[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Stable-8.2.3 29/87] nbd/server: introduce NBDClient->lock to protect fi
From: |
Michael Tokarev |
Subject: |
[Stable-8.2.3 29/87] nbd/server: introduce NBDClient->lock to protect fields |
Date: |
Wed, 10 Apr 2024 10:22:02 +0300 |
From: Stefan Hajnoczi <stefanha@redhat.com>
NBDClient has a number of fields that are accessed by both the export
AioContext and the main loop thread. When the AioContext lock is removed
these fields will need another form of protection.
Add NBDClient->lock and protect fields that are accessed by both
threads. Also add assertions where possible and otherwise add doc
comments stating assumptions about which thread and lock holding.
Note this patch moves the client->recv_coroutine assertion from
nbd_co_receive_request() to nbd_trip() where client->lock is held.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20231221192452.1785567-7-stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
(cherry picked from commit 7075d235114b40b30524cf1c5b61906c0bbf5f4d)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
diff --git a/nbd/server.c b/nbd/server.c
index 65ec99fa92..f8738f6d27 100644
--- a/nbd/server.c
+++ b/nbd/server.c
@@ -125,23 +125,25 @@ struct NBDClient {
int refcount; /* atomic */
void (*close_fn)(NBDClient *client, bool negotiated);
+ QemuMutex lock;
+
NBDExport *exp;
QCryptoTLSCreds *tlscreds;
char *tlsauthz;
QIOChannelSocket *sioc; /* The underlying data channel */
QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
- Coroutine *recv_coroutine;
+ Coroutine *recv_coroutine; /* protected by lock */
CoMutex send_lock;
Coroutine *send_coroutine;
- bool read_yielding;
- bool quiescing;
+ bool read_yielding; /* protected by lock */
+ bool quiescing; /* protected by lock */
QTAILQ_ENTRY(NBDClient) next;
- int nb_requests;
- bool closing;
+ int nb_requests; /* protected by lock */
+ bool closing; /* protected by lock */
uint32_t check_align; /* If non-zero, check for aligned client requests */
@@ -1415,11 +1417,18 @@ nbd_read_eof(NBDClient *client, void *buffer, size_t
size, Error **errp)
len = qio_channel_readv(client->ioc, &iov, 1, errp);
if (len == QIO_CHANNEL_ERR_BLOCK) {
- client->read_yielding = true;
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ client->read_yielding = true;
+
+ /* Prompt main loop thread to re-run nbd_drained_poll() */
+ aio_wait_kick();
+ }
qio_channel_yield(client->ioc, G_IO_IN);
- client->read_yielding = false;
- if (client->quiescing) {
- return -EAGAIN;
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ client->read_yielding = false;
+ if (client->quiescing) {
+ return -EAGAIN;
+ }
}
continue;
} else if (len < 0) {
@@ -1528,6 +1537,7 @@ void nbd_client_put(NBDClient *client)
blk_exp_unref(&client->exp->common);
}
g_free(client->contexts.bitmaps);
+ qemu_mutex_destroy(&client->lock);
g_free(client);
}
}
@@ -1561,11 +1571,13 @@ static void client_close(NBDClient *client, bool
negotiated)
{
assert(qemu_in_main_thread());
- if (client->closing) {
- return;
- }
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ if (client->closing) {
+ return;
+ }
- client->closing = true;
+ client->closing = true;
+ }
/* Force requests to finish. They will drop their own references,
* then we'll close the socket and free the NBDClient.
