2 * Network Block Device - server
4 * Copyright 1996-1998 Pavel Machek, distribute under GPL
5 * <pavel@atrey.karlin.mff.cuni.cz>
6 * Copyright 2001-2004 Wouter Verhelst <wouter@debian.org>
7 * Copyright 2002 Anton Altaparmakov <aia21@cam.ac.uk>
9 * Version 1.0 - hopefully 64-bit-clean
10 * Version 1.1 - merging enhancements from Josh Parsons, <josh@coombs.anu.edu.au>
11 * Version 1.2 - autodetect size of block devices, thanx to Peter T. Breuer" <ptb@it.uc3m.es>
12 * Version 1.5 - can compile on Unix systems that don't have 64 bit integer
13 * type, or don't have 64 bit file offsets by defining FS_32BIT
14 * in compile options for nbd-server *only*. This can be done
15 * with make FSCHOICE=-DFS_32BIT nbd-server. (I don't have the
16 * original autoconf input file, or I would make it a configure
17 * option.) Ken Yap <ken@nlc.net.au>.
18 * Version 1.6 - fix autodetection of block device size and really make 64 bit
19 * clean on 32 bit machines. Anton Altaparmakov <aia21@cam.ac.uk>
20 * Version 2.0 - Version synchronised with client
21 * Version 2.1 - Reap zombie client processes when they exit. Removed
22 * (uncommented) the _IO magic, it's no longer necessary. Wouter
23 * Verhelst <wouter@debian.org>
24 * Version 2.2 - Auto switch to read-only mode (usefull for floppies).
25 * Version 2.3 - Fixed code so that Large File Support works. This
26 * removes the FS_32BIT compile-time directive; define
27 * _FILE_OFFSET_BITS=64 and _LARGEFILE_SOURCE if you used to be
28 * using FS_32BIT. This will allow you to use files >2GB instead of
29 * having to use the -m option. Wouter Verhelst <wouter@debian.org>
30 * Version 2.4 - Added code to keep track of children, so that we can
31 * properly kill them from initscripts. Add a call to daemon(),
32 * so that processes don't think they have to wait for us, which is
33 * interesting for initscripts as well. Wouter Verhelst
35 * Version 2.5 - Bugfix release: forgot to reset child_arraysize to
36 * zero after fork()ing, resulting in nbd-server going berserk
37 * when it receives a signal with at least one child open. Wouter
38 * Verhelst <wouter@debian.org>
39 * 10/10/2003 - Added socket option SO_KEEPALIVE (sf.net bug 819235);
40 * rectified type of mainloop::size_host (sf.net bugs 814435 and
41 * 817385); close the PID file after writing to it, so that the
42 * daemon can actually be found. Wouter Verhelst
44 * 10/10/2003 - Size of the data "size_host" was wrong and so was not
45 * correctly put in network endianness. Many types were corrected
46 * (size_t and off_t instead of int). <vspaceg@sourceforge.net>
47 * Version 2.6 - Some code cleanup.
48 * Version 2.7 - Better build system.
49 * 11/02/2004 - Doxygenified the source, modularized it a bit. Needs a
50 * lot more work, but this is a start. Wouter Verhelst
52 * 16/03/2010 - Add IPv6 support.
53 * Kitt Tientanopajai <kitt@kitty.in.th>
54 * Neutron Soutmun <neo.neutron@gmail.com>
55 * Suriya Soutmun <darksolar@gmail.com>
58 /* Includes LFS defines, which defines behaviours of some of the following
59 * headers, so must come before those */
62 #include <sys/types.h>
63 #include <sys/socket.h>
65 #include <sys/select.h> /* select */
66 #include <sys/wait.h> /* wait */
67 #ifdef HAVE_SYS_IOCTL_H
68 #include <sys/ioctl.h>
70 #include <sys/param.h>
71 #ifdef HAVE_SYS_MOUNT_H
72 #include <sys/mount.h> /* For BLKGETSIZE */
74 #include <signal.h> /* sigaction */
76 #include <netinet/tcp.h>
77 #include <netinet/in.h>
85 #include <arpa/inet.h>
95 /* used in cliserv.h, so must come first */
96 #define MY_NAME "nbd_server"
100 #include <sdp_inet.h>
103 /** Default position of the config file */
105 #define SYSCONFDIR "/etc"
107 #define CFILE SYSCONFDIR "/nbd-server/config"
109 /** Where our config file actually is */
110 gchar* config_file_pos;
112 /** What user we're running as */
114 /** What group we're running as */
115 gchar* rungroup=NULL;
116 /** whether to export using the old negotiation protocol (port-based) */
117 gboolean do_oldstyle=FALSE;
119 /** Logging macros, now nothing goes to syslog unless you say ISSERVER */
121 #define msg2(a,b) syslog(a,b)
122 #define msg3(a,b,c) syslog(a,b,c)
123 #define msg4(a,b,c,d) syslog(a,b,c,d)
125 #define msg2(a,b) g_message(b)
126 #define msg3(a,b,c) g_message(b,c)
127 #define msg4(a,b,c,d) g_message(b,c,d)
130 /* Debugging macros */
133 #define DEBUG( a ) printf( a )
134 #define DEBUG2( a,b ) printf( a,b )
135 #define DEBUG3( a,b,c ) printf( a,b,c )
136 #define DEBUG4( a,b,c,d ) printf( a,b,c,d )
139 #define DEBUG2( a,b )
140 #define DEBUG3( a,b,c )
141 #define DEBUG4( a,b,c,d )
143 #ifndef PACKAGE_VERSION
144 #define PACKAGE_VERSION ""
147 * The highest value a variable of type off_t can reach. This is a signed
148 * integer, so set all bits except for the leftmost one.
150 #define OFFT_MAX ~((off_t)1<<(sizeof(off_t)*8-1))
151 #define LINELEN 256 /**< Size of static buffer used to read the
152 authorization file (yuck) */
153 #define BUFSIZE (1024*1024) /**< Size of buffer that can hold requests */
154 #define DIFFPAGESIZE 4096 /**< diff file uses those chunks */
155 #define F_READONLY 1 /**< flag to tell us a file is readonly */
156 #define F_MULTIFILE 2 /**< flag to tell us a file is exported using -m */
157 #define F_COPYONWRITE 4 /**< flag to tell us a file is exported using
159 #define F_AUTOREADONLY 8 /**< flag to tell us a file is set to autoreadonly */
160 #define F_SPARSE 16 /**< flag to tell us copyronwrite should use a sparse file */
161 #define F_SDP 32 /**< flag to tell us the export should be done using the Socket Direct Protocol for RDMA */
162 #define F_SYNC 64 /**< Whether to fsync() after a write */
163 GHashTable *children;
164 char pidfname[256]; /**< name of our PID file */
165 char pidftemplate[256]; /**< template to be used for the filename of the PID file */
166 char default_authname[] = SYSCONFDIR "/nbd-server/allow"; /**< default name of allow file */
168 int modernsock=0; /**< Socket for the modern handler. Not used
169 if a client was only specified on the
170 command line; only port used if
171 oldstyle is set to false (and then the
172 command-line client isn't used, gna gna) */
173 char* modern_listen; /**< listenaddr value for modernsock */
176 * Types of virtuatlization
179 VIRT_NONE=0, /**< No virtualization */
180 VIRT_IPLIT, /**< Literal IP address as part of the filename */
181 VIRT_IPHASH, /**< Replacing all dots in an ip address by a / before
182 doing the same as in IPLIT */
183 VIRT_CIDR, /**< Every subnet in its own directory */
187 * Variables associated with a server.
