block: make bi_phys_segments an unsigned int instead of short
[linux-2.6.git] / drivers / md / raid5.c
index 4426220..37e5465 100644 (file)
 const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256)));
 #endif
 
+/*
+ * We maintain a biased count of active stripes in the bottom 16 bits of
+ * bi_phys_segments, and a count of processed stripes in the upper 16 bits
+ */
+static inline int raid5_bi_phys_segments(struct bio *bio)
+{
+       return bio->bi_phys_segments & 0xffff;
+}
+
+static inline int raid5_bi_hw_segments(struct bio *bio)
+{
+       return (bio->bi_phys_segments >> 16) & 0xffff;
+}
+
+static inline int raid5_dec_bi_phys_segments(struct bio *bio)
+{
+       --bio->bi_phys_segments;
+       return raid5_bi_phys_segments(bio);
+}
+
+static inline int raid5_dec_bi_hw_segments(struct bio *bio)
+{
+       unsigned short val = raid5_bi_hw_segments(bio);
+
+       --val;
+       bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
+       return val;
+}
+
+static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
+{
+       bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16);
+}
+
 static inline int raid6_next_disk(int disk, int raid_disks)
 {
        disk++;
@@ -507,7 +541,7 @@ static void ops_complete_biofill(void *stripe_head_ref)
                        while (rbi && rbi->bi_sector <
                                dev->sector + STRIPE_SECTORS) {
                                rbi2 = r5_next_bio(rbi, dev->sector);
-                               if (--rbi->bi_phys_segments == 0) {
+                               if (!raid5_dec_bi_phys_segments(rbi)) {
                                        rbi->bi_next = return_bi;
                                        return_bi = rbi;
                                }
@@ -911,14 +945,16 @@ static int resize_stripes(raid5_conf_t *conf, int newsize)
        struct stripe_head *osh, *nsh;
        LIST_HEAD(newstripes);
        struct disk_info *ndisks;
-       int err = 0;
+       int err;
        struct kmem_cache *sc;
        int i;
 
        if (newsize <= conf->pool_size)
                return 0; /* never bother to shrink */
 
-       md_allow_write(conf->mddev);
+       err = md_allow_write(conf->mddev);
+       if (err)
+               return err;
 
        /* Step 1 */
        sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
@@ -1723,7 +1759,7 @@ static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, in
        if (*bip)
                bi->bi_next = *bip;
        *bip = bi;
-       bi->bi_phys_segments ++;
+       bi->bi_phys_segments++;
        spin_unlock_irq(&conf->device_lock);
        spin_unlock(&sh->lock);
 
@@ -1817,7 +1853,7 @@ handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
                        sh->dev[i].sector + STRIPE_SECTORS) {
                        struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
                        clear_bit(BIO_UPTODATE, &bi->bi_flags);
-                       if (--bi->bi_phys_segments == 0) {
+                       if (!raid5_dec_bi_phys_segments(bi)) {
                                md_write_end(conf->mddev);
                                bi->bi_next = *return_bi;
                                *return_bi = bi;
@@ -1832,7 +1868,7 @@ handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
                       sh->dev[i].sector + STRIPE_SECTORS) {
                        struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
                        clear_bit(BIO_UPTODATE, &bi->bi_flags);
-                       if (--bi->bi_phys_segments == 0) {
+                       if (!raid5_dec_bi_phys_segments(bi)) {
                                md_write_end(conf->mddev);
                                bi->bi_next = *return_bi;
                                *return_bi = bi;
@@ -1856,7 +1892,7 @@ handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
                                struct bio *nextbi =
                                        r5_next_bio(bi, sh->dev[i].sector);
                                clear_bit(BIO_UPTODATE, &bi->bi_flags);
-                               if (--bi->bi_phys_segments == 0) {
+                               if (!raid5_dec_bi_phys_segments(bi)) {
                                        bi->bi_next = *return_bi;
                                        *return_bi = bi;
                                }
@@ -2031,7 +2067,7 @@ static void handle_stripe_clean_event(raid5_conf_t *conf,
                                while (wbi && wbi->bi_sector <
                                        dev->sector + STRIPE_SECTORS) {
                                        wbi2 = r5_next_bio(wbi, dev->sector);
-                                       if (--wbi->bi_phys_segments == 0) {
+                                       if (!raid5_dec_bi_phys_segments(wbi)) {
                                                md_write_end(conf->mddev);
                                                wbi->bi_next = *return_bi;
                                                *return_bi = wbi;
@@ -2505,7 +2541,7 @@ static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
  *
  */
 