@@ -1579,6 +1591,7 @@ static void client_close(NBDClient *client, bool
negotiated)
}
}
+/* Runs in export AioContext with client->lock held */
static NBDRequestData *nbd_request_get(NBDClient *client)
{
NBDRequestData *req;
@@ -1592,6 +1605,7 @@ static NBDRequestData *nbd_request_get(NBDClient *client)
return req;
}
+/* Runs in export AioContext with client->lock held */
static void nbd_request_put(NBDRequestData *req)
{
NBDClient *client = req->client;
@@ -1617,14 +1631,18 @@ static void blk_aio_attached(AioContext *ctx, void
*opaque)
NBDExport *exp = opaque;
NBDClient *client;
+ assert(qemu_in_main_thread());
+
trace_nbd_blk_aio_attached(exp->name, ctx);
exp->common.ctx = ctx;
QTAILQ_FOREACH(client, &exp->clients, next) {
- assert(client->nb_requests == 0);
- assert(client->recv_coroutine == NULL);
- assert(client->send_coroutine == NULL);
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ assert(client->nb_requests == 0);
+ assert(client->recv_coroutine == NULL);
+ assert(client->send_coroutine == NULL);
+ }
}
}
@@ -1632,6 +1650,8 @@ static void blk_aio_detach(void *opaque)
{
NBDExport *exp = opaque;
+ assert(qemu_in_main_thread());
+
trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
exp->common.ctx = NULL;
@@ -1642,8 +1662,12 @@ static void nbd_drained_begin(void *opaque)
NBDExport *exp = opaque;
NBDClient *client;
+ assert(qemu_in_main_thread());
+
QTAILQ_FOREACH(client, &exp->clients, next) {
- client->quiescing = true;
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ client->quiescing = true;
+ }
}
}
@@ -1652,28 +1676,48 @@ static void nbd_drained_end(void *opaque)
NBDExport *exp = opaque;
NBDClient *client;
+ assert(qemu_in_main_thread());
+
QTAILQ_FOREACH(client, &exp->clients, next) {
- client->quiescing = false;
- nbd_client_receive_next_request(client);
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ client->quiescing = false;
+ nbd_client_receive_next_request(client);
+ }
}
}
+/* Runs in export AioContext */
+static void nbd_wake_read_bh(void *opaque)
+{
+ NBDClient *client = opaque;
+ qio_channel_wake_read(client->ioc);
+}
+
static bool nbd_drained_poll(void *opaque)
{
NBDExport *exp = opaque;
NBDClient *client;
+ assert(qemu_in_main_thread());
+
QTAILQ_FOREACH(client, &exp->clients, next) {
- if (client->nb_requests != 0) {
- /*
- * If there's a coroutine waiting for a request on nbd_read_eof()
- * enter it here so we don't depend on the client to wake it up.
- */
- if (client->recv_coroutine != NULL && client->read_yielding) {
- qio_channel_wake_read(client->ioc);
- }
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ if (client->nb_requests != 0) {
+ /*
+ * If there's a coroutine waiting for a request on
nbd_read_eof()
+ * enter it here so we don't depend on the client to wake it
up.
+ *
+ * Schedule a BH in the export AioContext to avoid missing the
+ * wake up due to the race between qio_channel_wake_read() and
+ * qio_channel_yield().
+ */
+ if (client->recv_coroutine != NULL && client->read_yielding) {
+
aio_bh_schedule_oneshot(nbd_export_aio_context(client->exp),
+ nbd_wake_read_bh, client);
+ }
- return true;
+ return true;
+ }
}
}
@@ -1684,6 +1728,8 @@ static void nbd_eject_notifier(Notifier *n, void *data)
{
NBDExport *exp = container_of(n, NBDExport, eject_notifier);
+ assert(qemu_in_main_thread());
+
blk_exp_request_shutdown(&exp->common);
}
@@ -2569,7 +2615,6 @@ static int coroutine_fn
nbd_co_receive_request(NBDRequestData *req,
int ret;
g_assert(qemu_in_coroutine());
- assert(client->recv_coroutine == qemu_coroutine_self());
ret = nbd_receive_request(client, request, errp);
if (ret < 0) {
return ret;
@@ -2978,6 +3023,9 @@ static coroutine_fn void nbd_trip(void *opaque)
*/
trace_nbd_trip();
+
+ qemu_mutex_lock(&client->lock);
+
if (client->closing) {
goto done;
}
@@ -2993,7 +3041,21 @@ static coroutine_fn void nbd_trip(void *opaque)
}
req = nbd_request_get(client);
- ret = nbd_co_receive_request(req, &request, &local_err);
+
+ /*
+ * nbd_co_receive_request() returns -EAGAIN when nbd_drained_begin() has
+ * set client->quiescing but by the time we get back nbd_drained_end() may
+ * have already cleared client->quiescing. In that case we try again
+ * because nothing else will spawn an nbd_trip() coroutine until we set
+ * client->recv_coroutine = NULL further down.
