The following changes since commit 2ef2f16781af9dee6ba6517755e9073ba5799fa2: Merge remote-tracking branch 'remotes/dgilbert/tags/pull-migration-20180615a' into staging (2018-06-15 18:13:35 +0100) are available in the git repository at: git://repo.or.cz/qemu/kevin.git tags/for-upstream for you to fetch changes up to 4c790afe2503eab12874508acab5b388d7babfd2: Merge remote-tracking branch 'mreitz/tags/pull-block-2018-06-18' into queue-block (2018-06-18 17:20:42 +0200) ---------------------------------------------------------------- Block layer patches: - Active mirror (blockdev-mirror copy-mode=write-blocking) - bdrv_drain_*() fixes and test cases - Fix crash with scsi-hd and drive_del ---------------------------------------------------------------- Greg Kurz (1): block: fix QEMU crash with scsi-hd and drive_del Kevin Wolf (20): test-bdrv-drain: bdrv_drain() works with cross-AioContext events block: Use bdrv_do_drain_begin/end in bdrv_drain_all() block: Remove 'recursive' parameter from bdrv_drain_invoke() block: Don't manually poll in bdrv_drain_all() tests/test-bdrv-drain: bdrv_drain_all() works in coroutines now block: Avoid unnecessary aio_poll() in AIO_WAIT_WHILE() block: Really pause block jobs on drain block: Remove bdrv_drain_recurse() block: Drain recursively with a single BDRV_POLL_WHILE() test-bdrv-drain: Test node deletion in subtree recursion block: Don't poll in parent drain callbacks test-bdrv-drain: Graph change through parent callback block: Defer .bdrv_drain_begin callback to polling phase test-bdrv-drain: Test that bdrv_drain_invoke() doesn't poll block: Allow AIO_WAIT_WHILE with NULL ctx block: Move bdrv_drain_all_begin() out of coroutine context block: ignore_bds_parents parameter for drain functions block: Allow graph changes in bdrv_drain_all_begin/end sections test-bdrv-drain: Test graph changes in drain_all section Merge remote-tracking branch 'mreitz/tags/pull-block-2018-06-18' into queue-block Max Reitz (15): test-bdrv-drain: Add test for node deletion block/mirror: Pull out mirror_perform() block/mirror: Convert to coroutines block/mirror: Use CoQueue to wait on in-flight ops block/mirror: Wait for in-flight op conflicts block/mirror: Use source as a BdrvChild block: Generalize should_update_child() rule hbitmap: Add @advance param to hbitmap_iter_next() test-hbitmap: Add non-advancing iter_next tests block/dirty-bitmap: Add bdrv_dirty_iter_next_area block/mirror: Add MirrorBDSOpaque job: Add job_progress_increase_remaining() block/mirror: Add active mirroring block/mirror: Add copy mode QAPI interface iotests: Add test for active mirroring qapi/block-core.json | 29 +- include/block/aio-wait.h | 25 +- include/block/block.h | 31 +- include/block/block_int.h | 18 +- include/block/blockjob_int.h | 8 + include/block/dirty-bitmap.h | 2 + include/qemu/hbitmap.h | 5 +- include/qemu/job.h | 15 + block.c | 96 +++++- block/backup.c | 2 +- block/block-backend.c | 5 + block/dirty-bitmap.c | 57 +++- block/io.c | 332 ++++++++++++-------- block/mirror.c | 613 +++++++++++++++++++++++++++++-------- block/vvfat.c | 1 + blockdev.c | 9 +- blockjob.c | 23 ++ job.c | 5 + tests/test-bdrv-drain.c | 705 +++++++++++++++++++++++++++++++++++++++++-- tests/test-hbitmap.c | 38 ++- util/hbitmap.c | 10 +- tests/qemu-iotests/151 | 120 ++++++++ tests/qemu-iotests/151.out | 5 + tests/qemu-iotests/group | 1 + 24 files changed, 1836 insertions(+), 319 deletions(-) create mode 100755 tests/qemu-iotests/151 create mode 100644 tests/qemu-iotests/151.out As long as nobody keeps the other I/O thread from working, there is no reason why bdrv_drain() wouldn't work with cross-AioContext events. The key is that the root request we're waiting for is in the AioContext we're polling (which it always is for bdrv_drain()) so that aio_poll() is woken up in the end. Add a test case that shows that it works. Remove the comment in bdrv_drain() that claims otherwise. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/io.c | 4 -- tests/test-bdrv-drain.c | 187 +++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 186 insertions(+), 5 deletions(-) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent) * * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState * AioContext. - * - * Only this BlockDriverState's AioContext is run, so in-flight requests must - * not depend on events in other AioContexts. In that case, use - * bdrv_drain_all() instead. */ void coroutine_fn bdrv_co_drain(BlockDriverState *bs) { diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ #include "block/blockjob_int.h" #include "sysemu/block-backend.h" #include "qapi/error.h" +#include "iothread.h" + +static QemuEvent done_event; typedef struct BDRVTestState { int drain_count; + AioContext *bh_indirection_ctx; } BDRVTestState; static void coroutine_fn bdrv_test_co_drain_begin(BlockDriverState *bs) @@ -XXX,XX +XXX,XX @@ static void bdrv_test_close(BlockDriverState *bs) g_assert_cmpint(s->drain_count, >, 0); } +static void co_reenter_bh(void *opaque) +{ + aio_co_wake(opaque); +} + static int coroutine_fn bdrv_test_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) { + BDRVTestState *s = bs->opaque; + /* We want this request to stay until the polling loop in drain waits for * it to complete. We need to sleep a while as bdrv_drain_invoke() comes * first and polls its result, too, but it shouldn't accidentally complete * this request yet. */ qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 100000); + if (s->bh_indirection_ctx) { + aio_bh_schedule_oneshot(s->bh_indirection_ctx, co_reenter_bh, + qemu_coroutine_self()); + qemu_coroutine_yield(); + } + return 0; } @@ -XXX,XX +XXX,XX @@ static void test_graph_change(void) blk_unref(blk_b); } +struct test_iothread_data { + BlockDriverState *bs; + enum drain_type drain_type; + int *aio_ret; +}; + +static void test_iothread_drain_entry(void *opaque) +{ + struct test_iothread_data *data = opaque; + + aio_context_acquire(bdrv_get_aio_context(data->bs)); + do_drain_begin(data->drain_type, data->bs); + g_assert_cmpint(*data->aio_ret, ==, 0); + do_drain_end(data->drain_type, data->bs); + aio_context_release(bdrv_get_aio_context(data->bs)); + + qemu_event_set(&done_event); +} + +static void test_iothread_aio_cb(void *opaque, int ret) +{ + int *aio_ret = opaque; + *aio_ret = ret; + qemu_event_set(&done_event); +} + +/* + * Starts an AIO request on a BDS that runs in the AioContext of iothread 1. + * The request involves a BH on iothread 2 before it can complete. + * + * @drain_thread = 0 means that do_drain_begin/end are called from the main + * thread, @drain_thread = 1 means that they are called from iothread 1. Drain + * for this BDS cannot be called from iothread 2 because only the main thread + * may do cross-AioContext polling. + */ +static void test_iothread_common(enum drain_type drain_type, int drain_thread) +{ + BlockBackend *blk; + BlockDriverState *bs; + BDRVTestState *s; + BlockAIOCB *acb; + int aio_ret; + struct test_iothread_data data; + + IOThread *a = iothread_new(); + IOThread *b = iothread_new(); + AioContext *ctx_a = iothread_get_aio_context(a); + AioContext *ctx_b = iothread_get_aio_context(b); + + QEMUIOVector qiov; + struct iovec iov = { + .iov_base = NULL, + .iov_len = 0, + }; + qemu_iovec_init_external(&qiov, &iov, 1); + + /* bdrv_drain_all() may only be called from the main loop thread */ + if (drain_type == BDRV_DRAIN_ALL && drain_thread != 0) { + goto out; + } + + blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL); + bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR, + &error_abort); + s = bs->opaque; + blk_insert_bs(blk, bs, &error_abort); + + blk_set_aio_context(blk, ctx_a); + aio_context_acquire(ctx_a); + + s->bh_indirection_ctx = ctx_b; + + aio_ret = -EINPROGRESS; + if (drain_thread == 0) { + acb = blk_aio_preadv(blk, 0, &qiov, 0, test_iothread_aio_cb, &aio_ret); + } else { + acb = blk_aio_preadv(blk, 0, &qiov, 0, aio_ret_cb, &aio_ret); + } + g_assert(acb != NULL); + g_assert_cmpint(aio_ret, ==, -EINPROGRESS); + + aio_context_release(ctx_a); + + data = (struct test_iothread_data) { + .bs = bs, + .drain_type = drain_type, + .aio_ret = &aio_ret, + }; + + switch (drain_thread) { + case 0: + if (drain_type != BDRV_DRAIN_ALL) { + aio_context_acquire(ctx_a); + } + + /* The request is running on the IOThread a. Draining its block device + * will make sure that it has completed as far as the BDS is concerned, + * but the drain in this thread can continue immediately after + * bdrv_dec_in_flight() and aio_ret might be assigned only slightly + * later. */ + qemu_event_reset(&done_event); + do_drain_begin(drain_type, bs); + g_assert_cmpint(bs->in_flight, ==, 0); + + if (drain_type != BDRV_DRAIN_ALL) { + aio_context_release(ctx_a); + } + qemu_event_wait(&done_event); + if (drain_type != BDRV_DRAIN_ALL) { + aio_context_acquire(ctx_a); + } + + g_assert_cmpint(aio_ret, ==, 0); + do_drain_end(drain_type, bs); + + if (drain_type != BDRV_DRAIN_ALL) { + aio_context_release(ctx_a); + } + break; + case 1: + qemu_event_reset(&done_event); + aio_bh_schedule_oneshot(ctx_a, test_iothread_drain_entry, &data); + qemu_event_wait(&done_event); + break; + default: + g_assert_not_reached(); + } + + aio_context_acquire(ctx_a); + blk_set_aio_context(blk, qemu_get_aio_context()); + aio_context_release(ctx_a); + + bdrv_unref(bs); + blk_unref(blk); + +out: + iothread_join(a); + iothread_join(b); +} + +static void test_iothread_drain_all(void) +{ + test_iothread_common(BDRV_DRAIN_ALL, 0); + test_iothread_common(BDRV_DRAIN_ALL, 1); +} + +static void test_iothread_drain(void) +{ + test_iothread_common(BDRV_DRAIN, 0); + test_iothread_common(BDRV_DRAIN, 1); +} + +static void test_iothread_drain_subtree(void) +{ + test_iothread_common(BDRV_SUBTREE_DRAIN, 0); + test_iothread_common(BDRV_SUBTREE_DRAIN, 1); +} + typedef struct TestBlockJob { BlockJob common; @@ -XXX,XX +XXX,XX @@ static void test_blockjob_drain_subtree(void) int main(int argc, char **argv) { + int ret; + bdrv_init(); qemu_init_main_loop(&error_abort); g_test_init(&argc, &argv, NULL); + qemu_event_init(&done_event, false); g_test_add_func("/bdrv-drain/driver-cb/drain_all", test_drv_cb_drain_all); g_test_add_func("/bdrv-drain/driver-cb/drain", test_drv_cb_drain); @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/multiparent", test_multiparent); g_test_add_func("/bdrv-drain/graph-change", test_graph_change); + g_test_add_func("/bdrv-drain/iothread/drain_all", test_iothread_drain_all); + g_test_add_func("/bdrv-drain/iothread/drain", test_iothread_drain); + g_test_add_func("/bdrv-drain/iothread/drain_subtree", + test_iothread_drain_subtree); + g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all); g_test_add_func("/bdrv-drain/blockjob/drain", test_blockjob_drain); g_test_add_func("/bdrv-drain/blockjob/drain_subtree", test_blockjob_drain_subtree); - return g_test_run(); + ret = g_test_run(); + qemu_event_destroy(&done_event); + return ret; } -- 2.13.6 bdrv_do_drain_begin/end() implement already everything that bdrv_drain_all_begin/end() need and currently still do manually: Disable external events, call parent drain callbacks, call block driver callbacks. It also does two more things: The first is incrementing bs->quiesce_counter. bdrv_drain_all() already stood out in the test case by behaving different from the other drain variants. Adding this is not only safe, but in fact a bug fix. The second is calling bdrv_drain_recurse(). We already do that later in the same function in a loop, so basically doing an early first iteration doesn't hurt. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> --- block/io.c | 10 ++-------- tests/test-bdrv-drain.c | 14 ++++---------- 2 files changed, 6 insertions(+), 18 deletions(-) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void) for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { AioContext *aio_context = bdrv_get_aio_context(bs); - /* Stop things in parent-to-child order */ aio_context_acquire(aio_context); - aio_disable_external(aio_context); - bdrv_parent_drained_begin(bs, NULL); - bdrv_drain_invoke(bs, true, true); + bdrv_do_drained_begin(bs, true, NULL); aio_context_release(aio_context); if (!g_slist_find(aio_ctxs, aio_context)) { @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void) for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { AioContext *aio_context = bdrv_get_aio_context(bs); - /* Re-enable things in child-to-parent order */ aio_context_acquire(aio_context); - bdrv_drain_invoke(bs, false, true); - bdrv_parent_drained_end(bs, NULL); - aio_enable_external(aio_context); + bdrv_do_drained_end(bs, true, NULL); aio_context_release(aio_context); } } diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static void test_quiesce_common(enum drain_type drain_type, bool recursive) static void test_quiesce_drain_all(void) { - // XXX drain_all doesn't quiesce - //test_quiesce_common(BDRV_DRAIN_ALL, true); + test_quiesce_common(BDRV_DRAIN_ALL, true); } static void test_quiesce_drain(void) @@ -XXX,XX +XXX,XX @@ static void test_nested(void) for (outer = 0; outer < DRAIN_TYPE_MAX; outer++) { for (inner = 0; inner < DRAIN_TYPE_MAX; inner++) { - /* XXX bdrv_drain_all() doesn't increase the quiesce_counter */ - int bs_quiesce = (outer != BDRV_DRAIN_ALL) + - (inner != BDRV_DRAIN_ALL); - int backing_quiesce = (outer == BDRV_SUBTREE_DRAIN) + - (inner == BDRV_SUBTREE_DRAIN); - int backing_cb_cnt = (outer != BDRV_DRAIN) + + int backing_quiesce = (outer != BDRV_DRAIN) + (inner != BDRV_DRAIN); g_assert_cmpint(bs->quiesce_counter, ==, 0); @@ -XXX,XX +XXX,XX @@ static void test_nested(void) do_drain_begin(outer, bs); do_drain_begin(inner, bs); - g_assert_cmpint(bs->quiesce_counter, ==, bs_quiesce); + g_assert_cmpint(bs->quiesce_counter, ==, 2); g_assert_cmpint(backing->quiesce_counter, ==, backing_quiesce); g_assert_cmpint(s->drain_count, ==, 2); - g_assert_cmpint(backing_s->drain_count, ==, backing_cb_cnt); + g_assert_cmpint(backing_s->drain_count, ==, backing_quiesce); do_drain_end(inner, bs); do_drain_end(outer, bs); -- 2.13.6 All callers pass false for the 'recursive' parameter now. Remove it. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> --- block/io.c | 13 +++---------- 1 file changed, 3 insertions(+), 10 deletions(-) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_drain_invoke_entry(void *opaque) } /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */ -static void bdrv_drain_invoke(BlockDriverState *bs, bool begin, bool recursive) +static void bdrv_drain_invoke(BlockDriverState *bs, bool begin) { - BdrvChild *child, *tmp; BdrvCoDrainData data = { .bs = bs, .done = false, .begin = begin}; if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) || @@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin, bool recursive) data.co = qemu_coroutine_create(bdrv_drain_invoke_entry, &data); bdrv_coroutine_enter(bs, data.co); BDRV_POLL_WHILE(bs, !data.done); - - if (recursive) { - QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) { - bdrv_drain_invoke(child->bs, begin, true); - } - } } static bool bdrv_drain_recurse(BlockDriverState *bs) @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, } bdrv_parent_drained_begin(bs, parent); - bdrv_drain_invoke(bs, true, false); + bdrv_drain_invoke(bs, true); bdrv_drain_recurse(bs); if (recursive) { @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter); /* Re-enable things in child-to-parent order */ - bdrv_drain_invoke(bs, false, false); + bdrv_drain_invoke(bs, false); bdrv_parent_drained_end(bs, parent); if (old_quiesce_counter == 1) { aio_enable_external(bdrv_get_aio_context(bs)); -- 2.13.6 All involved nodes are already idle, we called bdrv_do_drain_begin() on them. The comment in the code suggested that this was not correct because the completion of a request on one node could spawn a new request on a different node (which might have been drained before, so we wouldn't drain the new request). In reality, new requests to different nodes aren't spawned out of nothing, but only in the context of a parent request, and they aren't submitted to random nodes, but only to child nodes. As long as we still poll for the completion of the parent request (which we do), draining each root node separately is good enough. Remove the additional polling code from bdrv_drain_all_begin() and replace it with an assertion that all nodes are already idle after we drained them separately. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> --- block/io.c | 41 ++++++++++++----------------------------- 1 file changed, 12 insertions(+), 29 deletions(-) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ void bdrv_drain(BlockDriverState *bs) bdrv_drained_end(bs); } +static void bdrv_drain_assert_idle(BlockDriverState *bs) +{ + BdrvChild *child, *next; + + assert(atomic_read(&bs->in_flight) == 0); + QLIST_FOREACH_SAFE(child, &bs->children, next, next) { + bdrv_drain_assert_idle(child->bs); + } +} + /* * Wait for pending requests to complete across all BlockDriverStates * @@ -XXX,XX +XXX,XX @@ void bdrv_drain(BlockDriverState *bs) */ void bdrv_drain_all_begin(void) { - /* Always run first iteration so any pending completion BHs run */ - bool waited = true; BlockDriverState *bs; BdrvNextIterator it; - GSList *aio_ctxs = NULL, *ctx; /* BDRV_POLL_WHILE() for a node can only be called from its own I/O thread * or the main loop AioContext. We potentially use BDRV_POLL_WHILE() on @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void) aio_context_acquire(aio_context); bdrv_do_drained_begin(bs, true, NULL); aio_context_release(aio_context); - - if (!g_slist_find(aio_ctxs, aio_context)) { - aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); - } } - /* Note that completion of an asynchronous I/O operation can trigger any - * number of other I/O operations on other devices---for example a - * coroutine can submit an I/O request to another device in response to - * request completion. Therefore we must keep looping until there was no - * more activity rather than simply draining each device independently. - */ - while (waited) { - waited = false; - - for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { - AioContext *aio_context = ctx->data; - - aio_context_acquire(aio_context); - for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { - if (aio_context == bdrv_get_aio_context(bs)) { - waited |= bdrv_drain_recurse(bs); - } - } - aio_context_release(aio_context); - } + for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { + bdrv_drain_assert_idle(bs); } - - g_slist_free(aio_ctxs); } void bdrv_drain_all_end(void) -- 2.13.6 Since we use bdrv_do_drained_begin/end() for bdrv_drain_all_begin/end(), coroutine context is automatically left with a BH, preventing the deadlocks that made bdrv_drain_all*() unsafe in coroutine context. Now that we even removed the old polling code as dead code, it's obvious that it's compatible now. Enable the coroutine test cases for bdrv_drain_all(). Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> --- tests/test-bdrv-drain.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-) diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static void test_drv_cb_drain_subtree(void) test_drv_cb_common(BDRV_SUBTREE_DRAIN, true); } +static void test_drv_cb_co_drain_all(void) +{ + call_in_coroutine(test_drv_cb_drain_all); +} + static void test_drv_cb_co_drain(void) { call_in_coroutine(test_drv_cb_drain); @@ -XXX,XX +XXX,XX @@ static void test_quiesce_drain_subtree(void) test_quiesce_common(BDRV_SUBTREE_DRAIN, true); } +static void test_quiesce_co_drain_all(void) +{ + call_in_coroutine(test_quiesce_drain_all); +} + static void test_quiesce_co_drain(void) { call_in_coroutine(test_quiesce_drain); @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/driver-cb/drain_subtree", test_drv_cb_drain_subtree); - // XXX bdrv_drain_all() doesn't work in coroutine context + g_test_add_func("/bdrv-drain/driver-cb/co/drain_all", + test_drv_cb_co_drain_all); g_test_add_func("/bdrv-drain/driver-cb/co/drain", test_drv_cb_co_drain); g_test_add_func("/bdrv-drain/driver-cb/co/drain_subtree", test_drv_cb_co_drain_subtree); @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/quiesce/drain_subtree", test_quiesce_drain_subtree); - // XXX bdrv_drain_all() doesn't work in coroutine context + g_test_add_func("/bdrv-drain/quiesce/co/drain_all", + test_quiesce_co_drain_all); g_test_add_func("/bdrv-drain/quiesce/co/drain", test_quiesce_co_drain); g_test_add_func("/bdrv-drain/quiesce/co/drain_subtree", test_quiesce_co_drain_subtree); -- 2.13.6 Commit 91af091f923 added an additional aio_poll() to BDRV_POLL_WHILE() in order to make sure that all pending BHs are executed on drain. This was the wrong place to make the fix, as it is useless overhead for all other users of the macro and unnecessarily complicates the mechanism. This patch effectively reverts said commit (the context has changed a bit and the code has moved to AIO_WAIT_WHILE()) and instead polls in the loop condition for drain. The effect is probably hard to measure in any real-world use case because actual I/O will dominate, but if I run only the initialisation part of 'qemu-img convert' where it calls bdrv_block_status() for the whole image to find out how much data there is copy, this phase actually needs only roughly half the time after this patch. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> --- include/block/aio-wait.h | 22 ++++++++-------------- block/io.c | 11 ++++++++++- 2 files changed, 18 insertions(+), 15 deletions(-) diff --git a/include/block/aio-wait.h b/include/block/aio-wait.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/aio-wait.h +++ b/include/block/aio-wait.h @@ -XXX,XX +XXX,XX @@ typedef struct { */ #define AIO_WAIT_WHILE(wait, ctx, cond) ({ \ bool waited_ = false; \ - bool busy_ = true; \ AioWait *wait_ = (wait); \ AioContext *ctx_ = (ctx); \ if (in_aio_context_home_thread(ctx_)) { \ - while ((cond) || busy_) { \ - busy_ = aio_poll(ctx_, (cond)); \ - waited_ |= !!(cond) | busy_; \ + while ((cond)) { \ + aio_poll(ctx_, true); \ + waited_ = true; \ } \ } else { \ assert(qemu_get_current_aio_context() == \ qemu_get_aio_context()); \ /* Increment wait_->num_waiters before evaluating cond. */ \ atomic_inc(&wait_->num_waiters); \ - while (busy_) { \ - if ((cond)) { \ - waited_ = busy_ = true; \ - aio_context_release(ctx_); \ - aio_poll(qemu_get_aio_context(), true); \ - aio_context_acquire(ctx_); \ - } else { \ - busy_ = aio_poll(ctx_, false); \ - waited_ |= busy_; \ - } \ + while ((cond)) { \ + aio_context_release(ctx_); \ + aio_poll(qemu_get_aio_context(), true); \ + aio_context_acquire(ctx_); \ + waited_ = true; \ } \ atomic_dec(&wait_->num_waiters); \ } \ diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin) BDRV_POLL_WHILE(bs, !data.done); } +/* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ +static bool bdrv_drain_poll(BlockDriverState *bs) +{ + /* Execute pending BHs first and check everything else only after the BHs + * have executed. */ + while (aio_poll(bs->aio_context, false)); + return atomic_read(&bs->in_flight); +} + static bool bdrv_drain_recurse(BlockDriverState *bs) { BdrvChild *child, *tmp; bool waited; /* Wait for drained requests to finish */ - waited = BDRV_POLL_WHILE(bs, atomic_read(&bs->in_flight) > 0); + waited = BDRV_POLL_WHILE(bs, bdrv_drain_poll(bs)); QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) { BlockDriverState *bs = child->bs; -- 2.13.6 We already requested that block jobs be paused in .bdrv_drained_begin, but no guarantee was made that the job was actually inactive at the point where bdrv_drained_begin() returned. This introduces a new callback BdrvChildRole.bdrv_drained_poll() and uses it to make bdrv_drain_poll() consider block jobs using the node to be drained. For the test case to work as expected, we have to switch from block_job_sleep_ns() to qemu_co_sleep_ns() so that the test job is even considered active and must be waited for when draining the node. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/block.h | 8 ++++++++ include/block/block_int.h | 7 +++++++ include/block/blockjob_int.h | 8 ++++++++ block.c | 9 +++++++++ block/io.c | 40 ++++++++++++++++++++++++++++++++++------ block/mirror.c | 8 ++++++++ blockjob.c | 23 +++++++++++++++++++++++ tests/test-bdrv-drain.c | 18 ++++++++++-------- 8 files changed, 107 insertions(+), 14 deletions(-) diff --git a/include/block/block.h b/include/block/block.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block.h +++ b/include/block/block.h @@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore); void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore); /** + * bdrv_drain_poll: + * + * Poll for pending requests in @bs and its parents (except for + * @ignore_parent). This is part of bdrv_drained_begin. + */ +bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent); + +/** * bdrv_drained_begin: * * Begin a quiesced section for exclusive access to the BDS, by disabling diff --git a/include/block/block_int.h b/include/block/block_int.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block_int.h +++ b/include/block/block_int.h @@ -XXX,XX +XXX,XX @@ struct BdrvChildRole { void (*drained_begin)(BdrvChild *child); void (*drained_end)(BdrvChild *child); + /* + * Returns whether the parent has pending requests for the child. This + * callback is polled after .drained_begin() has been called until all + * activity on the child has stopped. + */ + bool (*drained_poll)(BdrvChild *child); + /* Notifies the parent that the child has been activated/inactivated (e.g. * when migration is completing) and it can start/stop requesting * permissions and doing I/O on it. */ diff --git a/include/block/blockjob_int.h b/include/block/blockjob_int.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/blockjob_int.h +++ b/include/block/blockjob_int.h @@ -XXX,XX +XXX,XX @@ struct BlockJobDriver { JobDriver job_driver; /* + * Returns whether the job has pending requests for the child or will + * submit new requests before the next pause point. This callback is polled + * in the context of draining a job node after requesting that the job be + * paused, until all activity on the child has stopped. + */ + bool (*drained_poll)(BlockJob *job); + + /* * If the callback is not NULL, it will be invoked before the job is * resumed in a new AioContext. This is the place to move any resources * besides job->blk to the new AioContext. diff --git a/block.c b/block.c index XXXXXXX..XXXXXXX 100644 --- a/block.c +++ b/block.c @@ -XXX,XX +XXX,XX @@ static void bdrv_child_cb_drained_begin(BdrvChild *child) bdrv_drained_begin(bs); } +static bool bdrv_child_cb_drained_poll(BdrvChild *child) +{ + BlockDriverState *bs = child->opaque; + return bdrv_drain_poll(bs, NULL); +} + static void bdrv_child_cb_drained_end(BdrvChild *child) { BlockDriverState *bs = child->opaque; @@ -XXX,XX +XXX,XX @@ const BdrvChildRole child_file = { .get_parent_desc = bdrv_child_get_parent_desc, .inherit_options = bdrv_inherited_options, .drained_begin = bdrv_child_cb_drained_begin, + .drained_poll = bdrv_child_cb_drained_poll, .drained_end = bdrv_child_cb_drained_end, .attach = bdrv_child_cb_attach, .detach = bdrv_child_cb_detach, @@ -XXX,XX +XXX,XX @@ const BdrvChildRole child_format = { .get_parent_desc = bdrv_child_get_parent_desc, .inherit_options = bdrv_inherited_fmt_options, .drained_begin = bdrv_child_cb_drained_begin, + .drained_poll = bdrv_child_cb_drained_poll, .drained_end = bdrv_child_cb_drained_end, .attach = bdrv_child_cb_attach, .detach = bdrv_child_cb_detach, @@ -XXX,XX +XXX,XX @@ const BdrvChildRole child_backing = { .detach = bdrv_backing_detach, .inherit_options = bdrv_backing_options, .drained_begin = bdrv_child_cb_drained_begin, + .drained_poll = bdrv_child_cb_drained_poll, .drained_end = bdrv_child_cb_drained_end, .inactivate = bdrv_child_cb_inactivate, .update_filename = bdrv_backing_update_filename, diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore) } } +static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore) +{ + BdrvChild *c, *next; + bool busy = false; + + QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { + if (c == ignore) { + continue; + } + if (c->role->drained_poll) { + busy |= c->role->drained_poll(c); + } + } + + return busy; +} + static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) { dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); @@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin) } /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ -static bool bdrv_drain_poll(BlockDriverState *bs) +bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent) +{ + if (bdrv_parent_drained_poll(bs, ignore_parent)) { + return true; + } + + return atomic_read(&bs->in_flight); +} + +static bool bdrv_drain_poll_top_level(BlockDriverState *bs, + BdrvChild *ignore_parent) { /* Execute pending BHs first and check everything else only after the BHs * have executed. */ while (aio_poll(bs->aio_context, false)); - return atomic_read(&bs->in_flight); + + return bdrv_drain_poll(bs, ignore_parent); } -static bool bdrv_drain_recurse(BlockDriverState *bs) +static bool bdrv_drain_recurse(BlockDriverState *bs, BdrvChild *parent) { BdrvChild *child, *tmp; bool waited; /* Wait for drained requests to finish */ - waited = BDRV_POLL_WHILE(bs, bdrv_drain_poll(bs)); + waited = BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) { BlockDriverState *bs = child->bs; @@ -XXX,XX +XXX,XX @@ static bool bdrv_drain_recurse(BlockDriverState *bs) */ bdrv_ref(bs); } - waited |= bdrv_drain_recurse(bs); + waited |= bdrv_drain_recurse(bs, child); if (in_main_loop) { bdrv_unref(bs); } @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, bdrv_parent_drained_begin(bs, parent); bdrv_drain_invoke(bs, true); - bdrv_drain_recurse(bs); + bdrv_drain_recurse(bs, parent); if (recursive) { bs->recursive_quiesce_counter++; diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ static void mirror_pause(Job *job) mirror_wait_for_all_io(s); } +static bool mirror_drained_poll(BlockJob *job) +{ + MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); + return !!s->in_flight; +} + static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) { MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); @@ -XXX,XX +XXX,XX @@ static const BlockJobDriver mirror_job_driver = { .pause = mirror_pause, .complete = mirror_complete, }, + .drained_poll = mirror_drained_poll, .attached_aio_context = mirror_attached_aio_context, .drain = mirror_drain, }; @@ -XXX,XX +XXX,XX @@ static const BlockJobDriver commit_active_job_driver = { .pause = mirror_pause, .complete = mirror_complete, }, + .drained_poll = mirror_drained_poll, .attached_aio_context = mirror_attached_aio_context, .drain = mirror_drain, }; diff --git a/blockjob.c b/blockjob.c index XXXXXXX..XXXXXXX 100644 --- a/blockjob.c +++ b/blockjob.c @@ -XXX,XX +XXX,XX @@ static void child_job_drained_begin(BdrvChild *c) job_pause(&job->job); } +static bool child_job_drained_poll(BdrvChild *c) +{ + BlockJob *bjob = c->opaque; + Job *job = &bjob->job; + const BlockJobDriver *drv = block_job_driver(bjob); + + /* An inactive or completed job doesn't have any pending requests. Jobs + * with !job->busy are either already paused or have a pause point after + * being reentered, so no job driver code will run before they pause. */ + if (!job->busy || job_is_completed(job) || job->deferred_to_main_loop) { + return false; + } + + /* Otherwise, assume that it isn't fully stopped yet, but allow the job to + * override this assumption. */ + if (drv->drained_poll) { + return drv->drained_poll(bjob); + } else { + return true; + } +} + static void child_job_drained_end(BdrvChild *c) { BlockJob *job = c->opaque; @@ -XXX,XX +XXX,XX @@ static void child_job_drained_end(BdrvChild *c) static const BdrvChildRole child_job = { .get_parent_desc = child_job_get_parent_desc, .drained_begin = child_job_drained_begin, + .drained_poll = child_job_drained_poll, .drained_end = child_job_drained_end, .stay_at_node = true, }; diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static void coroutine_fn test_job_start(void *opaque) job_transition_to_ready(&s->common.job); while (!s->should_complete) { - job_sleep_ns(&s->common.job, 100000); + /* Avoid block_job_sleep_ns() because it marks the job as !busy. We + * want to emulate some actual activity (probably some I/O) here so + * that drain has to wait for this acitivity to stop. */ + qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 100000); + job_pause_point(&s->common.job); } job_defer_to_main_loop(&s->common.job, test_job_completed, NULL); @@ -XXX,XX +XXX,XX @@ static void test_blockjob_common(enum drain_type drain_type) g_assert_cmpint(job->job.pause_count, ==, 0); g_assert_false(job->job.paused); - g_assert_false(job->job.busy); /* We're in job_sleep_ns() */ + g_assert_true(job->job.busy); /* We're in job_sleep_ns() */ do_drain_begin(drain_type, src); @@ -XXX,XX +XXX,XX @@ static void test_blockjob_common(enum drain_type drain_type) } else { g_assert_cmpint(job->job.pause_count, ==, 1); } - /* XXX We don't wait until the job is actually paused. Is this okay? */ - /* g_assert_true(job->job.paused); */ + g_assert_true(job->job.paused); g_assert_false(job->job.busy); /* The job is paused */ do_drain_end(drain_type, src); g_assert_cmpint(job->job.pause_count, ==, 0); g_assert_false(job->job.paused); - g_assert_false(job->job.busy); /* We're in job_sleep_ns() */ + g_assert_true(job->job.busy); /* We're in qemu_co_sleep_ns() */ do_drain_begin(drain_type, target); @@ -XXX,XX +XXX,XX @@ static void test_blockjob_common(enum drain_type drain_type) } else { g_assert_cmpint(job->job.pause_count, ==, 1); } - /* XXX We don't wait until the job is actually paused. Is this okay? */ - /* g_assert_true(job->job.paused); */ + g_assert_true(job->job.paused); g_assert_false(job->job.busy); /* The job is paused */ do_drain_end(drain_type, target); g_assert_cmpint(job->job.pause_count, ==, 0); g_assert_false(job->job.paused); - g_assert_false(job->job.busy); /* We're in job_sleep_ns() */ + g_assert_true(job->job.busy); /* We're in job_sleep_ns() */ ret = job_complete_sync(&job->job, &error_abort); g_assert_cmpint(ret, ==, 0); -- 2.13.6 For bdrv_drain(), recursively waiting for child node requests is pointless because we didn't quiesce their parents, so new requests could come in anyway. Letting the function work only on a single node makes it more consistent. For subtree drains and drain_all, we already have the recursion in bdrv_do_drained_begin(), so the extra recursion doesn't add anything either. Remove the useless code. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> --- block/io.c | 36 +++--------------------------------- 1 file changed, 3 insertions(+), 33 deletions(-) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static bool bdrv_drain_poll_top_level(BlockDriverState *bs, return bdrv_drain_poll(bs, ignore_parent); } -static bool bdrv_drain_recurse(BlockDriverState *bs, BdrvChild *parent) -{ - BdrvChild *child, *tmp; - bool waited; - - /* Wait for drained requests to finish */ - waited = BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); - - QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) { - BlockDriverState *bs = child->bs; - bool in_main_loop = - qemu_get_current_aio_context() == qemu_get_aio_context(); - assert(bs->refcnt > 0); - if (in_main_loop) { - /* In case the recursive bdrv_drain_recurse processes a - * block_job_defer_to_main_loop BH and modifies the graph, - * let's hold a reference to bs until we are done. - * - * IOThread doesn't have such a BH, and it is not safe to call - * bdrv_unref without BQL, so skip doing it there. - */ - bdrv_ref(bs); - } - waited |= bdrv_drain_recurse(bs, child); - if (in_main_loop) { - bdrv_unref(bs); - } - } - - return waited; -} - static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, BdrvChild *parent); static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, bdrv_parent_drained_begin(bs, parent); bdrv_drain_invoke(bs, true); - bdrv_drain_recurse(bs, parent); + + /* Wait for drained requests to finish */ + BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); if (recursive) { bs->recursive_quiesce_counter++; -- 2.13.6 From: Max Reitz <mreitz@redhat.com> This patch adds two bdrv-drain tests for what happens if some BDS goes away during the drainage. The basic idea is that you have a parent BDS with some child nodes. Then, you drain one of the children. Because of that, the party who actually owns the parent decides to (A) delete it, or (B) detach all its children from it -- both while the child is still being drained. A real-world case where this can happen is the mirror block job, which may exit if you drain one of its children. Signed-off-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/test-bdrv-drain.c | 169 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 169 insertions(+) diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static void test_blockjob_drain_subtree(void) test_blockjob_common(BDRV_SUBTREE_DRAIN); } + +typedef struct BDRVTestTopState { + BdrvChild *wait_child; +} BDRVTestTopState; + +static void bdrv_test_top_close(BlockDriverState *bs) +{ + BdrvChild *c, *next_c; + QLIST_FOREACH_SAFE(c, &bs->children, next, next_c) { + bdrv_unref_child(bs, c); + } +} + +static int coroutine_fn bdrv_test_top_co_preadv(BlockDriverState *bs, + uint64_t offset, uint64_t bytes, + QEMUIOVector *qiov, int flags) +{ + BDRVTestTopState *tts = bs->opaque; + return bdrv_co_preadv(tts->wait_child, offset, bytes, qiov, flags); +} + +static BlockDriver bdrv_test_top_driver = { + .format_name = "test_top_driver", + .instance_size = sizeof(BDRVTestTopState), + + .bdrv_close = bdrv_test_top_close, + .bdrv_co_preadv = bdrv_test_top_co_preadv, + + .bdrv_child_perm = bdrv_format_default_perms, +}; + +typedef struct TestCoDeleteByDrainData { + BlockBackend *blk; + bool detach_instead_of_delete; + bool done; +} TestCoDeleteByDrainData; + +static void coroutine_fn test_co_delete_by_drain(void *opaque) +{ + TestCoDeleteByDrainData *dbdd = opaque; + BlockBackend *blk = dbdd->blk; + BlockDriverState *bs = blk_bs(blk); + BDRVTestTopState *tts = bs->opaque; + void *buffer = g_malloc(65536); + QEMUIOVector qiov; + struct iovec iov = { + .iov_base = buffer, + .iov_len = 65536, + }; + + qemu_iovec_init_external(&qiov, &iov, 1); + + /* Pretend some internal write operation from parent to child. + * Important: We have to read from the child, not from the parent! + * Draining works by first propagating it all up the tree to the + * root and then waiting for drainage from root to the leaves + * (protocol nodes). If we have a request waiting on the root, + * everything will be drained before we go back down the tree, but + * we do not want that. We want to be in the middle of draining + * when this following requests returns. */ + bdrv_co_preadv(tts->wait_child, 0, 65536, &qiov, 0); + + g_assert_cmpint(bs->refcnt, ==, 1); + + if (!dbdd->detach_instead_of_delete) { + blk_unref(blk); + } else { + BdrvChild *c, *next_c; + QLIST_FOREACH_SAFE(c, &bs->children, next, next_c) { + bdrv_unref_child(bs, c); + } + } + + dbdd->done = true; +} + +/** + * Test what happens when some BDS has some children, you drain one of + * them and this results in the BDS being deleted. + * + * If @detach_instead_of_delete is set, the BDS is not going to be + * deleted but will only detach all of its children. + */ +static void do_test_delete_by_drain(bool detach_instead_of_delete) +{ + BlockBackend *blk; + BlockDriverState *bs, *child_bs, *null_bs; + BDRVTestTopState *tts; + TestCoDeleteByDrainData dbdd; + Coroutine *co; + + bs = bdrv_new_open_driver(&bdrv_test_top_driver, "top", BDRV_O_RDWR, + &error_abort); + bs->total_sectors = 65536 >> BDRV_SECTOR_BITS; + tts = bs->opaque; + + null_bs = bdrv_open("null-co://", NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, + &error_abort); + bdrv_attach_child(bs, null_bs, "null-child", &child_file, &error_abort); + + /* This child will be the one to pass to requests through to, and + * it will stall until a drain occurs */ + child_bs = bdrv_new_open_driver(&bdrv_test, "child", BDRV_O_RDWR, + &error_abort); + child_bs->total_sectors = 65536 >> BDRV_SECTOR_BITS; + /* Takes our reference to child_bs */ + tts->wait_child = bdrv_attach_child(bs, child_bs, "wait-child", &child_file, + &error_abort); + + /* This child is just there to be deleted + * (for detach_instead_of_delete == true) */ + null_bs = bdrv_open("null-co://", NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, + &error_abort); + bdrv_attach_child(bs, null_bs, "null-child", &child_file, &error_abort); + + blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL); + blk_insert_bs(blk, bs, &error_abort); + + /* Referenced by blk now */ + bdrv_unref(bs); + + g_assert_cmpint(bs->refcnt, ==, 1); + g_assert_cmpint(child_bs->refcnt, ==, 1); + g_assert_cmpint(null_bs->refcnt, ==, 1); + + + dbdd = (TestCoDeleteByDrainData){ + .blk = blk, + .detach_instead_of_delete = detach_instead_of_delete, + .done = false, + }; + co = qemu_coroutine_create(test_co_delete_by_drain, &dbdd); + qemu_coroutine_enter(co); + + /* Drain the child while the read operation is still pending. + * This should result in the operation finishing and + * test_co_delete_by_drain() resuming. Thus, @bs will be deleted + * and the coroutine will exit while this drain operation is still + * in progress. */ + bdrv_ref(child_bs); + bdrv_drain(child_bs); + bdrv_unref(child_bs); + + while (!dbdd.done) { + aio_poll(qemu_get_aio_context(), true); + } + + if (detach_instead_of_delete) { + /* Here, the reference has not passed over to the coroutine, + * so we have to delete the BB ourselves */ + blk_unref(blk); + } +} + + +static void test_delete_by_drain(void) +{ + do_test_delete_by_drain(false); +} + +static void test_detach_by_drain(void) +{ + do_test_delete_by_drain(true); +} + + int main(int argc, char **argv) { int ret; @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/blockjob/drain_subtree", test_blockjob_drain_subtree); + g_test_add_func("/bdrv-drain/deletion", test_delete_by_drain); + g_test_add_func("/bdrv-drain/detach", test_detach_by_drain); + ret = g_test_run(); qemu_event_destroy(&done_event); return ret; -- 2.13.6 Anything can happen inside BDRV_POLL_WHILE(), including graph changes that may interfere with its callers (e.g. child list iteration in recursive callers of bdrv_do_drained_begin). Switch to a single BDRV_POLL_WHILE() call for the whole subtree at the end of bdrv_do_drained_begin() to avoid such effects. The recursion happens now inside the loop condition. As the graph can only change between bdrv_drain_poll() calls, but not inside of it, doing the recursion here is safe. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/block.h | 9 +++++--- block.c | 2 +- block/io.c | 63 ++++++++++++++++++++++++++++++++++++--------------- 3 files changed, 52 insertions(+), 22 deletions(-) diff --git a/include/block/block.h b/include/block/block.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block.h +++ b/include/block/block.h @@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore); /** * bdrv_drain_poll: * - * Poll for pending requests in @bs and its parents (except for - * @ignore_parent). This is part of bdrv_drained_begin. + * Poll for pending requests in @bs, its parents (except for @ignore_parent), + * and if @recursive is true its children as well. + * + * This is part of bdrv_drained_begin. */ -bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent); +bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, + BdrvChild *ignore_parent); /** * bdrv_drained_begin: diff --git a/block.c b/block.c index XXXXXXX..XXXXXXX 100644 --- a/block.c +++ b/block.c @@ -XXX,XX +XXX,XX @@ static void bdrv_child_cb_drained_begin(BdrvChild *child) static bool bdrv_child_cb_drained_poll(BdrvChild *child) { BlockDriverState *bs = child->opaque; - return bdrv_drain_poll(bs, NULL); + return bdrv_drain_poll(bs, false, NULL); } static void bdrv_child_cb_drained_end(BdrvChild *child) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ typedef struct { bool done; bool begin; bool recursive; + bool poll; BdrvChild *parent; } BdrvCoDrainData; @@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin) } /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ -bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent) +bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, + BdrvChild *ignore_parent) { + BdrvChild *child, *next; + if (bdrv_parent_drained_poll(bs, ignore_parent)) { return true; } - return atomic_read(&bs->in_flight); + if (atomic_read(&bs->in_flight)) { + return true; + } + + if (recursive) { + QLIST_FOREACH_SAFE(child, &bs->children, next, next) { + if (bdrv_drain_poll(child->bs, recursive, child)) { + return true; + } + } + } + + return false; } -static bool bdrv_drain_poll_top_level(BlockDriverState *bs, +static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive, BdrvChild *ignore_parent) { /* Execute pending BHs first and check everything else only after the BHs * have executed. */ while (aio_poll(bs->aio_context, false)); - return bdrv_drain_poll(bs, ignore_parent); + return bdrv_drain_poll(bs, recursive, ignore_parent); } static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, - BdrvChild *parent); + BdrvChild *parent, bool poll); static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, BdrvChild *parent); @@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque) bdrv_dec_in_flight(bs); if (data->begin) { - bdrv_do_drained_begin(bs, data->recursive, data->parent); + bdrv_do_drained_begin(bs, data->recursive, data->parent, data->poll); } else { bdrv_do_drained_end(bs, data->recursive, data->parent); } @@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque) static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, bool begin, bool recursive, - BdrvChild *parent) + BdrvChild *parent, bool poll) { BdrvCoDrainData data; @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, .begin = begin, .recursive = recursive, .parent = parent, + .poll = poll, }; bdrv_inc_in_flight(bs); aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, } void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, - BdrvChild *parent) + BdrvChild *parent, bool poll) { BdrvChild *child, *next; if (qemu_in_coroutine()) { - bdrv_co_yield_to_drain(bs, true, recursive, parent); + bdrv_co_yield_to_drain(bs, true, recursive, parent, poll); return; } @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, bdrv_parent_drained_begin(bs, parent); bdrv_drain_invoke(bs, true); - /* Wait for drained requests to finish */ - BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); - if (recursive) { bs->recursive_quiesce_counter++; QLIST_FOREACH_SAFE(child, &bs->children, next, next) { - bdrv_do_drained_begin(child->bs, true, child); + bdrv_do_drained_begin(child->bs, true, child, false); } } + + /* + * Wait for drained requests to finish. + * + * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The + * call is needed so things in this AioContext can make progress even + * though we don't return to the main AioContext loop - this automatically + * includes other nodes in the same AioContext and therefore all child + * nodes. + */ + if (poll) { + BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent)); + } } void bdrv_drained_begin(BlockDriverState *bs) { - bdrv_do_drained_begin(bs, false, NULL); + bdrv_do_drained_begin(bs, false, NULL, true); } void bdrv_subtree_drained_begin(BlockDriverState *bs) { - bdrv_do_drained_begin(bs, true, NULL); + bdrv_do_drained_begin(bs, true, NULL, true); } void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, int old_quiesce_counter; if (qemu_in_coroutine()) { - bdrv_co_yield_to_drain(bs, false, recursive, parent); + bdrv_co_yield_to_drain(bs, false, recursive, parent, false); return; } assert(bs->quiesce_counter > 0); @@ -XXX,XX +XXX,XX @@ void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent) int i; for (i = 0; i < new_parent->recursive_quiesce_counter; i++) { - bdrv_do_drained_begin(child->bs, true, child); + bdrv_do_drained_begin(child->bs, true, child, true); } } @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void) AioContext *aio_context = bdrv_get_aio_context(bs); aio_context_acquire(aio_context); - bdrv_do_drained_begin(bs, true, NULL); + bdrv_do_drained_begin(bs, true, NULL, true); aio_context_release(aio_context); } -- 2.13.6 If bdrv_do_drained_begin() polls during its subtree recursion, the graph can change and mess up the bs->children iteration. Test that this doesn't happen. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/test-bdrv-drain.c | 38 +++++++++++++++++++++++++++++--------- 1 file changed, 29 insertions(+), 9 deletions(-) diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static void coroutine_fn test_co_delete_by_drain(void *opaque) * If @detach_instead_of_delete is set, the BDS is not going to be * deleted but will only detach all of its children. */ -static void do_test_delete_by_drain(bool detach_instead_of_delete) +static void do_test_delete_by_drain(bool detach_instead_of_delete, + enum drain_type drain_type) { BlockBackend *blk; BlockDriverState *bs, *child_bs, *null_bs; @@ -XXX,XX +XXX,XX @@ static void do_test_delete_by_drain(bool detach_instead_of_delete) * test_co_delete_by_drain() resuming. Thus, @bs will be deleted * and the coroutine will exit while this drain operation is still * in progress. */ - bdrv_ref(child_bs); - bdrv_drain(child_bs); - bdrv_unref(child_bs); + switch (drain_type) { + case BDRV_DRAIN: + bdrv_ref(child_bs); + bdrv_drain(child_bs); + bdrv_unref(child_bs); + break; + case BDRV_SUBTREE_DRAIN: + /* Would have to ref/unref bs here for !detach_instead_of_delete, but + * then the whole test becomes pointless because the graph changes + * don't occur during the drain any more. */ + assert(detach_instead_of_delete); + bdrv_subtree_drained_begin(bs); + bdrv_subtree_drained_end(bs); + break; + default: + g_assert_not_reached(); + } while (!dbdd.done) { aio_poll(qemu_get_aio_context(), true); @@ -XXX,XX +XXX,XX @@ static void do_test_delete_by_drain(bool detach_instead_of_delete) } } - static void test_delete_by_drain(void) { - do_test_delete_by_drain(false); + do_test_delete_by_drain(false, BDRV_DRAIN); } static void test_detach_by_drain(void) { - do_test_delete_by_drain(true); + do_test_delete_by_drain(true, BDRV_DRAIN); +} + +static void test_detach_by_drain_subtree(void) +{ + do_test_delete_by_drain(true, BDRV_SUBTREE_DRAIN); } @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/blockjob/drain_subtree", test_blockjob_drain_subtree); - g_test_add_func("/bdrv-drain/deletion", test_delete_by_drain); - g_test_add_func("/bdrv-drain/detach", test_detach_by_drain); + g_test_add_func("/bdrv-drain/deletion/drain", test_delete_by_drain); + g_test_add_func("/bdrv-drain/detach/drain", test_detach_by_drain); + g_test_add_func("/bdrv-drain/detach/drain_subtree", test_detach_by_drain_subtree); ret = g_test_run(); qemu_event_destroy(&done_event); -- 2.13.6 bdrv_do_drained_begin() is only safe if we have a single BDRV_POLL_WHILE() after quiescing all affected nodes. We cannot allow that parent callbacks introduce a nested polling loop that could cause graph changes while we're traversing the graph. Split off bdrv_do_drained_begin_quiesce(), which only quiesces a single node without waiting for its requests to complete. These requests will be waited for in the BDRV_POLL_WHILE() call down the call chain. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/block.h | 9 +++++++++ block.c | 2 +- block/io.c | 24 ++++++++++++++++-------- 3 files changed, 26 insertions(+), 9 deletions(-) diff --git a/include/block/block.h b/include/block/block.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block.h +++ b/include/block/block.h @@ -XXX,XX +XXX,XX @@ bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, void bdrv_drained_begin(BlockDriverState *bs); /** + * bdrv_do_drained_begin_quiesce: + * + * Quiesces a BDS like bdrv_drained_begin(), but does not wait for already + * running requests to complete. + */ +void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, + BdrvChild *parent); + +/** * Like bdrv_drained_begin, but recursively begins a quiesced section for * exclusive access to all child nodes as well. */ diff --git a/block.c b/block.c index XXXXXXX..XXXXXXX 100644 --- a/block.c +++ b/block.c @@ -XXX,XX +XXX,XX @@ static char *bdrv_child_get_parent_desc(BdrvChild *c) static void bdrv_child_cb_drained_begin(BdrvChild *child) { BlockDriverState *bs = child->opaque; - bdrv_drained_begin(bs); + bdrv_do_drained_begin_quiesce(bs, NULL); } static bool bdrv_child_cb_drained_poll(BdrvChild *child) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, assert(data.done); } -void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, - BdrvChild *parent, bool poll) +void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, + BdrvChild *parent) { - BdrvChild *child, *next; - - if (qemu_in_coroutine()) { - bdrv_co_yield_to_drain(bs, true, recursive, parent, poll); - return; - } + assert(!qemu_in_coroutine()); /* Stop things in parent-to-child order */ if (atomic_fetch_inc(&bs->quiesce_counter) == 0) { @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, bdrv_parent_drained_begin(bs, parent); bdrv_drain_invoke(bs, true); +} + +static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, + BdrvChild *parent, bool poll) +{ + BdrvChild *child, *next; + + if (qemu_in_coroutine()) { + bdrv_co_yield_to_drain(bs, true, recursive, parent, poll); + return; + } + + bdrv_do_drained_begin_quiesce(bs, parent); if (recursive) { bs->recursive_quiesce_counter++; -- 2.13.6 Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/test-bdrv-drain.c | 130 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 130 insertions(+) diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static void test_detach_by_drain_subtree(void) } +struct detach_by_parent_data { + BlockDriverState *parent_b; + BdrvChild *child_b; + BlockDriverState *c; + BdrvChild *child_c; +}; + +static void detach_by_parent_aio_cb(void *opaque, int ret) +{ + struct detach_by_parent_data *data = opaque; + + g_assert_cmpint(ret, ==, 0); + bdrv_unref_child(data->parent_b, data->child_b); + + bdrv_ref(data->c); + data->child_c = bdrv_attach_child(data->parent_b, data->c, "PB-C", + &child_file, &error_abort); +} + +/* + * Initial graph: + * + * PA PB + * \ / \ + * A B C + * + * PA has a pending write request whose callback changes the child nodes of PB: + * It removes B and adds C instead. The subtree of PB is drained, which will + * indirectly drain the write request, too. + */ +static void test_detach_by_parent_cb(void) +{ + BlockBackend *blk; + BlockDriverState *parent_a, *parent_b, *a, *b, *c; + BdrvChild *child_a, *child_b; + BlockAIOCB *acb; + struct detach_by_parent_data data; + + QEMUIOVector qiov; + struct iovec iov = { + .iov_base = NULL, + .iov_len = 0, + }; + qemu_iovec_init_external(&qiov, &iov, 1); + + /* Create all involved nodes */ + parent_a = bdrv_new_open_driver(&bdrv_test, "parent-a", BDRV_O_RDWR, + &error_abort); + parent_b = bdrv_new_open_driver(&bdrv_test, "parent-b", 0, + &error_abort); + + a = bdrv_new_open_driver(&bdrv_test, "a", BDRV_O_RDWR, &error_abort); + b = bdrv_new_open_driver(&bdrv_test, "b", BDRV_O_RDWR, &error_abort); + c = bdrv_new_open_driver(&bdrv_test, "c", BDRV_O_RDWR, &error_abort); + + /* blk is a BB for parent-a */ + blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL); + blk_insert_bs(blk, parent_a, &error_abort); + bdrv_unref(parent_a); + + /* Set child relationships */ + bdrv_ref(b); + bdrv_ref(a); + child_b = bdrv_attach_child(parent_b, b, "PB-B", &child_file, &error_abort); + child_a = bdrv_attach_child(parent_b, a, "PB-A", &child_backing, &error_abort); + + bdrv_ref(a); + bdrv_attach_child(parent_a, a, "PA-A", &child_file, &error_abort); + + g_assert_cmpint(parent_a->refcnt, ==, 1); + g_assert_cmpint(parent_b->refcnt, ==, 1); + g_assert_cmpint(a->refcnt, ==, 3); + g_assert_cmpint(b->refcnt, ==, 2); + g_assert_cmpint(c->refcnt, ==, 1); + + g_assert(QLIST_FIRST(&parent_b->children) == child_a); + g_assert(QLIST_NEXT(child_a, next) == child_b); + g_assert(QLIST_NEXT(child_b, next) == NULL); + + /* Start the evil write request */ + data = (struct detach_by_parent_data) { + .parent_b = parent_b, + .child_b = child_b, + .c = c, + }; + acb = blk_aio_preadv(blk, 0, &qiov, 0, detach_by_parent_aio_cb, &data); + g_assert(acb != NULL); + + /* Drain and check the expected result */ + bdrv_subtree_drained_begin(parent_b); + + g_assert(data.child_c != NULL); + + g_assert_cmpint(parent_a->refcnt, ==, 1); + g_assert_cmpint(parent_b->refcnt, ==, 1); + g_assert_cmpint(a->refcnt, ==, 3); + g_assert_cmpint(b->refcnt, ==, 1); + g_assert_cmpint(c->refcnt, ==, 2); + + g_assert(QLIST_FIRST(&parent_b->children) == data.child_c); + g_assert(QLIST_NEXT(data.child_c, next) == child_a); + g_assert(QLIST_NEXT(child_a, next) == NULL); + + g_assert_cmpint(parent_a->quiesce_counter, ==, 1); + g_assert_cmpint(parent_b->quiesce_counter, ==, 1); + g_assert_cmpint(a->quiesce_counter, ==, 1); + g_assert_cmpint(b->quiesce_counter, ==, 0); + g_assert_cmpint(c->quiesce_counter, ==, 1); + + bdrv_subtree_drained_end(parent_b); + + bdrv_unref(parent_b); + blk_unref(blk); + + /* XXX Once bdrv_close() unref's children instead of just detaching them, + * this won't be necessary any more. */ + bdrv_unref(a); + bdrv_unref(a); + bdrv_unref(c); + + g_assert_cmpint(a->refcnt, ==, 1); + g_assert_cmpint(b->refcnt, ==, 1); + g_assert_cmpint(c->refcnt, ==, 1); + bdrv_unref(a); + bdrv_unref(b); + bdrv_unref(c); +} + + int main(int argc, char **argv) { int ret; @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/deletion/drain", test_delete_by_drain); g_test_add_func("/bdrv-drain/detach/drain", test_detach_by_drain); g_test_add_func("/bdrv-drain/detach/drain_subtree", test_detach_by_drain_subtree); + g_test_add_func("/bdrv-drain/detach/parent_cb", test_detach_by_parent_cb); ret = g_test_run(); qemu_event_destroy(&done_event); -- 2.13.6 We cannot allow aio_poll() in bdrv_drain_invoke(begin=true) until we're done with propagating the drain through the graph and are doing the single final BDRV_POLL_WHILE(). Just schedule the coroutine with the callback and increase bs->in_flight to make sure that the polling phase will wait for it. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/io.c | 28 +++++++++++++++++++++++----- 1 file changed, 23 insertions(+), 5 deletions(-) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_drain_invoke_entry(void *opaque) /* Set data->done before reading bs->wakeup. */ atomic_mb_set(&data->done, true); - bdrv_wakeup(bs); + bdrv_dec_in_flight(bs); + + if (data->begin) { + g_free(data); + } } /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin) { - BdrvCoDrainData data = { .bs = bs, .done = false, .begin = begin}; + BdrvCoDrainData *data; if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) || (!begin && !bs->drv->bdrv_co_drain_end)) { return; } - data.co = qemu_coroutine_create(bdrv_drain_invoke_entry, &data); - bdrv_coroutine_enter(bs, data.co); - BDRV_POLL_WHILE(bs, !data.done); + data = g_new(BdrvCoDrainData, 1); + *data = (BdrvCoDrainData) { + .bs = bs, + .done = false, + .begin = begin + }; + + /* Make sure the driver callback completes during the polling phase for + * drain_begin. */ + bdrv_inc_in_flight(bs); + data->co = qemu_coroutine_create(bdrv_drain_invoke_entry, data); + aio_co_schedule(bdrv_get_aio_context(bs), data->co); + + if (!begin) { + BDRV_POLL_WHILE(bs, !data->done); + g_free(data); + } } /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ -- 2.13.6 This adds a test case that goes wrong if bdrv_drain_invoke() calls aio_poll(). Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/test-bdrv-drain.c | 102 +++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 88 insertions(+), 14 deletions(-) diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static QemuEvent done_event; typedef struct BDRVTestState { int drain_count; AioContext *bh_indirection_ctx; + bool sleep_in_drain_begin; } BDRVTestState; static void coroutine_fn bdrv_test_co_drain_begin(BlockDriverState *bs) { BDRVTestState *s = bs->opaque; s->drain_count++; + if (s->sleep_in_drain_begin) { + qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 100000); + } } static void coroutine_fn bdrv_test_co_drain_end(BlockDriverState *bs) @@ -XXX,XX +XXX,XX @@ static int coroutine_fn bdrv_test_co_preadv(BlockDriverState *bs, return 0; } +static void bdrv_test_child_perm(BlockDriverState *bs, BdrvChild *c, + const BdrvChildRole *role, + BlockReopenQueue *reopen_queue, + uint64_t perm, uint64_t shared, + uint64_t *nperm, uint64_t *nshared) +{ + /* bdrv_format_default_perms() accepts only these two, so disguise + * detach_by_driver_cb_role as one of them. */ + if (role != &child_file && role != &child_backing) { + role = &child_file; + } + + bdrv_format_default_perms(bs, c, role, reopen_queue, perm, shared, + nperm, nshared); +} + static BlockDriver bdrv_test = { .format_name = "test", .instance_size = sizeof(BDRVTestState), @@ -XXX,XX +XXX,XX @@ static BlockDriver bdrv_test = { .bdrv_co_drain_begin = bdrv_test_co_drain_begin, .bdrv_co_drain_end = bdrv_test_co_drain_end, - .bdrv_child_perm = bdrv_format_default_perms, + .bdrv_child_perm = bdrv_test_child_perm, }; static void aio_ret_cb(void *opaque, int ret) @@ -XXX,XX +XXX,XX @@ struct detach_by_parent_data { BdrvChild *child_b; BlockDriverState *c; BdrvChild *child_c; + bool by_parent_cb; }; +static struct detach_by_parent_data detach_by_parent_data; -static void detach_by_parent_aio_cb(void *opaque, int ret) +static void detach_indirect_bh(void *opaque) { struct detach_by_parent_data *data = opaque; - g_assert_cmpint(ret, ==, 0); bdrv_unref_child(data->parent_b, data->child_b); bdrv_ref(data->c); @@ -XXX,XX +XXX,XX @@ static void detach_by_parent_aio_cb(void *opaque, int ret) &child_file, &error_abort); } +static void detach_by_parent_aio_cb(void *opaque, int ret) +{ + struct detach_by_parent_data *data = &detach_by_parent_data; + + g_assert_cmpint(ret, ==, 0); + if (data->by_parent_cb) { + detach_indirect_bh(data); + } +} + +static void detach_by_driver_cb_drained_begin(BdrvChild *child) +{ + aio_bh_schedule_oneshot(qemu_get_current_aio_context(), + detach_indirect_bh, &detach_by_parent_data); + child_file.drained_begin(child); +} + +static BdrvChildRole detach_by_driver_cb_role; + /* * Initial graph: * @@ -XXX,XX +XXX,XX @@ static void detach_by_parent_aio_cb(void *opaque, int ret) * \ / \ * A B C * - * PA has a pending write request whose callback changes the child nodes of PB: - * It removes B and adds C instead. The subtree of PB is drained, which will - * indirectly drain the write request, too. + * by_parent_cb == true: Test that parent callbacks don't poll + * + * PA has a pending write request whose callback changes the child nodes of + * PB: It removes B and adds C instead. The subtree of PB is drained, which + * will indirectly drain the write request, too. + * + * by_parent_cb == false: Test that bdrv_drain_invoke() doesn't poll + * + * PA's BdrvChildRole has a .drained_begin callback that schedules a BH + * that does the same graph change. If bdrv_drain_invoke() calls it, the + * state is messed up, but if it is only polled in the single + * BDRV_POLL_WHILE() at the end of the drain, this should work fine. */ -static void test_detach_by_parent_cb(void) +static void test_detach_indirect(bool by_parent_cb) { BlockBackend *blk; BlockDriverState *parent_a, *parent_b, *a, *b, *c; BdrvChild *child_a, *child_b; BlockAIOCB *acb; - struct detach_by_parent_data data; QEMUIOVector qiov; struct iovec iov = { @@ -XXX,XX +XXX,XX @@ static void test_detach_by_parent_cb(void) }; qemu_iovec_init_external(&qiov, &iov, 1); + if (!by_parent_cb) { + detach_by_driver_cb_role = child_file; + detach_by_driver_cb_role.drained_begin = + detach_by_driver_cb_drained_begin; + } + /* Create all involved nodes */ parent_a = bdrv_new_open_driver(&bdrv_test, "parent-a", BDRV_O_RDWR, &error_abort); @@ -XXX,XX +XXX,XX @@ static void test_detach_by_parent_cb(void) blk_insert_bs(blk, parent_a, &error_abort); bdrv_unref(parent_a); + /* If we want to get bdrv_drain_invoke() to call aio_poll(), the driver + * callback must not return immediately. */ + if (!by_parent_cb) { + BDRVTestState *s = parent_a->opaque; + s->sleep_in_drain_begin = true; + } + /* Set child relationships */ bdrv_ref(b); bdrv_ref(a); @@ -XXX,XX +XXX,XX @@ static void test_detach_by_parent_cb(void) child_a = bdrv_attach_child(parent_b, a, "PB-A", &child_backing, &error_abort); bdrv_ref(a); - bdrv_attach_child(parent_a, a, "PA-A", &child_file, &error_abort); + bdrv_attach_child(parent_a, a, "PA-A", + by_parent_cb ? &child_file : &detach_by_driver_cb_role, + &error_abort); g_assert_cmpint(parent_a->refcnt, ==, 1); g_assert_cmpint(parent_b->refcnt, ==, 1); @@ -XXX,XX +XXX,XX @@ static void test_detach_by_parent_cb(void) g_assert(QLIST_NEXT(child_b, next) == NULL); /* Start the evil write request */ - data = (struct detach_by_parent_data) { + detach_by_parent_data = (struct detach_by_parent_data) { .parent_b = parent_b, .child_b = child_b, .c = c, + .by_parent_cb = by_parent_cb, }; - acb = blk_aio_preadv(blk, 0, &qiov, 0, detach_by_parent_aio_cb, &data); + acb = blk_aio_preadv(blk, 0, &qiov, 0, detach_by_parent_aio_cb, NULL); g_assert(acb != NULL); /* Drain and check the expected result */ bdrv_subtree_drained_begin(parent_b); - g_assert(data.child_c != NULL); + g_assert(detach_by_parent_data.child_c != NULL); g_assert_cmpint(parent_a->refcnt, ==, 1); g_assert_cmpint(parent_b->refcnt, ==, 1); @@ -XXX,XX +XXX,XX @@ static void test_detach_by_parent_cb(void) g_assert_cmpint(b->refcnt, ==, 1); g_assert_cmpint(c->refcnt, ==, 2); - g_assert(QLIST_FIRST(&parent_b->children) == data.child_c); - g_assert(QLIST_NEXT(data.child_c, next) == child_a); + g_assert(QLIST_FIRST(&parent_b->children) == detach_by_parent_data.child_c); + g_assert(QLIST_NEXT(detach_by_parent_data.child_c, next) == child_a); g_assert(QLIST_NEXT(child_a, next) == NULL); g_assert_cmpint(parent_a->quiesce_counter, ==, 1); @@ -XXX,XX +XXX,XX @@ static void test_detach_by_parent_cb(void) bdrv_unref(c); } +static void test_detach_by_parent_cb(void) +{ + test_detach_indirect(true); +} + +static void test_detach_by_driver_cb(void) +{ + test_detach_indirect(false); +} int main(int argc, char **argv) { @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/detach/drain", test_detach_by_drain); g_test_add_func("/bdrv-drain/detach/drain_subtree", test_detach_by_drain_subtree); g_test_add_func("/bdrv-drain/detach/parent_cb", test_detach_by_parent_cb); + g_test_add_func("/bdrv-drain/detach/driver_cb", test_detach_by_driver_cb); ret = g_test_run(); qemu_event_destroy(&done_event); -- 2.13.6 bdrv_drain_all() wants to have a single polling loop for draining the in-flight requests of all nodes. This means that the AIO_WAIT_WHILE() condition relies on activity in multiple AioContexts, which is polled from the mainloop context. We must therefore call AIO_WAIT_WHILE() from the mainloop thread and use the AioWait notification mechanism. Just randomly picking the AioContext of any non-mainloop thread would work, but instead of bothering to find such a context in the caller, we can just as well accept NULL for ctx. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/aio-wait.h | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-) diff --git a/include/block/aio-wait.h b/include/block/aio-wait.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/aio-wait.h +++ b/include/block/aio-wait.h @@ -XXX,XX +XXX,XX @@ typedef struct { /** * AIO_WAIT_WHILE: * @wait: the aio wait object - * @ctx: the aio context + * @ctx: the aio context, or NULL if multiple aio contexts (for which the + * caller does not hold a lock) are involved in the polling condition. * @cond: wait while this conditional expression is true * * Wait while a condition is true. Use this to implement synchronous @@ -XXX,XX +XXX,XX @@ typedef struct { bool waited_ = false; \ AioWait *wait_ = (wait); \ AioContext *ctx_ = (ctx); \ - if (in_aio_context_home_thread(ctx_)) { \ + if (ctx_ && in_aio_context_home_thread(ctx_)) { \ while ((cond)) { \ aio_poll(ctx_, true); \ waited_ = true; \ @@ -XXX,XX +XXX,XX @@ typedef struct { /* Increment wait_->num_waiters before evaluating cond. */ \ atomic_inc(&wait_->num_waiters); \ while ((cond)) { \ - aio_context_release(ctx_); \ + if (ctx_) { \ + aio_context_release(ctx_); \ + } \ aio_poll(qemu_get_aio_context(), true); \ - aio_context_acquire(ctx_); \ + if (ctx_) { \ + aio_context_acquire(ctx_); \ + } \ waited_ = true; \ } \ atomic_dec(&wait_->num_waiters); \ -- 2.13.6 Before we can introduce a single polling loop for all nodes in bdrv_drain_all_begin(), we must make sure to run it outside of coroutine context like we already do for bdrv_do_drained_begin(). Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/io.c | 22 +++++++++++++++++----- 1 file changed, 17 insertions(+), 5 deletions(-) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque) Coroutine *co = data->co; BlockDriverState *bs = data->bs; - bdrv_dec_in_flight(bs); - if (data->begin) { - bdrv_do_drained_begin(bs, data->recursive, data->parent, data->poll); + if (bs) { + bdrv_dec_in_flight(bs); + if (data->begin) { + bdrv_do_drained_begin(bs, data->recursive, data->parent, data->poll); + } else { + bdrv_do_drained_end(bs, data->recursive, data->parent); + } } else { - bdrv_do_drained_end(bs, data->recursive, data->parent); + assert(data->begin); + bdrv_drain_all_begin(); } data->done = true; @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, .parent = parent, .poll = poll, }; - bdrv_inc_in_flight(bs); + if (bs) { + bdrv_inc_in_flight(bs); + } aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), bdrv_co_drain_bh_cb, &data); @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void) BlockDriverState *bs; BdrvNextIterator it; + if (qemu_in_coroutine()) { + bdrv_co_yield_to_drain(NULL, true, false, NULL, true); + return; + } + /* BDRV_POLL_WHILE() for a node can only be called from its own I/O thread * or the main loop AioContext. We potentially use BDRV_POLL_WHILE() on * nodes in several different AioContexts, so make sure we're in the main -- 2.13.6 In the future, bdrv_drained_all_begin/end() will drain all invidiual nodes separately rather than whole subtrees. This means that we don't want to propagate the drain to all parents any more: If the parent is a BDS, it will already be drained separately. Recursing to all parents is unnecessary work and would make it an O(n²) operation. Prepare the drain function for the changed drain_all by adding an ignore_bds_parents parameter to the internal implementation that prevents the propagation of the drain to BDS parents. We still (have to) propagate it to non-BDS parents like BlockBackends or Jobs because those are not drained separately. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/block.h | 16 ++++++--- include/block/block_int.h | 6 ++++ block.c | 11 +++--- block/io.c | 88 ++++++++++++++++++++++++++++------------------- block/vvfat.c | 1 + 5 files changed, 78 insertions(+), 44 deletions(-) diff --git a/include/block/block.h b/include/block/block.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block.h +++ b/include/block/block.h @@ -XXX,XX +XXX,XX @@ void bdrv_io_unplug(BlockDriverState *bs); * Begin a quiesced section of all users of @bs. This is part of * bdrv_drained_begin. */ -void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore); +void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore, + bool ignore_bds_parents); /** * bdrv_parent_drained_end: @@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore); * End a quiesced section of all users of @bs. This is part of * bdrv_drained_end. */ -void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore); +void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore, + bool ignore_bds_parents); /** * bdrv_drain_poll: * * Poll for pending requests in @bs, its parents (except for @ignore_parent), - * and if @recursive is true its children as well. + * and if @recursive is true its children as well (used for subtree drain). + * + * If @ignore_bds_parents is true, parents that are BlockDriverStates must + * ignore the drain request because they will be drained separately (used for + * drain_all). * * This is part of bdrv_drained_begin. */ bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, - BdrvChild *ignore_parent); + BdrvChild *ignore_parent, bool ignore_bds_parents); /** * bdrv_drained_begin: @@ -XXX,XX +XXX,XX @@ void bdrv_drained_begin(BlockDriverState *bs); * running requests to complete. */ void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, - BdrvChild *parent); + BdrvChild *parent, bool ignore_bds_parents); /** * Like bdrv_drained_begin, but recursively begins a quiesced section for diff --git a/include/block/block_int.h b/include/block/block_int.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block_int.h +++ b/include/block/block_int.h @@ -XXX,XX +XXX,XX @@ struct BdrvChildRole { * points to. */ bool stay_at_node; + /* If true, the parent is a BlockDriverState and bdrv_next_all_states() + * will return it. This information is used for drain_all, where every node + * will be drained separately, so the drain only needs to be propagated to + * non-BDS parents. */ + bool parent_is_bds; + void (*inherit_options)(int *child_flags, QDict *child_options, int parent_flags, QDict *parent_options); diff --git a/block.c b/block.c index XXXXXXX..XXXXXXX 100644 --- a/block.c +++ b/block.c @@ -XXX,XX +XXX,XX @@ static char *bdrv_child_get_parent_desc(BdrvChild *c) static void bdrv_child_cb_drained_begin(BdrvChild *child) { BlockDriverState *bs = child->opaque; - bdrv_do_drained_begin_quiesce(bs, NULL); + bdrv_do_drained_begin_quiesce(bs, NULL, false); } static bool bdrv_child_cb_drained_poll(BdrvChild *child) { BlockDriverState *bs = child->opaque; - return bdrv_drain_poll(bs, false, NULL); + return bdrv_drain_poll(bs, false, NULL, false); } static void bdrv_child_cb_drained_end(BdrvChild *child) @@ -XXX,XX +XXX,XX @@ static void bdrv_inherited_options(int *child_flags, QDict *child_options, } const BdrvChildRole child_file = { + .parent_is_bds = true, .get_parent_desc = bdrv_child_get_parent_desc, .inherit_options = bdrv_inherited_options, .drained_begin = bdrv_child_cb_drained_begin, @@ -XXX,XX +XXX,XX @@ static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options, } const BdrvChildRole child_format = { + .parent_is_bds = true, .get_parent_desc = bdrv_child_get_parent_desc, .inherit_options = bdrv_inherited_fmt_options, .drained_begin = bdrv_child_cb_drained_begin, @@ -XXX,XX +XXX,XX @@ static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, } const BdrvChildRole child_backing = { + .parent_is_bds = true, .get_parent_desc = bdrv_child_get_parent_desc, .attach = bdrv_backing_attach, .detach = bdrv_backing_detach, @@ -XXX,XX +XXX,XX @@ void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context) AioContext *ctx = bdrv_get_aio_context(bs); aio_disable_external(ctx); - bdrv_parent_drained_begin(bs, NULL); + bdrv_parent_drained_begin(bs, NULL, false); bdrv_drain(bs); /* ensure there are no in-flight requests */ while (aio_poll(ctx, false)) { @@ -XXX,XX +XXX,XX @@ void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context) */ aio_context_acquire(new_context); bdrv_attach_aio_context(bs, new_context); - bdrv_parent_drained_end(bs, NULL); + bdrv_parent_drained_end(bs, NULL, false); aio_enable_external(ctx); aio_context_release(new_context); } diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, BdrvRequestFlags flags); -void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore) +void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore, + bool ignore_bds_parents) { BdrvChild *c, *next; QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { - if (c == ignore) { + if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { continue; } if (c->role->drained_begin) { @@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore) } } -void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore) +void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore, + bool ignore_bds_parents) { BdrvChild *c, *next; QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { - if (c == ignore) { + if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { continue; } if (c->role->drained_end) { @@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore) } } -static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore) +static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore, + bool ignore_bds_parents) { BdrvChild *c, *next; bool busy = false; QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { - if (c == ignore) { + if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { continue; } if (c->role->drained_poll) { @@ -XXX,XX +XXX,XX @@ typedef struct { bool recursive; bool poll; BdrvChild *parent; + bool ignore_bds_parents; } BdrvCoDrainData; static void coroutine_fn bdrv_drain_invoke_entry(void *opaque) @@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin) /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, - BdrvChild *ignore_parent) + BdrvChild *ignore_parent, bool ignore_bds_parents) { BdrvChild *child, *next; - if (bdrv_parent_drained_poll(bs, ignore_parent)) { + if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { return true; } @@ -XXX,XX +XXX,XX @@ bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, } if (recursive) { + assert(!ignore_bds_parents); QLIST_FOREACH_SAFE(child, &bs->children, next, next) { - if (bdrv_drain_poll(child->bs, recursive, child)) { + if (bdrv_drain_poll(child->bs, recursive, child, false)) { return true; } } @@ -XXX,XX +XXX,XX @@ static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive, * have executed. */ while (aio_poll(bs->aio_context, false)); - return bdrv_drain_poll(bs, recursive, ignore_parent); + return bdrv_drain_poll(bs, recursive, ignore_parent, false); } static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, - BdrvChild *parent, bool poll); + BdrvChild *parent, bool ignore_bds_parents, + bool poll); static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, - BdrvChild *parent); + BdrvChild *parent, bool ignore_bds_parents); static void bdrv_co_drain_bh_cb(void *opaque) { @@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque) if (bs) { bdrv_dec_in_flight(bs); if (data->begin) { - bdrv_do_drained_begin(bs, data->recursive, data->parent, data->poll); + bdrv_do_drained_begin(bs, data->recursive, data->parent, + data->ignore_bds_parents, data->poll); } else { - bdrv_do_drained_end(bs, data->recursive, data->parent); + bdrv_do_drained_end(bs, data->recursive, data->parent, + data->ignore_bds_parents); } } else { assert(data->begin); @@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque) static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, bool begin, bool recursive, - BdrvChild *parent, bool poll) + BdrvChild *parent, + bool ignore_bds_parents, + bool poll) { BdrvCoDrainData data; @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, .begin = begin, .recursive = recursive, .parent = parent, + .ignore_bds_parents = ignore_bds_parents, .poll = poll, }; if (bs) { @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, } void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, - BdrvChild *parent) + BdrvChild *parent, bool ignore_bds_parents) { assert(!qemu_in_coroutine()); @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, aio_disable_external(bdrv_get_aio_context(bs)); } - bdrv_parent_drained_begin(bs, parent); + bdrv_parent_drained_begin(bs, parent, ignore_bds_parents); bdrv_drain_invoke(bs, true); } static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, - BdrvChild *parent, bool poll) + BdrvChild *parent, bool ignore_bds_parents, + bool poll) { BdrvChild *child, *next; if (qemu_in_coroutine()) { - bdrv_co_yield_to_drain(bs, true, recursive, parent, poll); + bdrv_co_yield_to_drain(bs, true, recursive, parent, ignore_bds_parents, + poll); return; } - bdrv_do_drained_begin_quiesce(bs, parent); + bdrv_do_drained_begin_quiesce(bs, parent, ignore_bds_parents); if (recursive) { + assert(!ignore_bds_parents); bs->recursive_quiesce_counter++; QLIST_FOREACH_SAFE(child, &bs->children, next, next) { - bdrv_do_drained_begin(child->bs, true, child, false); + bdrv_do_drained_begin(child->bs, true, child, ignore_bds_parents, + false); } } @@ -XXX,XX +XXX,XX @@ static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, * nodes. */ if (poll) { + assert(!ignore_bds_parents); BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent)); } } void bdrv_drained_begin(BlockDriverState *bs) { - bdrv_do_drained_begin(bs, false, NULL, true); + bdrv_do_drained_begin(bs, false, NULL, false, true); } void bdrv_subtree_drained_begin(BlockDriverState *bs) { - bdrv_do_drained_begin(bs, true, NULL, true); + bdrv_do_drained_begin(bs, true, NULL, false, true); } -void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, - BdrvChild *parent) +static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, + BdrvChild *parent, bool ignore_bds_parents) { BdrvChild *child, *next; int old_quiesce_counter; if (qemu_in_coroutine()) { - bdrv_co_yield_to_drain(bs, false, recursive, parent, false); + bdrv_co_yield_to_drain(bs, false, recursive, parent, ignore_bds_parents, + false); return; } assert(bs->quiesce_counter > 0); @@ -XXX,XX +XXX,XX @@ void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, /* Re-enable things in child-to-parent order */ bdrv_drain_invoke(bs, false); - bdrv_parent_drained_end(bs, parent); + bdrv_parent_drained_end(bs, parent, ignore_bds_parents); if (old_quiesce_counter == 1) { aio_enable_external(bdrv_get_aio_context(bs)); } if (recursive) { + assert(!ignore_bds_parents); bs->recursive_quiesce_counter--; QLIST_FOREACH_SAFE(child, &bs->children, next, next) { - bdrv_do_drained_end(child->bs, true, child); + bdrv_do_drained_end(child->bs, true, child, ignore_bds_parents); } } } void bdrv_drained_end(BlockDriverState *bs) { - bdrv_do_drained_end(bs, false, NULL); + bdrv_do_drained_end(bs, false, NULL, false); } void bdrv_subtree_drained_end(BlockDriverState *bs) { - bdrv_do_drained_end(bs, true, NULL); + bdrv_do_drained_end(bs, true, NULL, false); } void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent) @@ -XXX,XX +XXX,XX @@ void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent) int i; for (i = 0; i < new_parent->recursive_quiesce_counter; i++) { - bdrv_do_drained_begin(child->bs, true, child, true); + bdrv_do_drained_begin(child->bs, true, child, false, true); } } @@ -XXX,XX +XXX,XX @@ void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent) int i; for (i = 0; i < old_parent->recursive_quiesce_counter; i++) { - bdrv_do_drained_end(child->bs, true, child); + bdrv_do_drained_end(child->bs, true, child, false); } } @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void) BdrvNextIterator it; if (qemu_in_coroutine()) { - bdrv_co_yield_to_drain(NULL, true, false, NULL, true); + bdrv_co_yield_to_drain(NULL, true, false, NULL, false, true); return; } @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void) AioContext *aio_context = bdrv_get_aio_context(bs); aio_context_acquire(aio_context); - bdrv_do_drained_begin(bs, true, NULL, true); + bdrv_do_drained_begin(bs, true, NULL, false, true); aio_context_release(aio_context); } @@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void) AioContext *aio_context = bdrv_get_aio_context(bs); aio_context_acquire(aio_context); - bdrv_do_drained_end(bs, true, NULL); + bdrv_do_drained_end(bs, true, NULL, false); aio_context_release(aio_context); } } diff --git a/block/vvfat.c b/block/vvfat.c index XXXXXXX..XXXXXXX 100644 --- a/block/vvfat.c +++ b/block/vvfat.c @@ -XXX,XX +XXX,XX @@ static void vvfat_qcow_options(int *child_flags, QDict *child_options, } static const BdrvChildRole child_vvfat_qcow = { + .parent_is_bds = true, .inherit_options = vvfat_qcow_options, }; -- 2.13.6 bdrv_drain_all_*() used bdrv_next() to iterate over all root nodes and did a subtree drain for each of them. This works fine as long as the graph is static, but sadly, reality looks different. If the graph changes so that root nodes are added or removed, we would have to compensate for this. bdrv_next() returns each root node only once even if it's the root node for multiple BlockBackends or for a monitor-owned block driver tree, which would only complicate things. The much easier and more obviously correct way is to fundamentally change the way the functions work: Iterate over all BlockDriverStates, no matter who owns them, and drain them individually. Compensation is only necessary when a new BDS is created inside a drain_all section. Removal of a BDS doesn't require any action because it's gone afterwards anyway. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/block.h | 1 + include/block/block_int.h | 1 + block.c | 34 ++++++++++++++++++++++++--- block/io.c | 60 ++++++++++++++++++++++++++++++++++++----------- 4 files changed, 79 insertions(+), 17 deletions(-) diff --git a/include/block/block.h b/include/block/block.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block.h +++ b/include/block/block.h @@ -XXX,XX +XXX,XX @@ BlockDriverState *bdrv_lookup_bs(const char *device, Error **errp); bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base); BlockDriverState *bdrv_next_node(BlockDriverState *bs); +BlockDriverState *bdrv_next_all_states(BlockDriverState *bs); typedef struct BdrvNextIterator { enum { diff --git a/include/block/block_int.h b/include/block/block_int.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block_int.h +++ b/include/block/block_int.h @@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_pwritev(BdrvChild *child, int64_t offset, unsigned int bytes, QEMUIOVector *qiov, BdrvRequestFlags flags); +extern unsigned int bdrv_drain_all_count; void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent); void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent); diff --git a/block.c b/block.c index XXXXXXX..XXXXXXX 100644 --- a/block.c +++ b/block.c @@ -XXX,XX +XXX,XX @@ BlockDriverState *bdrv_new(void) qemu_co_queue_init(&bs->flush_queue); + for (i = 0; i < bdrv_drain_all_count; i++) { + bdrv_drained_begin(bs); + } + QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); return bs; @@ -XXX,XX +XXX,XX @@ static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, int open_flags, Error **errp) { Error *local_err = NULL; - int ret; + int i, ret; bdrv_assign_node_name(bs, node_name, &local_err); if (local_err) { @@ -XXX,XX +XXX,XX @@ static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, assert(bdrv_min_mem_align(bs) != 0); assert(is_power_of_2(bs->bl.request_alignment)); + for (i = 0; i < bs->quiesce_counter; i++) { + if (drv->bdrv_co_drain_begin) { + drv->bdrv_co_drain_begin(bs); + } + } + return 0; open_failed: bs->drv = NULL; @@ -XXX,XX +XXX,XX @@ static void bdrv_replace_child_noperm(BdrvChild *child, child->role->detach(child); } if (old_bs->quiesce_counter && child->role->drained_end) { - for (i = 0; i < old_bs->quiesce_counter; i++) { + int num = old_bs->quiesce_counter; + if (child->role->parent_is_bds) { + num -= bdrv_drain_all_count; + } + assert(num >= 0); + for (i = 0; i < num; i++) { child->role->drained_end(child); } } @@ -XXX,XX +XXX,XX @@ static void bdrv_replace_child_noperm(BdrvChild *child, if (new_bs) { QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); if (new_bs->quiesce_counter && child->role->drained_begin) { - for (i = 0; i < new_bs->quiesce_counter; i++) { + int num = new_bs->quiesce_counter; + if (child->role->parent_is_bds) { + num -= bdrv_drain_all_count; + } + assert(num >= 0); + for (i = 0; i < num; i++) { child->role->drained_begin(child); } } @@ -XXX,XX +XXX,XX @@ BlockDriverState *bdrv_next_node(BlockDriverState *bs) return QTAILQ_NEXT(bs, node_list); } +BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) +{ + if (!bs) { + return QTAILQ_FIRST(&all_bdrv_states); + } + return QTAILQ_NEXT(bs, bs_list); +} + const char *bdrv_get_node_name(const BlockDriverState *bs) { return bs->node_name; diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */ #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS) +static AioWait drain_all_aio_wait; + static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, BdrvRequestFlags flags); @@ -XXX,XX +XXX,XX @@ static void bdrv_drain_assert_idle(BlockDriverState *bs) } } +unsigned int bdrv_drain_all_count = 0; + +static bool bdrv_drain_all_poll(void) +{ + BlockDriverState *bs = NULL; + bool result = false; + + /* Execute pending BHs first (may modify the graph) and check everything + * else only after the BHs have executed. */ + while (aio_poll(qemu_get_aio_context(), false)); + + /* bdrv_drain_poll() can't make changes to the graph and we are holding the + * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ + while ((bs = bdrv_next_all_states(bs))) { + AioContext *aio_context = bdrv_get_aio_context(bs); + aio_context_acquire(aio_context); + result |= bdrv_drain_poll(bs, false, NULL, true); + aio_context_release(aio_context); + } + + return result; +} + /* * Wait for pending requests to complete across all BlockDriverStates * @@ -XXX,XX +XXX,XX @@ static void bdrv_drain_assert_idle(BlockDriverState *bs) */ void bdrv_drain_all_begin(void) { - BlockDriverState *bs; - BdrvNextIterator it; + BlockDriverState *bs = NULL; if (qemu_in_coroutine()) { - bdrv_co_yield_to_drain(NULL, true, false, NULL, false, true); + bdrv_co_yield_to_drain(NULL, true, false, NULL, true, true); return; } - /* BDRV_POLL_WHILE() for a node can only be called from its own I/O thread - * or the main loop AioContext. We potentially use BDRV_POLL_WHILE() on - * nodes in several different AioContexts, so make sure we're in the main - * context. */ + /* AIO_WAIT_WHILE() with a NULL context can only be called from the main + * loop AioContext, so make sure we're in the main context. */ assert(qemu_get_current_aio_context() == qemu_get_aio_context()); + assert(bdrv_drain_all_count < INT_MAX); + bdrv_drain_all_count++; - for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { + /* Quiesce all nodes, without polling in-flight requests yet. The graph + * cannot change during this loop. */ + while ((bs = bdrv_next_all_states(bs))) { AioContext *aio_context = bdrv_get_aio_context(bs); aio_context_acquire(aio_context); - bdrv_do_drained_begin(bs, true, NULL, false, true); + bdrv_do_drained_begin(bs, false, NULL, true, false); aio_context_release(aio_context); } - for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { + /* Now poll the in-flight requests */ + AIO_WAIT_WHILE(&drain_all_aio_wait, NULL, bdrv_drain_all_poll()); + + while ((bs = bdrv_next_all_states(bs))) { bdrv_drain_assert_idle(bs); } } void bdrv_drain_all_end(void) { - BlockDriverState *bs; - BdrvNextIterator it; + BlockDriverState *bs = NULL; - for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { + while ((bs = bdrv_next_all_states(bs))) { AioContext *aio_context = bdrv_get_aio_context(bs); aio_context_acquire(aio_context); - bdrv_do_drained_end(bs, true, NULL, false); + bdrv_do_drained_end(bs, false, NULL, true); aio_context_release(aio_context); } + + assert(bdrv_drain_all_count > 0); + bdrv_drain_all_count--; } void bdrv_drain_all(void) @@ -XXX,XX +XXX,XX @@ void bdrv_inc_in_flight(BlockDriverState *bs) void bdrv_wakeup(BlockDriverState *bs) { aio_wait_kick(bdrv_get_aio_wait(bs)); + aio_wait_kick(&drain_all_aio_wait); } void bdrv_dec_in_flight(BlockDriverState *bs) -- 2.13.6 This tests both adding and remove a node between bdrv_drain_all_begin() and bdrv_drain_all_end(), and enabled the existing detach test for drain_all. