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/>.
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"
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 QTAILQ_ENTRY(NBDExport) next;
104 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
108 void (*close)(NBDClient *client);
113 Coroutine *recv_coroutine;
116 Coroutine *send_coroutine;
118 QTAILQ_ENTRY(NBDClient) next;
123 /* That's all folks */
125 ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
130 if (qemu_in_coroutine()) {
132 return qemu_co_recv(fd, buffer, size);
134 return qemu_co_send(fd, buffer, size);
138 while (offset < size) {
142 len = qemu_recv(fd, buffer + offset, size - offset, 0);
144 len = send(fd, buffer + offset, size - offset, 0);
148 err = socket_error();
150 /* recoverable error */
151 if (err == EINTR || (offset > 0 && err == EAGAIN)) {
155 /* unrecoverable error */
170 static ssize_t read_sync(int fd, void *buffer, size_t size)
172 /* Sockets are kept in blocking mode in the negotiation phase. After
173 * that, a non-readable socket simply means that another thread stole
174 * our request/reply. Synchronization is done with recv_coroutine, so
175 * that this is coroutine-safe.
177 return nbd_wr_sync(fd, buffer, size, true);
180 static ssize_t write_sync(int fd, void *buffer, size_t size)
184 /* For writes, we do expect the socket to be writable. */
185 ret = nbd_wr_sync(fd, buffer, size, false);
186 } while (ret == -EAGAIN);
190 static void combine_addr(char *buf, size_t len, const char* address,
193 /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
194 if (strstr(address, ":")) {
195 snprintf(buf, len, "[%s]:%u", address, port);
197 snprintf(buf, len, "%s:%u", address, port);
201 int tcp_socket_outgoing_opts(QemuOpts *opts)
203 Error *local_err = NULL;
204 int fd = inet_connect_opts(opts, &local_err, NULL, NULL);
205 if (local_err != NULL) {
206 qerror_report_err(local_err);
207 error_free(local_err);
213 int tcp_socket_incoming(const char *address, uint16_t port)
215 char address_and_port[128];
216 combine_addr(address_and_port, 128, address, port);
217 return tcp_socket_incoming_spec(address_and_port);
220 int tcp_socket_incoming_spec(const char *address_and_port)
222 Error *local_err = NULL;
223 int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
225 if (local_err != NULL) {
226 qerror_report_err(local_err);
227 error_free(local_err);
232 int unix_socket_incoming(const char *path)
234 Error *local_err = NULL;
235 int fd = unix_listen(path, NULL, 0, &local_err);
237 if (local_err != NULL) {
238 qerror_report_err(local_err);
239 error_free(local_err);
244 int unix_socket_outgoing(const char *path)
246 Error *local_err = NULL;
247 int fd = unix_connect(path, &local_err);
249 if (local_err != NULL) {
250 qerror_report_err(local_err);
251 error_free(local_err);
256 /* Basic flow for negotiation
283 static int nbd_receive_options(NBDClient *client)
285 int csock = client->sock;
287 uint32_t tmp, length;
292 [ 0 .. 3] reserved (0)
293 [ 4 .. 11] NBD_OPTS_MAGIC
294 [12 .. 15] NBD_OPT_EXPORT_NAME
296 [20 .. xx] export name (length bytes)
300 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
304 TRACE("Checking reserved");
306 LOG("Bad reserved received");
310 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
314 TRACE("Checking reserved");
315 if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) {
316 LOG("Bad magic received");
320 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
324 TRACE("Checking option");
325 if (tmp != be32_to_cpu(NBD_OPT_EXPORT_NAME)) {
326 LOG("Bad option received");
330 if (read_sync(csock, &length, sizeof(length)) != sizeof(length)) {
334 TRACE("Checking length");
335 length = be32_to_cpu(length);
337 LOG("Bad length received");
340 if (read_sync(csock, name, length) != length) {
346 client->exp = nbd_export_find(name);
348 LOG("export not found");
352 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
353 nbd_export_get(client->exp);
355 TRACE("Option negotiation succeeded.");
361 static int nbd_send_negotiate(NBDClient *client)
363 int csock = client->sock;
364 char buf[8 + 8 + 8 + 128];
366 const int myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
367 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
