2 * Network Block Device - server
4 * Copyright 1996-1998 Pavel Machek, distribute under GPL
5 * <pavel@atrey.karlin.mff.cuni.cz>
6 * Copyright 2001-2004 Wouter Verhelst <wouter@debian.org>
7 * Copyright 2002 Anton Altaparmakov <aia21@cam.ac.uk>
9 * Version 1.0 - hopefully 64-bit-clean
10 * Version 1.1 - merging enhancements from Josh Parsons, <josh@coombs.anu.edu.au>
11 * Version 1.2 - autodetect size of block devices, thanx to Peter T. Breuer" <ptb@it.uc3m.es>
12 * Version 1.5 - can compile on Unix systems that don't have 64 bit integer
13 * type, or don't have 64 bit file offsets by defining FS_32BIT
14 * in compile options for nbd-server *only*. This can be done
15 * with make FSCHOICE=-DFS_32BIT nbd-server. (I don't have the
16 * original autoconf input file, or I would make it a configure
17 * option.) Ken Yap <ken@nlc.net.au>.
18 * Version 1.6 - fix autodetection of block device size and really make 64 bit
19 * clean on 32 bit machines. Anton Altaparmakov <aia21@cam.ac.uk>
20 * Version 2.0 - Version synchronised with client
21 * Version 2.1 - Reap zombie client processes when they exit. Removed
22 * (uncommented) the _IO magic, it's no longer necessary. Wouter
23 * Verhelst <wouter@debian.org>
24 * Version 2.2 - Auto switch to read-only mode (usefull for floppies).
25 * Version 2.3 - Fixed code so that Large File Support works. This
26 * removes the FS_32BIT compile-time directive; define
27 * _FILE_OFFSET_BITS=64 and _LARGEFILE_SOURCE if you used to be
28 * using FS_32BIT. This will allow you to use files >2GB instead of
29 * having to use the -m option. Wouter Verhelst <wouter@debian.org>
30 * Version 2.4 - Added code to keep track of children, so that we can
31 * properly kill them from initscripts. Add a call to daemon(),
32 * so that processes don't think they have to wait for us, which is
33 * interesting for initscripts as well. Wouter Verhelst
35 * Version 2.5 - Bugfix release: forgot to reset child_arraysize to
36 * zero after fork()ing, resulting in nbd-server going berserk
37 * when it receives a signal with at least one child open. Wouter
38 * Verhelst <wouter@debian.org>
39 * 10/10/2003 - Added socket option SO_KEEPALIVE (sf.net bug 819235);
40 * rectified type of mainloop::size_host (sf.net bugs 814435 and
41 * 817385); close the PID file after writing to it, so that the
42 * daemon can actually be found. Wouter Verhelst
44 * 10/10/2003 - Size of the data "size_host" was wrong and so was not
45 * correctly put in network endianness. Many types were corrected
46 * (size_t and off_t instead of int). <vspaceg@sourceforge.net>
47 * Version 2.6 - Some code cleanup.
48 * Version 2.7 - Better build system.
49 * 11/02/2004 - Doxygenified the source, modularized it a bit. Needs a
50 * lot more work, but this is a start. Wouter Verhelst
54 /* Includes LFS defines, which defines behaviours of some of the following
55 * headers, so must come before those */
58 #include <sys/types.h>
59 #include <sys/socket.h>
61 #include <sys/select.h> /* select */
62 #include <sys/wait.h> /* wait */
63 #ifdef HAVE_SYS_IOCTL_H
64 #include <sys/ioctl.h>
66 #include <sys/param.h>
67 #ifdef HAVE_SYS_MOUNT_H
68 #include <sys/mount.h> /* For BLKGETSIZE */
70 #include <signal.h> /* sigaction */
71 #include <netinet/tcp.h>
72 #include <netinet/in.h> /* sockaddr_in, htons, in_addr */
73 #include <netdb.h> /* hostent, gethostby*, getservby* */
80 #include <arpa/inet.h>
88 /* used in cliserv.h, so must come first */
89 #define MY_NAME "nbd_server"
92 /** Default position of the config file */
94 #define SYSCONFDIR "/etc"
96 #define CFILE SYSCONFDIR "/nbd-server/config"
98 /** Where our config file actually is */
99 gchar* config_file_pos;
101 /** Logging macros, now nothing goes to syslog unless you say ISSERVER */
103 #define msg2(a,b) syslog(a,b)
104 #define msg3(a,b,c) syslog(a,b,c)
105 #define msg4(a,b,c,d) syslog(a,b,c,d)
107 #define msg2(a,b) g_message(b)
108 #define msg3(a,b,c) g_message(b,c)
109 #define msg4(a,b,c,d) g_message(b,c,d)
112 /* Debugging macros */
115 #define DEBUG( a ) printf( a )
116 #define DEBUG2( a,b ) printf( a,b )
117 #define DEBUG3( a,b,c ) printf( a,b,c )
118 #define DEBUG4( a,b,c,d ) printf( a,b,c,d )
121 #define DEBUG2( a,b )
122 #define DEBUG3( a,b,c )
123 #define DEBUG4( a,b,c,d )
125 #ifndef PACKAGE_VERSION
126 #define PACKAGE_VERSION ""
129 * The highest value a variable of type off_t can reach. This is a signed
130 * integer, so set all bits except for the leftmost one.
