summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--drivers/md/bitmap.c6
-rw-r--r--drivers/md/md.c12
-rw-r--r--drivers/md/raid5.c158
-rw-r--r--drivers/md/raid5.h4
4 files changed, 138 insertions, 42 deletions
diff --git a/drivers/md/bitmap.c b/drivers/md/bitmap.c
index 9a8e66ae04f5..67f8b31e2054 100644
--- a/drivers/md/bitmap.c
+++ b/drivers/md/bitmap.c
@@ -669,17 +669,13 @@ static inline unsigned long file_page_offset(struct bitmap_storage *store,
669/* 669/*
670 * return a pointer to the page in the filemap that contains the given bit 670 * return a pointer to the page in the filemap that contains the given bit
671 * 671 *
672 * this lookup is complicated by the fact that the bitmap sb might be exactly
673 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
674 * 0 or page 1
675 */ 672 */
676static inline struct page *filemap_get_page(struct bitmap_storage *store, 673static inline struct page *filemap_get_page(struct bitmap_storage *store,
677 unsigned long chunk) 674 unsigned long chunk)
678{ 675{
679 if (file_page_index(store, chunk) >= store->file_pages) 676 if (file_page_index(store, chunk) >= store->file_pages)
680 return NULL; 677 return NULL;
681 return store->filemap[file_page_index(store, chunk) 678 return store->filemap[file_page_index(store, chunk)];
682 - file_page_index(store, 0)];
683} 679}
684 680
685static int bitmap_storage_alloc(struct bitmap_storage *store, 681static int bitmap_storage_alloc(struct bitmap_storage *store,
diff --git a/drivers/md/md.c b/drivers/md/md.c
index 2382cfc9bb3f..34846856dbc6 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -3448,6 +3448,8 @@ level_store(struct mddev *mddev, const char *buf, size_t len)
3448 mddev->level = LEVEL_NONE; 3448 mddev->level = LEVEL_NONE;
3449 return rv; 3449 return rv;
3450 } 3450 }
3451 if (mddev->ro)
3452 return -EROFS;
3451 3453
3452 /* request to change the personality. Need to ensure: 3454 /* request to change the personality. Need to ensure:
3453 * - array is not engaged in resync/recovery/reshape 3455 * - array is not engaged in resync/recovery/reshape
@@ -3634,6 +3636,8 @@ layout_store(struct mddev *mddev, const char *buf, size_t len)
3634 int err; 3636 int err;
3635 if (mddev->pers->check_reshape == NULL) 3637 if (mddev->pers->check_reshape == NULL)
3636 return -EBUSY; 3638 return -EBUSY;
3639 if (mddev->ro)
3640 return -EROFS;
3637 mddev->new_layout = n; 3641 mddev->new_layout = n;
3638 err = mddev->pers->check_reshape(mddev); 3642 err = mddev->pers->check_reshape(mddev);
3639 if (err) { 3643 if (err) {
@@ -3723,6 +3727,8 @@ chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3723 int err; 3727 int err;
3724 if (mddev->pers->check_reshape == NULL) 3728 if (mddev->pers->check_reshape == NULL)
3725 return -EBUSY; 3729 return -EBUSY;
3730 if (mddev->ro)
3731 return -EROFS;
3726 mddev->new_chunk_sectors = n >> 9; 3732 mddev->new_chunk_sectors = n >> 9;
3727 err = mddev->pers->check_reshape(mddev); 3733 err = mddev->pers->check_reshape(mddev);
3728 if (err) { 3734 if (err) {
@@ -6135,6 +6141,8 @@ static int update_size(struct mddev *mddev, sector_t num_sectors)
6135 */ 6141 */
6136 if (mddev->sync_thread) 6142 if (mddev->sync_thread)
6137 return -EBUSY; 6143 return -EBUSY;
6144 if (mddev->ro)
6145 return -EROFS;
6138 6146
6139 rdev_for_each(rdev, mddev) { 6147 rdev_for_each(rdev, mddev) {
6140 sector_t avail = rdev->sectors; 6148 sector_t avail = rdev->sectors;
@@ -6157,6 +6165,8 @@ static int update_raid_disks(struct mddev *mddev, int raid_disks)
