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 void blk_recalc_rq_sectors(struct request *rq, int nsect)
  13 {
  14         if (blk_fs_request(rq)) {
  15                 rq->hard_sector += nsect;
  16                 rq->hard_nr_sectors -= nsect;
  17
  18                 /*
  19                  * Move the I/O submission pointers ahead if required.
  20                  */
  21                 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
  22                     (rq->sector <= rq->hard_sector)) {
  23                         rq->sector = rq->hard_sector;
  24                         rq->nr_sectors = rq->hard_nr_sectors;
  25                         rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
  26                         rq->current_nr_sectors = rq->hard_cur_sectors;
  27                         rq->buffer = bio_data(rq->bio);
  28                 }
  29
  30                 /*
  31                  * if total number of sectors is less than the first segment
  32                  * size, something has gone terribly wrong
  33                  */
  34                 if (rq->nr_sectors < rq->current_nr_sectors) {
  35                         printk(KERN_ERR "blk: request botched\n");
  36                         rq->nr_sectors = rq->current_nr_sectors;
  37                 }
  38         }
  39 }
  40
  41 void blk_recalc_rq_segments(struct request *rq)
  42 {
  43         int nr_phys_segs;
  44         int nr_hw_segs;
  45         unsigned int phys_size;
  46         unsigned int hw_size;
  47         struct bio_vec *bv, *bvprv = NULL;
  48         int seg_size;
  49         int hw_seg_size;
  50         int cluster;
  51         struct req_iterator iter;
  52         int high, highprv = 1;
  53         struct request_queue *q = rq->q;
  54
  55         if (!rq->bio)
  56                 return;
  57
  58         cluster = test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
  59         hw_seg_size = seg_size = 0;
  60         phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
  61         rq_for_each_segment(bv, rq, iter) {
  62                 /*
  63                  * the trick here is making sure that a high page is never
  64                  * considered part of another segment, since that might
  65                  * change with the bounce page.
  66                  */
  67                 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
  68                 if (high || highprv)
  69                         goto new_hw_segment;
  70                 if (cluster) {
  71                         if (seg_size + bv->bv_len > q->max_segment_size)
  72                                 goto new_segment;
  73                         if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
  74                                 goto new_segment;
  75                         if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
  76                                 goto new_segment;
  77                         if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
  78                                 goto new_hw_segment;
  79
  80                         seg_size += bv->bv_len;
  81                         hw_seg_size += bv->bv_len;
  82                         bvprv = bv;
  83                         continue;
  84                 }
  85 new_segment:
  86                 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
  87                     !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
  88                         hw_seg_size += bv->bv_len;
  89                 else {
  90 new_hw_segment:
  91                         if (nr_hw_segs == 1 &&
  92                             hw_seg_size > rq->bio->bi_hw_front_size)
  93                                 rq->bio->bi_hw_front_size = hw_seg_size;
  94                         hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
  95                         nr_hw_segs++;
  96                 }
  97
  98                 if (nr_phys_segs == 1 && seg_size > rq->bio->bi_seg_front_size)
  99                         rq->bio->bi_seg_front_size = seg_size;
 100
 101                 nr_phys_segs++;
 102                 bvprv = bv;
 103                 seg_size = bv->bv_len;
 104                 highprv = high;
 105         }
 106
 107         if (nr_hw_segs == 1 &&
 108             hw_seg_size > rq->bio->bi_hw_front_size)
 109                 rq->bio->bi_hw_front_size = hw_seg_size;
 110         if (hw_seg_size > rq->biotail->bi_hw_back_size)
 111                 rq->biotail->bi_hw_back_size = hw_seg_size;
 112         if (nr_phys_segs == 1 && seg_size > rq->bio->bi_seg_front_size)
 113                 rq->bio->bi_seg_front_size = seg_size;
 114         if (seg_size > rq->biotail->bi_seg_back_size)
 115                 rq->biotail->bi_seg_back_size = seg_size;
 116         rq->nr_phys_segments = nr_phys_segs;
 117         rq->nr_hw_segments = nr_hw_segs;
 118 }
 119
 120 void blk_recount_segments(struct request_queue *q, struct bio *bio)
 121 {
 122         struct request rq;
 123         struct bio *nxt = bio->bi_next;
 124         rq.q = q;
 125         rq.bio = rq.biotail = bio;
 126         bio->bi_next = NULL;
 127         blk_recalc_rq_segments(&rq);
 128         bio->bi_next = nxt;
 129         bio->bi_phys_segments = rq.nr_phys_segments;
 130         bio->bi_hw_segments = rq.nr_hw_segments;
 131         bio->bi_flags |= (1 << BIO_SEG_VALID);
 132 }
 133 EXPORT_SYMBOL(blk_recount_segments);
 134
 135 static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
 136                                    struct bio *nxt)
