drbd: Bitmap IO functions can now return prematurely if the disk breaks
authorPhilipp Reisner <philipp.reisner@linbit.com>
Mon, 27 Jun 2011 13:29:16 +0000 (15:29 +0200)
committerPhilipp Reisner <philipp.reisner@linbit.com>
Wed, 9 May 2012 13:10:33 +0000 (15:10 +0200)
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
drivers/block/drbd/drbd_bitmap.c

index a2c337b38d6e598e1745579abdcdb9d1527f08f6..e5e756dbc4343402d0a2a0a324a7d00b9743c35c 100644 (file)
@@ -886,7 +886,7 @@ void drbd_bm_clear_all(struct drbd_conf *mdev)
 struct bm_aio_ctx {
        struct drbd_conf *mdev;
        atomic_t in_flight;
-       struct completion done;
+       unsigned int done;
        unsigned flags;
 #define BM_AIO_COPY_PAGES      1
        int error;
@@ -897,6 +897,7 @@ static void bm_aio_ctx_destroy(struct kref *kref)
 {
        struct bm_aio_ctx *ctx = container_of(kref, struct bm_aio_ctx, kref);
 
+       put_ldev(ctx->mdev);
        kfree(ctx);
 }
 
@@ -945,7 +946,8 @@ static void bm_async_io_complete(struct bio *bio, int error)
        bio_put(bio);
 
        if (atomic_dec_and_test(&ctx->in_flight)) {
-               complete(&ctx->done);
+               ctx->done = 1;
+               wake_up(&mdev->misc_wait);
                kref_put(&ctx->kref, &bm_aio_ctx_destroy);
        }
 }
@@ -1034,12 +1036,18 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
        *ctx = (struct bm_aio_ctx) {
                .mdev = mdev,
                .in_flight = ATOMIC_INIT(1),
-               .done = COMPLETION_INITIALIZER(ctx->done),
+               .done = 0,
                .flags = lazy_writeout_upper_idx ? BM_AIO_COPY_PAGES : 0,
                .error = 0,
                .kref = { ATOMIC_INIT(2) },
        };
 
+       if (!get_ldev_if_state(mdev, D_ATTACHING)) {  /* put is in bm_aio_ctx_destroy() */
+               dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in bm_rw()\n");
+               kfree(ctx);
+               return -ENODEV;
+       }
+
        if (!ctx->flags)
                WARN_ON(!(BM_LOCKED_MASK & b->bm_flags));
 
@@ -1073,11 +1081,16 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
 
        /*
         * We initialize ctx->in_flight to one to make sure bm_async_io_complete
-        * will not complete() early, and decrement / test it here.  If there
+        * will not set ctx->done early, and decrement / test it here.  If there
         * are still some bios in flight, we need to wait for them here.
+        * If all IO is done already (or nothing had been submitted), there is
+        * no need to wait.  Still, we need to put the kref associated with the
+        * "in_flight reached zero, all done" event.
         */
        if (!atomic_dec_and_test(&ctx->in_flight))
-               wait_for_completion(&ctx->done);
+               wait_until_done_or_disk_failure(mdev, &ctx->done);
+       else
+               kref_put(&ctx->kref, &bm_aio_ctx_destroy);
 
        dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n",
                        rw == WRITE ? "WRITE" : "READ",
@@ -1089,6 +1102,9 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
                err = -EIO; /* ctx->error ? */
        }
 
+       if (atomic_read(&ctx->in_flight))
+               err = -EIO; /* Disk failed during IO... */
+
        now = jiffies;
        if (rw == WRITE) {
                drbd_md_flush(mdev);
@@ -1103,7 +1119,6 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
             ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now);
 
        kref_put(&ctx->kref, &bm_aio_ctx_destroy);
-
        return err;
 }
 
@@ -1167,14 +1182,20 @@ int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(loc
        *ctx = (struct bm_aio_ctx) {
                .mdev = mdev,
                .in_flight = ATOMIC_INIT(1),
-               .done = COMPLETION_INITIALIZER(ctx->done),
+               .done = 0,
                .flags = BM_AIO_COPY_PAGES,
                .error = 0,
                .kref = { ATOMIC_INIT(2) },
        };
 
+       if (!get_ldev_if_state(mdev, D_ATTACHING)) {  /* put is in bm_aio_ctx_destroy() */
+               dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in drbd_bm_write_page()\n");
+               kfree(ctx);
+               return -ENODEV;
+       }
+
        bm_page_io_async(ctx, idx, WRITE_SYNC);
-       wait_for_completion(&ctx->done);
+       wait_until_done_or_disk_failure(mdev, &ctx->done);
 
        if (ctx->error)
                drbd_chk_io_error(mdev, 1, true);
@@ -1182,9 +1203,8 @@ int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(loc
                 * gone in a moment as well. */
 
        mdev->bm_writ_cnt++;
-       err = ctx->error;
+       err = atomic_read(&ctx->in_flight) ? -EIO : ctx->error;
        kref_put(&ctx->kref, &bm_aio_ctx_destroy);
-
        return err;
 }