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) {
246 if ((f=fopen(opts->server->authname,"r"))==NULL) {
247 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
248 opts->server->authname,strerror(errno)) ;
252 while (fgets(line,LINELEN,f)!=NULL) {
253 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
263 * Read data from a file descriptor into a buffer
265 * @param f a file descriptor
266 * @param buf a buffer
267 * @param len the number of bytes to be read
269 inline void readit(int f, void *buf, size_t len) {
273 if ((res = read(f, buf, len)) <= 0)
274 err("Read failed: %m");
281 * Write data from a buffer into a filedescriptor
283 * @param f a file descriptor
284 * @param buf a buffer containing data
285 * @param len the number of bytes to be written
287 inline void writeit(int f, void *buf, size_t len) {
291 if ((res = write(f, buf, len)) <= 0)
292 err("Send failed: %m");
299 * Print out a message about how to use nbd-server. Split out to a separate
300 * function so that we can call it from multiple places
303 printf("This is nbd-server version " VERSION "\n");
304 printf("Usage: port file_to_export [size][kKmM] [-l authorize_file] [-r] [-m] [-c] [-a timeout_sec] [-C configuration file]\n"
305 "\t-r|--read-only\t\tread only\n"
306 "\t-m|--multi-file\t\tmultiple file\n"
307 "\t-c|--copy-on-write\tcopy on write\n"
308 "\t-C|--config-file\tspecify an alternat configuration file\n"
309 "\t-l|--authorize-file\tfile with list of hosts that are allowed to\n\t\t\t\tconnect.\n"
310 "\t-a|--idle-time\t\tmaximum idle seconds; server terminates when\n\t\t\t\tidle time exceeded\n\n"
311 "\tif port is set to 0, stdin is used (for running from inetd)\n"
312 "\tif file_to_export contains '%%s', it is substituted with the IP\n"
313 "\t\taddress of the machine trying to connect\n" );
314 printf("Using configuration file %s\n", CFILE);
318 * Parse the command line.
320 * @param argc the argc argument to main()
321 * @param argv the argv argument to main()
323 SERVER* cmdline(int argc, char *argv[]) {
327 struct option long_options[] = {
328 {"read-only", no_argument, NULL, 'r'},
329 {"multi-file", no_argument, NULL, 'm'},
330 {"copy-on-write", no_argument, NULL, 'c'},
331 {"authorize-file", required_argument, NULL, 'l'},
332 {"idle-time", required_argument, NULL, 'a'},
333 {"config-file", required_argument, NULL, 'C'},
344 serve=g_new0(SERVER, 1);
345 serve->authname = g_strdup(default_authname);
346 while((c=getopt_long(argc, argv, "-a:C:cl:mr", long_options, &i))>=0) {
349 /* non-option argument */
350 switch(nonspecial++) {
352 serve->port=strtol(optarg, NULL, 0);
355 serve->exportname = g_strdup(optarg);
356 if(serve->exportname[0] != '/') {
357 fprintf(stderr, "E: The to be exported file needs to be an absolute filename!\n");
362 last=strlen(optarg)-1;
364 if (suffix == 'k' || suffix == 'K' ||
365 suffix == 'm' || suffix == 'M')
367 es = (off_t)atol(optarg);
375 serve->expected_size = es;
380 serve->flags |= F_READONLY;
383 serve->flags |= F_MULTIFILE;
386 serve->flags |=F_COPYONWRITE;
389 g_free(config_file_pos);
390 config_file_pos=g_strdup(optarg);
393 g_free(serve->authname);
394 serve->authname=g_strdup(optarg);
397 serve->timeout=strtol(optarg, NULL, 0);
405 /* What's left: the port to export, the name of the to be exported
406 * file, and, optionally, the size of the file, in that order. */
415 * Error codes for config file parsing
418 CFILE_NOTFOUND, /**< The configuration file is not found */
419 CFILE_MISSING_GENERIC, /**< The (required) group "generic" is missing */
420 CFILE_KEY_MISSING, /**< A (required) key is missing */
421 CFILE_VALUE_INVALID, /**< A value is syntactically invalid */
422 CFILE_PROGERR /**< Programmer error */
426 * Remove a SERVER from memory. Used from the hash table
428 void remove_server(gpointer s) {
432 g_free(server->exportname);
434 g_free(server->authname);
439 * Parse the config file.
