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>
90 /* used in cliserv.h, so must come first */
91 #define MY_NAME "nbd_server"
94 /** Default position of the config file */
96 #define SYSCONFDIR "/etc"
98 #define CFILE SYSCONFDIR "/nbd-server/config"
100 /** Where our config file actually is */
101 gchar* config_file_pos;
103 /** What user we're running as */
105 /** What group we're running as */
106 gchar* rungroup=NULL;
108 /** Logging macros, now nothing goes to syslog unless you say ISSERVER */
110 #define msg2(a,b) syslog(a,b)
111 #define msg3(a,b,c) syslog(a,b,c)
112 #define msg4(a,b,c,d) syslog(a,b,c,d)
114 #define msg2(a,b) g_message(b)
115 #define msg3(a,b,c) g_message(b,c)
116 #define msg4(a,b,c,d) g_message(b,c,d)
119 /* Debugging macros */
122 #define DEBUG( a ) printf( a )
123 #define DEBUG2( a,b ) printf( a,b )
124 #define DEBUG3( a,b,c ) printf( a,b,c )
125 #define DEBUG4( a,b,c,d ) printf( a,b,c,d )
128 #define DEBUG2( a,b )
129 #define DEBUG3( a,b,c )
130 #define DEBUG4( a,b,c,d )
132 #ifndef PACKAGE_VERSION
133 #define PACKAGE_VERSION ""
136 * The highest value a variable of type off_t can reach. This is a signed
137 * integer, so set all bits except for the leftmost one.
139 #define OFFT_MAX ~((off_t)1<<(sizeof(off_t)*8-1))
140 #define LINELEN 256 /**< Size of static buffer used to read the
141 authorization file (yuck) */
142 #define BUFSIZE (1024*1024) /**< Size of buffer that can hold requests */
143 #define DIFFPAGESIZE 4096 /**< diff file uses those chunks */
144 #define F_READONLY 1 /**< flag to tell us a file is readonly */
145 #define F_MULTIFILE 2 /**< flag to tell us a file is exported using -m */
146 #define F_COPYONWRITE 4 /**< flag to tell us a file is exported using
148 #define F_AUTOREADONLY 8 /**< flag to tell us a file is set to autoreadonly */
150 GHashTable *children;
151 char pidfname[256]; /**< name of our PID file */
152 char default_authname[] = SYSCONFDIR "/nbd-server/allow"; /**< default name of allow file */
155 * Types of virtuatlization
158 VIRT_NONE=0, /**< No virtualization */
159 VIRT_IPLIT, /**< Literal IP address as part of the filename */
160 VIRT_IPHASH, /**< Replacing all dots in an ip address by a / before
161 doing the same as in IPLIT */
162 VIRT_CIDR, /**< Every subnet in its own directory */
166 * Variables associated with a server.
169 gchar* exportname; /**< (unprocessed) filename of the file we're exporting */
170 off_t expected_size; /**< size of the exported file as it was told to
171 us through configuration */
172 unsigned int port; /**< port we're exporting this file at */
173 char* authname; /**< filename of the authorization file */
174 int flags; /**< flags associated with this exported file */
175 unsigned int timeout;/**< how long a connection may be idle
177 int socket; /**< The socket of this server. */
178 VIRT_STYLE virtstyle;/**< The style of virtualization, if any */
179 uint8_t cidrlen; /**< The length of the mask when we use
180 CIDR-style virtualization */
184 * Variables associated with a client socket.
187 int fhandle; /**< file descriptor */
188 off_t startoff; /**< starting offset of this file */
192 off_t exportsize; /**< size of the file we're exporting */
193 char *clientname; /**< peer */
194 char *exportname; /**< (processed) filename of the file we're exporting */
195 GArray *export; /**< array of FILE_INFO of exported files;
196 array size is always 1 unless we're
197 doing the multiple file option */
198 int net; /**< The actual client socket */
199 SERVER *server; /**< The server this client is getting data from */
200 char* difffilename; /**< filename of the copy-on-write file, if any */
201 int difffile; /**< filedescriptor of copyonwrite file. @todo
202 shouldn't this be an array too? (cfr export) Or
203 make -m and -c mutually exclusive */
204 u32 difffilelen; /**< number of pages in difffile */
205 u32 *difmap; /**< see comment on the global difmap for this one */
209 * Type of configuration file values
212 PARAM_INT, /**< This parameter is an integer */
213 PARAM_STRING, /**< This parameter is a string */
214 PARAM_BOOL, /**< This parameter is a boolean */
218 * Configuration file values
221 gchar *paramname; /**< Name of the parameter, as it appears in
223 gboolean required; /**< Whether this is a required (as opposed to
224 optional) parameter */
225 PARAM_TYPE ptype; /**< Type of the parameter. */
226 gpointer target; /**< Pointer to where the data of this
227 parameter should be written. If ptype is
228 PARAM_BOOL, the data is or'ed rather than
230 gint flagval; /**< Flag mask for this parameter in case ptype
235 * Check whether a client is allowed to connect. Works with an authorization
236 * file which contains one line per machine, no wildcards.
