6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; under version 2 of the License.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
22 #include "qemu-coroutine.h"
27 #include <sys/ioctl.h>
29 #if defined(__sun__) || defined(__HAIKU__)
30 #include <sys/ioccom.h>
39 #include "qemu_socket.h"
40 #include "qemu-queue.h"
45 #define TRACE(msg, ...) do { \
46 LOG(msg, ## __VA_ARGS__); \
49 #define TRACE(msg, ...) \
53 #define LOG(msg, ...) do { \
54 fprintf(stderr, "%s:%s():L%d: " msg "\n", \
55 __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \
58 /* This is all part of the "official" NBD API */
60 #define NBD_REPLY_SIZE (4 + 4 + 8)
61 #define NBD_REQUEST_MAGIC 0x25609513
62 #define NBD_REPLY_MAGIC 0x67446698
64 #define NBD_SET_SOCK _IO(0xab, 0)
65 #define NBD_SET_BLKSIZE _IO(0xab, 1)
66 #define NBD_SET_SIZE _IO(0xab, 2)
67 #define NBD_DO_IT _IO(0xab, 3)
68 #define NBD_CLEAR_SOCK _IO(0xab, 4)
69 #define NBD_CLEAR_QUE _IO(0xab, 5)
70 #define NBD_PRINT_DEBUG _IO(0xab, 6)
71 #define NBD_SET_SIZE_BLOCKS _IO(0xab, 7)
72 #define NBD_DISCONNECT _IO(0xab, 8)
73 #define NBD_SET_TIMEOUT _IO(0xab, 9)
74 #define NBD_SET_FLAGS _IO(0xab, 10)
76 #define NBD_OPT_EXPORT_NAME (1 << 0)
78 /* That's all folks */
80 ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
85 if (qemu_in_coroutine()) {
87 return qemu_co_recv(fd, buffer, size);
89 return qemu_co_send(fd, buffer, size);
93 while (offset < size) {
97 len = qemu_recv(fd, buffer + offset, size - offset, 0);
99 len = send(fd, buffer + offset, size - offset, 0);
103 err = socket_error();
105 /* recoverable error */
106 if (err == EINTR || (offset > 0 && err == EAGAIN)) {
110 /* unrecoverable error */
125 static ssize_t read_sync(int fd, void *buffer, size_t size)
127 /* Sockets are kept in blocking mode in the negotiation phase. After
128 * that, a non-readable socket simply means that another thread stole
129 * our request/reply. Synchronization is done with recv_coroutine, so
130 * that this is coroutine-safe.
132 return nbd_wr_sync(fd, buffer, size, true);
135 static ssize_t write_sync(int fd, void *buffer, size_t size)
139 /* For writes, we do expect the socket to be writable. */
140 ret = nbd_wr_sync(fd, buffer, size, false);
141 } while (ret == -EAGAIN);
145 static void combine_addr(char *buf, size_t len, const char* address,
148 /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
149 if (strstr(address, ":")) {
150 snprintf(buf, len, "[%s]:%u", address, port);
152 snprintf(buf, len, "%s:%u", address, port);
156 int tcp_socket_outgoing(const char *address, uint16_t port)
158 char address_and_port[128];
159 combine_addr(address_and_port, 128, address, port);
160 return tcp_socket_outgoing_spec(address_and_port);
163 int tcp_socket_outgoing_spec(const char *address_and_port)
165 return inet_connect(address_and_port, true, NULL);
168 int tcp_socket_incoming(const char *address, uint16_t port)
170 char address_and_port[128];
171 combine_addr(address_and_port, 128, address, port);
172 return tcp_socket_incoming_spec(address_and_port);
175 int tcp_socket_incoming_spec(const char *address_and_port)
179 return inet_listen(address_and_port, ostr, olen, SOCK_STREAM, 0, NULL);
182 int unix_socket_incoming(const char *path)
187 return unix_listen(path, ostr, olen);
190 int unix_socket_outgoing(const char *path)
192 return unix_connect(path);
209 static int nbd_send_negotiate(int csock, off_t size, uint32_t flags)
211 char buf[8 + 8 + 8 + 128];
215 [ 0 .. 7] passwd ("NBDMAGIC")
216 [ 8 .. 15] magic (0x00420281861253)
219 [28 .. 151] reserved (0)
222 socket_set_block(csock);
225 TRACE("Beginning negotiation.");
226 memcpy(buf, "NBDMAGIC", 8);
227 cpu_to_be64w((uint64_t*)(buf + 8), 0x00420281861253LL);
228 cpu_to_be64w((uint64_t*)(buf + 16), size);
229 cpu_to_be32w((uint32_t*)(buf + 24),
230 flags | NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
231 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
232 memset(buf + 28, 0, 124);
234 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
239 TRACE("Negotiation succeeded.");
242 socket_set_nonblock(csock);
246 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
247 off_t *size, size_t *blocksize)
254 TRACE("Receiving negotiation.");
