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
54 /* Includes LFS defines, which defines behaviours of some of the following
55 * headers, so must come before those */
58 #include <sys/types.h>
59 #include <sys/socket.h>
61 #include <sys/select.h> /* select */
62 #include <sys/wait.h> /* wait */
63 #ifdef HAVE_SYS_IOCTL_H
64 #include <sys/ioctl.h>
66 #include <sys/param.h>
67 #ifdef HAVE_SYS_MOUNT_H
68 #include <sys/mount.h> /* For BLKGETSIZE */
70 #include <signal.h> /* sigaction */
71 #include <netinet/tcp.h>
72 #include <netinet/in.h> /* sockaddr_in, htons, in_addr */
73 #include <netdb.h> /* hostent, gethostby*, getservby* */
80 #include <arpa/inet.h>
88 /* used in cliserv.h, so must come first */
89 #define MY_NAME "nbd_server"
92 /** Default position of the config file */
94 #define SYSCONFDIR "/etc"
96 #define CFILE SYSCONFDIR "/nbd-server/config"
98 /** Where our config file actually is */
99 gchar* config_file_pos;
101 /** Logging macros, now nothing goes to syslog unless you say ISSERVER */
103 #define msg2(a,b) syslog(a,b)
104 #define msg3(a,b,c) syslog(a,b,c)
105 #define msg4(a,b,c,d) syslog(a,b,c,d)
107 #define msg2(a,b) g_message(b)
108 #define msg3(a,b,c) g_message(b,c)
109 #define msg4(a,b,c,d) g_message(b,c,d)
112 /* Debugging macros */
115 #define DEBUG( a ) printf( a )
116 #define DEBUG2( a,b ) printf( a,b )
117 #define DEBUG3( a,b,c ) printf( a,b,c )
120 #define DEBUG2( a,b )
121 #define DEBUG3( a,b,c )
123 #ifndef PACKAGE_VERSION
124 #define PACKAGE_VERSION ""
127 * The highest value a variable of type off_t can reach. This is a signed
128 * integer, so set all bits except for the leftmost one.
130 #define OFFT_MAX ~((off_t)1<<(sizeof(off_t)*8-1))
131 #define LINELEN 256 /**< Size of static buffer used to read the
132 authorization file (yuck) */
133 #define BUFSIZE (1024*1024) /**< Size of buffer that can hold requests */
134 #define GIGA (1*1024*1024*1024) /**< 1 Gigabyte. Used as hunksize when doing
135 the multiple file thingy. @todo: make this a
136 configuration option. */
137 #define DIFFPAGESIZE 4096 /**< diff file uses those chunks */
138 #define F_READONLY 1 /**< flag to tell us a file is readonly */
139 #define F_MULTIFILE 2 /**< flag to tell us a file is exported using -m */
140 #define F_COPYONWRITE 4 /**< flag to tell us a file is exported using
142 #define F_AUTOREADONLY 8 /**< flag to tell us a file is set to autoreadonly */
143 GHashTable *children;
144 char pidfname[256]; /**< name of our PID file */
145 char default_authname[] = "/etc/nbd_server.allow"; /**< default name of allow file */
148 * Variables associated with a server.
151 gchar* exportname; /**< (unprocessed) filename of the file we're exporting */
152 off_t hunksize; /**< size of a hunk of an exported file */
153 off_t expected_size; /**< size of the exported file as it was told to
154 us through configuration */
155 unsigned int port; /**< port we're exporting this file at */
156 char* authname; /**< filename of the authorization file */
157 int flags; /**< flags associated with this exported file */
158 unsigned int timeout;/**< how long a connection may be idle
160 int socket; /**< The socket of this server. */
164 * Variables associated with a client socket.