-static void handle_stripe5(struct stripe_head *sh)
+static bool handle_stripe5(struct stripe_head *sh)
 {
        raid5_conf_t *conf = sh->raid_conf;
        int disks = sh->disks, i;
@@ -2566,10 +2602,10 @@ static void handle_stripe5(struct stripe_head *sh)
                if (dev->written)
                        s.written++;
                rdev = rcu_dereference(conf->disks[i].rdev);
-               if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
+               if (blocked_rdev == NULL &&
+                   rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
                        blocked_rdev = rdev;
                        atomic_inc(&rdev->nr_pending);
-                       break;
                }
                if (!rdev || !test_bit(In_sync, &rdev->flags)) {
                        /* The ReadError flag will just be confusing now */
@@ -2586,8 +2622,14 @@ static void handle_stripe5(struct stripe_head *sh)
        rcu_read_unlock();
 
        if (unlikely(blocked_rdev)) {
-               set_bit(STRIPE_HANDLE, &sh->state);
-               goto unlock;
+               if (s.syncing || s.expanding || s.expanded ||
+                   s.to_write || s.written) {
+                       set_bit(STRIPE_HANDLE, &sh->state);
+                       goto unlock;
+               }
+               /* There is nothing for the blocked_rdev to block */
+               rdev_dec_pending(blocked_rdev, conf->mddev);
+               blocked_rdev = NULL;
        }
 
        if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
@@ -2715,10 +2757,11 @@ static void handle_stripe5(struct stripe_head *sh)
        if (sh->reconstruct_state == reconstruct_state_result) {
                sh->reconstruct_state = reconstruct_state_idle;
                clear_bit(STRIPE_EXPANDING, &sh->state);
-               for (i = conf->raid_disks; i--; )
+               for (i = conf->raid_disks; i--; ) {
                        set_bit(R5_Wantwrite, &sh->dev[i].flags);
-                       set_bit(R5_LOCKED, &dev->flags);
+                       set_bit(R5_LOCKED, &sh->dev[i].flags);
                        s.locked++;
+               }
        }
 
        if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
@@ -2752,9 +2795,11 @@ static void handle_stripe5(struct stripe_head *sh)
        ops_run_io(sh, &s);
 
        return_io(return_bi);
+
+       return blocked_rdev == NULL;
 }
 
-static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
+static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
 {
        raid6_conf_t *conf = sh->raid_conf;
        int disks = sh->disks;
@@ -2803,7 +2848,7 @@ static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
                                copy_data(0, rbi, dev->page, dev->sector);
                                rbi2 = r5_next_bio(rbi, dev->sector);
                                spin_lock_irq(&conf->device_lock);
-                               if (--rbi->bi_phys_segments == 0) {
+                               if (!raid5_dec_bi_phys_segments(rbi)) {
                                        rbi->bi_next = return_bi;
                                        return_bi = rbi;
                                }
@@ -2827,10 +2872,10 @@ static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
                if (dev->written)
                        s.written++;
                rdev = rcu_dereference(conf->disks[i].rdev);
-               if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
+               if (blocked_rdev == NULL &&
+                   rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
                        blocked_rdev = rdev;
                        atomic_inc(&rdev->nr_pending);
-                       break;
                }
                if (!rdev || !test_bit(In_sync, &rdev->flags)) {
                        /* The ReadError flag will just be confusing now */
@@ -2848,9 +2893,16 @@ static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
        rcu_read_unlock();
 
        if (unlikely(blocked_rdev)) {
-               set_bit(STRIPE_HANDLE, &sh->state);
-               goto unlock;
+               if (s.syncing || s.expanding || s.expanded ||
+                   s.to_write || s.written) {
+                       set_bit(STRIPE_HANDLE, &sh->state);
+                       goto unlock;
+               }
+               /* There is nothing for the blocked_rdev to block */
+               rdev_dec_pending(blocked_rdev, conf->mddev);
+               blocked_rdev = NULL;
        }
+
        pr_debug("locked=%d uptodate=%d to_read=%d"
               " to_write=%d failed=%d failed_num=%d,%d\n",
               s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
@@ -2965,14 +3017,17 @@ static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
        ops_run_io(sh, &s);
 
        return_io(return_bi);
+
+       return blocked_rdev == NULL;
 }
 
-static void handle_stripe(struct stripe_head *sh, struct page *tmp_page)
+/* returns true if the stripe was handled */
+static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page)
 {
        if (sh->raid_conf->level == 6)
-               handle_stripe6(sh, tmp_page);
+               return handle_stripe6(sh, tmp_page);
        else
-               handle_stripe5(sh);
+               return handle_stripe5(sh);
 }
 
