Merge remote-tracking branch 'remotes/maxreitz/tags/pull-block-2020-12-18' into staging

Block patches:
- New block filter: preallocate (which, on writes beyond an image file's
  end, allocates big chunks of data so that such post-EOF writes will
  occur less frequently)
- write-zeroes and block-status support for Quorum
- Implementation of truncate for the nvme block driver similarly to the
  existing implementations for host block devices and iscsi devices
- Block layer refactoring: Drop the tighten_restrictions concept in the
  block permission functions
- iotest fixes

# gpg: Signature made Fri 18 Dec 2020 14:45:30 GMT
# gpg:                using RSA key 91BEB60A30DB3E8857D11829F407DB0061D5CF40
# gpg:                issuer "mreitz@redhat.com"
# gpg: Good signature from "Max Reitz <mreitz@redhat.com>" [full]
# Primary key fingerprint: 91BE B60A 30DB 3E88 57D1  1829 F407 DB00 61D5 CF40

* remotes/maxreitz/tags/pull-block-2020-12-18: (30 commits)
  iotests: Fix _send_qemu_cmd with bash 5.1
  iotests/102: Pass $QEMU_HANDLE to _send_qemu_cmd
  block/nvme: Implement fake truncate() coroutine
  quorum: Implement bdrv_co_pwrite_zeroes()
  quorum: Implement bdrv_co_block_status()
  scripts/simplebench: add bench_prealloc.py
  simplebench/results_to_text: make executable
  simplebench/results_to_text: add difference line to the table
  simplebench/results_to_text: improve view of the table
  simplebench: move results_to_text() into separate file
  simplebench: rename ascii() to results_to_text()
  scripts/simplebench: use standard deviation for +- error
  scripts/simplebench: support iops
  scripts/simplebench: fix grammar: s/successed/succeeded/
  iotests: add 298 to test new preallocate filter driver
  iotests.py: execute_setup_common(): add required_fmts argument
  iotests: qemu_io_silent: support --image-opts
  qemu-io: add preallocate mode parameter for truncate command
  block: introduce preallocate filter
  block: bdrv_check_perm(): process children anyway
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

6 files changed, 747 insertions, 57 deletions

diff --git a/block/file-posix.c b/block/file-posix.c
index 9804681..00cdaaa 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -2953,7 +2953,7 @@ raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
 
         assert(bdrv_check_request(req->offset, req->bytes) == 0);
 
-        bdrv_mark_request_serialising(req, bs->bl.request_alignment);
+        bdrv_make_request_serialising(req, bs->bl.request_alignment);
     }
 #endif
 
diff --git a/block/io.c b/block/io.c
index 24205f5..95b1c56 100644
--- a/block/io.c
+++ b/block/io.c
@@ -754,55 +754,65 @@ static bool tracked_request_overlaps(BdrvTrackedRequest *req,
     return true;
 }
 
-static bool coroutine_fn
-bdrv_wait_serialising_requests_locked(BlockDriverState *bs,
-                                      BdrvTrackedRequest *self)
+/* Called with self->bs->reqs_lock held */
+static BdrvTrackedRequest *
+bdrv_find_conflicting_request(BdrvTrackedRequest *self)
 {
     BdrvTrackedRequest *req;
-    bool retry;
-    bool waited = false;
 
-    do {
-        retry = false;
-        QLIST_FOREACH(req, &bs->tracked_requests, list) {
-            if (req == self || (!req->serialising && !self->serialising)) {
-                continue;
-            }
-            if (tracked_request_overlaps(req, self->overlap_offset,
-                                         self->overlap_bytes))
-            {
-                /* Hitting this means there was a reentrant request, for
-                 * example, a block driver issuing nested requests.  This must
-                 * never happen since it means deadlock.
-                 */
-                assert(qemu_coroutine_self() != req->co);
-
-                /* If the request is already (indirectly) waiting for us, or
-                 * will wait for us as soon as it wakes up, then just go on
-                 * (instead of producing a deadlock in the former case). */
-                if (!req->waiting_for) {
-                    self->waiting_for = req;
-                    qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock);
-                    self->waiting_for = NULL;
-                    retry = true;
-                    waited = true;
-                    break;
-                }
+    QLIST_FOREACH(req, &self->bs->tracked_requests, list) {
+        if (req == self || (!req->serialising && !self->serialising)) {
+            continue;
+        }
+        if (tracked_request_overlaps(req, self->overlap_offset,
+                                     self->overlap_bytes))
+        {
+            /*
+             * Hitting this means there was a reentrant request, for
+             * example, a block driver issuing nested requests.  This must
+             * never happen since it means deadlock.
+             */
+            assert(qemu_coroutine_self() != req->co);
+
+            /*
+             * If the request is already (indirectly) waiting for us, or
+             * will wait for us as soon as it wakes up, then just go on
+             * (instead of producing a deadlock in the former case).
+             */
+            if (!req->waiting_for) {
+                return req;
             }
         }
-    } while (retry);
+    }
+
+    return NULL;
+}
+
+/* Called with self->bs->reqs_lock held */
+static bool coroutine_fn
+bdrv_wait_serialising_requests_locked(BdrvTrackedRequest *self)
+{
+    BdrvTrackedRequest *req;
+    bool waited = false;
+
+    while ((req = bdrv_find_conflicting_request(self))) {
+        self->waiting_for = req;
+        qemu_co_queue_wait(&req->wait_queue, &self->bs->reqs_lock);
+        self->waiting_for = NULL;
+        waited = true;
+    }
+
     return waited;
 }
 