190 gchar* exportname; /**< (unprocessed) filename of the file we're exporting */
191 off_t expected_size; /**< size of the exported file as it was told to
192 us through configuration */
193 gchar* listenaddr; /**< The IP address we're listening on */
194 unsigned int port; /**< port we're exporting this file at */
195 char* authname; /**< filename of the authorization file */
196 int flags; /**< flags associated with this exported file */
197 int socket; /**< The socket of this server. */
198 int socket_family; /**< family of the socket */
199 VIRT_STYLE virtstyle;/**< The style of virtualization, if any */
200 uint8_t cidrlen; /**< The length of the mask when we use
201 CIDR-style virtualization */
202 gchar* prerun; /**< command to be ran after connecting a client,
203 but before starting to serve */
204 gchar* postrun; /**< command that will be ran after the client
206 gchar* servename; /**< name of the export as selected by nbd-client */
210 * Variables associated with a client socket.
213 int fhandle; /**< file descriptor */
214 off_t startoff; /**< starting offset of this file */
218 off_t exportsize; /**< size of the file we're exporting */
219 char *clientname; /**< peer */
220 char *exportname; /**< (processed) filename of the file we're exporting */
221 GArray *export; /**< array of FILE_INFO of exported files;
222 array size is always 1 unless we're
223 doing the multiple file option */
224 int net; /**< The actual client socket */
225 SERVER *server; /**< The server this client is getting data from */
226 char* difffilename; /**< filename of the copy-on-write file, if any */
227 int difffile; /**< filedescriptor of copyonwrite file. @todo
228 shouldn't this be an array too? (cfr export) Or
229 make -m and -c mutually exclusive */
230 u32 difffilelen; /**< number of pages in difffile */
231 u32 *difmap; /**< see comment on the global difmap for this one */
232 gboolean modern; /**< client was negotiated using modern negotiation protocol */
236 * Type of configuration file values
239 PARAM_INT, /**< This parameter is an integer */
240 PARAM_STRING, /**< This parameter is a string */
241 PARAM_BOOL, /**< This parameter is a boolean */
245 * Configuration file values
248 gchar *paramname; /**< Name of the parameter, as it appears in
250 gboolean required; /**< Whether this is a required (as opposed to
251 optional) parameter */
252 PARAM_TYPE ptype; /**< Type of the parameter. */
253 gpointer target; /**< Pointer to where the data of this
254 parameter should be written. If ptype is
255 PARAM_BOOL, the data is or'ed rather than
257 gint flagval; /**< Flag mask for this parameter in case ptype
262 * Check whether a client is allowed to connect. Works with an authorization
263 * file which contains one line per machine, no wildcards.
265 * @param opts The client who's trying to connect.
266 * @return 0 - authorization refused, 1 - OK
268 int authorized_client(CLIENT *opts) {
269 const char *ERRMSG="Invalid entry '%s' in authfile '%s', so, refusing all connections.";
274 struct in_addr client;
275 struct in_addr cltemp;
278 if ((f=fopen(opts->server->authname,"r"))==NULL) {
279 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
280 opts->server->authname,strerror(errno)) ;
284 inet_aton(opts->clientname, &client);
285 while (fgets(line,LINELEN,f)!=NULL) {
286 if((tmp=index(line, '/'))) {
287 if(strlen(line)<=tmp-line) {
288 msg4(LOG_CRIT, ERRMSG, line, opts->server->authname);
292 if(!inet_aton(line,&addr)) {
293 msg4(LOG_CRIT, ERRMSG, line, opts->server->authname);
296 len=strtol(tmp, NULL, 0);
297 addr.s_addr>>=32-len;
298 addr.s_addr<<=32-len;
299 memcpy(&cltemp,&client,sizeof(client));
300 cltemp.s_addr>>=32-len;
301 cltemp.s_addr<<=32-len;
302 if(addr.s_addr == cltemp.s_addr) {
306 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
316 * Read data from a file descriptor into a buffer
318 * @param f a file descriptor
319 * @param buf a buffer
320 * @param len the number of bytes to be read
322 inline void readit(int f, void *buf, size_t len) {
326 if ((res = read(f, buf, len)) <= 0) {
327 if(errno != EAGAIN) {
328 err("Read failed: %m");
338 * Write data from a buffer into a filedescriptor
340 * @param f a file descriptor
341 * @param buf a buffer containing data
342 * @param len the number of bytes to be written
344 inline void writeit(int f, void *buf, size_t len) {
348 if ((res = write(f, buf, len)) <= 0)
349 err("Send failed: %m");
356 * Print out a message about how to use nbd-server. Split out to a separate
357 * function so that we can call it from multiple places
360 printf("This is nbd-server version " VERSION "\n");
361 printf("Usage: [ip:|ip6@]port file_to_export [size][kKmM] [-l authorize_file] [-r] [-m] [-c] [-C configuration file] [-p PID file name] [-o section name]\n"
362 "\t-r|--read-only\t\tread only\n"
363 "\t-m|--multi-file\t\tmultiple file\n"
364 "\t-c|--copy-on-write\tcopy on write\n"
365 "\t-C|--config-file\tspecify an alternate configuration file\n"
366 "\t-l|--authorize-file\tfile with list of hosts that are allowed to\n\t\t\t\tconnect.\n"
367 "\t-p|--pid-file\t\tspecify a filename to write our PID to\n"
368 "\t-o|--output-config\toutput a config file section for what you\n\t\t\t\tspecified on the command line, with the\n\t\t\t\tspecified section name\n\n"
369 "\tif port is set to 0, stdin is used (for running from inetd)\n"
370 "\tif file_to_export contains '%%s', it is substituted with the IP\n"
371 "\t\taddress of the machine trying to connect\n"
372 "\tif ip is set, it contains the local IP address on which we're listening.\n\tif not, the server will listen on all local IP addresses\n");
373 printf("Using configuration file %s\n", CFILE);
376 /* Dumps a config file section of the given SERVER*, and exits. */
377 void dump_section(SERVER* serve, gchar* section_header) {
378 printf("[%s]\n", section_header);
379 printf("\texportname = %s\n", serve->exportname);
380 printf("\tlistenaddr = %s\n", serve->listenaddr);
381 printf("\tport = %d\n", serve->port);
382 if(serve->flags & F_READONLY) {
383 printf("\treadonly = true\n");
385 if(serve->flags & F_MULTIFILE) {
386 printf("\tmultifile = true\n");
388 if(serve->flags & F_COPYONWRITE) {
389 printf("\tcopyonwrite = true\n");
391 if(serve->expected_size) {
392 printf("\tfilesize = %lld\n", (long long int)serve->expected_size);
394 if(serve->authname) {
395 printf("\tauthfile = %s\n", serve->authname);
401 * Parse the command line.
