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 /* Definitions for opaque data types */
83 typedef struct NBDRequest NBDRequest;
86 QSIMPLEQ_ENTRY(NBDRequest) entry;
93 void (*close)(NBDExport *exp);
100 QTAILQ_HEAD(, NBDClient) clients;
101 QSIMPLEQ_HEAD(, NBDRequest) requests;
102 QTAILQ_ENTRY(NBDExport) next;
105 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
109 void (*close)(NBDClient *client);
114 Coroutine *recv_coroutine;
117 Coroutine *send_coroutine;
119 QTAILQ_ENTRY(NBDClient) next;
124 /* That's all folks */
126 ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
131 if (qemu_in_coroutine()) {
133 return qemu_co_recv(fd, buffer, size);
135 return qemu_co_send(fd, buffer, size);
139 while (offset < size) {
143 len = qemu_recv(fd, buffer + offset, size - offset, 0);
145 len = send(fd, buffer + offset, size - offset, 0);
149 err = socket_error();
151 /* recoverable error */
152 if (err == EINTR || (offset > 0 && err == EAGAIN)) {
156 /* unrecoverable error */
171 static ssize_t read_sync(int fd, void *buffer, size_t size)
173 /* Sockets are kept in blocking mode in the negotiation phase. After
174 * that, a non-readable socket simply means that another thread stole
175 * our request/reply. Synchronization is done with recv_coroutine, so
176 * that this is coroutine-safe.
178 return nbd_wr_sync(fd, buffer, size, true);
181 static ssize_t write_sync(int fd, void *buffer, size_t size)
185 /* For writes, we do expect the socket to be writable. */
186 ret = nbd_wr_sync(fd, buffer, size, false);
187 } while (ret == -EAGAIN);
191 static void combine_addr(char *buf, size_t len, const char* address,
194 /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
195 if (strstr(address, ":")) {
196 snprintf(buf, len, "[%s]:%u", address, port);
198 snprintf(buf, len, "%s:%u", address, port);
202 int tcp_socket_outgoing(const char *address, uint16_t port)
204 char address_and_port[128];
205 combine_addr(address_and_port, 128, address, port);
206 return tcp_socket_outgoing_spec(address_and_port);
209 int tcp_socket_outgoing_spec(const char *address_and_port)
211 return inet_connect(address_and_port, NULL);
214 int tcp_socket_incoming(const char *address, uint16_t port)
216 char address_and_port[128];
217 combine_addr(address_and_port, 128, address, port);
218 return tcp_socket_incoming_spec(address_and_port);
221 int tcp_socket_incoming_spec(const char *address_and_port)
225 return inet_listen(address_and_port, ostr, olen, SOCK_STREAM, 0, NULL);
228 int unix_socket_incoming(const char *path)
233 return unix_listen(path, ostr, olen);
236 int unix_socket_outgoing(const char *path)
238 return unix_connect(path);
241 /* Basic flow for negotiation
268 static int nbd_receive_options(NBDClient *client)
270 int csock = client->sock;
272 uint32_t tmp, length;
277 [ 0 .. 3] reserved (0)
278 [ 4 .. 11] NBD_OPTS_MAGIC
279 [12 .. 15] NBD_OPT_EXPORT_NAME
281 [20 .. xx] export name (length bytes)
285 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
289 TRACE("Checking reserved");
291 LOG("Bad reserved received");
295 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
299 TRACE("Checking reserved");
300 if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) {
301 LOG("Bad magic received");
305 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
309 TRACE("Checking option");
310 if (tmp != be32_to_cpu(NBD_OPT_EXPORT_NAME)) {
311 LOG("Bad option received");
315 if (read_sync(csock, &length, sizeof(length)) != sizeof(length)) {
319 TRACE("Checking length");
320 length = be32_to_cpu(length);
322 LOG("Bad length received");
325 if (read_sync(csock, name, length) != length) {
331 client->exp = nbd_export_find(name);
333 LOG("export not found");
337 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
338 nbd_export_get(client->exp);
340 TRACE("Option negotiation succeeded.");
346 static int nbd_send_negotiate(NBDClient *client)
348 int csock = client->sock;
349 char buf[8 + 8 + 8 + 128];
351 const int myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
352 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
