block/blk-timeout.c

   1 /*
   2  * Functions related to generic timeout handling of requests.
   3  */
   4 #include <linux/kernel.h>
   5 #include <linux/module.h>
   6 #include <linux/blkdev.h>
   7 #include <linux/fault-inject.h>
   8
   9 #include "blk.h"
  10 #include "blk-mq.h"
  11
  12 #ifdef CONFIG_FAIL_IO_TIMEOUT
  13
  14 static DECLARE_FAULT_ATTR(fail_io_timeout);
  15
  16 static int __init setup_fail_io_timeout(char *str)
  17 {
  18         return setup_fault_attr(&fail_io_timeout, str);
  19 }
  20 __setup("fail_io_timeout=", setup_fail_io_timeout);
  21
  22 int blk_should_fake_timeout(struct request_queue *q)
  23 {
  24         if (!test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
  25                 return 0;
  26
  27         return should_fail(&fail_io_timeout, 1);
  28 }
  29
  30 static int __init fail_io_timeout_debugfs(void)
  31 {
  32         struct dentry *dir = fault_create_debugfs_attr("fail_io_timeout",
  33                                                 NULL, &fail_io_timeout);
  34
  35         return PTR_ERR_OR_ZERO(dir);
  36 }
  37
  38 late_initcall(fail_io_timeout_debugfs);
  39
  40 ssize_t part_timeout_show(struct device *dev, struct device_attribute *attr,
  41                           char *buf)
  42 {
  43         struct gendisk *disk = dev_to_disk(dev);
  44         int set = test_bit(QUEUE_FLAG_FAIL_IO, &disk->queue->queue_flags);
  45
  46         return sprintf(buf, "%d\n", set != 0);
  47 }
  48
  49 ssize_t part_timeout_store(struct device *dev, struct device_attribute *attr,
  50                            const char *buf, size_t count)
  51 {
  52         struct gendisk *disk = dev_to_disk(dev);
  53         int val;
  54
  55         if (count) {
  56                 struct request_queue *q = disk->queue;
  57                 char *p = (char *) buf;
  58
  59                 val = simple_strtoul(p, &p, 10);
  60                 spin_lock_irq(q->queue_lock);
  61                 if (val)
  62                         queue_flag_set(QUEUE_FLAG_FAIL_IO, q);
  63                 else
  64                         queue_flag_clear(QUEUE_FLAG_FAIL_IO, q);
  65                 spin_unlock_irq(q->queue_lock);
  66         }
  67
  68         return count;
  69 }
  70
  71 #endif /* CONFIG_FAIL_IO_TIMEOUT */
  72
  73 /*
  74  * blk_delete_timer - Delete/cancel timer for a given function.
  75  * @req:        request that we are canceling timer for
  76  *
  77  */
  78 void blk_delete_timer(struct request *req)
  79 {
  80         list_del_init(&req->timeout_list);
  81 }
  82
  83 static void blk_rq_timed_out(struct request *req)
  84 {
  85         struct request_queue *q = req->q;
  86         enum blk_eh_timer_return ret = BLK_EH_RESET_TIMER;
  87
  88         if (q->rq_timed_out_fn)
  89                 ret = q->rq_timed_out_fn(req);
  90         switch (ret) {
  91         case BLK_EH_HANDLED:
  92                 __blk_complete_request(req);
  93                 break;
  94         case BLK_EH_RESET_TIMER:
  95                 blk_add_timer(req);
  96                 blk_clear_rq_complete(req);
  97                 break;
  98         case BLK_EH_NOT_HANDLED:
  99                 /*
 100                  * LLD handles this for now but in the future
 101                  * we can send a request msg to abort the command
 102                  * and we can move more of the generic scsi eh code to
 103                  * the blk layer.
