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/>.
19 #include "block/nbd.h"
20 #include "block/block.h"
22 #include "block/coroutine.h"
27 #include <sys/ioctl.h>
29 #if defined(__sun__) || defined(__HAIKU__)
30 #include <sys/ioccom.h>
39 #include "qemu/sockets.h"
40 #include "qemu/queue.h"
41 #include "qemu/main-loop.h"
46 #define TRACE(msg, ...) do { \
47 LOG(msg, ## __VA_ARGS__); \
50 #define TRACE(msg, ...) \
54 #define LOG(msg, ...) do { \
55 fprintf(stderr, "%s:%s():L%d: " msg "\n", \
56 __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \
59 /* This is all part of the "official" NBD API */
61 #define NBD_REQUEST_SIZE (4 + 4 + 8 + 8 + 4)
62 #define NBD_REPLY_SIZE (4 + 4 + 8)
63 #define NBD_REQUEST_MAGIC 0x25609513
64 #define NBD_REPLY_MAGIC 0x67446698
65 #define NBD_OPTS_MAGIC 0x49484156454F5054LL
66 #define NBD_CLIENT_MAGIC 0x0000420281861253LL
68 #define NBD_SET_SOCK _IO(0xab, 0)
69 #define NBD_SET_BLKSIZE _IO(0xab, 1)
70 #define NBD_SET_SIZE _IO(0xab, 2)
71 #define NBD_DO_IT _IO(0xab, 3)
72 #define NBD_CLEAR_SOCK _IO(0xab, 4)
73 #define NBD_CLEAR_QUE _IO(0xab, 5)
74 #define NBD_PRINT_DEBUG _IO(0xab, 6)
75 #define NBD_SET_SIZE_BLOCKS _IO(0xab, 7)
76 #define NBD_DISCONNECT _IO(0xab, 8)
77 #define NBD_SET_TIMEOUT _IO(0xab, 9)
78 #define NBD_SET_FLAGS _IO(0xab, 10)
80 #define NBD_OPT_EXPORT_NAME (1 << 0)
82 /* Definitions for opaque data types */
84 typedef struct NBDRequest NBDRequest;
87 QSIMPLEQ_ENTRY(NBDRequest) entry;
94 void (*close)(NBDExport *exp);
101 QTAILQ_HEAD(, NBDClient) clients;
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_opts(QemuOpts *opts)
204 Error *local_err = NULL;
205 int fd = inet_connect_opts(opts, &local_err, NULL, NULL);
206 if (local_err != NULL) {
207 qerror_report_err(local_err);
208 error_free(local_err);
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)
223 Error *local_err = NULL;
224 int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
226 if (local_err != NULL) {
227 qerror_report_err(local_err);
228 error_free(local_err);
233 int unix_socket_incoming(const char *path)
235 Error *local_err = NULL;
236 int fd = unix_listen(path, NULL, 0, &local_err);
238 if (local_err != NULL) {
239 qerror_report_err(local_err);
240 error_free(local_err);
245 int unix_socket_outgoing(const char *path)
247 Error *local_err = NULL;
248 int fd = unix_connect(path, &local_err);
250 if (local_err != NULL) {
251 qerror_report_err(local_err);
252 error_free(local_err);
257 /* Basic flow for negotiation
284 static int nbd_receive_options(NBDClient *client)
286 int csock = client->sock;
288 uint32_t tmp, length;
293 [ 0 .. 3] reserved (0)
294 [ 4 .. 11] NBD_OPTS_MAGIC
295 [12 .. 15] NBD_OPT_EXPORT_NAME
297 [20 .. xx] export name (length bytes)
301 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
305 TRACE("Checking reserved");
307 LOG("Bad reserved received");
311 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
315 TRACE("Checking reserved");
316 if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) {
317 LOG("Bad magic received");
321 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
325 TRACE("Checking option");
326 if (tmp != be32_to_cpu(NBD_OPT_EXPORT_NAME)) {
327 LOG("Bad option received");
331 if (read_sync(csock, &length, sizeof(length)) != sizeof(length)) {
335 TRACE("Checking length");
336 length = be32_to_cpu(length);
338 LOG("Bad length received");
341 if (read_sync(csock, name, length) != length) {
347 client->exp = nbd_export_find(name);
349 LOG("export not found");
353 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
354 nbd_export_get(client->exp);
356 TRACE("Option negotiation succeeded.");
362 static int nbd_send_negotiate(NBDClient *client)
364 int csock = client->sock;
365 char buf[8 + 8 + 8 + 128];
367 const int myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
368 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
370 /* Negotiation header without options:
371 [ 0 .. 7] passwd ("NBDMAGIC")
