md/raid10: collect some geometry fields into a dedicated structure.
authorNeilBrown <neilb@suse.de>
Sun, 20 May 2012 23:28:20 +0000 (09:28 +1000)
committerNeilBrown <neilb@suse.de>
Sun, 20 May 2012 23:28:20 +0000 (09:28 +1000)
We will shortly be adding reshape support for RAID10 which will
require it having 2 concurrent geometries (before and after).
To make that easier, collect most geometry fields into 'struct geom'
and access them from there.  Then we will more easily be able to add
a second set of fields.

Note that 'copies' is not in this struct and so cannot be changed.
There is little need to change this number and doing so is a lot
more difficult as it requires reallocating more things.
So leave it out for now.

Signed-off-by: NeilBrown <neilb@suse.de>
drivers/md/raid10.c
drivers/md/raid10.h

index 832fb4d..36f445f 100644 (file)
@@ -511,42 +511,43 @@ static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio)
        sector_t chunk;
        sector_t stripe;
        int dev;
+       struct geom *geo = &conf->geo;
 
        int slot = 0;
 
        /* now calculate first sector/dev */
-       chunk = r10bio->sector >> conf->chunk_shift;
-       sector = r10bio->sector & conf->chunk_mask;
+       chunk = r10bio->sector >> geo->chunk_shift;
+       sector = r10bio->sector & geo->chunk_mask;
 
-       chunk *= conf->near_copies;
+       chunk *= geo->near_copies;
        stripe = chunk;
-       dev = sector_div(stripe, conf->raid_disks);
-       if (conf->far_offset)
-               stripe *= conf->far_copies;
+       dev = sector_div(stripe, geo->raid_disks);
+       if (geo->far_offset)
+               stripe *= geo->far_copies;
 
-       sector += stripe << conf->chunk_shift;
+       sector += stripe << geo->chunk_shift;
 
        /* and calculate all the others */
-       for (n=0; n < conf->near_copies; n++) {
+       for (n = 0; n < geo->near_copies; n++) {
                int d = dev;
                sector_t s = sector;
                r10bio->devs[slot].addr = sector;
                r10bio->devs[slot].devnum = d;
                slot++;
 
-               for (f = 1; f < conf->far_copies; f++) {
-                       d += conf->near_copies;
-                       if (d >= conf->raid_disks)
-                               d -= conf->raid_disks;
-                       s += conf->stride;
+               for (f = 1; f < geo->far_copies; f++) {
+                       d += geo->near_copies;
+                       if (d >= geo->raid_disks)
+                               d -= geo->raid_disks;
+                       s += geo->stride;
                        r10bio->devs[slot].devnum = d;
                        r10bio->devs[slot].addr = s;
                        slot++;
                }
                dev++;
-               if (dev >= conf->raid_disks) {
+               if (dev >= geo->raid_disks) {
                        dev = 0;
-                       sector += (conf->chunk_mask + 1);
+                       sector += (geo->chunk_mask + 1);
                }
        }
        BUG_ON(slot != conf->copies);
@@ -555,28 +556,29 @@ static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio)
 static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev)
 {
        sector_t offset, chunk, vchunk;
+       struct geom *geo = &conf->geo;
 
-       offset = sector & conf->chunk_mask;
-       if (conf->far_offset) {
+       offset = sector & geo->chunk_mask;
+       if (geo->far_offset) {
                int fc;
-               chunk = sector >> conf->chunk_shift;
-               fc = sector_div(chunk, conf->far_copies);
-               dev -= fc * conf->near_copies;
+               chunk = sector >> geo->chunk_shift;
+               fc = sector_div(chunk, geo->far_copies);
+               dev -= fc * geo->near_copies;
                if (dev < 0)
-                       dev += conf->raid_disks;
+                       dev += geo->raid_disks;
        } else {
-               while (sector >= conf->stride) {
-                       sector -= conf->stride;
-                       if (dev < conf->near_copies)
-                               dev += conf->raid_disks - conf->near_copies;
+               while (sector >= geo->stride) {
+                       sector -= geo->stride;
+                       if (dev < geo->near_copies)
+                               dev += geo->raid_disks - geo->near_copies;
                        else
-                               dev -= conf->near_copies;
+                               dev -= geo->near_copies;
                }
-               chunk = sector >> conf->chunk_shift;
+               chunk = sector >> geo->chunk_shift;
        }
-       vchunk = chunk * conf->raid_disks + dev;
-       sector_div(vchunk, conf->near_copies);
-       return (vchunk << conf->chunk_shift) + offset;
+       vchunk = chunk * geo->raid_disks + dev;
+       sector_div(vchunk, geo->near_copies);
+       return (vchunk << geo->chunk_shift) + offset;
 }
 