-bool bdrv_mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
+/* Called with req->bs->reqs_lock held */
+static void tracked_request_set_serialising(BdrvTrackedRequest *req,
+                                            uint64_t align)
 {
-    BlockDriverState *bs = req->bs;
     int64_t overlap_offset = req->offset & ~(align - 1);
     uint64_t overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
                                - overlap_offset;
-    bool waited;
 
-    qemu_co_mutex_lock(&bs->reqs_lock);
     if (!req->serialising) {
         qatomic_inc(&req->bs->serialising_in_flight);
         req->serialising = true;
@@ -810,9 +820,6 @@ bool bdrv_mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
 
     req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
     req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
-    waited = bdrv_wait_serialising_requests_locked(bs, req);
-    qemu_co_mutex_unlock(&bs->reqs_lock);
-    return waited;
 }
 
 /**
@@ -892,12 +899,27 @@ static bool coroutine_fn bdrv_wait_serialising_requests(BdrvTrackedRequest *self
     }
 
     qemu_co_mutex_lock(&bs->reqs_lock);
-    waited = bdrv_wait_serialising_requests_locked(bs, self);
+    waited = bdrv_wait_serialising_requests_locked(self);
     qemu_co_mutex_unlock(&bs->reqs_lock);
 
     return waited;
 }
 
+bool coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req,
+                                                uint64_t align)
+{
+    bool waited;
+
+    qemu_co_mutex_lock(&req->bs->reqs_lock);
+
+    tracked_request_set_serialising(req, align);
+    waited = bdrv_wait_serialising_requests_locked(req);
+
+    qemu_co_mutex_unlock(&req->bs->reqs_lock);
+
+    return waited;
+}
+
 int bdrv_check_request(int64_t offset, int64_t bytes)
 {
     if (offset < 0 || bytes < 0) {
@@ -1423,7 +1445,7 @@ static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child,
          * with each other for the same cluster.  For example, in copy-on-read
          * it ensures that the CoR read and write operations are atomic and
          * guest writes cannot interleave between them. */
-        bdrv_mark_request_serialising(req, bdrv_get_cluster_size(bs));
+        bdrv_make_request_serialising(req, bdrv_get_cluster_size(bs));
     } else {
         bdrv_wait_serialising_requests(req);
     }
@@ -1827,7 +1849,6 @@ bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes,
                           BdrvTrackedRequest *req, int flags)
 {
     BlockDriverState *bs = child->bs;
-    bool waited;
     int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
 
     if (bs->read_only) {
@@ -1837,17 +1858,18 @@ bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes,
     assert(!(bs->open_flags & BDRV_O_INACTIVE));
     assert((bs->open_flags & BDRV_O_NO_IO) == 0);
     assert(!(flags & ~BDRV_REQ_MASK));
+    assert(!((flags & BDRV_REQ_NO_WAIT) && !(flags & BDRV_REQ_SERIALISING)));
 
     if (flags & BDRV_REQ_SERIALISING) {
-        waited = bdrv_mark_request_serialising(req, bdrv_get_cluster_size(bs));
-        /*
-         * For a misaligned request we should have already waited earlier,
-         * because we come after bdrv_padding_rmw_read which must be called
-         * with the request already marked as serialising.
-         */
-        assert(!waited ||
-               (req->offset == req->overlap_offset &&
-                req->bytes == req->overlap_bytes));
+        QEMU_LOCK_GUARD(&bs->reqs_lock);
+
+        tracked_request_set_serialising(req, bdrv_get_cluster_size(bs));
+
+        if ((flags & BDRV_REQ_NO_WAIT) && bdrv_find_conflicting_request(req)) {
+            return -EBUSY;
+        }
+
+        bdrv_wait_serialising_requests_locked(req);
     } else {
         bdrv_wait_serialising_requests(req);
     }
@@ -2013,7 +2035,7 @@ static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child,
 
     padding = bdrv_init_padding(bs, offset, bytes, &pad);
     if (padding) {
-        bdrv_mark_request_serialising(req, align);
+        bdrv_make_request_serialising(req, align);
 
         bdrv_padding_rmw_read(child, req, &pad, true);
 