132 #define OFFT_MAX ~((off_t)1<<(sizeof(off_t)*8-1))
133 #define LINELEN 256 /**< Size of static buffer used to read the
134 authorization file (yuck) */
135 #define BUFSIZE (1024*1024) /**< Size of buffer that can hold requests */
136 #define DIFFPAGESIZE 4096 /**< diff file uses those chunks */
137 #define F_READONLY 1 /**< flag to tell us a file is readonly */
138 #define F_MULTIFILE 2 /**< flag to tell us a file is exported using -m */
139 #define F_COPYONWRITE 4 /**< flag to tell us a file is exported using
141 #define F_AUTOREADONLY 8 /**< flag to tell us a file is set to autoreadonly */
142 GHashTable *children;
143 char pidfname[256]; /**< name of our PID file */
144 char default_authname[] = "/etc/nbd_server.allow"; /**< default name of allow file */
147 * Variables associated with a server.
150 gchar* exportname; /**< (unprocessed) filename of the file we're exporting */
151 off_t expected_size; /**< size of the exported file as it was told to
152 us through configuration */
153 unsigned int port; /**< port we're exporting this file at */
154 char* authname; /**< filename of the authorization file */
155 int flags; /**< flags associated with this exported file */
156 unsigned int timeout;/**< how long a connection may be idle
158 int socket; /**< The socket of this server. */
162 * Variables associated with a client socket.
165 int fhandle; /**< file descriptor */
166 off_t startoff; /**< starting offset of this file */
170 off_t exportsize; /**< size of the file we're exporting */
171 char *clientname; /**< peer */
172 char *exportname; /**< (processed) filename of the file we're exporting */
173 GArray *export; /**< array of FILE_INFO of exported files;
174 array size is always 1 unless we're
175 doing the multiple file option */
176 int net; /**< The actual client socket */
177 SERVER *server; /**< The server this client is getting data from */
178 char* difffilename; /**< filename of the copy-on-write file, if any */
179 int difffile; /**< filedescriptor of copyonwrite file. @todo
180 shouldn't this be an array too? (cfr export) Or
181 make -m and -c mutually exclusive */
182 u32 difffilelen; /**< number of pages in difffile */
183 u32 *difmap; /**< see comment on the global difmap for this one */
187 * Type of configuration file values
190 PARAM_INT, /**< This parameter is an integer */
191 PARAM_STRING, /**< This parameter is a string */
192 PARAM_BOOL, /**< This parameter is a boolean */
195 * Configuration file values
198 gchar *paramname; /**< Name of the parameter, as it appears in
200 gboolean required; /**< Whether this is a required (as opposed to
201 optional) parameter */
202 PARAM_TYPE ptype; /**< Type of the parameter. */
203 gpointer target; /**< Pointer to where the data of this
204 parameter should be written. If ptype is
205 PARAM_BOOL, the data is or'ed rather than
207 gint flagval; /**< Flag mask for this parameter in case ptype
212 * Check whether a client is allowed to connect. Works with an authorization
213 * file which contains one line per machine, no wildcards.
215 * @param opts The client who's trying to connect.
216 * @return 0 - authorization refused, 1 - OK
218 int authorized_client(CLIENT *opts) {
223 if ((f=fopen(opts->server->authname,"r"))==NULL) {
224 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
225 opts->server->authname,strerror(errno)) ;
229 while (fgets(line,LINELEN,f)!=NULL) {
230 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
240 * Read data from a file descriptor into a buffer
242 * @param f a file descriptor
243 * @param buf a buffer
244 * @param len the number of bytes to be read
246 inline void readit(int f, void *buf, size_t len) {
250 if ((res = read(f, buf, len)) <= 0)
251 err("Read failed: %m");
258 * Write data from a buffer into a filedescriptor
260 * @param f a file descriptor
261 * @param buf a buffer containing data
262 * @param len the number of bytes to be written
264 inline void writeit(int f, void *buf, size_t len) {
268 if ((res = write(f, buf, len)) <= 0)
269 err("Send failed: %m");
276 * Print out a message about how to use nbd-server. Split out to a separate
277 * function so that we can call it from multiple places
280 printf("This is nbd-server version " VERSION "\n");
281 printf("Usage: port file_to_export [size][kKmM] [-l authorize_file] [-r] [-m] [-c] [-a timeout_sec] [-C configuration file]\n"
282 "\t-r|--read-only\t\tread only\n"
283 "\t-m|--multi-file\t\tmultiple file\n"
284 "\t-c|--copy-on-write\tcopy on write\n"
285 "\t-C|--config-file\tspecify an alternat configuration file\n"
286 "\t-l|--authorize-file\tfile with list of hosts that are allowed to\n\t\t\t\tconnect.\n"
287 "\t-a|--idle-time\t\tmaximum idle seconds; server terminates when\n\t\t\t\tidle time exceeded\n\n"
288 "\tif port is set to 0, stdin is used (for running from inetd)\n"
289 "\tif file_to_export contains '%%s', it is substituted with the IP\n"
290 "\t\taddress of the machine trying to connect\n" );
291 printf("Using configuration file %s\n", CFILE);
295 * Parse the command line.
