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 pidftemplate[256]; /**< template to be used for the filename of the PID file */
153 char default_authname[] = SYSCONFDIR "/nbd-server/allow"; /**< default name of allow file */
156 * Types of virtuatlization
159 VIRT_NONE=0, /**< No virtualization */
160 VIRT_IPLIT, /**< Literal IP address as part of the filename */
161 VIRT_IPHASH, /**< Replacing all dots in an ip address by a / before
162 doing the same as in IPLIT */
163 VIRT_CIDR, /**< Every subnet in its own directory */
167 * Variables associated with a server.
170 gchar* exportname; /**< (unprocessed) filename of the file we're exporting */
171 off_t expected_size; /**< size of the exported file as it was told to
172 us through configuration */
173 unsigned int port; /**< port we're exporting this file at */
174 char* authname; /**< filename of the authorization file */
175 int flags; /**< flags associated with this exported file */
176 unsigned int timeout;/**< how long a connection may be idle
178 int socket; /**< The socket of this server. */
179 VIRT_STYLE virtstyle;/**< The style of virtualization, if any */
180 uint8_t cidrlen; /**< The length of the mask when we use
181 CIDR-style virtualization */
185 * Variables associated with a client socket.
188 int fhandle; /**< file descriptor */
189 off_t startoff; /**< starting offset of this file */
193 off_t exportsize; /**< size of the file we're exporting */
194 char *clientname; /**< peer */
195 char *exportname; /**< (processed) filename of the file we're exporting */
196 GArray *export; /**< array of FILE_INFO of exported files;
197 array size is always 1 unless we're
198 doing the multiple file option */
199 int net; /**< The actual client socket */
200 SERVER *server; /**< The server this client is getting data from */
201 char* difffilename; /**< filename of the copy-on-write file, if any */
202 int difffile; /**< filedescriptor of copyonwrite file. @todo
203 shouldn't this be an array too? (cfr export) Or
204 make -m and -c mutually exclusive */
205 u32 difffilelen; /**< number of pages in difffile */
206 u32 *difmap; /**< see comment on the global difmap for this one */
210 * Type of configuration file values
213 PARAM_INT, /**< This parameter is an integer */
214 PARAM_STRING, /**< This parameter is a string */
215 PARAM_BOOL, /**< This parameter is a boolean */
219 * Configuration file values
222 gchar *paramname; /**< Name of the parameter, as it appears in
224 gboolean required; /**< Whether this is a required (as opposed to
225 optional) parameter */
226 PARAM_TYPE ptype; /**< Type of the parameter. */
227 gpointer target; /**< Pointer to where the data of this
228 parameter should be written. If ptype is
229 PARAM_BOOL, the data is or'ed rather than
231 gint flagval; /**< Flag mask for this parameter in case ptype
236 * Check whether a client is allowed to connect. Works with an authorization
237 * file which contains one line per machine, no wildcards.
239 * @param opts The client who's trying to connect.
240 * @return 0 - authorization refused, 1 - OK
242 int authorized_client(CLIENT *opts) {
243 const char *ERRMSG="Invalid entry '%s' in authfile '%s', so, refusing all connections.";
248 struct in_addr client;
249 struct in_addr cltemp;
252 if ((f=fopen(opts->server->authname,"r"))==NULL) {
253 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
254 opts->server->authname,strerror(errno)) ;
258 inet_aton(opts->clientname, &client);
259 while (fgets(line,LINELEN,f)!=NULL) {
260 if((tmp=index(line, '/'))) {
261 if(strlen(line)<=tmp-line) {
262 msg4(LOG_CRIT, ERRMSG, line, opts->server->authname);
266 if(inet_aton(line,&addr)) {
267 msg4(LOG_CRIT, ERRMSG, line, opts->server->authname);
270 len=strtol(tmp, NULL, 0);
271 addr.s_addr>>=32-len;
272 addr.s_addr<<=32-len;
273 memcpy(&cltemp,&client,sizeof(client));
274 cltemp.s_addr>>=32-len;
275 cltemp.s_addr<<=32-len;
276 if(addr.s_addr == cltemp.s_addr) {
280 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
290 * Read data from a file descriptor into a buffer
292 * @param f a file descriptor
293 * @param buf a buffer
294 * @param len the number of bytes to be read
296 inline void readit(int f, void *buf, size_t len) {
300 if ((res = read(f, buf, len)) <= 0)
301 err("Read failed: %m");
308 * Write data from a buffer into a filedescriptor
310 * @param f a file descriptor
311 * @param buf a buffer containing data
312 * @param len the number of bytes to be written
314 inline void writeit(int f, void *buf, size_t len) {
318 if ((res = write(f, buf, len)) <= 0)
319 err("Send failed: %m");
326 * Print out a message about how to use nbd-server. Split out to a separate
327 * function so that we can call it from multiple places
330 printf("This is nbd-server version " VERSION "\n");
331 printf("Usage: port file_to_export [size][kKmM] [-l authorize_file] [-r] [-m] [-c] [-a timeout_sec] [-C configuration file] [-p PID file name]\n"
332 "\t-r|--read-only\t\tread only\n"
333 "\t-m|--multi-file\t\tmultiple file\n"
334 "\t-c|--copy-on-write\tcopy on write\n"
335 "\t-C|--config-file\tspecify an alternate configuration file\n"
336 "\t-l|--authorize-file\tfile with list of hosts that are allowed to\n\t\t\t\tconnect.\n"
337 "\t-a|--idle-time\t\tmaximum idle seconds; server terminates when\n\t\t\t\tidle time exceeded\n"
338 "\t-p|--pid-file\t\tspecify a filename to write our PID to\n\n"
339 "\tif port is set to 0, stdin is used (for running from inetd)\n"
340 "\tif file_to_export contains '%%s', it is substituted with the IP\n"
341 "\t\taddress of the machine trying to connect\n" );
342 printf("Using configuration file %s\n", CFILE);
346 * Parse the command line.