167 off_t exportsize; /**< size of the file we're exporting */
168 char *clientname; /**< peer */
169 char *exportname; /**< (processed) filename of the file we're exporting */
170 GArray *export; /**< array of filedescriptors of exported files;
171 only the first is actually used unless we're
172 doing the multiple file option */
173 int net; /**< The actual client socket */
174 SERVER *server; /**< The server this client is getting data from */
175 char* difffilename; /**< filename of the copy-on-write file, if any */
176 int difffile; /**< filedescriptor of copyonwrite file. @todo
177 shouldn't this be an array too? (cfr export) Or
178 make -m and -c mutually exclusive */
179 u32 difffilelen; /**< number of pages in difffile */
180 u32 *difmap; /**< see comment on the global difmap for this one */
184 * Type of configuration file values
187 PARAM_INT, /**< This parameter is an integer */
188 PARAM_STRING, /**< This parameter is a string */
189 PARAM_BOOL, /**< This parameter is a boolean */
192 * Configuration file values
195 gchar *paramname; /**< Name of the parameter, as it appears in
197 gboolean required; /**< Whether this is a required (as opposed to
198 optional) parameter */
199 PARAM_TYPE ptype; /**< Type of the parameter. */
200 gpointer target; /**< Pointer to where the data of this
201 parameter should be written. If ptype is
202 PARAM_BOOL, the data is or'ed rather than
204 gint flagval; /**< Flag mask for this parameter in case ptype
209 * Check whether a client is allowed to connect. Works with an authorization
210 * file which contains one line per machine, no wildcards.
212 * @param opts The client who's trying to connect.
213 * @return 0 - authorization refused, 1 - OK
215 int authorized_client(CLIENT *opts) {
220 if ((f=fopen(opts->server->authname,"r"))==NULL) {
221 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
222 opts->server->authname,strerror(errno)) ;
226 while (fgets(line,LINELEN,f)!=NULL) {
227 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
237 * Read data from a file descriptor into a buffer
239 * @param f a file descriptor
240 * @param buf a buffer
241 * @param len the number of bytes to be read
243 inline void readit(int f, void *buf, size_t len) {
247 if ((res = read(f, buf, len)) <= 0)
248 err("Read failed: %m");
255 * Write data from a buffer into a filedescriptor
257 * @param f a file descriptor
258 * @param buf a buffer containing data
259 * @param len the number of bytes to be written
261 inline void writeit(int f, void *buf, size_t len) {
265 if ((res = write(f, buf, len)) <= 0)
266 err("Send failed: %m");
273 * Print out a message about how to use nbd-server. Split out to a separate
274 * function so that we can call it from multiple places
277 printf("This is nbd-server version " VERSION "\n");
278 printf("Usage: port file_to_export [size][kKmM] [-l authorize_file] [-r] [-m] [-c] [-a timeout_sec] [-C configuration file]\n"
279 "\t-r|--read-only\t\tread only\n"
280 "\t-m|--multi-file\t\tmultiple file\n"
281 "\t-c|--copy-on-write\tcopy on write\n"
282 "\t-C|--config-file\tspecify an alternat configuration file\n"
283 "\t-l|--authorize-file\tfile with list of hosts that are allowed to\n\t\t\t\tconnect.\n"
284 "\t-a|--idle-time\t\tmaximum idle seconds; server terminates when\n\t\t\t\tidle time exceeded\n\n"
285 "\tif port is set to 0, stdin is used (for running from inetd)\n"
286 "\tif file_to_export contains '%%s', it is substituted with the IP\n"
287 "\t\taddress of the machine trying to connect\n" );
288 printf("Using configuration file %s\n", CFILE);
292 * Parse the command line.
