nfs: new subdir Documentation/filesystems/nfs
[safe/jmp/linux-2.6] / block / blk-merge.c
index bef1b4d..99cb5cf 100644 (file)
 
 #include "blk.h"
 
-void blk_recalc_rq_sectors(struct request *rq, int nsect)
+static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
+                                            struct bio *bio)
 {
-       if (blk_fs_request(rq)) {
-               rq->hard_sector += nsect;
-               rq->hard_nr_sectors -= nsect;
-
-               /*
-                * Move the I/O submission pointers ahead if required.
-                */
-               if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
-                   (rq->sector <= rq->hard_sector)) {
-                       rq->sector = rq->hard_sector;
-                       rq->nr_sectors = rq->hard_nr_sectors;
-                       rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
-                       rq->current_nr_sectors = rq->hard_cur_sectors;
-                       rq->buffer = bio_data(rq->bio);
-               }
-
-               /*
-                * if total number of sectors is less than the first segment
-                * size, something has gone terribly wrong
-                */
-               if (rq->nr_sectors < rq->current_nr_sectors) {
-                       printk(KERN_ERR "blk: request botched\n");
-                       rq->nr_sectors = rq->current_nr_sectors;
-               }
-       }
-}
-
-void blk_recalc_rq_segments(struct request *rq)
-{
-       int nr_phys_segs;
-       int nr_hw_segs;
        unsigned int phys_size;
-       unsigned int hw_size;
        struct bio_vec *bv, *bvprv = NULL;
-       int seg_size;
-       int hw_seg_size;
-       int cluster;
-       struct req_iterator iter;
-       int high, highprv = 1;
-       struct request_queue *q = rq->q;
+       int cluster, i, high, highprv = 1;
+       unsigned int seg_size, nr_phys_segs;
+       struct bio *fbio, *bbio;
 
-       if (!rq->bio)
-               return;
+       if (!bio)
+               return 0;
 
-       cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
-       hw_seg_size = seg_size = 0;
-       phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
-       rq_for_each_segment(bv, rq, iter) {
-               /*
-                * the trick here is making sure that a high page is never
-                * considered part of another segment, since that might
-                * change with the bounce page.
-                */
-               high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
-               if (high || highprv)
-                       goto new_hw_segment;
-               if (cluster) {
-                       if (seg_size + bv->bv_len > q->max_segment_size)
-                               goto new_segment;
-                       if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
+       fbio = bio;
+       cluster = test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
+       seg_size = 0;
+       phys_size = nr_phys_segs = 0;
+       for_each_bio(bio) {
+               bio_for_each_segment(bv, bio, i) {
+                       /*
+                        * the trick here is making sure that a high page is
+                        * never considered part of another segment, since that
+                        * might change with the bounce page.
+                        */
+                       high = page_to_pfn(bv->bv_page) > queue_bounce_pfn(q);
+                       if (high || highprv)
                                goto new_segment;
-                       if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
-                               goto new_segment;
-                       if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
-                               goto new_hw_segment;
+                       if (cluster) {
+                               if (seg_size + bv->bv_len
+                                   > queue_max_segment_size(q))
+                                       goto new_segment;
+                               if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
+                                       goto new_segment;
+                               if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
+                                       goto new_segment;
+
+                               seg_size += bv->bv_len;
+                               bvprv = bv;
+                               continue;
+                       }
+new_segment:
+                       if (nr_phys_segs == 1 && seg_size >
+                           fbio->bi_seg_front_size)
+                               fbio->bi_seg_front_size = seg_size;
 
-                       seg_size += bv->bv_len;
-                       hw_seg_size += bv->bv_len;
+                       nr_phys_segs++;
                        bvprv = bv;
-                       continue;
+                       seg_size = bv->bv_len;
+                       highprv = high;
                }
-new_segment:
-               if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
-                   !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
-                       hw_seg_size += bv->bv_len;
-               else {
-new_hw_segment:
-                       if (nr_hw_segs == 1 &&
-                           hw_seg_size > rq->bio->bi_hw_front_size)
-                               rq->bio->bi_hw_front_size = hw_seg_size;
-                       hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
-                       nr_hw_segs++;
-               }
-
-               nr_phys_segs++;
-               bvprv = bv;
-               seg_size = bv->bv_len;
-               highprv = high;
+               bbio = bio;
        }
 