441 * @param f the name of the config file
442 * @param e a GError. @see CFILE_ERRORS for what error values this function can
444 * @return a Array of SERVER* pointers, If the config file is empty or does not
445 * exist, returns an empty GHashTable; if the config file contains an
446 * error, returns NULL, and e is set appropriately
448 GArray* parse_cfile(gchar* f, GError** e) {
449 const char* DEFAULT_ERROR = "Could not parse %s in group %s: %s";
450 const char* MISSING_REQUIRED_ERROR = "Could not find required value %s in group %s: %s";
452 gchar *virtstyle=NULL;
454 { "exportname", TRUE, PARAM_STRING, NULL, 0 },
455 { "port", TRUE, PARAM_INT, NULL, 0 },
456 { "authfile", FALSE, PARAM_STRING, NULL, 0 },
457 { "timeout", FALSE, PARAM_INT, NULL, 0 },
458 { "filesize", FALSE, PARAM_INT, NULL, 0 },
459 { "virtstyle", FALSE, PARAM_STRING, NULL, 0 },
460 { "readonly", FALSE, PARAM_BOOL, NULL, F_READONLY },
461 { "multifile", FALSE, PARAM_BOOL, NULL, F_MULTIFILE },
462 { "copyonwrite", FALSE, PARAM_BOOL, NULL, F_COPYONWRITE },
463 { "autoreadonly", FALSE, PARAM_BOOL, NULL, F_AUTOREADONLY },
464 { "sparse_cow", FALSE, PARAM_BOOL, NULL, F_SPARSE },
466 const int lp_size=11;
468 { "user", FALSE, PARAM_STRING, &runuser, 0 },
469 { "group", FALSE, PARAM_STRING, &rungroup, 0 },
475 const char *err_msg=NULL;
482 errdomain = g_quark_from_string("parse_cfile");
483 cfile = g_key_file_new();
484 retval = g_array_new(FALSE, TRUE, sizeof(SERVER));
485 if(!g_key_file_load_from_file(cfile, f, G_KEY_FILE_KEEP_COMMENTS |
486 G_KEY_FILE_KEEP_TRANSLATIONS, &err)) {
487 g_set_error(e, errdomain, CFILE_NOTFOUND, "Could not open config file.");
488 g_key_file_free(cfile);
491 if(strcmp(g_key_file_get_start_group(cfile), "generic")) {
492 g_set_error(e, errdomain, CFILE_MISSING_GENERIC, "Config file does not contain the [generic] group!");
493 g_key_file_free(cfile);
496 groups = g_key_file_get_groups(cfile, NULL);
497 for(i=0;groups[i];i++) {
498 memset(&s, '\0', sizeof(SERVER));
499 lp[0].target=&(s.exportname);
500 lp[1].target=&(s.port);
501 lp[2].target=&(s.authname);
502 lp[3].target=&(s.timeout);
503 lp[4].target=&(s.expected_size);
504 lp[5].target=&(virtstyle);
505 lp[6].target=lp[7].target=lp[8].target=
506 lp[9].target=lp[10].target=&(s.flags);
507 /* After the [generic] group, start parsing exports */
512 for(j=0;j<p_size;j++) {
513 g_assert(p[j].target != NULL);
514 g_assert(p[j].ptype==PARAM_INT||p[j].ptype==PARAM_STRING||p[j].ptype==PARAM_BOOL);
517 *((gint*)p[j].target) =
518 g_key_file_get_integer(cfile,
524 *((gchar**)p[j].target) =
525 g_key_file_get_string(cfile,
531 value = g_key_file_get_boolean(cfile,
533 p[j].paramname, &err);
536 *((gint*)p[j].target) |= p[j].flagval;
538 *((gint*)p[j].target) &= ~(p[j].flagval);
544 if(err->code == G_KEY_FILE_ERROR_KEY_NOT_FOUND) {
546 /* Ignore not-found error for optional values */
550 err_msg = MISSING_REQUIRED_ERROR;
553 err_msg = DEFAULT_ERROR;
555 g_set_error(e, errdomain, CFILE_VALUE_INVALID, err_msg, p[j].paramname, groups[i], err->message);