294 * @param argc the argc argument to main()
295 * @param argv the argv argument to main()
297 SERVER* cmdline(int argc, char *argv[]) {
301 struct option long_options[] = {
302 {"read-only", no_argument, NULL, 'r'},
303 {"multi-file", no_argument, NULL, 'm'},
304 {"copy-on-write", no_argument, NULL, 'c'},
305 {"authorize-file", required_argument, NULL, 'l'},
306 {"idle-time", required_argument, NULL, 'a'},
307 {"config-file", required_argument, NULL, 'C'},
318 serve=g_new0(SERVER, 1);
319 serve->hunksize=OFFT_MAX;
320 serve->authname = g_strdup(default_authname);
321 while((c=getopt_long(argc, argv, "-a:C:cl:mr", long_options, &i))>=0) {
324 /* non-option argument */
325 switch(nonspecial++) {
327 serve->port=strtol(optarg, NULL, 0);
330 serve->exportname = g_strdup(optarg);
331 if(serve->exportname[0] != '/') {
332 fprintf(stderr, "E: The to be exported file needs to be an absolute filename!\n");
337 last=strlen(optarg)-1;
339 if (suffix == 'k' || suffix == 'K' ||
340 suffix == 'm' || suffix == 'M')
342 es = (off_t)atol(optarg);
350 serve->expected_size = es;
355 serve->flags |= F_READONLY;
358 serve->flags |= F_MULTIFILE;
359 serve->hunksize = 1*GIGA;
362 serve->flags |=F_COPYONWRITE;
365 g_free(config_file_pos);
366 config_file_pos=g_strdup(optarg);
369 g_free(serve->authname);
370 serve->authname=g_strdup(optarg);
373 serve->timeout=strtol(optarg, NULL, 0);
381 /* What's left: the port to export, the name of the to be exported
382 * file, and, optionally, the size of the file, in that order. */
391 * Error codes for config file parsing
394 CFILE_NOTFOUND, /**< The configuration file is not found */
395 CFILE_MISSING_GENERIC, /**< The (required) group "generic" is missing */
396 CFILE_KEY_MISSING, /**< A (required) key is missing */
397 CFILE_VALUE_INVALID, /**< A value is syntactically invalid */
398 CFILE_PROGERR /**< Programmer error */
402 * Remove a SERVER from memory. Used from the hash table
404 void remove_server(gpointer s) {
408 g_free(server->exportname);
410 g_free(server->authname);
415 * Parse the config file.
417 * @param f the name of the config file
418 * @param e a GError. @see CFILE_ERRORS for what error values this function can
420 * @return a GHashTable of SERVER* pointers, with the port number as the hash
421 * key. If the config file is empty or does not exist, returns an empty
422 * GHashTable; if the config file contains an error, returns NULL, and
423 * e is set appropriately
425 GArray* parse_cfile(gchar* f, GError** e) {
428 { "exportname", TRUE, PARAM_STRING, NULL, 0 },
429 { "port", TRUE, PARAM_INT, NULL, 0 },
430 { "authfile", FALSE, PARAM_STRING, NULL, 0 },
431 { "timeout", FALSE, PARAM_INT, NULL, 0 },
432 { "filesize", FALSE, PARAM_INT, NULL, 0 },
433 { "readonly", FALSE, PARAM_BOOL, NULL, F_READONLY },
434 { "multifile", FALSE, PARAM_BOOL, NULL, F_MULTIFILE },
435 { "copyonwrite", FALSE, PARAM_BOOL, NULL, F_COPYONWRITE },
446 memset(&s, '\0', sizeof(SERVER));
447 errdomain = g_quark_from_string("parse_cfile");
448 cfile = g_key_file_new();
449 retval = g_array_new(FALSE, TRUE, sizeof(SERVER));
450 if(!g_key_file_load_from_file(cfile, f, G_KEY_FILE_KEEP_COMMENTS |
451 G_KEY_FILE_KEEP_TRANSLATIONS, &err)) {
452 g_set_error(e, errdomain, CFILE_NOTFOUND, "Could not open config file.");
453 g_key_file_free(cfile);
456 if(strcmp(g_key_file_get_start_group(cfile), "generic")) {
457 g_set_error(e, errdomain, CFILE_MISSING_GENERIC, "Config file does not contain the [generic] group!");
458 g_key_file_free(cfile);
461 groups = g_key_file_get_groups(cfile, NULL);
462 for(i=1;groups[i];i++) {
463 p[0].target=&(s.exportname);
464 p[1].target=&(s.port);
465 p[2].target=&(s.authname);
466 p[3].target=&(s.timeout);
467 p[4].target=&(s.expected_size);
468 p[5].target=p[6].target=p[7].target=p[8].target=&(s.flags);
469 for(j=0;j<p_size;j++) {
470 g_assert(p[j].target != NULL);
471 g_assert(p[j].ptype==PARAM_INT||p[j].ptype==PARAM_STRING||p[j].ptype==PARAM_BOOL);