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/test-bdrv-drain.c | 75 +++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 73 insertions(+), 2 deletions(-) diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-bdrv-drain.c +++ b/tests/test-bdrv-drain.c @@ -XXX,XX +XXX,XX @@ static void test_multiparent(void) blk_unref(blk_b); } -static void test_graph_change(void) +static void test_graph_change_drain_subtree(void) { BlockBackend *blk_a, *blk_b; BlockDriverState *bs_a, *bs_b, *backing; @@ -XXX,XX +XXX,XX @@ static void test_graph_change(void) blk_unref(blk_b); } +static void test_graph_change_drain_all(void) +{ + BlockBackend *blk_a, *blk_b; + BlockDriverState *bs_a, *bs_b; + BDRVTestState *a_s, *b_s; + + /* Create node A with a BlockBackend */ + blk_a = blk_new(BLK_PERM_ALL, BLK_PERM_ALL); + bs_a = bdrv_new_open_driver(&bdrv_test, "test-node-a", BDRV_O_RDWR, + &error_abort); + a_s = bs_a->opaque; + blk_insert_bs(blk_a, bs_a, &error_abort); + + g_assert_cmpint(bs_a->quiesce_counter, ==, 0); + g_assert_cmpint(a_s->drain_count, ==, 0); + + /* Call bdrv_drain_all_begin() */ + bdrv_drain_all_begin(); + + g_assert_cmpint(bs_a->quiesce_counter, ==, 1); + g_assert_cmpint(a_s->drain_count, ==, 1); + + /* Create node B with a BlockBackend */ + blk_b = blk_new(BLK_PERM_ALL, BLK_PERM_ALL); + bs_b = bdrv_new_open_driver(&bdrv_test, "test-node-b", BDRV_O_RDWR, + &error_abort); + b_s = bs_b->opaque; + blk_insert_bs(blk_b, bs_b, &error_abort); + + g_assert_cmpint(bs_a->quiesce_counter, ==, 1); + g_assert_cmpint(bs_b->quiesce_counter, ==, 1); + g_assert_cmpint(a_s->drain_count, ==, 1); + g_assert_cmpint(b_s->drain_count, ==, 1); + + /* Unref and finally delete node A */ + blk_unref(blk_a); + + g_assert_cmpint(bs_a->quiesce_counter, ==, 1); + g_assert_cmpint(bs_b->quiesce_counter, ==, 1); + g_assert_cmpint(a_s->drain_count, ==, 1); + g_assert_cmpint(b_s->drain_count, ==, 1); + + bdrv_unref(bs_a); + + g_assert_cmpint(bs_b->quiesce_counter, ==, 1); + g_assert_cmpint(b_s->drain_count, ==, 1); + + /* End the drained section */ + bdrv_drain_all_end(); + + g_assert_cmpint(bs_b->quiesce_counter, ==, 0); + g_assert_cmpint(b_s->drain_count, ==, 0); + + bdrv_unref(bs_b); + blk_unref(blk_b); +} + struct test_iothread_data { BlockDriverState *bs; enum drain_type drain_type; @@ -XXX,XX +XXX,XX @@ static void do_test_delete_by_drain(bool detach_instead_of_delete, bdrv_subtree_drained_begin(bs); bdrv_subtree_drained_end(bs); break; + case BDRV_DRAIN_ALL: + bdrv_drain_all_begin(); + bdrv_drain_all_end(); + break; default: g_assert_not_reached(); } @@ -XXX,XX +XXX,XX @@ static void test_delete_by_drain(void) do_test_delete_by_drain(false, BDRV_DRAIN); } +static void test_detach_by_drain_all(void) +{ + do_test_delete_by_drain(true, BDRV_DRAIN_ALL); +} + static void test_detach_by_drain(void) { do_test_delete_by_drain(true, BDRV_DRAIN); @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) g_test_add_func("/bdrv-drain/nested", test_nested); g_test_add_func("/bdrv-drain/multiparent", test_multiparent); - g_test_add_func("/bdrv-drain/graph-change", test_graph_change); + + g_test_add_func("/bdrv-drain/graph-change/drain_subtree", + test_graph_change_drain_subtree); + g_test_add_func("/bdrv-drain/graph-change/drain_all", + test_graph_change_drain_all); g_test_add_func("/bdrv-drain/iothread/drain_all", test_iothread_drain_all); g_test_add_func("/bdrv-drain/iothread/drain", test_iothread_drain); @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) test_blockjob_drain_subtree); g_test_add_func("/bdrv-drain/deletion/drain", test_delete_by_drain); + g_test_add_func("/bdrv-drain/detach/drain_all", test_detach_by_drain_all); g_test_add_func("/bdrv-drain/detach/drain", test_detach_by_drain); g_test_add_func("/bdrv-drain/detach/drain_subtree", test_detach_by_drain_subtree); g_test_add_func("/bdrv-drain/detach/parent_cb", test_detach_by_parent_cb); -- 2.13.6 From: Greg Kurz <groug@kaod.org> Removing a drive with drive_del while it is being used to run an I/O intensive workload can cause QEMU to crash. An AIO flush can yield at some point: blk_aio_flush_entry() blk_co_flush(blk) bdrv_co_flush(blk->root->bs) ... qemu_coroutine_yield() and let the HMP command to run, free blk->root and give control back to the AIO flush: hmp_drive_del() blk_remove_bs() bdrv_root_unref_child(blk->root) child_bs = blk->root->bs bdrv_detach_child(blk->root) bdrv_replace_child(blk->root, NULL) blk->root->bs = NULL g_free(blk->root) <============== blk->root becomes stale bdrv_unref(child_bs) bdrv_delete(child_bs) bdrv_close() bdrv_drained_begin() bdrv_do_drained_begin() bdrv_drain_recurse() aio_poll() ... qemu_coroutine_switch() and the AIO flush completion ends up dereferencing blk->root: blk_aio_complete() scsi_aio_complete() blk_get_aio_context(blk) bs = blk_bs(blk) ie, bs = blk->root ? blk->root->bs : NULL ^^^^^ stale The problem is that we should avoid making block driver graph changes while we have in-flight requests. Let's drain all I/O for this BB before calling bdrv_root_unref_child(). Signed-off-by: Greg Kurz <groug@kaod.org> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/block-backend.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/block/block-backend.c b/block/block-backend.c index XXXXXXX..XXXXXXX 100644 --- a/block/block-backend.c +++ b/block/block-backend.c @@ -XXX,XX +XXX,XX @@ void blk_remove_bs(BlockBackend *blk) blk_update_root_state(blk); + /* bdrv_root_unref_child() will cause blk->root to become stale and may + * switch to a completion coroutine later on. Let's drain all I/O here + * to avoid that and a potential QEMU crash. + */ + blk_drain(blk); bdrv_root_unref_child(blk->root); blk->root = NULL; } -- 2.13.6 From: Max Reitz <mreitz@redhat.com> When converting mirror's I/O to coroutines, we are going to need a point where these coroutines are created. mirror_perform() is going to be that point. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Reviewed-by: Jeff Cody <jcody@redhat.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Message-id: 20180613181823.13618-2-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- block/mirror.c | 51 +++++++++++++++++++++++++++++---------------------- 1 file changed, 29 insertions(+), 22 deletions(-) diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ typedef struct MirrorOp { uint64_t bytes; } MirrorOp; +typedef enum MirrorMethod { + MIRROR_METHOD_COPY, + MIRROR_METHOD_ZERO, + MIRROR_METHOD_DISCARD, +} MirrorMethod; + static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, int error) { @@ -XXX,XX +XXX,XX @@ static void mirror_do_zero_or_discard(MirrorBlockJob *s, } } +static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, + unsigned bytes, MirrorMethod mirror_method) +{ + switch (mirror_method) { + case MIRROR_METHOD_COPY: + return mirror_do_read(s, offset, bytes); + case MIRROR_METHOD_ZERO: + case MIRROR_METHOD_DISCARD: + mirror_do_zero_or_discard(s, offset, bytes, + mirror_method == MIRROR_METHOD_DISCARD); + return bytes; + default: + abort(); + } +} + static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) { BlockDriverState *source = s->source; @@ -XXX,XX +XXX,XX @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) int ret; int64_t io_bytes; int64_t io_bytes_acct; - enum MirrorMethod { - MIRROR_METHOD_COPY, - MIRROR_METHOD_ZERO, - MIRROR_METHOD_DISCARD - } mirror_method = MIRROR_METHOD_COPY; + MirrorMethod mirror_method = MIRROR_METHOD_COPY; assert(!(offset % s->granularity)); ret = bdrv_block_status_above(source, NULL, offset, @@ -XXX,XX +XXX,XX @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) } io_bytes = mirror_clip_bytes(s, offset, io_bytes); - switch (mirror_method) { - case MIRROR_METHOD_COPY: - io_bytes = io_bytes_acct = mirror_do_read(s, offset, io_bytes); - break; - case MIRROR_METHOD_ZERO: - case MIRROR_METHOD_DISCARD: - mirror_do_zero_or_discard(s, offset, io_bytes, - mirror_method == MIRROR_METHOD_DISCARD); - if (write_zeroes_ok) { - io_bytes_acct = 0; - } else { - io_bytes_acct = io_bytes; - } - break; - default: - abort(); + io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); + if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { + io_bytes_acct = 0; + } else { + io_bytes_acct = io_bytes; } assert(io_bytes); offset += io_bytes; @@ -XXX,XX +XXX,XX @@ static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) continue; } - mirror_do_zero_or_discard(s, offset, bytes, false); + mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); offset += bytes; } -- 2.13.6 From: Max Reitz <mreitz@redhat.com> In order to talk to the source BDS (and maybe in the future to the target BDS as well) directly, we need to convert our existing AIO requests into coroutine I/O requests. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Message-id: 20180613181823.13618-3-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- block/mirror.c | 152 ++++++++++++++++++++++++++++++++++----------------------- 1 file changed, 90 insertions(+), 62 deletions(-) diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ typedef struct MirrorOp { QEMUIOVector qiov; int64_t offset; uint64_t bytes; + + /* The pointee is set by mirror_co_read(), mirror_co_zero(), and + * mirror_co_discard() before yielding for the first time */ + int64_t *bytes_handled; } MirrorOp; typedef enum MirrorMethod { @@ -XXX,XX +XXX,XX @@ static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, } } -static void mirror_iteration_done(MirrorOp *op, int ret) +static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) { MirrorBlockJob *s = op->s; struct iovec *iov; @@ -XXX,XX +XXX,XX @@ static void mirror_iteration_done(MirrorOp *op, int ret) } } -static void mirror_write_complete(void *opaque, int ret) +static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) { - MirrorOp *op = opaque; MirrorBlockJob *s = op->s; aio_context_acquire(blk_get_aio_context(s->common.blk)); @@ -XXX,XX +XXX,XX @@ static void mirror_write_complete(void *opaque, int ret) aio_context_release(blk_get_aio_context(s->common.blk)); } -static void mirror_read_complete(void *opaque, int ret) +static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret) { - MirrorOp *op = opaque; MirrorBlockJob *s = op->s; aio_context_acquire(blk_get_aio_context(s->common.blk)); @@ -XXX,XX +XXX,XX @@ static void mirror_read_complete(void *opaque, int ret) mirror_iteration_done(op, ret); } else { - blk_aio_pwritev(s->target, op->offset, &op->qiov, - 0, mirror_write_complete, op); + ret = blk_co_pwritev(s->target, op->offset, + op->qiov.size, &op->qiov, 0); + mirror_write_complete(op, ret); } aio_context_release(blk_get_aio_context(s->common.blk)); } @@ -XXX,XX +XXX,XX @@ static inline void mirror_wait_for_io(MirrorBlockJob *s) s->waiting_for_io = false; } -/* Submit async read while handling COW. - * Returns: The number of bytes copied after and including offset, - * excluding any bytes copied prior to offset due to alignment. - * This will be @bytes if no alignment is necessary, or - * (new_end - offset) if tail is rounded up or down due to - * alignment or buffer limit. +/* Perform a mirror copy operation. + * + * *op->bytes_handled is set to the number of bytes copied after and + * including offset, excluding any bytes copied prior to offset due + * to alignment. This will be op->bytes if no alignment is necessary, + * or (new_end - op->offset) if the tail is rounded up or down due to + * alignment or buffer limit. */ -static uint64_t mirror_do_read(MirrorBlockJob *s, int64_t offset, - uint64_t bytes) +static void coroutine_fn mirror_co_read(void *opaque) { + MirrorOp *op = opaque; + MirrorBlockJob *s = op->s; BlockBackend *source = s->common.blk; int nb_chunks; uint64_t ret; - MirrorOp *op; uint64_t max_bytes; max_bytes = s->granularity * s->max_iov; /* We can only handle as much as buf_size at a time. */ - bytes = MIN(s->buf_size, MIN(max_bytes, bytes)); - assert(bytes); - assert(bytes < BDRV_REQUEST_MAX_BYTES); - ret = bytes; + op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes)); + assert(op->bytes); + assert(op->bytes < BDRV_REQUEST_MAX_BYTES); + *op->bytes_handled = op->bytes; if (s->cow_bitmap) { - ret += mirror_cow_align(s, &offset, &bytes); + *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes); } - assert(bytes <= s->buf_size); + /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */ + assert(*op->bytes_handled <= UINT_MAX); + assert(op->bytes <= s->buf_size); /* The offset is granularity-aligned because: * 1) Caller passes in aligned values; * 2) mirror_cow_align is used only when target cluster is larger. */ - assert(QEMU_IS_ALIGNED(offset, s->granularity)); + assert(QEMU_IS_ALIGNED(op->offset, s->granularity)); /* The range is sector-aligned, since bdrv_getlength() rounds up. */ - assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); - nb_chunks = DIV_ROUND_UP(bytes, s->granularity); + assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE)); + nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); while (s->buf_free_count < nb_chunks) { - trace_mirror_yield_in_flight(s, offset, s->in_flight); + trace_mirror_yield_in_flight(s, op->offset, s->in_flight); mirror_wait_for_io(s); } - /* Allocate a MirrorOp that is used as an AIO callback. */ - op = g_new(MirrorOp, 1); - op->s = s; - op->offset = offset; - op->bytes = bytes; - /* Now make a QEMUIOVector taking enough granularity-sized chunks * from s->buf_free. */ qemu_iovec_init(&op->qiov, nb_chunks); while (nb_chunks-- > 0) { MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); - size_t remaining = bytes - op->qiov.size; + size_t remaining = op->bytes - op->qiov.size; QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); s->buf_free_count--; @@ -XXX,XX +XXX,XX @@ static uint64_t mirror_do_read(MirrorBlockJob *s, int64_t offset, /* Copy the dirty cluster. */ s->in_flight++; - s->bytes_in_flight += bytes; - trace_mirror_one_iteration(s, offset, bytes); + s->bytes_in_flight += op->bytes; + trace_mirror_one_iteration(s, op->offset, op->bytes); - blk_aio_preadv(source, offset, &op->qiov, 0, mirror_read_complete, op); - return ret; + ret = blk_co_preadv(source, op->offset, op->bytes, &op->qiov, 0); + mirror_read_complete(op, ret); } -static void mirror_do_zero_or_discard(MirrorBlockJob *s, - int64_t offset, - uint64_t bytes, - bool is_discard) +static void coroutine_fn mirror_co_zero(void *opaque) { - MirrorOp *op; + MirrorOp *op = opaque; + int ret; - /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed - * so the freeing in mirror_iteration_done is nop. */ - op = g_new0(MirrorOp, 1); - op->s = s; - op->offset = offset; - op->bytes = bytes; + op->s->in_flight++; + op->s->bytes_in_flight += op->bytes; + *op->bytes_handled = op->bytes; - s->in_flight++; - s->bytes_in_flight += bytes; - if (is_discard) { - blk_aio_pdiscard(s->target, offset, - op->bytes, mirror_write_complete, op); - } else { - blk_aio_pwrite_zeroes(s->target, offset, - op->bytes, s->unmap ? BDRV_REQ_MAY_UNMAP : 0, - mirror_write_complete, op); - } + ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes, + op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0); + mirror_write_complete(op, ret); +} + +static void coroutine_fn mirror_co_discard(void *opaque) +{ + MirrorOp *op = opaque; + int ret; + + op->s->in_flight++; + op->s->bytes_in_flight += op->bytes; + *op->bytes_handled = op->bytes; + + ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes); + mirror_write_complete(op, ret); } static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, unsigned bytes, MirrorMethod mirror_method) { + MirrorOp *op; + Coroutine *co; + int64_t bytes_handled = -1; + + op = g_new(MirrorOp, 1); + *op = (MirrorOp){ + .s = s, + .offset = offset, + .bytes = bytes, + .bytes_handled = &bytes_handled, + }; + switch (mirror_method) { case MIRROR_METHOD_COPY: - return mirror_do_read(s, offset, bytes); + co = qemu_coroutine_create(mirror_co_read, op); + break; case MIRROR_METHOD_ZERO: + co = qemu_coroutine_create(mirror_co_zero, op); + break; case MIRROR_METHOD_DISCARD: - mirror_do_zero_or_discard(s, offset, bytes, - mirror_method == MIRROR_METHOD_DISCARD); - return bytes; + co = qemu_coroutine_create(mirror_co_discard, op); + break; default: abort(); } + + qemu_coroutine_enter(co); + /* At this point, ownership of op has been moved to the coroutine + * and the object may already be freed */ + + /* Assert that this value has been set */ + assert(bytes_handled >= 0); + + /* Same assertion as in mirror_co_read() (and for mirror_co_read() + * and mirror_co_discard(), bytes_handled == op->bytes, which + * is the @bytes parameter given to this function) */ + assert(bytes_handled <= UINT_MAX); + return bytes_handled; } static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) -- 2.13.6 From: Max Reitz <mreitz@redhat.com> Attach a CoQueue to each in-flight operation so if we need to wait for any we can use it to wait instead of just blindly yielding and hoping for some operation to wake us. A later patch will use this infrastructure to allow requests accessing the same area of the virtual disk to specifically wait for each other. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Message-id: 20180613181823.13618-4-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- block/mirror.c | 34 +++++++++++++++++++++++----------- 1 file changed, 23 insertions(+), 11 deletions(-) diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ #include "qemu/osdep.h" #include "qemu/cutils.h" +#include "qemu/coroutine.h" #include "trace.h" #include "block/blockjob_int.h" #include "block/block_int.h" @@ -XXX,XX +XXX,XX @@ typedef struct MirrorBuffer { QSIMPLEQ_ENTRY(MirrorBuffer) next; } MirrorBuffer; +typedef struct MirrorOp MirrorOp; + typedef struct MirrorBlockJob { BlockJob common; BlockBackend *target; @@ -XXX,XX +XXX,XX @@ typedef struct MirrorBlockJob { unsigned long *in_flight_bitmap; int in_flight; int64_t bytes_in_flight; + QTAILQ_HEAD(MirrorOpList, MirrorOp) ops_in_flight; int ret; bool unmap; - bool waiting_for_io; int target_cluster_size; int max_iov; bool initial_zeroing_ongoing; } MirrorBlockJob; -typedef struct MirrorOp { +struct MirrorOp { MirrorBlockJob *s; QEMUIOVector qiov; int64_t offset; @@ -XXX,XX +XXX,XX @@ typedef struct MirrorOp { /* The pointee is set by mirror_co_read(), mirror_co_zero(), and * mirror_co_discard() before yielding for the first time */ int64_t *bytes_handled; -} MirrorOp; + + CoQueue waiting_requests; + + QTAILQ_ENTRY(MirrorOp) next; +}; typedef enum MirrorMethod { MIRROR_METHOD_COPY, @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) chunk_num = op->offset / s->granularity; nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); + bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); + QTAILQ_REMOVE(&s->ops_in_flight, op, next); if (ret >= 0) { if (s->cow_bitmap) { bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) } } qemu_iovec_destroy(&op->qiov); - g_free(op); - if (s->waiting_for_io) { - qemu_coroutine_enter(s->common.job.co); - } + qemu_co_queue_restart_all(&op->waiting_requests); + g_free(op); } static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) @@ -XXX,XX +XXX,XX @@ static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, static inline void mirror_wait_for_io(MirrorBlockJob *s) { - assert(!s->waiting_for_io); - s->waiting_for_io = true; - qemu_coroutine_yield(); - s->waiting_for_io = false; + MirrorOp *op; + + op = QTAILQ_FIRST(&s->ops_in_flight); + assert(op); + qemu_co_queue_wait(&op->waiting_requests, NULL); } /* Perform a mirror copy operation. @@ -XXX,XX +XXX,XX @@ static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, .bytes = bytes, .bytes_handled = &bytes_handled, }; + qemu_co_queue_init(&op->waiting_requests); switch (mirror_method) { case MIRROR_METHOD_COPY: @@ -XXX,XX +XXX,XX @@ static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, abort(); } + QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); qemu_coroutine_enter(co); /* At this point, ownership of op has been moved to the coroutine * and the object may already be freed */ @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, } } + QTAILQ_INIT(&s->ops_in_flight); + trace_mirror_start(bs, s, opaque); job_start(&s->common.job); return; -- 2.13.6 From: Max Reitz <mreitz@redhat.com> This patch makes the mirror code differentiate between simply waiting for any operation to complete (mirror_wait_for_free_in_flight_slot()) and specifically waiting for all operations touching a certain range of the virtual disk to complete (mirror_wait_on_conflicts()). Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Message-id: 20180613181823.13618-5-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- block/mirror.c | 102 +++++++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 84 insertions(+), 18 deletions(-) diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ #include "qemu/osdep.h" #include "qemu/cutils.h" #include "qemu/coroutine.h" +#include "qemu/range.h" #include "trace.h" #include "block/blockjob_int.h" #include "block/block_int.h" @@ -XXX,XX +XXX,XX @@ struct MirrorOp { * mirror_co_discard() before yielding for the first time */ int64_t *bytes_handled; + bool is_pseudo_op; CoQueue waiting_requests; QTAILQ_ENTRY(MirrorOp) next; @@ -XXX,XX +XXX,XX @@ static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, } } +static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self, + MirrorBlockJob *s, + uint64_t offset, + uint64_t bytes) +{ + uint64_t self_start_chunk = offset / s->granularity; + uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); + uint64_t self_nb_chunks = self_end_chunk - self_start_chunk; + + while (find_next_bit(s->in_flight_bitmap, self_end_chunk, + self_start_chunk) < self_end_chunk && + s->ret >= 0) + { + MirrorOp *op; + + QTAILQ_FOREACH(op, &s->ops_in_flight, next) { + uint64_t op_start_chunk = op->offset / s->granularity; + uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes, + s->granularity) - + op_start_chunk; + + if (op == self) { + continue; + } + + if (ranges_overlap(self_start_chunk, self_nb_chunks, + op_start_chunk, op_nb_chunks)) + { + qemu_co_queue_wait(&op->waiting_requests, NULL); + break; + } + } + } +} + static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) { MirrorBlockJob *s = op->s; @@ -XXX,XX +XXX,XX @@ static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, return ret; } -static inline void mirror_wait_for_io(MirrorBlockJob *s) +static inline void mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) { MirrorOp *op; - op = QTAILQ_FIRST(&s->ops_in_flight); - assert(op); - qemu_co_queue_wait(&op->waiting_requests, NULL); + QTAILQ_FOREACH(op, &s->ops_in_flight, next) { + /* Do not wait on pseudo ops, because it may in turn wait on + * some other operation to start, which may in fact be the + * caller of this function. Since there is only one pseudo op + * at any given time, we will always find some real operation + * to wait on. */ + if (!op->is_pseudo_op) { + qemu_co_queue_wait(&op->waiting_requests, NULL); + return; + } + } + abort(); } /* Perform a mirror copy operation. @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_co_read(void *opaque) while (s->buf_free_count < nb_chunks) { trace_mirror_yield_in_flight(s, op->offset, s->in_flight); - mirror_wait_for_io(s); + mirror_wait_for_free_in_flight_slot(s); } /* Now make a QEMUIOVector taking enough granularity-sized chunks @@ -XXX,XX +XXX,XX @@ static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) { BlockDriverState *source = s->source; - int64_t offset, first_chunk; - uint64_t delay_ns = 0; + MirrorOp *pseudo_op; + int64_t offset; + uint64_t delay_ns = 0, ret = 0; /* At least the first dirty chunk is mirrored in one iteration. */ int nb_chunks = 1; bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); @@ -XXX,XX +XXX,XX @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) } bdrv_dirty_bitmap_unlock(s->dirty_bitmap); - first_chunk = offset / s->granularity; - while (test_bit(first_chunk, s->in_flight_bitmap)) { - trace_mirror_yield_in_flight(s, offset, s->in_flight); - mirror_wait_for_io(s); - } + mirror_wait_on_conflicts(NULL, s, offset, 1); job_pause_point(&s->common.job); @@ -XXX,XX +XXX,XX @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) nb_chunks * s->granularity); bdrv_dirty_bitmap_unlock(s->dirty_bitmap); + /* Before claiming an area in the in-flight bitmap, we have to + * create a MirrorOp for it so that conflicting requests can wait + * for it. mirror_perform() will create the real MirrorOps later, + * for now we just create a pseudo operation that will wake up all + * conflicting requests once all real operations have been + * launched. */ + pseudo_op = g_new(MirrorOp, 1); + *pseudo_op = (MirrorOp){ + .offset = offset, + .bytes = nb_chunks * s->granularity, + .is_pseudo_op = true, + }; + qemu_co_queue_init(&pseudo_op->waiting_requests); + QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next); + bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); while (nb_chunks > 0 && offset < s->bdev_length) { int ret; @@ -XXX,XX +XXX,XX @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) while (s->in_flight >= MAX_IN_FLIGHT) { trace_mirror_yield_in_flight(s, offset, s->in_flight); - mirror_wait_for_io(s); + mirror_wait_for_free_in_flight_slot(s); } if (s->ret < 0) { - return 0; + ret = 0; + goto fail; } io_bytes = mirror_clip_bytes(s, offset, io_bytes); @@ -XXX,XX +XXX,XX @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct); } - return delay_ns; + + ret = delay_ns; +fail: + QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next); + qemu_co_queue_restart_all(&pseudo_op->waiting_requests); + g_free(pseudo_op); + + return ret; } static void mirror_free_init(MirrorBlockJob *s) @@ -XXX,XX +XXX,XX @@ static void mirror_free_init(MirrorBlockJob *s) static void mirror_wait_for_all_io(MirrorBlockJob *s) { while (s->in_flight > 0) { - mirror_wait_for_io(s); + mirror_wait_for_free_in_flight_slot(s); } } @@ -XXX,XX +XXX,XX @@ static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) if (s->in_flight >= MAX_IN_FLIGHT) { trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, s->in_flight); - mirror_wait_for_io(s); + mirror_wait_for_free_in_flight_slot(s); continue; } @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_run(void *opaque) if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || (cnt == 0 && s->in_flight > 0)) { trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); - mirror_wait_for_io(s); + mirror_wait_for_free_in_flight_slot(s); continue; } else if (cnt != 0) { delay_ns = mirror_iteration(s); -- 2.13.6 From: Max Reitz <mreitz@redhat.com> With this, the mirror_top_bs is no longer just a technically required node in the BDS graph but actually represents the block job operation. Also, drop MirrorBlockJob.source, as we can reach it through mirror_top_bs->backing. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Message-id: 20180613181823.13618-6-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- block/mirror.c | 14 ++++++-------- 1 file changed, 6 insertions(+), 8 deletions(-) diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ typedef struct MirrorBlockJob { BlockJob common; BlockBackend *target; BlockDriverState *mirror_top_bs; - BlockDriverState *source; BlockDriverState *base; /* The name of the graph node to replace */ @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_co_read(void *opaque) { MirrorOp *op = opaque; MirrorBlockJob *s = op->s; - BlockBackend *source = s->common.blk; int nb_chunks; uint64_t ret; uint64_t max_bytes; @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_co_read(void *opaque) s->bytes_in_flight += op->bytes; trace_mirror_one_iteration(s, op->offset, op->bytes); - ret = blk_co_preadv(source, op->offset, op->bytes, &op->qiov, 0); + ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes, + &op->qiov, 0); mirror_read_complete(op, ret); } @@ -XXX,XX +XXX,XX @@ static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) { - BlockDriverState *source = s->source; + BlockDriverState *source = s->mirror_top_bs->backing->bs; MirrorOp *pseudo_op; int64_t offset; uint64_t delay_ns = 0, ret = 0; @@ -XXX,XX +XXX,XX @@ static void mirror_exit(Job *job, void *opaque) BlockJob *bjob = &s->common; MirrorExitData *data = opaque; AioContext *replace_aio_context = NULL; - BlockDriverState *src = s->source; + BlockDriverState *src = s->mirror_top_bs->backing->bs; BlockDriverState *target_bs = blk_bs(s->target); BlockDriverState *mirror_top_bs = s->mirror_top_bs; Error *local_err = NULL; @@ -XXX,XX +XXX,XX @@ static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) { int64_t offset; BlockDriverState *base = s->base; - BlockDriverState *bs = s->source; + BlockDriverState *bs = s->mirror_top_bs->backing->bs; BlockDriverState *target_bs = blk_bs(s->target); int ret; int64_t count; @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_run(void *opaque) { MirrorBlockJob *s = opaque; MirrorExitData *data; - BlockDriverState *bs = s->source; + BlockDriverState *bs = s->mirror_top_bs->backing->bs; BlockDriverState *target_bs = blk_bs(s->target); bool need_drain = true; int64_t length; @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, /* The block job now has a reference to this node */ bdrv_unref(mirror_top_bs); - s->source = bs; s->mirror_top_bs = mirror_top_bs; /* No resize for the target either; while the mirror is still running, a -- 2.13.6 From: Max Reitz <mreitz@redhat.com> Currently, bdrv_replace_node() refuses to create loops from one BDS to itself if the BDS to be replaced is the backing node of the BDS to replace it: Say there is a node A and a node B. Replacing B by A means making all references to B point to A. If B is a child of A (i.e. A has a reference to B), that would mean we would have to make this reference point to A itself -- so we'd create a loop. bdrv_replace_node() (through should_update_child()) refuses to do so if B is the backing node of A. There is no reason why we should create loops if B is not the backing node of A, though. The BDS graph should never contain loops, so we should always refuse to create them. If B is a child of A and B is to be replaced by A, we should simply leave B in place there because it is the most sensible choice. A more specific argument would be: Putting filter drivers into the BDS graph is basically the same as appending an overlay to a backing chain. But the main child BDS of a filter driver is not "backing" but "file", so restricting the no-loop rule to backing nodes would fail here. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Message-id: 20180613181823.13618-7-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- block.c | 44 ++++++++++++++++++++++++++++++++++---------- 1 file changed, 34 insertions(+), 10 deletions(-) diff --git a/block.c b/block.c index XXXXXXX..XXXXXXX 100644 --- a/block.c +++ b/block.c @@ -XXX,XX +XXX,XX @@ static bool should_update_child(BdrvChild *c, BlockDriverState *to) return false; } - if (c->role == &child_backing) { - /* If @from is a backing file of @to, ignore the child to avoid - * creating a loop. We only want to change the pointer of other - * parents. */ - QLIST_FOREACH(to_c, &to->children, next) { - if (to_c == c) { - break; - } - } - if (to_c) { + /* If the child @c belongs to the BDS @to, replacing the current + * c->bs by @to would mean to create a loop. + * + * Such a case occurs when appending a BDS to a backing chain. + * For instance, imagine the following chain: + * + * guest device -> node A -> further backing chain... + * + * Now we create a new BDS B which we want to put on top of this + * chain, so we first attach A as its backing node: + * + * node B + * | + * v + * guest device -> node A -> further backing chain... + * + * Finally we want to replace A by B. When doing that, we want to + * replace all pointers to A by pointers to B -- except for the + * pointer from B because (1) that would create a loop, and (2) + * that pointer should simply stay intact: + * + * guest device -> node B + * | + * v + * node A -> further backing chain... + * + * In general, when replacing a node A (c->bs) by a node B (@to), + * if A is a child of B, that means we cannot replace A by B there + * because that would create a loop. Silently detaching A from B + * is also not really an option. So overall just leaving A in + * place there is the most sensible choice. */ + QLIST_FOREACH(to_c, &to->children, next) { + if (to_c == c) { return false; } } @@ -XXX,XX +XXX,XX @@ void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, /* Put all parents into @list and calculate their cumulative permissions */ QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { + assert(c->bs == from); if (!should_update_child(c, to)) { continue; } -- 2.13.6 From: Max Reitz <mreitz@redhat.com> This new parameter allows the caller to just query the next dirty position without moving the iterator. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Reviewed-by: John Snow <jsnow@redhat.com> Message-id: 20180613181823.13618-8-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- include/qemu/hbitmap.h | 5 ++++- block/backup.c | 2 +- block/dirty-bitmap.c | 2 +- tests/test-hbitmap.c | 26 +++++++++++++------------- util/hbitmap.c | 10 +++++++--- 5 files changed, 26 insertions(+), 19 deletions(-) diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h index XXXXXXX..XXXXXXX 100644 --- a/include/qemu/hbitmap.h +++ b/include/qemu/hbitmap.h @@ -XXX,XX +XXX,XX @@ void hbitmap_free_meta(HBitmap *hb); /** * hbitmap_iter_next: * @hbi: HBitmapIter to operate on. + * @advance: If true, advance the iterator. Otherwise, the next call + * of this function will return the same result (if that + * position is still dirty). * * Return the next bit that is set in @hbi's associated HBitmap, * or -1 if all remaining bits are zero. */ -int64_t hbitmap_iter_next(HBitmapIter *hbi); +int64_t hbitmap_iter_next(HBitmapIter *hbi, bool advance); /** * hbitmap_iter_next_word: diff --git a/block/backup.c b/block/backup.c index XXXXXXX..XXXXXXX 100644 --- a/block/backup.c +++ b/block/backup.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn backup_run_incremental(BackupBlockJob *job) HBitmapIter hbi; hbitmap_iter_init(&hbi, job->copy_bitmap, 0); - while ((cluster = hbitmap_iter_next(&hbi)) != -1) { + while ((cluster = hbitmap_iter_next(&hbi, true)) != -1) { do { if (yield_and_check(job)) { return 0; diff --git a/block/dirty-bitmap.c b/block/dirty-bitmap.c index XXXXXXX..XXXXXXX 100644 --- a/block/dirty-bitmap.c +++ b/block/dirty-bitmap.c @@ -XXX,XX +XXX,XX @@ void bdrv_dirty_iter_free(BdrvDirtyBitmapIter *iter) int64_t bdrv_dirty_iter_next(BdrvDirtyBitmapIter *iter) { - return hbitmap_iter_next(&iter->hbi); + return hbitmap_iter_next(&iter->hbi, true); } /* Called within bdrv_dirty_bitmap_lock..unlock */ diff --git a/tests/test-hbitmap.c b/tests/test-hbitmap.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-hbitmap.c +++ b/tests/test-hbitmap.c @@ -XXX,XX +XXX,XX @@ static void hbitmap_test_check(TestHBitmapData *data, i = first; for (;;) { - next = hbitmap_iter_next(&hbi); + next = hbitmap_iter_next(&hbi, true); if (next < 0) { next = data->size; } @@ -XXX,XX +XXX,XX @@ static void test_hbitmap_iter_granularity(TestHBitmapData *data, /* Note that hbitmap_test_check has to be invoked manually in this test. */ hbitmap_test_init(data, 131072 << 7, 7); hbitmap_iter_init(&hbi, data->hb, 0); - g_assert_cmpint(hbitmap_iter_next(&hbi), <, 0); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); hbitmap_test_set(data, ((L2 + L1 + 1) << 7) + 8, 8); hbitmap_iter_init(&hbi, data->hb, 0); - g_assert_cmpint(hbitmap_iter_next(&hbi), ==, (L2 + L1 + 1) << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi), <, 0); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, (L2 + L1 + 1) << 7); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); hbitmap_iter_init(&hbi, data->hb, (L2 + L1 + 2) << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi), <, 0); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); hbitmap_test_set(data, (131072 << 7) - 8, 8); hbitmap_iter_init(&hbi, data->hb, 0); - g_assert_cmpint(hbitmap_iter_next(&hbi), ==, (L2 + L1 + 1) << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi), ==, 131071 << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi), <, 0); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, (L2 + L1 + 1) << 7); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, 131071 << 7); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); hbitmap_iter_init(&hbi, data->hb, (L2 + L1 + 2) << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi), ==, 131071 << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi), <, 0); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, 131071 << 7); + g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); } static void hbitmap_test_set_boundary_bits(TestHBitmapData *data, ssize_t diff) @@ -XXX,XX +XXX,XX @@ static void test_hbitmap_serialize_zeroes(TestHBitmapData *data, for (i = 0; i < num_positions; i++) { hbitmap_deserialize_zeroes(data->hb, positions[i], min_l1, true); hbitmap_iter_init(&iter, data->hb, 0); - next = hbitmap_iter_next(&iter); + next = hbitmap_iter_next(&iter, true); if (i == num_positions - 1) { g_assert_cmpint(next, ==, -1); } else { @@ -XXX,XX +XXX,XX @@ static void test_hbitmap_iter_and_reset(TestHBitmapData *data, hbitmap_iter_init(&hbi, data->hb, BITS_PER_LONG - 1); - hbitmap_iter_next(&hbi); + hbitmap_iter_next(&hbi, true); hbitmap_reset_all(data->hb); - hbitmap_iter_next(&hbi); + hbitmap_iter_next(&hbi, true); } static void test_hbitmap_next_zero_check(TestHBitmapData *data, int64_t start) diff --git a/util/hbitmap.c b/util/hbitmap.c index XXXXXXX..XXXXXXX 100644 --- a/util/hbitmap.c +++ b/util/hbitmap.c @@ -XXX,XX +XXX,XX @@ unsigned long hbitmap_iter_skip_words(HBitmapIter *hbi) return cur; } -int64_t hbitmap_iter_next(HBitmapIter *hbi) +int64_t hbitmap_iter_next(HBitmapIter *hbi, bool advance) { unsigned long cur = hbi->cur[HBITMAP_LEVELS - 1] & hbi->hb->levels[HBITMAP_LEVELS - 1][hbi->pos]; @@ -XXX,XX +XXX,XX @@ int64_t hbitmap_iter_next(HBitmapIter *hbi) } } - /* The next call will resume work from the next bit. */ - hbi->cur[HBITMAP_LEVELS - 1] = cur & (cur - 1); + if (advance) { + /* The next call will resume work from the next bit. */ + hbi->cur[HBITMAP_LEVELS - 1] = cur & (cur - 1); + } else { + hbi->cur[HBITMAP_LEVELS - 1] = cur; + } item = ((uint64_t)hbi->pos << BITS_PER_LEVEL) + ctzl(cur); return item << hbi->granularity; -- 2.13.6 From: Max Reitz <mreitz@redhat.com> Add a function that wraps hbitmap_iter_next() and always calls it in non-advancing mode first, and in advancing mode next. The result should always be the same. By using this function everywhere we called hbitmap_iter_next() before, we should get good test coverage for non-advancing hbitmap_iter_next(). Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Reviewed-by: John Snow <jsnow@redhat.com> Message-id: 20180613181823.13618-9-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- tests/test-hbitmap.c | 36 ++++++++++++++++++++++++------------ 1 file changed, 24 insertions(+), 12 deletions(-) diff --git a/tests/test-hbitmap.c b/tests/test-hbitmap.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-hbitmap.c +++ b/tests/test-hbitmap.c @@ -XXX,XX +XXX,XX @@ typedef struct TestHBitmapData { } TestHBitmapData; +static int64_t check_hbitmap_iter_next(HBitmapIter *hbi) +{ + int next0, next1; + + next0 = hbitmap_iter_next(hbi, false); + next1 = hbitmap_iter_next(hbi, true); + + g_assert_cmpint(next0, ==, next1); + + return next0; +} + /* Check that the HBitmap and the shadow bitmap contain the same data, * ignoring the same "first" bits. */ @@ -XXX,XX +XXX,XX @@ static void hbitmap_test_check(TestHBitmapData *data, i = first; for (;;) { - next = hbitmap_iter_next(&hbi, true); + next = check_hbitmap_iter_next(&hbi); if (next < 0) { next = data->size; } @@ -XXX,XX +XXX,XX @@ static void test_hbitmap_iter_granularity(TestHBitmapData *data, /* Note that hbitmap_test_check has to be invoked manually in this test. */ hbitmap_test_init(data, 131072 << 7, 7); hbitmap_iter_init(&hbi, data->hb, 0); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), <, 0); hbitmap_test_set(data, ((L2 + L1 + 1) << 7) + 8, 8); hbitmap_iter_init(&hbi, data->hb, 0); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, (L2 + L1 + 1) << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), ==, (L2 + L1 + 1) << 7); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), <, 0); hbitmap_iter_init(&hbi, data->hb, (L2 + L1 + 2) << 7); g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); hbitmap_test_set(data, (131072 << 7) - 8, 8); hbitmap_iter_init(&hbi, data->hb, 0); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, (L2 + L1 + 1) << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, 131071 << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), ==, (L2 + L1 + 1) << 7); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), ==, 131071 << 7); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), <, 0); hbitmap_iter_init(&hbi, data->hb, (L2 + L1 + 2) << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), ==, 131071 << 7); - g_assert_cmpint(hbitmap_iter_next(&hbi, true), <, 0); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), ==, 131071 << 7); + g_assert_cmpint(check_hbitmap_iter_next(&hbi), <, 0); } static void hbitmap_test_set_boundary_bits(TestHBitmapData *data, ssize_t diff) @@ -XXX,XX +XXX,XX @@ static void test_hbitmap_serialize_zeroes(TestHBitmapData *data, for (i = 0; i < num_positions; i++) { hbitmap_deserialize_zeroes(data->hb, positions[i], min_l1, true); hbitmap_iter_init(&iter, data->hb, 0); - next = hbitmap_iter_next(&iter, true); + next = check_hbitmap_iter_next(&iter); if (i == num_positions - 1) { g_assert_cmpint(next, ==, -1); } else { @@ -XXX,XX +XXX,XX @@ static void test_hbitmap_iter_and_reset(TestHBitmapData *data, hbitmap_iter_init(&hbi, data->hb, BITS_PER_LONG - 1); - hbitmap_iter_next(&hbi, true); + check_hbitmap_iter_next(&hbi); hbitmap_reset_all(data->hb); - hbitmap_iter_next(&hbi, true); + check_hbitmap_iter_next(&hbi); } static void test_hbitmap_next_zero_check(TestHBitmapData *data, int64_t start) -- 2.13.6 From: Max Reitz <mreitz@redhat.com> This new function allows to look for a consecutively dirty area in a dirty bitmap. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Reviewed-by: John Snow <jsnow@redhat.com> Message-id: 20180613181823.13618-10-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- include/block/dirty-bitmap.h | 2 ++ block/dirty-bitmap.c | 55 ++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 57 insertions(+) diff --git a/include/block/dirty-bitmap.h b/include/block/dirty-bitmap.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/dirty-bitmap.h +++ b/include/block/dirty-bitmap.h @@ -XXX,XX +XXX,XX @@ void bdrv_set_dirty_bitmap_locked(BdrvDirtyBitmap *bitmap, void bdrv_reset_dirty_bitmap_locked(BdrvDirtyBitmap *bitmap, int64_t offset, int64_t bytes); int64_t bdrv_dirty_iter_next(BdrvDirtyBitmapIter *iter); +bool bdrv_dirty_iter_next_area(BdrvDirtyBitmapIter *iter, uint64_t max_offset, + uint64_t *offset, int *bytes); void bdrv_set_dirty_iter(BdrvDirtyBitmapIter *hbi, int64_t offset); int64_t bdrv_get_dirty_count(BdrvDirtyBitmap *bitmap); int64_t bdrv_get_meta_dirty_count(BdrvDirtyBitmap *bitmap); diff --git a/block/dirty-bitmap.c b/block/dirty-bitmap.c index XXXXXXX..XXXXXXX 100644 --- a/block/dirty-bitmap.c +++ b/block/dirty-bitmap.c @@ -XXX,XX +XXX,XX @@ int64_t bdrv_dirty_iter_next(BdrvDirtyBitmapIter *iter) return hbitmap_iter_next(&iter->hbi, true); } +/** + * Return the next consecutively dirty area in the dirty bitmap + * belonging to the given iterator @iter. + * + * @max_offset: Maximum value that may be returned for + * *offset + *bytes + * @offset: Will contain the start offset of the next dirty area + * @bytes: Will contain the length of the next dirty area + * + * Returns: True if a dirty area could be found before max_offset + * (which means that *offset and *bytes then contain valid + * values), false otherwise. + * + * Note that @iter is never advanced if false is returned. If an area + * is found (which means that true is returned), it will be advanced + * past that area. + */ +bool bdrv_dirty_iter_next_area(BdrvDirtyBitmapIter *iter, uint64_t max_offset, + uint64_t *offset, int *bytes) +{ + uint32_t granularity = bdrv_dirty_bitmap_granularity(iter->bitmap); + uint64_t gran_max_offset; + int64_t ret; + int size; + + if (max_offset == iter->bitmap->size) { + /* If max_offset points to the image end, round it up by the + * bitmap granularity */ + gran_max_offset = ROUND_UP(max_offset, granularity); + } else { + gran_max_offset = max_offset; + } + + ret = hbitmap_iter_next(&iter->hbi, false); + if (ret < 0 || ret + granularity > gran_max_offset) { + return false; + } + + *offset = ret; + size = 0; + + assert(granularity <= INT_MAX); + + do { + /* Advance iterator */ + ret = hbitmap_iter_next(&iter->hbi, true); + size += granularity; + } while (ret + granularity <= gran_max_offset && + hbitmap_iter_next(&iter->hbi, false) == ret + granularity && + size <= INT_MAX - granularity); + + *bytes = MIN(size, max_offset - *offset); + return true; +} + /* Called within bdrv_dirty_bitmap_lock..unlock */ void bdrv_set_dirty_bitmap_locked(BdrvDirtyBitmap *bitmap, int64_t offset, int64_t bytes) -- 2.13.6 From: Max Reitz <mreitz@redhat.com> This will allow us to access the block job data when the mirror block driver becomes more complex. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Message-id: 20180613181823.13618-11-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- block/mirror.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ typedef struct MirrorBlockJob { bool initial_zeroing_ongoing; } MirrorBlockJob; +typedef struct MirrorBDSOpaque { + MirrorBlockJob *job; +} MirrorBDSOpaque; + struct MirrorOp { MirrorBlockJob *s; QEMUIOVector qiov; @@ -XXX,XX +XXX,XX @@ static void mirror_exit(Job *job, void *opaque) MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); BlockJob *bjob = &s->common; MirrorExitData *data = opaque; + MirrorBDSOpaque *bs_opaque = s->mirror_top_bs->opaque; AioContext *replace_aio_context = NULL; BlockDriverState *src = s->mirror_top_bs->backing->bs; BlockDriverState *target_bs = blk_bs(s->target); @@ -XXX,XX +XXX,XX @@ static void mirror_exit(Job *job, void *opaque) blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); blk_insert_bs(bjob->blk, mirror_top_bs, &error_abort); + bs_opaque->job = NULL; job_completed(job, data->ret, NULL); g_free(data); @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, Error **errp) { MirrorBlockJob *s; + MirrorBDSOpaque *bs_opaque; BlockDriverState *mirror_top_bs; bool target_graph_mod; bool target_is_backing; @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, mirror_top_bs->total_sectors = bs->total_sectors; mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED; + bs_opaque = g_new0(MirrorBDSOpaque, 1); + mirror_top_bs->opaque = bs_opaque; bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, if (!s) { goto fail; } + bs_opaque->job = s; + /* The block job now has a reference to this node */ bdrv_unref(mirror_top_bs); @@ -XXX,XX +XXX,XX @@ fail: g_free(s->replaces); blk_unref(s->target); + bs_opaque->job = NULL; job_early_fail(&s->common.job); } -- 2.13.6 From: Max Reitz <mreitz@redhat.com> Signed-off-by: Max Reitz <mreitz@redhat.com> Message-id: 20180613181823.13618-12-mreitz@redhat.com Reviewed-by: Kevin Wolf <kwolf@redhat.com> Signed-off-by: Max Reitz <mreitz@redhat.com> --- include/qemu/job.h | 15 +++++++++++++++ job.c | 5 +++++ 2 files changed, 20 insertions(+) diff --git a/include/qemu/job.h b/include/qemu/job.h index XXXXXXX..XXXXXXX 100644 --- a/include/qemu/job.h +++ b/include/qemu/job.h @@ -XXX,XX +XXX,XX @@ void job_progress_update(Job *job, uint64_t done); */ void job_progress_set_remaining(Job *job, uint64_t remaining); +/** + * @job: The job whose expected progress end value is updated + * @delta: Value which is to be added to the current expected end + * value + * + * Increases the expected end value of the progress counter of a job. + * This is useful for parenthesis operations: If a job has to + * conditionally perform a high-priority operation as part of its + * progress, it calls this function with the expected operation's + * length before, and job_progress_update() afterwards. + * (So the operation acts as a parenthesis in regards to the main job + * operation running in background.) + */ +void job_progress_increase_remaining(Job *job, uint64_t delta); + /** To be called when a cancelled job is finalised. */ void job_event_cancelled(Job *job); diff --git a/job.c b/job.c index XXXXXXX..XXXXXXX 100644 --- a/job.c +++ b/job.c @@ -XXX,XX +XXX,XX @@ void job_progress_set_remaining(Job *job, uint64_t remaining) job->progress_total = job->progress_current + remaining; } +void job_progress_increase_remaining(Job *job, uint64_t delta) +{ + job->progress_total += delta; +} + void job_event_cancelled(Job *job) { notifier_list_notify(&job->on_finalize_cancelled, job); -- 2.13.6 From: Max Reitz <mreitz@redhat.com> This patch implements active synchronous mirroring. In active mode, the passive mechanism will still be in place and is used to copy all initially dirty clusters off the source disk; but every write request will write data both to the source and the target disk, so the source cannot be dirtied faster than data is mirrored to the target. Also, once the block job has converged (BLOCK_JOB_READY sent), source and target are guaranteed to stay in sync (unless an error occurs). Active mode is completely optional and currently disabled at runtime. A later patch will add a way for users to enable it. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Message-id: 20180613181823.13618-13-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- qapi/block-core.json | 18 ++++ block/mirror.c | 252 ++++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 265 insertions(+), 5 deletions(-) diff --git a/qapi/block-core.json b/qapi/block-core.json index XXXXXXX..XXXXXXX 100644 --- a/qapi/block-core.json +++ b/qapi/block-core.json @@ -XXX,XX +XXX,XX @@ 'data': ['top', 'full', 'none', 'incremental'] } ## +# @MirrorCopyMode: +# +# An enumeration whose values tell the mirror block job when to +# trigger writes to the target. +# +# @background: copy data in background only. +# +# @write-blocking: when data is written to the source, write it +# (synchronously) to the target as well. In +# addition, data is copied in background just like in +# @background mode. +# +# Since: 3.0 +## +{ 'enum': 'MirrorCopyMode', + 'data': ['background', 'write-blocking'] } + +## # @BlockJobInfo: # # Information about a long-running block device operation. diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ typedef struct MirrorBlockJob { Error *replace_blocker; bool is_none_mode; BlockMirrorBackingMode backing_mode; + MirrorCopyMode copy_mode; BlockdevOnError on_source_error, on_target_error; bool synced; + /* Set when the target is synced (dirty bitmap is clean, nothing + * in flight) and the job is running in active mode */ + bool actively_synced; bool should_complete; int64_t granularity; size_t buf_size; @@ -XXX,XX +XXX,XX @@ typedef struct MirrorBlockJob { int target_cluster_size; int max_iov; bool initial_zeroing_ongoing; + int in_active_write_counter; } MirrorBlockJob; typedef struct MirrorBDSOpaque { @@ -XXX,XX +XXX,XX @@ struct MirrorOp { int64_t *bytes_handled; bool is_pseudo_op; + bool is_active_write; CoQueue waiting_requests; QTAILQ_ENTRY(MirrorOp) next; @@ -XXX,XX +XXX,XX @@ static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, int error) { s->synced = false; + s->actively_synced = false; if (read) { return block_job_error_action(&s->common, s->on_source_error, true, error); @@ -XXX,XX +XXX,XX @@ static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, return ret; } -static inline void mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) +static inline void mirror_wait_for_any_operation(MirrorBlockJob *s, bool active) { MirrorOp *op; @@ -XXX,XX +XXX,XX @@ static inline void mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) * caller of this function. Since there is only one pseudo op * at any given time, we will always find some real operation * to wait on. */ - if (!op->is_pseudo_op) { + if (!op->is_pseudo_op && op->is_active_write == active) { qemu_co_queue_wait(&op->waiting_requests, NULL); return; } @@ -XXX,XX +XXX,XX @@ static inline void mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) abort(); } +static inline void mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) +{ + /* Only non-active operations use up in-flight slots */ + mirror_wait_for_any_operation(s, false); +} + /* Perform a mirror copy operation. * * *op->bytes_handled is set to the number of bytes copied after and @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_run(void *opaque) /* Transition to the READY state and wait for complete. */ job_transition_to_ready(&s->common.job); s->synced = true; + s->actively_synced = true; while (!job_is_cancelled(&s->common.job) && !s->should_complete) { job_yield(&s->common.job); } @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_run(void *opaque) int64_t cnt, delta; bool should_complete; + /* Do not start passive operations while there are active + * writes in progress */ + while (s->in_active_write_counter) { + mirror_wait_for_any_operation(s, true); + } + if (s->ret < 0) { ret = s->ret; goto immediate_exit; @@ -XXX,XX +XXX,XX @@ static void coroutine_fn mirror_run(void *opaque) */ job_transition_to_ready(&s->common.job); s->synced = true; + if (s->copy_mode != MIRROR_COPY_MODE_BACKGROUND) { + s->actively_synced = true; + } } should_complete = s->should_complete || @@ -XXX,XX +XXX,XX @@ static const BlockJobDriver commit_active_job_driver = { .drain = mirror_drain, }; +static void do_sync_target_write(MirrorBlockJob *job, MirrorMethod method, + uint64_t offset, uint64_t bytes, + QEMUIOVector *qiov, int flags) +{ + BdrvDirtyBitmapIter *iter; + QEMUIOVector target_qiov; + uint64_t dirty_offset; + int dirty_bytes; + + if (qiov) { + qemu_iovec_init(&target_qiov, qiov->niov); + } + + iter = bdrv_dirty_iter_new(job->dirty_bitmap); + bdrv_set_dirty_iter(iter, offset); + + while (true) { + bool valid_area; + int ret; + + bdrv_dirty_bitmap_lock(job->dirty_bitmap); + valid_area = bdrv_dirty_iter_next_area(iter, offset + bytes, + &dirty_offset, &dirty_bytes); + if (!valid_area) { + bdrv_dirty_bitmap_unlock(job->dirty_bitmap); + break; + } + + bdrv_reset_dirty_bitmap_locked(job->dirty_bitmap, + dirty_offset, dirty_bytes); + bdrv_dirty_bitmap_unlock(job->dirty_bitmap); + + job_progress_increase_remaining(&job->common.job, dirty_bytes); + + assert(dirty_offset - offset <= SIZE_MAX); + if (qiov) { + qemu_iovec_reset(&target_qiov); + qemu_iovec_concat(&target_qiov, qiov, + dirty_offset - offset, dirty_bytes); + } + + switch (method) { + case MIRROR_METHOD_COPY: + ret = blk_co_pwritev(job->target, dirty_offset, dirty_bytes, + qiov ? &target_qiov : NULL, flags); + break; + + case MIRROR_METHOD_ZERO: + assert(!qiov); + ret = blk_co_pwrite_zeroes(job->target, dirty_offset, dirty_bytes, + flags); + break; + + case MIRROR_METHOD_DISCARD: + assert(!qiov); + ret = blk_co_pdiscard(job->target, dirty_offset, dirty_bytes); + break; + + default: + abort(); + } + + if (ret >= 0) { + job_progress_update(&job->common.job, dirty_bytes); + } else { + BlockErrorAction action; + + bdrv_set_dirty_bitmap(job->dirty_bitmap, dirty_offset, dirty_bytes); + job->actively_synced = false; + + action = mirror_error_action(job, false, -ret); + if (action == BLOCK_ERROR_ACTION_REPORT) { + if (!job->ret) { + job->ret = ret; + } + break; + } + } + } + + bdrv_dirty_iter_free(iter); + if (qiov) { + qemu_iovec_destroy(&target_qiov); + } +} + +static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s, + uint64_t offset, + uint64_t bytes) +{ + MirrorOp *op; + uint64_t start_chunk = offset / s->granularity; + uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); + + op = g_new(MirrorOp, 1); + *op = (MirrorOp){ + .s = s, + .offset = offset, + .bytes = bytes, + .is_active_write = true, + }; + qemu_co_queue_init(&op->waiting_requests); + QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); + + s->in_active_write_counter++; + + mirror_wait_on_conflicts(op, s, offset, bytes); + + bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); + + return op; +} + +static void coroutine_fn active_write_settle(MirrorOp *op) +{ + uint64_t start_chunk = op->offset / op->s->granularity; + uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes, + op->s->granularity); + + if (!--op->s->in_active_write_counter && op->s->actively_synced) { + BdrvChild *source = op->s->mirror_top_bs->backing; + + if (QLIST_FIRST(&source->bs->parents) == source && + QLIST_NEXT(source, next_parent) == NULL) + { + /* Assert that we are back in sync once all active write + * operations are settled. + * Note that we can only assert this if the mirror node + * is the source node's only parent. */ + assert(!bdrv_get_dirty_count(op->s->dirty_bitmap)); + } + } + bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); + QTAILQ_REMOVE(&op->s->ops_in_flight, op, next); + qemu_co_queue_restart_all(&op->waiting_requests); + g_free(op); +} + static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) { return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); } +static int coroutine_fn bdrv_mirror_top_do_write(BlockDriverState *bs, + MirrorMethod method, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, + int flags) +{ + MirrorOp *op = NULL; + MirrorBDSOpaque *s = bs->opaque; + int ret = 0; + bool copy_to_target; + + copy_to_target = s->job->ret >= 0 && + s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; + + if (copy_to_target) { + op = active_write_prepare(s->job, offset, bytes); + } + + switch (method) { + case MIRROR_METHOD_COPY: + ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); + break; + + case MIRROR_METHOD_ZERO: + ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); + break; + + case MIRROR_METHOD_DISCARD: + ret = bdrv_co_pdiscard(bs->backing->bs, offset, bytes); + break; + + default: + abort(); + } + + if (ret < 0) { + goto out; + } + + if (copy_to_target) { + do_sync_target_write(s->job, method, offset, bytes, qiov, flags); + } + +out: + if (copy_to_target) { + active_write_settle(op); + } + return ret; +} + static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) { - return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); + MirrorBDSOpaque *s = bs->opaque; + QEMUIOVector bounce_qiov; + void *bounce_buf; + int ret = 0; + bool copy_to_target; + + copy_to_target = s->job->ret >= 0 && + s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; + + if (copy_to_target) { + /* The guest might concurrently modify the data to write; but + * the data on source and destination must match, so we have + * to use a bounce buffer if we are going to write to the + * target now. */ + bounce_buf = qemu_blockalign(bs, bytes); + iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes); + + qemu_iovec_init(&bounce_qiov, 1); + qemu_iovec_add(&bounce_qiov, bounce_buf, bytes); + qiov = &bounce_qiov; + } + + ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, offset, bytes, qiov, + flags); + + if (copy_to_target) { + qemu_iovec_destroy(&bounce_qiov); + qemu_vfree(bounce_buf); + } + + return ret; } static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) @@ -XXX,XX +XXX,XX @@ static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, BdrvRequestFlags flags) { - return bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); + return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, offset, bytes, NULL, + flags); } static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) { - return bdrv_co_pdiscard(bs->backing->bs, offset, bytes); + return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes, + NULL, 0); } static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, s->on_target_error = on_target_error; s->is_none_mode = is_none_mode; s->backing_mode = backing_mode; + s->copy_mode = MIRROR_COPY_MODE_BACKGROUND; s->base = base; s->granularity = granularity; s->buf_size = ROUND_UP(buf_size, granularity); -- 2.13.6 From: Max Reitz <mreitz@redhat.com> This patch allows the user to specify whether to use active or only background mode for mirror block jobs. Currently, this setting will remain constant for the duration of the entire block job. Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Message-id: 20180613181823.13618-14-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- qapi/block-core.json | 11 +++++++++-- include/block/block_int.h | 4 +++- block/mirror.c | 12 +++++++----- blockdev.c | 9 ++++++++- 4 files changed, 27 insertions(+), 9 deletions(-) diff --git a/qapi/block-core.json b/qapi/block-core.json index XXXXXXX..XXXXXXX 100644 --- a/qapi/block-core.json +++ b/qapi/block-core.json @@ -XXX,XX +XXX,XX @@ # written. Both will result in identical contents. # Default is true. (Since 2.4) # +# @copy-mode: when to copy data to the destination; defaults to 'background' +# (Since: 3.0) +# # Since: 1.3 ## { 'struct': 'DriveMirror', @@ -XXX,XX +XXX,XX @@ '*speed': 'int', '*granularity': 'uint32', '*buf-size': 'int', '*on-source-error': 'BlockdevOnError', '*on-target-error': 'BlockdevOnError', - '*unmap': 'bool' } } + '*unmap': 'bool', '*copy-mode': 'MirrorCopyMode' } } ## # @BlockDirtyBitmap: @@ -XXX,XX +XXX,XX @@ # above @device. If this option is not given, a node name is # autogenerated. (Since: 2.9) # +# @copy-mode: when to copy data to the destination; defaults to 'background' +# (Since: 3.0) +# # Returns: nothing on success. # # Since: 2.6 @@ -XXX,XX +XXX,XX @@ '*speed': 'int', '*granularity': 'uint32', '*buf-size': 'int', '*on-source-error': 'BlockdevOnError', '*on-target-error': 'BlockdevOnError', - '*filter-node-name': 'str' } } + '*filter-node-name': 'str', + '*copy-mode': 'MirrorCopyMode' } } ## # @block_set_io_throttle: diff --git a/include/block/block_int.h b/include/block/block_int.