369 /* Negotiation header without options:
370 [ 0 .. 7] passwd ("NBDMAGIC")
371 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
373 [24 .. 25] server flags (0)
374 [24 .. 27] export flags
375 [28 .. 151] reserved (0)
377 Negotiation header with options, part 1:
378 [ 0 .. 7] passwd ("NBDMAGIC")
379 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
380 [16 .. 17] server flags (0)
382 part 2 (after options are sent):
384 [26 .. 27] export flags
385 [28 .. 151] reserved (0)
388 qemu_set_block(csock);
391 TRACE("Beginning negotiation.");
392 memset(buf, 0, sizeof(buf));
393 memcpy(buf, "NBDMAGIC", 8);
395 assert ((client->exp->nbdflags & ~65535) == 0);
396 cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC);
397 cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size);
398 cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
400 cpu_to_be64w((uint64_t*)(buf + 8), NBD_OPTS_MAGIC);
404 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
409 if (write_sync(csock, buf, 18) != 18) {
413 rc = nbd_receive_options(client);
415 LOG("option negotiation failed");
419 assert ((client->exp->nbdflags & ~65535) == 0);
420 cpu_to_be64w((uint64_t*)(buf + 18), client->exp->size);
421 cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
422 if (write_sync(csock, buf + 18, sizeof(buf) - 18) != sizeof(buf) - 18) {
428 TRACE("Negotiation succeeded.");
431 qemu_set_nonblock(csock);
435 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
436 off_t *size, size_t *blocksize)
443 TRACE("Receiving negotiation.");
445 qemu_set_block(csock);
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)");
560 qemu_set_nonblock(csock);
565 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
567 TRACE("Setting NBD socket");
569 if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
571 LOG("Failed to set NBD socket");
575 TRACE("Setting block size to %lu", (unsigned long)blocksize);
577 if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
579 LOG("Failed setting NBD block size");
583 TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
585 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
587 LOG("Failed setting size (in blocks)");
591 if (ioctl(fd, NBD_SET_FLAGS, flags) < 0) {
592 if (errno == ENOTTY) {
593 int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
594 TRACE("Setting readonly attribute");
596 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
598 LOG("Failed setting read-only attribute");
603 LOG("Failed setting flags");
608 TRACE("Negotiation ended");
613 int nbd_disconnect(int fd)
615 ioctl(fd, NBD_CLEAR_QUE);
616 ioctl(fd, NBD_DISCONNECT);
617 ioctl(fd, NBD_CLEAR_SOCK);
621 int nbd_client(int fd)
626 TRACE("Doing NBD loop");
628 ret = ioctl(fd, NBD_DO_IT);
629 if (ret < 0 && errno == EPIPE) {
630 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
631 * the socket via NBD_DISCONNECT. We do not want to return 1 in
638 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
640 TRACE("Clearing NBD queue");
641 ioctl(fd, NBD_CLEAR_QUE);
643 TRACE("Clearing NBD socket");
644 ioctl(fd, NBD_CLEAR_SOCK);
650 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
655 int nbd_disconnect(int fd)
660 int nbd_client(int fd)
666 ssize_t nbd_send_request(int csock, struct nbd_request *request)
668 uint8_t buf[NBD_REQUEST_SIZE];
671 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
672 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
673 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
674 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
675 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
677 TRACE("Sending request to client: "
678 "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
679 request->from, request->len, request->handle, request->type);
681 ret = write_sync(csock, buf, sizeof(buf));
686 if (ret != sizeof(buf)) {
687 LOG("writing to socket failed");
693 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
695 uint8_t buf[NBD_REQUEST_SIZE];
699 ret = read_sync(csock, buf, sizeof(buf));
704 if (ret != sizeof(buf)) {
710 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
711 [ 4 .. 7] type (0 == READ, 1 == WRITE)
717 magic = be32_to_cpup((uint32_t*)buf);
718 request->type = be32_to_cpup((uint32_t*)(buf + 4));