556 g_array_free(retval, TRUE);
558 g_key_file_free(cfile);
563 if(!strncmp(virtstyle, "none", 4)) {
564 s.virtstyle=VIRT_NONE;
565 } else if(!strncmp(virtstyle, "ipliteral", 9)) {
566 s.virtstyle=VIRT_IPLIT;
567 } else if(!strncmp(virtstyle, "iphash", 6)) {
568 s.virtstyle=VIRT_IPHASH;
569 } else if(!strncmp(virtstyle, "cidrhash", 8)) {
570 s.virtstyle=VIRT_CIDR;
571 if(strlen(virtstyle)<10) {
572 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s: missing length", virtstyle, groups[i]);
573 g_array_free(retval, TRUE);
574 g_key_file_free(cfile);
577 s.cidrlen=strtol(virtstyle+8, NULL, 0);
579 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Invalid value %s for parameter virtstyle in group %s", virtstyle, groups[i]);
580 g_array_free(retval, TRUE);
581 g_key_file_free(cfile);
585 s.virtstyle=VIRT_IPLIT;
587 /* Don't need to free this, it's not our string */
589 /* Don't append values for the [generic] group */
591 g_array_append_val(retval, s);
598 * Signal handler for SIGCHLD
599 * @param s the signal we're handling (must be SIGCHLD, or something
602 void sigchld_handler(int s) {
607 while((pid=waitpid(-1, &status, WNOHANG)) > 0) {
608 if(WIFEXITED(&status)) {
609 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
611 i=g_hash_table_lookup(children, &pid);
613 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
615 DEBUG2("Removing %d from the list of children", pid);
616 g_hash_table_remove(children, &pid);
622 * Kill a child. Called from sigterm_handler::g_hash_table_foreach.
625 * @param value the value corresponding to the above key
626 * @param user_data a pointer which we always set to 1, so that we know what
629 void killchild(gpointer key, gpointer value, gpointer user_data) {
631 int *parent=user_data;
638 * Handle SIGTERM and dispatch it to our children
639 * @param s the signal we're handling (must be SIGTERM, or something
640 * is severely wrong).
642 void sigterm_handler(int s) {
645 g_hash_table_foreach(children, killchild, &parent);
655 * Detect the size of a file.
657 * @param fhandle An open filedescriptor
658 * @return the size of the file, or OFFT_MAX if detection was
661 off_t size_autodetect(int fhandle) {
664 struct stat stat_buf;
667 #ifdef HAVE_SYS_MOUNT_H
668 #ifdef HAVE_SYS_IOCTL_H
670 DEBUG("looking for fhandle size with ioctl BLKGETSIZE\n");
671 if (!ioctl(fhandle, BLKGETSIZE, &es32) && es32) {
672 es = (off_t)es32 * (off_t)512;
675 #endif /* BLKGETSIZE */
676 #endif /* HAVE_SYS_IOCTL_H */
677 #endif /* HAVE_SYS_MOUNT_H */
679 DEBUG("looking for fhandle size with fstat\n");
680 stat_buf.st_size = 0;
681 error = fstat(fhandle, &stat_buf);
683 if(stat_buf.st_size > 0)
684 return (off_t)stat_buf.st_size;
686 err("fstat failed: %m");
689 DEBUG("looking for fhandle size with lseek SEEK_END\n");
690 es = lseek(fhandle, (off_t)0, SEEK_END);
691 if (es > ((off_t)0)) {
694 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
697 err("Could not find size of exported block device: %m");
702 * Get the file handle and offset, given an export offset.