372 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
374 [24 .. 25] server flags (0)
375 [24 .. 27] export flags
376 [28 .. 151] reserved (0)
378 Negotiation header with options, part 1:
379 [ 0 .. 7] passwd ("NBDMAGIC")
380 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
381 [16 .. 17] server flags (0)
383 part 2 (after options are sent):
385 [26 .. 27] export flags
386 [28 .. 151] reserved (0)
389 qemu_set_block(csock);
392 TRACE("Beginning negotiation.");
393 memset(buf, 0, sizeof(buf));
394 memcpy(buf, "NBDMAGIC", 8);
396 assert ((client->exp->nbdflags & ~65535) == 0);
397 cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC);
398 cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size);
399 cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
401 cpu_to_be64w((uint64_t*)(buf + 8), NBD_OPTS_MAGIC);
405 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
410 if (write_sync(csock, buf, 18) != 18) {
414 rc = nbd_receive_options(client);
416 LOG("option negotiation failed");
420 assert ((client->exp->nbdflags & ~65535) == 0);
421 cpu_to_be64w((uint64_t*)(buf + 18), client->exp->size);
422 cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
423 if (write_sync(csock, buf + 18, sizeof(buf) - 18) != sizeof(buf) - 18) {
429 TRACE("Negotiation succeeded.");
432 qemu_set_nonblock(csock);
436 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
437 off_t *size, size_t *blocksize)
444 TRACE("Receiving negotiation.");
448 if (read_sync(csock, buf, 8) != 8) {
454 if (strlen(buf) == 0) {
455 LOG("server connection closed");
459 TRACE("Magic is %c%c%c%c%c%c%c%c",
460 qemu_isprint(buf[0]) ? buf[0] : '.',
461 qemu_isprint(buf[1]) ? buf[1] : '.',
462 qemu_isprint(buf[2]) ? buf[2] : '.',
463 qemu_isprint(buf[3]) ? buf[3] : '.',
464 qemu_isprint(buf[4]) ? buf[4] : '.',
465 qemu_isprint(buf[5]) ? buf[5] : '.',
466 qemu_isprint(buf[6]) ? buf[6] : '.',
467 qemu_isprint(buf[7]) ? buf[7] : '.');
469 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
470 LOG("Invalid magic received");
474 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
478 magic = be64_to_cpu(magic);
479 TRACE("Magic is 0x%" PRIx64, magic);
482 uint32_t reserved = 0;
486 TRACE("Checking magic (opts_magic)");
487 if (magic != NBD_OPTS_MAGIC) {
488 LOG("Bad magic received");
491 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
492 LOG("flags read failed");
495 *flags = be16_to_cpu(tmp) << 16;
496 /* reserved for future use */
497 if (write_sync(csock, &reserved, sizeof(reserved)) !=
499 LOG("write failed (reserved)");
502 /* write the export name */
503 magic = cpu_to_be64(magic);
504 if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
505 LOG("write failed (magic)");
508 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
509 if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
510 LOG("write failed (opt)");
513 namesize = cpu_to_be32(strlen(name));
514 if (write_sync(csock, &namesize, sizeof(namesize)) !=
516 LOG("write failed (namesize)");
519 if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
520 LOG("write failed (name)");
524 TRACE("Checking magic (cli_magic)");
526 if (magic != NBD_CLIENT_MAGIC) {
527 LOG("Bad magic received");
532 if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
536 *size = be64_to_cpu(s);
538 TRACE("Size is %" PRIu64, *size);
541 if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
542 LOG("read failed (flags)");
545 *flags = be32_to_cpup(flags);
547 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
548 LOG("read failed (tmp)");
551 *flags |= be32_to_cpu(tmp);
553 if (read_sync(csock, &buf, 124) != 124) {
554 LOG("read failed (buf)");
564 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
566 TRACE("Setting NBD socket");
568 if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
570 LOG("Failed to set NBD socket");
574 TRACE("Setting block size to %lu", (unsigned long)blocksize);
576 if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
578 LOG("Failed setting NBD block size");
