block/blk-barrier.c

   1 /*
   2  * Functions related to barrier IO handling
   3  */
   4 #include <linux/kernel.h>
   5 #include <linux/module.h>
   6 #include <linux/bio.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/gfp.h>
   9
  10 #include "blk.h"
  11
  12 /*
  13  * Cache flushing for ordered writes handling
  14  */
  15 unsigned blk_ordered_cur_seq(struct request_queue *q)
  16 {
  17         if (!q->ordseq)
  18                 return 0;
  19         return 1 << ffz(q->ordseq);
  20 }
  21
  22 unsigned blk_ordered_req_seq(struct request *rq)
  23 {
  24         struct request_queue *q = rq->q;
  25
  26         BUG_ON(q->ordseq == 0);
  27
  28         if (rq == &q->pre_flush_rq)
  29                 return QUEUE_ORDSEQ_PREFLUSH;
  30         if (rq == &q->bar_rq)
  31                 return QUEUE_ORDSEQ_BAR;
  32         if (rq == &q->post_flush_rq)
  33                 return QUEUE_ORDSEQ_POSTFLUSH;
  34
  35         /*
  36          * !fs requests don't need to follow barrier ordering.  Always
  37          * put them at the front.  This fixes the following deadlock.
  38          *
  39          * http://thread.gmane.org/gmane.linux.kernel/537473
  40          */
  41         if (rq->cmd_type != REQ_TYPE_FS)
  42                 return QUEUE_ORDSEQ_DRAIN;
  43
  44         if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
  45             (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
  46                 return QUEUE_ORDSEQ_DRAIN;
  47         else
  48                 return QUEUE_ORDSEQ_DONE;
  49 }
  50
  51 bool blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error)
  52 {
  53         struct request *rq;
  54
  55         if (error && !q->orderr)
  56                 q->orderr = error;
  57
  58         BUG_ON(q->ordseq & seq);
  59         q->ordseq |= seq;
  60
  61         if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
  62                 return false;
  63
  64         /*
  65          * Okay, sequence complete.
  66          */
  67         q->ordseq = 0;
  68         rq = q->orig_bar_rq;
  69         __blk_end_request_all(rq, q->orderr);
  70         return true;
  71 }
  72
  73 static void pre_flush_end_io(struct request *rq, int error)
  74 {
  75         elv_completed_request(rq->q, rq);
  76         blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
  77 }
  78
  79 static void bar_end_io(struct request *rq, int error)
  80 {
  81         elv_completed_request(rq->q, rq);
  82         blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
  83 }
  84
  85 static void post_flush_end_io(struct request *rq, int error)
  86 {
  87         elv_completed_request(rq->q, rq);
  88         blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
  89 }
  90
  91 static void queue_flush(struct request_queue *q, unsigned which)
  92 {
  93         struct request *rq;
  94         rq_end_io_fn *end_io;
  95
  96         if (which == QUEUE_ORDERED_DO_PREFLUSH) {
  97                 rq = &q->pre_flush_rq;
  98                 end_io = pre_flush_end_io;
  99         } else {
 100                 rq = &q->post_flush_rq;
 101                 end_io = post_flush_end_io;
 102         }
 103
 104         blk_rq_init(q, rq);
 105         rq->cmd_type = REQ_TYPE_FS;
 106         rq->cmd_flags = REQ_HARDBARRIER | REQ_FLUSH;
 107         rq->rq_disk = q->orig_bar_rq->rq_disk;
 108         rq->end_io = end_io;
 109
 110         elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
 111 }
 112
 113 static inline bool start_ordered(struct request_queue *q, struct request **rqp)
 114 {
 115         struct request *rq = *rqp;
 116         unsigned skip = 0;
 117
 118         q->orderr = 0;
 119         q->ordered = q->next_ordered;
 120         q->ordseq |= QUEUE_ORDSEQ_STARTED;
 121
 122         /*
 123          * For an empty barrier, there's no actual BAR request, which
 124          * in turn makes POSTFLUSH unnecessary.  Mask them off.
 125          */
 126         if (!blk_rq_sectors(rq))
 127                 q->ordered &= ~(QUEUE_ORDERED_DO_BAR |
 128                                 QUEUE_ORDERED_DO_POSTFLUSH);
 129
 130         /* stash away the original request */
 131         blk_dequeue_request(rq);
 132         q->orig_bar_rq = rq;
 133         rq = NULL;
 134
 135         /*
 136          * Queue ordered sequence.  As we stack them at the head, we
 137          * need to queue in reverse order.  Note that we rely on that
 138          * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
 139          * request gets inbetween ordered sequence.