6157 /* change the number of raid disks */ 6165 /* change the number of raid disks */
6158 if (mddev->pers->check_reshape == NULL) 6166 if (mddev->pers->check_reshape == NULL)
6159 return -EINVAL; 6167 return -EINVAL;
6168 if (mddev->ro)
6169 return -EROFS;
6160 if (raid_disks <= 0 || 6170 if (raid_disks <= 0 ||
6161 (mddev->max_disks && raid_disks >= mddev->max_disks)) 6171 (mddev->max_disks && raid_disks >= mddev->max_disks))
6162 return -EINVAL; 6172 return -EINVAL;
@@ -8333,7 +8343,7 @@ static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8333 if (a < s) { 8343 if (a < s) {
8334 /* we need to split this range */ 8344 /* we need to split this range */
8335 if (bb->count >= MD_MAX_BADBLOCKS) { 8345 if (bb->count >= MD_MAX_BADBLOCKS) {
8336 rv = 0; 8346 rv = -ENOSPC;
8337 goto out; 8347 goto out;
8338 } 8348 }
8339 memmove(p+lo+1, p+lo, (bb->count - lo) * 8); 8349 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 2afef4ec9312..6234b2e84587 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -292,9 +292,12 @@ static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
292 BUG_ON(atomic_read(&conf->active_stripes)==0); 292 BUG_ON(atomic_read(&conf->active_stripes)==0);
293 if (test_bit(STRIPE_HANDLE, &sh->state)) { 293 if (test_bit(STRIPE_HANDLE, &sh->state)) {
294 if (test_bit(STRIPE_DELAYED, &sh->state) && 294 if (test_bit(STRIPE_DELAYED, &sh->state) &&
295 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) 295 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
296 list_add_tail(&sh->lru, &conf->delayed_list); 296 list_add_tail(&sh->lru, &conf->delayed_list);
297 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && 297 if (atomic_read(&conf->preread_active_stripes)
298 < IO_THRESHOLD)
299 md_wakeup_thread(conf->mddev->thread);
300 } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
298 sh->bm_seq - conf->seq_write > 0) 301 sh->bm_seq - conf->seq_write > 0)
299 list_add_tail(&sh->lru, &conf->bitmap_list); 302 list_add_tail(&sh->lru, &conf->bitmap_list);
300 else { 303 else {
@@ -413,6 +416,11 @@ static void release_stripe(struct stripe_head *sh)
413 int hash; 416 int hash;
414 bool wakeup; 417 bool wakeup;
415 418
419 /* Avoid release_list until the last reference.
420 */
421 if (atomic_add_unless(&sh->count, -1, 1))
422 return;
423
416 if (unlikely(!conf->mddev->thread) || 424 if (unlikely(!conf->mddev->thread) ||
417 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) 425 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
418 goto slow_path; 426 goto slow_path;
@@ -479,6 +487,7 @@ static void shrink_buffers(struct stripe_head *sh)
479 int num = sh->raid_conf->pool_size; 487 int num = sh->raid_conf->pool_size;
480 488
481 for (i = 0; i < num ; i++) { 489 for (i = 0; i < num ; i++) {
490 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
482 p = sh->dev[i].page; 491 p = sh->dev[i].page;
483 if (!p) 492 if (!p)
484 continue; 493 continue;
@@ -499,6 +508,7 @@ static int grow_buffers(struct stripe_head *sh)
499 return 1; 508 return 1;
500 } 509 }
501 sh->dev[i].page = page; 510 sh->dev[i].page = page;
511 sh->dev[i].orig_page = page;
502 } 512 }
503 return 0; 513 return 0;
504} 514}
@@ -855,6 +865,9 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
855 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) 865 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
856 bi->bi_rw |= REQ_NOMERGE; 866 bi->bi_rw |= REQ_NOMERGE;
857 867