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block_int.h +++ b/include/block/block_int.h @@ -XXX,XX +XXX,XX @@ void commit_active_start(const char *job_id, BlockDriverState *bs, * @filter_node_name: The node name that should be assigned to the filter * driver that the mirror job inserts into the graph above @bs. NULL means that * a node name should be autogenerated. + * @copy_mode: When to trigger writes to the target. * @errp: Error object. * * Start a mirroring operation on @bs. Clusters that are allocated @@ -XXX,XX +XXX,XX @@ void mirror_start(const char *job_id, BlockDriverState *bs, MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, BlockdevOnError on_source_error, BlockdevOnError on_target_error, - bool unmap, const char *filter_node_name, Error **errp); + bool unmap, const char *filter_node_name, + MirrorCopyMode copy_mode, Error **errp); /* * backup_job_create: diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, const BlockJobDriver *driver, bool is_none_mode, BlockDriverState *base, bool auto_complete, const char *filter_node_name, - bool is_mirror, + bool is_mirror, MirrorCopyMode copy_mode, Error **errp) { MirrorBlockJob *s; @@ -XXX,XX +XXX,XX @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, s->on_target_error = on_target_error; s->is_none_mode = is_none_mode; s->backing_mode = backing_mode; - s->copy_mode = MIRROR_COPY_MODE_BACKGROUND; + s->copy_mode = copy_mode; s->base = base; s->granularity = granularity; s->buf_size = ROUND_UP(buf_size, granularity); @@ -XXX,XX +XXX,XX @@ void mirror_start(const char *job_id, BlockDriverState *bs, MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, BlockdevOnError on_source_error, BlockdevOnError on_target_error, - bool unmap, const char *filter_node_name, Error **errp) + bool unmap, const char *filter_node_name, + MirrorCopyMode copy_mode, Error **errp) { bool is_none_mode; BlockDriverState *base; @@ -XXX,XX +XXX,XX @@ void mirror_start(const char *job_id, BlockDriverState *bs, speed, granularity, buf_size, backing_mode, on_source_error, on_target_error, unmap, NULL, NULL, &mirror_job_driver, is_none_mode, base, false, - filter_node_name, true, errp); + filter_node_name, true, copy_mode, errp); } void commit_active_start(const char *job_id, BlockDriverState *bs, @@ -XXX,XX +XXX,XX @@ void commit_active_start(const char *job_id, BlockDriverState *bs, MIRROR_LEAVE_BACKING_CHAIN, on_error, on_error, true, cb, opaque, &commit_active_job_driver, false, base, auto_complete, - filter_node_name, false, &local_err); + filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND, + &local_err); if (local_err) { error_propagate(errp, local_err); goto error_restore_flags; diff --git a/blockdev.c b/blockdev.c index XXXXXXX..XXXXXXX 100644 --- a/blockdev.c +++ b/blockdev.c @@ -XXX,XX +XXX,XX @@ static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs, bool has_unmap, bool unmap, bool has_filter_node_name, const char *filter_node_name, + bool has_copy_mode, MirrorCopyMode copy_mode, Error **errp) { @@ -XXX,XX +XXX,XX @@ static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs, if (!has_filter_node_name) { filter_node_name = NULL; } + if (!has_copy_mode) { + copy_mode = MIRROR_COPY_MODE_BACKGROUND; + } if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) { error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity", @@ -XXX,XX +XXX,XX @@ static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs, has_replaces ? replaces : NULL, speed, granularity, buf_size, sync, backing_mode, on_source_error, on_target_error, unmap, filter_node_name, - errp); + copy_mode, errp); } void qmp_drive_mirror(DriveMirror *arg, Error **errp) @@ -XXX,XX +XXX,XX @@ void qmp_drive_mirror(DriveMirror *arg, Error **errp) arg->has_on_target_error, arg->on_target_error, arg->has_unmap, arg->unmap, false, NULL, + arg->has_copy_mode, arg->copy_mode, &local_err); bdrv_unref(target_bs); error_propagate(errp, local_err); @@ -XXX,XX +XXX,XX @@ void qmp_blockdev_mirror(bool has_job_id, const char *job_id, BlockdevOnError on_target_error, bool has_filter_node_name, const char *filter_node_name, + bool has_copy_mode, MirrorCopyMode copy_mode, Error **errp) { BlockDriverState *bs; @@ -XXX,XX +XXX,XX @@ void qmp_blockdev_mirror(bool has_job_id, const char *job_id, has_on_target_error, on_target_error, true, true, has_filter_node_name, filter_node_name, + has_copy_mode, copy_mode, &local_err); error_propagate(errp, local_err); -- 2.13.6 From: Max Reitz <mreitz@redhat.com> Signed-off-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Fam Zheng <famz@redhat.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Message-id: 20180613181823.13618-15-mreitz@redhat.com Signed-off-by: Max Reitz <mreitz@redhat.com> --- tests/qemu-iotests/151 | 120 +++++++++++++++++++++++++++++++++++++++++++++ tests/qemu-iotests/151.out | 5 ++ tests/qemu-iotests/group | 1 + 3 files changed, 126 insertions(+) create mode 100755 tests/qemu-iotests/151 create mode 100644 tests/qemu-iotests/151.out diff --git a/tests/qemu-iotests/151 b/tests/qemu-iotests/151 new file mode 100755 index XXXXXXX..XXXXXXX --- /dev/null +++ b/tests/qemu-iotests/151 @@ -XXX,XX +XXX,XX @@ +#!/usr/bin/env python +# +# Tests for active mirroring +# +# Copyright (C) 2018 Red Hat, Inc. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +# + +import os +import iotests +from iotests import qemu_img + +source_img = os.path.join(iotests.test_dir, 'source.' + iotests.imgfmt) +target_img = os.path.join(iotests.test_dir, 'target.' + iotests.imgfmt) + +class TestActiveMirror(iotests.QMPTestCase): + image_len = 128 * 1024 * 1024 # MB + potential_writes_in_flight = True + + def setUp(self): + qemu_img('create', '-f', iotests.imgfmt, source_img, '128M') + qemu_img('create', '-f', iotests.imgfmt, target_img, '128M') + + blk_source = {'id': 'source', + 'if': 'none', + 'node-name': 'source-node', + 'driver': iotests.imgfmt, + 'file': {'driver': 'file', + 'filename': source_img}} + + blk_target = {'node-name': 'target-node', + 'driver': iotests.imgfmt, + 'file': {'driver': 'file', + 'filename': target_img}} + + self.vm = iotests.VM() + self.vm.add_drive_raw(self.vm.qmp_to_opts(blk_source)) + self.vm.add_blockdev(self.vm.qmp_to_opts(blk_target)) + self.vm.add_device('virtio-blk,drive=source') + self.vm.launch() + + def tearDown(self): + self.vm.shutdown() + + if not self.potential_writes_in_flight: + self.assertTrue(iotests.compare_images(source_img, target_img), + 'mirror target does not match source') + + os.remove(source_img) + os.remove(target_img) + + def doActiveIO(self, sync_source_and_target): + # Fill the source image + self.vm.hmp_qemu_io('source', + 'write -P 1 0 %i' % self.image_len); + + # Start some background requests + for offset in range(1 * self.image_len / 8, 3 * self.image_len / 8, 1024 * 1024): + self.vm.hmp_qemu_io('source', 'aio_write -P 2 %i 1M' % offset) + for offset in range(2 * self.image_len / 8, 3 * self.image_len / 8, 1024 * 1024): + self.vm.hmp_qemu_io('source', 'aio_write -z %i 1M' % offset) + + # Start the block job + result = self.vm.qmp('blockdev-mirror', + job_id='mirror', + filter_node_name='mirror-node', + device='source-node', + target='target-node', + sync='full', + copy_mode='write-blocking') + self.assert_qmp(result, 'return', {}) + + # Start some more requests + for offset in range(3 * self.image_len / 8, 5 * self.image_len / 8, 1024 * 1024): + self.vm.hmp_qemu_io('source', 'aio_write -P 3 %i 1M' % offset) + for offset in range(4 * self.image_len / 8, 5 * self.image_len / 8, 1024 * 1024): + self.vm.hmp_qemu_io('source', 'aio_write -z %i 1M' % offset) + + # Wait for the READY event + self.wait_ready(drive='mirror') + + # Now start some final requests; all of these (which land on + # the source) should be settled using the active mechanism. + # The mirror code itself asserts that the source BDS's dirty + # bitmap will stay clean between READY and COMPLETED. + for offset in range(5 * self.image_len / 8, 7 * self.image_len / 8, 1024 * 1024): + self.vm.hmp_qemu_io('source', 'aio_write -P 3 %i 1M' % offset) + for offset in range(6 * self.image_len / 8, 7 * self.image_len / 8, 1024 * 1024): + self.vm.hmp_qemu_io('source', 'aio_write -z %i 1M' % offset) + + if sync_source_and_target: + # If source and target should be in sync after the mirror, + # we have to flush before completion + self.vm.hmp_qemu_io('source', 'aio_flush') + self.potential_writes_in_flight = False + + self.complete_and_wait(drive='mirror', wait_ready=False) + + def testActiveIO(self): + self.doActiveIO(False) + + def testActiveIOFlushed(self): + self.doActiveIO(True) + + + +if __name__ == '__main__': + iotests.main(supported_fmts=['qcow2', 'raw']) diff --git a/tests/qemu-iotests/151.out b/tests/qemu-iotests/151.out new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/tests/qemu-iotests/151.out @@ -XXX,XX +XXX,XX @@ +.. +---------------------------------------------------------------------- +Ran 2 tests + +OK diff --git a/tests/qemu-iotests/group b/tests/qemu-iotests/group index XXXXXXX..XXXXXXX 100644 --- a/tests/qemu-iotests/group +++ b/tests/qemu-iotests/group @@ -XXX,XX +XXX,XX @@ 148 rw auto quick 149 rw auto sudo 150 rw auto quick +151 rw auto 152 rw auto quick 153 rw auto quick 154 rw auto backing quick -- 2.13.6 The following changes since commit 16aaacb307ed607b9780c12702c44f0fe52edc7e: Merge remote-tracking branch 'remotes/cohuck/tags/s390x-20200430' into staging (2020-04-30 14:00:36 +0100) are available in the Git repository at: git://repo.or.cz/qemu/kevin.git tags/for-upstream for you to fetch changes up to eaae29ef89d498d0eac553c77b554f310a47f809: qemu-storage-daemon: Fix non-string --object properties (2020-04-30 17:51:07 +0200) ---------------------------------------------------------------- Block layer patches: - Fix resize (extending) of short overlays - nvme: introduce PMR support from NVMe 1.4 spec - qemu-storage-daemon: Fix non-string --object properties ---------------------------------------------------------------- Alberto Garcia (1): qcow2: Add incompatibility note between backing files and raw external data files Andrzej Jakowski (1): nvme: introduce PMR support from NVMe 1.4 spec Kevin Wolf (12): block: Add flags to BlockDriver.bdrv_co_truncate() block: Add flags to bdrv(_co)_truncate() block-backend: Add flags to blk_truncate() qcow2: Support BDRV_REQ_ZERO_WRITE for truncate raw-format: Support BDRV_REQ_ZERO_WRITE for truncate file-posix: Support BDRV_REQ_ZERO_WRITE for truncate block: truncate: Don't make backing file data visible iotests: Filter testfiles out in filter_img_info() iotests: Test committing to short backing file qcow2: Forward ZERO_WRITE flag for full preallocation qom: Factor out user_creatable_add_dict() qemu-storage-daemon: Fix non-string --object properties Paolo Bonzini (1): qemu-iotests: allow qcow2 external discarded clusters to contain stale data docs/interop/qcow2.txt | 3 + hw/block/nvme.h | 2 + include/block/block.h | 5 +- include/block/block_int.h | 10 +- include/block/nvme.h | 172 ++++++++++++++++++++++++++ include/qom/object_interfaces.h | 16 +++ include/sysemu/block-backend.h | 2 +- block.c | 3 +- block/block-backend.c | 4 +- block/commit.c | 4 +- block/crypto.c | 7 +- block/file-posix.c | 6 +- block/file-win32.c | 2 +- block/gluster.c | 1 + block/io.c | 43 ++++++- block/iscsi.c | 2 +- block/mirror.c | 2 +- block/nfs.c | 3 +- block/parallels.c | 6 +- block/qcow.c | 4 +- block/qcow2-cluster.c | 2 +- block/qcow2-refcount.c | 2 +- block/qcow2.c | 73 +++++++++-- block/qed.c | 3 +- block/raw-format.c | 6 +- block/rbd.c | 1 + block/sheepdog.c | 4 +- block/ssh.c | 2 +- block/vdi.c | 2 +- block/vhdx-log.c | 2 +- block/vhdx.c | 6 +- block/vmdk.c | 8 +- block/vpc.c | 2 +- blockdev.c | 2 +- hw/block/nvme.c | 109 ++++++++++++++++ qemu-img.c | 2 +- qemu-io-cmds.c | 2 +- qemu-storage-daemon.c | 4 +- qom/object_interfaces.c | 31 +++++ qom/qom-qmp-cmds.c | 24 +--- tests/test-block-iothread.c | 9 +- tests/qemu-iotests/iotests.py | 5 +- hw/block/Makefile.objs | 2 +- hw/block/trace-events | 4 + tests/qemu-iotests/244 | 10 +- tests/qemu-iotests/244.out | 9 +- tests/qemu-iotests/274 | 155 +++++++++++++++++++++++ tests/qemu-iotests/274.out | 268 ++++++++++++++++++++++++++++++++++++++++ tests/qemu-iotests/group | 1 + 49 files changed, 951 insertions(+), 96 deletions(-) create mode 100755 tests/qemu-iotests/274 create mode 100644 tests/qemu-iotests/274.out From: Alberto Garcia <berto@igalia.com> Backing files and raw external data files are mutually exclusive. The documentation of the raw external data bit (in autoclear_features) already indicates that, but we should also mention it on the other side. Suggested-by: Eric Blake <eblake@redhat.com> Signed-off-by: Alberto Garcia <berto@igalia.com> Message-Id: <20200410121816.8334-1-berto@igalia.com> Reviewed-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- docs/interop/qcow2.txt | 3 +++ 1 file changed, 3 insertions(+) diff --git a/docs/interop/qcow2.txt b/docs/interop/qcow2.txt index XXXXXXX..XXXXXXX 100644 --- a/docs/interop/qcow2.txt +++ b/docs/interop/qcow2.txt @@ -XXX,XX +XXX,XX @@ The first cluster of a qcow2 image contains the file header: is stored (NB: The string is not null terminated). 0 if the image doesn't have a backing file. + Note: backing files are incompatible with raw external data + files (auto-clear feature bit 1). + 16 - 19: backing_file_size Length of the backing file name in bytes. Must not be longer than 1023 bytes. Undefined if the image doesn't have -- 2.25.3 From: Paolo Bonzini <pbonzini@redhat.com> Test 244 checks the expected behavior of qcow2 external data files with respect to zero and discarded clusters. Filesystems however are free to ignore discard requests, and this seems to be the case for overlayfs. Relax the tests to skip checks on the external data file for discarded areas, which implies not using qemu-img compare in the data_file_raw=on case. This fixes docker tests on RHEL8. Cc: Kevin Wolf <kwolf@redhat.com> Cc: qemu-block@nongnu.org Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Message-Id: <20200409191006.24429-1-pbonzini@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/qemu-iotests/244 | 10 ++++++++-- tests/qemu-iotests/244.out | 9 ++++++--- 2 files changed, 14 insertions(+), 5 deletions(-) diff --git a/tests/qemu-iotests/244 b/tests/qemu-iotests/244 index XXXXXXX..XXXXXXX 100755 --- a/tests/qemu-iotests/244 +++ b/tests/qemu-iotests/244 @@ -XXX,XX +XXX,XX @@ $QEMU_IO -c 'read -P 0 0 1M' \ echo $QEMU_IO -c 'read -P 0 0 1M' \ -c 'read -P 0x11 1M 1M' \ - -c 'read -P 0 2M 2M' \ -c 'read -P 0x11 4M 1M' \ -c 'read -P 0 5M 1M' \ -f raw "$TEST_IMG.data" | @@ -XXX,XX +XXX,XX @@ $QEMU_IO -c 'read -P 0 0 1M' \ -f $IMGFMT "$TEST_IMG" | _filter_qemu_io +# Discarded clusters are only marked as such in the qcow2 metadata, but +# they can contain stale data in the external data file. Instead, zero +# clusters must be zeroed in the external data file too. echo -$QEMU_IMG compare "$TEST_IMG" "$TEST_IMG.data" +$QEMU_IO -c 'read -P 0 0 1M' \ + -c 'read -P 0x11 1M 1M' \ + -c 'read -P 0 3M 3M' \ + -f raw "$TEST_IMG".data | + _filter_qemu_io echo -n "qcow2 file size after I/O: " du -b $TEST_IMG | cut -f1 diff --git a/tests/qemu-iotests/244.out b/tests/qemu-iotests/244.out index XXXXXXX..XXXXXXX 100644 --- a/tests/qemu-iotests/244.out +++ b/tests/qemu-iotests/244.out @@ -XXX,XX +XXX,XX @@ read 1048576/1048576 bytes at offset 0 1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) read 1048576/1048576 bytes at offset 1048576 1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) -read 2097152/2097152 bytes at offset 2097152 -2 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) read 1048576/1048576 bytes at offset 4194304 1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) read 1048576/1048576 bytes at offset 5242880 @@ -XXX,XX +XXX,XX @@ read 1048576/1048576 bytes at offset 1048576 read 4194304/4194304 bytes at offset 2097152 4 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) -Images are identical. +read 1048576/1048576 bytes at offset 0 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) +read 1048576/1048576 bytes at offset 1048576 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) +read 3145728/3145728 bytes at offset 3145728 +3 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) qcow2 file size after I/O: 327680 === bdrv_co_block_status test for file and offset=0 === -- 2.25.3 This adds a new BdrvRequestFlags parameter to the .bdrv_co_truncate() driver callbacks, and a supported_truncate_flags field in BlockDriverState that allows drivers to advertise support for request flags in the context of truncate. For now, we always pass 0 and no drivers declare support for any flag. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Message-Id: <20200424125448.63318-2-kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/block_int.h | 10 +++++++++- block/crypto.c | 3 ++- block/file-posix.c | 2 +- block/file-win32.c | 2 +- block/gluster.c | 1 + block/io.c | 8 +++++++- block/iscsi.c | 2 +- block/nfs.c | 3 ++- block/qcow2.c | 2 +- block/qed.c | 1 + block/raw-format.c | 2 +- block/rbd.c | 1 + block/sheepdog.c | 4 ++-- block/ssh.c | 2 +- tests/test-block-iothread.c | 3 ++- 15 files changed, 33 insertions(+), 13 deletions(-) diff --git a/include/block/block_int.h b/include/block/block_int.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block_int.h +++ b/include/block/block_int.h @@ -XXX,XX +XXX,XX @@ struct BlockDriver { */ int coroutine_fn (*bdrv_co_truncate)(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp); + BdrvRequestFlags flags, Error **errp); int64_t (*bdrv_getlength)(BlockDriverState *bs); bool has_variable_length; @@ -XXX,XX +XXX,XX @@ struct BlockDriverState { /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA, * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED) */ unsigned int supported_zero_flags; + /* + * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE). + * + * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure + * that any added space reads as all zeros. If this can't be guaranteed, + * the operation must fail. + */ + unsigned int supported_truncate_flags; /* the following member gives a name to every node on the bs graph. */ char node_name[32]; diff --git a/block/crypto.c b/block/crypto.c index XXXXXXX..XXXXXXX 100644 --- a/block/crypto.c +++ b/block/crypto.c @@ -XXX,XX +XXX,XX @@ static int block_crypto_co_create_generic(BlockDriverState *bs, static int coroutine_fn block_crypto_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp) + PreallocMode prealloc, BdrvRequestFlags flags, + Error **errp) { BlockCrypto *crypto = bs->opaque; uint64_t payload_offset = diff --git a/block/file-posix.c b/block/file-posix.c index XXXXXXX..XXXXXXX 100644 --- a/block/file-posix.c +++ b/block/file-posix.c @@ -XXX,XX +XXX,XX @@ raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp) + BdrvRequestFlags flags, Error **errp) { BDRVRawState *s = bs->opaque; struct stat st; diff --git a/block/file-win32.c b/block/file-win32.c index XXXXXXX..XXXXXXX 100644 --- a/block/file-win32.c +++ b/block/file-win32.c @@ -XXX,XX +XXX,XX @@ static void raw_close(BlockDriverState *bs) static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp) + BdrvRequestFlags flags, Error **errp) { BDRVRawState *s = bs->opaque; LONG low, high; diff --git a/block/gluster.c b/block/gluster.c index XXXXXXX..XXXXXXX 100644 --- a/block/gluster.c +++ b/block/gluster.c @@ -XXX,XX +XXX,XX @@ static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, + BdrvRequestFlags flags, Error **errp) { BDRVGlusterState *s = bs->opaque; diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, BlockDriverState *bs = child->bs; BlockDriver *drv = bs->drv; BdrvTrackedRequest req; + BdrvRequestFlags flags = 0; int64_t old_size, new_bytes; int ret; @@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, } if (drv->bdrv_co_truncate) { - ret = drv->bdrv_co_truncate(bs, offset, exact, prealloc, errp); + if (flags & ~bs->supported_truncate_flags) { + error_setg(errp, "Block driver does not support requested flags"); + ret = -ENOTSUP; + goto out; + } + ret = drv->bdrv_co_truncate(bs, offset, exact, prealloc, flags, errp); } else if (bs->file && drv->is_filter) { ret = bdrv_co_truncate(bs->file, offset, exact, prealloc, errp); } else { diff --git a/block/iscsi.c b/block/iscsi.c index XXXXXXX..XXXXXXX 100644 --- a/block/iscsi.c +++ b/block/iscsi.c @@ -XXX,XX +XXX,XX @@ static void iscsi_reopen_commit(BDRVReopenState *reopen_state) static int coroutine_fn iscsi_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp) + BdrvRequestFlags flags, Error **errp) { IscsiLun *iscsilun = bs->opaque; int64_t cur_length; diff --git a/block/nfs.c b/block/nfs.c index XXXXXXX..XXXXXXX 100644 --- a/block/nfs.c +++ b/block/nfs.c @@ -XXX,XX +XXX,XX @@ static int64_t nfs_get_allocated_file_size(BlockDriverState *bs) static int coroutine_fn nfs_file_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp) + PreallocMode prealloc, BdrvRequestFlags flags, + Error **errp) { NFSClient *client = bs->opaque; int ret; diff --git a/block/qcow2.c b/block/qcow2.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow2.c +++ b/block/qcow2.c @@ -XXX,XX +XXX,XX @@ fail: static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp) + BdrvRequestFlags flags, Error **errp) { BDRVQcow2State *s = bs->opaque; uint64_t old_length; diff --git a/block/qed.c b/block/qed.c index XXXXXXX..XXXXXXX 100644 --- a/block/qed.c +++ b/block/qed.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn bdrv_qed_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, + BdrvRequestFlags flags, Error **errp) { BDRVQEDState *s = bs->opaque; diff --git a/block/raw-format.c b/block/raw-format.c index XXXXXXX..XXXXXXX 100644 --- a/block/raw-format.c +++ b/block/raw-format.c @@ -XXX,XX +XXX,XX @@ static void raw_refresh_limits(BlockDriverState *bs, Error **errp) static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp) + BdrvRequestFlags flags, Error **errp) { BDRVRawState *s = bs->opaque; diff --git a/block/rbd.c b/block/rbd.c index XXXXXXX..XXXXXXX 100644 --- a/block/rbd.c +++ b/block/rbd.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn qemu_rbd_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, + BdrvRequestFlags flags, Error **errp) { int r; diff --git a/block/sheepdog.c b/block/sheepdog.c index XXXXXXX..XXXXXXX 100644 --- a/block/sheepdog.c +++ b/block/sheepdog.c @@ -XXX,XX +XXX,XX @@ static int64_t sd_getlength(BlockDriverState *bs) static int coroutine_fn sd_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp) + BdrvRequestFlags flags, Error **errp) { BDRVSheepdogState *s = bs->opaque; int ret, fd; @@ -XXX,XX +XXX,XX @@ static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, assert(!flags); if (offset > s->inode.vdi_size) { - ret = sd_co_truncate(bs, offset, false, PREALLOC_MODE_OFF, NULL); + ret = sd_co_truncate(bs, offset, false, PREALLOC_MODE_OFF, 0, NULL); if (ret < 0) { return ret; } diff --git a/block/ssh.c b/block/ssh.c index XXXXXXX..XXXXXXX 100644 --- a/block/ssh.c +++ b/block/ssh.c @@ -XXX,XX +XXX,XX @@ static int64_t ssh_getlength(BlockDriverState *bs) static int coroutine_fn ssh_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, PreallocMode prealloc, - Error **errp) + BdrvRequestFlags flags, Error **errp) { BDRVSSHState *s = bs->opaque; diff --git a/tests/test-block-iothread.c b/tests/test-block-iothread.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-block-iothread.c +++ b/tests/test-block-iothread.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn bdrv_test_co_pdiscard(BlockDriverState *bs, static int coroutine_fn bdrv_test_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp) + PreallocMode prealloc, BdrvRequestFlags flags, + Error **errp) { return 0; } -- 2.25.3 Now that block drivers can support flags for .bdrv_co_truncate, expose the parameter in the node level interfaces bdrv_co_truncate() and bdrv_truncate(). Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Message-Id: <20200424125448.63318-3-kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/block/block.h | 5 +++-- block/block-backend.c | 2 +- block/crypto.c | 2 +- block/io.c | 12 +++++++----- block/parallels.c | 6 +++--- block/qcow.c | 4 ++-- block/qcow2-refcount.c | 2 +- block/qcow2.c | 15 +++++++++------ block/raw-format.c | 2 +- block/vhdx-log.c | 2 +- block/vhdx.c | 2 +- block/vmdk.c | 2 +- tests/test-block-iothread.c | 6 +++--- 13 files changed, 34 insertions(+), 28 deletions(-) diff --git a/include/block/block.h b/include/block/block.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/block.h +++ b/include/block/block.h @@ -XXX,XX +XXX,XX @@ BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, void bdrv_refresh_filename(BlockDriverState *bs); int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp); + PreallocMode prealloc, BdrvRequestFlags flags, + Error **errp); int bdrv_truncate(BdrvChild *child, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp); + PreallocMode prealloc, BdrvRequestFlags flags, Error **errp); int64_t bdrv_nb_sectors(BlockDriverState *bs); int64_t bdrv_getlength(BlockDriverState *bs); diff --git a/block/block-backend.c b/block/block-backend.c index XXXXXXX..XXXXXXX 100644 --- a/block/block-backend.c +++ b/block/block-backend.c @@ -XXX,XX +XXX,XX @@ int blk_truncate(BlockBackend *blk, int64_t offset, bool exact, return -ENOMEDIUM; } - return bdrv_truncate(blk->root, offset, exact, prealloc, errp); + return bdrv_truncate(blk->root, offset, exact, prealloc, 0, errp); } int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf, diff --git a/block/crypto.c b/block/crypto.c index XXXXXXX..XXXXXXX 100644 --- a/block/crypto.c +++ b/block/crypto.c @@ -XXX,XX +XXX,XX @@ block_crypto_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, offset += payload_offset; - return bdrv_co_truncate(bs->file, offset, exact, prealloc, errp); + return bdrv_co_truncate(bs->file, offset, exact, prealloc, 0, errp); } static void block_crypto_close(BlockDriverState *bs) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ static void bdrv_parent_cb_resize(BlockDriverState *bs) * 'offset' bytes in length. */ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp) + PreallocMode prealloc, BdrvRequestFlags flags, + Error **errp) { BlockDriverState *bs = child->bs; BlockDriver *drv = bs->drv; BdrvTrackedRequest req; - BdrvRequestFlags flags = 0; int64_t old_size, new_bytes; int ret; @@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, } ret = drv->bdrv_co_truncate(bs, offset, exact, prealloc, flags, errp); } else if (bs->file && drv->is_filter) { - ret = bdrv_co_truncate(bs->file, offset, exact, prealloc, errp); + ret = bdrv_co_truncate(bs->file, offset, exact, prealloc, flags, errp); } else { error_setg(errp, "Image format driver does not support resize"); ret = -ENOTSUP; @@ -XXX,XX +XXX,XX @@ typedef struct TruncateCo { int64_t offset; bool exact; PreallocMode prealloc; + BdrvRequestFlags flags; Error **errp; int ret; } TruncateCo; @@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_truncate_co_entry(void *opaque) { TruncateCo *tco = opaque; tco->ret = bdrv_co_truncate(tco->child, tco->offset, tco->exact, - tco->prealloc, tco->errp); + tco->prealloc, tco->flags, tco->errp); aio_wait_kick(); } int bdrv_truncate(BdrvChild *child, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp) + PreallocMode prealloc, BdrvRequestFlags flags, Error **errp) { Coroutine *co; TruncateCo tco = { @@ -XXX,XX +XXX,XX @@ int bdrv_truncate(BdrvChild *child, int64_t offset, bool exact, .offset = offset, .exact = exact, .prealloc = prealloc, + .flags = flags, .errp = errp, .ret = NOT_DONE, }; diff --git a/block/parallels.c b/block/parallels.c index XXXXXXX..XXXXXXX 100644 --- a/block/parallels.c +++ b/block/parallels.c @@ -XXX,XX +XXX,XX @@ static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num, } else { ret = bdrv_truncate(bs->file, (s->data_end + space) << BDRV_SECTOR_BITS, - false, PREALLOC_MODE_OFF, NULL); + false, PREALLOC_MODE_OFF, 0, NULL); } if (ret < 0) { return ret; @@ -XXX,XX +XXX,XX @@ static int coroutine_fn parallels_co_check(BlockDriverState *bs, * That means we have to pass exact=true. */ ret = bdrv_truncate(bs->file, res->image_end_offset, true, - PREALLOC_MODE_OFF, &local_err); + PREALLOC_MODE_OFF, 0, &local_err); if (ret < 0) { error_report_err(local_err); res->check_errors++; @@ -XXX,XX +XXX,XX @@ static void parallels_close(BlockDriverState *bs) /* errors are ignored, so we might as well pass exact=true */ bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true, - PREALLOC_MODE_OFF, NULL); + PREALLOC_MODE_OFF, 0, NULL); } g_free(s->bat_dirty_bmap); diff --git a/block/qcow.c b/block/qcow.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow.c +++ b/block/qcow.c @@ -XXX,XX +XXX,XX @@ static int get_cluster_offset(BlockDriverState *bs, return -E2BIG; } ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size, - false, PREALLOC_MODE_OFF, NULL); + false, PREALLOC_MODE_OFF, 0, NULL); if (ret < 0) { return ret; } @@ -XXX,XX +XXX,XX @@ static int qcow_make_empty(BlockDriverState *bs) l1_length) < 0) return -1; ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false, - PREALLOC_MODE_OFF, NULL); + PREALLOC_MODE_OFF, 0, NULL); if (ret < 0) return ret; diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow2-refcount.c +++ b/block/qcow2-refcount.c @@ -XXX,XX +XXX,XX @@ static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res, } ret = bdrv_truncate(bs->file, offset + s->cluster_size, false, - PREALLOC_MODE_OFF, &local_err); + PREALLOC_MODE_OFF, 0, &local_err); if (ret < 0) { error_report_err(local_err); goto resize_fail; diff --git a/block/qcow2.c b/block/qcow2.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow2.c +++ b/block/qcow2.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, mode = PREALLOC_MODE_OFF; } ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false, - mode, errp); + mode, 0, errp); if (ret < 0) { return ret; } @@ -XXX,XX +XXX,XX @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, * always fulfilled, so there is no need to pass it on.) */ bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, - false, PREALLOC_MODE_OFF, &local_err); + false, PREALLOC_MODE_OFF, 0, &local_err); if (local_err) { warn_reportf_err(local_err, "Failed to truncate the tail of the image: "); @@ -XXX,XX +XXX,XX @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, * file should be resized to the exact target size, too, * so we pass @exact here. */ - ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, errp); + ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0, + errp); if (ret < 0) { goto fail; } @@ -XXX,XX +XXX,XX @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, new_file_size = allocation_start + nb_new_data_clusters * s->cluster_size; /* Image file grows, so @exact does not matter */ - ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, errp); + ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, + errp); if (ret < 0) { error_prepend(errp, "Failed to resize underlying file: "); qcow2_free_clusters(bs, allocation_start, @@ -XXX,XX +XXX,XX @@ qcow2_co_pwritev_compressed_part(BlockDriverState *bs, if (len < 0) { return len; } - return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, NULL); + return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0, + NULL); } if (offset_into_cluster(s, offset)) { @@ -XXX,XX +XXX,XX @@ static int make_completely_empty(BlockDriverState *bs) } ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false, - PREALLOC_MODE_OFF, &local_err); + PREALLOC_MODE_OFF, 0, &local_err); if (ret < 0) { error_report_err(local_err); goto fail; diff --git a/block/raw-format.c b/block/raw-format.c index XXXXXXX..XXXXXXX 100644 --- a/block/raw-format.c +++ b/block/raw-format.