719 request->handle = be64_to_cpup((uint64_t*)(buf + 8));
720 request->from = be64_to_cpup((uint64_t*)(buf + 16));
721 request->len = be32_to_cpup((uint32_t*)(buf + 24));
723 TRACE("Got request: "
724 "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
725 magic, request->type, request->from, request->len);
727 if (magic != NBD_REQUEST_MAGIC) {
728 LOG("invalid magic (got 0x%x)", magic);
734 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
736 uint8_t buf[NBD_REPLY_SIZE];
740 ret = read_sync(csock, buf, sizeof(buf));
745 if (ret != sizeof(buf)) {
751 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
752 [ 4 .. 7] error (0 == no error)
756 magic = be32_to_cpup((uint32_t*)buf);
757 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
758 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
761 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
762 magic, reply->error, reply->handle);
764 if (magic != NBD_REPLY_MAGIC) {
765 LOG("invalid magic (got 0x%x)", magic);
771 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
773 uint8_t buf[NBD_REPLY_SIZE];
777 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
778 [ 4 .. 7] error (0 == no error)
781 cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
782 cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
783 cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
785 TRACE("Sending response to client");
787 ret = write_sync(csock, buf, sizeof(buf));
792 if (ret != sizeof(buf)) {
793 LOG("writing to socket failed");
799 #define MAX_NBD_REQUESTS 16
801 void nbd_client_get(NBDClient *client)
806 void nbd_client_put(NBDClient *client)
808 if (--client->refcount == 0) {
809 /* The last reference should be dropped by client->close,
810 * which is called by nbd_client_close.
812 assert(client->closing);
814 qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
818 QTAILQ_REMOVE(&client->exp->clients, client, next);
819 nbd_export_put(client->exp);
825 void nbd_client_close(NBDClient *client)
827 if (client->closing) {
831 client->closing = true;
833 /* Force requests to finish. They will drop their own references,
834 * then we'll close the socket and free the NBDClient.
836 shutdown(client->sock, 2);
838 /* Also tell the client, so that they release their reference. */
840 client->close(client);
844 static NBDRequest *nbd_request_get(NBDClient *client)
848 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
849 client->nb_requests++;
851 req = g_slice_new0(NBDRequest);
852 nbd_client_get(client);
853 req->client = client;
857 static void nbd_request_put(NBDRequest *req)
859 NBDClient *client = req->client;
862 qemu_vfree(req->data);
864 g_slice_free(NBDRequest, req);
866 if (client->nb_requests-- == MAX_NBD_REQUESTS) {
869 nbd_client_put(client);
872 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
873 off_t size, uint32_t nbdflags,
874 void (*close)(NBDExport *))
876 NBDExport *exp = g_malloc0(sizeof(NBDExport));
878 QTAILQ_INIT(&exp->clients);
880 exp->dev_offset = dev_offset;
881 exp->nbdflags = nbdflags;
882 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);
932 void nbd_export_get(NBDExport *exp)
934 assert(exp->refcount > 0);
938 void nbd_export_put(NBDExport *exp)
940 assert(exp->refcount > 0);
941 if (exp->refcount == 1) {
942 nbd_export_close(exp);
945 if (--exp->refcount == 0) {
946 assert(exp->name == NULL);
956 BlockDriverState *nbd_export_get_blockdev(NBDExport *exp)
961 void nbd_export_close_all(void)
963 NBDExport *exp, *next;
965 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
966 nbd_export_close(exp);
970 static int nbd_can_read(void *opaque);
971 static void nbd_read(void *opaque);
972 static void nbd_restart_write(void *opaque);
974 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
977 NBDClient *client = req->client;
978 int csock = client->sock;
981 qemu_co_mutex_lock(&client->send_lock);
982 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
983 nbd_restart_write, client);
984 client->send_coroutine = qemu_coroutine_self();
987 rc = nbd_send_reply(csock, reply);
989 socket_set_cork(csock, 1);
990 rc = nbd_send_reply(csock, reply);
992 ret = qemu_co_send(csock, req->data, len);
997 socket_set_cork(csock, 0);
1000 client->send_coroutine = NULL;