 140          */
 141         if (q->ordered & QUEUE_ORDERED_DO_POSTFLUSH) {
 142                 queue_flush(q, QUEUE_ORDERED_DO_POSTFLUSH);
 143                 rq = &q->post_flush_rq;
 144         } else
 145                 skip |= QUEUE_ORDSEQ_POSTFLUSH;
 146
 147         if (q->ordered & QUEUE_ORDERED_DO_BAR) {
 148                 rq = &q->bar_rq;
 149
 150                 /* initialize proxy request and queue it */
 151                 blk_rq_init(q, rq);
 152                 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
 153                         rq->cmd_flags |= REQ_WRITE;
 154                 if (q->ordered & QUEUE_ORDERED_DO_FUA)
 155                         rq->cmd_flags |= REQ_FUA;
 156                 init_request_from_bio(rq, q->orig_bar_rq->bio);
 157                 rq->end_io = bar_end_io;
 158
 159                 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
 160         } else
 161                 skip |= QUEUE_ORDSEQ_BAR;
 162
 163         if (q->ordered & QUEUE_ORDERED_DO_PREFLUSH) {
 164                 queue_flush(q, QUEUE_ORDERED_DO_PREFLUSH);
 165                 rq = &q->pre_flush_rq;
 166         } else
 167                 skip |= QUEUE_ORDSEQ_PREFLUSH;
 168
 169         if (queue_in_flight(q))
 170                 rq = NULL;
 171         else
 172                 skip |= QUEUE_ORDSEQ_DRAIN;
 173
 174         *rqp = rq;
 175
 176         /*
 177          * Complete skipped sequences.  If whole sequence is complete,
 178          * return false to tell elevator that this request is gone.
 179          */
 180         return !blk_ordered_complete_seq(q, skip, 0);
 181 }
 182
 183 bool blk_do_ordered(struct request_queue *q, struct request **rqp)
 184 {
 185         struct request *rq = *rqp;
 186         const int is_barrier = rq->cmd_type == REQ_TYPE_FS &&
 187                                 (rq->cmd_flags & REQ_HARDBARRIER);
 188
 189         if (!q->ordseq) {
 190                 if (!is_barrier)
 191                         return true;
 192
 193                 if (q->next_ordered != QUEUE_ORDERED_NONE)
 194                         return start_ordered(q, rqp);
 195                 else {
 196                         /*
 197                          * Queue ordering not supported.  Terminate
 198                          * with prejudice.
 199                          */
 200                         blk_dequeue_request(rq);
 201                         __blk_end_request_all(rq, -EOPNOTSUPP);
 202                         *rqp = NULL;
 203                         return false;
 204                 }
 205         }
 206
 207         /*
 208          * Ordered sequence in progress
 209          */
 210
 211         /* Special requests are not subject to ordering rules. */
 212         if (rq->cmd_type != REQ_TYPE_FS &&
 213             rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
 214                 return true;
 215
 216         /* Ordered by draining.  Wait for turn. */
 217         WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
 218         if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
 219                 *rqp = NULL;
 220
 221         return true;
 222 }
 223
 224 static void bio_end_empty_barrier(struct bio *bio, int err)
 225 {
 226         if (err) {
 227                 if (err == -EOPNOTSUPP)
 228                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
 229                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
 230         }
 231         if (bio->bi_private)
 232                 complete(bio->bi_private);
 233         bio_put(bio);
 234 }
 235
 236 /**
 237  * blkdev_issue_flush - queue a flush
 238  * @bdev:       blockdev to issue flush for
 239  * @gfp_mask:   memory allocation flags (for bio_alloc)
 240  * @error_sector:       error sector
 241  * @flags:      BLKDEV_IFL_* flags to control behaviour
 242  *
 243  * Description:
 244  *    Issue a flush for the block device in question. Caller can supply
 245  *    room for storing the error offset in case of a flush error, if they
 246  *    wish to. If WAIT flag is not passed then caller may check only what
 247  *    request was pushed in some internal queue for later handling.
 248  */
 249 int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask,
 250                 sector_t *error_sector, unsigned long flags)
 251 {
 252         DECLARE_COMPLETION_ONSTACK(wait);
 253         struct request_queue *q;
 254         struct bio *bio;
 255         int ret = 0;
 256
 257         if (bdev->bd_disk == NULL)
 258                 return -ENXIO;
 259
 260         q = bdev_get_queue(bdev);
 261         if (!q)
 262                 return -ENXIO;
 263
 264         /*
 265          * some block devices may not have their queue correctly set up here
 266          * (e.g. loop device without a backing file) and so issuing a flush
 267          * here will panic. Ensure there is a request function before issuing
 268          * the barrier.
 269          */
 270         if (!q->make_request_fn)
 271                 return -ENXIO;
 272
 273         bio = bio_alloc(gfp_mask, 0);
 274         bio->bi_end_io = bio_end_empty_barrier;
 275         bio->bi_bdev = bdev;
 276         if (test_bit(BLKDEV_WAIT, &flags))
 277                 bio->bi_private = &wait;
 278
 279         bio_get(bio);
 280         submit_bio(WRITE_BARRIER, bio);
 281         if (test_bit(BLKDEV_WAIT, &flags)) {
 282                 wait_for_completion(&wait);
 283                 /*
 284                  * The driver must store the error location in ->bi_sector, if
 285                  * it supports it. For non-stacked drivers, this should be
 286                  * copied from blk_rq_pos(rq).
 287                  */
 288                 if (error_sector)
 289                         *error_sector = bio->bi_sector;
 290         }
 291
 292         if (bio_flagged(bio, BIO_EOPNOTSUPP))
 293                 ret = -EOPNOTSUPP;
 294         else if (!bio_flagged(bio, BIO_UPTODATE))
 295                 ret = -EIO;
 296
 297         bio_put(bio);
 298         return ret;
 299 }
 300 EXPORT_SYMBOL(blkdev_issue_flush);