From 242f9dcb8ba6f68fcd217a119a7648a4f69290e9 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Sun, 14 Sep 2008 05:55:09 -0700 Subject: block: unify request timeout handling Right now SCSI and others do their own command timeout handling. Move those bits to the block layer. Instead of having a timer per command, we try to be a bit more clever and simply have one per-queue. This avoids the overhead of having to tear down and setup a timer for each command, so it will result in a lot less timer fiddling. Signed-off-by: Mike Anderson Signed-off-by: Jens Axboe --- block/blk-timeout.c | 155 ++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 155 insertions(+) create mode 100644 block/blk-timeout.c (limited to 'block/blk-timeout.c') diff --git a/block/blk-timeout.c b/block/blk-timeout.c new file mode 100644 index 0000000..b36d07bf --- /dev/null +++ b/block/blk-timeout.c @@ -0,0 +1,155 @@ +/* + * Functions related to generic timeout handling of requests. + */ +#include +#include +#include + +#include "blk.h" + +/* + * blk_delete_timer - Delete/cancel timer for a given function. + * @req: request that we are canceling timer for + * + */ +void blk_delete_timer(struct request *req) +{ + struct request_queue *q = req->q; + + /* + * Nothing to detach + */ + if (!q->rq_timed_out_fn || !req->deadline) + return; + + list_del_init(&req->timeout_list); + + if (list_empty(&q->timeout_list)) + del_timer(&q->timeout); +} + +static void blk_rq_timed_out(struct request *req) +{ + struct request_queue *q = req->q; + enum blk_eh_timer_return ret; + + ret = q->rq_timed_out_fn(req); + switch (ret) { + case BLK_EH_HANDLED: + __blk_complete_request(req); + break; + case BLK_EH_RESET_TIMER: + blk_clear_rq_complete(req); + blk_add_timer(req); + break; + case BLK_EH_NOT_HANDLED: + /* + * LLD handles this for now but in the future + * we can send a request msg to abort the command + * and we can move more of the generic scsi eh code to + * the blk layer. + */ + break; + default: + printk(KERN_ERR "block: bad eh return: %d\n", ret); + break; + } +} + +void blk_rq_timed_out_timer(unsigned long data) +{ + struct request_queue *q = (struct request_queue *) data; + unsigned long flags, uninitialized_var(next), next_set = 0; + struct request *rq, *tmp; + + spin_lock_irqsave(q->queue_lock, flags); + + list_for_each_entry_safe(rq, tmp, &q->timeout_list, timeout_list) { + if (time_after_eq(jiffies, rq->deadline)) { + list_del_init(&rq->timeout_list); + + /* + * Check if we raced with end io completion + */ + if (blk_mark_rq_complete(rq)) + continue; + blk_rq_timed_out(rq); + } + if (!next_set) { + next = rq->deadline; + next_set = 1; + } else if (time_after(next, rq->deadline)) + next = rq->deadline; + } + + if (next_set && !list_empty(&q->timeout_list)) + mod_timer(&q->timeout, round_jiffies(next)); + + spin_unlock_irqrestore(q->queue_lock, flags); +} + +/** + * blk_abort_request -- Request request recovery for the specified command + * @req: pointer to the request of interest + * + * This function requests that the block layer start recovery for the + * request by deleting the timer and calling the q's timeout function. + * LLDDs who implement their own error recovery MAY ignore the timeout + * event if they generated blk_abort_req. Must hold queue lock. + */ +void blk_abort_request(struct request *req) +{ + blk_delete_timer(req); + blk_rq_timed_out(req); +} +EXPORT_SYMBOL_GPL(blk_abort_request); + +/** + * blk_add_timer - Start timeout timer for a single request + * @req: request that is about to start running. + * + * Notes: + * Each request has its own timer, and as it is added to the queue, we + * set up the timer. When the request completes, we cancel the timer. + */ +void blk_add_timer(struct request *req) +{ + struct request_queue *q = req->q; + unsigned long expiry; + + if (!q->rq_timed_out_fn) + return; + + BUG_ON(!list_empty(&req->timeout_list)); + BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); + + if (req->timeout) + req->deadline = jiffies + req->timeout; + else { + req->deadline = jiffies + q->rq_timeout; + /* + * Some LLDs, like scsi, peek at the timeout to prevent + * a command from being retried forever. + */ + req->timeout = q->rq_timeout; + } + list_add_tail(&req->timeout_list, &q->timeout_list); + + /* + * If the timer isn't already pending or this timeout is earlier + * than an existing one, modify the timer. Round to next nearest + * second. + */ + expiry = round_jiffies(req->deadline); + + /* + * We use ->deadline == 0 to detect whether a timer was added or + * not, so just increase to next jiffy for that specific case + */ + if (unlikely(!req->deadline)) + req->deadline = 1; + + if (!timer_pending(&q->timeout) || + time_before(expiry, q->timeout.expires)) + mod_timer(&q->timeout, expiry); +} -- cgit v1.1 From 11914a53d2ec2974a565311af327b8983d8c820d Mon Sep 17 00:00:00 2001 From: Mike Anderson Date: Sat, 13 Sep 2008 20:31:27 +0200 Subject: block: Add interface to abort queued requests Signed-off-by: Mike Anderson Signed-off-by: Jens Axboe --- block/blk-timeout.