block/blk-merge.c

   1 /*
   2  * Functions related to segment and merge 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/scatterlist.h>
   9
  10 #include "blk.h"
  11
  12 static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
  13                                              struct bio *bio)
  14 {
  15         struct bio_vec *bv, *bvprv = NULL;
  16         int cluster, i, high, highprv = 1;
  17         unsigned int seg_size, nr_phys_segs;
  18         struct bio *fbio, *bbio;
  19
  20         if (!bio)
  21                 return 0;
  22
  23         fbio = bio;
  24         cluster = blk_queue_cluster(q);
  25         seg_size = 0;
  26         nr_phys_segs = 0;
  27         for_each_bio(bio) {
  28                 bio_for_each_segment(bv, bio, i) {
  29                         /*
  30                          * the trick here is making sure that a high page is
  31                          * never considered part of another segment, since that
  32                          * might change with the bounce page.
  33                          */
  34                         high = page_to_pfn(bv->bv_page) > queue_bounce_pfn(q);
  35                         if (high || highprv)
  36                                 goto new_segment;
  37                         if (cluster) {
  38                                 if (seg_size + bv->bv_len
  39                                     > queue_max_segment_size(q))
  40                                         goto new_segment;
  41                                 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
  42                                         goto new_segment;
  43                                 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
  44                                         goto new_segment;
  45
  46                                 seg_size += bv->bv_len;
  47                                 bvprv = bv;
  48                                 continue;
  49                         }
  50 new_segment:
  51                         if (nr_phys_segs == 1 && seg_size >
  52                             fbio->bi_seg_front_size)
  53                                 fbio->bi_seg_front_size = seg_size;
  54
  55                         nr_phys_segs++;
  56                         bvprv = bv;
  57                         seg_size = bv->bv_len;
  58                         highprv = high;
  59                 }
  60                 bbio = bio;
  61         }
  62
  63         if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
  64                 fbio->bi_seg_front_size = seg_size;
  65         if (seg_size > bbio->bi_seg_back_size)
  66                 bbio->bi_seg_back_size = seg_size;
  67
  68         return nr_phys_segs;
  69 }
  70
  71 void blk_recalc_rq_segments(struct request *rq)
  72 {
  73         rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio);
  74 }
  75
  76 void blk_recount_segments(struct request_queue *q, struct bio *bio)
  77 {
  78         struct bio *nxt = bio->bi_next;
  79
  80         bio->bi_next = NULL;
  81         bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio);
  82         bio->bi_next = nxt;
  83         bio->bi_flags |= (1 << BIO_SEG_VALID);
  84 }
  85 EXPORT_SYMBOL(blk_recount_segments);
  86
  87 static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
  88                                    struct bio *nxt)
  89 {
  90         if (!blk_queue_cluster(q))
  91                 return 0;
  92
  93         if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
  94             queue_max_segment_size(q))
  95                 return 0;
  96
  97         if (!bio_has_data(bio))
  98                 return 1;
  99
 100         if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
 101                 return 0;
 102
 103         /*
 104          * bio and nxt are contiguous in memory; check if the queue allows
 105          * these two to be merged into one
 106          */
 107         if (BIO_SEG_BOUNDARY(q, bio, nxt))
 108                 return 1;
 109
 110         return 0;
 111 }
 112
 113 static void
 114 __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
 115                      struct scatterlist *sglist, struct bio_vec **bvprv,
 116                      struct scatterlist **sg, int *nsegs, int *cluster)
 117 {
 118
 119         int nbytes = bvec->bv_len;
 120
 121         if (*bvprv && *cluster) {
 122                 if ((*sg)->length + nbytes > queue_max_segment_size(q))
 123                         goto new_segment;
 124
 125                 if (!BIOVEC_PHYS_MERGEABLE(*bvprv, bvec))
 126                         goto new_segment;
 127                 if (!BIOVEC_SEG_BOUNDARY(q, *bvprv, bvec))
 128                         goto new_segment;
 129
 130                 (*sg)->length += nbytes;
 131         } else {
 132 new_segment:
 133                 if (!*sg)
 134                         *sg = sglist;
 135                 else {
 136                         /*
 137                          * If the driver previously mapped a shorter
 138                          * list, we could see a termination bit
 139                          * prematurely unless it fully inits the sg
 140                          * table on each mapping. We KNOW that there
 141                          * must be more entries here or the driver
 142                          * would be buggy, so force clear the
 143                          * termination bit to avoid doing a full
 144                          * sg_init_table() in drivers for each command.