238 * @param opts The client who's trying to connect.
239 * @return 0 - authorization refused, 1 - OK
241 int authorized_client(CLIENT *opts) {
242 const char ERRMSG="Invalid entry '%s' in authfile '%s', so, refusing all connections.";
247 struct in_addr client;
248 struct in_addr cltemp;
251 if ((f=fopen(opts->server->authname,"r"))==NULL) {
252 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
253 opts->server->authname,strerror(errno)) ;
257 inet_aton(opts->clientname, &client);
258 while (fgets(line,LINELEN,f)!=NULL) {
259 if((tmp=index(line, '/'))) {
260 if(strlen(line)<=tmp-line) {
261 msg4(LOG_CRIT, ERRMSG, line, opts->server->authname);
265 if(inet_aton(line,&addr)) {
266 msg4(LOG_CRIT, ERRMSG, line, opts->server->authname);
269 len=strtol(tmp, NULL, 0);
270 addr.sin_addr.s_addr>>=32-len;
271 addr.sin_addr.s_addr<<=32-len;
272 memcpy(&cltemp,&client,sizeof(client));
273 cltemp.sin_addr.s_addr>>=32-len;
274 cltemp.sin_addr.s_addr<<=32-len;
275 if(addr.sin_addr.s_addr == cltemp.sin_addr.s_addr) {
279 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
289 * Read data from a file descriptor into a buffer
291 * @param f a file descriptor
292 * @param buf a buffer
293 * @param len the number of bytes to be read
295 inline void readit(int f, void *buf, size_t len) {
299 if ((res = read(f, buf, len)) <= 0)
300 err("Read failed: %m");
307 * Write data from a buffer into a filedescriptor
309 * @param f a file descriptor
310 * @param buf a buffer containing data
311 * @param len the number of bytes to be written
313 inline void writeit(int f, void *buf, size_t len) {
317 if ((res = write(f, buf, len)) <= 0)
318 err("Send failed: %m");
325 * Print out a message about how to use nbd-server. Split out to a separate
326 * function so that we can call it from multiple places
329 printf("This is nbd-server version " VERSION "\n");
330 printf("Usage: port file_to_export [size][kKmM] [-l authorize_file] [-r] [-m] [-c] [-a timeout_sec] [-C configuration file]\n"
331 "\t-r|--read-only\t\tread only\n"
332 "\t-m|--multi-file\t\tmultiple file\n"
333 "\t-c|--copy-on-write\tcopy on write\n"
334 "\t-C|--config-file\tspecify an alternat configuration file\n"
335 "\t-l|--authorize-file\tfile with list of hosts that are allowed to\n\t\t\t\tconnect.\n"
336 "\t-a|--idle-time\t\tmaximum idle seconds; server terminates when\n\t\t\t\tidle time exceeded\n\n"
337 "\tif port is set to 0, stdin is used (for running from inetd)\n"
338 "\tif file_to_export contains '%%s', it is substituted with the IP\n"
339 "\t\taddress of the machine trying to connect\n" );
340 printf("Using configuration file %s\n", CFILE);
344 * Parse the command line.
346 * @param argc the argc argument to main()
347 * @param argv the argv argument to main()
349 SERVER* cmdline(int argc, char *argv[]) {
353 struct option long_options[] = {
354 {"read-only", no_argument, NULL, 'r'},
355 {"multi-file", no_argument, NULL, 'm'},
356 {"copy-on-write", no_argument, NULL, 'c'},
357 {"authorize-file", required_argument, NULL, 'l'},
358 {"idle-time", required_argument, NULL, 'a'},
359 {"config-file", required_argument, NULL, 'C'},
370 serve=g_new0(SERVER, 1);
371 serve->authname = g_strdup(default_authname);
372 while((c=getopt_long(argc, argv, "-a:C:cl:mr", long_options, &i))>=0) {
375 /* non-option argument */
376 switch(nonspecial++) {
378 serve->port=strtol(optarg, NULL, 0);
381 serve->exportname = g_strdup(optarg);
382 if(serve->exportname[0] != '/') {
383 fprintf(stderr, "E: The to be exported file needs to be an absolute filename!\n");
388 last=strlen(optarg)-1;
390 if (suffix == 'k' || suffix == 'K' ||
391 suffix == 'm' || suffix == 'M')
393 es = (off_t)atol(optarg);
401 serve->expected_size = es;
406 serve->flags |= F_READONLY;
409 serve->flags |= F_MULTIFILE;
412 serve->flags |=F_COPYONWRITE;
415 g_free(config_file_pos);
416 config_file_pos=g_strdup(optarg);
419 g_free(serve->authname);
420 serve->authname=g_strdup(optarg);
423 serve->timeout=strtol(optarg, NULL, 0);
431 /* What's left: the port to export, the name of the to be exported
432 * file, and, optionally, the size of the file, in that order. */
441 * Error codes for config file parsing
444 CFILE_NOTFOUND, /**< The configuration file is not found */
445 CFILE_MISSING_GENERIC, /**< The (required) group "generic" is missing */
446 CFILE_KEY_MISSING, /**< A (required) key is missing */
447 CFILE_VALUE_INVALID, /**< A value is syntactically invalid */
448 CFILE_PROGERR /**< Programmer error */
452 * Remove a SERVER from memory. Used from the hash table
454 void remove_server(gpointer s) {
458 g_free(server->exportname);
460 g_free(server->authname);
465 * Parse the config file.