348 * @param argc the argc argument to main()
349 * @param argv the argv argument to main()
351 SERVER* cmdline(int argc, char *argv[]) {
355 struct option long_options[] = {
356 {"read-only", no_argument, NULL, 'r'},
357 {"multi-file", no_argument, NULL, 'm'},
358 {"copy-on-write", no_argument, NULL, 'c'},
359 {"authorize-file", required_argument, NULL, 'l'},
360 {"idle-time", required_argument, NULL, 'a'},
361 {"config-file", required_argument, NULL, 'C'},
362 {"pid-file", required_argument, NULL, 'p'},
373 serve=g_new0(SERVER, 1);
374 serve->authname = g_strdup(default_authname);
375 while((c=getopt_long(argc, argv, "-a:C:cl:mrp:", long_options, &i))>=0) {
378 /* non-option argument */
379 switch(nonspecial++) {
381 serve->port=strtol(optarg, NULL, 0);
384 serve->exportname = g_strdup(optarg);
385 if(serve->exportname[0] != '/') {
386 fprintf(stderr, "E: The to be exported file needs to be an absolute filename!\n");
391 last=strlen(optarg)-1;
393 if (suffix == 'k' || suffix == 'K' ||
394 suffix == 'm' || suffix == 'M')
396 es = (off_t)atol(optarg);
404 serve->expected_size = es;
409 serve->flags |= F_READONLY;
412 serve->flags |= F_MULTIFILE;
415 strncpy(pidftemplate, optarg, 256);
418 serve->flags |=F_COPYONWRITE;
421 g_free(config_file_pos);
422 config_file_pos=g_strdup(optarg);
425 g_free(serve->authname);
426 serve->authname=g_strdup(optarg);
429 serve->timeout=strtol(optarg, NULL, 0);
437 /* What's left: the port to export, the name of the to be exported
438 * file, and, optionally, the size of the file, in that order. */
447 * Error codes for config file parsing
450 CFILE_NOTFOUND, /**< The configuration file is not found */
451 CFILE_MISSING_GENERIC, /**< The (required) group "generic" is missing */
452 CFILE_KEY_MISSING, /**< A (required) key is missing */
453 CFILE_VALUE_INVALID, /**< A value is syntactically invalid */
454 CFILE_PROGERR /**< Programmer error */
458 * Remove a SERVER from memory. Used from the hash table
460 void remove_server(gpointer s) {
464 g_free(server->exportname);
466 g_free(server->authname);
471 * Parse the config file.
