]>
git.proxmox.com Git - qemu.git/blob - nbd.c
2 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
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_REQUEST_SIZE (4 + 4 + 8 + 8 + 4)
61 #define NBD_REPLY_SIZE (4 + 4 + 8)
62 #define NBD_REQUEST_MAGIC 0x25609513
63 #define NBD_REPLY_MAGIC 0x67446698
64 #define NBD_OPTS_MAGIC 0x49484156454F5054LL
65 #define NBD_CLIENT_MAGIC 0x0000420281861253LL
67 #define NBD_SET_SOCK _IO(0xab, 0)
68 #define NBD_SET_BLKSIZE _IO(0xab, 1)
69 #define NBD_SET_SIZE _IO(0xab, 2)
70 #define NBD_DO_IT _IO(0xab, 3)
71 #define NBD_CLEAR_SOCK _IO(0xab, 4)
72 #define NBD_CLEAR_QUE _IO(0xab, 5)
73 #define NBD_PRINT_DEBUG _IO(0xab, 6)
74 #define NBD_SET_SIZE_BLOCKS _IO(0xab, 7)
75 #define NBD_DISCONNECT _IO(0xab, 8)
76 #define NBD_SET_TIMEOUT _IO(0xab, 9)
77 #define NBD_SET_FLAGS _IO(0xab, 10)
79 #define NBD_OPT_EXPORT_NAME (1 << 0)
81 /* That's all folks */
83 ssize_t
nbd_wr_sync(int fd
, void *buffer
, size_t size
, bool do_read
)
88 if (qemu_in_coroutine()) {
90 return qemu_co_recv(fd
, buffer
, size
);
92 return qemu_co_send(fd
, buffer
, size
);
96 while (offset
< size
) {
100 len
= qemu_recv(fd
, buffer
+ offset
, size
- offset
, 0);
102 len
= send(fd
, buffer
+ offset
, size
- offset
, 0);
106 err
= socket_error();
108 /* recoverable error */
109 if (err
== EINTR
|| (offset
> 0 && err
== EAGAIN
)) {
113 /* unrecoverable error */
128 static ssize_t
read_sync(int fd
, void *buffer
, size_t size
)
130 /* Sockets are kept in blocking mode in the negotiation phase. After
131 * that, a non-readable socket simply means that another thread stole
132 * our request/reply. Synchronization is done with recv_coroutine, so
133 * that this is coroutine-safe.
135 return nbd_wr_sync(fd
, buffer
, size
, true);
138 static ssize_t
write_sync(int fd
, void *buffer
, size_t size
)
142 /* For writes, we do expect the socket to be writable. */
143 ret
= nbd_wr_sync(fd
, buffer
, size
, false);
144 } while (ret
== -EAGAIN
);
148 static void combine_addr(char *buf
, size_t len
, const char* address
,
151 /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
152 if (strstr(address
, ":")) {
153 snprintf(buf
, len
, "[%s]:%u", address
, port
);
155 snprintf(buf
, len
, "%s:%u", address
, port
);
159 int tcp_socket_outgoing(const char *address
, uint16_t port
)
161 char address_and_port
[128];
162 combine_addr(address_and_port
, 128, address
, port
);
163 return tcp_socket_outgoing_spec(address_and_port
);
166 int tcp_socket_outgoing_spec(const char *address_and_port
)
168 return inet_connect(address_and_port
, true, NULL
, NULL
);
171 int tcp_socket_incoming(const char *address
, uint16_t port
)
173 char address_and_port
[128];
174 combine_addr(address_and_port
, 128, address
, port
);
175 return tcp_socket_incoming_spec(address_and_port
);
178 int tcp_socket_incoming_spec(const char *address_and_port
)
182 return inet_listen(address_and_port