 145                          */
 146                         sg_unmark_end(*sg);
 147                         *sg = sg_next(*sg);
 148                 }
 149
 150                 sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
 151                 (*nsegs)++;
 152         }
 153         *bvprv = bvec;
 154 }
 155
 156 /*
 157  * map a request to scatterlist, return number of sg entries setup. Caller
 158  * must make sure sg can hold rq->nr_phys_segments entries
 159  */
 160 int blk_rq_map_sg(struct request_queue *q, struct request *rq,
 161                   struct scatterlist *sglist)
 162 {
 163         struct bio_vec *bvec, *bvprv;
 164         struct req_iterator iter;
 165         struct scatterlist *sg;
 166         int nsegs, cluster;
 167
 168         nsegs = 0;
 169         cluster = blk_queue_cluster(q);
 170
 171         /*
 172          * for each bio in rq
 173          */
 174         bvprv = NULL;
 175         sg = NULL;
 176         rq_for_each_segment(bvec, rq, iter) {
 177                 __blk_segment_map_sg(q, bvec, sglist, &bvprv, &sg,
 178                                      &nsegs, &cluster);
 179         } /* segments in rq */
 180
 181
 182         if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
 183             (blk_rq_bytes(rq) & q->dma_pad_mask)) {
 184                 unsigned int pad_len =
 185                         (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
 186
 187                 sg->length += pad_len;
 188                 rq->extra_len += pad_len;
 189         }
 190
 191         if (q->dma_drain_size && q->dma_drain_needed(rq)) {
 192                 if (rq->cmd_flags & REQ_WRITE)
 193                         memset(q->dma_drain_buffer, 0, q->dma_drain_size);
 194
 195                 sg->page_link &= ~0x02;
 196                 sg = sg_next(sg);
 197                 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
 198                             q->dma_drain_size,
 199                             ((unsigned long)q->dma_drain_buffer) &
 200                             (PAGE_SIZE - 1));
 201                 nsegs++;
 202                 rq->extra_len += q->dma_drain_size;
 203         }
 204
 205         if (sg)
 206                 sg_mark_end(sg);
 207
 208         return nsegs;
 209 }
 210 EXPORT_SYMBOL(blk_rq_map_sg);
 211
 212 /**
 213  * blk_bio_map_sg - map a bio to a scatterlist
 214  * @q: request_queue in question
 215  * @bio: bio being mapped
 216  * @sglist: scatterlist being mapped
 217  *
 218  * Note:
 219  *    Caller must make sure sg can hold bio->bi_phys_segments entries
 220  *
 221  * Will return the number of sg entries setup
 222  */
 223 int blk_bio_map_sg(struct request_queue *q, struct bio *bio,
 224                    struct scatterlist *sglist)
 225 {
 226         struct bio_vec *bvec, *bvprv;
 227         struct scatterlist *sg;
 228         int nsegs, cluster;
 229         unsigned long i;
 230
 231         nsegs = 0;
 232         cluster = blk_queue_cluster(q);
 233
 234         bvprv = NULL;
 235         sg = NULL;
 236         bio_for_each_segment(bvec, bio, i) {
 237                 __blk_segment_map_sg(q, bvec, sglist, &bvprv, &sg,
 238                                      &nsegs, &cluster);
 239         } /* segments in bio */
 240
 241         if (sg)
 242                 sg_mark_end(sg);
 243
 244         BUG_ON(bio->bi_phys_segments && nsegs > bio->bi_phys_segments);
 245         return nsegs;
 246 }
 247 EXPORT_SYMBOL(blk_bio_map_sg);
 248
 249 static inline int ll_new_hw_segment(struct request_queue *q,
 250                                     struct request *req,
 251                                     struct bio *bio)
 252 {
 253         int nr_phys_segs = bio_phys_segments(q, bio);
 254
 255         if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
 256                 goto no_merge;
 257
 258         if (bio_integrity(bio) && blk_integrity_merge_bio(q, req, bio))
 259                 goto no_merge;
 260
 261         /*
 262          * This will form the start of a new hw segment.  Bump both
 263          * counters.