+ */
+ do {
+ assert(client->recv_coroutine == qemu_coroutine_self());
+ qemu_mutex_unlock(&client->lock);
+ ret = nbd_co_receive_request(req, &request, &local_err);
+ qemu_mutex_lock(&client->lock);
+ } while (ret == -EAGAIN && !client->quiescing);
+
client->recv_coroutine = NULL;
if (client->closing) {
@@ -3005,15 +3067,16 @@ static coroutine_fn void nbd_trip(void *opaque)
}
if (ret == -EAGAIN) {
- assert(client->quiescing);
goto done;
}
nbd_client_receive_next_request(client);
+
if (ret == -EIO) {
goto disconnect;
}
+ qemu_mutex_unlock(&client->lock);
qio_channel_set_cork(client->ioc, true);
if (ret < 0) {
@@ -3033,6 +3096,10 @@ static coroutine_fn void nbd_trip(void *opaque)
g_free(request.contexts->bitmaps);
g_free(request.contexts);
}
+
+ qio_channel_set_cork(client->ioc, false);
+ qemu_mutex_lock(&client->lock);
+
if (ret < 0) {
error_prepend(&local_err, "Failed to send reply: ");
goto disconnect;
@@ -3047,11 +3114,13 @@ static coroutine_fn void nbd_trip(void *opaque)
goto disconnect;
}
- qio_channel_set_cork(client->ioc, false);
done:
if (req) {
nbd_request_put(req);
}
+
+ qemu_mutex_unlock(&client->lock);
+
if (!nbd_client_put_nonzero(client)) {
aio_co_reschedule_self(qemu_get_aio_context());
nbd_client_put(client);
@@ -3062,13 +3131,19 @@ disconnect:
if (local_err) {
error_reportf_err(local_err, "Disconnect client, due to: ");
}
+
nbd_request_put(req);
+ qemu_mutex_unlock(&client->lock);
aio_co_reschedule_self(qemu_get_aio_context());
client_close(client, true);
nbd_client_put(client);
}
+/*
+ * Runs in export AioContext and main loop thread. Caller must hold
+ * client->lock.
+ */
static void nbd_client_receive_next_request(NBDClient *client)
{
if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
@@ -3094,7 +3169,9 @@ static coroutine_fn void nbd_co_client_start(void *opaque)
return;
}
- nbd_client_receive_next_request(client);
+ WITH_QEMU_LOCK_GUARD(&client->lock) {
+ nbd_client_receive_next_request(client);
+ }
}
/*
@@ -3111,6 +3188,7 @@ void nbd_client_new(QIOChannelSocket *sioc,
Coroutine *co;
client = g_new0(NBDClient, 1);
+ qemu_mutex_init(&client->lock);
client->refcount = 1;
client->tlscreds = tlscreds;
if (tlscreds) {
--
2.39.2
- [Stable-8.2.3 17/87] xen: Drop out of coroutine context xen_invalidate_map_cache_entry, (continued)
- [Stable-8.2.3 17/87] xen: Drop out of coroutine context xen_invalidate_map_cache_entry, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 20/87] hw/nvme: generalize the mbar size helper, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 18/87] hw/nvme: separate 'serial' property for VFs, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 22/87] hw/nvme: Use pcie_sriov_num_vfs(), Michael Tokarev, 2024/04/10
- [Stable-8.2.3 25/87] hw/audio/virtio-sound: return correct command response size, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 21/87] hw/nvme: add machine compatibility parameter to enable msix exclusive bar, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 23/87] pcie_sriov: Validate NumVFs, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 24/87] hmat acpi: Fix out of bounds access due to missing use of indirection, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 26/87] migration: Skip only empty block devices, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 30/87] nbd/server: Fix race in draining the export, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 29/87] nbd/server: introduce NBDClient->lock to protect fields,
Michael Tokarev <=
- [Stable-8.2.3 31/87] iotests: Add test for reset/AioContext switches with NBD exports, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 27/87] mirror: Don't call job_pause_point() under graph lock, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 28/87] nbd/server: only traverse NBDExport->clients from main loop thread, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 37/87] target/i386: fix direction of "32-bit MMU" test, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 32/87] tests/unit: Bump test-aio-multithread test timeout to 2 minutes, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 34/87] tests/unit: Bump test-replication timeout to 60 seconds, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 36/87] target/i386: use separate MMU indexes for 32-bit accesses, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 35/87] target/i386: introduce function to query MMU indices, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 33/87] tests/unit: Bump test-crypto-block test timeout to 5 minutes, Michael Tokarev, 2024/04/10
- [Stable-8.2.3 38/87] Revert "chardev/char-socket: Fix TLS io channels sending too much data to the backend", Michael Tokarev, 2024/04/10