354 /* Negotiation header without options:
355 [ 0 .. 7] passwd ("NBDMAGIC")
356 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
358 [24 .. 25] server flags (0)
359 [24 .. 27] export flags
360 [28 .. 151] reserved (0)
362 Negotiation header with options, part 1:
363 [ 0 .. 7] passwd ("NBDMAGIC")
364 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
365 [16 .. 17] server flags (0)
367 part 2 (after options are sent):
369 [26 .. 27] export flags
370 [28 .. 151] reserved (0)
373 socket_set_block(csock);
376 TRACE("Beginning negotiation.");
377 memcpy(buf, "NBDMAGIC", 8);
379 assert ((client->exp->nbdflags & ~65535) == 0);
380 cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC);
381 cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size);
382 cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
384 cpu_to_be64w((uint64_t*)(buf + 8), NBD_OPTS_MAGIC);
386 memset(buf + 28, 0, 124);
389 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
394 if (write_sync(csock, buf, 18) != 18) {
398 rc = nbd_receive_options(client);
400 LOG("option negotiation failed");
404 assert ((client->exp->nbdflags & ~65535) == 0);
405 cpu_to_be64w((uint64_t*)(buf + 18), client->exp->size);
406 cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
407 if (write_sync(csock, buf + 18, sizeof(buf) - 18) != sizeof(buf) - 18) {
413 TRACE("Negotiation succeeded.");
416 socket_set_nonblock(csock);
420 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
421 off_t *size, size_t *blocksize)
428 TRACE("Receiving negotiation.");
430 socket_set_block(csock);
433 if (read_sync(csock, buf, 8) != 8) {
439 if (strlen(buf) == 0) {
440 LOG("server connection closed");
444 TRACE("Magic is %c%c%c%c%c%c%c%c",
445 qemu_isprint(buf[0]) ? buf[0] : '.',
446 qemu_isprint(buf[1]) ? buf[1] : '.',
447 qemu_isprint(buf[2]) ? buf[2] : '.',
448 qemu_isprint(buf[3]) ? buf[3] : '.',
449 qemu_isprint(buf[4]) ? buf[4] : '.',
450 qemu_isprint(buf[5]) ? buf[5] : '.',
451 qemu_isprint(buf[6]) ? buf[6] : '.',
452 qemu_isprint(buf[7]) ? buf[7] : '.');
454 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
455 LOG("Invalid magic received");
459 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
463 magic = be64_to_cpu(magic);
464 TRACE("Magic is 0x%" PRIx64, magic);
467 uint32_t reserved = 0;
471 TRACE("Checking magic (opts_magic)");
472 if (magic != NBD_OPTS_MAGIC) {
473 LOG("Bad magic received");
476 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
477 LOG("flags read failed");
480 *flags = be16_to_cpu(tmp) << 16;
481 /* reserved for future use */
482 if (write_sync(csock, &reserved, sizeof(reserved)) !=
484 LOG("write failed (reserved)");
487 /* write the export name */
488 magic = cpu_to_be64(magic);
489 if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
490 LOG("write failed (magic)");
493 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
494 if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
495 LOG("write failed (opt)");
498 namesize = cpu_to_be32(strlen(name));
499 if (write_sync(csock, &namesize, sizeof(namesize)) !=
501 LOG("write failed (namesize)");
504 if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
505 LOG("write failed (name)");
509 TRACE("Checking magic (cli_magic)");
511 if (magic != NBD_CLIENT_MAGIC) {
512 LOG("Bad magic received");
517 if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
521 *size = be64_to_cpu(s);
523 TRACE("Size is %" PRIu64, *size);
526 if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
527 LOG("read failed (flags)");
530 *flags = be32_to_cpup(flags);
532 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
533 LOG("read failed (tmp)");
536 *flags |= be32_to_cpu(tmp);
538 if (read_sync(csock, &buf, 124) != 124) {
539 LOG("read failed (buf)");
545 socket_set_nonblock(csock);
550 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
552 TRACE("Setting NBD socket");
554 if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
556 LOG("Failed to set NBD socket");
560 TRACE("Setting block size to %lu", (unsigned long)blocksize);