467 * @param f the name of the config file
468 * @param e a GError. @see CFILE_ERRORS for what error values this function can
470 * @return a Array of SERVER* pointers, If the config file is empty or does not
471 * exist, returns an empty GHashTable; if the config file contains an
472 * error, returns NULL, and e is set appropriately
474 GArray* parse_cfile(gchar* f, GError** e) {
475 const char* DEFAULT_ERROR = "Could not parse %s in group %s: %s";
476 const char* MISSING_REQUIRED_ERROR = "Could not find required value %s in group %s: %s";
478 gchar *virtstyle=NULL;
480 { "exportname", TRUE, PARAM_STRING, NULL, 0 },
481 { "port", TRUE, PARAM_INT, NULL, 0 },
482 { "authfile", FALSE, PARAM_STRING, NULL, 0 },
483 { "timeout", FALSE, PARAM_INT, NULL, 0 },
484 { "filesize", FALSE, PARAM_INT, NULL, 0 },
485 { "virtstyle", FALSE, PARAM_STRING, NULL, 0 },
486 { "readonly", FALSE, PARAM_BOOL, NULL, F_READONLY },
487 { "multifile", FALSE, PARAM_BOOL, NULL, F_MULTIFILE },
488 { "copyonwrite", FALSE, PARAM_BOOL, NULL, F_COPYONWRITE },
489 { "autoreadonly", FALSE, PARAM_BOOL, NULL, F_AUTOREADONLY },
490 { "sparse_cow", FALSE, PARAM_BOOL, NULL, F_SPARSE },
492 const int lp_size=11;
494 { "user", FALSE, PARAM_STRING, &runuser, 0 },
495 { "group", FALSE, PARAM_STRING, &rungroup, 0 },
501 const char *err_msg=NULL;
508 errdomain = g_quark_from_string("parse_cfile");
509 cfile = g_key_file_new();
510 retval = g_array_new(FALSE, TRUE, sizeof(SERVER));
511 if(!g_key_file_load_from_file(cfile, f, G_KEY_FILE_KEEP_COMMENTS |
512 G_KEY_FILE_KEEP_TRANSLATIONS, &err)) {
513 g_set_error(e, errdomain, CFILE_NOTFOUND, "Could not open config file.");
514 g_key_file_free(cfile);
517 if(strcmp(g_key_file_get_start_group(cfile), "generic")) {
518 g_set_error(e, errdomain, CFILE_MISSING_GENERIC, "Config file does not contain the [generic] group!");
519 g_key_file_free(cfile);
522 groups = g_key_file_get_groups(cfile, NULL);
523 for(i=0;groups[i];i++) {
524 memset(&s, '\0', sizeof(SERVER));
525 lp[0].target=&(s.exportname);
526 lp[1].target=&(s.port);
527 lp[2].target=&(s.authname);
528 lp[3].target=&(s.timeout);
529 lp[4].target=&(s.expected_size);
530 lp[5].target=&(virtstyle);
531 lp[6].target=lp[7].target=lp[8].target=
532 lp[9].target=lp[10].target=&(s.flags);
533 /* After the [generic] group, start parsing exports */
538 for(j=0;j<p_size;j++) {
539 g_assert(p[j].target != NULL);
540 g_assert(p[j].ptype==PARAM_INT||p[j].ptype==PARAM_STRING||p[j].ptype==PARAM_BOOL);
543 *((gint*)p[j].target) =
544 g_key_file_get_integer(cfile,
550 *((gchar**)p[j].target) =
551 g_key_file_get_string(cfile,
557 value = g_key_file_get_boolean(cfile,
559 p[j].paramname, &err);
562 *((gint*)p[j].target) |= p[j].flagval;
564 *((gint*)p[j].target) &= ~(p[j].flagval);
570 if(err->code == G_KEY_FILE_ERROR_KEY_NOT_FOUND) {
572 /* Ignore not-found error for optional values */
576 err_msg = MISSING_REQUIRED_ERROR;
579 err_msg = DEFAULT_ERROR;
581 g_set_error(e, errdomain, CFILE_VALUE_INVALID, err_msg, p[j].paramname, groups[i], err->message);
582 g_array_free(retval, TRUE);
584 g_key_file_free(cfile);
589 if(!strncmp(virtstyle, "none", 4)) {
590 s.virtstyle=VIRT_NONE;
591 } else if(!strncmp(virtstyle, "ipliteral", 9)) {
592 s.virtstyle=VIRT_IPLIT;
593 } else if(!strncmp(virtstyle, "iphash", 6)) {
594 s.virtstyle=VIRT_IPHASH;
595 } else if(!strncmp(virtstyle, "cidrhash", 8)) {
596 s.virtstyle=VIRT_CIDR;
597 if(strlen(virtstyle)<10) {
598 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s: missing length", virtstyle, groups[i]);
599 g_array_free(retval, TRUE);
600 g_key_file_free(cfile);