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, s->size = offset; offset += s->offset; - return bdrv_co_truncate(bs->file, offset, exact, prealloc, errp); + return bdrv_co_truncate(bs->file, offset, exact, prealloc, 0, errp); } static void raw_eject(BlockDriverState *bs, bool eject_flag) diff --git a/block/vhdx-log.c b/block/vhdx-log.c index XXXXXXX..XXXXXXX 100644 --- a/block/vhdx-log.c +++ b/block/vhdx-log.c @@ -XXX,XX +XXX,XX @@ static int vhdx_log_flush(BlockDriverState *bs, BDRVVHDXState *s, goto exit; } ret = bdrv_truncate(bs->file, new_file_size, false, - PREALLOC_MODE_OFF, NULL); + PREALLOC_MODE_OFF, 0, NULL); if (ret < 0) { goto exit; } diff --git a/block/vhdx.c b/block/vhdx.c index XXXXXXX..XXXXXXX 100644 --- a/block/vhdx.c +++ b/block/vhdx.c @@ -XXX,XX +XXX,XX @@ static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s, } return bdrv_truncate(bs->file, *new_offset + s->block_size, false, - PREALLOC_MODE_OFF, NULL); + PREALLOC_MODE_OFF, 0, NULL); } /* diff --git a/block/vmdk.c b/block/vmdk.c index XXXXXXX..XXXXXXX 100644 --- a/block/vmdk.c +++ b/block/vmdk.c @@ -XXX,XX +XXX,XX @@ vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, } length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE); ret = bdrv_truncate(s->extents[i].file, length, false, - PREALLOC_MODE_OFF, NULL); + PREALLOC_MODE_OFF, 0, NULL); if (ret < 0) { return ret; } diff --git a/tests/test-block-iothread.c b/tests/test-block-iothread.c index XXXXXXX..XXXXXXX 100644 --- a/tests/test-block-iothread.c +++ b/tests/test-block-iothread.c @@ -XXX,XX +XXX,XX @@ static void test_sync_op_truncate(BdrvChild *c) int ret; /* Normal success path */ - ret = bdrv_truncate(c, 65536, false, PREALLOC_MODE_OFF, NULL); + ret = bdrv_truncate(c, 65536, false, PREALLOC_MODE_OFF, 0, NULL); g_assert_cmpint(ret, ==, 0); /* Early error: Negative offset */ - ret = bdrv_truncate(c, -2, false, PREALLOC_MODE_OFF, NULL); + ret = bdrv_truncate(c, -2, false, PREALLOC_MODE_OFF, 0, NULL); g_assert_cmpint(ret, ==, -EINVAL); /* Error: Read-only image */ c->bs->read_only = true; c->bs->open_flags &= ~BDRV_O_RDWR; - ret = bdrv_truncate(c, 65536, false, PREALLOC_MODE_OFF, NULL); + ret = bdrv_truncate(c, 65536, false, PREALLOC_MODE_OFF, 0, NULL); g_assert_cmpint(ret, ==, -EACCES); c->bs->read_only = false; -- 2.25.3 Now that node level interface bdrv_truncate() supports passing request flags to the block driver, expose this on the BlockBackend level, too. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Message-Id: <20200424125448.63318-4-kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/sysemu/block-backend.h | 2 +- block.c | 3 ++- block/block-backend.c | 4 ++-- block/commit.c | 4 ++-- block/crypto.c | 2 +- block/mirror.c | 2 +- block/qcow2.c | 4 ++-- block/qed.c | 2 +- block/vdi.c | 2 +- block/vhdx.c | 4 ++-- block/vmdk.c | 6 +++--- block/vpc.c | 2 +- blockdev.c | 2 +- qemu-img.c | 2 +- qemu-io-cmds.c | 2 +- 15 files changed, 22 insertions(+), 21 deletions(-) diff --git a/include/sysemu/block-backend.h b/include/sysemu/block-backend.h index XXXXXXX..XXXXXXX 100644 --- a/include/sysemu/block-backend.h +++ b/include/sysemu/block-backend.h @@ -XXX,XX +XXX,XX @@ int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset, int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf, int bytes); int blk_truncate(BlockBackend *blk, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp); + PreallocMode prealloc, BdrvRequestFlags flags, Error **errp); int blk_pdiscard(BlockBackend *blk, int64_t offset, int bytes); int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf, int64_t pos, int size); diff --git a/block.c b/block.c index XXXXXXX..XXXXXXX 100644 --- a/block.c +++ b/block.c @@ -XXX,XX +XXX,XX @@ static int64_t create_file_fallback_truncate(BlockBackend *blk, int64_t size; int ret; - ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, &local_err); + ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0, + &local_err); if (ret < 0 && ret != -ENOTSUP) { error_propagate(errp, local_err); return ret; diff --git a/block/block-backend.c b/block/block-backend.c index XXXXXXX..XXXXXXX 100644 --- a/block/block-backend.c +++ b/block/block-backend.c @@ -XXX,XX +XXX,XX @@ int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf, } int blk_truncate(BlockBackend *blk, int64_t offset, bool exact, - PreallocMode prealloc, Error **errp) + PreallocMode prealloc, BdrvRequestFlags flags, Error **errp) { if (!blk_is_available(blk)) { error_setg(errp, "No medium inserted"); return -ENOMEDIUM; } - return bdrv_truncate(blk->root, offset, exact, prealloc, 0, errp); + return bdrv_truncate(blk->root, offset, exact, prealloc, flags, errp); } int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf, diff --git a/block/commit.c b/block/commit.c index XXXXXXX..XXXXXXX 100644 --- a/block/commit.c +++ b/block/commit.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn commit_run(Job *job, Error **errp) } if (base_len < len) { - ret = blk_truncate(s->base, len, false, PREALLOC_MODE_OFF, NULL); + ret = blk_truncate(s->base, len, false, PREALLOC_MODE_OFF, 0, NULL); if (ret) { goto out; } @@ -XXX,XX +XXX,XX @@ int bdrv_commit(BlockDriverState *bs) * grow the backing file image if possible. If not possible, * we must return an error */ if (length > backing_length) { - ret = blk_truncate(backing, length, false, PREALLOC_MODE_OFF, + ret = blk_truncate(backing, length, false, PREALLOC_MODE_OFF, 0, &local_err); if (ret < 0) { error_report_err(local_err); diff --git a/block/crypto.c b/block/crypto.c index XXXXXXX..XXXXXXX 100644 --- a/block/crypto.c +++ b/block/crypto.c @@ -XXX,XX +XXX,XX @@ static ssize_t block_crypto_init_func(QCryptoBlock *block, * which will be used by the crypto header */ return blk_truncate(data->blk, data->size + headerlen, false, - data->prealloc, errp); + data->prealloc, 0, errp); } diff --git a/block/mirror.c b/block/mirror.c index XXXXXXX..XXXXXXX 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn mirror_run(Job *job, Error **errp) if (s->bdev_length > base_length) { ret = blk_truncate(s->target, s->bdev_length, false, - PREALLOC_MODE_OFF, NULL); + PREALLOC_MODE_OFF, 0, NULL); if (ret < 0) { goto immediate_exit; } diff --git a/block/qcow2.c b/block/qcow2.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow2.c +++ b/block/qcow2.c @@ -XXX,XX +XXX,XX @@ qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) /* Okay, now that we have a valid image, let's give it the right size */ ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation, - errp); + 0, errp); if (ret < 0) { error_prepend(errp, "Could not resize image: "); goto out; @@ -XXX,XX +XXX,XX @@ static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, * Amending image options should ensure that the image has * exactly the given new values, so pass exact=true here. */ - ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, errp); + ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp); blk_unref(blk); if (ret < 0) { return ret; diff --git a/block/qed.c b/block/qed.c index XXXXXXX..XXXXXXX 100644 --- a/block/qed.c +++ b/block/qed.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn bdrv_qed_co_create(BlockdevCreateOptions *opts, * The QED format associates file length with allocation status, * so a new file (which is empty) must have a length of 0. */ - ret = blk_truncate(blk, 0, true, PREALLOC_MODE_OFF, errp); + ret = blk_truncate(blk, 0, true, PREALLOC_MODE_OFF, 0, errp); if (ret < 0) { goto out; } diff --git a/block/vdi.c b/block/vdi.c index XXXXXXX..XXXXXXX 100644 --- a/block/vdi.c +++ b/block/vdi.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options, if (image_type == VDI_TYPE_STATIC) { ret = blk_truncate(blk, offset + blocks * block_size, false, - PREALLOC_MODE_OFF, errp); + PREALLOC_MODE_OFF, 0, errp); if (ret < 0) { error_prepend(errp, "Failed to statically allocate file"); goto exit; diff --git a/block/vhdx.c b/block/vhdx.c index XXXXXXX..XXXXXXX 100644 --- a/block/vhdx.c +++ b/block/vhdx.c @@ -XXX,XX +XXX,XX @@ static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s, /* All zeroes, so we can just extend the file - the end of the BAT * is the furthest thing we have written yet */ ret = blk_truncate(blk, data_file_offset, false, PREALLOC_MODE_OFF, - errp); + 0, errp); if (ret < 0) { goto exit; } } else if (type == VHDX_TYPE_FIXED) { ret = blk_truncate(blk, data_file_offset + image_size, false, - PREALLOC_MODE_OFF, errp); + PREALLOC_MODE_OFF, 0, errp); if (ret < 0) { goto exit; } diff --git a/block/vmdk.c b/block/vmdk.c index XXXXXXX..XXXXXXX 100644 --- a/block/vmdk.c +++ b/block/vmdk.c @@ -XXX,XX +XXX,XX @@ static int vmdk_init_extent(BlockBackend *blk, int gd_buf_size; if (flat) { - ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, errp); + ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp); goto exit; } magic = cpu_to_be32(VMDK4_MAGIC); @@ -XXX,XX +XXX,XX @@ static int vmdk_init_extent(BlockBackend *blk, } ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false, - PREALLOC_MODE_OFF, errp); + PREALLOC_MODE_OFF, 0, errp); if (ret < 0) { goto exit; } @@ -XXX,XX +XXX,XX @@ static int coroutine_fn vmdk_co_do_create(int64_t size, /* bdrv_pwrite write padding zeros to align to sector, we don't need that * for description file */ if (desc_offset == 0) { - ret = blk_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, errp); + ret = blk_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp); if (ret < 0) { goto exit; } diff --git a/block/vpc.c b/block/vpc.c index XXXXXXX..XXXXXXX 100644 --- a/block/vpc.c +++ b/block/vpc.c @@ -XXX,XX +XXX,XX @@ static int create_fixed_disk(BlockBackend *blk, uint8_t *buf, /* Add footer to total size */ total_size += HEADER_SIZE; - ret = blk_truncate(blk, total_size, false, PREALLOC_MODE_OFF, errp); + ret = blk_truncate(blk, total_size, false, PREALLOC_MODE_OFF, 0, errp); if (ret < 0) { return ret; } diff --git a/blockdev.c b/blockdev.c index XXXXXXX..XXXXXXX 100644 --- a/blockdev.c +++ b/blockdev.c @@ -XXX,XX +XXX,XX @@ void qmp_block_resize(bool has_device, const char *device, } bdrv_drained_begin(bs); - ret = blk_truncate(blk, size, false, PREALLOC_MODE_OFF, errp); + ret = blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp); bdrv_drained_end(bs); out: diff --git a/qemu-img.c b/qemu-img.c index XXXXXXX..XXXXXXX 100644 --- a/qemu-img.c +++ b/qemu-img.c @@ -XXX,XX +XXX,XX @@ static int img_resize(int argc, char **argv) * resizing, so pass @exact=true. It is of no use to report * success when the image has not actually been resized. */ - ret = blk_truncate(blk, total_size, true, prealloc, &err); + ret = blk_truncate(blk, total_size, true, prealloc, 0, &err); if (!ret) { qprintf(quiet, "Image resized.\n"); } else { diff --git a/qemu-io-cmds.c b/qemu-io-cmds.c index XXXXXXX..XXXXXXX 100644 --- a/qemu-io-cmds.c +++ b/qemu-io-cmds.c @@ -XXX,XX +XXX,XX @@ static int truncate_f(BlockBackend *blk, int argc, char **argv) * exact=true. It is better to err on the "emit more errors" side * than to be overly permissive. */ - ret = blk_truncate(blk, offset, true, PREALLOC_MODE_OFF, &local_err); + ret = blk_truncate(blk, offset, true, PREALLOC_MODE_OFF, 0, &local_err); if (ret < 0) { error_report_err(local_err); return ret; -- 2.25.3 If BDRV_REQ_ZERO_WRITE is set and we're extending the image, calling qcow2_cluster_zeroize() with flags=0 does the right thing: It doesn't undo any previous preallocation, but just adds the zero flag to all relevant L2 entries. If an external data file is in use, a write_zeroes request to the data file is made instead. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Message-Id: <20200424125448.63318-5-kwolf@redhat.com> Reviewed-by: Eric Blake <eblake@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/qcow2-cluster.c | 2 +- block/qcow2.c | 34 ++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 1 deletion(-) diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow2-cluster.c +++ b/block/qcow2-cluster.c @@ -XXX,XX +XXX,XX @@ int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, /* Caller must pass aligned values, except at image end */ assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || - end_offset == bs->total_sectors << BDRV_SECTOR_BITS); + end_offset >= bs->total_sectors << BDRV_SECTOR_BITS); /* The zero flag is only supported by version 3 and newer */ if (s->qcow_version < 3) { diff --git a/block/qcow2.c b/block/qcow2.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow2.c +++ b/block/qcow2.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, bs->supported_zero_flags = header.version >= 3 ? BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0; + bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; /* Repair image if dirty */ if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && @@ -XXX,XX +XXX,XX @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, g_assert_not_reached(); } + if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) { + uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->cluster_size); + + /* + * Use zero clusters as much as we can. qcow2_cluster_zeroize() + * requires a cluster-aligned start. The end may be unaligned if it is + * at the end of the image (which it is here). + */ + ret = qcow2_cluster_zeroize(bs, zero_start, offset - zero_start, 0); + if (ret < 0) { + error_setg_errno(errp, -ret, "Failed to zero out new clusters"); + goto fail; + } + + /* Write explicit zeros for the unaligned head */ + if (zero_start > old_length) { + uint64_t len = zero_start - old_length; + uint8_t *buf = qemu_blockalign0(bs, len); + QEMUIOVector qiov; + qemu_iovec_init_buf(&qiov, buf, len); + + qemu_co_mutex_unlock(&s->lock); + ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0); + qemu_co_mutex_lock(&s->lock); + + qemu_vfree(buf); + if (ret < 0) { + error_setg_errno(errp, -ret, "Failed to zero out the new area"); + goto fail; + } + } + } + if (prealloc != PREALLOC_MODE_OFF) { /* Flush metadata before actually changing the image size */ ret = qcow2_write_caches(bs); -- 2.25.3 The raw format driver can simply forward the flag and let its bs->file child take care of actually providing the zeros. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Eric Blake <eblake@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Message-Id: <20200424125448.63318-6-kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/raw-format.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/block/raw-format.c b/block/raw-format.c index XXXXXXX..XXXXXXX 100644 --- a/block/raw-format.c +++ b/block/raw-format.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, s->size = offset; offset += s->offset; - return bdrv_co_truncate(bs->file, offset, exact, prealloc, 0, errp); + return bdrv_co_truncate(bs->file, offset, exact, prealloc, flags, errp); } static void raw_eject(BlockDriverState *bs, bool eject_flag) @@ -XXX,XX +XXX,XX @@ static int raw_open(BlockDriverState *bs, QDict *options, int flags, bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED | ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) & bs->file->bs->supported_zero_flags); + bs->supported_truncate_flags = bs->file->bs->supported_truncate_flags & + BDRV_REQ_ZERO_WRITE; if (bs->probed && !bdrv_is_read_only(bs)) { bdrv_refresh_filename(bs->file->bs); -- 2.25.3 For regular files, we always get BDRV_REQ_ZERO_WRITE behaviour from the OS, so we can advertise the flag and just ignore it. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Reviewed-by: Alberto Garcia <berto@igalia.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Message-Id: <20200424125448.63318-7-kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/file-posix.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/block/file-posix.c b/block/file-posix.c index XXXXXXX..XXXXXXX 100644 --- a/block/file-posix.c +++ b/block/file-posix.c @@ -XXX,XX +XXX,XX @@ static int raw_open_common(BlockDriverState *bs, QDict *options, #endif bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; + if (S_ISREG(st.st_mode)) { + /* When extending regular files, we get zeros from the OS */ + bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; + } ret = 0; fail: if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { -- 2.25.3 When extending the size of an image that has a backing file larger than its old size, make sure that the backing file data doesn't become visible in the guest, but the added area is properly zeroed out. Consider the following scenario where the overlay is shorter than its backing file: base.qcow2: AAAAAAAA overlay.qcow2: BBBB When resizing (extending) overlay.qcow2, the new blocks should not stay unallocated and make the additional As from base.qcow2 visible like before this patch, but zeros should be read. A similar case happens with the various variants of a commit job when an intermediate file is short (- for unallocated): base.qcow2: A-A-AAAA mid.qcow2: BB-B top.qcow2: C--C--C- After commit top.qcow2 to mid.qcow2, the following happens: mid.qcow2: CB-C00C0 (correct result) mid.qcow2: CB-C--C- (before this fix) Without the fix, blocks that previously read as zeros on top.qcow2 suddenly turn into A. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Message-Id: <20200424125448.63318-8-kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/io.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) diff --git a/block/io.c b/block/io.c index XXXXXXX..XXXXXXX 100644 --- a/block/io.c +++ b/block/io.c @@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, goto out; } + /* + * If the image has a backing file that is large enough that it would + * provide data for the new area, we cannot leave it unallocated because + * then the backing file content would become visible. Instead, zero-fill + * the new area. + * + * Note that if the image has a backing file, but was opened without the + * backing file, taking care of keeping things consistent with that backing + * file is the user's responsibility. + */ + if (new_bytes && bs->backing) { + int64_t backing_len; + + backing_len = bdrv_getlength(backing_bs(bs)); + if (backing_len < 0) { + ret = backing_len; + error_setg_errno(errp, -ret, "Could not get backing file size"); + goto out; + } + + if (backing_len > old_size) { + flags |= BDRV_REQ_ZERO_WRITE; + } + } + if (drv->bdrv_co_truncate) { if (flags & ~bs->supported_truncate_flags) { error_setg(errp, "Block driver does not support requested flags"); -- 2.25.3 We want to keep TEST_IMG for the full path of the main test image, but filter_testfiles() must be called for other test images before replacing other things like the image format because the test directory path could contain the format as a substring. Insert a filter_testfiles() call between both. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Message-Id: <20200424125448.63318-9-kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/qemu-iotests/iotests.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/tests/qemu-iotests/iotests.py b/tests/qemu-iotests/iotests.py index XXXXXXX..XXXXXXX 100644 --- a/tests/qemu-iotests/iotests.py +++ b/tests/qemu-iotests/iotests.py @@ -XXX,XX +XXX,XX @@ def filter_img_info(output, filename): for line in output.split('\n'): if 'disk size' in line or 'actual-size' in line: continue - line = line.replace(filename, 'TEST_IMG') \ - .replace(imgfmt, 'IMGFMT') + line = line.replace(filename, 'TEST_IMG') + line = filter_testfiles(line) + line = line.replace(imgfmt, 'IMGFMT') line = re.sub('iters: [0-9]+', 'iters: XXX', line) line = re.sub('uuid: [-a-f0-9]+', 'uuid: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX', line) line = re.sub('cid: [0-9]+', 'cid: XXXXXXXXXX', line) -- 2.25.3 Signed-off-by: Kevin Wolf <kwolf@redhat.com> Message-Id: <20200424125448.63318-10-kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- tests/qemu-iotests/274 | 155 +++++++++++++++++++++ tests/qemu-iotests/274.out | 268 +++++++++++++++++++++++++++++++++++++ tests/qemu-iotests/group | 1 + 3 files changed, 424 insertions(+) create mode 100755 tests/qemu-iotests/274 create mode 100644 tests/qemu-iotests/274.out diff --git a/tests/qemu-iotests/274 b/tests/qemu-iotests/274 new file mode 100755 index XXXXXXX..XXXXXXX --- /dev/null +++ b/tests/qemu-iotests/274 @@ -XXX,XX +XXX,XX @@ +#!/usr/bin/env python3 +# +# Copyright (C) 2019 Red Hat, Inc. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +# +# Creator/Owner: Kevin Wolf <kwolf@redhat.com> +# +# Some tests for short backing files and short overlays + +import iotests + +iotests.verify_image_format(supported_fmts=['qcow2']) +iotests.verify_platform(['linux']) + +size_short = 1 * 1024 * 1024 +size_long = 2 * 1024 * 1024 +size_diff = size_long - size_short + +def create_chain() -> None: + iotests.qemu_img_log('create', '-f', iotests.imgfmt, base, + str(size_long)) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, '-b', base, mid, + str(size_short)) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, '-b', mid, top, + str(size_long)) + + iotests.qemu_io_log('-c', 'write -P 1 0 %d' % size_long, base) + +def create_vm() -> iotests.VM: + vm = iotests.VM() + vm.add_blockdev('file,filename=%s,node-name=base-file' % base) + vm.add_blockdev('%s,file=base-file,node-name=base' % iotests.imgfmt) + vm.add_blockdev('file,filename=%s,node-name=mid-file' % mid) + vm.add_blockdev('%s,file=mid-file,node-name=mid,backing=base' + % iotests.imgfmt) + vm.add_drive(top, 'backing=mid,node-name=top') + return vm + +with iotests.FilePath('base') as base, \ + iotests.FilePath('mid') as mid, \ + iotests.FilePath('top') as top: + + iotests.log('== Commit tests ==') + + create_chain() + + iotests.log('=== Check visible data ===') + + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, top) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), top) + + iotests.log('=== Checking allocation status ===') + + iotests.qemu_io_log('-c', 'alloc 0 %d' % size_short, + '-c', 'alloc %d %d' % (size_short, size_diff), + base) + + iotests.qemu_io_log('-c', 'alloc 0 %d' % size_short, + '-c', 'alloc %d %d' % (size_short, size_diff), + mid) + + iotests.qemu_io_log('-c', 'alloc 0 %d' % size_short, + '-c', 'alloc %d %d' % (size_short, size_diff), + top) + + iotests.log('=== Checking map ===') + + iotests.qemu_img_log('map', '--output=json', base) + iotests.qemu_img_log('map', '--output=human', base) + iotests.qemu_img_log('map', '--output=json', mid) + iotests.qemu_img_log('map', '--output=human', mid) + iotests.qemu_img_log('map', '--output=json', top) + iotests.qemu_img_log('map', '--output=human', top) + + iotests.log('=== Testing qemu-img commit (top -> mid) ===') + + iotests.qemu_img_log('commit', top) + iotests.img_info_log(mid) + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, mid) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), mid) + + iotests.log('=== Testing HMP commit (top -> mid) ===') + + create_chain() + with create_vm() as vm: + vm.launch() + vm.qmp_log('human-monitor-command', command_line='commit drive0') + + iotests.img_info_log(mid) + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, mid) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), mid) + + iotests.log('=== Testing QMP active commit (top -> mid) ===') + + create_chain() + with create_vm() as vm: + vm.launch() + vm.qmp_log('block-commit', device='top', base_node='mid', + job_id='job0', auto_dismiss=False) + vm.run_job('job0', wait=5) + + iotests.img_info_log(mid) + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, mid) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), mid) + + + iotests.log('== Resize tests ==') + + # Use different sizes for different allocation modes: + # + # We want to have at least one test where 32 bit truncation in the size of + # the overlapping area becomes visible. This is covered by the + # prealloc='off' case (1G to 6G is an overlap of 5G). + # + # However, we can only do this for modes that don't preallocate data + # because otherwise we might run out of space on the test host. + # + # We also want to test some unaligned combinations. + for (prealloc, base_size, top_size_old, top_size_new, off) in [ + ('off', '6G', '1G', '8G', '5G'), + ('metadata', '32G', '30G', '33G', '31G'), + ('falloc', '10M', '5M', '15M', '9M'), + ('full', '16M', '8M', '12M', '11M'), + ('off', '384k', '253k', '512k', '253k'), + ('off', '400k', '256k', '512k', '336k'), + ('off', '512k', '256k', '500k', '436k')]: + + iotests.log('=== preallocation=%s ===' % prealloc) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, base, base_size) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, '-b', base, top, + top_size_old) + iotests.qemu_io_log('-c', 'write -P 1 %s 64k' % off, base) + + # After this, top_size_old to base_size should be allocated/zeroed. + # + # In theory, leaving base_size to top_size_new unallocated would be + # correct, but in practice, if we zero out anything, we zero out + # everything up to top_size_new. + iotests.qemu_img_log('resize', '-f', iotests.imgfmt, + '--preallocation', prealloc, top, top_size_new) + iotests.qemu_io_log('-c', 'read -P 0 %s 64k' % off, top) + iotests.qemu_io_log('-c', 'map', top) + iotests.qemu_img_log('map', '--output=json', top) diff --git a/tests/qemu-iotests/274.out b/tests/qemu-iotests/274.out new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/tests/qemu-iotests/274.out @@ -XXX,XX +XXX,XX @@ +== Commit tests == +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=2097152 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-mid', fmt=qcow2 size=1048576 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=2097152 backing_file=TEST_DIR/PID-mid cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 2097152/2097152 bytes at offset 0 +2 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +=== Check visible data === +read 1048576/1048576 bytes at offset 0 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +read 1048576/1048576 bytes at offset 1048576 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +=== Checking allocation status === +1048576/1048576 bytes allocated at offset 0 bytes +1048576/1048576 bytes allocated at offset 1 MiB + +0/1048576 bytes allocated at offset 0 bytes +0/0 bytes allocated at offset 1 MiB + +0/1048576 bytes allocated at offset 0 bytes +0/1048576 bytes allocated at offset 1 MiB + +=== Checking map === +[{ "start": 0, "length": 2097152, "depth": 0, "zero": false, "data": true, "offset": 327680}] + +Offset Length Mapped to File +0 0x200000 0x50000 TEST_DIR/PID-base + +[{ "start": 0, "length": 1048576, "depth": 1, "zero": false, "data": true, "offset": 327680}] + +Offset Length Mapped to File +0 0x100000 0x50000 TEST_DIR/PID-base + +[{ "start": 0, "length": 1048576, "depth": 2, "zero": false, "data": true, "offset": 327680}, +{ "start": 1048576, "length": 1048576, "depth": 0, "zero": true, "data": false}] + +Offset Length Mapped to File +0 0x100000 0x50000 TEST_DIR/PID-base + +=== Testing qemu-img commit (top -> mid) === +Image committed. + +image: TEST_IMG +file format: IMGFMT +virtual size: 2 MiB (2097152 bytes) +cluster_size: 65536 +backing file: TEST_DIR/PID-base +Format specific information: + compat: 1.1 + lazy refcounts: false + refcount bits: 16 + corrupt: false + +read 1048576/1048576 bytes at offset 0 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +read 1048576/1048576 bytes at offset 1048576 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +=== Testing HMP commit (top -> mid) === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=2097152 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-mid', fmt=qcow2 size=1048576 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=2097152 backing_file=TEST_DIR/PID-mid cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 2097152/2097152 bytes at offset 0 +2 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +{"execute": "human-monitor-command", "arguments": {"command-line": "commit drive0"}} +{"return": ""} +image: TEST_IMG +file format: IMGFMT +virtual size: 2 MiB (2097152 bytes) +cluster_size: 65536 +backing file: TEST_DIR/PID-base +Format specific information: + compat: 1.1 + lazy refcounts: false + refcount bits: 16 + corrupt: false + +read 1048576/1048576 bytes at offset 0 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +read 1048576/1048576 bytes at offset 1048576 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +=== Testing QMP active commit (top -> mid) === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=2097152 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-mid', fmt=qcow2 size=1048576 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=2097152 backing_file=TEST_DIR/PID-mid cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 2097152/2097152 bytes at offset 0 +2 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +{"execute": "block-commit", "arguments": {"auto-dismiss": false, "base-node": "mid", "device": "top", "job-id": "job0"}} +{"return": {}} +{"execute": "job-complete", "arguments": {"id": "job0"}} +{"return": {}} +{"data": {"device": "job0", "len": 0, "offset": 0, "speed": 0, "type": "commit"}, "event": "BLOCK_JOB_READY", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}} +{"data": {"device": "job0", "len": 0, "offset": 0, "speed": 0, "type": "commit"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}} +{"execute": "job-dismiss", "arguments": {"id": "job0"}} +{"return": {}} +image: TEST_IMG +file format: IMGFMT +virtual size: 2 MiB (2097152 bytes) +cluster_size: 65536 +backing file: TEST_DIR/PID-base +Format specific information: + compat: 1.1 + lazy refcounts: false + refcount bits: 16 + corrupt: false + +read 1048576/1048576 bytes at offset 0 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +read 1048576/1048576 bytes at offset 1048576 +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +== Resize tests == +=== preallocation=off === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=6442450944 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=1073741824 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 65536/65536 bytes at offset 5368709120 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +Image resized. + +read 65536/65536 bytes at offset 5368709120 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +1 GiB (0x40000000) bytes not allocated at offset 0 bytes (0x0) +7 GiB (0x1c0000000) bytes allocated at offset 1 GiB (0x40000000) + +[{ "start": 0, "length": 1073741824, "depth": 1, "zero": true, "data": false}, +{ "start": 1073741824, "length": 7516192768, "depth": 0, "zero": true, "data": false}] + +=== preallocation=metadata === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=34359738368 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=32212254720 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 65536/65536 bytes at offset 33285996544 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +Image resized. + +read 65536/65536 bytes at offset 33285996544 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +30 GiB (0x780000000) bytes not allocated at offset 0 bytes (0x0) +3 GiB (0xc0000000) bytes allocated at offset 30 GiB (0x780000000) + +[{ "start": 0, "length": 32212254720, "depth": 1, "zero": true, "data": false}, +{ "start": 32212254720, "length": 536870912, "depth": 0, "zero": true, "data": false, "offset": 327680}, +{ "start": 32749125632, "length": 536870912, "depth": 0, "zero": true, "data": false, "offset": 537264128}, +{ "start": 33285996544, "length": 536870912, "depth": 0, "zero": true, "data": false, "offset": 1074200576}, +{ "start": 33822867456, "length": 536870912, "depth": 0, "zero": true, "data": false, "offset": 1611137024}, +{ "start": 34359738368, "length": 536870912, "depth": 0, "zero": true, "data": false, "offset": 2148139008}, +{ "start": 34896609280, "length": 536870912, "depth": 0, "zero": true, "data": false, "offset": 2685075456}] + +=== preallocation=falloc === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=10485760 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=5242880 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 65536/65536 bytes at offset 9437184 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +Image resized. + +read 65536/65536 bytes at offset 9437184 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +5 MiB (0x500000) bytes not allocated at offset 0 bytes (0x0) +10 MiB (0xa00000) bytes allocated at offset 5 MiB (0x500000) + +[{ "start": 0, "length": 5242880, "depth": 1, "zero": true, "data": false}, +{ "start": 5242880, "length": 10485760, "depth": 0, "zero": true, "data": false, "offset": 327680}] + +=== preallocation=full === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=16777216 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=8388608 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 65536/65536 bytes at offset 11534336 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +Image resized. + +read 65536/65536 bytes at offset 11534336 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +8 MiB (0x800000) bytes not allocated at offset 0 bytes (0x0) +4 MiB (0x400000) bytes allocated at offset 8 MiB (0x800000) + +[{ "start": 0, "length": 8388608, "depth": 1, "zero": true, "data": false}, +{ "start": 8388608, "length": 4194304, "depth": 0, "zero": true, "data": false, "offset": 327680}] + +=== preallocation=off === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=393216 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=259072 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 65536/65536 bytes at offset 259072 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +Image resized. + +read 65536/65536 bytes at offset 259072 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +192 KiB (0x30000) bytes not allocated at offset 0 bytes (0x0) +320 KiB (0x50000) bytes allocated at offset 192 KiB (0x30000) + +[{ "start": 0, "length": 196608, "depth": 1, "zero": true, "data": false}, +{ "start": 196608, "length": 65536, "depth": 0, "zero": false, "data": true, "offset": 327680}, +{ "start": 262144, "length": 262144, "depth": 0, "zero": true, "data": false}] + +=== preallocation=off === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=409600 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=262144 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 65536/65536 bytes at offset 344064 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +Image resized. + +read 65536/65536 bytes at offset 344064 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +256 KiB (0x40000) bytes not allocated at offset 0 bytes (0x0) +256 KiB (0x40000) bytes allocated at offset 256 KiB (0x40000) + +[{ "start": 0, "length": 262144, "depth": 1, "zero": true, "data": false}, +{ "start": 262144, "length": 262144, "depth": 0, "zero": true, "data": false}] + +=== preallocation=off === +Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=524288 cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +Formatting 'TEST_DIR/PID-top', fmt=qcow2 size=262144 backing_file=TEST_DIR/PID-base cluster_size=65536 lazy_refcounts=off refcount_bits=16 + +wrote 65536/65536 bytes at offset 446464 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +Image resized. + +read 65536/65536 bytes at offset 446464 +64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) + +256 KiB (0x40000) bytes not allocated at offset 0 bytes (0x0) +244 KiB (0x3d000) bytes allocated at offset 256 KiB (0x40000) + +[{ "start": 0, "length": 262144, "depth": 1, "zero": true, "data": false}, +{ "start": 262144, "length": 249856, "depth": 0, "zero": true, "data": false}] + diff --git a/tests/qemu-iotests/group b/tests/qemu-iotests/group index XXXXXXX..XXXXXXX 100644 --- a/tests/qemu-iotests/group +++ b/tests/qemu-iotests/group @@ -XXX,XX +XXX,XX @@ 270 rw backing quick 272 rw 273 backing quick +274 rw backing 277 rw quick 279 rw backing quick 280 rw migration quick -- 2.25.3 The BDRV_REQ_ZERO_WRITE is currently implemented in a way that first the image is possibly preallocated and then the zero flag is added to all clusters. This means that a copy-on-write operation may be needed when writing to these clusters, despite having used preallocation, negating one of the major benefits of preallocation. Instead, try to forward the BDRV_REQ_ZERO_WRITE to the protocol driver, and if the protocol driver can ensure that the new area reads as zeros, we can skip setting the zero flag in the qcow2 layer. Unfortunately, the same approach doesn't work for metadata preallocation, so we'll still set the zero flag there. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Message-Id: <20200424142701.67053-1-kwolf@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/qcow2.c | 22 +++++++++++++++++++--- tests/qemu-iotests/274.out | 4 ++-- 2 files changed, 21 insertions(+), 5 deletions(-) diff --git a/block/qcow2.c b/block/qcow2.c index XXXXXXX..XXXXXXX 100644 --- a/block/qcow2.c +++ b/block/qcow2.c @@ -XXX,XX +XXX,XX @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, /* Allocate the data area */ new_file_size = allocation_start + nb_new_data_clusters * s->cluster_size; - /* Image file grows, so @exact does not matter */ - ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, - errp); + /* + * Image file grows, so @exact does not matter. + * + * If we need to zero out the new area, try first whether the protocol + * driver can already take care of this. + */ + if (flags & BDRV_REQ_ZERO_WRITE) { + ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, + BDRV_REQ_ZERO_WRITE, NULL); + if (ret >= 0) { + flags &= ~BDRV_REQ_ZERO_WRITE; + } + } else { + ret = -1; + } + if (ret < 0) { + ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, + errp); + } if (ret < 0) { error_prepend(errp, "Failed to resize underlying file: "); qcow2_free_clusters(bs, allocation_start, diff --git a/tests/qemu-iotests/274.out b/tests/qemu-iotests/274.out index XXXXXXX..XXXXXXX 100644 --- a/tests/qemu-iotests/274.out +++ b/tests/qemu-iotests/274.out @@ -XXX,XX +XXX,XX @@ read 65536/65536 bytes at offset 9437184 10 MiB (0xa00000) bytes allocated at offset 5 MiB (0x500000) [{ "start": 0, "length": 5242880, "depth": 1, "zero": true, "data": false}, -{ "start": 5242880, "length": 10485760, "depth": 0, "zero": true, "data": false, "offset": 327680}] +{ "start": 5242880, "length": 10485760, "depth": 0, "zero": false, "data": true, "offset": 327680}] === preallocation=full === Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=16777216 cluster_size=65536 lazy_refcounts=off refcount_bits=16 @@ -XXX,XX +XXX,XX @@ read 65536/65536 bytes at offset 11534336 4 MiB (0x400000) bytes allocated at offset 8 MiB (0x800000) [{ "start": 0, "length": 8388608, "depth": 1, "zero": true, "data": false}, -{ "start": 8388608, "length": 4194304, "depth": 0, "zero": true, "data": false, "offset": 327680}] +{ "start": 8388608, "length": 4194304, "depth": 0, "zero": false, "data": true, "offset": 327680}] === preallocation=off === Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=393216 cluster_size=65536 lazy_refcounts=off refcount_bits=16 -- 2.25.3 From: Andrzej Jakowski <andrzej.jakowski@linux.intel.com> This patch introduces support for PMR that has been defined as part of NVMe 1.4 spec. User can now specify a pmrdev option that should point to HostMemoryBackend. pmrdev memory region will subsequently be exposed as PCI BAR 2 in emulated NVMe device. Guest OS can perform mmio read and writes to the PMR region that will stay persistent across system reboot. Signed-off-by: Andrzej Jakowski <andrzej.jakowski@linux.intel.com> Reviewed-by: Klaus Jensen <k.jensen@samsung.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> Message-Id: <20200330164656.9348-1-andrzej.jakowski@linux.intel.com> Reviewed-by: Keith Busch <kbusch@kernel.org> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- hw/block/nvme.h | 2 + include/block/nvme.h | 172 +++++++++++++++++++++++++++++++++++++++++ hw/block/nvme.c | 109 ++++++++++++++++++++++++++ hw/block/Makefile.objs | 2 +- hw/block/trace-events | 4 + 5 files changed, 288 insertions(+), 1 deletion(-) diff --git a/hw/block/nvme.h b/hw/block/nvme.h index XXXXXXX..XXXXXXX 100644 --- a/hw/block/nvme.h +++ b/hw/block/nvme.h @@ -XXX,XX +XXX,XX @@ typedef struct NvmeCtrl { uint64_t timestamp_set_qemu_clock_ms; /* QEMU clock time */ char *serial; + HostMemoryBackend *pmrdev; + NvmeNamespace *namespaces; NvmeSQueue **sq; NvmeCQueue **cq; diff --git a/include/block/nvme.h b/include/block/nvme.h index XXXXXXX..XXXXXXX 100644 --- a/include/block/nvme.h +++ b/include/block/nvme.h @@ -XXX,XX +XXX,XX @@ typedef struct NvmeBar { uint64_t acq; uint32_t cmbloc; uint32_t cmbsz; + uint8_t padding[3520]; /* not used by QEMU */ + uint32_t pmrcap; + uint32_t pmrctl; + uint32_t pmrsts; + uint32_t pmrebs; + uint32_t pmrswtp; + uint32_t pmrmsc; } NvmeBar; enum NvmeCapShift { @@ -XXX,XX +XXX,XX @@ enum NvmeCapShift { CAP_CSS_SHIFT = 37, CAP_MPSMIN_SHIFT = 48, CAP_MPSMAX_SHIFT = 52, + CAP_PMR_SHIFT = 56, }; enum NvmeCapMask { @@ -XXX,XX +XXX,XX @@ enum NvmeCapMask { CAP_CSS_MASK = 0xff, CAP_MPSMIN_MASK = 0xf, CAP_MPSMAX_MASK = 0xf, + CAP_PMR_MASK = 0x1, }; #define NVME_CAP_MQES(cap) (((cap) >> CAP_MQES_SHIFT) & CAP_MQES_MASK) @@ -XXX,XX +XXX,XX @@ enum NvmeCapMask { << CAP_MPSMIN_SHIFT) #define NVME_CAP_SET_MPSMAX(cap, val) (cap |= (uint64_t)(val & CAP_MPSMAX_MASK)\ << CAP_MPSMAX_SHIFT) +#define NVME_CAP_SET_PMRS(cap, val) (cap |= (uint64_t)(val & CAP_PMR_MASK)\ + << CAP_PMR_SHIFT) enum NvmeCcShift { CC_EN_SHIFT = 0, @@ -XXX,XX +XXX,XX @@ enum NvmeCmbszMask { #define NVME_CMBSZ_GETSIZE(cmbsz) \ (NVME_CMBSZ_SZ(cmbsz) * (1 << (12 + 4 * NVME_CMBSZ_SZU(cmbsz)))) +enum NvmePmrcapShift { + PMRCAP_RDS_SHIFT = 3, + PMRCAP_WDS_SHIFT = 4, + PMRCAP_BIR_SHIFT = 5, + PMRCAP_PMRTU_SHIFT = 8, + PMRCAP_PMRWBM_SHIFT = 10, + PMRCAP_PMRTO_SHIFT = 16, + PMRCAP_CMSS_SHIFT = 24, +}; + +enum NvmePmrcapMask { + PMRCAP_RDS_MASK = 0x1, + PMRCAP_WDS_MASK = 0x1, + PMRCAP_BIR_MASK = 0x7, + PMRCAP_PMRTU_MASK = 0x3, + PMRCAP_PMRWBM_MASK = 0xf, + PMRCAP_PMRTO_MASK = 0xff, + PMRCAP_CMSS_MASK = 0x1, +}; + +#define NVME_PMRCAP_RDS(pmrcap) \ + ((pmrcap >> PMRCAP_RDS_SHIFT) & PMRCAP_RDS_MASK) +#define NVME_PMRCAP_WDS(pmrcap) \ + ((pmrcap >> PMRCAP_WDS_SHIFT) & PMRCAP_WDS_MASK) +#define NVME_PMRCAP_BIR(pmrcap) \ + ((pmrcap >> PMRCAP_BIR_SHIFT) & PMRCAP_BIR_MASK) +#define NVME_PMRCAP_PMRTU(pmrcap) \ + ((pmrcap >> PMRCAP_PMRTU_SHIFT) & PMRCAP_PMRTU_MASK) +#define NVME_PMRCAP_PMRWBM(pmrcap) \ + ((pmrcap >> PMRCAP_PMRWBM_SHIFT) & PMRCAP_PMRWBM_MASK) +#define NVME_PMRCAP_PMRTO(pmrcap) \ + ((pmrcap >> PMRCAP_PMRTO_SHIFT) & PMRCAP_PMRTO_MASK) +#define NVME_PMRCAP_CMSS(pmrcap) \ + ((pmrcap >> PMRCAP_CMSS_SHIFT) & PMRCAP_CMSS_MASK) + +#define NVME_PMRCAP_SET_RDS(pmrcap, val) \ + (pmrcap |= (uint64_t)(val & PMRCAP_RDS_MASK) << PMRCAP_RDS_SHIFT) +#define NVME_PMRCAP_SET_WDS(pmrcap, val) \ + (pmrcap |= (uint64_t)(val & PMRCAP_WDS_MASK) << PMRCAP_WDS_SHIFT) +#define NVME_PMRCAP_SET_BIR(pmrcap, val) \ + (pmrcap |= (uint64_t)(val & PMRCAP_BIR_MASK) << PMRCAP_BIR_SHIFT) +#define NVME_PMRCAP_SET_PMRTU(pmrcap, val) \ + (pmrcap |= (uint64_t)(val & PMRCAP_PMRTU_MASK) << PMRCAP_PMRTU_SHIFT) +#define NVME_PMRCAP_SET_PMRWBM(pmrcap, val) \ + (pmrcap |= (uint64_t)(val & PMRCAP_PMRWBM_MASK) << PMRCAP_PMRWBM_SHIFT) +#define NVME_PMRCAP_SET_PMRTO(pmrcap, val) \ + (pmrcap |= (uint64_t)(val & PMRCAP_PMRTO_MASK) << PMRCAP_PMRTO_SHIFT) +#define NVME_PMRCAP_SET_CMSS(pmrcap, val) \ + (pmrcap |= (uint64_t)(val & PMRCAP_CMSS_MASK) << PMRCAP_CMSS_SHIFT) + +enum NvmePmrctlShift { + PMRCTL_EN_SHIFT = 0, +}; + +enum NvmePmrctlMask { + PMRCTL_EN_MASK = 0x1, +}; + +#define NVME_PMRCTL_EN(pmrctl) ((pmrctl >> PMRCTL_EN_SHIFT) & PMRCTL_EN_MASK) + +#define NVME_PMRCTL_SET_EN(pmrctl, val) \ + (pmrctl |= (uint64_t)(val & PMRCTL_EN_MASK) << PMRCTL_EN_SHIFT) + +enum NvmePmrstsShift { + PMRSTS_ERR_SHIFT = 0, + PMRSTS_NRDY_SHIFT = 8, + PMRSTS_HSTS_SHIFT = 9, + PMRSTS_CBAI_SHIFT = 12, +}; + +enum NvmePmrstsMask { + PMRSTS_ERR_MASK = 0xff, + PMRSTS_NRDY_MASK = 0x1, + PMRSTS_HSTS_MASK = 0x7, + PMRSTS_CBAI_MASK = 0x1, +}; + +#define NVME_PMRSTS_ERR(pmrsts) \ + ((pmrsts >> PMRSTS_ERR_SHIFT) & PMRSTS_ERR_MASK) +#define NVME_PMRSTS_NRDY(pmrsts) \ + ((pmrsts >> PMRSTS_NRDY_SHIFT) & PMRSTS_NRDY_MASK) +#define NVME_PMRSTS_HSTS(pmrsts) \ + ((pmrsts >> PMRSTS_HSTS_SHIFT) & PMRSTS_HSTS_MASK) +#define NVME_PMRSTS_CBAI(pmrsts) \ + ((pmrsts >> PMRSTS_CBAI_SHIFT) & PMRSTS_CBAI_MASK) + +#define NVME_PMRSTS_SET_ERR(pmrsts, val) \ + (pmrsts |= (uint64_t)(val & PMRSTS_ERR_MASK) << PMRSTS_ERR_SHIFT) +#define NVME_PMRSTS_SET_NRDY(pmrsts, val) \ + (pmrsts |= (uint64_t)(val & PMRSTS_NRDY_MASK) << PMRSTS_NRDY_SHIFT) +#define NVME_PMRSTS_SET_HSTS(pmrsts, val) \ + (pmrsts |= (uint64_t)(val & PMRSTS_HSTS_MASK) << PMRSTS_HSTS_SHIFT) +#define NVME_PMRSTS_SET_CBAI(pmrsts, val) \ + (pmrsts |= (uint64_t)(val & PMRSTS_CBAI_MASK) << PMRSTS_CBAI_SHIFT) + +enum NvmePmrebsShift { + PMREBS_PMRSZU_SHIFT = 0, + PMREBS_RBB_SHIFT = 4, + PMREBS_PMRWBZ_SHIFT = 8, +}; + +enum NvmePmrebsMask { + PMREBS_PMRSZU_MASK = 0xf, + PMREBS_RBB_MASK = 0x1, + PMREBS_PMRWBZ_MASK = 0xffffff, +}; + +#define NVME_PMREBS_PMRSZU(pmrebs) \ + ((pmrebs >> PMREBS_PMRSZU_SHIFT) & PMREBS_PMRSZU_MASK) +#define NVME_PMREBS_RBB(pmrebs) \ + ((pmrebs >> PMREBS_RBB_SHIFT) & PMREBS_RBB_MASK) +#define NVME_PMREBS_PMRWBZ(pmrebs) \ + ((pmrebs >> PMREBS_PMRWBZ_SHIFT) & PMREBS_PMRWBZ_MASK) + +#define NVME_PMREBS_SET_PMRSZU(pmrebs, val) \ + (pmrebs |= (uint64_t)(val & PMREBS_PMRSZU_MASK) << PMREBS_PMRSZU_SHIFT) +#define NVME_PMREBS_SET_RBB(pmrebs, val) \ + (pmrebs |= (uint64_t)(val & PMREBS_RBB_MASK) << PMREBS_RBB_SHIFT) +#define NVME_PMREBS_SET_PMRWBZ(pmrebs, val) \ + (pmrebs |= (uint64_t)(val & PMREBS_PMRWBZ_MASK) << PMREBS_PMRWBZ_SHIFT) + +enum NvmePmrswtpShift { + PMRSWTP_PMRSWTU_SHIFT = 0, + PMRSWTP_PMRSWTV_SHIFT = 8, +}; + +enum NvmePmrswtpMask { + PMRSWTP_PMRSWTU_MASK = 0xf, + PMRSWTP_PMRSWTV_MASK = 0xffffff, +}; + +#define NVME_PMRSWTP_PMRSWTU(pmrswtp) \ + ((pmrswtp >> PMRSWTP_PMRSWTU_SHIFT) & PMRSWTP_PMRSWTU_MASK) +#define NVME_PMRSWTP_PMRSWTV(pmrswtp) \ + ((pmrswtp >> PMRSWTP_PMRSWTV_SHIFT) & PMRSWTP_PMRSWTV_MASK) + +#define NVME_PMRSWTP_SET_PMRSWTU(pmrswtp, val) \ + (pmrswtp |= (uint64_t)(val & PMRSWTP_PMRSWTU_MASK) << PMRSWTP_PMRSWTU_SHIFT) +#define NVME_PMRSWTP_SET_PMRSWTV(pmrswtp, val) \ + (pmrswtp |= (uint64_t)(val & PMRSWTP_PMRSWTV_MASK) << PMRSWTP_PMRSWTV_SHIFT) + +enum NvmePmrmscShift { + PMRMSC_CMSE_SHIFT = 1, + PMRMSC_CBA_SHIFT = 12, +}; + +enum NvmePmrmscMask { + PMRMSC_CMSE_MASK = 0x1, + PMRMSC_CBA_MASK = 0xfffffffffffff, +}; + +#define NVME_PMRMSC_CMSE(pmrmsc) \ + ((pmrmsc >> PMRMSC_CMSE_SHIFT) & PMRMSC_CMSE_MASK) +#define NVME_PMRMSC_CBA(pmrmsc) \ + ((pmrmsc >> PMRMSC_CBA_SHIFT) & PMRMSC_CBA_MASK) + +#define NVME_PMRMSC_SET_CMSE(pmrmsc, val) \ + (pmrmsc |= (uint64_t)(val & PMRMSC_CMSE_MASK) << PMRMSC_CMSE_SHIFT) +#define NVME_PMRMSC_SET_CBA(pmrmsc, val) \ + (pmrmsc |= (uint64_t)(val & PMRMSC_CBA_MASK) << PMRMSC_CBA_SHIFT) + typedef struct NvmeCmd { uint8_t opcode; uint8_t fuse; diff --git a/hw/block/nvme.c b/hw/block/nvme.c index XXXXXXX..XXXXXXX 100644 --- a/hw/block/nvme.c +++ b/hw/block/nvme.c @@ -XXX,XX +XXX,XX @@ * -drive file=<file>,if=none,id=<drive_id> * -device nvme,drive=<drive_id>,serial=<serial>,id=<id[optional]>, \ * cmb_size_mb=<cmb_size_mb[optional]>, \ + * [pmrdev=<mem_backend_file_id>,] \ * num_queues=<N[optional]> * * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. + * + * cmb_size_mb= and pmrdev= options are mutually exclusive due to limitation + * in available BAR's. cmb_size_mb= will take precedence over pmrdev= when + * both provided. + * Enabling pmr emulation can be achieved by pointing to memory-backend-file. + * For example: + * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \ + * size=<size> .... -device nvme,...,pmrdev=<mem_id> */ #include "qemu/osdep.h" @@ -XXX,XX +XXX,XX @@ #include "sysemu/sysemu.h" #include "qapi/error.h" #include "qapi/visitor.h" +#include "sysemu/hostmem.h" #include "sysemu/block-backend.h" +#include "exec/ram_addr.h" #include "qemu/log.h" #include "qemu/module.h" @@ -XXX,XX +XXX,XX @@ static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data, NVME_GUEST_ERR(nvme_ub_mmiowr_cmbsz_readonly, "invalid write to read only CMBSZ, ignored"); return; + case 0xE00: /* PMRCAP */ + NVME_GUEST_ERR(nvme_ub_mmiowr_pmrcap_readonly, + "invalid write to PMRCAP register, ignored"); + return; + case 0xE04: /* TODO PMRCTL */ + break; + case 0xE08: /* PMRSTS */ + NVME_GUEST_ERR(nvme_ub_mmiowr_pmrsts_readonly, + "invalid write to PMRSTS register, ignored"); + return; + case 0xE0C: /* PMREBS */ + NVME_GUEST_ERR(nvme_ub_mmiowr_pmrebs_readonly, + "invalid write to PMREBS register, ignored"); + return; + case 0xE10: /* PMRSWTP */ + NVME_GUEST_ERR(nvme_ub_mmiowr_pmrswtp_readonly, + "invalid write to PMRSWTP register, ignored"); + return; + case 0xE14: /* TODO PMRMSC */ + break; default: NVME_GUEST_ERR(nvme_ub_mmiowr_invalid, "invalid MMIO write," @@ -XXX,XX +XXX,XX @@ static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size) } if (addr < sizeof(n->bar)) { + /* + * When PMRWBM bit 1 is set then read from + * from PMRSTS should ensure prior writes + * made it to persistent media + */ + if (addr == 0xE08 && + (NVME_PMRCAP_PMRWBM(n->bar.pmrcap) & 0x02)) { + qemu_ram_writeback(n->pmrdev->mr.ram_block, + 0, n->pmrdev->size); + } memcpy(&val, ptr + addr, size); } else { NVME_GUEST_ERR(nvme_ub_mmiord_invalid_ofs, @@ -XXX,XX +XXX,XX @@ static void nvme_realize(PCIDevice *pci_dev, Error **errp) error_setg(errp, "serial property not set"); return; } + + if (!n->cmb_size_mb && n->pmrdev) { + if (host_memory_backend_is_mapped(n->pmrdev)) { + char *path = object_get_canonical_path_component(OBJECT(n->pmrdev)); + error_setg(errp, "can't use already busy memdev: %s", path); + g_free(path); + return; + } + + if (!is_power_of_2(n->pmrdev->size)) { + error_setg(errp, "pmr backend size needs to be power of 2 in size"); + return; + } + + host_memory_backend_set_mapped(n->pmrdev, true); + } + blkconf_blocksizes(&n->conf); if (!blkconf_apply_backend_options(&n->conf, blk_is_read_only(n->conf.blk), false, errp)) { @@ -XXX,XX +XXX,XX @@ static void nvme_realize(PCIDevice *pci_dev, Error **errp) PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64 | PCI_BASE_ADDRESS_MEM_PREFETCH, &n->ctrl_mem); + } else if (n->pmrdev) { + /* Controller Capabilities register */ + NVME_CAP_SET_PMRS(n->bar.cap, 1); + + /* PMR Capabities register */ + n->bar.pmrcap = 0; + NVME_PMRCAP_SET_RDS(n->bar.pmrcap, 0); + NVME_PMRCAP_SET_WDS(n->bar.pmrcap, 0); + NVME_PMRCAP_SET_BIR(n->bar.pmrcap, 2); + NVME_PMRCAP_SET_PMRTU(n->bar.pmrcap, 0); + /* Turn on bit 1 support */ + NVME_PMRCAP_SET_PMRWBM(n->bar.pmrcap, 0x02); + NVME_PMRCAP_SET_PMRTO(n->bar.pmrcap, 0); + NVME_PMRCAP_SET_CMSS(n->bar.pmrcap, 0); + + /* PMR Control register */ + n->bar.pmrctl = 0; + NVME_PMRCTL_SET_EN(n->bar.pmrctl, 0); + + /* PMR Status register */ + n->bar.pmrsts = 0; + NVME_PMRSTS_SET_ERR(n->bar.pmrsts, 0); + NVME_PMRSTS_SET_NRDY(n->bar.pmrsts, 0); + NVME_PMRSTS_SET_HSTS(n->bar.pmrsts, 0); + NVME_PMRSTS_SET_CBAI(n->bar.pmrsts, 0); + + /* PMR Elasticity Buffer Size register */ + n->bar.pmrebs = 0; + NVME_PMREBS_SET_PMRSZU(n->bar.pmrebs, 0); + NVME_PMREBS_SET_RBB(n->bar.pmrebs, 0); + NVME_PMREBS_SET_PMRWBZ(n->bar.pmrebs, 0); + + /* PMR Sustained Write Throughput register */ + n->bar.pmrswtp = 0; + NVME_PMRSWTP_SET_PMRSWTU(n->bar.pmrswtp, 0); + NVME_PMRSWTP_SET_PMRSWTV(n->bar.pmrswtp, 0); + + /* PMR Memory Space Control register */ + n->bar.pmrmsc = 0; + NVME_PMRMSC_SET_CMSE(n->bar.pmrmsc, 0); + NVME_PMRMSC_SET_CBA(n->bar.pmrmsc, 0); + + pci_register_bar(pci_dev, NVME_PMRCAP_BIR(n->bar.pmrcap), + PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64 | + PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmrdev->mr); } for (i = 0; i < n->num_namespaces; i++) { @@ -XXX,XX +XXX,XX @@ static void nvme_exit(PCIDevice *pci_dev) if (n->cmb_size_mb) { g_free(n->cmbuf); } + + if (n->pmrdev) { + host_memory_backend_set_mapped(n->pmrdev, false); + } msix_uninit_exclusive_bar(pci_dev); } static Property nvme_props[] = { DEFINE_BLOCK_PROPERTIES(NvmeCtrl, conf), + DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmrdev, TYPE_MEMORY_BACKEND, + HostMemoryBackend *), DEFINE_PROP_STRING("serial", NvmeCtrl, serial), DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, cmb_size_mb, 0), DEFINE_PROP_UINT32("num_queues", NvmeCtrl, num_queues, 64), diff --git a/hw/block/Makefile.objs b/hw/block/Makefile.objs index XXXXXXX..XXXXXXX 100644 --- a/hw/block/Makefile.objs +++ b/hw/block/Makefile.objs @@ -XXX,XX +XXX,XX @@ common-obj-$(CONFIG_PFLASH_CFI02) += pflash_cfi02.o common-obj-$(CONFIG_XEN) += xen-block.o common-obj-$(CONFIG_ECC) += ecc.o common-obj-$(CONFIG_ONENAND) += onenand.o -common-obj-$(CONFIG_NVME_PCI) += nvme.o common-obj-$(CONFIG_SWIM) += swim.o common-obj-$(CONFIG_SH4) += tc58128.o obj-$(CONFIG_VIRTIO_BLK) += virtio-blk.o obj-$(CONFIG_VHOST_USER_BLK) += vhost-user-blk.o +obj-$(CONFIG_NVME_PCI) += nvme.o obj-y += dataplane/ diff --git a/hw/block/trace-events b/hw/block/trace-events index XXXXXXX..XXXXXXX 100644 --- a/hw/block/trace-events +++ b/hw/block/trace-events @@ -XXX,XX +XXX,XX @@ nvme_ub_mmiowr_ssreset_w1c_unsupported(void) "attempted to W1C CSTS.NSSRO but CA nvme_ub_mmiowr_ssreset_unsupported(void) "attempted NVM subsystem reset but CAP.NSSRS is zero (not supported)" nvme_ub_mmiowr_cmbloc_reserved(void) "invalid write to reserved CMBLOC when CMBSZ is zero, ignored" nvme_ub_mmiowr_cmbsz_readonly(void) "invalid write to read only CMBSZ, ignored" +nvme_ub_mmiowr_pmrcap_readonly(void) "invalid write to read only PMRCAP, ignored" +nvme_ub_mmiowr_pmrsts_readonly(void) "invalid write to read only PMRSTS, ignored" +nvme_ub_mmiowr_pmrebs_readonly(void) "invalid write to read only PMREBS, ignored" +nvme_ub_mmiowr_pmrswtp_readonly(void) "invalid write to read only PMRSWTP, ignored" nvme_ub_mmiowr_invalid(uint64_t offset, uint64_t data) "invalid MMIO write, offset=0x%"PRIx64", data=0x%"PRIx64"" nvme_ub_mmiord_misaligned32(uint64_t offset) "MMIO read not 32-bit aligned, offset=0x%"PRIx64"" nvme_ub_mmiord_toosmall(uint64_t offset) "MMIO read smaller than 32-bits, offset=0x%"PRIx64"" -- 2.25.3 The QMP handler qmp_object_add() and the implementation of --object in qemu-storage-daemon can share most of the code. Currently, qemu-storage-daemon calls qmp_object_add(), but this is not correct because different visitors need to be used. As a first step towards a fix, make qmp_object_add() a wrapper around a new function user_creatable_add_dict() that can get an additional parameter. The handling of "props" is only required for compatibility and not required for the qemu-storage-daemon command line, so it stays in qmp_object_add(). Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/qom/object_interfaces.h | 12 ++++++++++++ qom/object_interfaces.c | 27 +++++++++++++++++++++++++++ qom/qom-qmp-cmds.c | 24 +----------------------- 3 files changed, 40 insertions(+), 23 deletions(-) diff --git a/include/qom/object_interfaces.h b/include/qom/object_interfaces.h index XXXXXXX..XXXXXXX 100644 --- a/include/qom/object_interfaces.h +++ b/include/qom/object_interfaces.h @@ -XXX,XX +XXX,XX @@ Object *user_creatable_add_type(const char *type, const char *id, const QDict *qdict, Visitor *v, Error **errp); +/** + * user_creatable_add_dict: + * @qdict: the object definition + * @errp: if an error occurs, a pointer to an area to store the error + * + * Create an instance of the user creatable object that is defined by + * @qdict. The object type is taken from the QDict key 'qom-type', its + * ID from the key 'id'. The remaining entries in @qdict are used to + * initialize the object properties. + */ +void user_creatable_add_dict(QDict *qdict, Error **errp); + /** * user_creatable_add_opts: * @opts: the object definition diff --git a/qom/object_interfaces.c b/qom/object_interfaces.c index XXXXXXX..XXXXXXX 100644 --- a/qom/object_interfaces.c +++ b/qom/object_interfaces.c @@ -XXX,XX +XXX,XX @@ #include "qapi/qmp/qerror.h" #include "qapi/qmp/qjson.h" #include "qapi/qmp/qstring.h" +#include "qapi/qobject-input-visitor.h" #include "qom/object_interfaces.h" #include "qemu/help_option.h" #include "qemu/module.h" @@ -XXX,XX +XXX,XX @@ out: return obj; } +void user_creatable_add_dict(QDict *qdict, Error **errp) +{ + Visitor *v; + Object *obj; + g_autofree char *type = NULL; + g_autofree char *id = NULL; + + type = g_strdup(qdict_get_try_str(qdict, "qom-type")); + if (!type) { + error_setg(errp, QERR_MISSING_PARAMETER, "qom-type"); + return; + } + qdict_del(qdict, "qom-type"); + + id = g_strdup(qdict_get_try_str(qdict, "id")); + if (!id) { + error_setg(errp, QERR_MISSING_PARAMETER, "id"); + return; + } + qdict_del(qdict, "id"); + + v = qobject_input_visitor_new(QOBJECT(qdict)); + obj = user_creatable_add_type(type, id, qdict, v, errp); + visit_free(v); + object_unref(obj); +} Object *user_creatable_add_opts(QemuOpts *opts, Error **errp) { diff --git a/qom/qom-qmp-cmds.c b/qom/qom-qmp-cmds.c index XXXXXXX..XXXXXXX 100644 --- a/qom/qom-qmp-cmds.c +++ b/qom/qom-qmp-cmds.c @@ -XXX,XX +XXX,XX @@ #include "qapi/qapi-commands-qom.h" #include "qapi/qmp/qdict.h" #include "qapi/qmp/qerror.h" -#include "qapi/qobject-input-visitor.h" #include "qemu/cutils.h" #include "qom/object_interfaces.h" #include "qom/qom-qobject.h" @@ -XXX,XX +XXX,XX @@ void qmp_object_add(QDict *qdict, QObject **ret_data, Error **errp) { QObject *props; QDict *pdict; - Visitor *v; - Object *obj; - g_autofree char *type = NULL; - g_autofree char *id = NULL; - - type = g_strdup(qdict_get_try_str(qdict, "qom-type")); - if (!type) { - error_setg(errp, QERR_MISSING_PARAMETER, "qom-type"); - return; - } - qdict_del(qdict, "qom-type"); - - id = g_strdup(qdict_get_try_str(qdict, "id")); - if (!id) { - error_setg(errp, QERR_MISSING_PARAMETER, "id"); - return; - } - qdict_del(qdict, "id"); props = qdict_get(qdict, "props"); if (props) { @@ -XXX,XX +XXX,XX @@ void qmp_object_add(QDict *qdict, QObject **ret_data, Error **errp) qobject_unref(pdict); } - v = qobject_input_visitor_new(QOBJECT(qdict)); - obj = user_creatable_add_type(type, id, qdict, v, errp); - visit_free(v); - object_unref(obj); + user_creatable_add_dict(qdict, errp); } void qmp_object_del(const char *id, Error **errp) -- 2.25.3 After processing the option string with the keyval parser, we get a QDict that contains only strings. This QDict must be fed to a keyval visitor which converts the strings into the right data types. qmp_object_add(), however, uses the normal QObject input visitor, which expects a QDict where all properties already have the QType that matches the data type required by the QOM object type. Change the --object implementation in qemu-storage-daemon so that it doesn't call qmp_object_add(), but calls user_creatable_add_dict() directly instead and pass it a new keyval boolean that decides which visitor must be used. Reported-by: Coiby Xu <coiby.xu@gmail.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- include/qom/object_interfaces.h | 6 +++++- qemu-storage-daemon.c | 4 +--- qom/object_interfaces.c | 8 ++++++-- qom/qom-qmp-cmds.c | 2 +- 4 files changed, 13 insertions(+), 7 deletions(-) diff --git a/include/qom/object_interfaces.h b/include/qom/object_interfaces.h index XXXXXXX..XXXXXXX 100644 --- a/include/qom/object_interfaces.h +++ b/include/qom/object_interfaces.h @@ -XXX,XX +XXX,XX @@ Object *user_creatable_add_type(const char *type, const char *id, /** * user_creatable_add_dict: * @qdict: the object definition + * @keyval: if true, use a keyval visitor for processing @qdict (i.e. + * assume that all @qdict values are strings); otherwise, use + * the normal QObject visitor (i.e. assume all @qdict values + * have the QType expected by the QOM object type) * @errp: if an error occurs, a pointer to an area to store the error * * Create an instance of the user creatable object that is defined by @@ -XXX,XX +XXX,XX @@ Object *user_creatable_add_type(const char *type, const char *id, * ID from the key 'id'. The remaining entries in @qdict are used to * initialize the object properties. */ -void user_creatable_add_dict(QDict *qdict, Error **errp); +void user_creatable_add_dict(QDict *qdict, bool keyval, Error **errp); /** * user_creatable_add_opts: diff --git a/qemu-storage-daemon.c b/qemu-storage-daemon.c index XXXXXXX..XXXXXXX 100644 --- a/qemu-storage-daemon.c +++ b/qemu-storage-daemon.c @@ -XXX,XX +XXX,XX @@ static void process_options(int argc, char *argv[]) QemuOpts *opts; const char *type; QDict *args; - QObject *ret_data = NULL; /* FIXME The keyval parser rejects 'help' arguments, so we must * unconditionall try QemuOpts first. */ @@ -XXX,XX +XXX,XX @@ static void process_options(int argc, char *argv[]) qemu_opts_del(opts); args = keyval_parse(optarg, "qom-type", &error_fatal); - qmp_object_add(args, &ret_data, &error_fatal); + user_creatable_add_dict(args, true, &error_fatal); qobject_unref(args); - qobject_unref(ret_data); break; } default: diff --git a/qom/object_interfaces.c b/qom/object_interfaces.c index XXXXXXX..XXXXXXX 100644 --- a/qom/object_interfaces.c +++ b/qom/object_interfaces.c @@ -XXX,XX +XXX,XX @@ out: return obj; } -void user_creatable_add_dict(QDict *qdict, Error **errp) +void user_creatable_add_dict(QDict *qdict, bool keyval, Error **errp) { Visitor *v; Object *obj; @@ -XXX,XX +XXX,XX @@ void user_creatable_add_dict(QDict *qdict, Error **errp) } qdict_del(qdict, "id"); - v = qobject_input_visitor_new(QOBJECT(qdict)); + if (keyval) { + v = qobject_input_visitor_new_keyval(QOBJECT(qdict)); + } else { + v = qobject_input_visitor_new(QOBJECT(qdict)); + } obj = user_creatable_add_type(type, id, qdict, v, errp); visit_free(v); object_unref(obj); diff --git a/qom/qom-qmp-cmds.c b/qom/qom-qmp-cmds.c index XXXXXXX..XXXXXXX 100644 --- a/qom/qom-qmp-cmds.c +++ b/qom/qom-qmp-cmds.c @@ -XXX,XX +XXX,XX @@ void qmp_object_add(QDict *qdict, QObject **ret_data, Error **errp) qobject_unref(pdict); } - user_creatable_add_dict(qdict, errp); + user_creatable_add_dict(qdict, false, errp); } void qmp_object_del(const char *id, Error **errp) -- 2.25.3