 264          */
 265         req->nr_phys_segments += nr_phys_segs;
 266         return 1;
 267
 268 no_merge:
 269         req->cmd_flags |= REQ_NOMERGE;
 270         if (req == q->last_merge)
 271                 q->last_merge = NULL;
 272         return 0;
 273 }
 274
 275 int ll_back_merge_fn(struct request_queue *q, struct request *req,
 276                      struct bio *bio)
 277 {
 278         if (blk_rq_sectors(req) + bio_sectors(bio) >
 279             blk_rq_get_max_sectors(req)) {
 280                 req->cmd_flags |= REQ_NOMERGE;
 281                 if (req == q->last_merge)
 282                         q->last_merge = NULL;
 283                 return 0;
 284         }
 285         if (!bio_flagged(req->biotail, BIO_SEG_VALID))
 286                 blk_recount_segments(q, req->biotail);
 287         if (!bio_flagged(bio, BIO_SEG_VALID))
 288                 blk_recount_segments(q, bio);
 289
 290         return ll_new_hw_segment(q, req, bio);
 291 }
 292
 293 int ll_front_merge_fn(struct request_queue *q, struct request *req,
 294                       struct bio *bio)
 295 {
 296         if (blk_rq_sectors(req) + bio_sectors(bio) >
 297             blk_rq_get_max_sectors(req)) {
 298                 req->cmd_flags |= REQ_NOMERGE;
 299                 if (req == q->last_merge)
 300                         q->last_merge = NULL;
 301                 return 0;
 302         }
 303         if (!bio_flagged(bio, BIO_SEG_VALID))
 304                 blk_recount_segments(q, bio);
 305         if (!bio_flagged(req->bio, BIO_SEG_VALID))
 306                 blk_recount_segments(q, req->bio);
 307
 308         return ll_new_hw_segment(q, req, bio);
 309 }
 310
 311 /*
 312  * blk-mq uses req->special to carry normal driver per-request payload, it
 313  * does not indicate a prepared command that we cannot merge with.
 314  */
 315 static bool req_no_special_merge(struct request *req)
 316 {
 317         struct request_queue *q = req->q;
 318
 319         return !q->mq_ops && req->special;
 320 }
 321
 322 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
 323                                 struct request *next)
 324 {
 325         int total_phys_segments;
 326         unsigned int seg_size =
 327                 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
 328
 329         /*
 330          * First check if the either of the requests are re-queued
 331          * requests.  Can't merge them if they are.
 332          */
 333         if (req_no_special_merge(req) || req_no_special_merge(next))
 334                 return 0;
 335
 336         /*
 337          * Will it become too large?
 338          */
 339         if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
 340             blk_rq_get_max_sectors(req))
 341                 return 0;
 342
 343         total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
 344         if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
 345                 if (req->nr_phys_segments == 1)
 346                         req->bio->bi_seg_front_size = seg_size;
 347                 if (next->nr_phys_segments == 1)
 348                         next->biotail->bi_seg_back_size = seg_size;
 349                 total_phys_segments--;
 350         }
 351
 352         if (total_phys_segments > queue_max_segments(q))
 353                 return 0;
 354
 355         if (blk_integrity_rq(req) && blk_integrity_merge_rq(q, req, next))
 356                 return 0;
 357
 358         /* Merge is OK... */
 359         req->nr_phys_segments = total_phys_segments;
 360         return 1;
 361 }
 362
 363 /**
 364  * blk_rq_set_mixed_merge - mark a request as mixed merge
 365  * @rq: request to mark as mixed merge
 366  *
 367  * Description:
 368  *     @rq is about to be mixed merged.  Make sure the attributes
 369  *     which can be mixed are set in each bio and mark @rq as mixed
 370  *     merged.
 371  */
 372 void blk_rq_set_mixed_merge(struct request *rq)
 373 {
 374         unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
 375         struct bio *bio;
 376
 377         if (rq->cmd_flags & REQ_MIXED_MERGE)
 378                 return;
 379
 380         /*
 381          * @rq will no longer represent mixable attributes for all the
 382          * contained bios.  It will just track those of the first one.
 383          * Distributes the attributs to each bio.
 384          */
 385         for (bio = rq->bio; bio; bio = bio->bi_next) {
 386                 WARN_ON_ONCE((bio->bi_rw & REQ_FAILFAST_MASK) &&
 387                              (bio->bi_rw & REQ_FAILFAST_MASK) != ff);
 388                 bio->bi_rw |= ff;
 389         }
 390         rq->cmd_flags |= REQ_MIXED_MERGE;
 391 }
 392
 393 static void blk_account_io_merge(struct request *req)
 394 {
 395         if (blk_do_io_stat(req)) {
 396                 struct hd_struct *part;
 397                 int cpu;
 398
 399                 cpu = part_stat_lock();
 400                 part = req->part;
 401
 402                 part_round_stats(cpu, part);
 403                 part_dec_in_flight(part, rq_data_dir(req));
 404
 405                 hd_struct_put(part);
 406                 part_stat_unlock();
 407         }
 408 }
 409
 410 /*
 411  * Has to be called with the request spinlock acquired
 412  */
 413 static int attempt_merge(struct request_queue *q, struct request *req,
 414                           struct request *next)
 415 {
 416         if (!rq_mergeable(req) || !rq_mergeable(next))
 417                 return 0;
 418
 419         if (!blk_check_merge_flags(req->cmd_flags, next->cmd_flags))
 420                 return 0;
 421
 422         /*
 423          * not contiguous
 424          */
 425         if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
 426                 return 0;
 427
 428         if (rq_data_dir(req) != rq_data_dir(next)
 429             || req->rq_disk != next->rq_disk
 430             || req_no_special_merge(next))
 431                 return 0;
 432
 433         if (req->cmd_flags & REQ_WRITE_SAME &&
 434             !blk_write_same_mergeable(req->bio, next->bio))
 435                 return 0;
 436
 437         /*
 438          * If we are allowed to merge, then append bio list
 439          * from next to rq and release next. merge_requests_fn
 440          * will have updated segment counts, update sector
 441          * counts here.