297 * @param argc the argc argument to main()
298 * @param argv the argv argument to main()
300 SERVER* cmdline(int argc, char *argv[]) {
304 struct option long_options[] = {
305 {"read-only", no_argument, NULL, 'r'},
306 {"multi-file", no_argument, NULL, 'm'},
307 {"copy-on-write", no_argument, NULL, 'c'},
308 {"authorize-file", required_argument, NULL, 'l'},
309 {"idle-time", required_argument, NULL, 'a'},
310 {"config-file", required_argument, NULL, 'C'},
321 serve=g_new0(SERVER, 1);
322 serve->authname = g_strdup(default_authname);
323 while((c=getopt_long(argc, argv, "-a:C:cl:mr", long_options, &i))>=0) {
326 /* non-option argument */
327 switch(nonspecial++) {
329 serve->port=strtol(optarg, NULL, 0);
332 serve->exportname = g_strdup(optarg);
333 if(serve->exportname[0] != '/') {
334 fprintf(stderr, "E: The to be exported file needs to be an absolute filename!\n");
339 last=strlen(optarg)-1;
341 if (suffix == 'k' || suffix == 'K' ||
342 suffix == 'm' || suffix == 'M')
344 es = (off_t)atol(optarg);
352 serve->expected_size = es;
357 serve->flags |= F_READONLY;
360 serve->flags |= F_MULTIFILE;
363 serve->flags |=F_COPYONWRITE;
366 g_free(config_file_pos);
367 config_file_pos=g_strdup(optarg);
370 g_free(serve->authname);
371 serve->authname=g_strdup(optarg);
374 serve->timeout=strtol(optarg, NULL, 0);
382 /* What's left: the port to export, the name of the to be exported
383 * file, and, optionally, the size of the file, in that order. */
392 * Error codes for config file parsing
395 CFILE_NOTFOUND, /**< The configuration file is not found */
396 CFILE_MISSING_GENERIC, /**< The (required) group "generic" is missing */
397 CFILE_KEY_MISSING, /**< A (required) key is missing */
398 CFILE_VALUE_INVALID, /**< A value is syntactically invalid */
399 CFILE_PROGERR /**< Programmer error */
403 * Remove a SERVER from memory. Used from the hash table
405 void remove_server(gpointer s) {
409 g_free(server->exportname);
411 g_free(server->authname);
416 * Parse the config file.
418 * @param f the name of the config file
419 * @param e a GError. @see CFILE_ERRORS for what error values this function can
421 * @return a GHashTable of SERVER* pointers, with the port number as the hash
422 * key. If the config file is empty or does not exist, returns an empty
423 * GHashTable; if the config file contains an error, returns NULL, and
424 * e is set appropriately
426 GArray* parse_cfile(gchar* f, GError** e) {
427 const char* DEFAULT_ERROR = "Could not parse %s in group %s: %s";
428 const char* MISSING_REQUIRED_ERROR = "Could not find required value %s in group %s: %s";
431 { "exportname", TRUE, PARAM_STRING, NULL, 0 },
432 { "port", TRUE, PARAM_INT, NULL, 0 },
433 { "authfile", FALSE, PARAM_STRING, NULL, 0 },
434 { "timeout", FALSE, PARAM_INT, NULL, 0 },
435 { "filesize", FALSE, PARAM_INT, NULL, 0 },
436 { "readonly", FALSE, PARAM_BOOL, NULL, F_READONLY },
437 { "multifile", FALSE, PARAM_BOOL, NULL, F_MULTIFILE },
438 { "copyonwrite", FALSE, PARAM_BOOL, NULL, F_COPYONWRITE },
443 const char *err_msg=NULL;
450 memset(&s, '\0', sizeof(SERVER));
451 errdomain = g_quark_from_string("parse_cfile");
452 cfile = g_key_file_new();
453 retval = g_array_new(FALSE, TRUE, sizeof(SERVER));
454 if(!g_key_file_load_from_file(cfile, f, G_KEY_FILE_KEEP_COMMENTS |
455 G_KEY_FILE_KEEP_TRANSLATIONS, &err)) {
456 g_set_error(e, errdomain, CFILE_NOTFOUND, "Could not open config file.");