, ostr
, olen
, SOCK_STREAM
, 0, NULL
);
185 int unix_socket_incoming(const char *path
)
190 return unix_listen(path
, ostr
, olen
);
193 int unix_socket_outgoing(const char *path
)
195 return unix_connect(path
);
212 static int nbd_send_negotiate(int csock
, off_t size
, uint32_t flags
)
214 char buf
[8 + 8 + 8 + 128];
218 [ 0 .. 7] passwd ("NBDMAGIC")
219 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
222 [28 .. 151] reserved (0)
225 socket_set_block(csock
);
228 TRACE("Beginning negotiation.");
229 memcpy(buf
, "NBDMAGIC", 8);
230 cpu_to_be64w((uint64_t*)(buf
+ 8), NBD_CLIENT_MAGIC
);
231 cpu_to_be64w((uint64_t*)(buf
+ 16), size
);
232 cpu_to_be32w((uint32_t*)(buf
+ 24),
233 flags
| NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_TRIM
|
234 NBD_FLAG_SEND_FLUSH
| NBD_FLAG_SEND_FUA
);
235 memset(buf
+ 28, 0, 124);
237 if (write_sync(csock
, buf
, sizeof(buf
)) != sizeof(buf
)) {
242 TRACE("Negotiation succeeded.");
245 socket_set_nonblock(csock
);
249 int nbd_receive_negotiate(int csock
, const char *name
, uint32_t *flags
,
250 off_t
*size
, size_t *blocksize
)
257 TRACE("Receiving negotiation.");
259 socket_set_block(csock
);
262 if (read_sync(csock
, buf
, 8) != 8) {
268 if (strlen(buf
) == 0) {
269 LOG("server connection closed");
273 TRACE("Magic is %c%c%c%c%c%c%c%c",
274 qemu_isprint(buf
[0]) ? buf
[0] : '.',
275 qemu_isprint(buf
[1]) ? buf
[1] : '.',
276 qemu_isprint(buf
[2]) ? buf
[2] : '.',
277 qemu_isprint(buf
[3]) ? buf
[3] : '.',
278 qemu_isprint(buf
[4]) ? buf
[4] : '.',
279 qemu_isprint(buf
[5]) ? buf
[5] : '.',
280 qemu_isprint(buf
[6]) ? buf
[6] : '.',
281 qemu_isprint(buf
[7]) ? buf
[7] : '.');
283 if (memcmp(buf
, "NBDMAGIC", 8) != 0) {
284 LOG("Invalid magic received");
288 if (read_sync(csock
, &magic
, sizeof(magic
)) != sizeof(magic
)) {
292 magic
= be64_to_cpu(magic
);
293 TRACE("Magic is 0x%" PRIx64
, magic
);
296 uint32_t reserved
= 0;
300 TRACE("Checking magic (opts_magic)");
301 if (magic
!= NBD_OPTS_MAGIC
) {
302 LOG("Bad magic received");
305 if (read_sync(csock
, &tmp
, sizeof(tmp
)) != sizeof(tmp
)) {
306 LOG("flags read failed");
309 *flags
= be16_to_cpu(tmp
) << 16;
310 /* reserved for future use */
311 if (write_sync(csock
, &reserved
, sizeof(reserved
)) !=
313 LOG("write failed (reserved)");
316 /* write the export name */
317 magic
= cpu_to_be64(magic
);
318 if (write_sync(csock
, &magic
, sizeof(magic
)) != sizeof(magic
)) {
319 LOG("write failed (magic)");
322 opt
= cpu_to_be32(NBD_OPT_EXPORT_NAME
);
323 if (write_sync(csock
, &opt
, sizeof(opt
)) != sizeof(opt
)) {
324 LOG("write failed (opt)");
327 namesize
= cpu_to_be32(strlen(name
));
328 if (write_sync(csock
, &namesize
, sizeof(namesize
)) !=
330 LOG("write failed (namesize)");
333 if (write_sync(csock
, (char*)name
, strlen(name
)) != strlen(name
)) {
334 LOG("write failed (name)");
338 TRACE("Checking magic (cli_magic)");
