]> git.proxmox.com Git - qemu.git/blobdiff - nbd.c
block: enforce constraints on block size properties
[qemu.git] / nbd.c
diff --git a/nbd.c b/nbd.c
index d383d854fe2ca65506de85b3a9173031261b9a37..567e94e27a6f24957659e2990ca419cf246dcc3a 100644 (file)
--- a/nbd.c
+++ b/nbd.c
@@ -18,6 +18,9 @@
 
 #include "nbd.h"
 #include "block.h"
+#include "block_int.h"
+
+#include "qemu-coroutine.h"
 
 #include <errno.h>
 #include <string.h>
@@ -35,6 +38,7 @@
 #endif
 
 #include "qemu_socket.h"
+#include "qemu-queue.h"
 
 //#define DEBUG_NBD
 
@@ -186,7 +190,7 @@ int unix_socket_outgoing(const char *path)
                   Request (type == 2)
 */
 
-int nbd_negotiate(int csock, off_t size, uint32_t flags)
+static int nbd_send_negotiate(int csock, off_t size, uint32_t flags)
 {
     char buf[8 + 8 + 8 + 128];
 
@@ -583,144 +587,369 @@ static int nbd_send_reply(int csock, struct nbd_reply *reply)
     return 0;
 }
 
-int nbd_trip(BlockDriverState *bs, int csock, off_t size,
-             uint64_t dev_offset, uint32_t nbdflags,
-             uint8_t *data)
+#define MAX_NBD_REQUESTS 16
+
+typedef struct NBDRequest NBDRequest;
+
+struct NBDRequest {
+    QSIMPLEQ_ENTRY(NBDRequest) entry;
+    NBDClient *client;
+    uint8_t *data;
+};
+
+struct NBDExport {
+    BlockDriverState *bs;
+    off_t dev_offset;
+    off_t size;
+    uint32_t nbdflags;
+    QSIMPLEQ_HEAD(, NBDRequest) requests;
+};
+
+struct NBDClient {
+    int refcount;
+    void (*close)(NBDClient *client);
+
+    NBDExport *exp;
+    int sock;
+
+    Coroutine *recv_coroutine;
+
+    CoMutex send_lock;
+    Coroutine *send_coroutine;
+
+    int nb_requests;
+};
+
+static void nbd_client_get(NBDClient *client)
 {
-    struct nbd_request request;
-    struct nbd_reply reply;
-    int ret;
+    client->refcount++;
+}
 
-    TRACE("Reading request.");
+static void nbd_client_put(NBDClient *client)
+{
+    if (--client->refcount == 0) {
+        g_free(client);
+    }
+}
 
-    if (nbd_receive_request(csock, &request) == -1)
-        return -1;
+static void nbd_client_close(NBDClient *client)
+{
+    qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
+    close(client->sock);
+    client->sock = -1;
+    if (client->close) {
+        client->close(client);
+    }
+    nbd_client_put(client);
+}
+
+static NBDRequest *nbd_request_get(NBDClient *client)
+{
+    NBDRequest *req;
+    NBDExport *exp = client->exp;
+
+    assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
+    client->nb_requests++;
+
+    if (QSIMPLEQ_EMPTY(&exp->requests)) {
+        req = g_malloc0(sizeof(NBDRequest));
+        req->data = qemu_blockalign(exp->bs, NBD_BUFFER_SIZE);
+    } else {
+        req = QSIMPLEQ_FIRST(&exp->requests);
+        QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
+    }
+    nbd_client_get(client);
+    req->client = client;
+    return req;
+}
+
+static void nbd_request_put(NBDRequest *req)
+{
+    NBDClient *client = req->client;
+    QSIMPLEQ_INSERT_HEAD(&client->exp->requests, req, entry);
+    if (client->nb_requests-- == MAX_NBD_REQUESTS) {
+        qemu_notify_event();
+    }
+    nbd_client_put(client);
+}
+
+NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
+                          off_t size, uint32_t nbdflags)
+{
+    NBDExport *exp = g_malloc0(sizeof(NBDExport));
+    QSIMPLEQ_INIT(&exp->requests);
+    exp->bs = bs;
+    exp->dev_offset = dev_offset;
+    exp->nbdflags = nbdflags;
+    exp->size = size == -1 ? exp->bs->total_sectors * 512 : size;
+    return exp;
+}
+
+void nbd_export_close(NBDExport *exp)
+{
+    while (!QSIMPLEQ_EMPTY(&exp->requests)) {
+        NBDRequest *first = QSIMPLEQ_FIRST(&exp->requests);
+        QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
+        qemu_vfree(first->data);
+        g_free(first);
+    }
+
+    bdrv_close(exp->bs);
+    g_free(exp);
+}
+
+static int nbd_can_read(void *opaque);
+static void nbd_read(void *opaque);
+static void nbd_restart_write(void *opaque);
+
+static int nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
+                             int len)
+{
+    NBDClient *client = req->client;
+    int csock = client->sock;
+    int rc, ret;
+
+    qemu_co_mutex_lock(&client->send_lock);
+    qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
+                         nbd_restart_write, client);
+    client->send_coroutine = qemu_coroutine_self();
+
+    if (!len) {
+        rc = nbd_send_reply(csock, reply);
+        if (rc == -1) {
+            rc = -errno;
+        }
+    } else {
+        socket_set_cork(csock, 1);
+        rc = nbd_send_reply(csock, reply);
+        if (rc != -1) {
+            ret = qemu_co_send(csock, req->data, len);
+            if (ret != len) {
+                errno = EIO;
+                rc = -1;
+            }
+        }
+        if (rc == -1) {
+            rc = -errno;
+        }
+        socket_set_cork(csock, 0);
+    }
+
+    client->send_coroutine = NULL;
+    qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
+    qemu_co_mutex_unlock(&client->send_lock);
+    return rc;
+}
 