457 g_key_file_free(cfile);
460 if(strcmp(g_key_file_get_start_group(cfile), "generic")) {
461 g_set_error(e, errdomain, CFILE_MISSING_GENERIC, "Config file does not contain the [generic] group!");
462 g_key_file_free(cfile);
465 groups = g_key_file_get_groups(cfile, NULL);
466 for(i=1;groups[i];i++) {
467 p[0].target=&(s.exportname);
468 p[1].target=&(s.port);
469 p[2].target=&(s.authname);
470 p[3].target=&(s.timeout);
471 p[4].target=&(s.expected_size);
472 p[5].target=p[6].target=p[7].target=&(s.flags);
473 for(j=0;j<p_size;j++) {
474 g_assert(p[j].target != NULL);
475 g_assert(p[j].ptype==PARAM_INT||p[j].ptype==PARAM_STRING||p[j].ptype==PARAM_BOOL);
478 *((gint*)p[j].target) =
479 g_key_file_get_integer(cfile,
485 *((gchar**)p[j].target) =
486 g_key_file_get_string(cfile,
492 value = g_key_file_get_boolean(cfile,
494 p[j].paramname, &err);
496 *((gint*)p[j].target) |= p[j].flagval;
501 if(err->code == G_KEY_FILE_ERROR_KEY_NOT_FOUND) {
503 /* Ignore not-found error for optional values */
507 err_msg = MISSING_REQUIRED_ERROR;
510 err_msg = DEFAULT_ERROR;
512 g_set_error(e, errdomain, CFILE_VALUE_INVALID, err_msg, p[j].paramname, groups[i], err->message);
513 g_array_free(retval, TRUE);
515 g_key_file_free(cfile);
519 g_array_append_val(retval, s);
525 * Signal handler for SIGCHLD
526 * @param s the signal we're handling (must be SIGCHLD, or something
529 void sigchld_handler(int s) {
534 while((pid=waitpid(-1, &status, WNOHANG)) > 0) {
535 if(WIFEXITED(&status)) {
536 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
538 i=g_hash_table_lookup(children, &pid);
540 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
542 DEBUG2("Removing %d from the list of children", pid);
543 g_hash_table_remove(children, &pid);
549 * Kill a child. Called from sigterm_handler::g_hash_table_foreach.
552 * @param value the value corresponding to the above key
553 * @param user_data a pointer which we always set to 1, so that we know what
556 void killchild(gpointer key, gpointer value, gpointer user_data) {
558 int *parent=user_data;
565 * Handle SIGTERM and dispatch it to our children
566 * @param s the signal we're handling (must be SIGTERM, or something
567 * is severely wrong).
569 void sigterm_handler(int s) {
572 g_hash_table_foreach(children, killchild, &parent);
582 * Detect the size of a file.
584 * @param fhandle An open filedescriptor
585 * @return the size of the file, or OFFT_MAX if detection was
588 off_t size_autodetect(int fhandle) {
591 struct stat stat_buf;
594 #ifdef HAVE_SYS_MOUNT_H
595 #ifdef HAVE_SYS_IOCTL_H
597 DEBUG("looking for fhandle size with ioctl BLKGETSIZE\n");
598 if (!ioctl(fhandle, BLKGETSIZE, &es32) && es32) {
599 es = (off_t)es32 * (off_t)512;
602 #endif /* BLKGETSIZE */
603 #endif /* HAVE_SYS_IOCTL_H */
604 #endif /* HAVE_SYS_MOUNT_H */
606 DEBUG("looking for fhandle size with fstat\n");
607 stat_buf.st_size = 0;
608 error = fstat(fhandle, &stat_buf);
610 if(stat_buf.st_size > 0)
611 return (off_t)stat_buf.st_size;
613 err("fstat failed: %m");
616 DEBUG("looking for fhandle size with lseek SEEK_END\n");
617 es = lseek(fhandle, (off_t)0, SEEK_END);
618 if (es > ((off_t)0)) {
621 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
624 err("Could not find size of exported block device: %m");
629 * Get the file handle and offset, given an export offset.