603 s.cidrlen=strtol(virtstyle+8, NULL, 0);
605 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s", virtstyle, groups[i]);
606 g_array_free(retval, TRUE);
607 g_key_file_free(cfile);
611 s.virtstyle=VIRT_IPLIT;
613 /* Don't need to free this, it's not our string */
615 /* Don't append values for the [generic] group */
617 g_array_append_val(retval, s);
624 * Signal handler for SIGCHLD
625 * @param s the signal we're handling (must be SIGCHLD, or something
628 void sigchld_handler(int s) {
633 while((pid=waitpid(-1, &status, WNOHANG)) > 0) {
634 if(WIFEXITED(&status)) {
635 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
637 i=g_hash_table_lookup(children, &pid);
639 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
641 DEBUG2("Removing %d from the list of children", pid);
642 g_hash_table_remove(children, &pid);
648 * Kill a child. Called from sigterm_handler::g_hash_table_foreach.
651 * @param value the value corresponding to the above key
652 * @param user_data a pointer which we always set to 1, so that we know what
655 void killchild(gpointer key, gpointer value, gpointer user_data) {
657 int *parent=user_data;
664 * Handle SIGTERM and dispatch it to our children
665 * @param s the signal we're handling (must be SIGTERM, or something
666 * is severely wrong).
668 void sigterm_handler(int s) {
671 g_hash_table_foreach(children, killchild, &parent);
681 * Detect the size of a file.
683 * @param fhandle An open filedescriptor
684 * @return the size of the file, or OFFT_MAX if detection was
687 off_t size_autodetect(int fhandle) {
690 struct stat stat_buf;
693 #ifdef HAVE_SYS_MOUNT_H
694 #ifdef HAVE_SYS_IOCTL_H
696 DEBUG("looking for fhandle size with ioctl BLKGETSIZE\n");
697 if (!ioctl(fhandle, BLKGETSIZE, &es32) && es32) {
698 es = (off_t)es32 * (off_t)512;
701 #endif /* BLKGETSIZE */
702 #endif /* HAVE_SYS_IOCTL_H */
703 #endif /* HAVE_SYS_MOUNT_H */
705 DEBUG("looking for fhandle size with fstat\n");
706 stat_buf.st_size = 0;
707 error = fstat(fhandle, &stat_buf);
709 if(stat_buf.st_size > 0)
710 return (off_t)stat_buf.st_size;
712 err("fstat failed: %m");
715 DEBUG("looking for fhandle size with lseek SEEK_END\n");
716 es = lseek(fhandle, (off_t)0, SEEK_END);
717 if (es > ((off_t)0)) {
720 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
723 err("Could not find size of exported block device: %m");
728 * Get the file handle and offset, given an export offset.
730 * @param export An array of export files
731 * @param a The offset to get corresponding file/offset for
732 * @param fhandle [out] File descriptor
733 * @param foffset [out] Offset into fhandle
734 * @param maxbytes [out] Tells how many bytes can be read/written
735 * from fhandle starting at foffset (0 if there is no limit)
736 * @return 0 on success, -1 on failure
738 int get_filepos(GArray* export, off_t a, int* fhandle, off_t* foffset, size_t* maxbytes ) {
739 /* Negative offset not allowed */
743 /* Binary search for last file with starting offset <= a */
746 int end = export->len - 1;
747 while( start <= end ) {
748 int mid = (start + end) / 2;
749 fi = g_array_index(export, FILE_INFO, mid);
750 if( fi.startoff < a ) {
752 } else if( fi.startoff > a ) {
760 /* end should never go negative, since first startoff is 0 and a >= 0 */
763 fi = g_array_index(export, FILE_INFO, end);
764 *fhandle = fi.fhandle;
765 *foffset = a - fi.startoff;
767 if( end+1 < export->len ) {
768 FILE_INFO fi_next = g_array_index(export, FILE_INFO, end+1);
769 *maxbytes = fi_next.startoff - a;
776 * seek to a position in a file, with error handling.