 
@@ -3071,15 +3126,17 @@ static int raid5_congested(void *data, int bits)
 /* We want read requests to align with chunks where possible,
  * but write requests don't need to.
  */
-static int raid5_mergeable_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *biovec)
+static int raid5_mergeable_bvec(struct request_queue *q,
+                               struct bvec_merge_data *bvm,
+                               struct bio_vec *biovec)
 {
        mddev_t *mddev = q->queuedata;
-       sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
+       sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
        int max;
        unsigned int chunk_sectors = mddev->chunk_size >> 9;
-       unsigned int bio_sectors = bio->bi_size >> 9;
+       unsigned int bio_sectors = bvm->bi_size >> 9;
 
-       if (bio_data_dir(bio) == WRITE)
+       if ((bvm->bi_rw & 1) == WRITE)
                return biovec->bv_len; /* always allow writes to be mergeable */
 
        max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
@@ -3132,8 +3189,11 @@ static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
        if(bi) {
                conf->retry_read_aligned_list = bi->bi_next;
                bi->bi_next = NULL;
+               /*
+                * this sets the active strip count to 1 and the processed
+                * strip count to zero (upper 8 bits)
+                */
                bi->bi_phys_segments = 1; /* biased count of active stripes */
-               bi->bi_hw_segments = 0; /* count of processed stripes */
        }
 
        return bi;
@@ -3183,8 +3243,7 @@ static int bio_fits_rdev(struct bio *bi)
        if ((bi->bi_size>>9) > q->max_sectors)
                return 0;
        blk_recount_segments(q, bi);
-       if (bi->bi_phys_segments > q->max_phys_segments ||
-           bi->bi_hw_segments > q->max_hw_segments)
+       if (bi->bi_phys_segments > q->max_phys_segments)
                return 0;
 
        if (q->merge_bvec_fn)
@@ -3445,7 +3504,7 @@ static int make_request(struct request_queue *q, struct bio * bi)
                        
        }
        spin_lock_irq(&conf->device_lock);
-       remaining = --bi->bi_phys_segments;
+       remaining = raid5_dec_bi_phys_segments(bi);
        spin_unlock_irq(&conf->device_lock);
        if (remaining == 0) {
 
@@ -3538,7 +3597,7 @@ static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped
                            j == raid6_next_disk(sh->pd_idx, sh->disks))
                                continue;
                        s = compute_blocknr(sh, j);
-                       if (s < (mddev->array_size<<1)) {
+                       if (s < mddev->array_sectors) {
                                skipped = 1;
                                continue;
                        }
@@ -3688,7 +3747,9 @@ static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *ski
        clear_bit(STRIPE_INSYNC, &sh->state);
        spin_unlock(&sh->lock);
 
-       handle_stripe(sh, NULL);
+       /* wait for any blocked device to be handled */
+       while(unlikely(!handle_stripe(sh, NULL)))
+               ;
        release_stripe(sh);
 
        return STRIPE_SECTORS;
@@ -3727,7 +3788,7 @@ static int  retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
                     sector += STRIPE_SECTORS,
                     scnt++) {
 
-               if (scnt < raid_bio->bi_hw_segments)
+               if (scnt < raid5_bi_hw_segments(raid_bio))
                        /* already done this stripe */
                        continue;
 
@@ -3735,7 +3796,7 @@ static int  retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
 
                if (!sh) {
                        /* failed to get a stripe - must wait */
-                       raid_bio->bi_hw_segments = scnt;
+                       raid5_set_bi_hw_segments(raid_bio, scnt);
                        conf->retry_read_aligned = raid_bio;
                        return handled;
                }
@@ -3743,7 +3804,7 @@ static int  retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
                set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
                if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
                        release_stripe(sh);
-                       raid_bio->bi_hw_segments = scnt;
+                       raid5_set_bi_hw_segments(raid_bio, scnt);
                        conf->retry_read_aligned = raid_bio;
                        return handled;
                }
@@ -3753,7 +3814,7 @@ static int  retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
                handled++;
        }
        spin_lock_irq(&conf->device_lock);
-       remaining = --raid_bio->bi_phys_segments;
+       remaining = raid5_dec_bi_phys_segments(raid_bio);
        spin_unlock_irq(&conf->device_lock);
        if (remaining == 0)
                bio_endio(raid_bio, 0);
@@ -3807,10 +3868,8 @@ static void raid5d(mddev_t *mddev)
 
                sh = __get_priority_stripe(conf);
 