403 * @param argc the argc argument to main()
404 * @param argv the argv argument to main()
406 SERVER* cmdline(int argc, char *argv[]) {
410 struct option long_options[] = {
411 {"read-only", no_argument, NULL, 'r'},
412 {"multi-file", no_argument, NULL, 'm'},
413 {"copy-on-write", no_argument, NULL, 'c'},
414 {"authorize-file", required_argument, NULL, 'l'},
415 {"config-file", required_argument, NULL, 'C'},
416 {"pid-file", required_argument, NULL, 'p'},
417 {"output-config", required_argument, NULL, 'o'},
424 gboolean do_output=FALSE;
425 gchar* section_header="";
431 serve=g_new0(SERVER, 1);
432 serve->authname = g_strdup(default_authname);
433 serve->virtstyle=VIRT_IPLIT;
434 while((c=getopt_long(argc, argv, "-C:cl:mo:rp:", long_options, &i))>=0) {
437 /* non-option argument */
438 switch(nonspecial++) {
440 if(strchr(optarg, ':') == strrchr(optarg, ':')) {
441 addr_port=g_strsplit(optarg, ":", 2);
443 /* Check for "@" - maybe user using this separator
446 g_strfreev(addr_port);
447 addr_port=g_strsplit(optarg, "@", 2);
450 addr_port=g_strsplit(optarg, "@", 2);
454 serve->port=strtol(addr_port[1], NULL, 0);
455 serve->listenaddr=g_strdup(addr_port[0]);
457 serve->listenaddr=NULL;
458 serve->port=strtol(addr_port[0], NULL, 0);
460 g_strfreev(addr_port);
463 serve->exportname = g_strdup(optarg);
464 if(serve->exportname[0] != '/') {
465 fprintf(stderr, "E: The to be exported file needs to be an absolute filename!\n");
470 last=strlen(optarg)-1;
472 if (suffix == 'k' || suffix == 'K' ||
473 suffix == 'm' || suffix == 'M')
475 es = (off_t)atoll(optarg);
483 serve->expected_size = es;
488 serve->flags |= F_READONLY;
491 serve->flags |= F_MULTIFILE;
495 section_header = g_strdup(optarg);
498 strncpy(pidftemplate, optarg, 256);
501 serve->flags |=F_COPYONWRITE;
504 g_free(config_file_pos);
505 config_file_pos=g_strdup(optarg);
508 g_free(serve->authname);
509 serve->authname=g_strdup(optarg);
517 /* What's left: the port to export, the name of the to be exported
518 * file, and, optionally, the size of the file, in that order. */
527 g_critical("Need a complete configuration on the command line to output a config file section!");
530 dump_section(serve, section_header);
536 * Error codes for config file parsing
539 CFILE_NOTFOUND, /**< The configuration file is not found */
540 CFILE_MISSING_GENERIC, /**< The (required) group "generic" is missing */
541 CFILE_KEY_MISSING, /**< A (required) key is missing */
542 CFILE_VALUE_INVALID, /**< A value is syntactically invalid */
543 CFILE_VALUE_UNSUPPORTED,/**< A value is not supported in this build */
544 CFILE_PROGERR, /**< Programmer error */
545 CFILE_NO_EXPORTS, /**< A config file was specified that does not
546 define any exports */
547 CFILE_INCORRECT_PORT, /**< The reserved port was specified for an
552 * Remove a SERVER from memory. Used from the hash table
554 void remove_server(gpointer s) {
558 g_free(server->exportname);
560 g_free(server->authname);
561 if(server->listenaddr)
562 g_free(server->listenaddr);
564 g_free(server->prerun);
566 g_free(server->postrun);
572 * @param s the old server we want to duplicate
573 * @return new duplicated server
575 SERVER* dup_serve(SERVER *s) {
576 SERVER *serve = NULL;
578 serve=g_new0(SERVER, 1);
583 serve->exportname = g_strdup(s->exportname);
585 serve->expected_size = s->expected_size;
588 serve->listenaddr = g_strdup(s->listenaddr);
590 serve->port = s->port;
593 serve->authname = strdup(s->authname);
595 serve->flags = s->flags;
596 serve->socket = serve->socket;
597 serve->socket_family = serve->socket_family;
598 serve->cidrlen = s->cidrlen;
601 serve->prerun = g_strdup(s->prerun);
604 serve->postrun = g_strdup(s->postrun);
607 serve->servename = g_strdup(s->servename);
613 * append new server to array
615 * @param a server array
616 * @return 0 success, -1 error
618 int append_serve(SERVER *s, GArray *a) {
620 struct addrinfo hints;
621 struct addrinfo *ai = NULL;
622 struct addrinfo *rp = NULL;
623 char host[NI_MAXHOST];
629 err("Invalid parsing server");
633 port = g_strdup_printf("%d", s->port);
635 memset(&hints,'\0',sizeof(hints));
636 hints.ai_family = AF_UNSPEC;
637 hints.ai_socktype = SOCK_STREAM;
638 hints.ai_flags = AI_ADDRCONFIG | AI_PASSIVE;
639 hints.ai_protocol = IPPROTO_TCP;
641 e = getaddrinfo(s->listenaddr, port, &hints, &ai);
647 for (rp = ai; rp != NULL; rp = rp->ai_next) {
648 e = getnameinfo(rp->ai_addr, rp->ai_addrlen, host, sizeof(host), NULL, 0, NI_NUMERICHOST);
650 if (e != 0) { // error
651 fprintf(stderr, "getnameinfo: %s\n", gai_strerror(e));
655 // duplicate server and set listenaddr to resolved IP address
658 ns->listenaddr = g_strdup(host);
659 ns->socket_family = rp->ai_family;
660 g_array_append_val(a, *ns);
668 fprintf(stderr, "getaddrinfo failed on listen host/address: %s (%s)\n", s->listenaddr ? s->listenaddr : "any", gai_strerror(e));
679 * Parse the config file.
681 * @param f the name of the config file
682 * @param e a GError. @see CFILE_ERRORS for what error values this function can
684 * @return a Array of SERVER* pointers, If the config file is empty or does not
685 * exist, returns an empty GHashTable; if the config file contains an
686 * error, returns NULL, and e is set appropriately
688 GArray* parse_cfile(gchar* f, GError** e) {
689 const char* DEFAULT_ERROR = "Could not parse %s in group %s: %s";
690 const char* MISSING_REQUIRED_ERROR = "Could not find required value %s in group %s: %s";
692 gchar *virtstyle=NULL;
694 { "exportname", TRUE, PARAM_STRING, NULL, 0 },
695 { "port", TRUE, PARAM_INT, NULL, 0 },
696 { "authfile", FALSE, PARAM_STRING, NULL, 0 },
697 { "filesize", FALSE, PARAM_INT, NULL, 0 },
698 { "virtstyle", FALSE, PARAM_STRING, NULL, 0 },
699 { "prerun", FALSE, PARAM_STRING, NULL, 0 },
700 { "postrun", FALSE, PARAM_STRING, NULL, 0 },
701 { "readonly", FALSE, PARAM_BOOL, NULL, F_READONLY },
702 { "multifile", FALSE, PARAM_BOOL, NULL, F_MULTIFILE },
703 { "copyonwrite", FALSE, PARAM_BOOL, NULL, F_COPYONWRITE },
704 { "sparse_cow", FALSE, PARAM_BOOL, NULL, F_SPARSE },
705 { "sdp", FALSE, PARAM_BOOL, NULL, F_SDP },