 137 {
 138         if (!test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags))
 139                 return 0;
 140
 141         if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
 142                 return 0;
 143         if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
 144             q->max_segment_size)
 145                 return 0;
 146
 147         /*
 148          * bio and nxt are contigous in memory, check if the queue allows
 149          * these two to be merged into one
 150          */
 151         if (BIO_SEG_BOUNDARY(q, bio, nxt))
 152                 return 1;
 153
 154         return 0;
 155 }
 156
 157 static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
 158                                  struct bio *nxt)
 159 {
 160         if (!bio_flagged(bio, BIO_SEG_VALID))
 161                 blk_recount_segments(q, bio);
 162         if (!bio_flagged(nxt, BIO_SEG_VALID))
 163                 blk_recount_segments(q, nxt);
 164         if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
 165             BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
 166                 return 0;
 167         if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
 168                 return 0;
 169
 170         return 1;
 171 }
 172
 173 /*
 174  * map a request to scatterlist, return number of sg entries setup. Caller
 175  * must make sure sg can hold rq->nr_phys_segments entries
 176  */
 177 int blk_rq_map_sg(struct request_queue *q, struct request *rq,
 178                   struct scatterlist *sglist)
 179 {
 180         struct bio_vec *bvec, *bvprv;
 181         struct req_iterator iter;
 182         struct scatterlist *sg;
 183         int nsegs, cluster;
 184
 185         nsegs = 0;
 186         cluster = test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
 187
 188         /*
 189          * for each bio in rq
 190          */
 191         bvprv = NULL;
 192         sg = NULL;
 193         rq_for_each_segment(bvec, rq, iter) {
 194                 int nbytes = bvec->bv_len;
 195
 196                 if (bvprv && cluster) {
 197                         if (sg->length + nbytes > q->max_segment_size)
 198                                 goto new_segment;
 199
 200                         if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
 201                                 goto new_segment;
 202                         if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
 203                                 goto new_segment;
 204
 205                         sg->length += nbytes;
 206                 } else {
 207 new_segment:
 208                         if (!sg)
 209                                 sg = sglist;
 210                         else {
 211                                 /*
 212                                  * If the driver previously mapped a shorter
 213                                  * list, we could see a termination bit
 214                                  * prematurely unless it fully inits the sg
 215                                  * table on each mapping. We KNOW that there
 216                                  * must be more entries here or the driver
 217                                  * would be buggy, so force clear the
 218                                  * termination bit to avoid doing a full
 219                                  * sg_init_table() in drivers for each command.
 220                                  */
 221                                 sg->page_link &= ~0x02;
 222                                 sg = sg_next(sg);
 223                         }
 224
 225                         sg_set_page(sg, bvec->bv_page, nbytes, bvec->bv_offset);
 226                         nsegs++;
 227                 }
 228                 bvprv = bvec;
 229         } /* segments in rq */
 230
 231
 232         if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
 233             (rq->data_len & q->dma_pad_mask)) {
 234                 unsigned int pad_len = (q->dma_pad_mask & ~rq->data_len) + 1;
 235
 236                 sg->length += pad_len;
 237                 rq->extra_len += pad_len;
 238         }
 239
 240         if (q->dma_drain_size && q->dma_drain_needed(rq)) {
 241                 if (rq->cmd_flags & REQ_RW)
 242                         memset(q->dma_drain_buffer, 0, q->dma_drain_size);
 243
 244                 sg->page_link &= ~0x02;
 245                 sg = sg_next(sg);
 246                 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
 247                             q->dma_drain_size,
 248                             ((unsigned long)q->dma_drain_buffer) &
 249                             (PAGE_SIZE - 1));
 250                 nsegs++;
 251                 rq->extra_len += q->dma_drain_size;
 252         }
 253
 254         if (sg)
 255                 sg_mark_end(sg);
 256
 257         return nsegs;
 258 }
 259 EXPORT_SYMBOL(blk_rq_map_sg);
 260
 261 static inline int ll_new_mergeable(struct request_queue *q,
 262                                    struct request *req,
 263                                    struct bio *bio)
 264 {
 265         int nr_phys_segs = bio_phys_segments(q, bio);
 266
 267         if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
 268                 req->cmd_flags |= REQ_NOMERGE;
 269                 if (req == q->last_merge)
 270                         q->last_merge = NULL;
 271                 return 0;
 272         }
 273
 274         /*
 275          * A hw segment is just getting larger, bump just the phys
 276          * counter.