868 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
869 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
870 sh->dev[i].vec.bv_page = sh->dev[i].page;
858 bi->bi_vcnt = 1; 871 bi->bi_vcnt = 1;
859 bi->bi_io_vec[0].bv_len = STRIPE_SIZE; 872 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
860 bi->bi_io_vec[0].bv_offset = 0; 873 bi->bi_io_vec[0].bv_offset = 0;
@@ -899,6 +912,9 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
899 else 912 else
900 rbi->bi_iter.bi_sector = (sh->sector 913 rbi->bi_iter.bi_sector = (sh->sector
901 + rrdev->data_offset); 914 + rrdev->data_offset);
915 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
916 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
917 sh->dev[i].rvec.bv_page = sh->dev[i].page;
902 rbi->bi_vcnt = 1; 918 rbi->bi_vcnt = 1;
903 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; 919 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
904 rbi->bi_io_vec[0].bv_offset = 0; 920 rbi->bi_io_vec[0].bv_offset = 0;
@@ -927,8 +943,9 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
927} 943}
928 944
929static struct dma_async_tx_descriptor * 945static struct dma_async_tx_descriptor *
930async_copy_data(int frombio, struct bio *bio, struct page *page, 946async_copy_data(int frombio, struct bio *bio, struct page **page,
931 sector_t sector, struct dma_async_tx_descriptor *tx) 947 sector_t sector, struct dma_async_tx_descriptor *tx,
948 struct stripe_head *sh)
932{ 949{
933 struct bio_vec bvl; 950 struct bio_vec bvl;
934 struct bvec_iter iter; 951 struct bvec_iter iter;
@@ -965,11 +982,16 @@ async_copy_data(int frombio, struct bio *bio, struct page *page,
965 if (clen > 0) { 982 if (clen > 0) {
966 b_offset += bvl.bv_offset; 983 b_offset += bvl.bv_offset;
967 bio_page = bvl.bv_page; 984 bio_page = bvl.bv_page;
968 if (frombio) 985 if (frombio) {
969 tx = async_memcpy(page, bio_page, page_offset, 986 if (sh->raid_conf->skip_copy &&
987 b_offset == 0 && page_offset == 0 &&
988 clen == STRIPE_SIZE)
989 *page = bio_page;
990 else
991 tx = async_memcpy(*page, bio_page, page_offset,
970 b_offset, clen, &submit); 992 b_offset, clen, &submit);
971 else 993 } else
972 tx = async_memcpy(bio_page, page, b_offset, 994 tx = async_memcpy(bio_page, *page, b_offset,
973 page_offset, clen, &submit); 995 page_offset, clen, &submit);
974 } 996 }
975 /* chain the operations */ 997 /* chain the operations */
@@ -1045,8 +1067,8 @@ static void ops_run_biofill(struct stripe_head *sh)
1045 spin_unlock_irq(&sh->stripe_lock); 1067 spin_unlock_irq(&sh->stripe_lock);
1046 while (rbi && rbi->bi_iter.bi_sector < 1068 while (rbi && rbi->bi_iter.bi_sector <
1047 dev->sector + STRIPE_SECTORS) { 1069 dev->sector + STRIPE_SECTORS) {
1048 tx = async_copy_data(0, rbi, dev->page, 1070 tx = async_copy_data(0, rbi, &dev->page,
1049 dev->sector, tx); 1071 dev->sector, tx, sh);
1050 rbi = r5_next_bio(rbi, dev->sector); 1072 rbi = r5_next_bio(rbi, dev->sector);
1051 } 1073 }
1052 } 1074 }
@@ -1384,6 +1406,7 @@ ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
1384 BUG_ON(dev->written); 1406 BUG_ON(dev->written);
1385 wbi = dev->written = chosen; 1407 wbi = dev->written = chosen;
1386 spin_unlock_irq(&sh->stripe_lock); 1408 spin_unlock_irq(&sh->stripe_lock);
1409 WARN_ON(dev->page != dev->orig_page);
1387 1410
1388 while (wbi && wbi->bi_iter.bi_sector < 1411 while (wbi && wbi->bi_iter.bi_sector <
1389 dev->sector + STRIPE_SECTORS) { 1412 dev->sector + STRIPE_SECTORS) {