474 *((gint*)p[j].target) =
475 g_key_file_get_integer(cfile,
481 *((gchar**)p[j].target) =
482 g_key_file_get_string(cfile,
488 value = g_key_file_get_boolean(cfile,
490 p[j].paramname, &err);
492 *((gint*)p[j].target) |= value;
497 if(err->code == G_KEY_FILE_ERROR_KEY_NOT_FOUND) {
499 g_set_error(e, errdomain, CFILE_KEY_MISSING, "Could not find required value %s in group %s: %s", p[j].paramname, groups[i], err->message);
500 g_array_free(retval, TRUE);
502 g_key_file_free(cfile);
508 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Could not parse %s in group %s: %s", p[j].paramname, groups[i], err->message);
509 g_array_free(retval, TRUE);
511 g_key_file_free(cfile);
516 g_array_append_val(retval, s);
522 * Signal handler for SIGCHLD
523 * @param s the signal we're handling (must be SIGCHLD, or something
526 void sigchld_handler(int s) {
531 while((pid=waitpid(-1, &status, WNOHANG)) > 0) {
532 if(WIFEXITED(&status)) {
533 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
535 i=g_hash_table_lookup(children, &pid);
537 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
539 DEBUG2("Removing %d from the list of children", pid);
540 g_hash_table_remove(children, &pid);
546 * Kill a child. Called from sigterm_handler::g_hash_table_foreach.
549 * @param value the value corresponding to the above key
550 * @param user_data a pointer which we always set to 1, so that we know what
553 void killchild(gpointer key, gpointer value, gpointer user_data) {
555 int *parent=user_data;
562 * Handle SIGTERM and dispatch it to our children
563 * @param s the signal we're handling (must be SIGTERM, or something
564 * is severely wrong).
566 void sigterm_handler(int s) {
569 g_hash_table_foreach(children, killchild, &parent);
579 * Detect the size of a file.
581 * @param export An open filedescriptor
582 * @return the size of the file, or OFFT_MAX if detection was
585 off_t size_autodetect(int export) {
588 struct stat stat_buf;
591 #ifdef HAVE_SYS_MOUNT_H
592 #ifdef HAVE_SYS_IOCTL_H
594 DEBUG("looking for export size with ioctl BLKGETSIZE\n");
595 if (!ioctl(export, BLKGETSIZE, &es32) && es32) {
596 es = (off_t)es32 * (off_t)512;
599 #endif /* BLKGETSIZE */
600 #endif /* HAVE_SYS_IOCTL_H */
601 #endif /* HAVE_SYS_MOUNT_H */
603 DEBUG("looking for export size with fstat\n");
604 stat_buf.st_size = 0;
605 error = fstat(export, &stat_buf);
607 if(stat_buf.st_size > 0)
608 return (off_t)stat_buf.st_size;
610 err("fstat failed: %m");
613 DEBUG("looking for export size with lseek SEEK_END\n");
614 es = lseek(export, (off_t)0, SEEK_END);
615 if (es > ((off_t)0)) {
618 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
621 err("Could not find size of exported block device: %m");
626 * seek to a position in a file, with error handling.
627 * @param handle a filedescriptor
628 * @param a position to seek to
629 * @todo get rid of this; lastpoint is a global variable right now, but it
630 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
633 void myseek(int handle,off_t a) {
634 if (lseek(handle, a, SEEK_SET) < 0) {
635 err("Can not seek locally!\n");
640 * Write an amount of bytes at a given offset to the right file. This
641 * abstracts the write-side of the multiple file option.
643 * @param a The offset where the write should start
644 * @param buf The buffer to write from
645 * @param len The length of buf
646 * @param client The client we're serving for
647 * @return The number of bytes actually written, or -1 in case of an error
649 int rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
652 myseek(g_array_index(client->export, int, (int)(a/client->server->hunksize)), a%client->server->hunksize);
654 res = write(g_array_index(client->export, int, (int)((off_t)a/(off_t)(client->server->hunksize))), buf, len);
655 return (res < 0 || (size_t)res != len);
659 * Read an amount of bytes at a given offset from the right file. This
660 * abstracts the read-side of the multiple files option.