 104                  */
 105                 break;
 106         default:
 107                 printk(KERN_ERR "block: bad eh return: %d\n", ret);
 108                 break;
 109         }
 110 }
 111
 112 static void blk_rq_check_expired(struct request *rq, unsigned long *next_timeout,
 113                           unsigned int *next_set)
 114 {
 115         if (time_after_eq(jiffies, rq->deadline)) {
 116                 list_del_init(&rq->timeout_list);
 117
 118                 /*
 119                  * Check if we raced with end io completion
 120                  */
 121                 if (!blk_mark_rq_complete(rq))
 122                         blk_rq_timed_out(rq);
 123         } else if (!*next_set || time_after(*next_timeout, rq->deadline)) {
 124                 *next_timeout = rq->deadline;
 125                 *next_set = 1;
 126         }
 127 }
 128
 129 void blk_timeout_work(struct work_struct *work)
 130 {
 131         struct request_queue *q =
 132                 container_of(work, struct request_queue, timeout_work);
 133         unsigned long flags, next = 0;
 134         struct request *rq, *tmp;
 135         int next_set = 0;
 136
 137         spin_lock_irqsave(q->queue_lock, flags);
 138
 139         list_for_each_entry_safe(rq, tmp, &q->timeout_list, timeout_list)
 140                 blk_rq_check_expired(rq, &next, &next_set);
 141
 142         if (next_set)
 143                 mod_timer(&q->timeout, round_jiffies_up(next));
 144
 145         spin_unlock_irqrestore(q->queue_lock, flags);
 146 }
 147
 148 /**
 149  * blk_abort_request -- Request request recovery for the specified command
 150  * @req:        pointer to the request of interest
 151  *
 152  * This function requests that the block layer start recovery for the
 153  * request by deleting the timer and calling the q's timeout function.
 154  * LLDDs who implement their own error recovery MAY ignore the timeout
 155  * event if they generated blk_abort_req. Must hold queue lock.
 156  */
 157 void blk_abort_request(struct request *req)
 158 {
 159         if (blk_mark_rq_complete(req))
 160                 return;
 161
 162         if (req->q->mq_ops) {
 163                 blk_mq_rq_timed_out(req, false);
 164         } else {
 165                 blk_delete_timer(req);
 166                 blk_rq_timed_out(req);
 167         }
 168 }
 169 EXPORT_SYMBOL_GPL(blk_abort_request);
 170
 171 unsigned long blk_rq_timeout(unsigned long timeout)
 172 {
 173         unsigned long maxt;
 174
 175         maxt = round_jiffies_up(jiffies + BLK_MAX_TIMEOUT);
 176         if (time_after(timeout, maxt))
 177                 timeout = maxt;
 178
 179         return timeout;
 180 }
 181
 182 /**
 183  * blk_add_timer - Start timeout timer for a single request
 184  * @req:        request that is about to start running.
 185  *
 186  * Notes:
 187  *    Each request has its own timer, and as it is added to the queue, we
 188  *    set up the timer. When the request completes, we cancel the timer.
 189  */
 190 void blk_add_timer(struct request *req)
 191 {
 192         struct request_queue *q = req->q;
 193         unsigned long expiry;
 194
 195         if (!q->mq_ops)
 196                 lockdep_assert_held(q->queue_lock);
 197
 198         /* blk-mq has its own handler, so we don't need ->rq_timed_out_fn */
 199         if (!q->mq_ops && !q->rq_timed_out_fn)
 200                 return;
 201
 202         BUG_ON(!list_empty(&req->timeout_list));
 203
 204         /*
 205          * Some LLDs, like scsi, peek at the timeout to prevent a
 206          * command from being retried forever.
 207          */
 208         if (!req->timeout)
 209                 req->timeout = q->rq_timeout;
 210
 211         WRITE_ONCE(req->deadline, jiffies + req->timeout);
 212
 213         /*
 214          * Only the non-mq case needs to add the request to a protected list.
 215          * For the mq case we simply scan the tag map.
 216          */
 217         if (!q->mq_ops)
 218                 list_add_tail(&req->timeout_list, &req->q->timeout_list);
 219
 220         /*
 221          * If the timer isn't already pending or this timeout is earlier
 222          * than an existing one, modify the timer. Round up to next nearest
 223          * second.
 224          */
 225         expiry = blk_rq_timeout(round_jiffies_up(req->deadline));
 226
 227         if (!timer_pending(&q->timeout) ||
 228             time_before(expiry, q->timeout.expires)) {
 229                 unsigned long diff = q->timeout.expires - expiry;
 230
 231                 /*
 232                  * Due to added timer slack to group timers, the timer
 233                  * will often be a little in front of what we asked for.
 234                  * So apply some tolerance here too, otherwise we keep
 235                  * modifying the timer because expires for value X
 236                  * will be X + something.
 237                  */
 238                 if (!timer_pending(&q->timeout) || (diff >= HZ / 2))
 239                         mod_timer(&q->timeout, expiry);
 240         }
 241
 242 }