 /**
@@ -599,8 +601,9 @@ static int raid10_mergeable_bvec(struct request_queue *q,
        int max;
        unsigned int chunk_sectors = mddev->chunk_sectors;
        unsigned int bio_sectors = bvm->bi_size >> 9;
+       struct geom *geo = &conf->geo;
 
-       if (conf->near_copies < conf->raid_disks) {
+       if (geo->near_copies < geo->raid_disks) {
                max = (chunk_sectors - ((sector & (chunk_sectors - 1))
                                        + bio_sectors)) << 9;
                if (max < 0)
@@ -681,6 +684,7 @@ static struct md_rdev *read_balance(struct r10conf *conf,
        struct md_rdev *rdev, *best_rdev;
        int do_balance;
        int best_slot;
+       struct geom *geo = &conf->geo;
 
        raid10_find_phys(conf, r10_bio);
        rcu_read_lock();
@@ -761,11 +765,11 @@ retry:
                 * sequential read speed for 'far copies' arrays.  So only
                 * keep it for 'near' arrays, and review those later.
                 */
-               if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending))
+               if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending))
                        break;
 
                /* for far > 1 always use the lowest address */
-               if (conf->far_copies > 1)
+               if (geo->far_copies > 1)
                        new_distance = r10_bio->devs[slot].addr;
                else
                        new_distance = abs(r10_bio->devs[slot].addr -
@@ -812,7 +816,7 @@ static int raid10_congested(void *data, int bits)
        if (mddev_congested(mddev, bits))
                return 1;
        rcu_read_lock();
-       for (i = 0; i < conf->raid_disks && ret == 0; i++) {
+       for (i = 0; i < conf->geo.raid_disks && ret == 0; i++) {
                struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
                if (rdev && !test_bit(Faulty, &rdev->flags)) {
                        struct request_queue *q = bdev_get_queue(rdev->bdev);
@@ -979,7 +983,8 @@ static void make_request(struct mddev *mddev, struct bio * bio)
        struct r10bio *r10_bio;
        struct bio *read_bio;
        int i;
-       int chunk_sects = conf->chunk_mask + 1;
+       sector_t chunk_mask = conf->geo.chunk_mask;
+       int chunk_sects = chunk_mask + 1;
        const int rw = bio_data_dir(bio);
        const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
        const unsigned long do_fua = (bio->bi_rw & REQ_FUA);
@@ -997,9 +1002,9 @@ static void make_request(struct mddev *mddev, struct bio * bio)
        /* If this request crosses a chunk boundary, we need to
         * split it.  This will only happen for 1 PAGE (or less) requests.
         */
-       if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9)
-                     > chunk_sects &&
-                   conf->near_copies < conf->raid_disks)) {
+       if (unlikely((bio->bi_sector & chunk_mask) + (bio->bi_size >> 9)
+                    > chunk_sects
+                    && conf->geo.near_copies < conf->geo.raid_disks)) {
                struct bio_pair *bp;
                /* Sanity check -- queue functions should prevent this happening */
                if (bio->bi_vcnt != 1 ||
@@ -1368,19 +1373,19 @@ static void status(struct seq_file *seq, struct mddev *mddev)
        struct r10conf *conf = mddev->private;
        int i;
 
-       if (conf->near_copies < conf->raid_disks)
+       if (conf->geo.near_copies < conf->geo.raid_disks)
                seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2);
-       if (conf->near_copies > 1)
-               seq_printf(seq, " %d near-copies", conf->near_copies);
-       if (conf->far_copies > 1) {
-               if (conf->far_offset)
-                       seq_printf(seq, " %d offset-copies", conf->far_copies);
+       if (conf->geo.near_copies > 1)
+               seq_printf(seq, " %d near-copies", conf->geo.near_copies);
+       if (conf->geo.far_copies > 1) {
+               if (conf->geo.far_offset)
+                       seq_printf(seq, " %d offset-copies", conf->geo.far_copies);
                else
-                       seq_printf(seq, " %d far-copies", conf->far_copies);
+                       seq_printf(seq, " %d far-copies", conf->geo.far_copies);
        }
-       seq_printf(seq, " [%d/%d] [", conf->raid_disks,
-                                       conf->raid_disks - mddev->degraded);
-       for (i = 0; i < conf->raid_disks; i++)
+       seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks,
+                                       conf->geo.raid_disks - mddev->degraded);
+       for (i = 0; i < conf->geo.raid_disks; i++)
                seq_printf(seq, "%s",
                              conf->mirrors[i].rdev &&
                              test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
@@ -1403,7 +1408,7 @@ static int enough(struct r10conf *conf, int ignore)
                        if (conf->mirrors[first].rdev &&
                            first != ignore)
                                cnt++;
-                       first = (first+1) % conf->raid_disks;
+                       first = (first+1) % conf->geo.raid_disks;
                }
                if (cnt == 0)
                        return 0;
@@ -1445,7 +1450,7 @@ static void error(struct mddev *mddev, struct md_rdev *rdev)
               "md/raid10:%s: Disk failure on %s, disabling device.\n"
               "md/raid10:%s: Operation continuing on %d devices.\n",
               mdname(mddev), bdevname(rdev->bdev, b),
-              mdname(mddev), conf->raid_disks - mddev->degraded);
+              mdname(mddev), conf->geo.raid_disks - mddev->degraded);
 }
 