256 socket_set_block(csock);
259 if (read_sync(csock, buf, 8) != 8) {
265 if (strlen(buf) == 0) {
266 LOG("server connection closed");
270 TRACE("Magic is %c%c%c%c%c%c%c%c",
271 qemu_isprint(buf[0]) ? buf[0] : '.',
272 qemu_isprint(buf[1]) ? buf[1] : '.',
273 qemu_isprint(buf[2]) ? buf[2] : '.',
274 qemu_isprint(buf[3]) ? buf[3] : '.',
275 qemu_isprint(buf[4]) ? buf[4] : '.',
276 qemu_isprint(buf[5]) ? buf[5] : '.',
277 qemu_isprint(buf[6]) ? buf[6] : '.',
278 qemu_isprint(buf[7]) ? buf[7] : '.');
280 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
281 LOG("Invalid magic received");
285 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
289 magic = be64_to_cpu(magic);
290 TRACE("Magic is 0x%" PRIx64, magic);
293 uint32_t reserved = 0;
297 TRACE("Checking magic (opts_magic)");
298 if (magic != 0x49484156454F5054LL) {
299 LOG("Bad magic received");
302 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
303 LOG("flags read failed");
306 *flags = be16_to_cpu(tmp) << 16;
307 /* reserved for future use */
308 if (write_sync(csock, &reserved, sizeof(reserved)) !=
310 LOG("write failed (reserved)");
313 /* write the export name */
314 magic = cpu_to_be64(magic);
315 if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
316 LOG("write failed (magic)");
319 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
320 if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
321 LOG("write failed (opt)");
324 namesize = cpu_to_be32(strlen(name));
325 if (write_sync(csock, &namesize, sizeof(namesize)) !=
327 LOG("write failed (namesize)");
330 if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
331 LOG("write failed (name)");
335 TRACE("Checking magic (cli_magic)");
337 if (magic != 0x00420281861253LL) {
338 LOG("Bad magic received");
343 if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
347 *size = be64_to_cpu(s);
349 TRACE("Size is %" PRIu64, *size);
352 if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
353 LOG("read failed (flags)");
356 *flags = be32_to_cpup(flags);
358 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
359 LOG("read failed (tmp)");
362 *flags |= be32_to_cpu(tmp);
364 if (read_sync(csock, &buf, 124) != 124) {
365 LOG("read failed (buf)");
371 socket_set_nonblock(csock);
376 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
378 TRACE("Setting NBD socket");
380 if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
382 LOG("Failed to set NBD socket");
386 TRACE("Setting block size to %lu", (unsigned long)blocksize);
388 if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
390 LOG("Failed setting NBD block size");
394 TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
396 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
398 LOG("Failed setting size (in blocks)");
402 if (flags & NBD_FLAG_READ_ONLY) {
404 TRACE("Setting readonly attribute");
406 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
408 LOG("Failed setting read-only attribute");
413 if (ioctl(fd, NBD_SET_FLAGS, flags) < 0
414 && errno != ENOTTY) {
416 LOG("Failed setting flags");
420 TRACE("Negotiation ended");
425 int nbd_disconnect(int fd)
427 ioctl(fd, NBD_CLEAR_QUE);
428 ioctl(fd, NBD_DISCONNECT);
429 ioctl(fd, NBD_CLEAR_SOCK);
433 int nbd_client(int fd)
438 TRACE("Doing NBD loop");
440 ret = ioctl(fd, NBD_DO_IT);
441 if (ret < 0 && errno == EPIPE) {
442 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
443 * the socket via NBD_DISCONNECT. We do not want to return 1 in
450 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
452 TRACE("Clearing NBD queue");
453 ioctl(fd, NBD_CLEAR_QUE);
455 TRACE("Clearing NBD socket");
456 ioctl(fd, NBD_CLEAR_SOCK);
462 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
467 int nbd_disconnect(int fd)
472 int nbd_client(int fd)
478 ssize_t nbd_send_request(int csock, struct nbd_request *request)
480 uint8_t buf[4 + 4 + 8 + 8 + 4];
483 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
484 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
485 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
486 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
487 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
489 TRACE("Sending request to client: "