-    if (request.len > NBD_BUFFER_SIZE) {
+static int nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
+{
+    NBDClient *client = req->client;
+    int csock = client->sock;
+    int rc;
+
+    client->recv_coroutine = qemu_coroutine_self();
+    if (nbd_receive_request(csock, request) == -1) {
+        rc = -EIO;
+        goto out;
+    }
+
+    if (request->len > NBD_BUFFER_SIZE) {
         LOG("len (%u) is larger than max len (%u)",
-            request.len, NBD_BUFFER_SIZE);
-        errno = EINVAL;
-        return -1;
+            request->len, NBD_BUFFER_SIZE);
+        rc = -EINVAL;
+        goto out;
     }
 
-    if ((request.from + request.len) < request.from) {
+    if ((request->from + request->len) < request->from) {
         LOG("integer overflow detected! "
             "you're probably being attacked");
-        errno = EINVAL;
-        return -1;
+        rc = -EINVAL;
+        goto out;
     }
 
-    if ((request.from + request.len) > size) {
-            LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
-            ", Offset: %" PRIu64 "\n",
-                    request.from, request.len, (uint64_t)size, dev_offset);
-        LOG("requested operation past EOF--bad client?");
-        errno = EINVAL;
-        return -1;
+    TRACE("Decoding type");
+
+    if ((request->type & NBD_CMD_MASK_COMMAND) == NBD_CMD_WRITE) {
+        TRACE("Reading %u byte(s)", request->len);
+
+        if (qemu_co_recv(csock, req->data, request->len) != request->len) {
+            LOG("reading from socket failed");
+            rc = -EIO;
+            goto out;
+        }
     }
+    rc = 0;
 
-    TRACE("Decoding type");
+out:
+    client->recv_coroutine = NULL;
+    return rc;
+}
+
+static void nbd_trip(void *opaque)
+{
+    NBDClient *client = opaque;
+    NBDRequest *req = nbd_request_get(client);
+    NBDExport *exp = client->exp;
+    struct nbd_request request;
+    struct nbd_reply reply;
+    int ret;
+
+    TRACE("Reading request.");
+
+    ret = nbd_co_receive_request(req, &request);
+    if (ret == -EIO) {
+        goto out;
+    }
 
     reply.handle = request.handle;
     reply.error = 0;
 
+    if (ret < 0) {
+        reply.error = -ret;
+        goto error_reply;
+    }
+
+    if ((request.from + request.len) > exp->size) {
+            LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
+            ", Offset: %" PRIu64 "\n",
+                    request.from, request.len,
+                    (uint64_t)exp->size, exp->dev_offset);
+        LOG("requested operation past EOF--bad client?");
+        goto invalid_request;
+    }
+
     switch (request.type & NBD_CMD_MASK_COMMAND) {
     case NBD_CMD_READ:
         TRACE("Request type is READ");
 
-        ret = bdrv_read(bs, (request.from + dev_offset) / 512,
-                        data, request.len / 512);
+        ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
+                        req->data, request.len / 512);
         if (ret < 0) {
             LOG("reading from file failed");
             reply.error = -ret;
-            request.len = 0;
+            goto error_reply;
         }
 
