#define DEBUG2( a,b ) printf( a,b )
#define DEBUG3( a,b,c ) printf( a,b,c )
#define DEBUG4( a,b,c,d ) printf( a,b,c,d )
+#define DEBUG5( a,b,c,d,e ) printf( a,b,c,d,e )
#else
#define DEBUG( a )
#define DEBUG2( a,b )
#define DEBUG3( a,b,c )
#define DEBUG4( a,b,c,d )
+#define DEBUG5( a,b,c,d,e )
#endif
#ifndef PACKAGE_VERSION
#define PACKAGE_VERSION ""
#define F_SPARSE 16 /**< flag to tell us copyronwrite should use a sparse file */
#define F_SDP 32 /**< flag to tell us the export should be done using the Socket Direct Protocol for RDMA */
#define F_SYNC 64 /**< Whether to fsync() after a write */
+#define F_FLUSH 128 /**< Whether server wants FLUSH to be sent by the client */
+#define F_FUA 256 /**< Whether server wants FUA to be sent by the client */
+#define F_ROTATIONAL 512 /**< Whether server wants the client to implement the elevator algorithm */
GHashTable *children;
char pidfname[256]; /**< name of our PID file */
char pidftemplate[256]; /**< template to be used for the filename of the PID file */
disconnects */
gchar* servename; /**< name of the export as selected by nbd-client */
int max_connections; /**< maximum number of opened connections */
+ gchar* transactionlog;/**< filename for transaction log */
} SERVER;
/**
u32 difffilelen; /**< number of pages in difffile */
u32 *difmap; /**< see comment on the global difmap for this one */
gboolean modern; /**< client was negotiated using modern negotiation protocol */
+ int transactionlogfd;/**< fd for transaction log */
} CLIENT;
/**
g_free(server->prerun);
if(server->postrun)
g_free(server->postrun);
+ if(server->transactionlog)
+ g_free(server->transactionlog);
g_free(server);
}
if(s->postrun)
serve->postrun = g_strdup(s->postrun);
+
+ if(s->transactionlog)
+ serve->transactionlog = g_strdup(s->transactionlog);
if(s->servename)
serve->servename = g_strdup(s->servename);
{ "virtstyle", FALSE, PARAM_STRING, &(virtstyle), 0 },
{ "prerun", FALSE, PARAM_STRING, &(s.prerun), 0 },
{ "postrun", FALSE, PARAM_STRING, &(s.postrun), 0 },
+ { "transactionlog", FALSE, PARAM_STRING, &(s.transactionlog), 0 },
{ "readonly", FALSE, PARAM_BOOL, &(s.flags), F_READONLY },
{ "multifile", FALSE, PARAM_BOOL, &(s.flags), F_MULTIFILE },
{ "copyonwrite", FALSE, PARAM_BOOL, &(s.flags), F_COPYONWRITE },
{ "sparse_cow", FALSE, PARAM_BOOL, &(s.flags), F_SPARSE },
{ "sdp", FALSE, PARAM_BOOL, &(s.flags), F_SDP },
{ "sync", FALSE, PARAM_BOOL, &(s.flags), F_SYNC },
+ { "flush", FALSE, PARAM_BOOL, &(s.flags), F_FLUSH },
+ { "fua", FALSE, PARAM_BOOL, &(s.flags), F_FUA },
+ { "rotational", FALSE, PARAM_BOOL, &(s.flags), F_ROTATIONAL },
{ "listenaddr", FALSE, PARAM_STRING, &(s.listenaddr), 0 },
{ "maxconnections", FALSE, PARAM_INT, &(s.max_connections), 0 },
};
* @param client The client we're serving for
* @return The number of bytes actually written, or -1 in case of an error
**/
-ssize_t rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
+ssize_t rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client, int fua) {
int fhandle;
off_t foffset;
size_t maxbytes;
if(maxbytes && len > maxbytes)
len = maxbytes;
- DEBUG4("(WRITE to fd %d offset %llu len %u), ", fhandle, foffset, len);
+ DEBUG5("(WRITE to fd %d offset %llu len %u fua %d), ", fhandle, foffset, len, fua);
myseek(fhandle, foffset);
retval = write(fhandle, buf, len);
if(client->server->flags & F_SYNC) {
fsync(fhandle);
+ } else if (fua) {
+#ifdef USE_SYNC_FILE_RANGE
+ sync_file_range(fhandle, foffset, len,
+ SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE |
+ SYNC_FILE_RANGE_WAIT_AFTER);
+#else
+ fdatasync(fhandle);
+#endif
}
return retval;
}
* Call rawexpwrite repeatedly until all data has been written.