562 if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
564 LOG("Failed setting NBD block size");
568 TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
570 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
572 LOG("Failed setting size (in blocks)");
576 if (flags & NBD_FLAG_READ_ONLY) {
578 TRACE("Setting readonly attribute");
580 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
582 LOG("Failed setting read-only attribute");
587 if (ioctl(fd, NBD_SET_FLAGS, flags) < 0
588 && errno != ENOTTY) {
590 LOG("Failed setting flags");
594 TRACE("Negotiation ended");
599 int nbd_disconnect(int fd)
601 ioctl(fd, NBD_CLEAR_QUE);
602 ioctl(fd, NBD_DISCONNECT);
603 ioctl(fd, NBD_CLEAR_SOCK);
607 int nbd_client(int fd)
612 TRACE("Doing NBD loop");
614 ret = ioctl(fd, NBD_DO_IT);
615 if (ret < 0 && errno == EPIPE) {
616 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
617 * the socket via NBD_DISCONNECT. We do not want to return 1 in
624 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
626 TRACE("Clearing NBD queue");
627 ioctl(fd, NBD_CLEAR_QUE);
629 TRACE("Clearing NBD socket");
630 ioctl(fd, NBD_CLEAR_SOCK);
636 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
641 int nbd_disconnect(int fd)
646 int nbd_client(int fd)
652 ssize_t nbd_send_request(int csock, struct nbd_request *request)
654 uint8_t buf[NBD_REQUEST_SIZE];
657 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
658 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
659 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
660 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
661 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
663 TRACE("Sending request to client: "
664 "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
665 request->from, request->len, request->handle, request->type);
667 ret = write_sync(csock, buf, sizeof(buf));
672 if (ret != sizeof(buf)) {
673 LOG("writing to socket failed");
679 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
681 uint8_t buf[NBD_REQUEST_SIZE];
685 ret = read_sync(csock, buf, sizeof(buf));
690 if (ret != sizeof(buf)) {
696 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
697 [ 4 .. 7] type (0 == READ, 1 == WRITE)
703 magic = be32_to_cpup((uint32_t*)buf);
704 request->type = be32_to_cpup((uint32_t*)(buf + 4));
705 request->handle = be64_to_cpup((uint64_t*)(buf + 8));
706 request->from = be64_to_cpup((uint64_t*)(buf + 16));
707 request->len = be32_to_cpup((uint32_t*)(buf + 24));
709 TRACE("Got request: "
710 "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
711 magic, request->type, request->from, request->len);
713 if (magic != NBD_REQUEST_MAGIC) {
714 LOG("invalid magic (got 0x%x)", magic);
720 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
722 uint8_t buf[NBD_REPLY_SIZE];
726 ret = read_sync(csock, buf, sizeof(buf));
731 if (ret != sizeof(buf)) {
737 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
738 [ 4 .. 7] error (0 == no error)
742 magic = be32_to_cpup((uint32_t*)buf);
743 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
744 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
747 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
748 magic, reply->error, reply->handle);
750 if (magic != NBD_REPLY_MAGIC) {
751 LOG("invalid magic (got 0x%x)", magic);
757 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
759 uint8_t buf[NBD_REPLY_SIZE];
763 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
764 [ 4 .. 7] error (0 == no error)
767 cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
768 cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
769 cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
771 TRACE("Sending response to client");
773 ret = write_sync(csock, buf, sizeof(buf));
778 if (ret != sizeof(buf)) {
779 LOG("writing to socket failed");
785 #define MAX_NBD_REQUESTS 16
787 void nbd_client_get(NBDClient *client)
792 void nbd_client_put(NBDClient *client)
794 if (--client->refcount == 0) {
795 /* The last reference should be dropped by client->close,
796 * which is called by nbd_client_close.