704 * @param export An array of export files
705 * @param a The offset to get corresponding file/offset for
706 * @param fhandle [out] File descriptor
707 * @param foffset [out] Offset into fhandle
708 * @param maxbytes [out] Tells how many bytes can be read/written
709 * from fhandle starting at foffset (0 if there is no limit)
710 * @return 0 on success, -1 on failure
712 int get_filepos(GArray* export, off_t a, int* fhandle, off_t* foffset, size_t* maxbytes ) {
713 /* Negative offset not allowed */
717 /* Binary search for last file with starting offset <= a */
720 int end = export->len - 1;
721 while( start <= end ) {
722 int mid = (start + end) / 2;
723 fi = g_array_index(export, FILE_INFO, mid);
724 if( fi.startoff < a ) {
726 } else if( fi.startoff > a ) {
734 /* end should never go negative, since first startoff is 0 and a >= 0 */
737 fi = g_array_index(export, FILE_INFO, end);
738 *fhandle = fi.fhandle;
739 *foffset = a - fi.startoff;
741 if( end+1 < export->len ) {
742 FILE_INFO fi_next = g_array_index(export, FILE_INFO, end+1);
743 *maxbytes = fi_next.startoff - a;
750 * seek to a position in a file, with error handling.
751 * @param handle a filedescriptor
752 * @param a position to seek to
753 * @todo get rid of this; lastpoint is a global variable right now, but it
754 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
757 void myseek(int handle,off_t a) {
758 if (lseek(handle, a, SEEK_SET) < 0) {
759 err("Can not seek locally!\n");
764 * Write an amount of bytes at a given offset to the right file. This
765 * abstracts the write-side of the multiple file option.
767 * @param a The offset where the write should start
768 * @param buf The buffer to write from
769 * @param len The length of buf
770 * @param client The client we're serving for
771 * @return The number of bytes actually written, or -1 in case of an error
773 ssize_t rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
778 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
780 if(maxbytes && len > maxbytes)
783 DEBUG4("(WRITE to fd %d offset %Lu len %u), ", fhandle, foffset, len);
785 myseek(fhandle, foffset);
786 return write(fhandle, buf, len);
790 * Call rawexpwrite repeatedly until all data has been written.
791 * @return 0 on success, nonzero on failure
793 int rawexpwrite_fully(off_t a, char *buf, size_t len, CLIENT *client) {
796 while(len > 0 && (ret=rawexpwrite(a, buf, len, client)) > 0 ) {
801 return (ret < 0 || len != 0);
805 * Read an amount of bytes at a given offset from the right file. This
806 * abstracts the read-side of the multiple files option.
808 * @param a The offset where the read should start
809 * @param buf A buffer to read into
810 * @param len The size of buf
811 * @param client The client we're serving for
812 * @return The number of bytes actually read, or -1 in case of an
815 ssize_t rawexpread(off_t a, char *buf, size_t len, CLIENT *client) {
820 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
822 if(maxbytes && len > maxbytes)
825 DEBUG4("(READ from fd %d offset %Lu len %u), ", fhandle, foffset, len);
827 myseek(fhandle, foffset);
828 return read(fhandle, buf, len);
832 * Call rawexpread repeatedly until all data has been read.
833 * @return 0 on success, nonzero on failure
835 int rawexpread_fully(off_t a, char *buf, size_t len, CLIENT *client) {
838 while(len > 0 && (ret=rawexpread(a, buf, len, client)) > 0 ) {
843 return (ret < 0 || len != 0);
847 * Read an amount of bytes at a given offset from the right file. This
848 * abstracts the read-side of the copyonwrite stuff, and calls
849 * rawexpread() with the right parameters to do the actual work.
850 * @param a The offset where the read should start
851 * @param buf A buffer to read into
852 * @param len The size of buf
853 * @param client The client we're going to read for
854 * @return 0 on success, nonzero on failure
856 int expread(off_t a, char *buf, size_t len, CLIENT *client) {
858 off_t mapcnt, mapl, maph, pagestart;
860 if (!(client->server->flags & F_COPYONWRITE))
861 return(rawexpread_fully(a, buf, len, client));
862 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
864 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
866 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
867 pagestart=mapcnt*DIFFPAGESIZE;
869 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
870 len : (size_t)DIFFPAGESIZE-offset;
871 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
872 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
873 (unsigned long)(client->difmap[mapcnt]));
874 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
875 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
876 } else { /* the block is not there */
877 DEBUG2("Page %Lu is not here, we read the original one\n",
878 (unsigned long long)mapcnt);
879 if(rawexpread_fully(a, buf, rdlen, client)) return -1;
881 len-=rdlen; a+=rdlen; buf+=rdlen;
887 * Write an amount of bytes at a given offset to the right file. This
888 * abstracts the write-side of the copyonwrite option, and calls
889 * rawexpwrite() with the right parameters to do the actual work.