-       if (nr_hw_segs == 1 &&
-           hw_seg_size > rq->bio->bi_hw_front_size)
-               rq->bio->bi_hw_front_size = hw_seg_size;
-       if (hw_seg_size > rq->biotail->bi_hw_back_size)
-               rq->biotail->bi_hw_back_size = hw_seg_size;
-       rq->nr_phys_segments = nr_phys_segs;
-       rq->nr_hw_segments = nr_hw_segs;
+       if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
+               fbio->bi_seg_front_size = seg_size;
+       if (seg_size > bbio->bi_seg_back_size)
+               bbio->bi_seg_back_size = seg_size;
+
+       return nr_phys_segs;
+}
+
+void blk_recalc_rq_segments(struct request *rq)
+{
+       rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio);
 }
 
 void blk_recount_segments(struct request_queue *q, struct bio *bio)
 {
-       struct request rq;
        struct bio *nxt = bio->bi_next;
-       rq.q = q;
-       rq.bio = rq.biotail = bio;
+
        bio->bi_next = NULL;
-       blk_recalc_rq_segments(&rq);
+       bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio);
        bio->bi_next = nxt;
-       bio->bi_phys_segments = rq.nr_phys_segments;
-       bio->bi_hw_segments = rq.nr_hw_segments;
        bio->bi_flags |= (1 << BIO_SEG_VALID);
 }
 EXPORT_SYMBOL(blk_recount_segments);
@@ -128,16 +88,21 @@ EXPORT_SYMBOL(blk_recount_segments);
 static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
                                   struct bio *nxt)
 {
-       if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
+       if (!test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags))
                return 0;
 
-       if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
+       if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
+           queue_max_segment_size(q))
                return 0;
-       if (bio->bi_size + nxt->bi_size > q->max_segment_size)
+
+       if (!bio_has_data(bio))
+               return 1;
+
+       if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
                return 0;
 
        /*
-        * bio and nxt are contigous in memory, check if the queue allows
+        * bio and nxt are contiguous in memory; check if the queue allows
         * these two to be merged into one
         */
        if (BIO_SEG_BOUNDARY(q, bio, nxt))
@@ -146,22 +111,6 @@ static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
        return 0;
 }
 
-static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
-                                struct bio *nxt)
-{
-       if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
-               blk_recount_segments(q, bio);
-       if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
-               blk_recount_segments(q, nxt);
-       if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
-           BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
-               return 0;
-       if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
-               return 0;
-
-       return 1;
-}
-
 /*
  * map a request to scatterlist, return number of sg entries setup. Caller
  * must make sure sg can hold rq->nr_phys_segments entries
@@ -175,7 +124,7 @@ int blk_rq_map_sg(struct request_queue *q, struct request *rq,
        int nsegs, cluster;
 
        nsegs = 0;
-       cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
+       cluster = test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
 
        /*
         * for each bio in rq
@@ -186,7 +135,7 @@ int blk_rq_map_sg(struct request_queue *q, struct request *rq,
                int nbytes = bvec->bv_len;
 
                if (bvprv && cluster) {
-                       if (sg->length + nbytes > q->max_segment_size)
+                       if (sg->length + nbytes > queue_max_segment_size(q))
                                goto new_segment;
 
                        if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
@@ -220,7 +169,20 @@ new_segment:
                bvprv = bvec;
        } /* segments in rq */
 
+
+       if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
+           (blk_rq_bytes(rq) & q->dma_pad_mask)) {
+               unsigned int pad_len =
+                       (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
+
+               sg->length += pad_len;
+               rq->extra_len += pad_len;
+       }
+
        if (q->dma_drain_size && q->dma_drain_needed(rq)) {
+               if (rq->cmd_flags & REQ_RW)
+                       memset(q->dma_drain_buffer, 0, q->dma_drain_size);
+
                sg->page_link &= ~0x02;
                sg = sg_next(sg);
                sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
@@ -228,7 +190,7 @@ new_segment:
                            ((unsigned long)q->dma_drain_buffer) &
                            (PAGE_SIZE - 1));
                nsegs++;
-               rq->data_len += q->dma_drain_size;
+               rq->extra_len += q->dma_drain_size;
        }
 
        if (sg)
@@ -238,36 +200,14 @@ new_segment:
 }
 EXPORT_SYMBOL(blk_rq_map_sg);
 
-static inline int ll_new_mergeable(struct request_queue *q,
-                                  struct request *req,
-                                  struct bio *bio)
-{
-       int nr_phys_segs = bio_phys_segments(q, bio);
-
-       if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
-               req->cmd_flags |= REQ_NOMERGE;
-               if (req == q->last_merge)
-                       q->last_merge = NULL;
-               return 0;
-       }
-
-       /*
-        * A hw segment is just getting larger, bump just the phys
-        * counter.
-        */
-       req->nr_phys_segments += nr_phys_segs;
-       return 1;
-}
-
 static inline int ll_new_hw_segment(struct request_queue *q,
                                    struct request *req,
                                    struct bio *bio)
 {
-       int nr_hw_segs = bio_hw_segments(q, bio);
        int nr_phys_segs = bio_phys_segments(q, bio);
 