706 { "sync", FALSE, PARAM_BOOL, NULL, F_SYNC },
707 { "listenaddr", FALSE, PARAM_STRING, NULL, 0 },
709 const int lp_size=sizeof(lp)/sizeof(PARAM);
711 { "user", FALSE, PARAM_STRING, &runuser, 0 },
712 { "group", FALSE, PARAM_STRING, &rungroup, 0 },
713 { "oldstyle", FALSE, PARAM_BOOL, &do_oldstyle, 1 },
714 { "listenaddr", FALSE, PARAM_STRING, &modern_listen, 0 },
717 int p_size=sizeof(gp)/sizeof(PARAM);
720 const char *err_msg=NULL;
729 errdomain = g_quark_from_string("parse_cfile");
730 cfile = g_key_file_new();
731 retval = g_array_new(FALSE, TRUE, sizeof(SERVER));
732 if(!g_key_file_load_from_file(cfile, f, G_KEY_FILE_KEEP_COMMENTS |
733 G_KEY_FILE_KEEP_TRANSLATIONS, &err)) {
734 g_set_error(e, errdomain, CFILE_NOTFOUND, "Could not open config file %s.", f);
735 g_key_file_free(cfile);
738 startgroup = g_key_file_get_start_group(cfile);
739 if(!startgroup || strcmp(startgroup, "generic")) {
740 g_set_error(e, errdomain, CFILE_MISSING_GENERIC, "Config file does not contain the [generic] group!");
741 g_key_file_free(cfile);
744 groups = g_key_file_get_groups(cfile, NULL);
745 for(i=0;groups[i];i++) {
746 memset(&s, '\0', sizeof(SERVER));
747 lp[0].target=&(s.exportname);
748 lp[1].target=&(s.port);
749 lp[2].target=&(s.authname);
750 lp[3].target=&(s.expected_size);
751 lp[4].target=&(virtstyle);
752 lp[5].target=&(s.prerun);
753 lp[6].target=&(s.postrun);
754 lp[7].target=lp[8].target=lp[9].target=
755 lp[10].target=lp[11].target=
756 lp[12].target=&(s.flags);
757 lp[13].target=&(s.listenaddr);
759 /* After the [generic] group, start parsing exports */
764 for(j=0;j<p_size;j++) {
765 g_assert(p[j].target != NULL);
766 g_assert(p[j].ptype==PARAM_INT||p[j].ptype==PARAM_STRING||p[j].ptype==PARAM_BOOL);
769 *((gint*)p[j].target) =
770 g_key_file_get_integer(cfile,
776 *((gchar**)p[j].target) =
777 g_key_file_get_string(cfile,
783 value = g_key_file_get_boolean(cfile,
785 p[j].paramname, &err);
788 *((gint*)p[j].target) |= p[j].flagval;
790 *((gint*)p[j].target) &= ~(p[j].flagval);
795 if(!strcmp(p[j].paramname, "port") && !strcmp(p[j].target, NBD_DEFAULT_PORT)) {
796 g_set_error(e, errdomain, CFILE_INCORRECT_PORT, "Config file specifies default port for oldstyle export");
797 g_key_file_free(cfile);
801 if(err->code == G_KEY_FILE_ERROR_KEY_NOT_FOUND) {
803 /* Ignore not-found error for optional values */
807 err_msg = MISSING_REQUIRED_ERROR;
810 err_msg = DEFAULT_ERROR;
812 g_set_error(e, errdomain, CFILE_VALUE_INVALID, err_msg, p[j].paramname, groups[i], err->message);
813 g_array_free(retval, TRUE);
815 g_key_file_free(cfile);
820 if(!strncmp(virtstyle, "none", 4)) {
821 s.virtstyle=VIRT_NONE;
822 } else if(!strncmp(virtstyle, "ipliteral", 9)) {
823 s.virtstyle=VIRT_IPLIT;
824 } else if(!strncmp(virtstyle, "iphash", 6)) {
825 s.virtstyle=VIRT_IPHASH;
826 } else if(!strncmp(virtstyle, "cidrhash", 8)) {
827 s.virtstyle=VIRT_CIDR;
828 if(strlen(virtstyle)<10) {
829 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s: missing length", virtstyle, groups[i]);
830 g_array_free(retval, TRUE);
831 g_key_file_free(cfile);
834 s.cidrlen=strtol(virtstyle+8, NULL, 0);
836 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s", virtstyle, groups[i]);
837 g_array_free(retval, TRUE);
838 g_key_file_free(cfile);
841 if(s.port && !do_oldstyle) {
842 g_warning("A port was specified, but oldstyle exports were not requested. This may not do what you expect.");
843 g_warning("Please read 'man 5 nbd-server' and search for oldstyle for more info");
846 s.virtstyle=VIRT_IPLIT;
848 /* Don't need to free this, it's not our string */
850 /* Don't append values for the [generic] group */
852 s.socket_family = AF_UNSPEC;
853 s.servename = groups[i];
855 append_serve(&s, retval);
862 if(s.flags & F_SDP) {
863 g_set_error(e, errdomain, CFILE_VALUE_UNSUPPORTED, "This nbd-server was built without support for SDP, yet group %s uses it", groups[i]);
864 g_array_free(retval, TRUE);
865 g_key_file_free(cfile);
871 g_set_error(e, errdomain, CFILE_NO_EXPORTS, "The config file does not specify any exports");
873 g_key_file_free(cfile);
878 * Signal handler for SIGCHLD
879 * @param s the signal we're handling (must be SIGCHLD, or something
882 void sigchld_handler(int s) {
887 while((pid=waitpid(-1, &status, WNOHANG)) > 0) {
888 if(WIFEXITED(status)) {
889 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
891 i=g_hash_table_lookup(children, &pid);
893 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
895 DEBUG2("Removing %d from the list of children", pid);
896 g_hash_table_remove(children, &pid);
902 * Kill a child. Called from sigterm_handler::g_hash_table_foreach.
905 * @param value the value corresponding to the above key
906 * @param user_data a pointer which we always set to 1, so that we know what
909 void killchild(gpointer key, gpointer value, gpointer user_data) {
911 int *parent=user_data;
918 * Handle SIGTERM and dispatch it to our children
919 * @param s the signal we're handling (must be SIGTERM, or something
920 * is severely wrong).
922 void sigterm_handler(int s) {
925 g_hash_table_foreach(children, killchild, &parent);
935 * Detect the size of a file.
937 * @param fhandle An open filedescriptor
938 * @return the size of the file, or OFFT_MAX if detection was
941 off_t size_autodetect(int fhandle) {
944 struct stat stat_buf;
947 #ifdef HAVE_SYS_MOUNT_H
948 #ifdef HAVE_SYS_IOCTL_H
950 DEBUG("looking for export size with ioctl BLKGETSIZE64\n");
951 if (!ioctl(fhandle, BLKGETSIZE64, &bytes) && bytes) {
954 #endif /* BLKGETSIZE64 */
955 #endif /* HAVE_SYS_IOCTL_H */
956 #endif /* HAVE_SYS_MOUNT_H */
958 DEBUG("looking for fhandle size with fstat\n");
959 stat_buf.st_size = 0;
960 error = fstat(fhandle, &stat_buf);
962 if(stat_buf.st_size > 0)
963 return (off_t)stat_buf.st_size;
965 err("fstat failed: %m");
968 DEBUG("looking for fhandle size with lseek SEEK_END\n");
969 es = lseek(fhandle, (off_t)0, SEEK_END);
970 if (es > ((off_t)0)) {
973 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
976 err("Could not find size of exported block device: %m");
981 * Get the file handle and offset, given an export offset.