891 * @param a The offset where the write should start
892 * @param buf The buffer to write from
893 * @param len The length of buf
894 * @param client The client we're going to write for.
895 * @return 0 on success, nonzero on failure
897 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
898 char pagebuf[DIFFPAGESIZE];
899 off_t mapcnt,mapl,maph;
904 if (!(client->server->flags & F_COPYONWRITE))
905 return(rawexpwrite_fully(a, buf, len, client));
906 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
908 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
910 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
911 pagestart=mapcnt*DIFFPAGESIZE ;
913 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
914 len : (size_t)DIFFPAGESIZE-offset;
916 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
917 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
918 (unsigned long)(client->difmap[mapcnt])) ;
919 myseek(client->difffile,
920 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
921 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
922 } else { /* the block is not there */
923 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
924 client->difmap[mapcnt]=(client->server->flags&F_SPARSE)?mapcnt:client->difffilelen++;
925 DEBUG3("Page %Lu is not here, we put it at %lu\n",
926 (unsigned long long)mapcnt,
927 (unsigned long)(client->difmap[mapcnt]));
929 if (rawexpread_fully(pagestart, pagebuf, rdlen, client))
931 memcpy(pagebuf+offset,buf,wrlen) ;
932 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
936 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
942 * Do the initial negotiation.
944 * @param client The client we're negotiating with.
946 void negotiate(CLIENT *client) {
950 memset(zeros, '\0', 290);
951 if (write(client->net, INIT_PASSWD, 8) < 0)
952 err("Negotiation failed: %m");
953 cliserv_magic = htonll(cliserv_magic);
954 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
955 err("Negotiation failed: %m");
956 size_host = htonll((u64)(client->exportsize));
957 if (write(client->net, &size_host, 8) < 0)
958 err("Negotiation failed: %m");
959 if (write(client->net, zeros, 128) < 0)
960 err("Negotiation failed: %m");
963 /** sending macro. */
964 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
966 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; }
968 * Serve a file to a single client.
970 * @todo This beast needs to be split up in many tiny little manageable
971 * pieces. Preferably with a chainsaw.
973 * @param client The client we're going to serve to.
976 int mainloop(CLIENT *client) {
977 struct nbd_request request;
978 struct nbd_reply reply;
984 DEBUG("Entering request loop!\n");
985 reply.magic = htonl(NBD_REPLY_MAGIC);
994 if (client->server->timeout)
995 alarm(client->server->timeout);
996 readit(client->net, &request, sizeof(request));
997 request.from = ntohll(request.from);
998 request.type = ntohl(request.type);
1000 if (request.type==NBD_CMD_DISC) {
1001 msg2(LOG_INFO, "Disconnect request received.");
1002 if (client->server->flags & F_COPYONWRITE) {
1003 if (client->difmap) g_free(client->difmap) ;
1004 close(client->difffile);
1005 unlink(client->difffilename);
1006 free(client->difffilename);
1012 len = ntohl(request.len);
1014 if (request.magic != htonl(NBD_REQUEST_MAGIC))
1015 err("Not enough magic.");
1016 if (len > BUFSIZE + sizeof(struct nbd_reply))
1017 err("Request too big!");
1019 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
1020 "READ", (unsigned long long)request.from,
1021 (unsigned long long)request.from / 512, len);
1023 memcpy(reply.handle, request.handle, sizeof(reply.handle));
1024 if ((request.from + len) > (OFFT_MAX)) {
1025 DEBUG("[Number too large!]");
1026 ERROR(client, reply);
1030 if (((ssize_t)((off_t)request.from + len) > client->exportsize)) {
1032 ERROR(client, reply);
1036 if (request.type==NBD_CMD_WRITE) {
1037 DEBUG("wr: net->buf, ");
1038 readit(client->net, buf, len);
1039 DEBUG("buf->exp, ");
1040 if ((client->server->flags & F_READONLY) ||
1041 (client->server->flags & F_AUTOREADONLY)) {
1042 DEBUG("[WRITE to READONLY!]");
1043 ERROR(client, reply);
1046 if (expwrite(request.from, buf, len, client)) {
1047 DEBUG("Write failed: %m" );
1048 ERROR(client, reply);
1051 SEND(client->net, reply);
1057 DEBUG("exp->buf, ");
1058 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
1059 DEBUG("Read failed: %m");
1060 ERROR(client, reply);
1064 DEBUG("buf->net, ");
1065 memcpy(buf, &reply, sizeof(struct nbd_reply));
1066 writeit(client->net, buf, len + sizeof(struct nbd_reply));
1073 * Set up client export array, which is an array of FILE_INFO.