-       if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
-           || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
+       if (req->nr_phys_segments + nr_phys_segs > queue_max_hw_segments(q) ||
+           req->nr_phys_segments + nr_phys_segs > queue_max_phys_segments(q)) {
                req->cmd_flags |= REQ_NOMERGE;
                if (req == q->last_merge)
                        q->last_merge = NULL;
@@ -278,7 +218,6 @@ static inline int ll_new_hw_segment(struct request_queue *q,
         * This will form the start of a new hw segment.  Bump both
         * counters.
         */
-       req->nr_hw_segments += nr_hw_segs;
        req->nr_phys_segments += nr_phys_segs;
        return 1;
 }
@@ -287,36 +226,22 @@ int ll_back_merge_fn(struct request_queue *q, struct request *req,
                     struct bio *bio)
 {
        unsigned short max_sectors;
-       int len;
 
        if (unlikely(blk_pc_request(req)))
-               max_sectors = q->max_hw_sectors;
+               max_sectors = queue_max_hw_sectors(q);
        else
-               max_sectors = q->max_sectors;
+               max_sectors = queue_max_sectors(q);
 
-       if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
+       if (blk_rq_sectors(req) + bio_sectors(bio) > max_sectors) {
                req->cmd_flags |= REQ_NOMERGE;
                if (req == q->last_merge)
                        q->last_merge = NULL;
                return 0;
        }
-       if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
+       if (!bio_flagged(req->biotail, BIO_SEG_VALID))
                blk_recount_segments(q, req->biotail);
-       if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
+       if (!bio_flagged(bio, BIO_SEG_VALID))
                blk_recount_segments(q, bio);
-       len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
-       if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio))
-           && !BIOVEC_VIRT_OVERSIZE(len)) {
-               int mergeable =  ll_new_mergeable(q, req, bio);
-
-               if (mergeable) {
-                       if (req->nr_hw_segments == 1)
-                               req->bio->bi_hw_front_size = len;
-                       if (bio->bi_hw_segments == 1)
-                               bio->bi_hw_back_size = len;
-               }
-               return mergeable;
-       }
 
        return ll_new_hw_segment(q, req, bio);
 }
@@ -325,37 +250,23 @@ int ll_front_merge_fn(struct request_queue *q, struct request *req,
                      struct bio *bio)
 {
        unsigned short max_sectors;
-       int len;
 
        if (unlikely(blk_pc_request(req)))
-               max_sectors = q->max_hw_sectors;
+               max_sectors = queue_max_hw_sectors(q);
        else
-               max_sectors = q->max_sectors;
+               max_sectors = queue_max_sectors(q);
 
 
-       if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
+       if (blk_rq_sectors(req) + bio_sectors(bio) > max_sectors) {
                req->cmd_flags |= REQ_NOMERGE;
                if (req == q->last_merge)
                        q->last_merge = NULL;
                return 0;
        }
-       len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
-       if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
+       if (!bio_flagged(bio, BIO_SEG_VALID))
                blk_recount_segments(q, bio);
-       if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
+       if (!bio_flagged(req->bio, BIO_SEG_VALID))
                blk_recount_segments(q, req->bio);
-       if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
-           !BIOVEC_VIRT_OVERSIZE(len)) {
-               int mergeable =  ll_new_mergeable(q, req, bio);
-
-               if (mergeable) {
-                       if (bio->bi_hw_segments == 1)
-                               bio->bi_hw_front_size = len;
-                       if (req->nr_hw_segments == 1)
-                               req->biotail->bi_hw_back_size = len;
-               }
-               return mergeable;
-       }
 
        return ll_new_hw_segment(q, req, bio);
 }
@@ -364,7 +275,8 @@ static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
                                struct request *next)
 {
        int total_phys_segments;
-       int total_hw_segments;
+       unsigned int seg_size =
+               req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
 
        /*
         * First check if the either of the requests are re-queued
@@ -376,39 +288,75 @@ static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
        /*
         * Will it become too large?
         */
-       if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
+       if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > queue_max_sectors(q))
                return 0;
 
        total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
-       if (blk_phys_contig_segment(q, req->biotail, next->bio))
+       if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
+               if (req->nr_phys_segments == 1)
+                       req->bio->bi_seg_front_size = seg_size;
+               if (next->nr_phys_segments == 1)
+                       next->biotail->bi_seg_back_size = seg_size;
                total_phys_segments--;
+       }
 
-       if (total_phys_segments > q->max_phys_segments)
+       if (total_phys_segments > queue_max_phys_segments(q))
                return 0;
 