662 * @param a The offset where the read should start
663 * @param buf A buffer to read into
664 * @param len The size of buf
665 * @param client The client we're serving for
666 * @return The number of bytes actually read, or -1 in case of an
669 int rawexpread(off_t a, char *buf, size_t len, CLIENT *client) {
672 myseek(g_array_index(client->export,int,(int)a/client->server->hunksize),
673 a%client->server->hunksize);
674 res = read(g_array_index(client->export,int,(int)a/client->server->hunksize), buf, len);
675 return (res < 0 || (size_t)res != len);
679 * Read an amount of bytes at a given offset from the right file. This
680 * abstracts the read-side of the copyonwrite stuff, and calls
681 * rawexpread() with the right parameters to do the actual work.
682 * @param a The offset where the read should start
683 * @param buf A buffer to read into
684 * @param len The size of buf
685 * @param client The client we're going to read for
686 * @return The number of bytes actually read, or -1 in case of an error
688 int expread(off_t a, char *buf, size_t len, CLIENT *client) {
690 off_t mapcnt, mapl, maph, pagestart;
692 if (!(client->server->flags & F_COPYONWRITE))
693 return rawexpread(a, buf, len, client);
694 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
696 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
698 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
699 pagestart=mapcnt*DIFFPAGESIZE;
701 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
702 len : (size_t)DIFFPAGESIZE-offset;
703 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
704 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
705 (unsigned long)(client->difmap[mapcnt]));
706 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
707 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
708 } else { /* the block is not there */
709 DEBUG2("Page %Lu is not here, we read the original one\n",
710 (unsigned long long)mapcnt);
711 if(rawexpread(a, buf, rdlen, client)) return -1;
713 len-=rdlen; a+=rdlen; buf+=rdlen;
719 * Write an amount of bytes at a given offset to the right file. This
720 * abstracts the write-side of the copyonwrite option, and calls
721 * rawexpwrite() with the right parameters to do the actual work.
723 * @param a The offset where the write should start
724 * @param buf The buffer to write from
725 * @param len The length of buf
726 * @param client The client we're going to write for.
727 * @return The number of bytes actually written, or -1 in case of an error
729 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
730 char pagebuf[DIFFPAGESIZE];
731 off_t mapcnt,mapl,maph;
736 if (!(client->server->flags & F_COPYONWRITE))
737 return(rawexpwrite(a,buf,len, client));
738 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
740 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
742 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
743 pagestart=mapcnt*DIFFPAGESIZE ;
745 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
746 len : (size_t)DIFFPAGESIZE-offset;
748 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
749 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
750 (unsigned long)(client->difmap[mapcnt])) ;
751 myseek(client->difffile,
752 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
753 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
754 } else { /* the block is not there */
755 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
756 client->difmap[mapcnt]=client->difffilelen++ ;
757 DEBUG3("Page %Lu is not here, we put it at %lu\n",
758 (unsigned long long)mapcnt,
759 (unsigned long)(client->difmap[mapcnt]));
761 if (rdlen+pagestart%(client->server->hunksize) >
762 (client->server->hunksize))
763 rdlen=client->server->hunksize -
764 (pagestart%client->server->hunksize);
765 if (rawexpread(pagestart, pagebuf, rdlen, client))
767 memcpy(pagebuf+offset,buf,wrlen) ;
768 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
772 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
778 * Do the initial negotiation.
780 * @param client The client we're negotiating with.
782 void negotiate(CLIENT *client) {
786 memset(zeros, 0, 290);
787 if (write(client->net, INIT_PASSWD, 8) < 0)
788 err("Negotiation failed: %m");
789 cliserv_magic = htonll(cliserv_magic);
790 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
791 err("Negotiation failed: %m");
792 size_host = htonll((u64)(client->exportsize));
793 if (write(client->net, &size_host, 8) < 0)
794 err("Negotiation failed: %m");
795 if (write(client->net, zeros, 128) < 0)
796 err("Negotiation failed: %m");
799 /** sending macro. */
800 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
802 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; }
804 * Serve a file to a single client.