-               if (!sh) {
-                       async_tx_issue_pending_all();
+               if (!sh)
                        break;
-               }
                spin_unlock_irq(&conf->device_lock);
                
                handled++;
@@ -3823,6 +3882,7 @@ static void raid5d(mddev_t *mddev)
 
        spin_unlock_irq(&conf->device_lock);
 
+       async_tx_issue_pending_all();
        unplug_slaves(mddev);
 
        pr_debug("--- raid5d inactive\n");
@@ -3843,6 +3903,8 @@ raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
 {
        raid5_conf_t *conf = mddev_to_conf(mddev);
        unsigned long new;
+       int err;
+
        if (len >= PAGE_SIZE)
                return -EINVAL;
        if (!conf)
@@ -3858,7 +3920,9 @@ raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
                else
                        break;
        }
-       md_allow_write(mddev);
+       err = md_allow_write(mddev);
+       if (err)
+               return err;
        while (new > conf->max_nr_stripes) {
                if (grow_one_stripe(conf))
                        conf->max_nr_stripes++;
@@ -4183,7 +4247,7 @@ static int run(mddev_t *mddev)
        mddev->queue->backing_dev_info.congested_data = mddev;
        mddev->queue->backing_dev_info.congested_fn = raid5_congested;
 
-       mddev->array_size =  mddev->size * (conf->previous_raid_disks -
+       mddev->array_sectors = 2 * mddev->size * (conf->previous_raid_disks -
                                            conf->max_degraded);
 
        blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
@@ -4407,8 +4471,9 @@ static int raid5_resize(mddev_t *mddev, sector_t sectors)
        raid5_conf_t *conf = mddev_to_conf(mddev);
 
        sectors &= ~((sector_t)mddev->chunk_size/512 - 1);
-       mddev->array_size = (sectors * (mddev->raid_disks-conf->max_degraded))>>1;
-       set_capacity(mddev->gendisk, mddev->array_size << 1);
+       mddev->array_sectors = sectors * (mddev->raid_disks
+                                         - conf->max_degraded);
+       set_capacity(mddev->gendisk, mddev->array_sectors);
        mddev->changed = 1;
        if (sectors/2  > mddev->size && mddev->recovery_cp == MaxSector) {
                mddev->recovery_cp = mddev->size << 1;
@@ -4430,6 +4495,9 @@ static int raid5_check_reshape(mddev_t *mddev)
                return -EINVAL; /* Cannot shrink array or change level yet */
        if (mddev->delta_disks == 0)
                return 0; /* nothing to do */
+       if (mddev->bitmap)
+               /* Cannot grow a bitmap yet */
+               return -EBUSY;
 
        /* Can only proceed if there are plenty of stripe_heads.
         * We need a minimum of one full stripe,, and for sensible progress
@@ -4541,15 +4609,16 @@ static void end_reshape(raid5_conf_t *conf)
        struct block_device *bdev;
 
        if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
-               conf->mddev->array_size = conf->mddev->size *
+               conf->mddev->array_sectors = 2 * conf->mddev->size *
                        (conf->raid_disks - conf->max_degraded);
-               set_capacity(conf->mddev->gendisk, conf->mddev->array_size << 1);
+               set_capacity(conf->mddev->gendisk, conf->mddev->array_sectors);
                conf->mddev->changed = 1;
 
                bdev = bdget_disk(conf->mddev->gendisk, 0);
                if (bdev) {
                        mutex_lock(&bdev->bd_inode->i_mutex);
-                       i_size_write(bdev->bd_inode, (loff_t)conf->mddev->array_size << 10);
+                       i_size_write(bdev->bd_inode,
+                                    (loff_t)conf->mddev->array_sectors << 9);
                        mutex_unlock(&bdev->bd_inode->i_mutex);
                        bdput(bdev);
                }