-       total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
-       if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
-               int len = req->biotail->bi_hw_back_size +
-                               next->bio->bi_hw_front_size;
-               /*
-                * propagate the combined length to the end of the requests
-                */
-               if (req->nr_hw_segments == 1)
-                       req->bio->bi_hw_front_size = len;
-               if (next->nr_hw_segments == 1)
-                       next->biotail->bi_hw_back_size = len;
-               total_hw_segments--;
-       }
-
-       if (total_hw_segments > q->max_hw_segments)
+       if (total_phys_segments > queue_max_hw_segments(q))
                return 0;
 
        /* Merge is OK... */
        req->nr_phys_segments = total_phys_segments;
-       req->nr_hw_segments = total_hw_segments;
        return 1;
 }
 
+/**
+ * blk_rq_set_mixed_merge - mark a request as mixed merge
+ * @rq: request to mark as mixed merge
+ *
+ * Description:
+ *     @rq is about to be mixed merged.  Make sure the attributes
+ *     which can be mixed are set in each bio and mark @rq as mixed
+ *     merged.
+ */
+void blk_rq_set_mixed_merge(struct request *rq)
+{
+       unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
+       struct bio *bio;
+
+       if (rq->cmd_flags & REQ_MIXED_MERGE)
+               return;
+
+       /*
+        * @rq will no longer represent mixable attributes for all the
+        * contained bios.  It will just track those of the first one.
+        * Distributes the attributs to each bio.
+        */
+       for (bio = rq->bio; bio; bio = bio->bi_next) {
+               WARN_ON_ONCE((bio->bi_rw & REQ_FAILFAST_MASK) &&
+                            (bio->bi_rw & REQ_FAILFAST_MASK) != ff);
+               bio->bi_rw |= ff;
+       }
+       rq->cmd_flags |= REQ_MIXED_MERGE;
+}
+
+static void blk_account_io_merge(struct request *req)
+{
+       if (blk_do_io_stat(req)) {
+               struct hd_struct *part;
+               int cpu;
+
+               cpu = part_stat_lock();
+               part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
+
+               part_round_stats(cpu, part);
+               part_dec_in_flight(part, rq_data_dir(req));
+
+               part_stat_unlock();
+       }
+}
+
 /*
  * Has to be called with the request spinlock acquired
  */
@@ -421,7 +369,7 @@ static int attempt_merge(struct request_queue *q, struct request *req,
        /*
         * not contiguous
         */
-       if (req->sector + req->nr_sectors != next->sector)
+       if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
                return 0;
 
        if (rq_data_dir(req) != rq_data_dir(next)
@@ -429,6 +377,9 @@ static int attempt_merge(struct request_queue *q, struct request *req,
            || next->special)
                return 0;
 
+       if (blk_integrity_rq(req) != blk_integrity_rq(next))
+               return 0;
+
        /*
         * If we are allowed to merge, then append bio list
         * from next to rq and release next. merge_requests_fn
@@ -439,6 +390,19 @@ static int attempt_merge(struct request_queue *q, struct request *req,
                return 0;
 
        /*
+        * If failfast settings disagree or any of the two is already
+        * a mixed merge, mark both as mixed before proceeding.  This
+        * makes sure that all involved bios have mixable attributes
+        * set properly.
+        */
+       if ((req->cmd_flags | next->cmd_flags) & REQ_MIXED_MERGE ||
+           (req->cmd_flags & REQ_FAILFAST_MASK) !=
+           (next->cmd_flags & REQ_FAILFAST_MASK)) {
+               blk_rq_set_mixed_merge(req);
+               blk_rq_set_mixed_merge(next);
+       }
+
+       /*
         * At this point we have either done a back merge
         * or front merge. We need the smaller start_time of
         * the merged requests to be the current request
@@ -450,23 +414,21 @@ static int attempt_merge(struct request_queue *q, struct request *req,
        req->biotail->bi_next = next->bio;
        req->biotail = next->biotail;
 
-       req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
+       req->__data_len += blk_rq_bytes(next);
 
        elv_merge_requests(q, req, next);
 
-       if (req->rq_disk) {
-               struct hd_struct *part
-                       = get_part(req->rq_disk, req->sector);
-               disk_round_stats(req->rq_disk);
-               req->rq_disk->in_flight--;
-               if (part) {
-                       part_round_stats(part);
-                       part->in_flight--;
-               }
-       }
+       /*
+        * 'next' is going away, so update stats accordingly
+        */
+       blk_account_io_merge(next);
 
        req->ioprio = ioprio_best(req->ioprio, next->ioprio);
+       if (blk_rq_cpu_valid(next))
+               req->cpu = next->cpu;
 
+       /* owner-ship of bio passed from next to req */
+       next->bio = NULL;
        __blk_put_request(q, next);
        return 1;
 }