From 0e8488ade26b4b16a9745aa15ecb88c3fb1cb953 Mon Sep 17 00:00:00 2001 From: Lars Ellenberg Date: Wed, 25 Apr 2012 23:06:45 +0200 Subject: [PATCH] drbd: allow bitmap to change during writeout from resync_finished Symptom: messages similar to "FIXME asender in bm_change_bits_to, bitmap locked for 'write from resync_finished' by worker" If a resync or verify is finished (or aborted), a full bitmap writeout is triggered. If we have ongoing local IO, the bitmap may still change during that writeout, pending and not yet processed acks may cause bits to be cleared, while new writes may cause bits to be to be set. To fix this, introduce the drbd_bm_write_copy_pages() variant. Signed-off-by: Philipp Reisner Signed-off-by: Lars Ellenberg --- drivers/block/drbd/drbd_bitmap.c | 26 +++++++++++++++++++++----- drivers/block/drbd/drbd_int.h | 15 +++++++++++---- drivers/block/drbd/drbd_main.c | 4 ++-- 3 files changed, 34 insertions(+), 11 deletions(-) diff --git a/drivers/block/drbd/drbd_bitmap.c b/drivers/block/drbd/drbd_bitmap.c index 49603bc67fe4..9dab3700ca2d 100644 --- a/drivers/block/drbd/drbd_bitmap.c +++ b/drivers/block/drbd/drbd_bitmap.c @@ -1011,7 +1011,7 @@ static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must /* * bm_rw: read/write the whole bitmap from/to its on disk location. */ -static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_idx) __must_hold(local) +static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_writeout_upper_idx) __must_hold(local) { struct bm_aio_ctx *ctx; struct drbd_bitmap *b = mdev->bitmap; @@ -1037,7 +1037,7 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id .mdev = mdev, .in_flight = ATOMIC_INIT(1), .done = 0, - .flags = lazy_writeout_upper_idx ? BM_AIO_COPY_PAGES : 0, + .flags = flags, .error = 0, .kref = { ATOMIC_INIT(2) }, }; @@ -1128,7 +1128,7 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id */ int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local) { - return bm_rw(mdev, READ, 0); + return bm_rw(mdev, READ, 0, 0); } /** @@ -1139,7 +1139,7 @@ int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local) */ int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local) { - return bm_rw(mdev, WRITE, 0); + return bm_rw(mdev, WRITE, 0, 0); } /** @@ -1149,7 +1149,23 @@ int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local) */ int drbd_bm_write_lazy(struct drbd_conf *mdev, unsigned upper_idx) __must_hold(local) { - return bm_rw(mdev, WRITE, upper_idx); + return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, upper_idx); +} + +/** + * drbd_bm_write_copy_pages() - Write the whole bitmap to its on disk location. + * @mdev: DRBD device. + * + * Will only write pages that have changed since last IO. + * In contrast to drbd_bm_write(), this will copy the bitmap pages + * to temporary writeout pages. It is intended to trigger a full write-out + * while still allowing the bitmap to change, for example if a resync or online + * verify is aborted due to a failed peer disk, while local IO continues, or + * pending resync acks are still being processed. + */ +int drbd_bm_write_copy_pages(struct drbd_conf *mdev) __must_hold(local) +{ + return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, 0); } diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h index 302a6e786f76..6586053429b4 100644 --- a/drivers/block/drbd/drbd_int.h +++ b/drivers/block/drbd/drbd_int.h @@ -862,22 +862,28 @@ enum bm_flag { BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */ /* currently locked for bulk operation */ - BM_LOCKED_MASK = 0x7, + BM_LOCKED_MASK = 0xf, /* in detail, that is: */ BM_DONT_CLEAR = 0x1, BM_DONT_SET = 0x2, BM_DONT_TEST = 0x4, + /* so we can mark it locked for bulk operation, + * and still allow all non-bulk operations */ + BM_IS_LOCKED = 0x8, + /* (test bit, count bit) allowed (common case) */ - BM_LOCKED_TEST_ALLOWED = 0x3, + BM_LOCKED_TEST_ALLOWED = BM_DONT_CLEAR | BM_DONT_SET | BM_IS_LOCKED, /* testing bits, as well as setting new bits allowed, but clearing bits * would be unexpected. Used during bitmap receive. Setting new bits * requires sending of "out-of-sync" information, though. */ - BM_LOCKED_SET_ALLOWED = 0x1, + BM_LOCKED_SET_ALLOWED = BM_DONT_CLEAR | BM_IS_LOCKED, - /* clear is not expected while bitmap is locked for bulk operation */ + /* for drbd_bm_write_copy_pages, everything is allowed, + * only concurrent bulk operations are locked out. */ + BM_LOCKED_CHANGE_ALLOWED = BM_IS_LOCKED, }; struct drbd_work_queue { @@ -1457,6 +1463,7 @@ extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr); extern int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local); extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local); extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local); +extern int drbd_bm_write_copy_pages(struct drbd_conf *mdev) __must_hold(local); extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, unsigned long al_enr); extern size_t drbd_bm_words(struct drbd_conf *mdev); diff --git a/drivers/block/drbd/drbd_main.c b/drivers/block/drbd/drbd_main.c index ab501b23b50e..d830116781f9 100644 --- a/drivers/block/drbd/drbd_main.c +++ b/drivers/block/drbd/drbd_main.c @@ -1698,8 +1698,8 @@ static void after_state_ch(struct drbd_conf *mdev, union drbd_state os, * No harm done if some bits change during this phase. */ if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) { - drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL, - "write from resync_finished", BM_LOCKED_SET_ALLOWED); + drbd_queue_bitmap_io(mdev, &drbd_bm_write_copy_pages, NULL, + "write from resync_finished", BM_LOCKED_CHANGE_ALLOWED); put_ldev(mdev); } -- 2.30.2