490 "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
491 request->from, request->len, request->handle, request->type);
493 ret = write_sync(csock, buf, sizeof(buf));
498 if (ret != sizeof(buf)) {
499 LOG("writing to socket failed");
505 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
507 uint8_t buf[4 + 4 + 8 + 8 + 4];
511 ret = read_sync(csock, buf, sizeof(buf));
516 if (ret != sizeof(buf)) {
522 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
523 [ 4 .. 7] type (0 == READ, 1 == WRITE)
529 magic = be32_to_cpup((uint32_t*)buf);
530 request->type = be32_to_cpup((uint32_t*)(buf + 4));
531 request->handle = be64_to_cpup((uint64_t*)(buf + 8));
532 request->from = be64_to_cpup((uint64_t*)(buf + 16));
533 request->len = be32_to_cpup((uint32_t*)(buf + 24));
535 TRACE("Got request: "
536 "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
537 magic, request->type, request->from, request->len);
539 if (magic != NBD_REQUEST_MAGIC) {
540 LOG("invalid magic (got 0x%x)", magic);
546 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
548 uint8_t buf[NBD_REPLY_SIZE];
552 ret = read_sync(csock, buf, sizeof(buf));
557 if (ret != sizeof(buf)) {
563 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
564 [ 4 .. 7] error (0 == no error)
568 magic = be32_to_cpup((uint32_t*)buf);
569 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
570 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
573 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
574 magic, reply->error, reply->handle);
576 if (magic != NBD_REPLY_MAGIC) {
577 LOG("invalid magic (got 0x%x)", magic);
583 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
585 uint8_t buf[4 + 4 + 8];
589 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
590 [ 4 .. 7] error (0 == no error)
593 cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
594 cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
595 cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
597 TRACE("Sending response to client");
599 ret = write_sync(csock, buf, sizeof(buf));
604 if (ret != sizeof(buf)) {
605 LOG("writing to socket failed");
611 #define MAX_NBD_REQUESTS 16
613 typedef struct NBDRequest NBDRequest;
616 QSIMPLEQ_ENTRY(NBDRequest) entry;
622 BlockDriverState *bs;
626 QSIMPLEQ_HEAD(, NBDRequest) requests;
631 void (*close)(NBDClient *client);
636 Coroutine *recv_coroutine;
639 Coroutine *send_coroutine;
644 static void nbd_client_get(NBDClient *client)
649 static void nbd_client_put(NBDClient *client)
651 if (--client->refcount == 0) {
656 static void nbd_client_close(NBDClient *client)
658 qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
662 client->close(client);
664 nbd_client_put(client);
667 static NBDRequest *nbd_request_get(NBDClient *client)
670 NBDExport *exp = client->exp;
672 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
673 client->nb_requests++;
675 if (QSIMPLEQ_EMPTY(&exp->requests)) {
676 req = g_malloc0(sizeof(NBDRequest));
677 req->data = qemu_blockalign(exp->bs, NBD_BUFFER_SIZE);
679 req = QSIMPLEQ_FIRST(&exp->requests);
680 QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
682 nbd_client_get(client);
683 req->client = client;
687 static void nbd_request_put(NBDRequest *req)
689 NBDClient *client = req->client;
690 QSIMPLEQ_INSERT_HEAD(&client->exp->requests, req, entry);
691 if (client->nb_requests-- == MAX_NBD_REQUESTS) {
694 nbd_client_put(client);
697 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
698 off_t size, uint32_t nbdflags)
700 NBDExport *exp = g_malloc0(sizeof(NBDExport));
701 QSIMPLEQ_INIT(&exp->requests);
703 exp->dev_offset = dev_offset;
704 exp->nbdflags = nbdflags;
705 exp->size = size == -1 ? bdrv_getlength(bs) : size;
709 void nbd_export_close(NBDExport *exp)
711 while (!QSIMPLEQ_EMPTY(&exp->requests)) {
712 NBDRequest *first = QSIMPLEQ_FIRST(&exp->requests);
713 QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
714 qemu_vfree(first->data);
722 static int nbd_can_read(void *opaque);
723 static void nbd_read(void *opaque);
724 static void nbd_restart_write(void *opaque);
726 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
729 NBDClient *client = req->client;
730 int csock = client->sock;
733 qemu_co_mutex_lock(&client->send_lock);