@@ -2127,7 +2149,7 @@ int coroutine_fn bdrv_co_pwritev_part(BdrvChild *child,
     }
 
     if (bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, &pad)) {
-        bdrv_mark_request_serialising(&req, align);
+        bdrv_make_request_serialising(&req, align);
         bdrv_padding_rmw_read(child, &req, &pad, false);
     }
 
@@ -3248,7 +3270,7 @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact,
      * new area, we need to make sure that no write requests are made to it
      * concurrently or they might be overwritten by preallocation. */
     if (new_bytes) {
-        bdrv_mark_request_serialising(&req, 1);
+        bdrv_make_request_serialising(&req, 1);
     }
     if (bs->read_only) {
         error_setg(errp, "Image is read-only");
diff --git a/block/meson.build b/block/meson.build
index 5dcc1e5..7595d86 100644
--- a/block/meson.build
+++ b/block/meson.build
@@ -12,6 +12,7 @@ block_ss.add(files(
   'block-copy.c',
   'commit.c',
   'copy-on-read.c',
+  'preallocate.c',
   'create.c',
   'crypto.c',
   'dirty-bitmap.c',
diff --git a/block/nvme.c b/block/nvme.c
index a06a188..5a6fbac 100644
--- a/block/nvme.c
+++ b/block/nvme.c
@@ -1389,6 +1389,29 @@ out:
 
 }
 
+static int coroutine_fn nvme_co_truncate(BlockDriverState *bs, int64_t offset,
+                                         bool exact, PreallocMode prealloc,
+                                         BdrvRequestFlags flags, Error **errp)
+{
+    int64_t cur_length;
+
+    if (prealloc != PREALLOC_MODE_OFF) {
+        error_setg(errp, "Unsupported preallocation mode '%s'",
+                   PreallocMode_str(prealloc));
+        return -ENOTSUP;
+    }
+
+    cur_length = nvme_getlength(bs);
+    if (offset != cur_length && exact) {
+        error_setg(errp, "Cannot resize NVMe devices");
+        return -ENOTSUP;
+    } else if (offset > cur_length) {
+        error_setg(errp, "Cannot grow NVMe devices");
+        return -EINVAL;
+    }
+
+    return 0;
+}
 
 static int nvme_reopen_prepare(BDRVReopenState *reopen_state,
                                BlockReopenQueue *queue, Error **errp)
@@ -1523,6 +1546,7 @@ static BlockDriver bdrv_nvme = {
     .bdrv_close               = nvme_close,
     .bdrv_getlength           = nvme_getlength,
     .bdrv_probe_blocksizes    = nvme_probe_blocksizes,
+    .bdrv_co_truncate         = nvme_co_truncate,
 