473 * @param f the name of the config file
474 * @param e a GError. @see CFILE_ERRORS for what error values this function can
476 * @return a Array of SERVER* pointers, If the config file is empty or does not
477 * exist, returns an empty GHashTable; if the config file contains an
478 * error, returns NULL, and e is set appropriately
480 GArray* parse_cfile(gchar* f, GError** e) {
481 const char* DEFAULT_ERROR = "Could not parse %s in group %s: %s";
482 const char* MISSING_REQUIRED_ERROR = "Could not find required value %s in group %s: %s";
484 gchar *virtstyle=NULL;
486 { "exportname", TRUE, PARAM_STRING, NULL, 0 },
487 { "port", TRUE, PARAM_INT, NULL, 0 },
488 { "authfile", FALSE, PARAM_STRING, NULL, 0 },
489 { "timeout", FALSE, PARAM_INT, NULL, 0 },
490 { "filesize", FALSE, PARAM_INT, NULL, 0 },
491 { "virtstyle", FALSE, PARAM_STRING, NULL, 0 },
492 { "readonly", FALSE, PARAM_BOOL, NULL, F_READONLY },
493 { "multifile", FALSE, PARAM_BOOL, NULL, F_MULTIFILE },
494 { "copyonwrite", FALSE, PARAM_BOOL, NULL, F_COPYONWRITE },
495 { "autoreadonly", FALSE, PARAM_BOOL, NULL, F_AUTOREADONLY },
496 { "sparse_cow", FALSE, PARAM_BOOL, NULL, F_SPARSE },
498 const int lp_size=11;
500 { "user", FALSE, PARAM_STRING, &runuser, 0 },
501 { "group", FALSE, PARAM_STRING, &rungroup, 0 },
507 const char *err_msg=NULL;
515 errdomain = g_quark_from_string("parse_cfile");
516 cfile = g_key_file_new();
517 retval = g_array_new(FALSE, TRUE, sizeof(SERVER));
518 if(!g_key_file_load_from_file(cfile, f, G_KEY_FILE_KEEP_COMMENTS |
519 G_KEY_FILE_KEEP_TRANSLATIONS, &err)) {
520 g_set_error(e, errdomain, CFILE_NOTFOUND, "Could not open config file.");
521 g_key_file_free(cfile);
524 if(strcmp(g_key_file_get_start_group(cfile), "generic")) {
525 g_set_error(e, errdomain, CFILE_MISSING_GENERIC, "Config file does not contain the [generic] group!");
526 g_key_file_free(cfile);
529 groups = g_key_file_get_groups(cfile, NULL);
530 for(i=0;groups[i];i++) {
531 memset(&s, '\0', sizeof(SERVER));
532 lp[0].target=&(s.exportname);
533 lp[1].target=&(s.port);
534 lp[2].target=&(s.authname);
535 lp[3].target=&(s.timeout);
536 lp[4].target=&(s.expected_size);
537 lp[5].target=&(virtstyle);
538 lp[6].target=lp[7].target=lp[8].target=
539 lp[9].target=lp[10].target=&(s.flags);
540 /* After the [generic] group, start parsing exports */
545 for(j=0;j<p_size;j++) {
546 g_assert(p[j].target != NULL);
547 g_assert(p[j].ptype==PARAM_INT||p[j].ptype==PARAM_STRING||p[j].ptype==PARAM_BOOL);
550 *((gint*)p[j].target) =
551 g_key_file_get_integer(cfile,
557 *((gchar**)p[j].target) =
558 g_key_file_get_string(cfile,
564 value = g_key_file_get_boolean(cfile,
566 p[j].paramname, &err);
569 *((gint*)p[j].target) |= p[j].flagval;
571 *((gint*)p[j].target) &= ~(p[j].flagval);
577 if(err->code == G_KEY_FILE_ERROR_KEY_NOT_FOUND) {
579 /* Ignore not-found error for optional values */
583 err_msg = MISSING_REQUIRED_ERROR;
586 err_msg = DEFAULT_ERROR;
588 g_set_error(e, errdomain, CFILE_VALUE_INVALID, err_msg, p[j].paramname, groups[i], err->message);
589 g_array_free(retval, TRUE);
591 g_key_file_free(cfile);
596 if(!strncmp(virtstyle, "none", 4)) {
597 s.virtstyle=VIRT_NONE;
598 } else if(!strncmp(virtstyle, "ipliteral", 9)) {
599 s.virtstyle=VIRT_IPLIT;
600 } else if(!strncmp(virtstyle, "iphash", 6)) {
601 s.virtstyle=VIRT_IPHASH;
602 } else if(!strncmp(virtstyle, "cidrhash", 8)) {
603 s.virtstyle=VIRT_CIDR;
604 if(strlen(virtstyle)<10) {
605 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s: missing length", virtstyle, groups[i]);
606 g_array_free(retval, TRUE);
607 g_key_file_free(cfile);
610 s.cidrlen=strtol(virtstyle+8, NULL, 0);
612 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s", virtstyle, groups[i]);
613 g_array_free(retval, TRUE);
614 g_key_file_free(cfile);
618 s.virtstyle=VIRT_IPLIT;
620 /* Don't need to free this, it's not our string */
622 /* Don't append values for the [generic] group */
624 g_array_append_val(retval, s);
631 * Signal handler for SIGCHLD
632 * @param s the signal we're handling (must be SIGCHLD, or something
635 void sigchld_handler(int s) {
640 while((pid=waitpid(-1, &status, WNOHANG)) > 0) {
641 if(WIFEXITED(&status)) {
642 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
644 i=g_hash_table_lookup(children, &pid);
646 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
648 DEBUG2("Removing %d from the list of children", pid);
649 g_hash_table_remove(children, &pid);
655 * Kill a child. Called from sigterm_handler::g_hash_table_foreach.