1001 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1002 qemu_co_mutex_unlock(&client->send_lock);
1006 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
1008 NBDClient *client = req->client;
1009 int csock = client->sock;
1013 client->recv_coroutine = qemu_coroutine_self();
1014 rc = nbd_receive_request(csock, request);
1016 if (rc != -EAGAIN) {
1022 if (request->len > NBD_MAX_BUFFER_SIZE) {
1023 LOG("len (%u) is larger than max len (%u)",
1024 request->len, NBD_MAX_BUFFER_SIZE);
1029 if ((request->from + request->len) < request->from) {
1030 LOG("integer overflow detected! "
1031 "you're probably being attacked");
1036 TRACE("Decoding type");
1038 command = request->type & NBD_CMD_MASK_COMMAND;
1039 if (command == NBD_CMD_READ || command == NBD_CMD_WRITE) {
1040 req->data = qemu_blockalign(client->exp->bs, request->len);
1042 if (command == NBD_CMD_WRITE) {
1043 TRACE("Reading %u byte(s)", request->len);
1045 if (qemu_co_recv(csock, req->data, request->len) != request->len) {
1046 LOG("reading from socket failed");
1054 client->recv_coroutine = NULL;
1058 static void nbd_trip(void *opaque)
1060 NBDClient *client = opaque;
1061 NBDExport *exp = client->exp;
1063 struct nbd_request request;
1064 struct nbd_reply reply;
1067 TRACE("Reading request.");
1068 if (client->closing) {
1072 req = nbd_request_get(client);
1073 ret = nbd_co_receive_request(req, &request);
1074 if (ret == -EAGAIN) {
1081 reply.handle = request.handle;
1089 if ((request.from + request.len) > exp->size) {
1090 LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
1091 ", Offset: %" PRIu64 "\n",
1092 request.from, request.len,
1093 (uint64_t)exp->size, (uint64_t)exp->dev_offset);
1094 LOG("requested operation past EOF--bad client?");
1095 goto invalid_request;
1098 switch (request.type & NBD_CMD_MASK_COMMAND) {
1100 TRACE("Request type is READ");
1102 if (request.type & NBD_CMD_FLAG_FUA) {
1103 ret = bdrv_co_flush(exp->bs);
1105 LOG("flush failed");
1111 ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
1112 req->data, request.len / 512);
1114 LOG("reading from file failed");
1119 TRACE("Read %u byte(s)", request.len);
1120 if (nbd_co_send_reply(req, &reply, request.len) < 0)
1124 TRACE("Request type is WRITE");
1126 if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1127 TRACE("Server is read-only, return error");
1128 reply.error = EROFS;
1132 TRACE("Writing to device");
1134 ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
1135 req->data, request.len / 512);
1137 LOG("writing to file failed");
1142 if (request.type & NBD_CMD_FLAG_FUA) {
1143 ret = bdrv_co_flush(exp->bs);
1145 LOG("flush failed");
1151 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1156 TRACE("Request type is DISCONNECT");
1160 TRACE("Request type is FLUSH");
1162 ret = bdrv_co_flush(exp->bs);
1164 LOG("flush failed");
1167 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1172 TRACE("Request type is TRIM");
1173 ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
1176 LOG("discard failed");
1179 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1184 LOG("invalid request type (%u) received", request.type);
1186 reply.error = -EINVAL;
1188 if (nbd_co_send_reply(req, &reply, 0) < 0) {
1194 TRACE("Request/Reply complete");
1197 nbd_request_put(req);
1201 nbd_request_put(req);
1202 nbd_client_close(client);
1205 static int nbd_can_read(void *opaque)
1207 NBDClient *client = opaque;
1209 return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
1212 static void nbd_read(void *opaque)
1214 NBDClient *client = opaque;
1216 if (client->recv_coroutine) {
1217 qemu_coroutine_enter(client->recv_coroutine, NULL);
1219 qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
1223 static void nbd_restart_write(void *opaque)
1225 NBDClient *client = opaque;
1227 qemu_coroutine_enter(client->send_coroutine, NULL);
1230 NBDClient *nbd_client_new(NBDExport *exp, int csock,
1231 void (*close)(NBDClient *))
1234 client = g_malloc0(sizeof(NBDClient));
1235 client->refcount = 1;
1237 client->sock = csock;
1238 if (nbd_send_negotiate(client) < 0) {
1242 client->close = close;
1243 qemu_co_mutex_init(&client->send_lock);
1244 qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1247 QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1248 nbd_export_get(exp);