 277          */
 278         req->nr_phys_segments += nr_phys_segs;
 279         return 1;
 280 }
 281
 282 static inline int ll_new_hw_segment(struct request_queue *q,
 283                                     struct request *req,
 284                                     struct bio *bio)
 285 {
 286         int nr_hw_segs = bio_hw_segments(q, bio);
 287         int nr_phys_segs = bio_phys_segments(q, bio);
 288
 289         if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
 290             || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
 291                 req->cmd_flags |= REQ_NOMERGE;
 292                 if (req == q->last_merge)
 293                         q->last_merge = NULL;
 294                 return 0;
 295         }
 296
 297         /*
 298          * This will form the start of a new hw segment.  Bump both
 299          * counters.
 300          */
 301         req->nr_hw_segments += nr_hw_segs;
 302         req->nr_phys_segments += nr_phys_segs;
 303         return 1;
 304 }
 305
 306 int ll_back_merge_fn(struct request_queue *q, struct request *req,
 307                      struct bio *bio)
 308 {
 309         unsigned short max_sectors;
 310         int len;
 311
 312         if (unlikely(blk_pc_request(req)))
 313                 max_sectors = q->max_hw_sectors;
 314         else
 315                 max_sectors = q->max_sectors;
 316
 317         if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
 318                 req->cmd_flags |= REQ_NOMERGE;
 319                 if (req == q->last_merge)
 320                         q->last_merge = NULL;
 321                 return 0;
 322         }
 323         if (!bio_flagged(req->biotail, BIO_SEG_VALID))
 324                 blk_recount_segments(q, req->biotail);
 325         if (!bio_flagged(bio, BIO_SEG_VALID))
 326                 blk_recount_segments(q, bio);
 327         len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
 328         if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio))
 329             && !BIOVEC_VIRT_OVERSIZE(len)) {
 330                 int mergeable =  ll_new_mergeable(q, req, bio);
 331
 332                 if (mergeable) {
 333                         if (req->nr_hw_segments == 1)
 334                                 req->bio->bi_hw_front_size = len;
 335                         if (bio->bi_hw_segments == 1)
 336                                 bio->bi_hw_back_size = len;
 337                 }
 338                 return mergeable;
 339         }
 340
 341         return ll_new_hw_segment(q, req, bio);
 342 }
 343
 344 int ll_front_merge_fn(struct request_queue *q, struct request *req,
 345                       struct bio *bio)
 346 {
 347         unsigned short max_sectors;
 348         int len;
 349
 350         if (unlikely(blk_pc_request(req)))
 351                 max_sectors = q->max_hw_sectors;
 352         else
 353                 max_sectors = q->max_sectors;
 354
 355
 356         if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
 357                 req->cmd_flags |= REQ_NOMERGE;
 358                 if (req == q->last_merge)
 359                         q->last_merge = NULL;
 360                 return 0;
 361         }
 362         len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
 363         if (!bio_flagged(bio, BIO_SEG_VALID))
 364                 blk_recount_segments(q, bio);
 365         if (!bio_flagged(req->bio, BIO_SEG_VALID))
 366                 blk_recount_segments(q, req->bio);
 367         if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
 368             !BIOVEC_VIRT_OVERSIZE(len)) {
 369                 int mergeable =  ll_new_mergeable(q, req, bio);
 370
 371                 if (mergeable) {
 372                         if (bio->bi_hw_segments == 1)
 373                                 bio->bi_hw_front_size = len;
 374                         if (req->nr_hw_segments == 1)
 375                                 req->biotail->bi_hw_back_size = len;
 376                 }
 377                 return mergeable;
 378         }
 379
 380         return ll_new_hw_segment(q, req, bio);
 381 }
 382
 383 static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
 384                                 struct request *next)
 385 {
 386         int total_phys_segments;
 387         int total_hw_segments;
 388         unsigned int seg_size =
 389                 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
 390
 391         /*
 392          * First check if the either of the requests are re-queued
 393          * requests.  Can't merge them if they are.