631 * @param export An array of export files
632 * @param a The offset to get corresponding file/offset for
633 * @param fhandle [out] File descriptor
634 * @param foffset [out] Offset into fhandle
635 * @param maxbytes [out] Tells how many bytes can be read/written
636 * from fhandle starting at foffset (0 if there is no limit)
637 * @return 0 on success, -1 on failure
639 int get_filepos(GArray* export, off_t a, int* fhandle, off_t* foffset, size_t* maxbytes ) {
640 /* Negative offset not allowed */
644 /* Binary search for last file with starting offset <= a */
647 int end = export->len - 1;
648 while( start <= end ) {
649 int mid = (start + end) / 2;
650 fi = g_array_index(export, FILE_INFO, mid);
651 if( fi.startoff < a ) {
653 } else if( fi.startoff > a ) {
661 /* end should never go negative, since first startoff is 0 and a >= 0 */
664 fi = g_array_index(export, FILE_INFO, end);
665 *fhandle = fi.fhandle;
666 *foffset = a - fi.startoff;
668 if( end+1 < export->len ) {
669 FILE_INFO fi_next = g_array_index(export, FILE_INFO, end+1);
670 *maxbytes = fi_next.startoff - a;
677 * seek to a position in a file, with error handling.
678 * @param handle a filedescriptor
679 * @param a position to seek to
680 * @todo get rid of this; lastpoint is a global variable right now, but it
681 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
684 void myseek(int handle,off_t a) {
685 if (lseek(handle, a, SEEK_SET) < 0) {
686 err("Can not seek locally!\n");
691 * Write an amount of bytes at a given offset to the right file. This
692 * abstracts the write-side of the multiple file option.
694 * @param a The offset where the write should start
695 * @param buf The buffer to write from
696 * @param len The length of buf
697 * @param client The client we're serving for
698 * @return The number of bytes actually written, or -1 in case of an error
700 ssize_t rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
705 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
707 if(maxbytes && len > maxbytes)
710 DEBUG4("(WRITE to fd %d offset %Lu len %u), ", fhandle, foffset, len);
712 myseek(fhandle, foffset);
713 return write(fhandle, buf, len);
717 * Call rawexpwrite repeatedly until all data has been written.
718 * @return 0 on success, nonzero on failure
720 int rawexpwrite_fully(off_t a, char *buf, size_t len, CLIENT *client) {
723 while(len > 0 && (ret=rawexpwrite(a, buf, len, client)) > 0 ) {
728 return (ret < 0 || len != 0);
732 * Read an amount of bytes at a given offset from the right file. This
733 * abstracts the read-side of the multiple files option.
735 * @param a The offset where the read should start
736 * @param buf A buffer to read into
737 * @param len The size of buf
738 * @param client The client we're serving for
739 * @return The number of bytes actually read, or -1 in case of an
742 ssize_t rawexpread(off_t a, char *buf, size_t len, CLIENT *client) {
747 if(get_filepos(client->export, a, &fhandle, &foffset, &maxbytes))
749 if(maxbytes && len > maxbytes)
752 DEBUG4("(READ from fd %d offset %Lu len %u), ", fhandle, foffset, len);
754 myseek(fhandle, foffset);
755 return read(fhandle, buf, len);
759 * Call rawexpread repeatedly until all data has been read.
760 * @return 0 on success, nonzero on failure
762 int rawexpread_fully(off_t a, char *buf, size_t len, CLIENT *client) {
765 while(len > 0 && (ret=rawexpread(a, buf, len, client)) > 0 ) {
770 return (ret < 0 || len != 0);
774 * Read an amount of bytes at a given offset from the right file. This
775 * abstracts the read-side of the copyonwrite stuff, and calls
776 * rawexpread() with the right parameters to do the actual work.
777 * @param a The offset where the read should start
778 * @param buf A buffer to read into
779 * @param len The size of buf
780 * @param client The client we're going to read for
781 * @return 0 on success, nonzero on failure
783 int expread(off_t a, char *buf, size_t len, CLIENT *client) {
785 off_t mapcnt, mapl, maph, pagestart;
787 if (!(client->server->flags & F_COPYONWRITE))
788 return(rawexpread_fully(a, buf, len, client));
789 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
791 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
793 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
794 pagestart=mapcnt*DIFFPAGESIZE;
796 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
797 len : (size_t)DIFFPAGESIZE-offset;
798 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
799 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
800 (unsigned long)(client->difmap[mapcnt]));
801 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
802 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
803 } else { /* the block is not there */
804 DEBUG2("Page %Lu is not here, we read the original one\n",
805 (unsigned long long)mapcnt);
806 if(rawexpread_fully(a, buf, rdlen, client)) return -1;
808 len-=rdlen; a+=rdlen; buf+=rdlen;
814 * Write an amount of bytes at a given offset to the right file. This
815 * abstracts the write-side of the copyonwrite option, and calls
816 * rawexpwrite() with the right parameters to do the actual work.
818 * @param a The offset where the write should start
819 * @param buf The buffer to write from
820 * @param len The length of buf
821 * @param client The client we're going to write for.