340 if (magic
!= NBD_CLIENT_MAGIC
) {
341 LOG("Bad magic received");
346 if (read_sync(csock
, &s
, sizeof(s
)) != sizeof(s
)) {
350 *size
= be64_to_cpu(s
);
352 TRACE("Size is %" PRIu64
, *size
);
355 if (read_sync(csock
, flags
, sizeof(*flags
)) != sizeof(*flags
)) {
356 LOG("read failed (flags)");
359 *flags
= be32_to_cpup(flags
);
361 if (read_sync(csock
, &tmp
, sizeof(tmp
)) != sizeof(tmp
)) {
362 LOG("read failed (tmp)");
365 *flags
|= be32_to_cpu(tmp
);
367 if (read_sync(csock
, &buf
, 124) != 124) {
368 LOG("read failed (buf)");
374 socket_set_nonblock(csock
);
379 int nbd_init(int fd
, int csock
, uint32_t flags
, off_t size
, size_t blocksize
)
381 TRACE("Setting NBD socket");
383 if (ioctl(fd
, NBD_SET_SOCK
, csock
) < 0) {
385 LOG("Failed to set NBD socket");
389 TRACE("Setting block size to %lu", (unsigned long)blocksize
);
391 if (ioctl(fd
, NBD_SET_BLKSIZE
, blocksize
) < 0) {
393 LOG("Failed setting NBD block size");
397 TRACE("Setting size to %zd block(s)", (size_t)(size
/ blocksize
));
399 if (ioctl(fd
, NBD_SET_SIZE_BLOCKS
, size
/ blocksize
) < 0) {
401 LOG("Failed setting size (in blocks)");
405 if (flags
& NBD_FLAG_READ_ONLY
) {
407 TRACE("Setting readonly attribute");
409 if (ioctl(fd
, BLKROSET
, (unsigned long) &read_only
) < 0) {
411 LOG("Failed setting read-only attribute");
416 if (ioctl(fd
, NBD_SET_FLAGS
, flags
) < 0
417 && errno
!= ENOTTY
) {
419 LOG("Failed setting flags");
423 TRACE("Negotiation ended");
428 int nbd_disconnect(int fd
)
430 ioctl(fd
, NBD_CLEAR_QUE
);
431 ioctl(fd
, NBD_DISCONNECT
);
432 ioctl(fd
, NBD_CLEAR_SOCK
);
436 int nbd_client(int fd
)
441 TRACE("Doing NBD loop");
443 ret
= ioctl(fd
, NBD_DO_IT
);
444 if (ret
< 0 && errno
== EPIPE
) {
445 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
446 * the socket via NBD_DISCONNECT. We do not want to return 1 in
453 TRACE("NBD loop returned %d: %s", ret
, strerror(serrno
));
455 TRACE("Clearing NBD queue");
456 ioctl(fd
, NBD_CLEAR_QUE
);
458 TRACE("Clearing NBD socket");
459 ioctl(fd
, NBD_CLEAR_SOCK
);
465 int nbd_init(int fd
, int csock
, uint32_t flags
, off_t size
, size_t blocksize
)
470 int nbd_disconnect(int fd
)
475 int nbd_client(int fd
)
481 ssize_t
nbd_send_request(int csock
, struct nbd_request
*request
)
483 uint8_t buf
[NBD_REQUEST_SIZE
];
486 cpu_to_be32w((uint32_t*)buf
, NBD_REQUEST_MAGIC
);
487 cpu_to_be32w((uint32_t*)(buf
+ 4), request
->type
);
488 cpu_to_be64w((uint64_t*)(buf
+ 8), request
->handle
);
489 cpu_to_be64w((uint64_t*)(buf
+ 16), request
->from
);
490 cpu_to_be32w((uint32_t*)(buf
+ 24), request
->len
);
492 TRACE("Sending request to client: "
493 "{ .from = %" PRIu64
", .len = %u, .handle = %" PRIu64
", .type=%i}",
494 request
->from
, request
->len
, request
->handle
, request
->type
);
496 ret
= write_sync(csock
, buf
, sizeof(buf
));
501 if (ret
!= sizeof(buf
)) {
502 LOG("writing to socket failed");