 394          */
 395         if (req->special || next->special)
 396                 return 0;
 397
 398         /*
 399          * Will it become too large?
 400          */
 401         if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
 402                 return 0;
 403
 404         total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
 405         if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
 406                 if (req->nr_phys_segments == 1)
 407                         req->bio->bi_seg_front_size = seg_size;
 408                 if (next->nr_phys_segments == 1)
 409                         next->biotail->bi_seg_back_size = seg_size;
 410                 total_phys_segments--;
 411         }
 412
 413         if (total_phys_segments > q->max_phys_segments)
 414                 return 0;
 415
 416         total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
 417         if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
 418                 int len = req->biotail->bi_hw_back_size +
 419                                 next->bio->bi_hw_front_size;
 420                 /*
 421                  * propagate the combined length to the end of the requests
 422                  */
 423                 if (req->nr_hw_segments == 1)
 424                         req->bio->bi_hw_front_size = len;
 425                 if (next->nr_hw_segments == 1)
 426                         next->biotail->bi_hw_back_size = len;
 427                 total_hw_segments--;
 428         }
 429
 430         if (total_hw_segments > q->max_hw_segments)
 431                 return 0;
 432
 433         /* Merge is OK... */
 434         req->nr_phys_segments = total_phys_segments;
 435         req->nr_hw_segments = total_hw_segments;
 436         return 1;
 437 }
 438
 439 /*
 440  * Has to be called with the request spinlock acquired
 441  */
 442 static int attempt_merge(struct request_queue *q, struct request *req,
 443                           struct request *next)
 444 {
 445         if (!rq_mergeable(req) || !rq_mergeable(next))
 446                 return 0;
 447
 448         /*
 449          * not contiguous
 450          */
 451         if (req->sector + req->nr_sectors != next->sector)
 452                 return 0;
 453
 454         if (rq_data_dir(req) != rq_data_dir(next)
 455             || req->rq_disk != next->rq_disk
 456             || next->special)
 457                 return 0;
 458
 459         if (blk_integrity_rq(req) != blk_integrity_rq(next))
 460                 return 0;
 461
 462         /*
 463          * If we are allowed to merge, then append bio list
 464          * from next to rq and release next. merge_requests_fn
 465          * will have updated segment counts, update sector
 466          * counts here.
 467          */
 468         if (!ll_merge_requests_fn(q, req, next))
 469                 return 0;
 470
 471         /*
 472          * At this point we have either done a back merge
 473          * or front merge. We need the smaller start_time of
 474          * the merged requests to be the current request
 475          * for accounting purposes.
 476          */
 477         if (time_after(req->start_time, next->start_time))
 478                 req->start_time = next->start_time;
 479
 480         req->biotail->bi_next = next->bio;
 481         req->biotail = next->biotail;
 482
 483         req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
 484
 485         elv_merge_requests(q, req, next);
 486
 487         if (req->rq_disk) {
 488                 struct hd_struct *part
 489                         = get_part(req->rq_disk, req->sector);
 490                 disk_round_stats(req->rq_disk);
 491                 req->rq_disk->in_flight--;
 492                 if (part) {
 493                         part_round_stats(part);
 494                         part->in_flight--;
 495                 }
 496         }
 497
 498         req->ioprio = ioprio_best(req->ioprio, next->ioprio);
 499
 500         __blk_put_request(q, next);
 501         return 1;
 502 }
 503
 504 int attempt_back_merge(struct request_queue *q, struct request *rq)
 505 {
 506         struct request *next = elv_latter_request(q, rq);
 507
 508         if (next)
 509                 return attempt_merge(q, rq, next);
 510
 511         return 0;
 512 }
 513
 514 int attempt_front_merge(struct request_queue *q, struct request *rq)
 515 {
 516         struct request *prev = elv_former_request(q, rq);
 517
 518         if (prev)
 519                 return attempt_merge(q, prev, rq);
 520
 521         return 0;
 522 }