983 * @param export An array of export files
984 * @param a The offset to get corresponding file/offset for
985 * @param fhandle [out] File descriptor
986 * @param foffset [out] Offset into fhandle
987 * @param maxbytes [out] Tells how many bytes can be read/written
988 * from fhandle starting at foffset (0 if there is no limit)
989 * @return 0 on success, -1 on failure
991 int get_filepos(GArray* export, off_t a, int* fhandle, off_t* foffset, size_t* maxbytes ) {
992 /* Negative offset not allowed */
996 /* Binary search for last file with starting offset <= a */
999 int end = export->len - 1;
1000 while( start <= end ) {
1001 int mid = (start + end) / 2;
1002 fi = g_array_index(export, FILE_INFO, mid);
1003 if( fi.startoff < a ) {
1005 } else if( fi.startoff > a ) {
1013 /* end should never go negative, since first startoff is 0 and a >= 0 */
1016 fi = g_array_index(export, FILE_INFO, end);
1017 *fhandle = fi.fhandle;
1018 *foffset = a - fi.startoff;
1020 if( end+1 < export->len ) {
1021 FILE_INFO fi_next = g_array_index(export, FILE_INFO, end+1);
1022 *maxbytes = fi_next.startoff - a;
1029 * seek to a position in a file, with error handling.
1030 * @param handle a filedescriptor
1031 * @param a position to seek to
1032 * @todo get rid of this; lastpoint is a global variable right now, but it
1033 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
1036 void myseek(int handle,off_t a) {
1037 if (lseek(handle, a, SEEK_SET) < 0) {
1038 err("Can not seek locally!\n");
1043 * Write an amount of bytes at a given offset to the right file. This
1044 * abstracts the write-side of the multiple file option.
1046 * @param a The offset where the write should start
1047 * @param buf The buffer to write from
1048 * @param len The length of buf
1049 * @param client The client we're serving for
1050 * @return The number of bytes actually written, or -1 in case of an error
1052 ssize_t rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
1058 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
1060 if(maxbytes && len > maxbytes)
1063 DEBUG4("(WRITE to fd %d offset %llu len %u), ", fhandle, foffset, len);
1065 myseek(fhandle, foffset);
1066 retval = write(fhandle, buf, len);
1067 if(client->server->flags & F_SYNC) {
1074 * Call rawexpwrite repeatedly until all data has been written.
1075 * @return 0 on success, nonzero on failure
1077 int rawexpwrite_fully(off_t a, char *buf, size_t len, CLIENT *client) {
1080 while(len > 0 && (ret=rawexpwrite(a, buf, len, client)) > 0 ) {
1085 return (ret < 0 || len != 0);
1089 * Read an amount of bytes at a given offset from the right file. This
1090 * abstracts the read-side of the multiple files option.
1092 * @param a The offset where the read should start
1093 * @param buf A buffer to read into
1094 * @param len The size of buf
1095 * @param client The client we're serving for
1096 * @return The number of bytes actually read, or -1 in case of an
1099 ssize_t rawexpread(off_t a, char *buf, size_t len, CLIENT *client) {
1104 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
1106 if(maxbytes && len > maxbytes)
1109 DEBUG4("(READ from fd %d offset %llu len %u), ", fhandle, foffset, len);
1111 myseek(fhandle, foffset);
1112 return read(fhandle, buf, len);
1116 * Call rawexpread repeatedly until all data has been read.
1117 * @return 0 on success, nonzero on failure
1119 int rawexpread_fully(off_t a, char *buf, size_t len, CLIENT *client) {
1122 while(len > 0 && (ret=rawexpread(a, buf, len, client)) > 0 ) {
1127 return (ret < 0 || len != 0);
1131 * Read an amount of bytes at a given offset from the right file. This
1132 * abstracts the read-side of the copyonwrite stuff, and calls
1133 * rawexpread() with the right parameters to do the actual work.
1134 * @param a The offset where the read should start
1135 * @param buf A buffer to read into
1136 * @param len The size of buf
1137 * @param client The client we're going to read for
1138 * @return 0 on success, nonzero on failure
1140 int expread(off_t a, char *buf, size_t len, CLIENT *client) {
1141 off_t rdlen, offset;
1142 off_t mapcnt, mapl, maph, pagestart;
1144 if (!(client->server->flags & F_COPYONWRITE))
1145 return(rawexpread_fully(a, buf, len, client));
1146 DEBUG3("Asked to read %d bytes at %llu.\n", len, (unsigned long long)a);
1148 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
1150 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
1151 pagestart=mapcnt*DIFFPAGESIZE;
1153 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
1154 len : (size_t)DIFFPAGESIZE-offset;
1155 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
1156 DEBUG3("Page %llu is at %lu\n", (unsigned long long)mapcnt,
1157 (unsigned long)(client->difmap[mapcnt]));
1158 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
1159 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
1160 } else { /* the block is not there */
1161 DEBUG2("Page %llu is not here, we read the original one\n",
1162 (unsigned long long)mapcnt);
1163 if(rawexpread_fully(a, buf, rdlen, client)) return -1;
1165 len-=rdlen; a+=rdlen; buf+=rdlen;
1171 * Write an amount of bytes at a given offset to the right file. This
1172 * abstracts the write-side of the copyonwrite option, and calls
1173 * rawexpwrite() with the right parameters to do the actual work.
1175 * @param a The offset where the write should start
1176 * @param buf The buffer to write from
1177 * @param len The length of buf
1178 * @param client The client we're going to write for.
1179 * @return 0 on success, nonzero on failure
1181 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
1182 char pagebuf[DIFFPAGESIZE];
1183 off_t mapcnt,mapl,maph;
1188 if (!(client->server->flags & F_COPYONWRITE))
1189 return(rawexpwrite_fully(a, buf, len, client));
1190 DEBUG3("Asked to write %d bytes at %llu.\n", len, (unsigned long long)a);
1192 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
1194 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
1195 pagestart=mapcnt*DIFFPAGESIZE ;
1196 offset=a-pagestart ;
1197 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
1198 len : (size_t)DIFFPAGESIZE-offset;
1200 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
1201 DEBUG3("Page %llu is at %lu\n", (unsigned long long)mapcnt,
1202 (unsigned long)(client->difmap[mapcnt])) ;
1203 myseek(client->difffile,
1204 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
1205 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
1206 } else { /* the block is not there */
1207 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
1208 client->difmap[mapcnt]=(client->server->flags&F_SPARSE)?mapcnt:client->difffilelen++;
1209 DEBUG3("Page %llu is not here, we put it at %lu\n",
1210 (unsigned long long)mapcnt,
1211 (unsigned long)(client->difmap[mapcnt]));
1212 rdlen=DIFFPAGESIZE ;
1213 if (rawexpread_fully(pagestart, pagebuf, rdlen, client))
1215 memcpy(pagebuf+offset,buf,wrlen) ;
1216 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
1220 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
1226 * Do the initial negotiation.
1228 * @param client The client we're negotiating with.
