ void dm_set_mdptr(struct mapped_device *md, void *ptr); diff -rupN 07-change-end-io/include/linux/dm-ioctl.h a1-rqdm-core/include/linux/dm-ioctl.h --- 07-change-end-io/include/linux/dm-ioctl.h 2007-08-13 00:25:24.000000000 -0400 +++ a1-rqdm-core/include/linux/dm-ioctl.h 2007-08-28 15:21:48.000000000 -0400 @@ -285,9 +285,9 @@ typedef char ioctl_struct[308]; #define DM_DEV_SET_GEOMETRY _IOWR(DM_IOCTL, DM_DEV_SET_GEOMETRY_CMD, struct dm_ioctl) #define DM_VERSION_MAJOR 4 -#define DM_VERSION_MINOR 11 +#define DM_VERSION_MINOR 12 #define DM_VERSION_PATCHLEVEL 0 -#define DM_VERSION_EXTRA "-ioctl (2006-10-12)" +#define DM_VERSION_EXTRA "-ioctl (2006-12-14)" /* Status bits */ #define DM_READONLY_FLAG (1 << 0) /* In/Out */ @@ -328,4 +328,9 @@ typedef char ioctl_struct[308]; */ #define DM_NOFLUSH_FLAG (1 << 11) /* In */ +/* + * Set this to create request based device-mapper device. + */ +#define DM_REQUEST_BASE_FLAG (1 << 12) /* In */ + #endif /* _LINUX_DM_IOCTL_H */ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/