806 * @todo This beast needs to be split up in many tiny little manageable
807 * pieces. Preferably with a chainsaw.
809 * @param client The client we're going to serve to.
812 int mainloop(CLIENT *client) {
813 struct nbd_request request;
814 struct nbd_reply reply;
820 DEBUG("Entering request loop!\n");
821 reply.magic = htonl(NBD_REPLY_MAGIC);
830 if (client->server->timeout)
831 alarm(client->server->timeout);
832 readit(client->net, &request, sizeof(request));
833 request.from = ntohll(request.from);
834 request.type = ntohl(request.type);
836 if (request.type==NBD_CMD_DISC) {
837 msg2(LOG_INFO, "Disconnect request received.");
838 if (client->difmap) g_free(client->difmap) ;
839 if (client->difffile>=0) {
840 close(client->difffile);
841 unlink(client->difffilename);
842 free(client->difffilename);
848 len = ntohl(request.len);
850 if (request.magic != htonl(NBD_REQUEST_MAGIC))
851 err("Not enough magic.");
852 if (len > BUFSIZE + sizeof(struct nbd_reply))
853 err("Request too big!");
855 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
856 "READ", (unsigned long long)request.from,
857 (unsigned long long)request.from / 512, len);
859 memcpy(reply.handle, request.handle, sizeof(reply.handle));
860 if ((request.from + len) > (OFFT_MAX)) {
861 DEBUG("[Number too large!]");
862 ERROR(client, reply);
866 if (((ssize_t)((off_t)request.from + len) > client->exportsize) ||
867 ((client->server->flags & F_READONLY) && request.type)) {
869 ERROR(client, reply);
873 if (request.type==NBD_CMD_WRITE) {
874 DEBUG("wr: net->buf, ");
875 readit(client->net, buf, len);
877 if ((client->server->flags & F_AUTOREADONLY) ||
878 expwrite(request.from, buf, len,
880 DEBUG("Write failed: %m" );
881 ERROR(client, reply);
884 SEND(client->net, reply);
891 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
892 DEBUG("Read failed: %m");
893 ERROR(client, reply);
898 memcpy(buf, &reply, sizeof(struct nbd_reply));
899 writeit(client->net, buf, len + sizeof(struct nbd_reply));
906 * Split a single exportfile into multiple ones, if that was asked.
907 * @return 0 on success, -1 on failure
908 * @param client information on the client which we want to split
910 int splitexport(CLIENT* client) {
914 client->export = g_array_new(TRUE, TRUE, sizeof(int));
915 for (i=0; i<client->exportsize; i+=client->server->hunksize) {
918 if(client->server->flags & F_MULTIFILE) {
919 tmpname=g_strdup_printf("%s.%d", client->exportname,
920 (int)(i/client->server->hunksize));
922 tmpname=g_strdup(client->exportname);
924 DEBUG2( "Opening %s\n", tmpname );
925 if((fhandle = open(tmpname, (client->server->flags & F_READONLY) ? O_RDONLY : O_RDWR)) == -1) {
926 /* Read WRITE ACCESS was requested by media is only read only */
927 client->server->flags |= F_AUTOREADONLY;
928 client->server->flags |= F_READONLY;
929 if((fhandle = open(tmpname, O_RDONLY)) == -1)
930 err("Could not open exported file: %m");
932 g_array_insert_val(client->export,i/client->server->hunksize,fhandle);
938 int copyonwrite_prepare(CLIENT* client) {
940 if ((client->difffilename = malloc(1024))==NULL)
941 err("Failed to allocate string for diff file name");
942 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
944 client->difffilename[1023]='\0';
945 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
946 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
947 if (client->difffile<0) err("Could not create diff file (%m)") ;
948 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
949 err("Could not allocate memory") ;
950 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
956 * Serve a connection.