822 * @return 0 on success, nonzero on failure
824 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
825 char pagebuf[DIFFPAGESIZE];
826 off_t mapcnt,mapl,maph;
831 if (!(client->server->flags & F_COPYONWRITE))
832 return(rawexpwrite_fully(a, buf, len, client));
833 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
835 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
837 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
838 pagestart=mapcnt*DIFFPAGESIZE ;
840 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
841 len : (size_t)DIFFPAGESIZE-offset;
843 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
844 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
845 (unsigned long)(client->difmap[mapcnt])) ;
846 myseek(client->difffile,
847 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
848 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
849 } else { /* the block is not there */
850 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
851 client->difmap[mapcnt]=client->difffilelen++ ;
852 DEBUG3("Page %Lu is not here, we put it at %lu\n",
853 (unsigned long long)mapcnt,
854 (unsigned long)(client->difmap[mapcnt]));
856 if (rawexpread_fully(pagestart, pagebuf, rdlen, client))
858 memcpy(pagebuf+offset,buf,wrlen) ;
859 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
863 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
869 * Do the initial negotiation.
871 * @param client The client we're negotiating with.
873 void negotiate(CLIENT *client) {
877 memset(zeros, 0, 290);
878 if (write(client->net, INIT_PASSWD, 8) < 0)
879 err("Negotiation failed: %m");
880 cliserv_magic = htonll(cliserv_magic);
881 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
882 err("Negotiation failed: %m");
883 size_host = htonll((u64)(client->exportsize));
884 if (write(client->net, &size_host, 8) < 0)
885 err("Negotiation failed: %m");
886 if (write(client->net, zeros, 128) < 0)
887 err("Negotiation failed: %m");
890 /** sending macro. */
891 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
893 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; }
895 * Serve a file to a single client.
897 * @todo This beast needs to be split up in many tiny little manageable
898 * pieces. Preferably with a chainsaw.
900 * @param client The client we're going to serve to.
903 int mainloop(CLIENT *client) {
904 struct nbd_request request;
905 struct nbd_reply reply;
911 DEBUG("Entering request loop!\n");
912 reply.magic = htonl(NBD_REPLY_MAGIC);
921 if (client->server->timeout)
922 alarm(client->server->timeout);
923 readit(client->net, &request, sizeof(request));
924 request.from = ntohll(request.from);
925 request.type = ntohl(request.type);
927 if (request.type==NBD_CMD_DISC) {
928 msg2(LOG_INFO, "Disconnect request received.");
929 if (client->difmap) g_free(client->difmap) ;
930 if (client->difffile>=0) {
931 close(client->difffile);
932 unlink(client->difffilename);
933 free(client->difffilename);
939 len = ntohl(request.len);
941 if (request.magic != htonl(NBD_REQUEST_MAGIC))
942 err("Not enough magic.");
943 if (len > BUFSIZE + sizeof(struct nbd_reply))
944 err("Request too big!");
946 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
947 "READ", (unsigned long long)request.from,
948 (unsigned long long)request.from / 512, len);
950 memcpy(reply.handle, request.handle, sizeof(reply.handle));
951 if ((request.from + len) > (OFFT_MAX)) {
952 DEBUG("[Number too large!]");
953 ERROR(client, reply);
957 if (((ssize_t)((off_t)request.from + len) > client->exportsize)) {
959 ERROR(client, reply);
963 if (request.type==NBD_CMD_WRITE) {
964 DEBUG("wr: net->buf, ");
965 readit(client->net, buf, len);
967 if ((client->server->flags & F_READONLY) ||
968 (client->server->flags & F_AUTOREADONLY)) {
969 DEBUG("[WRITE to READONLY!]");
970 ERROR(client, reply);
973 if (expwrite(request.from, buf, len, client)) {
974 DEBUG("Write failed: %m" );
975 ERROR(client, reply);
978 SEND(client->net, reply);
985 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
986 DEBUG("Read failed: %m");
987 ERROR(client, reply);
992 memcpy(buf, &reply, sizeof(struct nbd_reply));
993 writeit(client->net, buf, len + sizeof(struct nbd_reply));
1000 * Set up client export array, which is an array of FILE_INFO.
1001 * Also, split a single exportfile into multiple ones, if that was asked.
1002 * @param client information on the client which we want to setup export for
1004 void setupexport(CLIENT* client) {
1006 off_t laststartoff = 0, lastsize = 0;
1007 int multifile = (client->server->flags & F_MULTIFILE);
1009 client->export = g_array_new(TRUE, TRUE, sizeof(FILE_INFO));
1011 /* If multi-file, open as many files as we can.
1012 * If not, open exactly one file.