     .bdrv_co_preadv           = nvme_co_preadv,
     .bdrv_co_pwritev          = nvme_co_pwritev,
diff --git a/block/preallocate.c b/block/preallocate.c
new file mode 100644
index 0000000..b619206
--- /dev/null
+++ b/block/preallocate.c
@@ -0,0 +1,559 @@
+/*
+ * preallocate filter driver
+ *
+ * The driver performs preallocate operation: it is injected above
+ * some node, and before each write over EOF it does additional preallocating
+ * write-zeroes request.
+ *
+ * Copyright (c) 2020 Virtuozzo International GmbH.
+ *
+ * Author:
+ *  Sementsov-Ogievskiy Vladimir <vsementsov@virtuozzo.com>
+ *
+ * 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/>.
+ */
+
+#include "qemu/osdep.h"
+
+#include "qapi/error.h"
+#include "qemu/module.h"
+#include "qemu/option.h"
+#include "qemu/units.h"
+#include "block/block_int.h"
+
+
+typedef struct PreallocateOpts {
+    int64_t prealloc_size;
+    int64_t prealloc_align;
+} PreallocateOpts;
+
+typedef struct BDRVPreallocateState {
+    PreallocateOpts opts;
+
+    /*
+     * Track real data end, to crop preallocation on close. If < 0 the status is
+     * unknown.
+     *
+     * @data_end is a maximum of file size on open (or when we get write/resize
+     * permissions) and all write request ends after it. So it's safe to
+     * truncate to data_end if it is valid.
+     */
+    int64_t data_end;
+
+    /*
+     * Start of trailing preallocated area which reads as zero. May be smaller
+     * than data_end, if user does over-EOF write zero operation. If < 0 the
+     * status is unknown.
+     *
+     * If both @zero_start and @file_end are valid, the region
+     * [@zero_start, @file_end) is known to be preallocated zeroes. If @file_end
+     * is not valid, @zero_start doesn't make much sense.
+     */
+    int64_t zero_start;
+
+    /*
+     * Real end of file. Actually the cache for bdrv_getlength(bs->file->bs),
+     * to avoid extra lseek() calls on each write operation. If < 0 the status
+     * is unknown.
+     */
+    int64_t file_end;
+
+    /*
+     * All three states @data_end, @zero_start and @file_end are guaranteed to
+     * be invalid (< 0) when we don't have both exclusive BLK_PERM_RESIZE and
+     * BLK_PERM_WRITE permissions on file child.
+     */
+} BDRVPreallocateState;
+
+#define PREALLOCATE_OPT_PREALLOC_ALIGN "prealloc-align"
+#define PREALLOCATE_OPT_PREALLOC_SIZE "prealloc-size"
+static QemuOptsList runtime_opts = {
+    .name = "preallocate",
+    .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
+    .desc = {
+        {
+            .name = PREALLOCATE_OPT_PREALLOC_ALIGN,
+            .type = QEMU_OPT_SIZE,
+            .help = "on preallocation, align file length to this number, "
+                "default 1M",
+        },
+        {
+            .name = PREALLOCATE_OPT_PREALLOC_SIZE,
+            .type = QEMU_OPT_SIZE,
+            .help = "how much to preallocate, default 128M",
+        },
+        { /* end of list */ }
+    },
+};
+
+static bool preallocate_absorb_opts(PreallocateOpts *dest, QDict *options,
+                                    BlockDriverState *child_bs, Error **errp)
+{
+    QemuOpts *opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
+
+    if (!qemu_opts_absorb_qdict(opts, options, errp)) {
+        return false;
+    }
+
+    dest->prealloc_align =
+        qemu_opt_get_size(opts, PREALLOCATE_OPT_PREALLOC_ALIGN, 1 * MiB);
+    dest->prealloc_size =
+        qemu_opt_get_size(opts, PREALLOCATE_OPT_PREALLOC_SIZE, 128 * MiB);
+
+    qemu_opts_del(opts);
+
+    if (!QEMU_IS_ALIGNED(dest->prealloc_align, BDRV_SECTOR_SIZE)) {
+        error_setg(errp, "prealloc-align parameter of preallocate filter "
+                   "is not aligned to %llu", BDRV_SECTOR_SIZE);
+        return false;
+    }
+
+    if (!QEMU_IS_ALIGNED(dest->prealloc_align,
+                         child_bs->bl.request_alignment)) {
+        error_setg(errp, "prealloc-align parameter of preallocate filter "
+                   "is not aligned to underlying node request alignment "
+                   "(%" PRIi32 ")", child_bs->bl.request_alignment);
+        return false;
+    }
+
+    return true;
+}
+
+static int preallocate_open(BlockDriverState *bs, QDict *options, int flags,
+                            Error **errp)
+{
+    BDRVPreallocateState *s = bs->opaque;
+
+    /*
+     * s->data_end and friends should be initialized on permission update.
+     * For this to work, mark them invalid.
+     */
+    s->file_end = s->zero_start = s->data_end = -EINVAL;
+
+    bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
+                               BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
+                               false, errp);
+    if (!bs->file) {
+        return -EINVAL;
+    }
+
+    if (!preallocate_absorb_opts(&s->opts, options, bs->file->bs, errp)) {
+        return -EINVAL;
+    }
+
+    bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
+        (BDRV_REQ_FUA & bs->file->bs->supported_write_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);
+
+    return 0;
+}
+
+static void preallocate_close(BlockDriverState *bs)
+{
+    int ret;
+    BDRVPreallocateState *s = bs->opaque;
+
+    if (s->data_end < 0) {
+        return;
+    }
+
+    if (s->file_end < 0) {
+        s->file_end = bdrv_getlength(bs->file->bs);
+        if (s->file_end < 0) {
+            return;
+        }
+    }
+
+    if (s->data_end < s->file_end) {
+        ret = bdrv_truncate(bs->file, s->data_end, true, PREALLOC_MODE_OFF, 0,
+                            NULL);
+        s->file_end = ret < 0 ? ret : s->data_end;
+    }
+}
+
+
+/*
+ * Handle reopen.
+ *
+ * We must implement reopen handlers, otherwise reopen just don't work. Handle
+ * new options and don't care about preallocation state, as it is handled in
+ * set/check permission handlers.
+ */
+
+static int preallocate_reopen_prepare(BDRVReopenState *reopen_state,