1074 * Also, split a single exportfile into multiple ones, if that was asked.
1075 * @param client information on the client which we want to setup export for
1077 void setupexport(CLIENT* client) {
1079 off_t laststartoff = 0, lastsize = 0;
1080 int multifile = (client->server->flags & F_MULTIFILE);
1082 client->export = g_array_new(TRUE, TRUE, sizeof(FILE_INFO));
1084 /* If multi-file, open as many files as we can.
1085 * If not, open exactly one file.
1086 * Calculate file sizes as we go to get total size. */
1090 mode_t mode = (client->server->flags & F_READONLY) ? O_RDONLY : O_RDWR;
1093 tmpname=g_strdup_printf("%s.%d", client->exportname, i);
1095 tmpname=g_strdup(client->exportname);
1097 DEBUG2( "Opening %s\n", tmpname );
1098 fi.fhandle = open(tmpname, mode);
1099 if(fi.fhandle == -1 && mode == O_RDWR) {
1100 /* Try again because maybe media was read-only */
1101 fi.fhandle = open(tmpname, O_RDONLY);
1102 if(fi.fhandle != -1) {
1103 client->server->flags |= F_AUTOREADONLY;
1104 client->server->flags |= F_READONLY;
1107 if(fi.fhandle == -1) {
1108 if(multifile && i>0)
1110 err("Could not open exported file: %m");
1112 fi.startoff = laststartoff + lastsize;
1113 g_array_append_val(client->export, fi);
1116 /* Starting offset and size of this file will be used to
1117 * calculate starting offset of next file */
1118 laststartoff = fi.startoff;
1119 lastsize = size_autodetect(fi.fhandle);
1125 /* Set export size to total calculated size */
1126 client->exportsize = laststartoff + lastsize;
1128 /* Export size may be overridden */
1129 if(client->server->expected_size) {
1130 /* desired size must be <= total calculated size */
1131 if(client->server->expected_size > client->exportsize) {
1132 err("Size of exported file is too big\n");
1135 client->exportsize = client->server->expected_size;
1138 msg3(LOG_INFO, "Size of exported file/device is %Lu", (unsigned long long)client->exportsize);
1140 msg3(LOG_INFO, "Total number of files: %d", i);
1144 int copyonwrite_prepare(CLIENT* client) {
1146 if ((client->difffilename = malloc(1024))==NULL)
1147 err("Failed to allocate string for diff file name");
1148 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
1150 client->difffilename[1023]='\0';
1151 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
1152 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
1153 if (client->difffile<0) err("Could not create diff file (%m)") ;
1154 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
1155 err("Could not allocate memory") ;
1156 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
1162 * Serve a connection.
1164 * @todo allow for multithreading, perhaps use libevent. Not just yet, though;
1165 * follow the road map.
1167 * @param client a connected client
1169 void serveconnection(CLIENT *client) {
1170 setupexport(client);
1172 if (client->server->flags & F_COPYONWRITE) {
1173 copyonwrite_prepare(client);
1176 setmysockopt(client->net);
1182 * Find the name of the file we have to serve. This will use g_strdup_printf
1183 * to put the IP address of the client inside a filename containing
1184 * "%s" (in the form as specified by the "virtstyle" option). That name
1185 * is then written to client->exportname.