@@ -1393,9 +1416,15 @@ ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
1393 set_bit(R5_SyncIO, &dev->flags); 1416 set_bit(R5_SyncIO, &dev->flags);
1394 if (wbi->bi_rw & REQ_DISCARD) 1417 if (wbi->bi_rw & REQ_DISCARD)
1395 set_bit(R5_Discard, &dev->flags); 1418 set_bit(R5_Discard, &dev->flags);
1396 else 1419 else {
1397 tx = async_copy_data(1, wbi, dev->page, 1420 tx = async_copy_data(1, wbi, &dev->page,
1398 dev->sector, tx); 1421 dev->sector, tx, sh);
1422 if (dev->page != dev->orig_page) {
1423 set_bit(R5_SkipCopy, &dev->flags);
1424 clear_bit(R5_UPTODATE, &dev->flags);
1425 clear_bit(R5_OVERWRITE, &dev->flags);
1426 }
1427 }
1399 wbi = r5_next_bio(wbi, dev->sector); 1428 wbi = r5_next_bio(wbi, dev->sector);
1400 } 1429 }
1401 } 1430 }
@@ -1426,7 +1455,7 @@ static void ops_complete_reconstruct(void *stripe_head_ref)
1426 struct r5dev *dev = &sh->dev[i]; 1455 struct r5dev *dev = &sh->dev[i];
1427 1456
1428 if (dev->written || i == pd_idx || i == qd_idx) { 1457 if (dev->written || i == pd_idx || i == qd_idx) {
1429 if (!discard) 1458 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
1430 set_bit(R5_UPTODATE, &dev->flags); 1459 set_bit(R5_UPTODATE, &dev->flags);
1431 if (fua) 1460 if (fua)
1432 set_bit(R5_WantFUA, &dev->flags); 1461 set_bit(R5_WantFUA, &dev->flags);
@@ -1839,8 +1868,10 @@ static int resize_stripes(struct r5conf *conf, int newsize)
1839 osh = get_free_stripe(conf, hash); 1868 osh = get_free_stripe(conf, hash);
1840 unlock_device_hash_lock(conf, hash); 1869 unlock_device_hash_lock(conf, hash);
1841 atomic_set(&nsh->count, 1); 1870 atomic_set(&nsh->count, 1);
1842 for(i=0; i<conf->pool_size; i++) 1871 for(i=0; i<conf->pool_size; i++) {
1843 nsh->dev[i].page = osh->dev[i].page; 1872 nsh->dev[i].page = osh->dev[i].page;
1873 nsh->dev[i].orig_page = osh->dev[i].page;
1874 }
1844 for( ; i<newsize; i++) 1875 for( ; i<newsize; i++)
1845 nsh->dev[i].page = NULL; 1876 nsh->dev[i].page = NULL;
1846 nsh->hash_lock_index = hash; 1877 nsh->hash_lock_index = hash;
@@ -1896,6 +1927,7 @@ static int resize_stripes(struct r5conf *conf, int newsize)
1896 if (nsh->dev[i].page == NULL) { 1927 if (nsh->dev[i].page == NULL) {
1897 struct page *p = alloc_page(GFP_NOIO); 1928 struct page *p = alloc_page(GFP_NOIO);
1898 nsh->dev[i].page = p; 1929 nsh->dev[i].page = p;
1930 nsh->dev[i].orig_page = p;
1899 if (!p) 1931 if (!p)
1900 err = -ENOMEM; 1932 err = -ENOMEM;
1901 } 1933 }
@@ -2133,24 +2165,20 @@ static void raid5_end_write_request(struct bio *bi, int error)
2133} 2165}
2134 2166
2135static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); 2167static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
2136 2168
2137static void raid5_build_block(struct stripe_head *sh, int i, int previous) 2169static void raid5_build_block(struct stripe_head *sh, int i, int previous)
2138{ 2170{
2139 struct r5dev *dev = &sh->dev[i]; 2171 struct r5dev *dev = &sh->dev[i];
2140 2172
2141 bio_init(&dev->req); 2173 bio_init(&dev->req);
2142 dev->req.bi_io_vec = &dev->vec; 2174 dev->req.bi_io_vec = &dev->vec;
2143 dev->req.bi_vcnt++; 2175 dev->req.bi_max_vecs = 1;
2144 dev->req.bi_max_vecs++;
2145 dev->req.bi_private = sh; 2176 dev->req.bi_private = sh;
2146 dev->vec.bv_page = dev->page;
2147 2177
2148 bio_init(&dev->rreq); 2178 bio_init(&dev->rreq);
2149 dev->rreq.bi_io_vec = &dev->rvec; 2179 dev->rreq.bi_io_vec = &dev->rvec;
2150 dev->rreq.bi_vcnt++; 2180 dev->rreq.bi_max_vecs = 1;
2151 dev->rreq.bi_max_vecs++;
2152 dev->rreq.bi_private = sh; 2181 dev->rreq.bi_private = sh;
