bcache: store disk name in struct cache and struct cached_dev
authorColy Li <colyli@suse.de>
Thu, 3 May 2018 10:51:32 +0000 (18:51 +0800)
committerJens Axboe <axboe@kernel.dk>
Thu, 3 May 2018 14:35:08 +0000 (08:35 -0600)
Current code uses bdevname() or bio_devname() to reference gendisk
disk name when bcache needs to display the disk names in kernel message.
It was safe before bcache device failure handling patch set merged in,
because when devices are failed, there was deadlock to prevent bcache
printing error messages with gendisk disk name. But after the failure
handling patch set merged, the deadlock is fixed, so it is possible
that the gendisk structure bdev->hd_disk is released when bdevname() is
called to reference bdev->bd_disk->disk_name[]. This is why I receive
bug report of NULL pointers deference panic.

This patch stores gendisk disk name in a buffer inside struct cache and
struct cached_dev, then print out the offline device name won't reference
bdev->hd_disk anymore. And this patch also avoids extra function calls
of bdevname() and bio_devnmae().

Changelog:
v3, add Reviewed-by from Hannes.
v2, call bdevname() earlier in register_bdev()
v1, first version with segguestion from Junhui Tang.

Fixes: c7b7bd07404c5 ("bcache: add io_disable to struct cached_dev")
Fixes: 5138ac6748e38 ("bcache: fix misleading error message in bch_count_io_errors()")
Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
drivers/md/bcache/bcache.h
drivers/md/bcache/debug.c
drivers/md/bcache/io.c
drivers/md/bcache/request.c
drivers/md/bcache/super.c

index d338b70..3a0cfb2 100644 (file)
@@ -392,6 +392,8 @@ struct cached_dev {
 #define DEFAULT_CACHED_DEV_ERROR_LIMIT 64
        atomic_t                io_errors;
        unsigned                error_limit;
+
+       char                    backing_dev_name[BDEVNAME_SIZE];
 };
 
 enum alloc_reserve {
@@ -464,6 +466,8 @@ struct cache {
        atomic_long_t           meta_sectors_written;
        atomic_long_t           btree_sectors_written;
        atomic_long_t           sectors_written;
+
+       char                    cache_dev_name[BDEVNAME_SIZE];
 };
 
 struct gc_stat {
index 028f7b3..4e63c6f 100644 (file)
@@ -106,7 +106,6 @@ void bch_btree_verify(struct btree *b)
 
 void bch_data_verify(struct cached_dev *dc, struct bio *bio)
 {
-       char name[BDEVNAME_SIZE];
        struct bio *check;
        struct bio_vec bv, cbv;
        struct bvec_iter iter, citer = { 0 };
@@ -134,7 +133,7 @@ void bch_data_verify(struct cached_dev *dc, struct bio *bio)
                                        bv.bv_len),
                                 dc->disk.c,
                                 "verify failed at dev %s sector %llu",
-                                bdevname(dc->bdev, name),
+                                dc->backing_dev_name,
                                 (uint64_t) bio->bi_iter.bi_sector);
 
                kunmap_atomic(p1);
index 7fac97a..2ddf851 100644 (file)
@@ -52,7 +52,6 @@ void bch_submit_bbio(struct bio *bio, struct cache_set *c,
 /* IO errors */
 void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio)
 {
-       char buf[BDEVNAME_SIZE];
        unsigned errors;
 
        WARN_ONCE(!dc, "NULL pointer of struct cached_dev");
@@ -60,7 +59,7 @@ void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio)
        errors = atomic_add_return(1, &dc->io_errors);
        if (errors < dc->error_limit)
                pr_err("%s: IO error on backing device, unrecoverable",
-                       bio_devname(bio, buf));
+                       dc->backing_dev_name);
        else
                bch_cached_dev_error(dc);
 }
@@ -105,19 +104,18 @@ void bch_count_io_errors(struct cache *ca,
        }
 
        if (error) {
-               char buf[BDEVNAME_SIZE];
                unsigned errors = atomic_add_return(1 << IO_ERROR_SHIFT,
                                                    &ca->io_errors);
                errors >>= IO_ERROR_SHIFT;
 
                if (errors < ca->set->error_limit)
                        pr_err("%s: IO error on %s%s",
-                              bdevname(ca->bdev, buf), m,
+                              ca->cache_dev_name, m,
                               is_read ? ", recovering." : ".");
                else
                        bch_cache_set_error(ca->set,
                                            "%s: too many IO errors %s",
-                                           bdevname(ca->bdev, buf), m);
+                                           ca->cache_dev_name, m);
        }
 }
 
index a65e336..8e3e865 100644 (file)
@@ -649,11 +649,8 @@ static void backing_request_endio(struct bio *bio)
                 */
                if (unlikely(s->iop.writeback &&
                             bio->bi_opf & REQ_PREFLUSH)) {
-                       char buf[BDEVNAME_SIZE];
-
-                       bio_devname(bio, buf);
                        pr_err("Can't flush %s: returned bi_status %i",
-                               buf, bio->bi_status);
+                               dc->backing_dev_name, bio->bi_status);
                } else {
                        /* set to orig_bio->bi_status in bio_complete() */
                        s->iop.status = bio->bi_status;
index d90d9e5..8196b19 100644 (file)
@@ -936,7 +936,6 @@ static void cancel_writeback_rate_update_dwork(struct cached_dev *dc)
 static void cached_dev_detach_finish(struct work_struct *w)
 {
        struct cached_dev *dc = container_of(w, struct cached_dev, detach);
-       char buf[BDEVNAME_SIZE];
        struct closure cl;
        closure_init_stack(&cl);
 
@@ -967,7 +966,7 @@ static void cached_dev_detach_finish(struct work_struct *w)
 
        mutex_unlock(&bch_register_lock);
 