582 TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
584 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
586 LOG("Failed setting size (in blocks)");
590 if (ioctl(fd, NBD_SET_FLAGS, flags) < 0) {
591 if (errno == ENOTTY) {
592 int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
593 TRACE("Setting readonly attribute");
595 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
597 LOG("Failed setting read-only attribute");
602 LOG("Failed setting flags");
607 TRACE("Negotiation ended");
612 int nbd_disconnect(int fd)
614 ioctl(fd, NBD_CLEAR_QUE);
615 ioctl(fd, NBD_DISCONNECT);
616 ioctl(fd, NBD_CLEAR_SOCK);
620 int nbd_client(int fd)
625 TRACE("Doing NBD loop");
627 ret = ioctl(fd, NBD_DO_IT);
628 if (ret < 0 && errno == EPIPE) {
629 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
630 * the socket via NBD_DISCONNECT. We do not want to return 1 in
637 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
639 TRACE("Clearing NBD queue");
640 ioctl(fd, NBD_CLEAR_QUE);
642 TRACE("Clearing NBD socket");
643 ioctl(fd, NBD_CLEAR_SOCK);
649 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
654 int nbd_disconnect(int fd)
659 int nbd_client(int fd)
665 ssize_t nbd_send_request(int csock, struct nbd_request *request)
667 uint8_t buf[NBD_REQUEST_SIZE];
670 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
671 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
672 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
673 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
674 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
676 TRACE("Sending request to client: "
677 "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
678 request->from, request->len, request->handle, request->type);
680 ret = write_sync(csock, buf, sizeof(buf));
685 if (ret != sizeof(buf)) {
686 LOG("writing to socket failed");
692 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
694 uint8_t buf[NBD_REQUEST_SIZE];
698 ret = read_sync(csock, buf, sizeof(buf));
703 if (ret != sizeof(buf)) {
709 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
710 [ 4 .. 7] type (0 == READ, 1 == WRITE)
716 magic = be32_to_cpup((uint32_t*)buf);
717 request->type = be32_to_cpup((uint32_t*)(buf + 4));
718 request->handle = be64_to_cpup((uint64_t*)(buf + 8));
719 request->from = be64_to_cpup((uint64_t*)(buf + 16));
720 request->len = be32_to_cpup((uint32_t*)(buf + 24));
722 TRACE("Got request: "
723 "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
724 magic, request->type, request->from, request->len);
726 if (magic != NBD_REQUEST_MAGIC) {
727 LOG("invalid magic (got 0x%x)", magic);
733 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
735 uint8_t buf[NBD_REPLY_SIZE];
739 ret = read_sync(csock, buf, sizeof(buf));
744 if (ret != sizeof(buf)) {
750 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
751 [ 4 .. 7] error (0 == no error)
755 magic = be32_to_cpup((uint32_t*)buf);
756 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
757 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
760 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
761 magic, reply->error, reply->handle);
763 if (magic != NBD_REPLY_MAGIC) {
764 LOG("invalid magic (got 0x%x)", magic);
770 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
772 uint8_t buf[NBD_REPLY_SIZE];
776 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
777 [ 4 .. 7] error (0 == no error)
780 cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
781 cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
782 cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
784 TRACE("Sending response to client");
786 ret = write_sync(csock, buf, sizeof(buf));
791 if (ret != sizeof(buf)) {
792 LOG("writing to socket failed");
798 #define MAX_NBD_REQUESTS 16
800 void nbd_client_get(NBDClient *client)
805 void nbd_client_put(NBDClient *client)
807 if (--client->refcount == 0) {
808 /* The last reference should be dropped by client->close,
809 * which is called by nbd_client_close.