2153 dev->rvec.bv_page = dev->page;
2154 2182
2155 dev->flags = 0; 2183 dev->flags = 0;
2156 dev->sector = compute_blocknr(sh, i, previous); 2184 dev->sector = compute_blocknr(sh, i, previous);
@@ -2750,6 +2778,11 @@ handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
2750 /* and fail all 'written' */ 2778 /* and fail all 'written' */
2751 bi = sh->dev[i].written; 2779 bi = sh->dev[i].written;
2752 sh->dev[i].written = NULL; 2780 sh->dev[i].written = NULL;
2781 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
2782 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
2783 sh->dev[i].page = sh->dev[i].orig_page;
2784 }
2785
2753 if (bi) bitmap_end = 1; 2786 if (bi) bitmap_end = 1;
2754 while (bi && bi->bi_iter.bi_sector < 2787 while (bi && bi->bi_iter.bi_sector <
2755 sh->dev[i].sector + STRIPE_SECTORS) { 2788 sh->dev[i].sector + STRIPE_SECTORS) {
@@ -2886,8 +2919,11 @@ static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2886 (s->failed >= 1 && fdev[0]->toread) || 2919 (s->failed >= 1 && fdev[0]->toread) ||
2887 (s->failed >= 2 && fdev[1]->toread) || 2920 (s->failed >= 2 && fdev[1]->toread) ||
2888 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite && 2921 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
2922 (!test_bit(R5_Insync, &dev->flags) || test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) &&
2889 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) || 2923 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
2890 (sh->raid_conf->level == 6 && s->failed && s->to_write))) { 2924 (sh->raid_conf->level == 6 && s->failed && s->to_write &&
2925 s->to_write < sh->raid_conf->raid_disks - 2 &&
2926 (!test_bit(R5_Insync, &dev->flags) || test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))))) {
2891 /* we would like to get this block, possibly by computing it, 2927 /* we would like to get this block, possibly by computing it,
2892 * otherwise read it if the backing disk is insync 2928 * otherwise read it if the backing disk is insync
2893 */ 2929 */
@@ -2991,12 +3027,17 @@ static void handle_stripe_clean_event(struct r5conf *conf,
2991 dev = &sh->dev[i]; 3027 dev = &sh->dev[i];
2992 if (!test_bit(R5_LOCKED, &dev->flags) && 3028 if (!test_bit(R5_LOCKED, &dev->flags) &&
2993 (test_bit(R5_UPTODATE, &dev->flags) || 3029 (test_bit(R5_UPTODATE, &dev->flags) ||
2994 test_bit(R5_Discard, &dev->flags))) { 3030 test_bit(R5_Discard, &dev->flags) ||
3031 test_bit(R5_SkipCopy, &dev->flags))) {
2995 /* We can return any write requests */ 3032 /* We can return any write requests */
2996 struct bio *wbi, *wbi2; 3033 struct bio *wbi, *wbi2;
2997 pr_debug("Return write for disc %d\n", i); 3034 pr_debug("Return write for disc %d\n", i);
2998 if (test_and_clear_bit(R5_Discard, &dev->flags)) 3035 if (test_and_clear_bit(R5_Discard, &dev->flags))
2999 clear_bit(R5_UPTODATE, &dev->flags); 3036 clear_bit(R5_UPTODATE, &dev->flags);
3037 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3038 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
3039 dev->page = dev->orig_page;
3040 }
3000 wbi = dev->written; 3041 wbi = dev->written;
3001 dev->written = NULL; 3042 dev->written = NULL;
3002 while (wbi && wbi->bi_iter.bi_sector < 3043 while (wbi && wbi->bi_iter.bi_sector <
@@ -3015,6 +3056,8 @@ static void handle_stripe_clean_event(struct r5conf *conf,
3015 0); 3056 0);
3016 } else if (test_bit(R5_Discard, &dev->flags)) 3057 } else if (test_bit(R5_Discard, &dev->flags))