 static void print_conf(struct r10conf *conf)
@@ -1458,10 +1463,10 @@ static void print_conf(struct r10conf *conf)
                printk(KERN_DEBUG "(!conf)\n");
                return;
        }
-       printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
-               conf->raid_disks);
+       printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded,
+               conf->geo.raid_disks);
 
-       for (i = 0; i < conf->raid_disks; i++) {
+       for (i = 0; i < conf->geo.raid_disks; i++) {
                char b[BDEVNAME_SIZE];
                tmp = conf->mirrors + i;
                if (tmp->rdev)
@@ -1493,7 +1498,7 @@ static int raid10_spare_active(struct mddev *mddev)
         * Find all non-in_sync disks within the RAID10 configuration
         * and mark them in_sync
         */
-       for (i = 0; i < conf->raid_disks; i++) {
+       for (i = 0; i < conf->geo.raid_disks; i++) {
                tmp = conf->mirrors + i;
                if (tmp->replacement
                    && tmp->replacement->recovery_offset == MaxSector
@@ -1535,7 +1540,7 @@ static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev)
        int err = -EEXIST;
        int mirror;
        int first = 0;
-       int last = conf->raid_disks - 1;
+       int last = conf->geo.raid_disks - 1;
        struct request_queue *q = bdev_get_queue(rdev->bdev);
 
        if (mddev->recovery_cp < MaxSector)
@@ -2603,7 +2608,7 @@ static int init_resync(struct r10conf *conf)
        buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
        BUG_ON(conf->r10buf_pool);
        conf->have_replacement = 0;
-       for (i = 0; i < conf->raid_disks; i++)
+       for (i = 0; i < conf->geo.raid_disks; i++)
                if (conf->mirrors[i].replacement)
                        conf->have_replacement = 1;
        conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf);
@@ -2657,6 +2662,7 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
        sector_t sync_blocks;
        sector_t sectors_skipped = 0;
        int chunks_skipped = 0;
+       sector_t chunk_mask = conf->geo.chunk_mask;
 
        if (!conf->r10buf_pool)
                if (init_resync(conf))
@@ -2680,7 +2686,7 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
                        if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
                                bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
                                                &sync_blocks, 1);
-                       else for (i=0; i<conf->raid_disks; i++) {
+                       else for (i = 0; i < conf->geo.raid_disks; i++) {
                                sector_t sect =
                                        raid10_find_virt(conf, mddev->curr_resync, i);
                                bitmap_end_sync(mddev->bitmap, sect,
@@ -2694,7 +2700,7 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
                                /* Completed a full sync so the replacements
                                 * are now fully recovered.
                                 */
-                               for (i = 0; i < conf->raid_disks; i++)
+                               for (i = 0; i < conf->geo.raid_disks; i++)
                                        if (conf->mirrors[i].replacement)
                                                conf->mirrors[i].replacement
                                                        ->recovery_offset
@@ -2707,7 +2713,7 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
                *skipped = 1;
                return sectors_skipped;
        }
-       if (chunks_skipped >= conf->raid_disks) {
+       if (chunks_skipped >= conf->geo.raid_disks) {
                /* if there has been nothing to do on any drive,
                 * then there is nothing to do at all..
                 */
@@ -2721,9 +2727,9 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
        /* make sure whole request will fit in a chunk - if chunks
         * are meaningful
         */
-       if (conf->near_copies < conf->raid_disks &&
-           max_sector > (sector_nr | conf->chunk_mask))
-               max_sector = (sector_nr | conf->chunk_mask) + 1;
+       if (conf->geo.near_copies < conf->geo.raid_disks &&
+           max_sector > (sector_nr | chunk_mask))
+               max_sector = (sector_nr | chunk_mask) + 1;
        /*
         * If there is non-resync activity waiting for us then
         * put in a delay to throttle resync.
@@ -2752,7 +2758,7 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
                int j;
                r10_bio = NULL;
 