958 * @todo allow for multithreading, perhaps use libevent. Not just yet, though;
959 * follow the road map.
961 * @param client a connected client
963 void serveconnection(CLIENT *client) {
966 if (!client->server->expected_size) {
967 client->exportsize = size_autodetect(g_array_index(client->export,int,0));
969 /* Perhaps we should check first. Not now. */
970 client->exportsize = client->server->expected_size;
972 if (client->exportsize > OFFT_MAX) {
973 /* uhm, well... In a parallel universe, this *might* be
975 err("Size of exported file is too big\n");
978 msg3(LOG_INFO, "size of exported file/device is %Lu", (unsigned long long)client->exportsize);
981 if (client->server->flags & F_COPYONWRITE) {
982 copyonwrite_prepare(client);
985 setmysockopt(client->net);
991 * Find the name of the file we have to serve. This will use g_strdup_printf
992 * to put the IP address of the client inside a filename containing
993 * "%s". That name is then written to client->exportname.
995 * @param net A socket connected to an nbd client
996 * @param client information about the client. The IP address in human-readable
997 * format will be written to a new char* buffer, the address of which will be
998 * stored in client->clientname.
1000 void set_peername(int net, CLIENT *client) {
1001 struct sockaddr_in addrin;
1002 int addrinlen = sizeof( addrin );
1005 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
1006 err("getsockname failed: %m");
1007 peername = inet_ntoa(addrin.sin_addr);
1008 client->exportname=g_strdup_printf(client->server->exportname, peername);
1010 msg4(LOG_INFO, "connect from %s, assigned file is %s",
1011 peername, client->exportname);
1012 client->clientname=g_strdup(peername);
1017 * @param data a pointer to pid_t which should be freed
1019 void destroy_pid_t(gpointer data) {
1024 * Go daemon (unless we specified at compile time that we didn't want this)
1025 * @param serve the first server of our configuration. If its port is zero,
1026 * then do not daemonize, because we're doing inetd then. This parameter
1027 * is only used to create a PID file of the form
1028 * /var/run/nbd-server.<port>.pid; it's not modified in any way.
1030 #if !defined(NODAEMON) && !defined(NOFORK)
1031 void daemonize(SERVER* serve) {
1038 snprintf(pidfname, sizeof(char)*255, "/var/run/nbd-server.%d.pid", serve->port);
1040 strncpy(pidfname, "/var/run/nbd-server.pid", sizeof(char)*255);
1042 pidf=fopen(pidfname, "w");
1044 fprintf(pidf,"%d\n", (int)getpid());
1048 fprintf(stderr, "Not fatal; continuing");
1052 #define daemonize(serve)
1053 #endif /* !defined(NODAEMON) && !defined(NOFORK) */
1056 * Connect a server's socket.
1058 * @param serve the server we want to connect.
1060 void setup_serve(SERVER *serve) {
1061 struct sockaddr_in addrin;
1062 struct sigaction sa;
1063 int addrinlen = sizeof(addrin);
1070 if ((serve->socket = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
1073 /* lose the pesky "Address already in use" error message */
1074 if (setsockopt(serve->socket,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
1075 err("setsockopt SO_REUSEADDR");
1077 if (setsockopt(serve->socket,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
1078 err("setsockopt SO_KEEPALIVE");
1081 /* make the listening socket non-blocking */
1082 if ((sock_flags = fcntl(serve->socket, F_GETFL, 0)) == -1) {
1083 err("fcntl F_GETFL");
1085 if (fcntl(serve->socket, F_SETFL, sock_flags | O_NONBLOCK) == -1) {
1086 err("fcntl F_SETFL O_NONBLOCK");
1089 DEBUG("Waiting for connections... bind, ");
1090 addrin.sin_family = AF_INET;
1091 addrin.sin_port = htons(serve->port);
1092 addrin.sin_addr.s_addr = 0;
1093 if (bind(serve->socket, (struct sockaddr *) &addrin, addrinlen) < 0)
1096 if (listen(serve->socket, 1) < 0)
1098 sa.sa_handler = sigchld_handler;
1099 sigemptyset(&sa.sa_mask);
1100 sa.sa_flags = SA_RESTART;
1101 if(sigaction(SIGCHLD, &sa, NULL) == -1)
1102 err("sigaction: %m");
1103 sa.sa_handler = sigterm_handler;
1104 sigemptyset(&sa.sa_mask);
1105 sa.sa_flags = SA_RESTART;
1106 if(sigaction(SIGTERM, &sa, NULL) == -1)
1107 err("sigaction: %m");
1108 children=g_hash_table_new_full(g_int_hash, g_int_equal, NULL, destroy_pid_t);
1112 * Connect our servers.