         TRACE("Read %u byte(s)", request.len);
-        socket_set_cork(csock, 1);
-        if (nbd_send_reply(csock, &reply) == -1)
-            return -1;
-
-        TRACE("Sending data to client");
-
-        if (write_sync(csock, data, request.len) != request.len) {
-            LOG("writing to socket failed");
-            errno = EINVAL;
-            return -1;
-        }
-        socket_set_cork(csock, 0);
+        if (nbd_co_send_reply(req, &reply, request.len) < 0)
+            goto out;
         break;
     case NBD_CMD_WRITE:
         TRACE("Request type is WRITE");
 
-        TRACE("Reading %u byte(s)", request.len);
-
-        if (read_sync(csock, data, request.len) != request.len) {
-            LOG("reading from socket failed");
-            errno = EINVAL;
-            return -1;
+        if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
+            TRACE("Server is read-only, return error");
+            reply.error = EROFS;
+            goto error_reply;
         }
 
-        if (nbdflags & NBD_FLAG_READ_ONLY) {
-            TRACE("Server is read-only, return error");
-            reply.error = 1;
-        } else {
-            TRACE("Writing to device");
+        TRACE("Writing to device");
 
-            ret = bdrv_write(bs, (request.from + dev_offset) / 512,
-                             data, request.len / 512);
+        ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
+                         req->data, request.len / 512);
+        if (ret < 0) {
+            LOG("writing to file failed");
+            reply.error = -ret;
+            goto error_reply;
+        }
+
+        if (request.type & NBD_CMD_FLAG_FUA) {
+            ret = bdrv_co_flush(exp->bs);
             if (ret < 0) {
-                LOG("writing to file failed");
+                LOG("flush failed");
                 reply.error = -ret;
-                request.len = 0;
-            }
-
-            if (request.type & NBD_CMD_FLAG_FUA) {
-                ret = bdrv_flush(bs);
-                if (ret < 0) {
-                    LOG("flush failed");
-                    reply.error = -ret;
-                }
+                goto error_reply;
             }
         }
 
-        if (nbd_send_reply(csock, &reply) == -1)
-            return -1;
+        if (nbd_co_send_reply(req, &reply, 0) < 0)
+            goto out;
         break;
     case NBD_CMD_DISC:
         TRACE("Request type is DISCONNECT");
         errno = 0;
-        return 1;
+        goto out;
     case NBD_CMD_FLUSH:
         TRACE("Request type is FLUSH");
 
-        ret = bdrv_flush(bs);
+        ret = bdrv_co_flush(exp->bs);
         if (ret < 0) {
             LOG("flush failed");
             reply.error = -ret;
         }
 
-        if (nbd_send_reply(csock, &reply) == -1)
-            return -1;
+        if (nbd_co_send_reply(req, &reply, 0) < 0)
+            goto out;
         break;
     case NBD_CMD_TRIM:
         TRACE("Request type is TRIM");
-        ret = bdrv_discard(bs, (request.from + dev_offset) / 512,
-                           request.len / 512);
+        ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
+                              request.len / 512);
         if (ret < 0) {
             LOG("discard failed");
             reply.error = -ret;
         }
-        if (nbd_send_reply(csock, &reply) == -1)
-            return -1;
+        if (nbd_co_send_reply(req, &reply, 0) < 0)
+            goto out;
         break;
     default:
         LOG("invalid request type (%u) received", request.type);
-        errno = EINVAL;
-        return -1;
+    invalid_request:
+        reply.error = -EINVAL;
+    error_reply:
+        if (nbd_co_send_reply(req, &reply, 0) == -1)
+            goto out;
+        break;
     }
 
     TRACE("Request/Reply complete");
 
-    return 0;
+    nbd_request_put(req);
+    return;
+
+out:
+    nbd_request_put(req);
+    nbd_client_close(client);
+}
+
+static int nbd_can_read(void *opaque)
+{
+    NBDClient *client = opaque;
+
+    return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
+}
+
+static void nbd_read(void *opaque)
+{
+    NBDClient *client = opaque;
+
+    if (client->recv_coroutine) {
+        qemu_coroutine_enter(client->recv_coroutine, NULL);
+    } else {
+        qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
+    }
+}
+
+static void nbd_restart_write(void *opaque)
+{
+    NBDClient *client = opaque;
+
+    qemu_coroutine_enter(client->send_coroutine, NULL);
+}
+
+NBDClient *nbd_client_new(NBDExport *exp, int csock,
+                          void (*close)(NBDClient *))
+{
+    NBDClient *client;
+    if (nbd_send_negotiate(csock, exp->size, exp->nbdflags) == -1) {
+        return NULL;
+    }
+    client = g_malloc0(sizeof(NBDClient));
+    client->refcount = 1;
+    client->exp = exp;
+    client->sock = csock;
+    client->close = close;
+    qemu_co_mutex_init(&client->send_lock);
+    qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
+    return client;
 }