811 assert(client->closing);
813 qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
817 QTAILQ_REMOVE(&client->exp->clients, client, next);
818 nbd_export_put(client->exp);
824 void nbd_client_close(NBDClient *client)
826 if (client->closing) {
830 client->closing = true;
832 /* Force requests to finish. They will drop their own references,
833 * then we'll close the socket and free the NBDClient.
835 shutdown(client->sock, 2);
837 /* Also tell the client, so that they release their reference. */
839 client->close(client);
843 static NBDRequest *nbd_request_get(NBDClient *client)
847 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
848 client->nb_requests++;
850 req = g_slice_new0(NBDRequest);
851 nbd_client_get(client);
852 req->client = client;
856 static void nbd_request_put(NBDRequest *req)
858 NBDClient *client = req->client;
861 qemu_vfree(req->data);
863 g_slice_free(NBDRequest, req);
865 if (client->nb_requests-- == MAX_NBD_REQUESTS) {
868 nbd_client_put(client);
871 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
872 off_t size, uint32_t nbdflags,
873 void (*close)(NBDExport *))
875 NBDExport *exp = g_malloc0(sizeof(NBDExport));
877 QTAILQ_INIT(&exp->clients);
879 exp->dev_offset = dev_offset;
880 exp->nbdflags = nbdflags;
881 exp->size = size == -1 ? bdrv_getlength(bs) : size;
887 NBDExport *nbd_export_find(const char *name)
890 QTAILQ_FOREACH(exp, &exports, next) {
891 if (strcmp(name, exp->name) == 0) {
899 void nbd_export_set_name(NBDExport *exp, const char *name)
901 if (exp->name == name) {
906 if (exp->name != NULL) {
909 QTAILQ_REMOVE(&exports, exp, next);
914 exp->name = g_strdup(name);
915 QTAILQ_INSERT_TAIL(&exports, exp, next);
920 void nbd_export_close(NBDExport *exp)
922 NBDClient *client, *next;
925 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
926 nbd_client_close(client);
928 nbd_export_set_name(exp, NULL);
936 void nbd_export_get(NBDExport *exp)
938 assert(exp->refcount > 0);
942 void nbd_export_put(NBDExport *exp)
944 assert(exp->refcount > 0);
945 if (exp->refcount == 1) {
946 nbd_export_close(exp);
949 if (--exp->refcount == 0) {
950 assert(exp->name == NULL);
960 BlockDriverState *nbd_export_get_blockdev(NBDExport *exp)
965 void nbd_export_close_all(void)
967 NBDExport *exp, *next;
969 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
970 nbd_export_close(exp);
974 static int nbd_can_read(void *opaque);
975 static void nbd_read(void *opaque);
976 static void nbd_restart_write(void *opaque);
978 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
981 NBDClient *client = req->client;
982 int csock = client->sock;
985 qemu_co_mutex_lock(&client->send_lock);
986 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
987 nbd_restart_write, client);
988 client->send_coroutine = qemu_coroutine_self();
991 rc = nbd_send_reply(csock, reply);
993 socket_set_cork(csock, 1);
994 rc = nbd_send_reply(csock, reply);
996 ret = qemu_co_send(csock, req->data, len);
1001 socket_set_cork(csock, 0);
1004 client->send_coroutine = NULL;
1005 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1006 qemu_co_mutex_unlock(&client->send_lock);
1010 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
1012 NBDClient *client = req->client;
1013 int csock = client->sock;
1017 client->recv_coroutine = qemu_coroutine_self();
1018 rc = nbd_receive_request(csock, request);
1020 if (rc != -EAGAIN) {
1026 if (request->len > NBD_MAX_BUFFER_SIZE) {
1027 LOG("len (%u) is larger than max len (%u)",
1028 request->len, NBD_MAX_BUFFER_SIZE);
1033 if ((request->from + request->len) < request->from) {
1034 LOG("integer overflow detected! "
1035 "you're probably being attacked");