+                                      BlockReopenQueue *queue, Error **errp)
+{
+    PreallocateOpts *opts = g_new0(PreallocateOpts, 1);
+
+    if (!preallocate_absorb_opts(opts, reopen_state->options,
+                                 reopen_state->bs->file->bs, errp)) {
+        g_free(opts);
+        return -EINVAL;
+    }
+
+    reopen_state->opaque = opts;
+
+    return 0;
+}
+
+static void preallocate_reopen_commit(BDRVReopenState *state)
+{
+    BDRVPreallocateState *s = state->bs->opaque;
+
+    s->opts = *(PreallocateOpts *)state->opaque;
+
+    g_free(state->opaque);
+    state->opaque = NULL;
+}
+
+static void preallocate_reopen_abort(BDRVReopenState *state)
+{
+    g_free(state->opaque);
+    state->opaque = NULL;
+}
+
+static coroutine_fn int preallocate_co_preadv_part(
+        BlockDriverState *bs, uint64_t offset, uint64_t bytes,
+        QEMUIOVector *qiov, size_t qiov_offset, int flags)
+{
+    return bdrv_co_preadv_part(bs->file, offset, bytes, qiov, qiov_offset,
+                               flags);
+}
+
+static int coroutine_fn preallocate_co_pdiscard(BlockDriverState *bs,
+                                               int64_t offset, int bytes)
+{
+    return bdrv_co_pdiscard(bs->file, offset, bytes);
+}
+
+static bool can_write_resize(uint64_t perm)
+{
+    return (perm & BLK_PERM_WRITE) && (perm & BLK_PERM_RESIZE);
+}
+
+static bool has_prealloc_perms(BlockDriverState *bs)
+{
+    BDRVPreallocateState *s = bs->opaque;
+
+    if (can_write_resize(bs->file->perm)) {
+        assert(!(bs->file->shared_perm & BLK_PERM_WRITE));
+        assert(!(bs->file->shared_perm & BLK_PERM_RESIZE));
+        return true;
+    }
+
+    assert(s->data_end < 0);
+    assert(s->zero_start < 0);
+    assert(s->file_end < 0);
+    return false;
+}
+
+/*
+ * Call on each write. Returns true if @want_merge_zero is true and the region
+ * [offset, offset + bytes) is zeroed (as a result of this call or earlier
+ * preallocation).
+ *
+ * want_merge_zero is used to merge write-zero request with preallocation in
+ * one bdrv_co_pwrite_zeroes() call.
+ */
+static bool coroutine_fn handle_write(BlockDriverState *bs, int64_t offset,
+                                      int64_t bytes, bool want_merge_zero)
+{
+    BDRVPreallocateState *s = bs->opaque;
+    int64_t end = offset + bytes;
+    int64_t prealloc_start, prealloc_end;
+    int ret;
+
+    if (!has_prealloc_perms(bs)) {
+        /* We don't have state neither should try to recover it */
+        return false;
+    }
+
+    if (s->data_end < 0) {
+        s->data_end = bdrv_getlength(bs->file->bs);
+        if (s->data_end < 0) {
+            return false;
+        }
+
+        if (s->file_end < 0) {
+            s->file_end = s->data_end;
+        }
+    }
+
+    if (end <= s->data_end) {
+        return false;
+    }
+
+    /* We have valid s->data_end, and request writes beyond it. */
+
+    s->data_end = end;
+    if (s->zero_start < 0 || !want_merge_zero) {
+        s->zero_start = end;
+    }
+
+    if (s->file_end < 0) {
+        s->file_end = bdrv_getlength(bs->file->bs);
+        if (s->file_end < 0) {
+            return false;
+        }
+    }
+
+    /* Now s->data_end, s->zero_start and s->file_end are valid. */
+
+    if (end <= s->file_end) {
+        /* No preallocation needed. */
+        return want_merge_zero && offset >= s->zero_start;
+    }
+
+    /* Now we want new preallocation, as request writes beyond s->file_end. */
+
+    prealloc_start = want_merge_zero ? MIN(offset, s->file_end) : s->file_end;
+    prealloc_end = QEMU_ALIGN_UP(end + s->opts.prealloc_size,
+                                 s->opts.prealloc_align);
+
+    ret = bdrv_co_pwrite_zeroes(
+            bs->file, prealloc_start, prealloc_end - prealloc_start,
+            BDRV_REQ_NO_FALLBACK | BDRV_REQ_SERIALISING | BDRV_REQ_NO_WAIT);
+    if (ret < 0) {
+        s->file_end = ret;
+        return false;
+    }
+
+    s->file_end = prealloc_end;
+    return want_merge_zero;
+}
+
+static int coroutine_fn preallocate_co_pwrite_zeroes(BlockDriverState *bs,
+        int64_t offset, int bytes, BdrvRequestFlags flags)
+{
+    bool want_merge_zero =
+        !(flags & ~(BDRV_REQ_ZERO_WRITE | BDRV_REQ_NO_FALLBACK));
+    if (handle_write(bs, offset, bytes, want_merge_zero)) {
+        return 0;
+    }
+
+    return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
+}
+
+static coroutine_fn int preallocate_co_pwritev_part(BlockDriverState *bs,
+                                                    uint64_t offset,
+                                                    uint64_t bytes,
+                                                    QEMUIOVector *qiov,
+                                                    size_t qiov_offset,
+                                                    int flags)
+{
+    handle_write(bs, offset, bytes, false);
+
+    return bdrv_co_pwritev_part(bs->file, offset, bytes, qiov, qiov_offset,
+                                flags);
+}
+
+static int coroutine_fn
+preallocate_co_truncate(BlockDriverState *bs, int64_t offset,
+                        bool exact, PreallocMode prealloc,
+                        BdrvRequestFlags flags, Error **errp)
+{
+    ERRP_GUARD();
+    BDRVPreallocateState *s = bs->opaque;
+    int ret;
+
+    if (s->data_end >= 0 && offset > s->data_end) {
+        if (s->file_end < 0) {
+            s->file_end = bdrv_getlength(bs->file->bs);
+            if (s->file_end < 0) {
+                error_setg(errp, "failed to get file length");
+                return s->file_end;
+            }
+        }
+
+        if (prealloc == PREALLOC_MODE_FALLOC) {
+            /*
+             * If offset <= s->file_end, the task is already done, just
+             * update s->data_end, to move part of "filter preallocation"
+             * to "preallocation requested by user".
+             * Otherwise just proceed to preallocate missing part.
+             */
+            if (offset <= s->file_end) {