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) (limited to 'block/blk-timeout.c') diff --git a/block/blk-timeout.c b/block/blk-timeout.c index b36d07bf..6e5c781 100644 --- a/block/blk-timeout.c +++ b/block/blk-timeout.c @@ -153,3 +153,25 @@ void blk_add_timer(struct request *req) time_before(expiry, q->timeout.expires)) mod_timer(&q->timeout, expiry); } + +/** + * blk_abort_queue -- Abort all request on given queue + * @queue: pointer to queue + * + */ +void blk_abort_queue(struct request_queue *q) +{ + unsigned long flags; + struct request *rq, *tmp; + + spin_lock_irqsave(q->queue_lock, flags); + + elv_abort_queue(q); + + list_for_each_entry_safe(rq, tmp, &q->timeout_list, timeout_list) + blk_abort_request(rq); + + spin_unlock_irqrestore(q->queue_lock, flags); + +} +EXPORT_SYMBOL_GPL(blk_abort_queue); -- cgit v1.1 From 581d4e28d9195aa8b2231383dbabc288988d615e Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Sun, 14 Sep 2008 05:56:33 -0700 Subject: block: add fault injection mechanism for faking request timeouts Only works for the generic request timer handling. Allows one to sporadically ignore request completions, thus exercising the timeout handling. Signed-off-by: Jens Axboe --- block/blk-timeout.c | 59 +++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 59 insertions(+) (limited to 'block/blk-timeout.c') diff --git a/block/blk-timeout.c b/block/blk-timeout.c index 6e5c781..9b4ad13 100644 --- a/block/blk-timeout.c +++ b/block/blk-timeout.c @@ -4,9 +4,68 @@ #include #include #include +#include #include "blk.h" +#ifdef CONFIG_FAIL_IO_TIMEOUT + +static DECLARE_FAULT_ATTR(fail_io_timeout); + +static int __init setup_fail_io_timeout(char *str) +{ + return setup_fault_attr(&fail_io_timeout, str); +} +__setup("fail_io_timeout=", setup_fail_io_timeout); + +int blk_should_fake_timeout(struct request_queue *q) +{ + if (!test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags)) + return 0; + + return should_fail(&fail_io_timeout, 1); +} + +static int __init fail_io_timeout_debugfs(void) +{ + return init_fault_attr_dentries(&fail_io_timeout, "fail_io_timeout"); +} + +late_initcall(fail_io_timeout_debugfs); + +ssize_t part_timeout_show(struct device *dev, struct device_attribute *attr, + char *buf) +{ + struct gendisk *disk = dev_to_disk(dev); + int set = test_bit(QUEUE_FLAG_FAIL_IO, &disk->queue->queue_flags); + + return sprintf(buf, "%d\n", set != 0); +} + +ssize_t part_timeout_store(struct device *dev, struct device_attribute *attr, + const char *buf, size_t count) +{ + struct gendisk *disk = dev_to_disk(dev); + int val; + + if (count) { + struct request_queue *q = disk->queue; + char *p = (char *) buf; + + val = simple_strtoul(p, &p, 10); + spin_lock_irq(q->queue_lock); + if (val) + queue_flag_set(QUEUE_FLAG_FAIL_IO, q); + else + queue_flag_clear(QUEUE_FLAG_FAIL_IO, q); + spin_unlock_irq(q->queue_lock); + } + + return count; +} + +#endif /* CONFIG_FAIL_IO_TIMEOUT */ + /* * blk_delete_timer - Delete/cancel timer for a given function. * @req: request that we are canceling timer for -- cgit v1.1 From 7ba1fbaa4a478f72fbaf5a56af9c82a77966b4c7 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Tue, 16 Sep 2008 09:54:11 -0700 Subject: block: use rq complete marking in blk_abort_request() We cannot abort a request if we raced with the timeout handler already, or with the IO completion. So make blk_abort_request() mark the request as complete, and only continue if we succeeded. Found and suggested by Mike Anderson Signed-off-by: Jens Axboe --- block/blk-timeout.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'block/blk-timeout.c') diff --git a/block/blk-timeout.c b/block/blk-timeout.c index 9b4ad13..972a63f 100644 --- a/block/blk-timeout.c +++ b/block/blk-timeout.c @@ -158,6 +158,8 @@ void blk_rq_timed_out_timer(unsigned long data) */ void blk_abort_request(struct request *req) { + if (blk_mark_rq_complete(req)) + return; blk_delete_timer(req); blk_rq_timed_out(req); } -- cgit v1.1