-       pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
+       pr_info("Caching disabled for %s", dc->backing_dev_name);
 
        /* Drop ref we took in cached_dev_detach() */
        closure_put(&dc->disk.cl);
@@ -999,29 +998,28 @@ int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
 {
        uint32_t rtime = cpu_to_le32(get_seconds());
        struct uuid_entry *u;
-       char buf[BDEVNAME_SIZE];
        struct cached_dev *exist_dc, *t;
 
-       bdevname(dc->bdev, buf);
-
        if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
            (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
                return -ENOENT;
 
        if (dc->disk.c) {
-               pr_err("Can't attach %s: already attached", buf);
+               pr_err("Can't attach %s: already attached",
+                      dc->backing_dev_name);
                return -EINVAL;
        }
 
        if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
-               pr_err("Can't attach %s: shutting down", buf);
+               pr_err("Can't attach %s: shutting down",
+                      dc->backing_dev_name);
                return -EINVAL;
        }
 
        if (dc->sb.block_size < c->sb.block_size) {
                /* Will die */
                pr_err("Couldn't attach %s: block size less than set's block size",
-                      buf);
+                      dc->backing_dev_name);
                return -EINVAL;
        }
 
@@ -1029,7 +1027,7 @@ int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
        list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
                if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
                        pr_err("Tried to attach %s but duplicate UUID already attached",
-                               buf);
+                               dc->backing_dev_name);
 
                        return -EINVAL;
                }
@@ -1047,13 +1045,15 @@ int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
 
        if (!u) {
                if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
-                       pr_err("Couldn't find uuid for %s in set", buf);
+                       pr_err("Couldn't find uuid for %s in set",
+                              dc->backing_dev_name);
                        return -ENOENT;
                }
 
                u = uuid_find_empty(c);
                if (!u) {
-                       pr_err("Not caching %s, no room for UUID", buf);
+                       pr_err("Not caching %s, no room for UUID",
+                              dc->backing_dev_name);
                        return -EINVAL;
                }
        }
@@ -1112,7 +1112,8 @@ int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
        up_write(&dc->writeback_lock);
 
        pr_info("Caching %s as %s on set %pU",
-               bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
+               dc->backing_dev_name,
+               dc->disk.disk->disk_name,
                dc->disk.c->sb.set_uuid);
        return 0;
 }
@@ -1225,10 +1226,10 @@ static void register_bdev(struct cache_sb *sb, struct page *sb_page,
                                 struct block_device *bdev,
                                 struct cached_dev *dc)
 {
-       char name[BDEVNAME_SIZE];
        const char *err = "cannot allocate memory";
        struct cache_set *c;
 
+       bdevname(bdev, dc->backing_dev_name);
        memcpy(&dc->sb, sb, sizeof(struct cache_sb));
        dc->bdev = bdev;
        dc->bdev->bd_holder = dc;
@@ -1237,6 +1238,7 @@ static void register_bdev(struct cache_sb *sb, struct page *sb_page,
        bio_first_bvec_all(&dc->sb_bio)->bv_page = sb_page;
        get_page(sb_page);
 
+
        if (cached_dev_init(dc, sb->block_size << 9))
                goto err;
 
@@ -1247,7 +1249,7 @@ static void register_bdev(struct cache_sb *sb, struct page *sb_page,
        if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
                goto err;
 
-       pr_info("registered backing device %s", bdevname(bdev, name));
+       pr_info("registered backing device %s", dc->backing_dev_name);
 
        list_add(&dc->list, &uncached_devices);
        list_for_each_entry(c, &bch_cache_sets, list)
@@ -1259,7 +1261,7 @@ static void register_bdev(struct cache_sb *sb, struct page *sb_page,
 
        return;
 err:
-       pr_notice("error %s: %s", bdevname(bdev, name), err);
+       pr_notice("error %s: %s", dc->backing_dev_name, err);
        bcache_device_stop(&dc->disk);
 }
 
@@ -1367,8 +1369,6 @@ int bch_flash_dev_create(struct cache_set *c, uint64_t size)
 
 bool bch_cached_dev_error(struct cached_dev *dc)
 {
-       char name[BDEVNAME_SIZE];
-
        if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
                return false;
 
@@ -1377,7 +1377,7 @@ bool bch_cached_dev_error(struct cached_dev *dc)
        smp_mb();
 
        pr_err("stop %s: too many IO errors on backing device %s\n",
-               dc->disk.disk->disk_name, bdevname(dc->bdev, name));
+               dc->disk.disk->disk_name, dc->backing_dev_name);
 
        bcache_device_stop(&dc->disk);
        return true;
@@ -2003,12 +2003,10 @@ static int cache_alloc(struct cache *ca)
 static int register_cache(struct cache_sb *sb, struct page *sb_page,
                                struct block_device *bdev, struct cache *ca)
 {
-       char name[BDEVNAME_SIZE];
        const char *err = NULL; /* must be set for any error case */
        int ret = 0;
 
-       bdevname(bdev, name);
-
+       bdevname(bdev, ca->cache_dev_name);
        memcpy(&ca->sb, sb, sizeof(struct cache_sb));
        ca->bdev = bdev;
        ca->bdev->bd_holder = ca;
@@ -2045,14 +2043,14 @@ static int register_cache(struct cache_sb *sb, struct page *sb_page,
                goto out;
        }
 
-       pr_info("registered cache device %s", name);
+       pr_info("registered cache device %s", ca->cache_dev_name);
 
 out:
        kobject_put(&ca->kobj);
 
 err:
        if (err)
-               pr_notice("error %s: %s", name, err);
+               pr_notice("error %s: %s", ca->cache_dev_name, err);
 
        return ret;
 }