777 * @param handle a filedescriptor
778 * @param a position to seek to
779 * @todo get rid of this; lastpoint is a global variable right now, but it
780 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
783 void myseek(int handle,off_t a) {
784 if (lseek(handle, a, SEEK_SET) < 0) {
785 err("Can not seek locally!\n");
790 * Write an amount of bytes at a given offset to the right file. This
791 * abstracts the write-side of the multiple file option.
793 * @param a The offset where the write should start
794 * @param buf The buffer to write from
795 * @param len The length of buf
796 * @param client The client we're serving for
797 * @return The number of bytes actually written, or -1 in case of an error
799 ssize_t rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
804 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
806 if(maxbytes && len > maxbytes)
809 DEBUG4("(WRITE to fd %d offset %Lu len %u), ", fhandle, foffset, len);
811 myseek(fhandle, foffset);
812 return write(fhandle, buf, len);
816 * Call rawexpwrite repeatedly until all data has been written.
817 * @return 0 on success, nonzero on failure
819 int rawexpwrite_fully(off_t a, char *buf, size_t len, CLIENT *client) {
822 while(len > 0 && (ret=rawexpwrite(a, buf, len, client)) > 0 ) {
827 return (ret < 0 || len != 0);
831 * Read an amount of bytes at a given offset from the right file. This
832 * abstracts the read-side of the multiple files option.
834 * @param a The offset where the read should start
835 * @param buf A buffer to read into
836 * @param len The size of buf
837 * @param client The client we're serving for
838 * @return The number of bytes actually read, or -1 in case of an
841 ssize_t rawexpread(off_t a, char *buf, size_t len, CLIENT *client) {
846 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
848 if(maxbytes && len > maxbytes)
851 DEBUG4("(READ from fd %d offset %Lu len %u), ", fhandle, foffset, len);
853 myseek(fhandle, foffset);
854 return read(fhandle, buf, len);
858 * Call rawexpread repeatedly until all data has been read.
859 * @return 0 on success, nonzero on failure
861 int rawexpread_fully(off_t a, char *buf, size_t len, CLIENT *client) {
864 while(len > 0 && (ret=rawexpread(a, buf, len, client)) > 0 ) {
869 return (ret < 0 || len != 0);
873 * Read an amount of bytes at a given offset from the right file. This
874 * abstracts the read-side of the copyonwrite stuff, and calls
875 * rawexpread() with the right parameters to do the actual work.
876 * @param a The offset where the read should start
877 * @param buf A buffer to read into
878 * @param len The size of buf
879 * @param client The client we're going to read for
880 * @return 0 on success, nonzero on failure
882 int expread(off_t a, char *buf, size_t len, CLIENT *client) {
884 off_t mapcnt, mapl, maph, pagestart;
886 if (!(client->server->flags & F_COPYONWRITE))
887 return(rawexpread_fully(a, buf, len, client));
888 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
890 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
892 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
893 pagestart=mapcnt*DIFFPAGESIZE;
895 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
896 len : (size_t)DIFFPAGESIZE-offset;
897 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
898 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
899 (unsigned long)(client->difmap[mapcnt]));
900 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
901 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
902 } else { /* the block is not there */
903 DEBUG2("Page %Lu is not here, we read the original one\n",
904 (unsigned long long)mapcnt);
905 if(rawexpread_fully(a, buf, rdlen, client)) return -1;
907 len-=rdlen; a+=rdlen; buf+=rdlen;
913 * Write an amount of bytes at a given offset to the right file. This
914 * abstracts the write-side of the copyonwrite option, and calls
915 * rawexpwrite() with the right parameters to do the actual work.
917 * @param a The offset where the write should start
918 * @param buf The buffer to write from
919 * @param len The length of buf
920 * @param client The client we're going to write for.
921 * @return 0 on success, nonzero on failure
923 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
924 char pagebuf[DIFFPAGESIZE];
925 off_t mapcnt,mapl,maph;
930 if (!(client->server->flags & F_COPYONWRITE))
931 return(rawexpwrite_fully(a, buf, len, client));
932 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
934 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
936 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
937 pagestart=mapcnt*DIFFPAGESIZE ;
939 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
940 len : (size_t)DIFFPAGESIZE-offset;
942 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
943 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
944 (unsigned long)(client->difmap[mapcnt])) ;
945 myseek(client->difffile,
946 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
947 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
948 } else { /* the block is not there */
949 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
950 client->difmap[mapcnt]=(client->server->flags&F_SPARSE)?mapcnt:client->difffilelen++;
951 DEBUG3("Page %Lu is not here, we put it at %lu\n",
952 (unsigned long long)mapcnt,
953 (unsigned long)(client->difmap[mapcnt]));
955 if (rawexpread_fully(pagestart, pagebuf, rdlen, client))
957 memcpy(pagebuf+offset,buf,wrlen) ;
958 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
962 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
968 * Do the initial negotiation.
970 * @param client The client we're negotiating with.