658 * @param value the value corresponding to the above key
659 * @param user_data a pointer which we always set to 1, so that we know what
662 void killchild(gpointer key, gpointer value, gpointer user_data) {
664 int *parent=user_data;
671 * Handle SIGTERM and dispatch it to our children
672 * @param s the signal we're handling (must be SIGTERM, or something
673 * is severely wrong).
675 void sigterm_handler(int s) {
678 g_hash_table_foreach(children, killchild, &parent);
688 * Detect the size of a file.
690 * @param fhandle An open filedescriptor
691 * @return the size of the file, or OFFT_MAX if detection was
694 off_t size_autodetect(int fhandle) {
697 struct stat stat_buf;
700 #ifdef HAVE_SYS_MOUNT_H
701 #ifdef HAVE_SYS_IOCTL_H
703 DEBUG("looking for fhandle size with ioctl BLKGETSIZE\n");
704 if (!ioctl(fhandle, BLKGETSIZE, &es32) && es32) {
705 es = (off_t)es32 * (off_t)512;
708 #endif /* BLKGETSIZE */
709 #endif /* HAVE_SYS_IOCTL_H */
710 #endif /* HAVE_SYS_MOUNT_H */
712 DEBUG("looking for fhandle size with fstat\n");
713 stat_buf.st_size = 0;
714 error = fstat(fhandle, &stat_buf);
716 if(stat_buf.st_size > 0)
717 return (off_t)stat_buf.st_size;
719 err("fstat failed: %m");
722 DEBUG("looking for fhandle size with lseek SEEK_END\n");
723 es = lseek(fhandle, (off_t)0, SEEK_END);
724 if (es > ((off_t)0)) {
727 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
730 err("Could not find size of exported block device: %m");
735 * Get the file handle and offset, given an export offset.
737 * @param export An array of export files
738 * @param a The offset to get corresponding file/offset for
739 * @param fhandle [out] File descriptor
740 * @param foffset [out] Offset into fhandle
741 * @param maxbytes [out] Tells how many bytes can be read/written
742 * from fhandle starting at foffset (0 if there is no limit)
743 * @return 0 on success, -1 on failure
745 int get_filepos(GArray* export, off_t a, int* fhandle, off_t* foffset, size_t* maxbytes ) {
746 /* Negative offset not allowed */
750 /* Binary search for last file with starting offset <= a */
753 int end = export->len - 1;
754 while( start <= end ) {
755 int mid = (start + end) / 2;
756 fi = g_array_index(export, FILE_INFO, mid);
757 if( fi.startoff < a ) {
759 } else if( fi.startoff > a ) {
767 /* end should never go negative, since first startoff is 0 and a >= 0 */
770 fi = g_array_index(export, FILE_INFO, end);
771 *fhandle = fi.fhandle;
772 *foffset = a - fi.startoff;
774 if( end+1 < export->len ) {
775 FILE_INFO fi_next = g_array_index(export, FILE_INFO, end+1);
776 *maxbytes = fi_next.startoff - a;
783 * seek to a position in a file, with error handling.
784 * @param handle a filedescriptor
785 * @param a position to seek to
786 * @todo get rid of this; lastpoint is a global variable right now, but it
787 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
790 void myseek(int handle,off_t a) {
791 if (lseek(handle, a, SEEK_SET) < 0) {
792 err("Can not seek locally!\n");
797 * Write an amount of bytes at a given offset to the right file. This
798 * abstracts the write-side of the multiple file option.
800 * @param a The offset where the write should start
801 * @param buf The buffer to write from
802 * @param len The length of buf
803 * @param client The client we're serving for
804 * @return The number of bytes actually written, or -1 in case of an error
806 ssize_t rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
811 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
813 if(maxbytes && len > maxbytes)
816 DEBUG4("(WRITE to fd %d offset %Lu len %u), ", fhandle, foffset, len);
818 myseek(fhandle, foffset);
819 return write(fhandle, buf, len);
823 * Call rawexpwrite repeatedly until all data has been written.
824 * @return 0 on success, nonzero on failure
826 int rawexpwrite_fully(off_t a, char *buf, size_t len, CLIENT *client) {
829 while(len > 0 && (ret=rawexpwrite(a, buf, len, client)) > 0 ) {
834 return (ret < 0 || len != 0);
838 * Read an amount of bytes at a given offset from the right file. This
839 * abstracts the read-side of the multiple files option.