1040 TRACE("Decoding type");
1042 command = request->type & NBD_CMD_MASK_COMMAND;
1043 if (command == NBD_CMD_READ || command == NBD_CMD_WRITE) {
1044 req->data = qemu_blockalign(client->exp->bs, request->len);
1046 if (command == NBD_CMD_WRITE) {
1047 TRACE("Reading %u byte(s)", request->len);
1049 if (qemu_co_recv(csock, req->data, request->len) != request->len) {
1050 LOG("reading from socket failed");
1058 client->recv_coroutine = NULL;
1062 static void nbd_trip(void *opaque)
1064 NBDClient *client = opaque;
1065 NBDExport *exp = client->exp;
1067 struct nbd_request request;
1068 struct nbd_reply reply;
1071 TRACE("Reading request.");
1072 if (client->closing) {
1076 req = nbd_request_get(client);
1077 ret = nbd_co_receive_request(req, &request);
1078 if (ret == -EAGAIN) {
1085 reply.handle = request.handle;
1093 if ((request.from + request.len) > exp->size) {
1094 LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
1095 ", Offset: %" PRIu64 "\n",
1096 request.from, request.len,
1097 (uint64_t)exp->size, (uint64_t)exp->dev_offset);
1098 LOG("requested operation past EOF--bad client?");
1099 goto invalid_request;
1102 switch (request.type & NBD_CMD_MASK_COMMAND) {
1104 TRACE("Request type is READ");
1106 if (request.type & NBD_CMD_FLAG_FUA) {
1107 ret = bdrv_co_flush(exp->bs);
1109 LOG("flush failed");
1115 ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
1116 req->data, request.len / 512);
1118 LOG("reading from file failed");
1123 TRACE("Read %u byte(s)", request.len);
1124 if (nbd_co_send_reply(req, &reply, request.len) < 0)
1128 TRACE("Request type is WRITE");
1130 if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1131 TRACE("Server is read-only, return error");
1132 reply.error = EROFS;
1136 TRACE("Writing to device");
1138 ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
1139 req->data, request.len / 512);
1141 LOG("writing to file failed");
1146 if (request.type & NBD_CMD_FLAG_FUA) {
1147 ret = bdrv_co_flush(exp->bs);
1149 LOG("flush failed");
1155 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1160 TRACE("Request type is DISCONNECT");
1164 TRACE("Request type is FLUSH");
1166 ret = bdrv_co_flush(exp->bs);
1168 LOG("flush failed");
1171 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1176 TRACE("Request type is TRIM");
1177 ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
1180 LOG("discard failed");
1183 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1188 LOG("invalid request type (%u) received", request.type);
1190 reply.error = -EINVAL;
1192 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1198 TRACE("Request/Reply complete");
1201 nbd_request_put(req);
1205 nbd_request_put(req);
1206 nbd_client_close(client);
1209 static int nbd_can_read(void *opaque)
1211 NBDClient *client = opaque;
1213 return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
1216 static void nbd_read(void *opaque)
1218 NBDClient *client = opaque;
1220 if (client->recv_coroutine) {
1221 qemu_coroutine_enter(client->recv_coroutine, NULL);
1223 qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
1227 static void nbd_restart_write(void *opaque)
1229 NBDClient *client = opaque;
1231 qemu_coroutine_enter(client->send_coroutine, NULL);
1234 NBDClient *nbd_client_new(NBDExport *exp, int csock,
1235 void (*close)(NBDClient *))
1238 client = g_malloc0(sizeof(NBDClient));
1239 client->refcount = 1;
1241 client->sock = csock;
1242 if (nbd_send_negotiate(client) < 0) {
1246 client->close = close;
1247 qemu_co_mutex_init(&client->send_lock);
1248 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1251 QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1252 nbd_export_get(exp);