972 void negotiate(CLIENT *client) {
976 memset(zeros, '\0', 290);
977 if (write(client->net, INIT_PASSWD, 8) < 0)
978 err("Negotiation failed: %m");
979 cliserv_magic = htonll(cliserv_magic);
980 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
981 err("Negotiation failed: %m");
982 size_host = htonll((u64)(client->exportsize));
983 if (write(client->net, &size_host, 8) < 0)
984 err("Negotiation failed: %m");
985 if (write(client->net, zeros, 128) < 0)
986 err("Negotiation failed: %m");
989 /** sending macro. */
990 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
992 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; }
994 * Serve a file to a single client.
996 * @todo This beast needs to be split up in many tiny little manageable
997 * pieces. Preferably with a chainsaw.
999 * @param client The client we're going to serve to.
1002 int mainloop(CLIENT *client) {
1003 struct nbd_request request;
1004 struct nbd_reply reply;
1005 gboolean go_on=TRUE;
1010 DEBUG("Entering request loop!\n");
1011 reply.magic = htonl(NBD_REPLY_MAGIC);
1020 if (client->server->timeout)
1021 alarm(client->server->timeout);
1022 readit(client->net, &request, sizeof(request));
1023 request.from = ntohll(request.from);
1024 request.type = ntohl(request.type);
1026 if (request.type==NBD_CMD_DISC) {
1027 msg2(LOG_INFO, "Disconnect request received.");
1028 if (client->server->flags & F_COPYONWRITE) {
1029 if (client->difmap) g_free(client->difmap) ;
1030 close(client->difffile);
1031 unlink(client->difffilename);
1032 free(client->difffilename);
1038 len = ntohl(request.len);
1040 if (request.magic != htonl(NBD_REQUEST_MAGIC))
1041 err("Not enough magic.");
1042 if (len > BUFSIZE + sizeof(struct nbd_reply))
1043 err("Request too big!");
1045 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
1046 "READ", (unsigned long long)request.from,
1047 (unsigned long long)request.from / 512, len);
1049 memcpy(reply.handle, request.handle, sizeof(reply.handle));
1050 if ((request.from + len) > (OFFT_MAX)) {
1051 DEBUG("[Number too large!]");
1052 ERROR(client, reply);
1056 if (((ssize_t)((off_t)request.from + len) > client->exportsize)) {
1058 ERROR(client, reply);
1062 if (request.type==NBD_CMD_WRITE) {
1063 DEBUG("wr: net->buf, ");
1064 readit(client->net, buf, len);
1065 DEBUG("buf->exp, ");
1066 if ((client->server->flags & F_READONLY) ||
1067 (client->server->flags & F_AUTOREADONLY)) {
1068 DEBUG("[WRITE to READONLY!]");
1069 ERROR(client, reply);
1072 if (expwrite(request.from, buf, len, client)) {
1073 DEBUG("Write failed: %m" );
1074 ERROR(client, reply);
1077 SEND(client->net, reply);
1083 DEBUG("exp->buf, ");
1084 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
1085 DEBUG("Read failed: %m");
1086 ERROR(client, reply);
1090 DEBUG("buf->net, ");
1091 memcpy(buf, &reply, sizeof(struct nbd_reply));
1092 writeit(client->net, buf, len + sizeof(struct nbd_reply));
1099 * Set up client export array, which is an array of FILE_INFO.
1100 * Also, split a single exportfile into multiple ones, if that was asked.
1101 * @param client information on the client which we want to setup export for
1103 void setupexport(CLIENT* client) {
1105 off_t laststartoff = 0, lastsize = 0;
1106 int multifile = (client->server->flags & F_MULTIFILE);
1108 client->export = g_array_new(TRUE, TRUE, sizeof(FILE_INFO));
1110 /* If multi-file, open as many files as we can.
1111 * If not, open exactly one file.
1112 * Calculate file sizes as we go to get total size. */
1116 mode_t mode = (client->server->flags & F_READONLY) ? O_RDONLY : O_RDWR;
1119 tmpname=g_strdup_printf("%s.%d", client->exportname, i);
1121 tmpname=g_strdup(client->exportname);
1123 DEBUG2( "Opening %s\n", tmpname );
1124 fi.fhandle = open(tmpname, mode);
1125 if(fi.fhandle == -1 && mode == O_RDWR) {
1126 /* Try again because maybe media was read-only */
1127 fi.fhandle = open(tmpname, O_RDONLY);
1128 if(fi.fhandle != -1) {
1129 client->server->flags |= F_AUTOREADONLY;
1130 client->server->flags |= F_READONLY;
1133 if(fi.fhandle == -1) {
1134 if(multifile && i>0)
1136 err("Could not open exported file: %m");
1138 fi.startoff = laststartoff + lastsize;
1139 g_array_append_val(client->export, fi);
1142 /* Starting offset and size of this file will be used to
1143 * calculate starting offset of next file */
1144 laststartoff = fi.startoff;
1145 lastsize = size_autodetect(fi.fhandle);
1151 /* Set export size to total calculated size */
1152 client->exportsize = laststartoff + lastsize;
1154 /* Export size may be overridden */
1155 if(client->server->expected_size) {
1156 /* desired size must be <= total calculated size */
1157 if(client->server->expected_size > client->exportsize) {
1158 err("Size of exported file is too big\n");
1161 client->exportsize = client->server->expected_size;
1164 msg3(LOG_INFO, "Size of exported file/device is %Lu", (unsigned long long)client->exportsize);
1166 msg3(LOG_INFO, "Total number of files: %d", i);
1170 int copyonwrite_prepare(CLIENT* client) {
1172 if ((client->difffilename = malloc(1024))==NULL)
1173 err("Failed to allocate string for diff file name");
1174 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
1176 client->difffilename[1023]='\0';
1177 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
1178 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
1179 if (client->difffile<0) err("Could not create diff file (%m)") ;
1180 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
1181 err("Could not allocate memory") ;
1182 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
1188 * Serve a connection.