3017 discard_pending = 1; 3058 discard_pending = 1;
3059 WARN_ON(test_bit(R5_SkipCopy, &dev->flags));
3060 WARN_ON(dev->page != dev->orig_page);
3018 } 3061 }
3019 if (!discard_pending && 3062 if (!discard_pending &&
3020 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { 3063 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
@@ -3086,7 +3129,8 @@ static void handle_stripe_dirtying(struct r5conf *conf,
3086 !test_bit(R5_LOCKED, &dev->flags) && 3129 !test_bit(R5_LOCKED, &dev->flags) &&
3087 !(test_bit(R5_UPTODATE, &dev->flags) || 3130 !(test_bit(R5_UPTODATE, &dev->flags) ||
3088 test_bit(R5_Wantcompute, &dev->flags))) { 3131 test_bit(R5_Wantcompute, &dev->flags))) {
3089 if (test_bit(R5_Insync, &dev->flags)) rcw++; 3132 if (test_bit(R5_Insync, &dev->flags))
3133 rcw++;
3090 else 3134 else
3091 rcw += 2*disks; 3135 rcw += 2*disks;
3092 } 3136 }
@@ -3107,10 +3151,10 @@ static void handle_stripe_dirtying(struct r5conf *conf,
3107 !(test_bit(R5_UPTODATE, &dev->flags) || 3151 !(test_bit(R5_UPTODATE, &dev->flags) ||
3108 test_bit(R5_Wantcompute, &dev->flags)) && 3152 test_bit(R5_Wantcompute, &dev->flags)) &&
3109 test_bit(R5_Insync, &dev->flags)) { 3153 test_bit(R5_Insync, &dev->flags)) {
3110 if ( 3154 if (test_bit(STRIPE_PREREAD_ACTIVE,
3111 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { 3155 &sh->state)) {
3112 pr_debug("Read_old block " 3156 pr_debug("Read_old block %d for r-m-w\n",
3113 "%d for r-m-w\n", i); 3157 i);
3114 set_bit(R5_LOCKED, &dev->flags); 3158 set_bit(R5_LOCKED, &dev->flags);
3115 set_bit(R5_Wantread, &dev->flags); 3159 set_bit(R5_Wantread, &dev->flags);
3116 s->locked++; 3160 s->locked++;
@@ -3133,10 +3177,9 @@ static void handle_stripe_dirtying(struct r5conf *conf,
3133 !(test_bit(R5_UPTODATE, &dev->flags) || 3177 !(test_bit(R5_UPTODATE, &dev->flags) ||
3134 test_bit(R5_Wantcompute, &dev->flags))) { 3178 test_bit(R5_Wantcompute, &dev->flags))) {
3135 rcw++; 3179 rcw++;
3136 if (!test_bit(R5_Insync, &dev->flags)) 3180 if (test_bit(R5_Insync, &dev->flags) &&
3137 continue; /* it's a failed drive */ 3181 test_bit(STRIPE_PREREAD_ACTIVE,
3138 if ( 3182 &sh->state)) {
3139 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
3140 pr_debug("Read_old block " 3183 pr_debug("Read_old block "
3141 "%d for Reconstruct\n", i); 3184 "%d for Reconstruct\n", i);
3142 set_bit(R5_LOCKED, &dev->flags); 3185 set_bit(R5_LOCKED, &dev->flags);
@@ -5031,8 +5074,8 @@ static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int
5031 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); 5074 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5032 5075
5033 set_bit(STRIPE_SYNC_REQUESTED, &sh->state); 5076 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
5077 set_bit(STRIPE_HANDLE, &sh->state);
5034 5078
5035 handle_stripe(sh);
5036 release_stripe(sh); 5079 release_stripe(sh);
5037 5080
5038 return STRIPE_SECTORS; 5081 return STRIPE_SECTORS;
@@ -5072,7 +5115,7 @@ static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
5072 /* already done this stripe */ 5115 /* already done this stripe */
5073 continue; 5116 continue;
5074 5117
5075 sh = get_active_stripe(conf, sector, 0, 1, 0); 5118 sh = get_active_stripe(conf, sector, 0, 1, 1);
5076 5119
5077 if (!sh) { 5120 if (!sh) {
5078 /* failed to get a stripe - must wait */ 5121 /* failed to get a stripe - must wait */
@@ -5355,6 +5398,50 @@ raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
5355 raid5_store_preread_threshold); 5398 raid5_store_preread_threshold);