-               for (i=0 ; i<conf->raid_disks; i++) {
+               for (i = 0 ; i < conf->geo.raid_disks; i++) {
                        int still_degraded;
                        struct r10bio *rb2;
                        sector_t sect;
@@ -2806,7 +2812,7 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
                        /* Need to check if the array will still be
                         * degraded
                         */
-                       for (j=0; j<conf->raid_disks; j++)
+                       for (j = 0; j < conf->geo.raid_disks; j++)
                                if (conf->mirrors[j].rdev == NULL ||
                                    test_bit(Faulty, &conf->mirrors[j].rdev->flags)) {
                                        still_degraded = 1;
@@ -2984,9 +2990,9 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
                r10_bio->sector = sector_nr;
                set_bit(R10BIO_IsSync, &r10_bio->state);
                raid10_find_phys(conf, r10_bio);
-               r10_bio->sectors = (sector_nr | conf->chunk_mask) - sector_nr +1;
+               r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1;
 
-               for (i=0; i<conf->copies; i++) {
+               for (i = 0; i < conf->copies; i++) {
                        int d = r10_bio->devs[i].devnum;
                        sector_t first_bad, sector;
                        int bad_sectors;
@@ -3152,16 +3158,16 @@ raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks)
        struct r10conf *conf = mddev->private;
 
        if (!raid_disks)
-               raid_disks = conf->raid_disks;
+               raid_disks = conf->geo.raid_disks;
        if (!sectors)
                sectors = conf->dev_sectors;
 
-       size = sectors >> conf->chunk_shift;
-       sector_div(size, conf->far_copies);
+       size = sectors >> conf->geo.chunk_shift;
+       sector_div(size, conf->geo.far_copies);
        size = size * raid_disks;
-       sector_div(size, conf->near_copies);
+       sector_div(size, conf->geo.near_copies);
 
-       return size << conf->chunk_shift;
+       return size << conf->geo.chunk_shift;
 }
 
 static void calc_sectors(struct r10conf *conf, sector_t size)
@@ -3171,10 +3177,10 @@ static void calc_sectors(struct r10conf *conf, sector_t size)
         * conf->stride
         */
 
-       size = size >> conf->chunk_shift;
-       sector_div(size, conf->far_copies);
-       size = size * conf->raid_disks;
-       sector_div(size, conf->near_copies);
+       size = size >> conf->geo.chunk_shift;
+       sector_div(size, conf->geo.far_copies);
+       size = size * conf->geo.raid_disks;
+       sector_div(size, conf->geo.near_copies);
        /* 'size' is now the number of chunks in the array */
        /* calculate "used chunks per device" */
        size = size * conf->copies;
@@ -3182,15 +3188,15 @@ static void calc_sectors(struct r10conf *conf, sector_t size)
        /* We need to round up when dividing by raid_disks to
         * get the stride size.
         */
-       size = DIV_ROUND_UP_SECTOR_T(size, conf->raid_disks);
+       size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks);
 
-       conf->dev_sectors = size << conf->chunk_shift;
+       conf->dev_sectors = size << conf->geo.chunk_shift;
 
-       if (conf->far_offset)
-               conf->stride = 1 << conf->chunk_shift;
+       if (conf->geo.far_offset)
+               conf->geo.stride = 1 << conf->geo.chunk_shift;
        else {
-               sector_div(size, conf->far_copies);
-               conf->stride = size << conf->chunk_shift;
+               sector_div(size, conf->geo.far_copies);
+               conf->geo.stride = size << conf->geo.chunk_shift;
        }
 }
 
@@ -3234,13 +3240,13 @@ static struct r10conf *setup_conf(struct mddev *mddev)
                goto out;
 