1190 * @todo allow for multithreading, perhaps use libevent. Not just yet, though;
1191 * follow the road map.
1193 * @param client a connected client
1195 void serveconnection(CLIENT *client) {
1196 setupexport(client);
1198 if (client->server->flags & F_COPYONWRITE) {
1199 copyonwrite_prepare(client);
1202 setmysockopt(client->net);
1208 * Find the name of the file we have to serve. This will use g_strdup_printf
1209 * to put the IP address of the client inside a filename containing
1210 * "%s" (in the form as specified by the "virtstyle" option). That name
1211 * is then written to client->exportname.
1213 * @param net A socket connected to an nbd client
1214 * @param client information about the client. The IP address in human-readable
1215 * format will be written to a new char* buffer, the address of which will be
1216 * stored in client->clientname.
1218 void set_peername(int net, CLIENT *client) {
1219 struct sockaddr_in addrin;
1220 struct sockaddr_in netaddr;
1221 size_t addrinlen = sizeof( addrin );
1227 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
1228 err("getsockname failed: %m");
1229 peername = inet_ntoa(addrin.sin_addr);
1230 switch(client->server->virtstyle) {
1232 client->exportname=g_strdup(client->server->exportname);
1235 for(i=0;i<strlen(peername);i++) {
1236 if(peername[i]=='.') {
1241 client->exportname=g_strdup_printf(client->server->exportname, peername);
1244 memcpy(&netaddr, &addrin, addrinlen);
1245 netaddr.sin_addr.s_addr>>=32-(client->server->cidrlen);
1246 netaddr.sin_addr.s_addr<<=32-(client->server->cidrlen);
1247 netname = inet_ntoa(netaddr.sin_addr);
1248 tmp=g_strdup_printf("%s/%s", netname, peername);
1249 client->exportname=g_strdup_printf(client->server->exportname, tmp);
1253 msg4(LOG_INFO, "connect from %s, assigned file is %s",
1254 peername, client->exportname);
1255 client->clientname=g_strdup(peername);
1260 * @param data a pointer to pid_t which should be freed
1262 void destroy_pid_t(gpointer data) {
1267 * Go daemon (unless we specified at compile time that we didn't want this)
1268 * @param serve the first server of our configuration. If its port is zero,
1269 * then do not daemonize, because we're doing inetd then. This parameter
1270 * is only used to create a PID file of the form
1271 * /var/run/nbd-server.<port>.pid; it's not modified in any way.
1273 #if !defined(NODAEMON) && !defined(NOFORK)
1274 void daemonize(SERVER* serve) {
1281 snprintf(pidfname, sizeof(char)*255, "/var/run/nbd-server.%d.pid", serve->port);
1283 strncpy(pidfname, "/var/run/nbd-server.pid", sizeof(char)*255);
1285 pidf=fopen(pidfname, "w");
1287 fprintf(pidf,"%d\n", (int)getpid());
1291 fprintf(stderr, "Not fatal; continuing");
1295 #define daemonize(serve)
1296 #endif /* !defined(NODAEMON) && !defined(NOFORK) */
1299 * Connect a server's socket.
1301 * @param serve the server we want to connect.