841 * @param a The offset where the read should start
842 * @param buf A buffer to read into
843 * @param len The size of buf
844 * @param client The client we're serving for
845 * @return The number of bytes actually read, or -1 in case of an
848 ssize_t rawexpread(off_t a, char *buf, size_t len, CLIENT *client) {
853 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
855 if(maxbytes && len > maxbytes)
858 DEBUG4("(READ from fd %d offset %Lu len %u), ", fhandle, foffset, len);
860 myseek(fhandle, foffset);
861 return read(fhandle, buf, len);
865 * Call rawexpread repeatedly until all data has been read.
866 * @return 0 on success, nonzero on failure
868 int rawexpread_fully(off_t a, char *buf, size_t len, CLIENT *client) {
871 while(len > 0 && (ret=rawexpread(a, buf, len, client)) > 0 ) {
876 return (ret < 0 || len != 0);
880 * Read an amount of bytes at a given offset from the right file. This
881 * abstracts the read-side of the copyonwrite stuff, and calls
882 * rawexpread() with the right parameters to do the actual work.
883 * @param a The offset where the read should start
884 * @param buf A buffer to read into
885 * @param len The size of buf
886 * @param client The client we're going to read for
887 * @return 0 on success, nonzero on failure
889 int expread(off_t a, char *buf, size_t len, CLIENT *client) {
891 off_t mapcnt, mapl, maph, pagestart;
893 if (!(client->server->flags & F_COPYONWRITE))
894 return(rawexpread_fully(a, buf, len, client));
895 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
897 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
899 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
900 pagestart=mapcnt*DIFFPAGESIZE;
902 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
903 len : (size_t)DIFFPAGESIZE-offset;
904 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
905 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
906 (unsigned long)(client->difmap[mapcnt]));
907 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
908 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
909 } else { /* the block is not there */
910 DEBUG2("Page %Lu is not here, we read the original one\n",
911 (unsigned long long)mapcnt);
912 if(rawexpread_fully(a, buf, rdlen, client)) return -1;
914 len-=rdlen; a+=rdlen; buf+=rdlen;
920 * Write an amount of bytes at a given offset to the right file. This
921 * abstracts the write-side of the copyonwrite option, and calls
922 * rawexpwrite() with the right parameters to do the actual work.
924 * @param a The offset where the write should start
925 * @param buf The buffer to write from
926 * @param len The length of buf
927 * @param client The client we're going to write for.
928 * @return 0 on success, nonzero on failure
930 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
931 char pagebuf[DIFFPAGESIZE];
932 off_t mapcnt,mapl,maph;
937 if (!(client->server->flags & F_COPYONWRITE))
938 return(rawexpwrite_fully(a, buf, len, client));
939 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
941 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
943 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
944 pagestart=mapcnt*DIFFPAGESIZE ;
946 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
947 len : (size_t)DIFFPAGESIZE-offset;
949 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
950 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
951 (unsigned long)(client->difmap[mapcnt])) ;
952 myseek(client->difffile,
953 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
954 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
955 } else { /* the block is not there */
956 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
957 client->difmap[mapcnt]=(client->server->flags&F_SPARSE)?mapcnt:client->difffilelen++;
958 DEBUG3("Page %Lu is not here, we put it at %lu\n",
959 (unsigned long long)mapcnt,
960 (unsigned long)(client->difmap[mapcnt]));
962 if (rawexpread_fully(pagestart, pagebuf, rdlen, client))
964 memcpy(pagebuf+offset,buf,wrlen) ;
965 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
969 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
975 * Do the initial negotiation.
977 * @param client The client we're negotiating with.
979 void negotiate(CLIENT *client) {
983 memset(zeros, '\0', 290);
984 if (write(client->net, INIT_PASSWD, 8) < 0)
985 err("Negotiation failed: %m");
986 cliserv_magic = htonll(cliserv_magic);
987 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
988 err("Negotiation failed: %m");
989 size_host = htonll((u64)(client->exportsize));
990 if (write(client->net, &size_host, 8) < 0)
991 err("Negotiation failed: %m");
992 if (write(client->net, zeros, 128) < 0)
993 err("Negotiation failed: %m");
996 /** sending macro. */
997 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
999 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; }
1001 * Serve a file to a single client.
1003 * @todo This beast needs to be split up in many tiny little manageable
1004 * pieces. Preferably with a chainsaw.
1006 * @param client The client we're going to serve to.