798 assert(client->closing);
800 qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
804 QTAILQ_REMOVE(&client->exp->clients, client, next);
805 nbd_export_put(client->exp);
811 void nbd_client_close(NBDClient *client)
813 if (client->closing) {
817 client->closing = true;
819 /* Force requests to finish. They will drop their own references,
820 * then we'll close the socket and free the NBDClient.
822 shutdown(client->sock, 2);
824 /* Also tell the client, so that they release their reference. */
826 client->close(client);
830 static NBDRequest *nbd_request_get(NBDClient *client)
833 NBDExport *exp = client->exp;
835 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
836 client->nb_requests++;
838 if (QSIMPLEQ_EMPTY(&exp->requests)) {
839 req = g_malloc0(sizeof(NBDRequest));
840 req->data = qemu_blockalign(exp->bs, NBD_BUFFER_SIZE);
842 req = QSIMPLEQ_FIRST(&exp->requests);
843 QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
845 nbd_client_get(client);
846 req->client = client;
850 static void nbd_request_put(NBDRequest *req)
852 NBDClient *client = req->client;
853 QSIMPLEQ_INSERT_HEAD(&client->exp->requests, req, entry);
854 if (client->nb_requests-- == MAX_NBD_REQUESTS) {
857 nbd_client_put(client);
860 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
861 off_t size, uint32_t nbdflags,
862 void (*close)(NBDExport *))
864 NBDExport *exp = g_malloc0(sizeof(NBDExport));
865 QSIMPLEQ_INIT(&exp->requests);
867 QTAILQ_INIT(&exp->clients);
869 exp->dev_offset = dev_offset;
870 exp->nbdflags = nbdflags;
871 exp->size = size == -1 ? bdrv_getlength(bs) : size;
876 NBDExport *nbd_export_find(const char *name)
879 QTAILQ_FOREACH(exp, &exports, next) {
880 if (strcmp(name, exp->name) == 0) {
888 void nbd_export_set_name(NBDExport *exp, const char *name)
890 if (exp->name == name) {
895 if (exp->name != NULL) {
898 QTAILQ_REMOVE(&exports, exp, next);
903 exp->name = g_strdup(name);
904 QTAILQ_INSERT_TAIL(&exports, exp, next);
909 void nbd_export_close(NBDExport *exp)
911 NBDClient *client, *next;
914 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
915 nbd_client_close(client);
917 nbd_export_set_name(exp, NULL);
921 void nbd_export_get(NBDExport *exp)
923 assert(exp->refcount > 0);
927 void nbd_export_put(NBDExport *exp)
929 assert(exp->refcount > 0);
930 if (exp->refcount == 1) {
931 nbd_export_close(exp);
934 if (--exp->refcount == 0) {
935 assert(exp->name == NULL);
941 while (!QSIMPLEQ_EMPTY(&exp->requests)) {
942 NBDRequest *first = QSIMPLEQ_FIRST(&exp->requests);
943 QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
944 qemu_vfree(first->data);
952 BlockDriverState *nbd_export_get_blockdev(NBDExport *exp)
957 void nbd_export_close_all(void)
959 NBDExport *exp, *next;
961 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
962 nbd_export_close(exp);
966 static int nbd_can_read(void *opaque);
967 static void nbd_read(void *opaque);
968 static void nbd_restart_write(void *opaque);
970 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
973 NBDClient *client = req->client;
974 int csock = client->sock;
977 qemu_co_mutex_lock(&client->send_lock);
978 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
979 nbd_restart_write, client);
980 client->send_coroutine = qemu_coroutine_self();
983 rc = nbd_send_reply(csock, reply);
985 socket_set_cork(csock, 1);
986 rc = nbd_send_reply(csock, reply);
988 ret = qemu_co_send(csock, req->data, len);
993 socket_set_cork(csock, 0);
996 client->send_coroutine = NULL;
997 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
998 qemu_co_mutex_unlock(&client->send_lock);
1002 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
1004 NBDClient *client = req->client;
1005 int csock = client->sock;
1008 client->recv_coroutine = qemu_coroutine_self();
1009 rc = nbd_receive_request(csock, request);