1230 CLIENT* negotiate(int net, CLIENT *client, GArray* servers) {
1233 uint32_t flags = NBD_FLAG_HAS_FLAGS;
1234 uint16_t smallflags = 0;
1237 memset(zeros, '\0', sizeof(zeros));
1238 if(!client || !client->modern) {
1240 if (write(net, INIT_PASSWD, 8) < 0) {
1241 err_nonfatal("Negotiation failed: %m");
1245 if(!client || client->modern) {
1247 magic = htonll(opts_magic);
1250 magic = htonll(cliserv_magic);
1252 if (write(net, &magic, sizeof(magic)) < 0) {
1253 err_nonfatal("Negotiation failed: %m");
1267 err("programmer error");
1268 if (write(net, &smallflags, sizeof(uint16_t)) < 0)
1269 err("Negotiation failed: %m");
1270 if (read(net, &reserved, sizeof(reserved)) < 0)
1271 err("Negotiation failed: %m");
1272 if (read(net, &magic, sizeof(magic)) < 0)
1273 err("Negotiation failed: %m");
1274 magic = ntohll(magic);
1275 if(magic != opts_magic) {
1279 if (read(net, &opt, sizeof(opt)) < 0)
1280 err("Negotiation failed: %m");
1282 if(opt != NBD_OPT_EXPORT_NAME) {
1286 if (read(net, &namelen, sizeof(namelen)) < 0)
1287 err("Negotiation failed: %m");
1288 namelen = ntohl(namelen);
1289 name = malloc(namelen+1);
1291 if (read(net, name, namelen) < 0)
1292 err("Negotiation failed: %m");
1293 for(i=0; i<servers->len; i++) {
1294 SERVER* serve = &(g_array_index(servers, SERVER, i));
1295 if(!strcmp(serve->servename, name)) {
1296 CLIENT* client = g_new0(CLIENT, 1);
1297 client->server = serve;
1298 client->exportsize = OFFT_MAX;
1300 client->modern = TRUE;
1307 size_host = htonll((u64)(client->exportsize));
1308 if (write(net, &size_host, 8) < 0)
1309 err("Negotiation failed: %m");
1310 if (client->server->flags & F_READONLY)
1311 flags |= NBD_FLAG_READ_ONLY;
1312 if (!client->modern) {
1314 flags = htonl(flags);
1315 if (write(client->net, &flags, 4) < 0)
1316 err("Negotiation failed: %m");
1319 smallflags = (uint16_t)(flags & ~((uint16_t)0));
1320 smallflags = htons(smallflags);
1321 if (write(client->net, &smallflags, sizeof(smallflags)) < 0) {
1322 err("Negotiation failed: %m");
1326 if (write(client->net, zeros, 124) < 0)
1327 err("Negotiation failed: %m");
1331 /** sending macro. */
1332 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
1334 #define ERROR(client,reply,errcode) { reply.error = htonl(errcode); SEND(client->net,reply); reply.error = 0; }
1336 * Serve a file to a single client.
1338 * @todo This beast needs to be split up in many tiny little manageable
1339 * pieces. Preferably with a chainsaw.
1341 * @param client The client we're going to serve to.
1342 * @return when the client disconnects
1344 int mainloop(CLIENT *client) {
1345 struct nbd_request request;
1346 struct nbd_reply reply;
1347 gboolean go_on=TRUE;
1351 negotiate(client->net, client, NULL);
1352 DEBUG("Entering request loop!\n");
1353 reply.magic = htonl(NBD_REPLY_MAGIC);
1362 readit(client->net, &request, sizeof(request));
1363 request.from = ntohll(request.from);
1364 request.type = ntohl(request.type);
1366 if (request.type==NBD_CMD_DISC) {
1367 msg2(LOG_INFO, "Disconnect request received.");
1368 if (client->server->flags & F_COPYONWRITE) {
1369 if (client->difmap) g_free(client->difmap) ;
1370 close(client->difffile);
1371 unlink(client->difffilename);
1372 free(client->difffilename);
1378 len = ntohl(request.len);
1380 if (request.magic != htonl(NBD_REQUEST_MAGIC))
1381 err("Not enough magic.");
1382 if (len > BUFSIZE + sizeof(struct nbd_reply))
1383 err("Request too big!");
1385 printf("%s from %llu (%llu) len %d, ", request.type ? "WRITE" :
1386 "READ", (unsigned long long)request.from,
1387 (unsigned long long)request.from / 512, len);
1389 memcpy(reply.handle, request.handle, sizeof(reply.handle));
1390 if ((request.from + len) > (OFFT_MAX)) {
1391 DEBUG("[Number too large!]");
1392 ERROR(client, reply, EINVAL);
1396 if (((ssize_t)((off_t)request.from + len) > client->exportsize)) {
1398 ERROR(client, reply, EINVAL);
1402 if (request.type==NBD_CMD_WRITE) {
1403 DEBUG("wr: net->buf, ");
1404 readit(client->net, buf, len);
1405 DEBUG("buf->exp, ");
1406 if ((client->server->flags & F_READONLY) ||
1407 (client->server->flags & F_AUTOREADONLY)) {
1408 DEBUG("[WRITE to READONLY!]");
1409 ERROR(client, reply, EPERM);
1412 if (expwrite(request.from, buf, len, client)) {
1413 DEBUG("Write failed: %m" );
1414 ERROR(client, reply, errno);
1417 SEND(client->net, reply);
1423 DEBUG("exp->buf, ");
1424 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
1425 DEBUG("Read failed: %m");
1426 ERROR(client, reply, errno);
1430 DEBUG("buf->net, ");
1431 memcpy(buf, &reply, sizeof(struct nbd_reply));
1432 writeit(client->net, buf, len + sizeof(struct nbd_reply));
1439 * Set up client export array, which is an array of FILE_INFO.
1440 * Also, split a single exportfile into multiple ones, if that was asked.
1441 * @param client information on the client which we want to setup export for
1443 void setupexport(CLIENT* client) {
1445 off_t laststartoff = 0, lastsize = 0;
1446 int multifile = (client->server->flags & F_MULTIFILE);
1448 client->export = g_array_new(TRUE, TRUE, sizeof(FILE_INFO));
1450 /* If multi-file, open as many files as we can.
1451 * If not, open exactly one file.
1452 * Calculate file sizes as we go to get total size. */
1456 gchar* error_string;
1457 mode_t mode = (client->server->flags & F_READONLY) ? O_RDONLY : O_RDWR;
1460 tmpname=g_strdup_printf("%s.%d", client->exportname, i);
1462 tmpname=g_strdup(client->exportname);
1464 DEBUG2( "Opening %s\n", tmpname );
1465 fi.fhandle = open(tmpname, mode);
1466 if(fi.fhandle == -1 && mode == O_RDWR) {
1467 /* Try again because maybe media was read-only */
1468 fi.fhandle = open(tmpname, O_RDONLY);
1469 if(fi.fhandle != -1) {
1470 /* Opening the base file in copyonwrite mode is
1472 if(!(client->server->flags & F_COPYONWRITE)) {
1473 client->server->flags |= F_AUTOREADONLY;
1474 client->server->flags |= F_READONLY;
1478 if(fi.fhandle == -1) {
1479 if(multifile && i>0)
1481 error_string=g_strdup_printf(
1482 "Could not open exported file %s: %%m",
1486 fi.startoff = laststartoff + lastsize;
1487 g_array_append_val(client->export, fi);
1490 /* Starting offset and size of this file will be used to
1491 * calculate starting offset of next file */
1492 laststartoff = fi.startoff;
1493 lastsize = size_autodetect(fi.fhandle);
1499 /* Set export size to total calculated size */
1500 client->exportsize = laststartoff + lastsize;
1502 /* Export size may be overridden */
1503 if(client->server->expected_size) {
1504 /* desired size must be <= total calculated size */
1505 if(client->server->expected_size > client->exportsize) {
1506 err("Size of exported file is too big\n");
1509 client->exportsize = client->server->expected_size;
1512 msg3(LOG_INFO, "Size of exported file/device is %llu", (unsigned long long)client->exportsize);
1514 msg3(LOG_INFO, "Total number of files: %d", i);
1518 int copyonwrite_prepare(CLIENT* client) {
1520 if ((client->difffilename = malloc(1024))==NULL)
1521 err("Failed to allocate string for diff file name");
1522 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
1524 client->difffilename[1023]='\0';
1525 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
1526 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
1527 if (client->difffile<0) err("Could not create diff file (%m)") ;
1528 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
1529 err("Could not allocate memory") ;
1530 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
1536 * Run a command. This is used for the ``prerun'' and ``postrun'' config file
1539 * @param command the command to be ran. Read from the config file
1540 * @param file the file name we're about to export
1542 int do_run(gchar* command, gchar* file) {
1546 if(command && *command) {
1547 cmd = g_strdup_printf(command, file);
1555 * Serve a connection.