508 static ssize_t
nbd_receive_request(int csock
, struct nbd_request
*request
)
510 uint8_t buf
[NBD_REQUEST_SIZE
];
514 ret
= read_sync(csock
, buf
, sizeof(buf
));
519 if (ret
!= sizeof(buf
)) {
525 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
526 [ 4 .. 7] type (0 == READ, 1 == WRITE)
532 magic
= be32_to_cpup((uint32_t*)buf
);
533 request
->type
= be32_to_cpup((uint32_t*)(buf
+ 4));
534 request
->handle
= be64_to_cpup((uint64_t*)(buf
+ 8));
535 request
->from
= be64_to_cpup((uint64_t*)(buf
+ 16));
536 request
->len
= be32_to_cpup((uint32_t*)(buf
+ 24));
538 TRACE("Got request: "
539 "{ magic = 0x%x, .type = %d, from = %" PRIu64
" , len = %u }",
540 magic
, request
->type
, request
->from
, request
->len
);
542 if (magic
!= NBD_REQUEST_MAGIC
) {
543 LOG("invalid magic (got 0x%x)", magic
);
549 ssize_t
nbd_receive_reply(int csock
, struct nbd_reply
*reply
)
551 uint8_t buf
[NBD_REPLY_SIZE
];
555 ret
= read_sync(csock
, buf
, sizeof(buf
));
560 if (ret
!= sizeof(buf
)) {
566 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
567 [ 4 .. 7] error (0 == no error)
571 magic
= be32_to_cpup((uint32_t*)buf
);
572 reply
->error
= be32_to_cpup((uint32_t*)(buf
+ 4));
573 reply
->handle
= be64_to_cpup((uint64_t*)(buf
+ 8));
576 "{ magic = 0x%x, .error = %d, handle = %" PRIu64
" }",
577 magic
, reply
->error
, reply
->handle
);
579 if (magic
!= NBD_REPLY_MAGIC
) {
580 LOG("invalid magic (got 0x%x)", magic
);
586 static ssize_t
nbd_send_reply(int csock
, struct nbd_reply
*reply
)
588 uint8_t buf
[NBD_REPLY_SIZE
];
592 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
593 [ 4 .. 7] error (0 == no error)
596 cpu_to_be32w((uint32_t*)buf
, NBD_REPLY_MAGIC
);
597 cpu_to_be32w((uint32_t*)(buf
+ 4), reply
->error
);
598 cpu_to_be64w((uint64_t*)(buf
+ 8), reply
->handle
);
600 TRACE("Sending response to client");
602 ret
= write_sync(csock
, buf
, sizeof(buf
));
607 if (ret
!= sizeof(buf
)) {
608 LOG("writing to socket failed");
614 #define MAX_NBD_REQUESTS 16
616 typedef struct NBDRequest NBDRequest
;
619 QSIMPLEQ_ENTRY(NBDRequest
) entry
;
625 BlockDriverState
*bs
;
629 QSIMPLEQ_HEAD(, NBDRequest
) requests
;
634 void (*close
)(NBDClient
*client
);
639 Coroutine
*recv_coroutine
;
642 Coroutine
*send_coroutine
;
647 static void nbd_client_get(NBDClient
*client
)
652 static void nbd_client_put(NBDClient
*client
)
654 if (--client
->refcount
== 0) {
659 static void nbd_client_close(NBDClient
*client
)
661 qemu_set_fd_handler2(client
->sock
, NULL
, NULL
, NULL
, NULL
);
665 client
->close(client
);
667 nbd_client_put(client
);
670 static NBDRequest
*nbd_request_get(NBDClient
*client
)
673 NBDExport
*exp
= client
->exp
;
675 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
676 client
->nb_requests
++;
678 if (QSIMPLEQ_EMPTY(&exp
->requests
)) {
679 req
= g_malloc0(sizeof(NBDRequest
));
680 req
->data
= qemu_blockalign(exp
->bs