1009 int mainloop(CLIENT *client) {
1010 struct nbd_request request;
1011 struct nbd_reply reply;
1012 gboolean go_on=TRUE;
1017 DEBUG("Entering request loop!\n");
1018 reply.magic = htonl(NBD_REPLY_MAGIC);
1027 if (client->server->timeout)
1028 alarm(client->server->timeout);
1029 readit(client->net, &request, sizeof(request));
1030 request.from = ntohll(request.from);
1031 request.type = ntohl(request.type);
1033 if (request.type==NBD_CMD_DISC) {
1034 msg2(LOG_INFO, "Disconnect request received.");
1035 if (client->server->flags & F_COPYONWRITE) {
1036 if (client->difmap) g_free(client->difmap) ;
1037 close(client->difffile);
1038 unlink(client->difffilename);
1039 free(client->difffilename);
1045 len = ntohl(request.len);
1047 if (request.magic != htonl(NBD_REQUEST_MAGIC))
1048 err("Not enough magic.");
1049 if (len > BUFSIZE + sizeof(struct nbd_reply))
1050 err("Request too big!");
1052 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
1053 "READ", (unsigned long long)request.from,
1054 (unsigned long long)request.from / 512, len);
1056 memcpy(reply.handle, request.handle, sizeof(reply.handle));
1057 if ((request.from + len) > (OFFT_MAX)) {
1058 DEBUG("[Number too large!]");
1059 ERROR(client, reply);
1063 if (((ssize_t)((off_t)request.from + len) > client->exportsize)) {
1065 ERROR(client, reply);
1069 if (request.type==NBD_CMD_WRITE) {
1070 DEBUG("wr: net->buf, ");
1071 readit(client->net, buf, len);
1072 DEBUG("buf->exp, ");
1073 if ((client->server->flags & F_READONLY) ||
1074 (client->server->flags & F_AUTOREADONLY)) {
1075 DEBUG("[WRITE to READONLY!]");
1076 ERROR(client, reply);
1079 if (expwrite(request.from, buf, len, client)) {
1080 DEBUG("Write failed: %m" );
1081 ERROR(client, reply);
1084 SEND(client->net, reply);
1090 DEBUG("exp->buf, ");
1091 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
1092 DEBUG("Read failed: %m");
1093 ERROR(client, reply);
1097 DEBUG("buf->net, ");
1098 memcpy(buf, &reply, sizeof(struct nbd_reply));
1099 writeit(client->net, buf, len + sizeof(struct nbd_reply));
1106 * Set up client export array, which is an array of FILE_INFO.
1107 * Also, split a single exportfile into multiple ones, if that was asked.
1108 * @param client information on the client which we want to setup export for
1110 void setupexport(CLIENT* client) {
1112 off_t laststartoff = 0, lastsize = 0;
1113 int multifile = (client->server->flags & F_MULTIFILE);
1115 client->export = g_array_new(TRUE, TRUE, sizeof(FILE_INFO));
1117 /* If multi-file, open as many files as we can.
1118 * If not, open exactly one file.
1119 * Calculate file sizes as we go to get total size. */
1123 mode_t mode = (client->server->flags & F_READONLY) ? O_RDONLY : O_RDWR;
1126 tmpname=g_strdup_printf("%s.%d", client->exportname, i);
1128 tmpname=g_strdup(client->exportname);
1130 DEBUG2( "Opening %s\n", tmpname );
1131 fi.fhandle = open(tmpname, mode);
1132 if(fi.fhandle == -1 && mode == O_RDWR) {
1133 /* Try again because maybe media was read-only */
1134 fi.fhandle = open(tmpname, O_RDONLY);
1135 if(fi.fhandle != -1) {
1136 client->server->flags |= F_AUTOREADONLY;
1137 client->server->flags |= F_READONLY;
1140 if(fi.fhandle == -1) {
1141 if(multifile && i>0)
1143 err("Could not open exported file: %m");
1145 fi.startoff = laststartoff + lastsize;
1146 g_array_append_val(client->export, fi);
1149 /* Starting offset and size of this file will be used to
1150 * calculate starting offset of next file */
1151 laststartoff = fi.startoff;
1152 lastsize = size_autodetect(fi.fhandle);
1158 /* Set export size to total calculated size */
1159 client->exportsize = laststartoff + lastsize;
1161 /* Export size may be overridden */
1162 if(client->server->expected_size) {
1163 /* desired size must be <= total calculated size */
1164 if(client->server->expected_size > client->exportsize) {
1165 err("Size of exported file is too big\n");
1168 client->exportsize = client->server->expected_size;
1171 msg3(LOG_INFO, "Size of exported file/device is %Lu", (unsigned long long)client->exportsize);
1173 msg3(LOG_INFO, "Total number of files: %d", i);
1177 int copyonwrite_prepare(CLIENT* client) {
1179 if ((client->difffilename = malloc(1024))==NULL)
1180 err("Failed to allocate string for diff file name");
1181 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
1183 client->difffilename[1023]='\0';
1184 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
1185 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
1186 if (client->difffile<0) err("Could not create diff file (%m)") ;
1187 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
1188 err("Could not allocate memory") ;
1189 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
1195 * Serve a connection.