+                s->data_end = offset;
+                return 0;
+            }
+        } else {
+            /*
+             * We have to drop our preallocation, to
+             * - avoid "Cannot use preallocation for shrinking files" in
+             *   case of offset < file_end
+             * - give PREALLOC_MODE_OFF a chance to keep small disk
+             *   usage
+             * - give PREALLOC_MODE_FULL a chance to actually write the
+             *   whole region as user expects
+             */
+            if (s->file_end > s->data_end) {
+                ret = bdrv_co_truncate(bs->file, s->data_end, true,
+                                       PREALLOC_MODE_OFF, 0, errp);
+                if (ret < 0) {
+                    s->file_end = ret;
+                    error_prepend(errp, "preallocate-filter: failed to drop "
+                                  "write-zero preallocation: ");
+                    return ret;
+                }
+                s->file_end = s->data_end;
+            }
+        }
+
+        s->data_end = offset;
+    }
+
+    ret = bdrv_co_truncate(bs->file, offset, exact, prealloc, flags, errp);
+    if (ret < 0) {
+        s->file_end = s->zero_start = s->data_end = ret;
+        return ret;
+    }
+
+    if (has_prealloc_perms(bs)) {
+        s->file_end = s->zero_start = s->data_end = offset;
+    }
+    return 0;
+}
+
+static int coroutine_fn preallocate_co_flush(BlockDriverState *bs)
+{
+    return bdrv_co_flush(bs->file->bs);
+}
+
+static int64_t preallocate_getlength(BlockDriverState *bs)
+{
+    int64_t ret;
+    BDRVPreallocateState *s = bs->opaque;
+
+    if (s->data_end >= 0) {
+        return s->data_end;
+    }
+
+    ret = bdrv_getlength(bs->file->bs);
+
+    if (has_prealloc_perms(bs)) {
+        s->file_end = s->zero_start = s->data_end = ret;
+    }
+
+    return ret;
+}
+
+static int preallocate_check_perm(BlockDriverState *bs,
+                                  uint64_t perm, uint64_t shared, Error **errp)
+{
+    BDRVPreallocateState *s = bs->opaque;
+
+    if (s->data_end >= 0 && !can_write_resize(perm)) {
+        /*
+         * Lose permissions.
+         * We should truncate in check_perm, as in set_perm bs->file->perm will
+         * be already changed, and we should not violate it.
+         */
+        if (s->file_end < 0) {
+            s->file_end = bdrv_getlength(bs->file->bs);
+            if (s->file_end < 0) {
+                error_setg(errp, "Failed to get file length");
+                return s->file_end;
+            }
+        }
+
+        if (s->data_end < s->file_end) {
+            int ret = bdrv_truncate(bs->file, s->data_end, true,
+                                    PREALLOC_MODE_OFF, 0, NULL);
+            if (ret < 0) {
+                error_setg(errp, "Failed to drop preallocation");
+                s->file_end = ret;
+                return ret;
+            }
+            s->file_end = s->data_end;
+        }
+    }
+
+    return 0;
+}
+
+static void preallocate_set_perm(BlockDriverState *bs,
+                                 uint64_t perm, uint64_t shared)
+{
+    BDRVPreallocateState *s = bs->opaque;
+
+    if (can_write_resize(perm)) {
+        if (s->data_end < 0) {
+            s->data_end = s->file_end = s->zero_start =
+                bdrv_getlength(bs->file->bs);
+        }
+    } else {
+        /*
+         * We drop our permissions, as well as allow shared
+         * permissions (see preallocate_child_perm), anyone will be able to
+         * change the child, so mark all states invalid. We'll regain control if
+         * get good permissions back.
+         */
+        s->data_end = s->file_end = s->zero_start = -EINVAL;
+    }
+}
+
+static void preallocate_child_perm(BlockDriverState *bs, BdrvChild *c,
+    BdrvChildRole role, BlockReopenQueue *reopen_queue,
+    uint64_t perm, uint64_t shared, uint64_t *nperm, uint64_t *nshared)
+{
+    bdrv_default_perms(bs, c, role, reopen_queue, perm, shared, nperm, nshared);
+
+    if (can_write_resize(perm)) {
+        /* This should come by default, but let's enforce: */
+        *nperm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
+
+        /*
+         * Don't share, to keep our states s->file_end, s->data_end and
+         * s->zero_start valid.
+         */
+        *nshared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
+    }
+}
+
+BlockDriver bdrv_preallocate_filter = {
+    .format_name = "preallocate",
+    .instance_size = sizeof(BDRVPreallocateState),
+
+    .bdrv_getlength = preallocate_getlength,
+    .bdrv_open = preallocate_open,
+    .bdrv_close = preallocate_close,
+
+    .bdrv_reopen_prepare  = preallocate_reopen_prepare,
+    .bdrv_reopen_commit   = preallocate_reopen_commit,
+    .bdrv_reopen_abort    = preallocate_reopen_abort,
+
+    .bdrv_co_preadv_part = preallocate_co_preadv_part,
+    .bdrv_co_pwritev_part = preallocate_co_pwritev_part,
+    .bdrv_co_pwrite_zeroes = preallocate_co_pwrite_zeroes,
+    .bdrv_co_pdiscard = preallocate_co_pdiscard,
+    .bdrv_co_flush = preallocate_co_flush,
+    .bdrv_co_truncate = preallocate_co_truncate,
+
+    .bdrv_check_perm = preallocate_check_perm,
+    .bdrv_set_perm = preallocate_set_perm,
+    .bdrv_child_perm = preallocate_child_perm,
+
+    .has_variable_length = true,
+    .is_filter = true,
+};
+
+static void bdrv_preallocate_init(void)
+{
+    bdrv_register(&bdrv_preallocate_filter);
+}
+
+block_init(bdrv_preallocate_init);
diff --git a/block/quorum.c b/block/quorum.c
index 4b08a19..0bd7545 100644
--- a/block/quorum.c
+++ b/block/quorum.c
@@ -18,6 +18,7 @@
 #include "qemu/module.h"
 #include "qemu/option.h"
 #include "block/block_int.h"
+#include "block/coroutines.h"
 #include "block/qdict.h"
 #include "qapi/error.h"
 #include "qapi/qapi-events-block.h"
@@ -691,8 +692,13 @@ static void write_quorum_entry(void *opaque)
     QuorumChildRequest *sacb = &acb->qcrs[i];
 