1557 * @todo allow for multithreading, perhaps use libevent. Not just yet, though;
1558 * follow the road map.
1560 * @param client a connected client
1562 void serveconnection(CLIENT *client) {
1563 if(do_run(client->server->prerun, client->exportname)) {
1566 setupexport(client);
1568 if (client->server->flags & F_COPYONWRITE) {
1569 copyonwrite_prepare(client);
1572 setmysockopt(client->net);
1575 do_run(client->server->postrun, client->exportname);
1579 * Find the name of the file we have to serve. This will use g_strdup_printf
1580 * to put the IP address of the client inside a filename containing
1581 * "%s" (in the form as specified by the "virtstyle" option). That name
1582 * is then written to client->exportname.
1584 * @param net A socket connected to an nbd client
1585 * @param client information about the client. The IP address in human-readable
1586 * format will be written to a new char* buffer, the address of which will be
1587 * stored in client->clientname.
1589 void set_peername(int net, CLIENT *client) {
1590 struct sockaddr_storage addrin;
1591 struct sockaddr_storage netaddr;
1592 struct sockaddr_in *netaddr4 = NULL;
1593 struct sockaddr_in6 *netaddr6 = NULL;
1594 size_t addrinlen = sizeof( addrin );
1595 struct addrinfo hints;
1596 struct addrinfo *ai = NULL;
1597 char peername[NI_MAXHOST];
1598 char netname[NI_MAXHOST];
1604 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
1605 err("getsockname failed: %m");
1607 getnameinfo((struct sockaddr *)&addrin, (socklen_t)addrinlen,
1608 peername, sizeof (peername), NULL, 0, NI_NUMERICHOST);
1610 memset(&hints, '\0', sizeof (hints));
1611 hints.ai_flags = AI_ADDRCONFIG;
1612 e = getaddrinfo(peername, NULL, &hints, &ai);
1615 fprintf(stderr, "getaddrinfo failed: %s\n", gai_strerror(e));
1620 switch(client->server->virtstyle) {
1622 client->exportname=g_strdup(client->server->exportname);
1625 for(i=0;i<strlen(peername);i++) {
1626 if(peername[i]=='.') {
1631 client->exportname=g_strdup_printf(client->server->exportname, peername);
1634 memcpy(&netaddr, &addrin, addrinlen);
1635 if(ai->ai_family == AF_INET) {
1636 netaddr4 = (struct sockaddr_in *)&netaddr;
1637 (netaddr4->sin_addr).s_addr>>=32-(client->server->cidrlen);
1638 (netaddr4->sin_addr).s_addr<<=32-(client->server->cidrlen);
1640 getnameinfo((struct sockaddr *) netaddr4, (socklen_t) addrinlen,
1641 netname, sizeof (netname), NULL, 0, NI_NUMERICHOST);
1642 tmp=g_strdup_printf("%s/%s", netname, peername);
1643 }else if(ai->ai_family == AF_INET6) {
1644 netaddr6 = (struct sockaddr_in6 *)&netaddr;
1646 shift = 128-(client->server->cidrlen);
1648 while(shift >= 32) {
1649 ((netaddr6->sin6_addr).s6_addr32[i])=0;
1653 (netaddr6->sin6_addr).s6_addr32[i]>>=shift;
1654 (netaddr6->sin6_addr).s6_addr32[i]<<=shift;
1656 getnameinfo((struct sockaddr *)netaddr6, (socklen_t)addrinlen,
1657 netname, sizeof(netname), NULL, 0, NI_NUMERICHOST);
1658 tmp=g_strdup_printf("%s/%s", netname, peername);
1662 client->exportname=g_strdup_printf(client->server->exportname, tmp);
1668 msg4(LOG_INFO, "connect from %s, assigned file is %s",
1669 peername, client->exportname);
1670 client->clientname=g_strdup(peername);
1675 * @param data a pointer to pid_t which should be freed
1677 void destroy_pid_t(gpointer data) {
1682 * Loop through the available servers, and serve them. Never returns.
1684 int serveloop(GArray* servers) {
1685 struct sockaddr_storage addrin;
1686 socklen_t addrinlen=sizeof(addrin);
1694 * Set up the master fd_set. The set of descriptors we need
1695 * to select() for never changes anyway and it buys us a *lot*
1696 * of time to only build this once. However, if we ever choose
1697 * to not fork() for clients anymore, we may have to revisit
1702 for(i=0;i<servers->len;i++) {
1703 if((sock=(g_array_index(servers, SERVER, i)).socket)) {
1704 FD_SET(sock, &mset);
1705 max=sock>max?sock:max;
1709 FD_SET(modernsock, &mset);
1710 max=modernsock>max?modernsock:max;
1713 CLIENT *client = NULL;
1716 memcpy(&rset, &mset, sizeof(fd_set));
1717 if(select(max+1, &rset, NULL, NULL, NULL)>0) {
1722 if(FD_ISSET(modernsock, &rset)) {
1723 if((net=accept(modernsock, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1725 client = negotiate(net, NULL, servers);
1727 err_nonfatal("negotiation failed");
1732 for(i=0;i<servers->len && !net;i++) {
1733 serve=&(g_array_index(servers, SERVER, i));
1734 if(FD_ISSET(serve->socket, &rset)) {
1735 if ((net=accept(serve->socket, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1742 if((sock_flags = fcntl(net, F_GETFL, 0))==-1) {
1743 err("fcntl F_GETFL");
1745 if(fcntl(net, F_SETFL, sock_flags &~O_NONBLOCK)==-1) {
1746 err("fcntl F_SETFL ~O_NONBLOCK");
1749 client = g_new0(CLIENT, 1);
1750 client->server=serve;
1751 client->exportsize=OFFT_MAX;
1754 set_peername(net, client);
1755 if (!authorized_client(client)) {
1756 msg2(LOG_INFO,"Unauthorized client") ;
1760 msg2(LOG_INFO,"Authorized client") ;
1761 pid=g_malloc(sizeof(pid_t));
1763 if ((*pid=fork())<0) {
1764 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
1768 if (*pid>0) { /* parent */
1770 g_hash_table_insert(children, pid, pid);
1774 g_hash_table_destroy(children);
1775 for(i=0;i<servers->len;i++) {
1776 serve=&g_array_index(servers, SERVER, i);
1777 close(serve->socket);
1779 /* FALSE does not free the
1780 actual data. This is required,
1781 because the client has a
1782 direct reference into that
1783 data, and otherwise we get a
1785 g_array_free(servers, FALSE);
1787 msg2(LOG_INFO,"Starting to serve");
1788 serveconnection(client);
1795 void dosockopts(int socket) {
1803 /* lose the pesky "Address already in use" error message */
1804 if (setsockopt(socket,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
1805 err("setsockopt SO_REUSEADDR");
1807 if (setsockopt(socket,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
1808 err("setsockopt SO_KEEPALIVE");
1811 /* make the listening socket non-blocking */
1812 if ((sock_flags = fcntl(socket, F_GETFL, 0)) == -1) {
1813 err("fcntl F_GETFL");
1815 if (fcntl(socket, F_SETFL, sock_flags | O_NONBLOCK) == -1) {
1816 err("fcntl F_SETFL O_NONBLOCK");
1821 * Connect a server's socket.
1823 * @param serve the server we want to connect.