* @return 0 on success, nonzero on failure
**/
-int rawexpwrite_fully(off_t a, char *buf, size_t len, CLIENT *client) {
+int rawexpwrite_fully(off_t a, char *buf, size_t len, CLIENT *client, int fua) {
ssize_t ret=0;
- while(len > 0 && (ret=rawexpwrite(a, buf, len, client)) > 0 ) {
+ while(len > 0 && (ret=rawexpwrite(a, buf, len, client, fua)) > 0 ) {
a += ret;
buf += ret;
len -= ret;
* @param client The client we're going to write for.
* @return 0 on success, nonzero on failure
**/
-int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
+int expwrite(off_t a, char *buf, size_t len, CLIENT *client, int fua) {
char pagebuf[DIFFPAGESIZE];
off_t mapcnt,mapl,maph;
off_t wrlen,rdlen;
off_t offset;
if (!(client->server->flags & F_COPYONWRITE))
- return(rawexpwrite_fully(a, buf, len, client));
+ return(rawexpwrite_fully(a, buf, len, client, fua));
DEBUG3("Asked to write %d bytes at %llu.\n", len, (unsigned long long)a);
mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
}
len-=wrlen ; a+=wrlen ; buf+=wrlen ;
}
+ if (client->server->flags & F_SYNC) {
+ fsync(client->difffile);
+ } else if (fua) {
+ /* open question: would it be cheaper to do multiple sync_file_ranges?
+ as we iterate through the above?
+ */
+ fdatasync(client->difffile);
+ }
+ return 0;
+}
+
+int expflush(CLIENT *client) {
+ int fhandle;
+ off_t foffset;
+ size_t maxbytes;
+ gint i;
+
+ if (client->server->flags & F_COPYONWRITE) {
+ return fsync(client->difffile);
+ }
+
+ for (i = 0; i < client->export->len; i++) {
+ FILE_INFO fi = g_array_index(client->export, FILE_INFO, i);
+ if (fsync(fi.fhandle) < 0)
+ return -1;
+ }
+
return 0;
}
client->exportsize = OFFT_MAX;
client->net = net;
client->modern = TRUE;
+ client->transactionlogfd = -1;
free(name);
return client;
}
err("Negotiation failed: %m");
if (client->server->flags & F_READONLY)
flags |= NBD_FLAG_READ_ONLY;
+ if (client->server->flags & F_FLUSH)
+ flags |= NBD_FLAG_SEND_FLUSH;
+ if (client->server->flags & F_FUA)
+ flags |= NBD_FLAG_SEND_FUA;
+ if (client->server->flags & F_ROTATIONAL)
+ flags |= NBD_FLAG_ROTATIONAL;
if (!client->modern) {
/* oldstyle */
flags = htonl(flags);
}
/** sending macro. */
-#define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
+#define SEND(net,reply) { writeit( net, &reply, sizeof( reply )); \
+ if (client->transactionlogfd != -1) \
+ writeit(client->transactionlogfd, &reply, sizeof(reply)); }
/** error macro. */
#define ERROR(client,reply,errcode) { reply.error = htonl(errcode); SEND(client->net,reply); reply.error = 0; }
/**
size_t len;
size_t currlen;
size_t writelen;
+ uint16_t command;
#ifdef DODBG
i++;
printf("%d: ", i);
#endif
readit(client->net, &request, sizeof(request));
+ if (client->transactionlogfd != -1)
+ writeit(client->transactionlogfd, &request, sizeof(request));
+
request.from = ntohll(request.from);
request.type = ntohl(request.type);
+ command = request.type & NBD_CMD_MASK_COMMAND;
- if (request.type==NBD_CMD_DISC) {
+ if (command==NBD_CMD_DISC) {
msg2(LOG_INFO, "Disconnect request received.");
if (client->server->flags & F_COPYONWRITE) {
if (client->difmap) g_free(client->difmap) ;
currlen = len;
}
#ifdef DODBG
- printf("%s from %llu (%llu) len %d, ", request.type ? "WRITE" :
+ printf("%s from %llu (%llu) len %d, ", command ? "WRITE" :
"READ", (unsigned long long)request.from,