 442          */
 443         if (!ll_merge_requests_fn(q, req, next))
 444                 return 0;
 445
 446         /*
 447          * If failfast settings disagree or any of the two is already
 448          * a mixed merge, mark both as mixed before proceeding.  This
 449          * makes sure that all involved bios have mixable attributes
 450          * set properly.
 451          */
 452         if ((req->cmd_flags | next->cmd_flags) & REQ_MIXED_MERGE ||
 453             (req->cmd_flags & REQ_FAILFAST_MASK) !=
 454             (next->cmd_flags & REQ_FAILFAST_MASK)) {
 455                 blk_rq_set_mixed_merge(req);
 456                 blk_rq_set_mixed_merge(next);
 457         }
 458
 459         /*
 460          * At this point we have either done a back merge
 461          * or front merge. We need the smaller start_time of
 462          * the merged requests to be the current request
 463          * for accounting purposes.
 464          */
 465         if (time_after(req->start_time, next->start_time))
 466                 req->start_time = next->start_time;
 467
 468         req->biotail->bi_next = next->bio;
 469         req->biotail = next->biotail;
 470
 471         req->__data_len += blk_rq_bytes(next);
 472
 473         elv_merge_requests(q, req, next);
 474
 475         /*
 476          * 'next' is going away, so update stats accordingly
 477          */
 478         blk_account_io_merge(next);
 479
 480         req->ioprio = ioprio_best(req->ioprio, next->ioprio);
 481         if (blk_rq_cpu_valid(next))
 482                 req->cpu = next->cpu;
 483
 484         /* owner-ship of bio passed from next to req */
 485         next->bio = NULL;
 486         __blk_put_request(q, next);
 487         return 1;
 488 }
 489
 490 int attempt_back_merge(struct request_queue *q, struct request *rq)
 491 {
 492         struct request *next = elv_latter_request(q, rq);
 493
 494         if (next)
 495                 return attempt_merge(q, rq, next);
 496
 497         return 0;
 498 }
 499
 500 int attempt_front_merge(struct request_queue *q, struct request *rq)
 501 {
 502         struct request *prev = elv_former_request(q, rq);
 503
 504         if (prev)
 505                 return attempt_merge(q, prev, rq);
 506
 507         return 0;
 508 }
 509
 510 int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
 511                           struct request *next)
 512 {
 513         return attempt_merge(q, rq, next);
 514 }
 515
 516 bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
 517 {
 518         if (!rq_mergeable(rq) || !bio_mergeable(bio))
 519                 return false;
 520
 521         if (!blk_check_merge_flags(rq->cmd_flags, bio->bi_rw))
 522                 return false;
 523
 524         /* different data direction or already started, don't merge */
 525         if (bio_data_dir(bio) != rq_data_dir(rq))
 526                 return false;
 527
 528         /* must be same device and not a special request */
 529         if (rq->rq_disk != bio->bi_bdev->bd_disk || req_no_special_merge(rq))
 530                 return false;
 531
 532         /* only merge integrity protected bio into ditto rq */
 533         if (bio_integrity(bio) != blk_integrity_rq(rq))
 534                 return false;
 535
 536         /* must be using the same buffer */
 537         if (rq->cmd_flags & REQ_WRITE_SAME &&
 538             !blk_write_same_mergeable(rq->bio, bio))
 539                 return false;
 540
 541         return true;
 542 }
 543
 544 int blk_try_merge(struct request *rq, struct bio *bio)
 545 {
 546         if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_sector)
 547                 return ELEVATOR_BACK_MERGE;
 548         else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_sector)
 549                 return ELEVATOR_FRONT_MERGE;
 550         return ELEVATOR_NO_MERGE;
 551 }