1825 int setup_serve(SERVER *serve) {
1826 struct addrinfo hints;
1827 struct addrinfo *ai = NULL;
1832 return serve->servename ? 1 : 0;
1834 memset(&hints,'\0',sizeof(hints));
1835 hints.ai_flags = AI_PASSIVE | AI_ADDRCONFIG | AI_NUMERICSERV;
1836 hints.ai_socktype = SOCK_STREAM;
1837 hints.ai_family = serve->socket_family;
1839 port = g_strdup_printf ("%d", serve->port);
1843 e = getaddrinfo(serve->listenaddr,port,&hints,&ai);
1848 fprintf(stderr, "getaddrinfo failed: %s\n", gai_strerror(e));
1854 if(serve->socket_family == AF_UNSPEC)
1855 serve->socket_family = ai->ai_family;
1858 if ((serve->flags) && F_SDP) {
1859 if (ai->ai_family == AF_INET)
1860 ai->ai_family = AF_INET_SDP;
1861 else (ai->ai_family == AF_INET6)
1862 ai->ai_family = AF_INET6_SDP;
1865 if ((serve->socket = socket(ai->ai_family, ai->ai_socktype, ai->ai_protocol)) < 0)
1868 dosockopts(serve->socket);
1870 DEBUG("Waiting for connections... bind, ");
1871 e = bind(serve->socket, ai->ai_addr, ai->ai_addrlen);
1872 if (e != 0 && errno != EADDRINUSE)
1875 if (listen(serve->socket, 1) < 0)
1879 if(serve->servename) {
1886 void open_modern(void) {
1887 struct addrinfo hints;
1888 struct addrinfo* ai = NULL;
1892 memset(&hints, '\0', sizeof(hints));
1893 hints.ai_flags = AI_PASSIVE | AI_ADDRCONFIG;
1894 hints.ai_socktype = SOCK_STREAM;
1895 hints.ai_family = AF_UNSPEC;
1896 hints.ai_protocol = IPPROTO_TCP;
1897 e = getaddrinfo(modern_listen, NBD_DEFAULT_PORT, &hints, &ai);
1899 fprintf(stderr, "getaddrinfo failed: %s\n", gai_strerror(e));
1902 if((modernsock = socket(ai->ai_family, ai->ai_socktype, ai->ai_protocol))<0) {
1906 dosockopts(modernsock);
1908 if(bind(modernsock, ai->ai_addr, ai->ai_addrlen)) {
1911 if(listen(modernsock, 10) <0) {
1919 * Connect our servers.
1921 void setup_servers(GArray* servers) {
1923 struct sigaction sa;
1926 for(i=0;i<servers->len;i++) {
1927 want_modern |= setup_serve(&(g_array_index(servers, SERVER, i)));
1932 children=g_hash_table_new_full(g_int_hash, g_int_equal, NULL, destroy_pid_t);
1934 sa.sa_handler = sigchld_handler;
1935 sigemptyset(&sa.sa_mask);
1936 sa.sa_flags = SA_RESTART;
1937 if(sigaction(SIGCHLD, &sa, NULL) == -1)
1938 err("sigaction: %m");
1939 sa.sa_handler = sigterm_handler;
1940 sigemptyset(&sa.sa_mask);
1941 sa.sa_flags = SA_RESTART;
1942 if(sigaction(SIGTERM, &sa, NULL) == -1)
1943 err("sigaction: %m");
1947 * Go daemon (unless we specified at compile time that we didn't want this)
1948 * @param serve the first server of our configuration. If its port is zero,
1949 * then do not daemonize, because we're doing inetd then. This parameter
1950 * is only used to create a PID file of the form
1951 * /var/run/nbd-server.<port>.pid; it's not modified in any way.
1953 #if !defined(NODAEMON) && !defined(NOFORK)
1954 void daemonize(SERVER* serve) {
1957 if(serve && !(serve->port)) {
1963 if(!*pidftemplate) {
1965 strncpy(pidftemplate, "/var/run/nbd-server.%d.pid", 255);
1967 strncpy(pidftemplate, "/var/run/nbd-server.pid", 255);
1970 snprintf(pidfname, 255, pidftemplate, serve ? serve->port : 0);
1971 pidf=fopen(pidfname, "w");
1973 fprintf(pidf,"%d\n", (int)getpid());
1977 fprintf(stderr, "Not fatal; continuing");
1981 #define daemonize(serve)
1982 #endif /* !defined(NODAEMON) && !defined(NOFORK) */
1985 * Everything beyond this point (in the file) is run in non-daemon mode.
1986 * The stuff above daemonize() isn't.
1989 void serve_err(SERVER* serve, const char* msg) G_GNUC_NORETURN;
1991 void serve_err(SERVER* serve, const char* msg) {
1992 g_message("Export of %s on port %d failed:", serve->exportname,
1998 * Set up user-ID and/or group-ID
2000 void dousers(void) {
2005 gr=getgrnam(rungroup);
2007 str = g_strdup_printf("Invalid group name: %s", rungroup);
2010 if(setgid(gr->gr_gid)<0) {
2011 err("Could not set GID: %m");
2015 pw=getpwnam(runuser);
2017 str = g_strdup_printf("Invalid user name: %s", runuser);
2020 if(setuid(pw->pw_uid)<0) {
2021 err("Could not set UID: %m");
2027 void glib_message_syslog_redirect(const gchar *log_domain,
2028 GLogLevelFlags log_level,
2029 const gchar *message,
2032 int level=LOG_DEBUG;
2036 case G_LOG_FLAG_FATAL:
2037 case G_LOG_LEVEL_CRITICAL:
2038 case G_LOG_LEVEL_ERROR:
2041 case G_LOG_LEVEL_WARNING:
2044 case G_LOG_LEVEL_MESSAGE:
2045 case G_LOG_LEVEL_INFO:
2048 case G_LOG_LEVEL_DEBUG:
2053 syslog(level, "%s", message);
2058 * Main entry point...
2060 int main(int argc, char *argv[]) {
2065 if (sizeof( struct nbd_request )!=28) {
2066 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
2067 exit(EXIT_FAILURE) ;
2070 memset(pidftemplate, '\0', 256);
2073 config_file_pos = g_strdup(CFILE);
2074 serve=cmdline(argc, argv);
2075 servers = parse_cfile(config_file_pos, &err);
2078 serve->socket_family = AF_UNSPEC;
2080 append_serve(serve, servers);
2082 if (!(serve->port)) {
2085 /* You really should define ISSERVER if you're going to use
2086 * inetd mode, but if you don't, closing stdout and stderr
2087 * (which inetd had connected to the client socket) will let it
2091 open("/dev/null", O_WRONLY);
2092 open("/dev/null", O_WRONLY);
2093 g_log_set_default_handler( glib_message_syslog_redirect, NULL );
2095 client=g_malloc(sizeof(CLIENT));
2096 client->server=serve;
2098 client->exportsize=OFFT_MAX;
2099 set_peername(0,client);
2100 serveconnection(client);
2105 if(!servers || !servers->len) {
2106 if(err && !(err->domain == g_quark_from_string("parse_cfile")
2107 && err->code == CFILE_NOTFOUND)) {
2108 g_warning("Could not parse config file: %s",
2109 err ? err->message : "Unknown error");
2113 g_warning("Specifying an export on the command line is deprecated.");
2114 g_warning("Please use a configuration file instead.");
2117 if((!serve) && (!servers||!servers->len)) {
2118 g_message("No configured exports; quitting.");
2122 setup_servers(servers);