]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
md/r5cache: handle R5LOG_PAYLOAD_FLUSH in recovery
authorSong Liu <songliubraving@fb.com>
Wed, 8 Mar 2017 01:44:21 +0000 (17:44 -0800)
committerShaohua Li <shli@fb.com>
Thu, 16 Mar 2017 23:55:57 +0000 (16:55 -0700)
This patch adds handling of R5LOG_PAYLOAD_FLUSH in journal recovery.
Next patch will add logic that generate R5LOG_PAYLOAD_FLUSH on flush
finish.

When R5LOG_PAYLOAD_FLUSH is seen in recovery, pending data and parity
will be dropped from recovery. This will reduce the number of stripes
to replay, and thus accelerate the recovery process.

Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Shaohua Li <shli@fb.com>
drivers/md/raid5-cache.c

index 5f82dabdda6f54b698dcdd752d1b353f9917be4d..f85ab2744bfbd354af16ecdb830da2ff7228ee64 100644 (file)
@@ -1959,6 +1959,7 @@ r5l_recovery_verify_data_checksum_for_mb(struct r5l_log *log,
        sector_t log_offset = r5l_ring_add(log, ctx->pos, BLOCK_SECTORS);
        struct page *page;
        struct r5l_payload_data_parity *payload;
+       struct r5l_payload_flush *payload_flush;
 
        page = alloc_page(GFP_KERNEL);
        if (!page)
@@ -1966,6 +1967,7 @@ r5l_recovery_verify_data_checksum_for_mb(struct r5l_log *log,
 
        while (mb_offset < le32_to_cpu(mb->meta_size)) {
                payload = (void *)mb + mb_offset;
+               payload_flush = (void *)mb + mb_offset;
 
                if (payload->header.type == R5LOG_PAYLOAD_DATA) {
                        if (r5l_recovery_verify_data_checksum(
@@ -1984,15 +1986,23 @@ r5l_recovery_verify_data_checksum_for_mb(struct r5l_log *log,
                                                 BLOCK_SECTORS),
                                    payload->checksum[1]) < 0)
                                goto mismatch;
-               } else /* not R5LOG_PAYLOAD_DATA or R5LOG_PAYLOAD_PARITY */
+               } else if (payload->header.type == R5LOG_PAYLOAD_FLUSH) {
+                       /* nothing to do for R5LOG_PAYLOAD_FLUSH here */
+               } else /* not R5LOG_PAYLOAD_DATA/PARITY/FLUSH */
                        goto mismatch;
 
-               log_offset = r5l_ring_add(log, log_offset,
-                                         le32_to_cpu(payload->size));
+               if (payload->header.type == R5LOG_PAYLOAD_FLUSH) {
+                       mb_offset += sizeof(struct r5l_payload_flush) +
+                               le32_to_cpu(payload_flush->size);
+               } else {
+                       /* DATA or PARITY payload */
+                       log_offset = r5l_ring_add(log, log_offset,
+                                                 le32_to_cpu(payload->size));
+                       mb_offset += sizeof(struct r5l_payload_data_parity) +
+                               sizeof(__le32) *
+                               (le32_to_cpu(payload->size) >> (PAGE_SHIFT - 9));
+               }
 
-               mb_offset += sizeof(struct r5l_payload_data_parity) +
-                       sizeof(__le32) *
-                       (le32_to_cpu(payload->size) >> (PAGE_SHIFT - 9));
        }
 
        put_page(page);
@@ -2020,6 +2030,7 @@ r5c_recovery_analyze_meta_block(struct r5l_log *log,
        struct r5conf *conf = mddev->private;
        struct r5l_meta_block *mb;
        struct r5l_payload_data_parity *payload;
+       struct r5l_payload_flush *payload_flush;
        int mb_offset;
        sector_t log_offset;
        sector_t stripe_sect;
@@ -2045,6 +2056,30 @@ r5c_recovery_analyze_meta_block(struct r5l_log *log,
                int dd;
 
                payload = (void *)mb + mb_offset;
+               payload_flush = (void *)mb + mb_offset;
+
+               if (payload->header.type == R5LOG_PAYLOAD_FLUSH) {
+                       int i, count;
+
+                       count = le32_to_cpu(payload_flush->size) / sizeof(__le64);
+                       for (i = 0; i < count; ++i) {
+                               stripe_sect = le64_to_cpu(payload_flush->flush_stripes[i]);
+                               sh = r5c_recovery_lookup_stripe(cached_stripe_list,
+                                                               stripe_sect);
+                               if (sh) {
+                                       WARN_ON(test_bit(STRIPE_R5C_CACHING, &sh->state));
+                                       r5l_recovery_reset_stripe(sh);
+                                       list_del_init(&sh->lru);
+                                       raid5_release_stripe(sh);
+                               }
+                       }
+
+                       mb_offset += sizeof(struct r5l_payload_flush) +
+                               le32_to_cpu(payload_flush->size);
+                       continue;
+               }
+
+               /* DATA or PARITY payload */
                stripe_sect = (payload->header.type == R5LOG_PAYLOAD_DATA) ?
                        raid5_compute_sector(
                                conf, le64_to_cpu(payload->location), 0, &dd,