, NBD_BUFFER_SIZE
);
682 req
= QSIMPLEQ_FIRST(&exp
->requests
);
683 QSIMPLEQ_REMOVE_HEAD(&exp
->requests
, entry
);
685 nbd_client_get(client
);
686 req
->client
= client
;
690 static void nbd_request_put(NBDRequest
*req
)
692 NBDClient
*client
= req
->client
;
693 QSIMPLEQ_INSERT_HEAD(&client
->exp
->requests
, req
, entry
);
694 if (client
->nb_requests
-- == MAX_NBD_REQUESTS
) {
697 nbd_client_put(client
);
700 NBDExport
*nbd_export_new(BlockDriverState
*bs
, off_t dev_offset
,
701 off_t size
, uint32_t nbdflags
)
703 NBDExport
*exp
= g_malloc0(sizeof(NBDExport
));
704 QSIMPLEQ_INIT(&exp
->requests
);
706 exp
->dev_offset
= dev_offset
;
707 exp
->nbdflags
= nbdflags
;
708 exp
->size
= size
== -1 ? bdrv_getlength(bs
) : size
;
712 void nbd_export_close(NBDExport
*exp
)
714 while (!QSIMPLEQ_EMPTY(&exp
->requests
)) {
715 NBDRequest
*first
= QSIMPLEQ_FIRST(&exp
->requests
);
716 QSIMPLEQ_REMOVE_HEAD(&exp
->requests
, entry
);
717 qemu_vfree(first
->data
);
725 static int nbd_can_read(void *opaque
);
726 static void nbd_read(void *opaque
);
727 static void nbd_restart_write(void *opaque
);
729 static ssize_t
nbd_co_send_reply(NBDRequest
*req
, struct nbd_reply
*reply
,
732 NBDClient
*client
= req
->client
;
733 int csock
= client
->sock
;
736 qemu_co_mutex_lock(&client
->send_lock
);
737 qemu_set_fd_handler2(csock
, nbd_can_read
, nbd_read
,
738 nbd_restart_write
, client
);
739 client
->send_coroutine
= qemu_coroutine_self();
742 rc
= nbd_send_reply(csock
, reply
);
744 socket_set_cork(csock
, 1);
745 rc
= nbd_send_reply(csock
, reply
);
747 ret
= qemu_co_send(csock
, req
->data
, len
);
752 socket_set_cork(csock
, 0);
755 client
->send_coroutine
= NULL
;
756 qemu_set_fd_handler2(csock
, nbd_can_read
, nbd_read
, NULL
, client
);
757 qemu_co_mutex_unlock(&client
->send_lock
);
761 static ssize_t
nbd_co_receive_request(NBDRequest
*req
, struct nbd_request
*request
)
763 NBDClient
*client
= req
->client
;
764 int csock
= client
->sock
;
767 client
->recv_coroutine
= qemu_coroutine_self();
768 rc
= nbd_receive_request(csock
, request
);
776 if (request
->len
> NBD_BUFFER_SIZE
) {
777 LOG("len (%u) is larger than max len (%u)",
778 request
->len
, NBD_BUFFER_SIZE
);
783 if ((request
->from
+ request
->len
) < request
->from
) {
784 LOG("integer overflow detected! "
785 "you're probably being attacked");
790 TRACE("Decoding type");
792 if ((request
->type
& NBD_CMD_MASK_COMMAND
) == NBD_CMD_WRITE
) {
793 TRACE("Reading %u byte(s)", request
->len
);
795 if (qemu_co_recv(csock
, req
->data
, request
->len
) != request
->len
) {
796 LOG("reading from socket failed");
804 client
->recv_coroutine
= NULL
;
808 static void nbd_trip(void *opaque
)
810 NBDClient
*client
= opaque
;
811 NBDRequest
*req
= nbd_request_get(client
);
812 NBDExport
*exp
= client
->exp
;
813 struct nbd_request request
;
814 struct nbd_reply reply
;
817 TRACE("Reading request.");
819 ret