 
-       conf->raid_disks = mddev->raid_disks;
-       conf->near_copies = nc;
-       conf->far_copies = fc;
+       conf->geo.raid_disks = mddev->raid_disks;
+       conf->geo.near_copies = nc;
+       conf->geo.far_copies = fc;
        conf->copies = nc*fc;
-       conf->far_offset = fo;
-       conf->chunk_mask = mddev->new_chunk_sectors - 1;
-       conf->chunk_shift = ffz(~mddev->new_chunk_sectors);
+       conf->geo.far_offset = fo;
+       conf->geo.chunk_mask = mddev->new_chunk_sectors - 1;
+       conf->geo.chunk_shift = ffz(~mddev->new_chunk_sectors);
 
        conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc,
                                           r10bio_pool_free, conf);
@@ -3304,16 +3310,16 @@ static int run(struct mddev *mddev)
 
        chunk_size = mddev->chunk_sectors << 9;
        blk_queue_io_min(mddev->queue, chunk_size);
-       if (conf->raid_disks % conf->near_copies)
-               blk_queue_io_opt(mddev->queue, chunk_size * conf->raid_disks);
+       if (conf->geo.raid_disks % conf->geo.near_copies)
+               blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks);
        else
                blk_queue_io_opt(mddev->queue, chunk_size *
-                                (conf->raid_disks / conf->near_copies));
+                                (conf->geo.raid_disks / conf->geo.near_copies));
 
        rdev_for_each(rdev, mddev) {
 
                disk_idx = rdev->raid_disk;
-               if (disk_idx >= conf->raid_disks
+               if (disk_idx >= conf->geo.raid_disks
                    || disk_idx < 0)
                        continue;
                disk = conf->mirrors + disk_idx;
@@ -3341,7 +3347,7 @@ static int run(struct mddev *mddev)
        }
 
        mddev->degraded = 0;
-       for (i = 0; i < conf->raid_disks; i++) {
+       for (i = 0; i < conf->geo.raid_disks; i++) {
 
                disk = conf->mirrors + i;
 
@@ -3368,8 +3374,8 @@ static int run(struct mddev *mddev)
                       mdname(mddev));
        printk(KERN_INFO
                "md/raid10:%s: active with %d out of %d devices\n",
-               mdname(mddev), conf->raid_disks - mddev->degraded,
-               conf->raid_disks);
+               mdname(mddev), conf->geo.raid_disks - mddev->degraded,
+               conf->geo.raid_disks);
        /*
         * Ok, everything is just fine now
         */
@@ -3386,9 +3392,9 @@ static int run(struct mddev *mddev)
         * maybe...
         */
        {
-               int stripe = conf->raid_disks *
+               int stripe = conf->geo.raid_disks *
                        ((mddev->chunk_sectors << 9) / PAGE_SIZE);
-               stripe /= conf->near_copies;
+               stripe /= conf->geo.near_copies;
                if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
                        mddev->queue->backing_dev_info.ra_pages = 2* stripe;
        }
@@ -3460,7 +3466,7 @@ static int raid10_resize(struct mddev *mddev, sector_t sectors)
        struct r10conf *conf = mddev->private;
        sector_t oldsize, size;
 
-       if (conf->far_copies > 1 && !conf->far_offset)
+       if (conf->geo.far_copies > 1 && !conf->geo.far_offset)
                return -EINVAL;
 
        oldsize = raid10_size(mddev, 0, 0);
index 7c61561..4c4942a 100644 (file)
@@ -14,33 +14,34 @@ struct mirror_info {
 struct r10conf {
        struct mddev            *mddev;
        struct mirror_info      *mirrors;
-       int                     raid_disks;
        spinlock_t              device_lock;
 
        /* geometry */
-       int                     near_copies;  /* number of copies laid out
+       struct geom {
+               int             raid_disks;
+               int             near_copies;  /* number of copies laid out
                                               * raid0 style */
-       int                     far_copies;   /* number of copies laid out
+               int             far_copies;   /* number of copies laid out
                                               * at large strides across drives
                                               */
-       int                     far_offset;   /* far_copies are offset by 1
+               int             far_offset;   /* far_copies are offset by 1
                                               * stripe instead of many
                                               */
-       int                     copies;       /* near_copies * far_copies.
-                                              * must be <= raid_disks
-                                              */
-       sector_t                stride;       /* distance between far copies.
+               sector_t        stride;       /* distance between far copies.
                                               * This is size / far_copies unless
                                               * far_offset, in which case it is
                                               * 1 stripe.
                                               */
+               int             chunk_shift; /* shift from chunks to sectors */
+               sector_t        chunk_mask;
+       } geo;
+       int                     copies;       /* near_copies * far_copies.
+                                              * must be <= raid_disks
+                                              */
 
        sector_t                dev_sectors;  /* temp copy of
                                               * mddev->dev_sectors */
 
-       int                     chunk_shift; /* shift from chunks to sectors */
-       sector_t                chunk_mask;
-
        struct list_head        retry_list;
        /* queue pending writes and submit them on unplug */
        struct bio_list         pending_bio_list;