1197 * @todo allow for multithreading, perhaps use libevent. Not just yet, though;
1198 * follow the road map.
1200 * @param client a connected client
1202 void serveconnection(CLIENT *client) {
1203 setupexport(client);
1205 if (client->server->flags & F_COPYONWRITE) {
1206 copyonwrite_prepare(client);
1209 setmysockopt(client->net);
1215 * Find the name of the file we have to serve. This will use g_strdup_printf
1216 * to put the IP address of the client inside a filename containing
1217 * "%s" (in the form as specified by the "virtstyle" option). That name
1218 * is then written to client->exportname.
1220 * @param net A socket connected to an nbd client
1221 * @param client information about the client. The IP address in human-readable
1222 * format will be written to a new char* buffer, the address of which will be
1223 * stored in client->clientname.
1225 void set_peername(int net, CLIENT *client) {
1226 struct sockaddr_in addrin;
1227 struct sockaddr_in netaddr;
1228 size_t addrinlen = sizeof( addrin );
1234 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
1235 err("getsockname failed: %m");
1236 peername = g_strdup(inet_ntoa(addrin.sin_addr));
1237 switch(client->server->virtstyle) {
1239 client->exportname=g_strdup(client->server->exportname);
1242 for(i=0;i<strlen(peername);i++) {
1243 if(peername[i]=='.') {
1248 client->exportname=g_strdup_printf(client->server->exportname, peername);
1251 memcpy(&netaddr, &addrin, addrinlen);
1252 netaddr.sin_addr.s_addr>>=32-(client->server->cidrlen);
1253 netaddr.sin_addr.s_addr<<=32-(client->server->cidrlen);
1254 netname = inet_ntoa(netaddr.sin_addr);
1255 tmp=g_strdup_printf("%s/%s", netname, peername);
1256 client->exportname=g_strdup_printf(client->server->exportname, tmp);
1261 msg4(LOG_INFO, "connect from %s, assigned file is %s",
1262 peername, client->exportname);
1263 client->clientname=g_strdup(peername);
1268 * @param data a pointer to pid_t which should be freed
1270 void destroy_pid_t(gpointer data) {
1275 * Go daemon (unless we specified at compile time that we didn't want this)
1276 * @param serve the first server of our configuration. If its port is zero,
1277 * then do not daemonize, because we're doing inetd then. This parameter
1278 * is only used to create a PID file of the form
1279 * /var/run/nbd-server.<port>.pid; it's not modified in any way.
1281 #if !defined(NODAEMON) && !defined(NOFORK)
1282 void daemonize(SERVER* serve) {
1288 if(!*pidftemplate) {
1290 strncpy(pidftemplate, "/var/run/server.%d.pid", 255);
1292 strncpy(pidftemplate, "/var/run/server.pid", 255);
1295 snprintf(pidfname, 255, pidftemplate, serve->port);
1296 pidf=fopen(pidfname, "w");
1298 fprintf(pidf,"%d\n", (int)getpid());
1302 fprintf(stderr, "Not fatal; continuing");
1306 #define daemonize(serve)
1307 #endif /* !defined(NODAEMON) && !defined(NOFORK) */
1310 * Connect a server's socket.
1312 * @param serve the server we want to connect.
1314 void setup_serve(SERVER *serve) {
1315 struct sockaddr_in addrin;
1316 struct sigaction sa;
1317 int addrinlen = sizeof(addrin);
1324 if ((serve->socket = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
1327 /* lose the pesky "Address already in use" error message */
1328 if (setsockopt(serve->socket,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
1329 err("setsockopt SO_REUSEADDR");
1331 if (setsockopt(serve->socket,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
1332 err("setsockopt SO_KEEPALIVE");
1335 /* make the listening socket non-blocking */
1336 if ((sock_flags = fcntl(serve->socket, F_GETFL, 0)) == -1) {
1337 err("fcntl F_GETFL");
1339 if (fcntl(serve->socket, F_SETFL, sock_flags | O_NONBLOCK) == -1) {
1340 err("fcntl F_SETFL O_NONBLOCK");
1343 DEBUG("Waiting for connections... bind, ");
1344 addrin.sin_family = AF_INET;
1345 addrin.sin_port = htons(serve->port);
1346 addrin.sin_addr.s_addr = 0;
1347 if (bind(serve->socket, (struct sockaddr *) &addrin, addrinlen) < 0)
1350 if (listen(serve->socket, 1) < 0)
1352 sa.sa_handler = sigchld_handler;
1353 sigemptyset(&sa.sa_mask);
1354 sa.sa_flags = SA_RESTART;
1355 if(sigaction(SIGCHLD, &sa, NULL) == -1)
1356 err("sigaction: %m");
1357 sa.sa_handler = sigterm_handler;
1358 sigemptyset(&sa.sa_mask);
1359 sa.sa_flags = SA_RESTART;
1360 if(sigaction(SIGTERM, &sa, NULL) == -1)
1361 err("sigaction: %m");
1362 children=g_hash_table_new_full(g_int_hash, g_int_equal, NULL, destroy_pid_t);
1366 * Connect our servers.