1011 if (rc != -EAGAIN) {
1017 if (request->len > NBD_BUFFER_SIZE) {
1018 LOG("len (%u) is larger than max len (%u)",
1019 request->len, NBD_BUFFER_SIZE);
1024 if ((request->from + request->len) < request->from) {
1025 LOG("integer overflow detected! "
1026 "you're probably being attacked");
1031 TRACE("Decoding type");
1033 if ((request->type & NBD_CMD_MASK_COMMAND) == NBD_CMD_WRITE) {
1034 TRACE("Reading %u byte(s)", request->len);
1036 if (qemu_co_recv(csock, req->data, request->len) != request->len) {
1037 LOG("reading from socket failed");
1045 client->recv_coroutine = NULL;
1049 static void nbd_trip(void *opaque)
1051 NBDClient *client = opaque;
1052 NBDExport *exp = client->exp;
1054 struct nbd_request request;
1055 struct nbd_reply reply;
1058 TRACE("Reading request.");
1059 if (client->closing) {
1063 req = nbd_request_get(client);
1064 ret = nbd_co_receive_request(req, &request);
1065 if (ret == -EAGAIN) {
1072 reply.handle = request.handle;
1080 if ((request.from + request.len) > exp->size) {
1081 LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
1082 ", Offset: %" PRIu64 "\n",
1083 request.from, request.len,
1084 (uint64_t)exp->size, (uint64_t)exp->dev_offset);
1085 LOG("requested operation past EOF--bad client?");
1086 goto invalid_request;
1089 switch (request.type & NBD_CMD_MASK_COMMAND) {
1091 TRACE("Request type is READ");
1093 if (request.type & NBD_CMD_FLAG_FUA) {
1094 ret = bdrv_co_flush(exp->bs);
1096 LOG("flush failed");
1102 ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
1103 req->data, request.len / 512);
1105 LOG("reading from file failed");
1110 TRACE("Read %u byte(s)", request.len);
1111 if (nbd_co_send_reply(req, &reply, request.len) < 0)
1115 TRACE("Request type is WRITE");
1117 if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1118 TRACE("Server is read-only, return error");
1119 reply.error = EROFS;
1123 TRACE("Writing to device");
1125 ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
1126 req->data, request.len / 512);
1128 LOG("writing to file failed");
1133 if (request.type & NBD_CMD_FLAG_FUA) {
1134 ret = bdrv_co_flush(exp->bs);
1136 LOG("flush failed");
1142 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1147 TRACE("Request type is DISCONNECT");
1151 TRACE("Request type is FLUSH");
1153 ret = bdrv_co_flush(exp->bs);
1155 LOG("flush failed");
1158 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1163 TRACE("Request type is TRIM");
1164 ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
1167 LOG("discard failed");
1170 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1175 LOG("invalid request type (%u) received", request.type);
1177 reply.error = -EINVAL;
1179 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1185 TRACE("Request/Reply complete");
1188 nbd_request_put(req);
1192 nbd_request_put(req);
1193 nbd_client_close(client);
1196 static int nbd_can_read(void *opaque)
1198 NBDClient *client = opaque;
1200 return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
1203 static void nbd_read(void *opaque)
1205 NBDClient *client = opaque;
1207 if (client->recv_coroutine) {
1208 qemu_coroutine_enter(client->recv_coroutine, NULL);
1210 qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
1214 static void nbd_restart_write(void *opaque)
1216 NBDClient *client = opaque;
1218 qemu_coroutine_enter(client->send_coroutine, NULL);
1221 NBDClient *nbd_client_new(NBDExport *exp, int csock,
1222 void (*close)(NBDClient *))
1225 client = g_malloc0(sizeof(NBDClient));
1226 client->refcount = 1;
1228 client->sock = csock;
1229 if (nbd_send_negotiate(client) < 0) {
1233 client->close = close;
1234 qemu_co_mutex_init(&client->send_lock);
1235 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1238 QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1239 nbd_export_get(exp);