1187 * @param net A socket connected to an nbd client
1188 * @param client information about the client. The IP address in human-readable
1189 * format will be written to a new char* buffer, the address of which will be
1190 * stored in client->clientname.
1192 void set_peername(int net, CLIENT *client) {
1193 struct sockaddr_in addrin;
1194 struct sockaddr_in netaddr;
1195 size_t addrinlen = sizeof( addrin );
1201 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
1202 err("getsockname failed: %m");
1203 peername = inet_ntoa(addrin.sin_addr);
1204 switch(client->server->virtstyle) {
1206 client->exportname=g_strdup(client->server->exportname);
1209 for(i=0;i<strlen(peername);i++) {
1210 if(peername[i]=='.') {
1215 client->exportname=g_strdup_printf(client->server->exportname, peername);
1218 memcpy(&netaddr, &addrin, addrinlen);
1219 netaddr.sin_addr.s_addr=netaddr.sin_addr.s_addr>>(client->server->cidrlen)<<(client->server->cidrlen);
1220 netname = inet_ntoa(netaddr.sin_addr);
1221 tmp=g_strdup_printf("%s/%s", netname, peername);
1222 client->exportname=g_strdup_printf(client->server->exportname, tmp);
1226 msg4(LOG_INFO, "connect from %s, assigned file is %s",
1227 peername, client->exportname);
1228 client->clientname=g_strdup(peername);
1233 * @param data a pointer to pid_t which should be freed
1235 void destroy_pid_t(gpointer data) {
1240 * Go daemon (unless we specified at compile time that we didn't want this)
1241 * @param serve the first server of our configuration. If its port is zero,
1242 * then do not daemonize, because we're doing inetd then. This parameter
1243 * is only used to create a PID file of the form
1244 * /var/run/nbd-server.<port>.pid; it's not modified in any way.
1246 #if !defined(NODAEMON) && !defined(NOFORK)
1247 void daemonize(SERVER* serve) {
1254 snprintf(pidfname, sizeof(char)*255, "/var/run/nbd-server.%d.pid", serve->port);
1256 strncpy(pidfname, "/var/run/nbd-server.pid", sizeof(char)*255);
1258 pidf=fopen(pidfname, "w");
1260 fprintf(pidf,"%d\n", (int)getpid());
1264 fprintf(stderr, "Not fatal; continuing");
1268 #define daemonize(serve)
1269 #endif /* !defined(NODAEMON) && !defined(NOFORK) */
1272 * Connect a server's socket.
1274 * @param serve the server we want to connect.
1276 void setup_serve(SERVER *serve) {
1277 struct sockaddr_in addrin;
1278 struct sigaction sa;
1279 int addrinlen = sizeof(addrin);
1286 if ((serve->socket = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
1289 /* lose the pesky "Address already in use" error message */
1290 if (setsockopt(serve->socket,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
1291 err("setsockopt SO_REUSEADDR");
1293 if (setsockopt(serve->socket,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
1294 err("setsockopt SO_KEEPALIVE");
1297 /* make the listening socket non-blocking */
1298 if ((sock_flags = fcntl(serve->socket, F_GETFL, 0)) == -1) {
1299 err("fcntl F_GETFL");
1301 if (fcntl(serve->socket, F_SETFL, sock_flags | O_NONBLOCK) == -1) {
1302 err("fcntl F_SETFL O_NONBLOCK");
1305 DEBUG("Waiting for connections... bind, ");
1306 addrin.sin_family = AF_INET;
1307 addrin.sin_port = htons(serve->port);
1308 addrin.sin_addr.s_addr = 0;
1309 if (bind(serve->socket, (struct sockaddr *) &addrin, addrinlen) < 0)
1312 if (listen(serve->socket, 1) < 0)
1314 sa.sa_handler = sigchld_handler;
1315 sigemptyset(&sa.sa_mask);
1316 sa.sa_flags = SA_RESTART;
1317 if(sigaction(SIGCHLD, &sa, NULL) == -1)
1318 err("sigaction: %m");
1319 sa.sa_handler = sigterm_handler;
1320 sigemptyset(&sa.sa_mask);
1321 sa.sa_flags = SA_RESTART;
1322 if(sigaction(SIGTERM, &sa, NULL) == -1)
1323 err("sigaction: %m");
1324 children=g_hash_table_new_full(g_int_hash, g_int_equal, NULL, destroy_pid_t);
1328 * Connect our servers.