1013 * Calculate file sizes as we go to get total size. */
1017 mode_t mode = (client->server->flags & F_READONLY) ? O_RDONLY : O_RDWR;
1020 tmpname=g_strdup_printf("%s.%d", client->exportname, i);
1022 tmpname=g_strdup(client->exportname);
1024 DEBUG2( "Opening %s\n", tmpname );
1025 fi.fhandle = open(tmpname, mode);
1026 if(fi.fhandle == -1 && mode == O_RDWR) {
1027 /* Try again because maybe media was read-only */
1028 fi.fhandle = open(tmpname, O_RDONLY);
1029 if(fi.fhandle != -1) {
1030 client->server->flags |= F_AUTOREADONLY;
1031 client->server->flags |= F_READONLY;
1034 if(fi.fhandle == -1) {
1035 if(multifile && i>0)
1037 err("Could not open exported file: %m");
1039 fi.startoff = laststartoff + lastsize;
1040 g_array_append_val(client->export, fi);
1043 /* Starting offset and size of this file will be used to
1044 * calculate starting offset of next file */
1045 laststartoff = fi.startoff;
1046 lastsize = size_autodetect(fi.fhandle);
1052 /* Set export size to total calculated size */
1053 client->exportsize = laststartoff + lastsize;
1055 /* Export size may be overridden */
1056 if(client->server->expected_size) {
1057 /* desired size must be <= total calculated size */
1058 if(client->server->expected_size > client->exportsize) {
1059 err("Size of exported file is too big\n");
1062 client->exportsize = client->server->expected_size;
1065 msg3(LOG_INFO, "Size of exported file/device is %Lu", (unsigned long long)client->exportsize);
1067 msg3(LOG_INFO, "Total number of files: %d", i);
1071 int copyonwrite_prepare(CLIENT* client) {
1073 if ((client->difffilename = malloc(1024))==NULL)
1074 err("Failed to allocate string for diff file name");
1075 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
1077 client->difffilename[1023]='\0';
1078 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
1079 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
1080 if (client->difffile<0) err("Could not create diff file (%m)") ;
1081 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
1082 err("Could not allocate memory") ;
1083 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
1089 * Serve a connection.
1091 * @todo allow for multithreading, perhaps use libevent. Not just yet, though;
1092 * follow the road map.
1094 * @param client a connected client
1096 void serveconnection(CLIENT *client) {
1097 setupexport(client);
1099 if (client->server->flags & F_COPYONWRITE) {
1100 copyonwrite_prepare(client);
1103 setmysockopt(client->net);
1109 * Find the name of the file we have to serve. This will use g_strdup_printf
1110 * to put the IP address of the client inside a filename containing
1111 * "%s". That name is then written to client->exportname.
1113 * @param net A socket connected to an nbd client
1114 * @param client information about the client. The IP address in human-readable
1115 * format will be written to a new char* buffer, the address of which will be
1116 * stored in client->clientname.
1118 void set_peername(int net, CLIENT *client) {
1119 struct sockaddr_in addrin;
1120 int addrinlen = sizeof( addrin );
1123 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
1124 err("getsockname failed: %m");
1125 peername = inet_ntoa(addrin.sin_addr);
1126 client->exportname=g_strdup_printf(client->server->exportname, peername);
1128 msg4(LOG_INFO, "connect from %s, assigned file is %s",
1129 peername, client->exportname);
1130 client->clientname=g_strdup(peername);
1135 * @param data a pointer to pid_t which should be freed
1137 void destroy_pid_t(gpointer data) {
1142 * Go daemon (unless we specified at compile time that we didn't want this)
1143 * @param serve the first server of our configuration. If its port is zero,
1144 * then do not daemonize, because we're doing inetd then. This parameter
1145 * is only used to create a PID file of the form
1146 * /var/run/nbd-server.<port>.pid; it's not modified in any way.
1148 #if !defined(NODAEMON) && !defined(NOFORK)
1149 void daemonize(SERVER* serve) {
1156 snprintf(pidfname, sizeof(char)*255, "/var/run/nbd-server.%d.pid", serve->port);
1158 strncpy(pidfname, "/var/run/nbd-server.pid", sizeof(char)*255);
1160 pidf=fopen(pidfname, "w");
1162 fprintf(pidf,"%d\n", (int)getpid());
1166 fprintf(stderr, "Not fatal; continuing");
1170 #define daemonize(serve)
1171 #endif /* !defined(NODAEMON) && !defined(NOFORK) */
1174 * Connect a server's socket.
1176 * @param serve the server we want to connect.