= nbd_co_receive_request(req
, &request
);
820 if (ret
== -EAGAIN
) {
827 reply
.handle
= request
.handle
;
835 if ((request
.from
+ request
.len
) > exp
->size
) {
836 LOG("From: %" PRIu64
", Len: %u, Size: %" PRIu64
837 ", Offset: %" PRIu64
"\n",
838 request
.from
, request
.len
,
839 (uint64_t)exp
->size
, (uint64_t)exp
->dev_offset
);
840 LOG("requested operation past EOF--bad client?");
841 goto invalid_request
;
844 switch (request
.type
& NBD_CMD_MASK_COMMAND
) {
846 TRACE("Request type is READ");
848 if (request
.type
& NBD_CMD_FLAG_FUA
) {
849 ret
= bdrv_co_flush(exp
->bs
);
857 ret
= bdrv_read(exp
->bs
, (request
.from
+ exp
->dev_offset
) / 512,
858 req
->data
, request
.len
/ 512);
860 LOG("reading from file failed");
865 TRACE("Read %u byte(s)", request
.len
);
866 if (nbd_co_send_reply(req
, &reply
, request
.len
) < 0)
870 TRACE("Request type is WRITE");
872 if (exp
->nbdflags
& NBD_FLAG_READ_ONLY
) {
873 TRACE("Server is read-only, return error");
878 TRACE("Writing to device");
880 ret
= bdrv_write(exp
->bs
, (request
.from
+ exp
->dev_offset
) / 512,
881 req
->data
, request
.len
/ 512);
883 LOG("writing to file failed");
888 if (request
.type
& NBD_CMD_FLAG_FUA
) {
889 ret
= bdrv_co_flush(exp
->bs
);
897 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
902 TRACE("Request type is DISCONNECT");
906 TRACE("Request type is FLUSH");
908 ret
= bdrv_co_flush(exp
->bs
);
913 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
918 TRACE("Request type is TRIM");
919 ret
= bdrv_co_discard(exp
->bs
, (request
.from
+ exp
->dev_offset
) / 512,
922 LOG("discard failed");
925 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
930 LOG("invalid request type (%u) received", request
.type
);
932 reply
.error
= -EINVAL
;
934 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
940 TRACE("Request/Reply complete");
943 nbd_request_put(req
);
947 nbd_request_put(req
);
948 nbd_client_close(client
);
951 static int nbd_can_read(void *opaque
)
953 NBDClient
*client
= opaque
;
955 return client
->recv_coroutine
|| client
->nb_requests
< MAX_NBD_REQUESTS
;
958 static void nbd_read(void *opaque
)
960 NBDClient
*client
= opaque
;
962 if (client
->recv_coroutine
) {
963 qemu_coroutine_enter(client
->recv_coroutine
, NULL
);
965 qemu_coroutine_enter(qemu_coroutine_create(nbd_trip
), client
);
969 static void nbd_restart_write(void *opaque
)
971 NBDClient
*client
= opaque
;
973 qemu_coroutine_enter(client
->send_coroutine
, NULL
);
976 NBDClient
*nbd_client_new(NBDExport
*exp
, int csock
,
977 void (*close
)(NBDClient
*))
980 if (nbd_send_negotiate(csock
, exp
->size
, exp
->nbdflags
) < 0) {
983 client
= g_malloc0(sizeof(NBDClient
));
984 client
->refcount
= 1;
986 client
->sock
= csock
;
987 client
->close
= close
;
988 qemu_co_mutex_init(&client
->send_lock
);
989 qemu_set_fd_handler2(csock
, nbd_can_read
, nbd_read
, NULL
, client
);