5356 5399
5357static ssize_t 5400static ssize_t
5401raid5_show_skip_copy(struct mddev *mddev, char *page)
5402{
5403 struct r5conf *conf = mddev->private;
5404 if (conf)
5405 return sprintf(page, "%d\n", conf->skip_copy);
5406 else
5407 return 0;
5408}
5409
5410static ssize_t
5411raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
5412{
5413 struct r5conf *conf = mddev->private;
5414 unsigned long new;
5415 if (len >= PAGE_SIZE)
5416 return -EINVAL;
5417 if (!conf)
5418 return -ENODEV;
5419
5420 if (kstrtoul(page, 10, &new))
5421 return -EINVAL;
5422 new = !!new;
5423 if (new == conf->skip_copy)
5424 return len;
5425
5426 mddev_suspend(mddev);
5427 conf->skip_copy = new;
5428 if (new)
5429 mddev->queue->backing_dev_info.capabilities |=
5430 BDI_CAP_STABLE_WRITES;
5431 else
5432 mddev->queue->backing_dev_info.capabilities &=
5433 ~BDI_CAP_STABLE_WRITES;
5434 mddev_resume(mddev);
5435 return len;
5436}
5437
5438static struct md_sysfs_entry
5439raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
5440 raid5_show_skip_copy,
5441 raid5_store_skip_copy);
5442
5443
5444static ssize_t
5358stripe_cache_active_show(struct mddev *mddev, char *page) 5445stripe_cache_active_show(struct mddev *mddev, char *page)
5359{ 5446{
5360 struct r5conf *conf = mddev->private; 5447 struct r5conf *conf = mddev->private;
@@ -5439,6 +5526,7 @@ static struct attribute *raid5_attrs[] = {
5439 &raid5_stripecache_active.attr, 5526 &raid5_stripecache_active.attr,
5440 &raid5_preread_bypass_threshold.attr, 5527 &raid5_preread_bypass_threshold.attr,
5441 &raid5_group_thread_cnt.attr, 5528 &raid5_group_thread_cnt.attr,
5529 &raid5_skip_copy.attr,
5442 NULL, 5530 NULL,
5443}; 5531};
5444static struct attribute_group raid5_attrs_group = { 5532static struct attribute_group raid5_attrs_group = {
diff --git a/drivers/md/raid5.h b/drivers/md/raid5.h
index 01ad8ae8f578..bc72cd4be5f8 100644
--- a/drivers/md/raid5.h
+++ b/drivers/md/raid5.h
@@ -232,7 +232,7 @@ struct stripe_head {
232 */ 232 */
233 struct bio req, rreq; 233 struct bio req, rreq;
234 struct bio_vec vec, rvec; 234 struct bio_vec vec, rvec;
235 struct page *page; 235 struct page *page, *orig_page;
236 struct bio *toread, *read, *towrite, *written; 236 struct bio *toread, *read, *towrite, *written;
237 sector_t sector; /* sector of this page */ 237 sector_t sector; /* sector of this page */
238 unsigned long flags; 238 unsigned long flags;
@@ -299,6 +299,7 @@ enum r5dev_flags {
299 * data in, and now is a good time to write it out. 299 * data in, and now is a good time to write it out.
300 */ 300 */
301 R5_Discard, /* Discard the stripe */ 301 R5_Discard, /* Discard the stripe */
302 R5_SkipCopy, /* Don't copy data from bio to stripe cache */
302}; 303};
303 304
304/* 305/*
@@ -436,6 +437,7 @@ struct r5conf {
436 atomic_t pending_full_writes; /* full write backlog */ 437 atomic_t pending_full_writes; /* full write backlog */
437 int bypass_count; /* bypassed prereads */ 438 int bypass_count; /* bypassed prereads */
438 int bypass_threshold; /* preread nice */ 439 int bypass_threshold; /* preread nice */
440 int skip_copy; /* Don't copy data from bio to stripe cache */
439 struct list_head *last_hold; /* detect hold_list promotions */ 441 struct list_head *last_hold; /* detect hold_list promotions */
440 442
441 atomic_t reshape_stripes; /* stripes with pending writes for reshape */ 443 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-30 18:48:27 -0500