1178 void setup_serve(SERVER *serve) {
1179 struct sockaddr_in addrin;
1180 struct sigaction sa;
1181 int addrinlen = sizeof(addrin);
1188 if ((serve->socket = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
1191 /* lose the pesky "Address already in use" error message */
1192 if (setsockopt(serve->socket,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
1193 err("setsockopt SO_REUSEADDR");
1195 if (setsockopt(serve->socket,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
1196 err("setsockopt SO_KEEPALIVE");
1199 /* make the listening socket non-blocking */
1200 if ((sock_flags = fcntl(serve->socket, F_GETFL, 0)) == -1) {
1201 err("fcntl F_GETFL");
1203 if (fcntl(serve->socket, F_SETFL, sock_flags | O_NONBLOCK) == -1) {
1204 err("fcntl F_SETFL O_NONBLOCK");
1207 DEBUG("Waiting for connections... bind, ");
1208 addrin.sin_family = AF_INET;
1209 addrin.sin_port = htons(serve->port);
1210 addrin.sin_addr.s_addr = 0;
1211 if (bind(serve->socket, (struct sockaddr *) &addrin, addrinlen) < 0)
1214 if (listen(serve->socket, 1) < 0)
1216 sa.sa_handler = sigchld_handler;
1217 sigemptyset(&sa.sa_mask);
1218 sa.sa_flags = SA_RESTART;
1219 if(sigaction(SIGCHLD, &sa, NULL) == -1)
1220 err("sigaction: %m");
1221 sa.sa_handler = sigterm_handler;
1222 sigemptyset(&sa.sa_mask);
1223 sa.sa_flags = SA_RESTART;
1224 if(sigaction(SIGTERM, &sa, NULL) == -1)
1225 err("sigaction: %m");
1226 children=g_hash_table_new_full(g_int_hash, g_int_equal, NULL, destroy_pid_t);
1230 * Connect our servers.
1232 void setup_servers(GArray* servers) {
1235 for(i=0;i<servers->len;i++) {
1236 setup_serve(&(g_array_index(servers, SERVER, i)));
1241 * Loop through the available servers, and serve them.
1243 int serveloop(GArray* servers) {
1244 struct sockaddr_in addrin;
1245 socklen_t addrinlen=sizeof(addrin);
1255 * Set up the master fd_set. The set of descriptors we need
1256 * to select() for never changes anyway and it buys us a *lot*
1257 * of time to only build this once. However, if we ever choose
1258 * to not fork() for clients anymore, we may have to revisit
1263 for(i=0;i<servers->len;i++) {
1264 sock=(g_array_index(servers, SERVER, i)).socket;
1265 FD_SET(sock, &mset);
1266 max=sock>max?sock:max;
1273 memcpy(&rset, &mset, sizeof(fd_set));
1276 if(select(max+1, &rset, NULL, NULL, &tv)>0) {
1278 for(i=0;i<servers->len;i++) {
1279 serve=&(g_array_index(servers, SERVER, i));
1280 if(FD_ISSET(serve->socket, &rset)) {
1281 if ((net=accept(serve->socket, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1284 client = g_malloc(sizeof(CLIENT));
1285 client->server=serve;
1286 client->exportsize=OFFT_MAX;
1288 set_peername(net, client);
1289 if (!authorized_client(client)) {
1290 msg2(LOG_INFO,"Unauthorized client") ;
1294 msg2(LOG_INFO,"Authorized client") ;
1295 pid=g_malloc(sizeof(pid_t));
1297 if ((*pid=fork())<0) {
1298 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
1302 if (*pid>0) { /* parent */
1304 g_hash_table_insert(children, pid, pid);
1308 g_hash_table_destroy(children);
1309 for(i=0;i<servers->len,serve=(g_array_index(servers, SERVER*, i));i++) {
1310 close(serve->socket);
1312 /* FALSE does not free the
1313 actual data. This is required,
1314 because the client has a
1315 direct reference into that
1316 data, and otherwise we get a
1318 g_array_free(servers, FALSE);
1320 msg2(LOG_INFO,"Starting to serve");
1321 serveconnection(client);
1329 * Main entry point...
1331 int main(int argc, char *argv[]) {
1336 if (sizeof( struct nbd_request )!=28) {
1337 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1342 config_file_pos = g_strdup(CFILE);
1343 serve=cmdline(argc, argv);
1344 servers = parse_cfile(config_file_pos, &err);
1345 if(!servers || !servers->len) {
1346 g_warning("Could not parse config file: %s", err->message);
1349 g_array_append_val(servers, *serve);
1352 /* We don't support this at this time */
1354 if (!(serve->port)) {
1357 /* You really should define ISSERVER if you're going to use
1358 * inetd mode, but if you don't, closing stdout and stderr
1359 * (which inetd had connected to the client socket) will let it
1363 open("/dev/null", O_WRONLY);
1364 open("/dev/null", O_WRONLY);
1366 client=g_malloc(sizeof(CLIENT));
1367 client->server=serve;
1369 client->exportsize=OFFT_MAX;
1370 set_peername(0,client);
1371 serveconnection(client);
1375 if((!serve) && (!servers||!servers->len)) {
1376 g_message("Nothing to do! Bye!");
1380 setup_servers(servers);