734 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
735 nbd_restart_write, client);
736 client->send_coroutine = qemu_coroutine_self();
739 rc = nbd_send_reply(csock, reply);
741 socket_set_cork(csock, 1);
742 rc = nbd_send_reply(csock, reply);
744 ret = qemu_co_send(csock, req->data, len);
749 socket_set_cork(csock, 0);
752 client->send_coroutine = NULL;
753 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
754 qemu_co_mutex_unlock(&client->send_lock);
758 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
760 NBDClient *client = req->client;
761 int csock = client->sock;
764 client->recv_coroutine = qemu_coroutine_self();
765 rc = nbd_receive_request(csock, request);
773 if (request->len > NBD_BUFFER_SIZE) {
774 LOG("len (%u) is larger than max len (%u)",
775 request->len, NBD_BUFFER_SIZE);
780 if ((request->from + request->len) < request->from) {
781 LOG("integer overflow detected! "
782 "you're probably being attacked");
787 TRACE("Decoding type");
789 if ((request->type & NBD_CMD_MASK_COMMAND) == NBD_CMD_WRITE) {
790 TRACE("Reading %u byte(s)", request->len);
792 if (qemu_co_recv(csock, req->data, request->len) != request->len) {
793 LOG("reading from socket failed");
801 client->recv_coroutine = NULL;
805 static void nbd_trip(void *opaque)
807 NBDClient *client = opaque;
808 NBDRequest *req = nbd_request_get(client);
809 NBDExport *exp = client->exp;
810 struct nbd_request request;
811 struct nbd_reply reply;
814 TRACE("Reading request.");
816 ret = nbd_co_receive_request(req, &request);
817 if (ret == -EAGAIN) {
824 reply.handle = request.handle;
832 if ((request.from + request.len) > exp->size) {
833 LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
834 ", Offset: %" PRIu64 "\n",
835 request.from, request.len,
836 (uint64_t)exp->size, (uint64_t)exp->dev_offset);
837 LOG("requested operation past EOF--bad client?");
838 goto invalid_request;
841 switch (request.type & NBD_CMD_MASK_COMMAND) {
843 TRACE("Request type is READ");
845 if (request.type & NBD_CMD_FLAG_FUA) {
846 ret = bdrv_co_flush(exp->bs);
854 ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
855 req->data, request.len / 512);
857 LOG("reading from file failed");
862 TRACE("Read %u byte(s)", request.len);
863 if (nbd_co_send_reply(req, &reply, request.len) < 0)
867 TRACE("Request type is WRITE");
869 if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
870 TRACE("Server is read-only, return error");
875 TRACE("Writing to device");
877 ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
878 req->data, request.len / 512);
880 LOG("writing to file failed");
885 if (request.type & NBD_CMD_FLAG_FUA) {
886 ret = bdrv_co_flush(exp->bs);
894 if (nbd_co_send_reply(req, &reply, 0) < 0) {
899 TRACE("Request type is DISCONNECT");
903 TRACE("Request type is FLUSH");
905 ret = bdrv_co_flush(exp->bs);
910 if (nbd_co_send_reply(req, &reply, 0) < 0) {
915 TRACE("Request type is TRIM");
916 ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
919 LOG("discard failed");
922 if (nbd_co_send_reply(req, &reply, 0) < 0) {
927 LOG("invalid request type (%u) received", request.type);
929 reply.error = -EINVAL;
931 if (nbd_co_send_reply(req, &reply, 0) < 0) {
937 TRACE("Request/Reply complete");
940 nbd_request_put(req);
944 nbd_request_put(req);
945 nbd_client_close(client);
948 static int nbd_can_read(void *opaque)
950 NBDClient *client = opaque;
952 return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
955 static void nbd_read(void *opaque)
957 NBDClient *client = opaque;
959 if (client->recv_coroutine) {
960 qemu_coroutine_enter(client->recv_coroutine, NULL);
962 qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
966 static void nbd_restart_write(void *opaque)
968 NBDClient *client = opaque;
970 qemu_coroutine_enter(client->send_coroutine, NULL);
973 NBDClient *nbd_client_new(NBDExport *exp, int csock,
974 void (*close)(NBDClient *))
977 if (nbd_send_negotiate(csock, exp->size, exp->nbdflags) < 0) {
980 client = g_malloc0(sizeof(NBDClient));
981 client->refcount = 1;
983 client->sock = csock;
984 client->close = close;
985 qemu_co_mutex_init(&client->send_lock);
986 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);