1330 void setup_servers(GArray* servers) {
1333 for(i=0;i<servers->len;i++) {
1334 setup_serve(&(g_array_index(servers, SERVER, i)));
1339 * Loop through the available servers, and serve them.
1341 int serveloop(GArray* servers) {
1342 struct sockaddr_in addrin;
1343 socklen_t addrinlen=sizeof(addrin);
1353 * Set up the master fd_set. The set of descriptors we need
1354 * to select() for never changes anyway and it buys us a *lot*
1355 * of time to only build this once. However, if we ever choose
1356 * to not fork() for clients anymore, we may have to revisit
1361 for(i=0;i<servers->len;i++) {
1362 sock=(g_array_index(servers, SERVER, i)).socket;
1363 FD_SET(sock, &mset);
1364 max=sock>max?sock:max;
1371 memcpy(&rset, &mset, sizeof(fd_set));
1374 if(select(max+1, &rset, NULL, NULL, &tv)>0) {
1376 for(i=0;i<servers->len;i++) {
1377 serve=&(g_array_index(servers, SERVER, i));
1378 if(FD_ISSET(serve->socket, &rset)) {
1379 if ((net=accept(serve->socket, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1382 client = g_malloc(sizeof(CLIENT));
1383 client->server=serve;
1384 client->exportsize=OFFT_MAX;
1386 set_peername(net, client);
1387 if (!authorized_client(client)) {
1388 msg2(LOG_INFO,"Unauthorized client") ;
1392 msg2(LOG_INFO,"Authorized client") ;
1393 pid=g_malloc(sizeof(pid_t));
1395 if ((*pid=fork())<0) {
1396 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
1400 if (*pid>0) { /* parent */
1402 g_hash_table_insert(children, pid, pid);
1406 g_hash_table_destroy(children);
1407 for(i=0;i<servers->len,serve=(g_array_index(servers, SERVER*, i));i++) {
1408 close(serve->socket);
1410 /* FALSE does not free the
1411 actual data. This is required,
1412 because the client has a
1413 direct reference into that
1414 data, and otherwise we get a
1416 g_array_free(servers, FALSE);
1418 msg2(LOG_INFO,"Starting to serve");
1419 serveconnection(client);
1427 * Set up user-ID and/or group-ID
1429 void dousers(void) {
1433 pw=getpwnam(runuser);
1434 if(setuid(pw->pw_uid)<0)
1435 msg3(LOG_DEBUG, "Could not set UID: %s", strerror(errno));
1438 gr=getgrnam(rungroup);
1439 if(setgid(gr->gr_gid)<0)
1440 msg3(LOG_DEBUG, "Could not set GID: %s", strerror(errno));
1445 * Main entry point...
1447 int main(int argc, char *argv[]) {
1452 if (sizeof( struct nbd_request )!=28) {
1453 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1458 config_file_pos = g_strdup(CFILE);
1459 serve=cmdline(argc, argv);
1460 servers = parse_cfile(config_file_pos, &err);
1461 if(!servers || !servers->len) {
1462 g_warning("Could not parse config file: %s", err->message);
1465 g_array_append_val(servers, *serve);
1468 /* We don't support this at this time */
1470 if (!(serve->port)) {
1473 /* You really should define ISSERVER if you're going to use
1474 * inetd mode, but if you don't, closing stdout and stderr
1475 * (which inetd had connected to the client socket) will let it
1479 open("/dev/null", O_WRONLY);
1480 open("/dev/null", O_WRONLY);
1482 client=g_malloc(sizeof(CLIENT));
1483 client->server=serve;
1485 client->exportsize=OFFT_MAX;
1486 set_peername(0,client);
1487 serveconnection(client);
1491 if((!serve) && (!servers||!servers->len)) {
1492 g_message("Nothing to do! Bye!");
1496 setup_servers(servers);