(unsigned long long)request.from / 512, len);
#endif
memcpy(reply.handle, request.handle, sizeof(reply.handle));
- if ((request.from + len) > (OFFT_MAX)) {
- DEBUG("[Number too large!]");
- ERROR(client, reply, EINVAL);
- continue;
- }
- if (((ssize_t)((off_t)request.from + len) > client->exportsize)) {
- DEBUG("[RANGE!]");
- ERROR(client, reply, EINVAL);
- continue;
+ if ((command==NBD_CMD_WRITE) || (command==NBD_CMD_READ)) {
+ if ((request.from + len) > (OFFT_MAX)) {
+ DEBUG("[Number too large!]");
+ ERROR(client, reply, EINVAL);
+ continue;
+ }
+
+ if (((ssize_t)((off_t)request.from + len) > client->exportsize)) {
+ DEBUG("[RANGE!]");
+ ERROR(client, reply, EINVAL);
+ continue;
+ }
}
- if (request.type==NBD_CMD_WRITE) {
+ if (command==NBD_CMD_WRITE) {
DEBUG("wr: net->buf, ");
while(len > 0) {
readit(client->net, buf, currlen);
ERROR(client, reply, EPERM);
continue;
}
- if (expwrite(request.from, buf, len, client)) {
+ if (expwrite(request.from, buf, len, client,
+ request.type & NBD_CMD_FLAG_FUA)) {
DEBUG("Write failed: %m" );
ERROR(client, reply, errno);
continue;
}
continue;
}
- /* READ */
-
- DEBUG("exp->buf, ");
- memcpy(buf, &reply, sizeof(struct nbd_reply));
- p = buf + sizeof(struct nbd_reply);
- writelen = currlen + sizeof(struct nbd_reply);
- while(len > 0) {
- if (expread(request.from, p, currlen, client)) {
- DEBUG("Read failed: %m");
+
+ if (command==NBD_CMD_FLUSH) {
+ DEBUG("fl: ");
+ if (expflush(client)) {
+ DEBUG("Flush failed: %m");
ERROR(client, reply, errno);
continue;
}
+ SEND(client->net, reply);
+ DEBUG("OK!\n");
+ continue;
+ }
- DEBUG("buf->net, ");
- writeit(client->net, buf, writelen);
- len -= currlen;
- currlen = (len < BUFSIZE) ? len : BUFSIZE;
- p = buf;
- writelen = currlen;
+ if (command==NBD_CMD_READ) {
+ DEBUG("exp->buf, ");
+ memcpy(buf, &reply, sizeof(struct nbd_reply));
+ if (client->transactionlogfd != -1)
+ writeit(client->transactionlogfd, &reply, sizeof(reply));
+ p = buf + sizeof(struct nbd_reply);
+ writelen = currlen + sizeof(struct nbd_reply);
+ while(len > 0) {
+ if (expread(request.from, p, currlen, client)) {
+ DEBUG("Read failed: %m");
+ ERROR(client, reply, errno);
+ continue;
+ }
+
+ DEBUG("buf->net, ");
+ writeit(client->net, buf, writelen);
+ len -= currlen;
+ request.from += currlen;
+ currlen = (len < BUFSIZE) ? len : BUFSIZE;
+ p = buf;
+ writelen = currlen;
+ }
+ DEBUG("OK!\n");
+ continue;
}
- DEBUG("OK!\n");
+
+ DEBUG ("Ignoring unknown command\n");
}
return 0;
}
* @param client a connected client
**/
void serveconnection(CLIENT *client) {
+ if (client->server->transactionlog && (client->transactionlogfd == -1))
+ {
+ if (-1 == (client->transactionlogfd = open(client->server->transactionlog,
+ O_WRONLY | O_CREAT,
+ S_IRUSR | S_IWUSR)))
+ g_warning("Could not open transaction log %s",
+ client->server->transactionlog);
+ }
+
if(do_run(client->server->prerun, client->exportname)) {
exit(EXIT_FAILURE);
}
mainloop(client);
do_run(client->server->postrun, client->exportname);
+
+ if (-1 != client->transactionlogfd)
+ {
+ close(client->transactionlogfd);
+ client->transactionlogfd = -1;
+ }
}
/**
client->server=serve;
client->exportsize=OFFT_MAX;
client->net=net;
+ client->transactionlogfd = -1;
}
set_peername(net, client);
if (!authorized_client(client)) {
projects / nbd.git / blobdiff