1114 void setup_servers(GArray* servers) {
1117 for(i=0;i<servers->len;i++) {
1118 setup_serve(&(g_array_index(servers, SERVER, i)));
1123 * Loop through the available servers, and serve them.
1125 int serveloop(GArray* servers) {
1126 struct sockaddr_in addrin;
1127 socklen_t addrinlen=sizeof(addrin);
1137 * Set up the master fd_set. The set of descriptors we need
1138 * to select() for never changes anyway and it buys us a *lot*
1139 * of time to only build this once. However, if we ever choose
1140 * to not fork() for clients anymore, we may have to revisit
1145 for(i=0;i<servers->len;i++) {
1146 sock=(g_array_index(servers, SERVER, i)).socket;
1147 FD_SET(sock, &mset);
1148 max=sock>max?sock:max;
1155 memcpy(&rset, &mset, sizeof(fd_set));
1158 if(select(max+1, &rset, NULL, NULL, &tv)>0) {
1160 for(i=0;i<servers->len;i++) {
1161 serve=&(g_array_index(servers, SERVER, i));
1162 if(FD_ISSET(serve->socket, &rset)) {
1163 if ((net=accept(serve->socket, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1166 client = g_malloc(sizeof(CLIENT));
1167 client->server=serve;
1168 client->exportsize=OFFT_MAX;
1170 set_peername(net, client);
1171 if (!authorized_client(client)) {
1172 msg2(LOG_INFO,"Unauthorized client") ;
1176 msg2(LOG_INFO,"Authorized client") ;
1177 pid=g_malloc(sizeof(pid_t));
1179 if ((*pid=fork())<0) {
1180 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
1184 if (*pid>0) { /* parent */
1186 g_hash_table_insert(children, pid, pid);
1190 g_hash_table_destroy(children);
1191 for(i=0;i<servers->len,serve=(g_array_index(servers, SERVER*, i));i++) {
1192 close(serve->socket);
1194 /* FALSE does not free the
1195 actual data. This is required,
1196 because the client has a
1197 direct reference into that
1198 data, and otherwise we get a
1200 g_array_free(servers, FALSE);
1202 msg2(LOG_INFO,"Starting to serve");
1203 serveconnection(client);
1211 * Main entry point...
1213 int main(int argc, char *argv[]) {
1218 if (sizeof( struct nbd_request )!=28) {
1219 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1224 config_file_pos = g_strdup(CFILE);
1225 serve=cmdline(argc, argv);
1226 servers = parse_cfile(config_file_pos, &err);
1227 if(!servers || !servers->len) {
1228 g_warning("Could not parse config file: %s", err->message);
1231 g_array_append_val(servers, *serve);
1234 /* We don't support this at this time */
1236 if (!(serve->port)) {
1239 /* You really should define ISSERVER if you're going to use
1240 * inetd mode, but if you don't, closing stdout and stderr
1241 * (which inetd had connected to the client socket) will let it
1245 open("/dev/null", O_WRONLY);
1246 open("/dev/null", O_WRONLY);
1248 client=g_malloc(sizeof(CLIENT));
1249 client->server=serve;
1251 client->exportsize=OFFT_MAX;
1252 set_peername(0,client);
1253 serveconnection(client);
1257 if((!serve) && (!servers||!servers->len)) {
1258 g_message("Nothing to do! Bye!");
1262 setup_servers(servers);