     sacb->bs = s->children[i]->bs;
-    sacb->ret = bdrv_co_pwritev(s->children[i], acb->offset, acb->bytes,
-                                acb->qiov, acb->flags);
+    if (acb->flags & BDRV_REQ_ZERO_WRITE) {
+        sacb->ret = bdrv_co_pwrite_zeroes(s->children[i], acb->offset,
+                                          acb->bytes, acb->flags);
+    } else {
+        sacb->ret = bdrv_co_pwritev(s->children[i], acb->offset, acb->bytes,
+                                    acb->qiov, acb->flags);
+    }
     if (sacb->ret == 0) {
         acb->success_count++;
     } else {
@@ -738,6 +744,14 @@ static int quorum_co_pwritev(BlockDriverState *bs, uint64_t offset,
     return ret;
 }
 
+static int quorum_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
+                                   int bytes, BdrvRequestFlags flags)
+
+{
+    return quorum_co_pwritev(bs, offset, bytes, NULL,
+                             flags | BDRV_REQ_ZERO_WRITE);
+}
+
 static int64_t quorum_getlength(BlockDriverState *bs)
 {
     BDRVQuorumState *s = bs->opaque;
@@ -896,6 +910,21 @@ static QemuOptsList quorum_runtime_opts = {
     },
 };
 