1303 void setup_serve(SERVER *serve) {
1304 struct sockaddr_in addrin;
1305 struct sigaction sa;
1306 int addrinlen = sizeof(addrin);
1313 if ((serve->socket = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
1316 /* lose the pesky "Address already in use" error message */
1317 if (setsockopt(serve->socket,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
1318 err("setsockopt SO_REUSEADDR");
1320 if (setsockopt(serve->socket,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
1321 err("setsockopt SO_KEEPALIVE");
1324 /* make the listening socket non-blocking */
1325 if ((sock_flags = fcntl(serve->socket, F_GETFL, 0)) == -1) {
1326 err("fcntl F_GETFL");
1328 if (fcntl(serve->socket, F_SETFL, sock_flags | O_NONBLOCK) == -1) {
1329 err("fcntl F_SETFL O_NONBLOCK");
1332 DEBUG("Waiting for connections... bind, ");
1333 addrin.sin_family = AF_INET;
1334 addrin.sin_port = htons(serve->port);
1335 addrin.sin_addr.s_addr = 0;
1336 if (bind(serve->socket, (struct sockaddr *) &addrin, addrinlen) < 0)
1339 if (listen(serve->socket, 1) < 0)
1341 sa.sa_handler = sigchld_handler;
1342 sigemptyset(&sa.sa_mask);
1343 sa.sa_flags = SA_RESTART;
1344 if(sigaction(SIGCHLD, &sa, NULL) == -1)
1345 err("sigaction: %m");
1346 sa.sa_handler = sigterm_handler;
1347 sigemptyset(&sa.sa_mask);
1348 sa.sa_flags = SA_RESTART;
1349 if(sigaction(SIGTERM, &sa, NULL) == -1)
1350 err("sigaction: %m");
1351 children=g_hash_table_new_full(g_int_hash, g_int_equal, NULL, destroy_pid_t);
1355 * Connect our servers.
1357 void setup_servers(GArray* servers) {
1360 for(i=0;i<servers->len;i++) {
1361 setup_serve(&(g_array_index(servers, SERVER, i)));
1366 * Loop through the available servers, and serve them.
1368 int serveloop(GArray* servers) {
1369 struct sockaddr_in addrin;
1370 socklen_t addrinlen=sizeof(addrin);
1380 * Set up the master fd_set. The set of descriptors we need
1381 * to select() for never changes anyway and it buys us a *lot*
1382 * of time to only build this once. However, if we ever choose
1383 * to not fork() for clients anymore, we may have to revisit
1388 for(i=0;i<servers->len;i++) {
1389 sock=(g_array_index(servers, SERVER, i)).socket;
1390 FD_SET(sock, &mset);
1391 max=sock>max?sock:max;
1398 memcpy(&rset, &mset, sizeof(fd_set));
1401 if(select(max+1, &rset, NULL, NULL, &tv)>0) {
1403 for(i=0;i<servers->len;i++) {
1404 serve=&(g_array_index(servers, SERVER, i));
1405 if(FD_ISSET(serve->socket, &rset)) {
1406 if ((net=accept(serve->socket, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1409 client = g_malloc(sizeof(CLIENT));
1410 client->server=serve;
1411 client->exportsize=OFFT_MAX;
1413 set_peername(net, client);
1414 if (!authorized_client(client)) {
1415 msg2(LOG_INFO,"Unauthorized client") ;
1419 msg2(LOG_INFO,"Authorized client") ;
1420 pid=g_malloc(sizeof(pid_t));
1422 if ((*pid=fork())<0) {
1423 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
1427 if (*pid>0) { /* parent */
1429 g_hash_table_insert(children, pid, pid);
1433 g_hash_table_destroy(children);
1434 for(i=0;i<servers->len,serve=(g_array_index(servers, SERVER*, i));i++) {
1435 close(serve->socket);
1437 /* FALSE does not free the
1438 actual data. This is required,
1439 because the client has a
1440 direct reference into that
1441 data, and otherwise we get a
1443 g_array_free(servers, FALSE);
1445 msg2(LOG_INFO,"Starting to serve");
1446 serveconnection(client);
1454 * Set up user-ID and/or group-ID
1456 void dousers(void) {
1460 pw=getpwnam(runuser);
1461 if(setuid(pw->pw_uid)<0)
1462 msg3(LOG_DEBUG, "Could not set UID: %s", strerror(errno));
1465 gr=getgrnam(rungroup);
1466 if(setgid(gr->gr_gid)<0)
1467 msg3(LOG_DEBUG, "Could not set GID: %s", strerror(errno));
1472 * Main entry point...
1474 int main(int argc, char *argv[]) {
1479 if (sizeof( struct nbd_request )!=28) {
1480 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1485 config_file_pos = g_strdup(CFILE);
1486 serve=cmdline(argc, argv);
1487 servers = parse_cfile(config_file_pos, &err);
1488 if(!servers || !servers->len) {
1489 g_warning("Could not parse config file: %s", err->message);
1492 g_array_append_val(servers, *serve);
1495 /* We don't support this at this time */
1497 if (!(serve->port)) {
1500 /* You really should define ISSERVER if you're going to use
1501 * inetd mode, but if you don't, closing stdout and stderr
1502 * (which inetd had connected to the client socket) will let it
1506 open("/dev/null", O_WRONLY);
1507 open("/dev/null", O_WRONLY);
1509 client=g_malloc(sizeof(CLIENT));
1510 client->server=serve;
1512 client->exportsize=OFFT_MAX;
1513 set_peername(0,client);
1514 serveconnection(client);
1518 if((!serve) && (!servers||!servers->len)) {
1519 g_message("Nothing to do! Bye!");
1523 setup_servers(servers);