1368 void setup_servers(GArray* servers) {
1371 for(i=0;i<servers->len;i++) {
1372 setup_serve(&(g_array_index(servers, SERVER, i)));
1377 * Loop through the available servers, and serve them.
1379 int serveloop(GArray* servers) {
1380 struct sockaddr_in addrin;
1381 socklen_t addrinlen=sizeof(addrin);
1391 * Set up the master fd_set. The set of descriptors we need
1392 * to select() for never changes anyway and it buys us a *lot*
1393 * of time to only build this once. However, if we ever choose
1394 * to not fork() for clients anymore, we may have to revisit
1399 for(i=0;i<servers->len;i++) {
1400 sock=(g_array_index(servers, SERVER, i)).socket;
1401 FD_SET(sock, &mset);
1402 max=sock>max?sock:max;
1409 memcpy(&rset, &mset, sizeof(fd_set));
1412 if(select(max+1, &rset, NULL, NULL, &tv)>0) {
1414 for(i=0;i<servers->len;i++) {
1415 serve=&(g_array_index(servers, SERVER, i));
1416 if(FD_ISSET(serve->socket, &rset)) {
1417 if ((net=accept(serve->socket, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1420 client = g_malloc(sizeof(CLIENT));
1421 client->server=serve;
1422 client->exportsize=OFFT_MAX;
1424 set_peername(net, client);
1425 if (!authorized_client(client)) {
1426 msg2(LOG_INFO,"Unauthorized client") ;
1430 msg2(LOG_INFO,"Authorized client") ;
1431 pid=g_malloc(sizeof(pid_t));
1433 if ((*pid=fork())<0) {
1434 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
1438 if (*pid>0) { /* parent */
1440 g_hash_table_insert(children, pid, pid);
1444 g_hash_table_destroy(children);
1445 for(i=0;i<servers->len,serve=(g_array_index(servers, SERVER*, i));i++) {
1446 close(serve->socket);
1448 /* FALSE does not free the
1449 actual data. This is required,
1450 because the client has a
1451 direct reference into that
1452 data, and otherwise we get a
1454 g_array_free(servers, FALSE);
1456 msg2(LOG_INFO,"Starting to serve");
1457 serveconnection(client);
1465 * Set up user-ID and/or group-ID
1467 void dousers(void) {
1471 pw=getpwnam(runuser);
1472 if(setuid(pw->pw_uid)<0)
1473 msg3(LOG_DEBUG, "Could not set UID: %s", strerror(errno));
1476 gr=getgrnam(rungroup);
1477 if(setgid(gr->gr_gid)<0)
1478 msg3(LOG_DEBUG, "Could not set GID: %s", strerror(errno));
1483 * Main entry point...
1485 int main(int argc, char *argv[]) {
1490 if (sizeof( struct nbd_request )!=28) {
1491 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1495 memset(pidftemplate, '\0', 256);
1498 config_file_pos = g_strdup(CFILE);
1499 serve=cmdline(argc, argv);
1500 servers = parse_cfile(config_file_pos, &err);
1501 if(!servers || !servers->len) {
1502 g_warning("Could not parse config file: %s", err->message);
1505 g_array_append_val(servers, *serve);
1508 /* We don't support this at this time */
1510 if (!(serve->port)) {
1513 /* You really should define ISSERVER if you're going to use
1514 * inetd mode, but if you don't, closing stdout and stderr
1515 * (which inetd had connected to the client socket) will let it
1519 open("/dev/null", O_WRONLY);
1520 open("/dev/null", O_WRONLY);
1522 client=g_malloc(sizeof(CLIENT));
1523 client->server=serve;
1525 client->exportsize=OFFT_MAX;
1526 set_peername(0,client);
1527 serveconnection(client);
1531 if((!serve) && (!servers||!servers->len)) {
1532 g_message("Nothing to do! Bye!");
1536 setup_servers(servers);