+static void quorum_refresh_flags(BlockDriverState *bs)
+{
+    BDRVQuorumState *s = bs->opaque;
+    int i;
+
+    bs->supported_zero_flags =
+        BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK;
+
+    for (i = 0; i < s->num_children; i++) {
+        bs->supported_zero_flags &= s->children[i]->bs->supported_zero_flags;
+    }
+
+    bs->supported_zero_flags |= BDRV_REQ_WRITE_UNCHANGED;
+}
+
 static int quorum_open(BlockDriverState *bs, QDict *options, int flags,
                        Error **errp)
 {
@@ -990,6 +1019,7 @@ static int quorum_open(BlockDriverState *bs, QDict *options, int flags,
     s->next_child_index = s->num_children;
 
     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED;
+    quorum_refresh_flags(bs);
 
     g_free(opened);
     goto exit;
@@ -1061,6 +1091,7 @@ static void quorum_add_child(BlockDriverState *bs, BlockDriverState *child_bs,
     }
     s->children = g_renew(BdrvChild *, s->children, s->num_children + 1);
     s->children[s->num_children++] = child;
+    quorum_refresh_flags(bs);
 
 out:
     bdrv_drained_end(bs);
@@ -1105,6 +1136,7 @@ static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
     s->children = g_renew(BdrvChild *, s->children, --s->num_children);
     bdrv_unref_child(bs, child);
 
+    quorum_refresh_flags(bs);
     bdrv_drained_end(bs);
 }
 
@@ -1179,6 +1211,56 @@ static void quorum_child_perm(BlockDriverState *bs, BdrvChild *c,
              | DEFAULT_PERM_UNCHANGED;
 }
 
+/*
+ * Each one of the children can report different status flags even
+ * when they contain the same data, so what this function does is
+ * return BDRV_BLOCK_ZERO if *all* children agree that a certain
+ * region contains zeroes, and BDRV_BLOCK_DATA otherwise.
+ */
+static int coroutine_fn quorum_co_block_status(BlockDriverState *bs,
+                                               bool want_zero,
+                                               int64_t offset, int64_t count,
+                                               int64_t *pnum, int64_t *map,
+                                               BlockDriverState **file)
+{
+    BDRVQuorumState *s = bs->opaque;
+    int i, ret;
+    int64_t pnum_zero = count;
+    int64_t pnum_data = 0;
+
+    for (i = 0; i < s->num_children; i++) {
+        int64_t bytes;
+        ret = bdrv_co_common_block_status_above(s->children[i]->bs, NULL, false,
+                                                want_zero, offset, count,
+                                                &bytes, NULL, NULL, NULL);
+        if (ret < 0) {
+            quorum_report_bad(QUORUM_OP_TYPE_READ, offset, count,
+                              s->children[i]->bs->node_name, ret);
+            pnum_data = count;
+            break;
+        }
+        /*
+         * Even if all children agree about whether there are zeroes
+         * or not at @offset they might disagree on the size, so use
+         * the smallest when reporting BDRV_BLOCK_ZERO and the largest
+         * when reporting BDRV_BLOCK_DATA.
+         */
+        if (ret & BDRV_BLOCK_ZERO) {
+            pnum_zero = MIN(pnum_zero, bytes);
+        } else {
+            pnum_data = MAX(pnum_data, bytes);
+        }
+    }
+
+    if (pnum_data) {
+        *pnum = pnum_data;
+        return BDRV_BLOCK_DATA;
+    } else {
+        *pnum = pnum_zero;
+        return BDRV_BLOCK_ZERO;
+    }
+}
+
 static const char *const quorum_strong_runtime_opts[] = {
     QUORUM_OPT_VOTE_THRESHOLD,
     QUORUM_OPT_BLKVERIFY,
@@ -1197,6 +1279,7 @@ static BlockDriver bdrv_quorum = {
     .bdrv_close                         = quorum_close,
     .bdrv_gather_child_options          = quorum_gather_child_options,
     .bdrv_dirname                       = quorum_dirname,
+    .bdrv_co_block_status               = quorum_co_block_status,
 
     .bdrv_co_flush_to_disk              = quorum_co_flush,
 
@@ -1204,6 +1287,7 @@ static BlockDriver bdrv_quorum = {
 
     .bdrv_co_preadv                     = quorum_co_preadv,
     .bdrv_co_pwritev                    = quorum_co_pwritev,
+    .bdrv_co_pwrite_zeroes              = quorum_co_pwrite_zeroes,
 
     .bdrv_add_child                     = quorum_add_child,
     .bdrv_del_child                     = quorum_del_child,